Bug Summary

File:clang/lib/CodeGen/CGBuiltin.cpp
Warning:line 11974, column 22
Division by zero

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name CGBuiltin.cpp -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=cplusplus -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -setup-static-analyzer -analyzer-config-compatibility-mode=true -mrelocation-model pic -pic-level 2 -mframe-pointer=none -relaxed-aliasing -fmath-errno -fno-rounding-math -mconstructor-aliases -munwind-tables -target-cpu x86-64 -tune-cpu generic -debugger-tuning=gdb -ffunction-sections -fdata-sections -fcoverage-compilation-dir=/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/build-llvm/tools/clang/lib/CodeGen -resource-dir /usr/lib/llvm-14/lib/clang/14.0.0 -D CLANG_ROUND_TRIP_CC1_ARGS=ON -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/build-llvm/tools/clang/lib/CodeGen -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/include -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/build-llvm/tools/clang/include -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/build-llvm/include -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include -D NDEBUG -U NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/x86_64-linux-gnu/c++/10 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10/backward -internal-isystem /usr/lib/llvm-14/lib/clang/14.0.0/include -internal-isystem /usr/local/include -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../x86_64-linux-gnu/include -internal-externc-isystem /usr/include/x86_64-linux-gnu -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -Wno-unused-parameter -Wwrite-strings -Wno-missing-field-initializers -Wno-long-long -Wno-maybe-uninitialized -Wno-class-memaccess -Wno-redundant-move -Wno-pessimizing-move -Wno-noexcept-type -Wno-comment -std=c++14 -fdeprecated-macro -fdebug-compilation-dir=/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/build-llvm/tools/clang/lib/CodeGen -fdebug-prefix-map=/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0=. -ferror-limit 19 -fvisibility-inlines-hidden -stack-protector 2 -fgnuc-version=4.2.1 -vectorize-loops -vectorize-slp -analyzer-output=html -analyzer-config stable-report-filename=true -faddrsig -D__GCC_HAVE_DWARF2_CFI_ASM=1 -o /tmp/scan-build-2021-08-28-193554-24367-1 -x c++ /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp
1//===---- CGBuiltin.cpp - Emit LLVM Code for builtins ---------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This contains code to emit Builtin calls as LLVM code.
10//
11//===----------------------------------------------------------------------===//
12
13#include "CGCUDARuntime.h"
14#include "CGCXXABI.h"
15#include "CGObjCRuntime.h"
16#include "CGOpenCLRuntime.h"
17#include "CGRecordLayout.h"
18#include "CodeGenFunction.h"
19#include "CodeGenModule.h"
20#include "ConstantEmitter.h"
21#include "PatternInit.h"
22#include "TargetInfo.h"
23#include "clang/AST/ASTContext.h"
24#include "clang/AST/Attr.h"
25#include "clang/AST/Decl.h"
26#include "clang/AST/OSLog.h"
27#include "clang/Basic/TargetBuiltins.h"
28#include "clang/Basic/TargetInfo.h"
29#include "clang/CodeGen/CGFunctionInfo.h"
30#include "llvm/ADT/APFloat.h"
31#include "llvm/ADT/APInt.h"
32#include "llvm/ADT/SmallPtrSet.h"
33#include "llvm/ADT/StringExtras.h"
34#include "llvm/Analysis/ValueTracking.h"
35#include "llvm/IR/DataLayout.h"
36#include "llvm/IR/InlineAsm.h"
37#include "llvm/IR/Intrinsics.h"
38#include "llvm/IR/IntrinsicsAArch64.h"
39#include "llvm/IR/IntrinsicsAMDGPU.h"
40#include "llvm/IR/IntrinsicsARM.h"
41#include "llvm/IR/IntrinsicsBPF.h"
42#include "llvm/IR/IntrinsicsHexagon.h"
43#include "llvm/IR/IntrinsicsNVPTX.h"
44#include "llvm/IR/IntrinsicsPowerPC.h"
45#include "llvm/IR/IntrinsicsR600.h"
46#include "llvm/IR/IntrinsicsRISCV.h"
47#include "llvm/IR/IntrinsicsS390.h"
48#include "llvm/IR/IntrinsicsWebAssembly.h"
49#include "llvm/IR/IntrinsicsX86.h"
50#include "llvm/IR/MDBuilder.h"
51#include "llvm/IR/MatrixBuilder.h"
52#include "llvm/Support/ConvertUTF.h"
53#include "llvm/Support/ScopedPrinter.h"
54#include "llvm/Support/X86TargetParser.h"
55#include <sstream>
56
57using namespace clang;
58using namespace CodeGen;
59using namespace llvm;
60
61static
62int64_t clamp(int64_t Value, int64_t Low, int64_t High) {
63 return std::min(High, std::max(Low, Value));
64}
65
66static void initializeAlloca(CodeGenFunction &CGF, AllocaInst *AI, Value *Size,
67 Align AlignmentInBytes) {
68 ConstantInt *Byte;
69 switch (CGF.getLangOpts().getTrivialAutoVarInit()) {
70 case LangOptions::TrivialAutoVarInitKind::Uninitialized:
71 // Nothing to initialize.
72 return;
73 case LangOptions::TrivialAutoVarInitKind::Zero:
74 Byte = CGF.Builder.getInt8(0x00);
75 break;
76 case LangOptions::TrivialAutoVarInitKind::Pattern: {
77 llvm::Type *Int8 = llvm::IntegerType::getInt8Ty(CGF.CGM.getLLVMContext());
78 Byte = llvm::dyn_cast<llvm::ConstantInt>(
79 initializationPatternFor(CGF.CGM, Int8));
80 break;
81 }
82 }
83 if (CGF.CGM.stopAutoInit())
84 return;
85 auto *I = CGF.Builder.CreateMemSet(AI, Byte, Size, AlignmentInBytes);
86 I->addAnnotationMetadata("auto-init");
87}
88
89/// getBuiltinLibFunction - Given a builtin id for a function like
90/// "__builtin_fabsf", return a Function* for "fabsf".
91llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD,
92 unsigned BuiltinID) {
93 assert(Context.BuiltinInfo.isLibFunction(BuiltinID))(static_cast <bool> (Context.BuiltinInfo.isLibFunction(
BuiltinID)) ? void (0) : __assert_fail ("Context.BuiltinInfo.isLibFunction(BuiltinID)"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 93, __extension__ __PRETTY_FUNCTION__))
;
94
95 // Get the name, skip over the __builtin_ prefix (if necessary).
96 StringRef Name;
97 GlobalDecl D(FD);
98
99 // If the builtin has been declared explicitly with an assembler label,
100 // use the mangled name. This differs from the plain label on platforms
101 // that prefix labels.
102 if (FD->hasAttr<AsmLabelAttr>())
103 Name = getMangledName(D);
104 else
105 Name = Context.BuiltinInfo.getName(BuiltinID) + 10;
106
107 llvm::FunctionType *Ty =
108 cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType()));
109
110 return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false);
111}
112
113/// Emit the conversions required to turn the given value into an
114/// integer of the given size.
115static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V,
116 QualType T, llvm::IntegerType *IntType) {
117 V = CGF.EmitToMemory(V, T);
118
119 if (V->getType()->isPointerTy())
120 return CGF.Builder.CreatePtrToInt(V, IntType);
121
122 assert(V->getType() == IntType)(static_cast <bool> (V->getType() == IntType) ? void
(0) : __assert_fail ("V->getType() == IntType", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 122, __extension__ __PRETTY_FUNCTION__))
;
123 return V;
124}
125
126static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V,
127 QualType T, llvm::Type *ResultType) {
128 V = CGF.EmitFromMemory(V, T);
129
130 if (ResultType->isPointerTy())
131 return CGF.Builder.CreateIntToPtr(V, ResultType);
132
133 assert(V->getType() == ResultType)(static_cast <bool> (V->getType() == ResultType) ? void
(0) : __assert_fail ("V->getType() == ResultType", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 133, __extension__ __PRETTY_FUNCTION__))
;
134 return V;
135}
136
137/// Utility to insert an atomic instruction based on Intrinsic::ID
138/// and the expression node.
139static Value *MakeBinaryAtomicValue(
140 CodeGenFunction &CGF, llvm::AtomicRMWInst::BinOp Kind, const CallExpr *E,
141 AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) {
142 QualType T = E->getType();
143 assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()->
isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 143, __extension__ __PRETTY_FUNCTION__))
;
144 assert(CGF.getContext().hasSameUnqualifiedType(T,(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(T, E->getArg(0)->getType()->getPointeeType())) ? void
(0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 145, __extension__ __PRETTY_FUNCTION__))
145 E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(T, E->getArg(0)->getType()->getPointeeType())) ? void
(0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 145, __extension__ __PRETTY_FUNCTION__))
;
146 assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(T, E->getArg(1)->getType())) ? void (0) : __assert_fail
("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 146, __extension__ __PRETTY_FUNCTION__))
;
147
148 llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
149 unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
150
151 llvm::IntegerType *IntType =
152 llvm::IntegerType::get(CGF.getLLVMContext(),
153 CGF.getContext().getTypeSize(T));
154 llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
155
156 llvm::Value *Args[2];
157 Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
158 Args[1] = CGF.EmitScalarExpr(E->getArg(1));
159 llvm::Type *ValueType = Args[1]->getType();
160 Args[1] = EmitToInt(CGF, Args[1], T, IntType);
161
162 llvm::Value *Result = CGF.Builder.CreateAtomicRMW(
163 Kind, Args[0], Args[1], Ordering);
164 return EmitFromInt(CGF, Result, T, ValueType);
165}
166
167static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) {
168 Value *Val = CGF.EmitScalarExpr(E->getArg(0));
169 Value *Address = CGF.EmitScalarExpr(E->getArg(1));
170
171 // Convert the type of the pointer to a pointer to the stored type.
172 Val = CGF.EmitToMemory(Val, E->getArg(0)->getType());
173 Value *BC = CGF.Builder.CreateBitCast(
174 Address, llvm::PointerType::getUnqual(Val->getType()), "cast");
175 LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType());
176 LV.setNontemporal(true);
177 CGF.EmitStoreOfScalar(Val, LV, false);
178 return nullptr;
179}
180
181static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) {
182 Value *Address = CGF.EmitScalarExpr(E->getArg(0));
183
184 LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType());
185 LV.setNontemporal(true);
186 return CGF.EmitLoadOfScalar(LV, E->getExprLoc());
187}
188
189static RValue EmitBinaryAtomic(CodeGenFunction &CGF,
190 llvm::AtomicRMWInst::BinOp Kind,
191 const CallExpr *E) {
192 return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E));
193}
194
195/// Utility to insert an atomic instruction based Intrinsic::ID and
196/// the expression node, where the return value is the result of the
197/// operation.
198static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF,
199 llvm::AtomicRMWInst::BinOp Kind,
200 const CallExpr *E,
201 Instruction::BinaryOps Op,
202 bool Invert = false) {
203 QualType T = E->getType();
204 assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()->
isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 204, __extension__ __PRETTY_FUNCTION__))
;
205 assert(CGF.getContext().hasSameUnqualifiedType(T,(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(T, E->getArg(0)->getType()->getPointeeType())) ? void
(0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 206, __extension__ __PRETTY_FUNCTION__))
206 E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(T, E->getArg(0)->getType()->getPointeeType())) ? void
(0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 206, __extension__ __PRETTY_FUNCTION__))
;
207 assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(T, E->getArg(1)->getType())) ? void (0) : __assert_fail
("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 207, __extension__ __PRETTY_FUNCTION__))
;
208
209 llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
210 unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
211
212 llvm::IntegerType *IntType =
213 llvm::IntegerType::get(CGF.getLLVMContext(),
214 CGF.getContext().getTypeSize(T));
215 llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
216
217 llvm::Value *Args[2];
218 Args[1] = CGF.EmitScalarExpr(E->getArg(1));
219 llvm::Type *ValueType = Args[1]->getType();
220 Args[1] = EmitToInt(CGF, Args[1], T, IntType);
221 Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
222
223 llvm::Value *Result = CGF.Builder.CreateAtomicRMW(
224 Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent);
225 Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]);
226 if (Invert)
227 Result =
228 CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result,
229 llvm::ConstantInt::getAllOnesValue(IntType));
230 Result = EmitFromInt(CGF, Result, T, ValueType);
231 return RValue::get(Result);
232}
233
234/// Utility to insert an atomic cmpxchg instruction.
235///
236/// @param CGF The current codegen function.
237/// @param E Builtin call expression to convert to cmpxchg.
238/// arg0 - address to operate on
239/// arg1 - value to compare with
240/// arg2 - new value
241/// @param ReturnBool Specifies whether to return success flag of
242/// cmpxchg result or the old value.
243///
244/// @returns result of cmpxchg, according to ReturnBool
245///
246/// Note: In order to lower Microsoft's _InterlockedCompareExchange* intrinsics
247/// invoke the function EmitAtomicCmpXchgForMSIntrin.
248static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E,
249 bool ReturnBool) {
250 QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType();
251 llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
252 unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
253
254 llvm::IntegerType *IntType = llvm::IntegerType::get(
255 CGF.getLLVMContext(), CGF.getContext().getTypeSize(T));
256 llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
257
258 Value *Args[3];
259 Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
260 Args[1] = CGF.EmitScalarExpr(E->getArg(1));
261 llvm::Type *ValueType = Args[1]->getType();
262 Args[1] = EmitToInt(CGF, Args[1], T, IntType);
263 Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType);
264
265 Value *Pair = CGF.Builder.CreateAtomicCmpXchg(
266 Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent,
267 llvm::AtomicOrdering::SequentiallyConsistent);
268 if (ReturnBool)
269 // Extract boolean success flag and zext it to int.
270 return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1),
271 CGF.ConvertType(E->getType()));
272 else
273 // Extract old value and emit it using the same type as compare value.
274 return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T,
275 ValueType);
276}
277
278/// This function should be invoked to emit atomic cmpxchg for Microsoft's
279/// _InterlockedCompareExchange* intrinsics which have the following signature:
280/// T _InterlockedCompareExchange(T volatile *Destination,
281/// T Exchange,
282/// T Comparand);
283///
284/// Whereas the llvm 'cmpxchg' instruction has the following syntax:
285/// cmpxchg *Destination, Comparand, Exchange.
286/// So we need to swap Comparand and Exchange when invoking
287/// CreateAtomicCmpXchg. That is the reason we could not use the above utility
288/// function MakeAtomicCmpXchgValue since it expects the arguments to be
289/// already swapped.
290
291static
292Value *EmitAtomicCmpXchgForMSIntrin(CodeGenFunction &CGF, const CallExpr *E,
293 AtomicOrdering SuccessOrdering = AtomicOrdering::SequentiallyConsistent) {
294 assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()->
isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 294, __extension__ __PRETTY_FUNCTION__))
;
295 assert(CGF.getContext().hasSameUnqualifiedType((static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
( E->getType(), E->getArg(0)->getType()->getPointeeType
())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 296, __extension__ __PRETTY_FUNCTION__))
296 E->getType(), E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
( E->getType(), E->getArg(0)->getType()->getPointeeType
())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 296, __extension__ __PRETTY_FUNCTION__))
;
297 assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(E->getType(), E->getArg(1)->getType())) ? void (0) :
__assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 298, __extension__ __PRETTY_FUNCTION__))
298 E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(E->getType(), E->getArg(1)->getType())) ? void (0) :
__assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 298, __extension__ __PRETTY_FUNCTION__))
;
299 assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(E->getType(), E->getArg(2)->getType())) ? void (0) :
__assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 300, __extension__ __PRETTY_FUNCTION__))
300 E->getArg(2)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType
(E->getType(), E->getArg(2)->getType())) ? void (0) :
__assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 300, __extension__ __PRETTY_FUNCTION__))
;
301
302 auto *Destination = CGF.EmitScalarExpr(E->getArg(0));
303 auto *Comparand = CGF.EmitScalarExpr(E->getArg(2));
304 auto *Exchange = CGF.EmitScalarExpr(E->getArg(1));
305
306 // For Release ordering, the failure ordering should be Monotonic.
307 auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release ?
308 AtomicOrdering::Monotonic :
309 SuccessOrdering;
310
311 // The atomic instruction is marked volatile for consistency with MSVC. This
312 // blocks the few atomics optimizations that LLVM has. If we want to optimize
313 // _Interlocked* operations in the future, we will have to remove the volatile
314 // marker.
315 auto *Result = CGF.Builder.CreateAtomicCmpXchg(
316 Destination, Comparand, Exchange,
317 SuccessOrdering, FailureOrdering);
318 Result->setVolatile(true);
319 return CGF.Builder.CreateExtractValue(Result, 0);
320}
321
322// 64-bit Microsoft platforms support 128 bit cmpxchg operations. They are
323// prototyped like this:
324//
325// unsigned char _InterlockedCompareExchange128...(
326// __int64 volatile * _Destination,
327// __int64 _ExchangeHigh,
328// __int64 _ExchangeLow,
329// __int64 * _ComparandResult);
330static Value *EmitAtomicCmpXchg128ForMSIntrin(CodeGenFunction &CGF,
331 const CallExpr *E,
332 AtomicOrdering SuccessOrdering) {
333 assert(E->getNumArgs() == 4)(static_cast <bool> (E->getNumArgs() == 4) ? void (0
) : __assert_fail ("E->getNumArgs() == 4", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 333, __extension__ __PRETTY_FUNCTION__))
;
334 llvm::Value *Destination = CGF.EmitScalarExpr(E->getArg(0));
335 llvm::Value *ExchangeHigh = CGF.EmitScalarExpr(E->getArg(1));
336 llvm::Value *ExchangeLow = CGF.EmitScalarExpr(E->getArg(2));
337 llvm::Value *ComparandPtr = CGF.EmitScalarExpr(E->getArg(3));
338
339 assert(Destination->getType()->isPointerTy())(static_cast <bool> (Destination->getType()->isPointerTy
()) ? void (0) : __assert_fail ("Destination->getType()->isPointerTy()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 339, __extension__ __PRETTY_FUNCTION__))
;
340 assert(!ExchangeHigh->getType()->isPointerTy())(static_cast <bool> (!ExchangeHigh->getType()->isPointerTy
()) ? void (0) : __assert_fail ("!ExchangeHigh->getType()->isPointerTy()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 340, __extension__ __PRETTY_FUNCTION__))
;
341 assert(!ExchangeLow->getType()->isPointerTy())(static_cast <bool> (!ExchangeLow->getType()->isPointerTy
()) ? void (0) : __assert_fail ("!ExchangeLow->getType()->isPointerTy()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 341, __extension__ __PRETTY_FUNCTION__))
;
342 assert(ComparandPtr->getType()->isPointerTy())(static_cast <bool> (ComparandPtr->getType()->isPointerTy
()) ? void (0) : __assert_fail ("ComparandPtr->getType()->isPointerTy()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 342, __extension__ __PRETTY_FUNCTION__))
;
343
344 // For Release ordering, the failure ordering should be Monotonic.
345 auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release
346 ? AtomicOrdering::Monotonic
347 : SuccessOrdering;
348
349 // Convert to i128 pointers and values.
350 llvm::Type *Int128Ty = llvm::IntegerType::get(CGF.getLLVMContext(), 128);
351 llvm::Type *Int128PtrTy = Int128Ty->getPointerTo();
352 Destination = CGF.Builder.CreateBitCast(Destination, Int128PtrTy);
353 Address ComparandResult(CGF.Builder.CreateBitCast(ComparandPtr, Int128PtrTy),
354 CGF.getContext().toCharUnitsFromBits(128));
355
356 // (((i128)hi) << 64) | ((i128)lo)
357 ExchangeHigh = CGF.Builder.CreateZExt(ExchangeHigh, Int128Ty);
358 ExchangeLow = CGF.Builder.CreateZExt(ExchangeLow, Int128Ty);
359 ExchangeHigh =
360 CGF.Builder.CreateShl(ExchangeHigh, llvm::ConstantInt::get(Int128Ty, 64));
361 llvm::Value *Exchange = CGF.Builder.CreateOr(ExchangeHigh, ExchangeLow);
362
363 // Load the comparand for the instruction.
364 llvm::Value *Comparand = CGF.Builder.CreateLoad(ComparandResult);
365
366 auto *CXI = CGF.Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange,
367 SuccessOrdering, FailureOrdering);
368
369 // The atomic instruction is marked volatile for consistency with MSVC. This
370 // blocks the few atomics optimizations that LLVM has. If we want to optimize
371 // _Interlocked* operations in the future, we will have to remove the volatile
372 // marker.
373 CXI->setVolatile(true);
374
375 // Store the result as an outparameter.
376 CGF.Builder.CreateStore(CGF.Builder.CreateExtractValue(CXI, 0),
377 ComparandResult);
378
379 // Get the success boolean and zero extend it to i8.
380 Value *Success = CGF.Builder.CreateExtractValue(CXI, 1);
381 return CGF.Builder.CreateZExt(Success, CGF.Int8Ty);
382}
383
384static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E,
385 AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) {
386 assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()->
isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 386, __extension__ __PRETTY_FUNCTION__))
;
387
388 auto *IntTy = CGF.ConvertType(E->getType());
389 auto *Result = CGF.Builder.CreateAtomicRMW(
390 AtomicRMWInst::Add,
391 CGF.EmitScalarExpr(E->getArg(0)),
392 ConstantInt::get(IntTy, 1),
393 Ordering);
394 return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1));
395}
396
397static Value *EmitAtomicDecrementValue(CodeGenFunction &CGF, const CallExpr *E,
398 AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) {
399 assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()->
isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 399, __extension__ __PRETTY_FUNCTION__))
;
400
401 auto *IntTy = CGF.ConvertType(E->getType());
402 auto *Result = CGF.Builder.CreateAtomicRMW(
403 AtomicRMWInst::Sub,
404 CGF.EmitScalarExpr(E->getArg(0)),
405 ConstantInt::get(IntTy, 1),
406 Ordering);
407 return CGF.Builder.CreateSub(Result, ConstantInt::get(IntTy, 1));
408}
409
410// Build a plain volatile load.
411static Value *EmitISOVolatileLoad(CodeGenFunction &CGF, const CallExpr *E) {
412 Value *Ptr = CGF.EmitScalarExpr(E->getArg(0));
413 QualType ElTy = E->getArg(0)->getType()->getPointeeType();
414 CharUnits LoadSize = CGF.getContext().getTypeSizeInChars(ElTy);
415 llvm::Type *ITy =
416 llvm::IntegerType::get(CGF.getLLVMContext(), LoadSize.getQuantity() * 8);
417 Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo());
418 llvm::LoadInst *Load = CGF.Builder.CreateAlignedLoad(ITy, Ptr, LoadSize);
419 Load->setVolatile(true);
420 return Load;
421}
422
423// Build a plain volatile store.
424static Value *EmitISOVolatileStore(CodeGenFunction &CGF, const CallExpr *E) {
425 Value *Ptr = CGF.EmitScalarExpr(E->getArg(0));
426 Value *Value = CGF.EmitScalarExpr(E->getArg(1));
427 QualType ElTy = E->getArg(0)->getType()->getPointeeType();
428 CharUnits StoreSize = CGF.getContext().getTypeSizeInChars(ElTy);
429 llvm::Type *ITy =
430 llvm::IntegerType::get(CGF.getLLVMContext(), StoreSize.getQuantity() * 8);
431 Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo());
432 llvm::StoreInst *Store =
433 CGF.Builder.CreateAlignedStore(Value, Ptr, StoreSize);
434 Store->setVolatile(true);
435 return Store;
436}
437
438// Emit a simple mangled intrinsic that has 1 argument and a return type
439// matching the argument type. Depending on mode, this may be a constrained
440// floating-point intrinsic.
441static Value *emitUnaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF,
442 const CallExpr *E, unsigned IntrinsicID,
443 unsigned ConstrainedIntrinsicID) {
444 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
445
446 if (CGF.Builder.getIsFPConstrained()) {
447 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
448 Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType());
449 return CGF.Builder.CreateConstrainedFPCall(F, { Src0 });
450 } else {
451 Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
452 return CGF.Builder.CreateCall(F, Src0);
453 }
454}
455
456// Emit an intrinsic that has 2 operands of the same type as its result.
457// Depending on mode, this may be a constrained floating-point intrinsic.
458static Value *emitBinaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF,
459 const CallExpr *E, unsigned IntrinsicID,
460 unsigned ConstrainedIntrinsicID) {
461 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
462 llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
463
464 if (CGF.Builder.getIsFPConstrained()) {
465 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
466 Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType());
467 return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1 });
468 } else {
469 Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
470 return CGF.Builder.CreateCall(F, { Src0, Src1 });
471 }
472}
473
474// Emit an intrinsic that has 3 operands of the same type as its result.
475// Depending on mode, this may be a constrained floating-point intrinsic.
476static Value *emitTernaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF,
477 const CallExpr *E, unsigned IntrinsicID,
478 unsigned ConstrainedIntrinsicID) {
479 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
480 llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
481 llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2));
482
483 if (CGF.Builder.getIsFPConstrained()) {
484 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
485 Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType());
486 return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1, Src2 });
487 } else {
488 Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
489 return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 });
490 }
491}
492
493// Emit an intrinsic where all operands are of the same type as the result.
494// Depending on mode, this may be a constrained floating-point intrinsic.
495static Value *emitCallMaybeConstrainedFPBuiltin(CodeGenFunction &CGF,
496 unsigned IntrinsicID,
497 unsigned ConstrainedIntrinsicID,
498 llvm::Type *Ty,
499 ArrayRef<Value *> Args) {
500 Function *F;
501 if (CGF.Builder.getIsFPConstrained())
502 F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Ty);
503 else
504 F = CGF.CGM.getIntrinsic(IntrinsicID, Ty);
505
506 if (CGF.Builder.getIsFPConstrained())
507 return CGF.Builder.CreateConstrainedFPCall(F, Args);
508 else
509 return CGF.Builder.CreateCall(F, Args);
510}
511
512// Emit a simple mangled intrinsic that has 1 argument and a return type
513// matching the argument type.
514static Value *emitUnaryBuiltin(CodeGenFunction &CGF,
515 const CallExpr *E,
516 unsigned IntrinsicID) {
517 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
518
519 Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
520 return CGF.Builder.CreateCall(F, Src0);
521}
522
523// Emit an intrinsic that has 2 operands of the same type as its result.
524static Value *emitBinaryBuiltin(CodeGenFunction &CGF,
525 const CallExpr *E,
526 unsigned IntrinsicID) {
527 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
528 llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
529
530 Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
531 return CGF.Builder.CreateCall(F, { Src0, Src1 });
532}
533
534// Emit an intrinsic that has 3 operands of the same type as its result.
535static Value *emitTernaryBuiltin(CodeGenFunction &CGF,
536 const CallExpr *E,
537 unsigned IntrinsicID) {
538 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
539 llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
540 llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2));
541
542 Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
543 return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 });
544}
545
546// Emit an intrinsic that has 1 float or double operand, and 1 integer.
547static Value *emitFPIntBuiltin(CodeGenFunction &CGF,
548 const CallExpr *E,
549 unsigned IntrinsicID) {
550 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
551 llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
552
553 Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
554 return CGF.Builder.CreateCall(F, {Src0, Src1});
555}
556
557// Emit an intrinsic that has overloaded integer result and fp operand.
558static Value *
559emitMaybeConstrainedFPToIntRoundBuiltin(CodeGenFunction &CGF, const CallExpr *E,
560 unsigned IntrinsicID,
561 unsigned ConstrainedIntrinsicID) {
562 llvm::Type *ResultType = CGF.ConvertType(E->getType());
563 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
564
565 if (CGF.Builder.getIsFPConstrained()) {
566 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
567 Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID,
568 {ResultType, Src0->getType()});
569 return CGF.Builder.CreateConstrainedFPCall(F, {Src0});
570 } else {
571 Function *F =
572 CGF.CGM.getIntrinsic(IntrinsicID, {ResultType, Src0->getType()});
573 return CGF.Builder.CreateCall(F, Src0);
574 }
575}
576
577/// EmitFAbs - Emit a call to @llvm.fabs().
578static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) {
579 Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType());
580 llvm::CallInst *Call = CGF.Builder.CreateCall(F, V);
581 Call->setDoesNotAccessMemory();
582 return Call;
583}
584
585/// Emit the computation of the sign bit for a floating point value. Returns
586/// the i1 sign bit value.
587static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) {
588 LLVMContext &C = CGF.CGM.getLLVMContext();
589
590 llvm::Type *Ty = V->getType();
591 int Width = Ty->getPrimitiveSizeInBits();
592 llvm::Type *IntTy = llvm::IntegerType::get(C, Width);
593 V = CGF.Builder.CreateBitCast(V, IntTy);
594 if (Ty->isPPC_FP128Ty()) {
595 // We want the sign bit of the higher-order double. The bitcast we just
596 // did works as if the double-double was stored to memory and then
597 // read as an i128. The "store" will put the higher-order double in the
598 // lower address in both little- and big-Endian modes, but the "load"
599 // will treat those bits as a different part of the i128: the low bits in
600 // little-Endian, the high bits in big-Endian. Therefore, on big-Endian
601 // we need to shift the high bits down to the low before truncating.
602 Width >>= 1;
603 if (CGF.getTarget().isBigEndian()) {
604 Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width);
605 V = CGF.Builder.CreateLShr(V, ShiftCst);
606 }
607 // We are truncating value in order to extract the higher-order
608 // double, which we will be using to extract the sign from.
609 IntTy = llvm::IntegerType::get(C, Width);
610 V = CGF.Builder.CreateTrunc(V, IntTy);
611 }
612 Value *Zero = llvm::Constant::getNullValue(IntTy);
613 return CGF.Builder.CreateICmpSLT(V, Zero);
614}
615
616static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD,
617 const CallExpr *E, llvm::Constant *calleeValue) {
618 CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD));
619 return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot());
620}
621
622/// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.*
623/// depending on IntrinsicID.
624///
625/// \arg CGF The current codegen function.
626/// \arg IntrinsicID The ID for the Intrinsic we wish to generate.
627/// \arg X The first argument to the llvm.*.with.overflow.*.
628/// \arg Y The second argument to the llvm.*.with.overflow.*.
629/// \arg Carry The carry returned by the llvm.*.with.overflow.*.
630/// \returns The result (i.e. sum/product) returned by the intrinsic.
631static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF,
632 const llvm::Intrinsic::ID IntrinsicID,
633 llvm::Value *X, llvm::Value *Y,
634 llvm::Value *&Carry) {
635 // Make sure we have integers of the same width.
636 assert(X->getType() == Y->getType() &&(static_cast <bool> (X->getType() == Y->getType()
&& "Arguments must be the same type. (Did you forget to make sure both "
"arguments have the same integer width?)") ? void (0) : __assert_fail
("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 638, __extension__ __PRETTY_FUNCTION__))
637 "Arguments must be the same type. (Did you forget to make sure both "(static_cast <bool> (X->getType() == Y->getType()
&& "Arguments must be the same type. (Did you forget to make sure both "
"arguments have the same integer width?)") ? void (0) : __assert_fail
("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 638, __extension__ __PRETTY_FUNCTION__))
638 "arguments have the same integer width?)")(static_cast <bool> (X->getType() == Y->getType()
&& "Arguments must be the same type. (Did you forget to make sure both "
"arguments have the same integer width?)") ? void (0) : __assert_fail
("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 638, __extension__ __PRETTY_FUNCTION__))
;
639
640 Function *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType());
641 llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y});
642 Carry = CGF.Builder.CreateExtractValue(Tmp, 1);
643 return CGF.Builder.CreateExtractValue(Tmp, 0);
644}
645
646static Value *emitRangedBuiltin(CodeGenFunction &CGF,
647 unsigned IntrinsicID,
648 int low, int high) {
649 llvm::MDBuilder MDHelper(CGF.getLLVMContext());
650 llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high));
651 Function *F = CGF.CGM.getIntrinsic(IntrinsicID, {});
652 llvm::Instruction *Call = CGF.Builder.CreateCall(F);
653 Call->setMetadata(llvm::LLVMContext::MD_range, RNode);
654 return Call;
655}
656
657namespace {
658 struct WidthAndSignedness {
659 unsigned Width;
660 bool Signed;
661 };
662}
663
664static WidthAndSignedness
665getIntegerWidthAndSignedness(const clang::ASTContext &context,
666 const clang::QualType Type) {
667 assert(Type->isIntegerType() && "Given type is not an integer.")(static_cast <bool> (Type->isIntegerType() &&
"Given type is not an integer.") ? void (0) : __assert_fail (
"Type->isIntegerType() && \"Given type is not an integer.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 667, __extension__ __PRETTY_FUNCTION__))
;
668 unsigned Width = Type->isBooleanType() ? 1
669 : Type->isExtIntType() ? context.getIntWidth(Type)
670 : context.getTypeInfo(Type).Width;
671 bool Signed = Type->isSignedIntegerType();
672 return {Width, Signed};
673}
674
675// Given one or more integer types, this function produces an integer type that
676// encompasses them: any value in one of the given types could be expressed in
677// the encompassing type.
678static struct WidthAndSignedness
679EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) {
680 assert(Types.size() > 0 && "Empty list of types.")(static_cast <bool> (Types.size() > 0 && "Empty list of types."
) ? void (0) : __assert_fail ("Types.size() > 0 && \"Empty list of types.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 680, __extension__ __PRETTY_FUNCTION__))
;
681
682 // If any of the given types is signed, we must return a signed type.
683 bool Signed = false;
684 for (const auto &Type : Types) {
685 Signed |= Type.Signed;
686 }
687
688 // The encompassing type must have a width greater than or equal to the width
689 // of the specified types. Additionally, if the encompassing type is signed,
690 // its width must be strictly greater than the width of any unsigned types
691 // given.
692 unsigned Width = 0;
693 for (const auto &Type : Types) {
694 unsigned MinWidth = Type.Width + (Signed && !Type.Signed);
695 if (Width < MinWidth) {
696 Width = MinWidth;
697 }
698 }
699
700 return {Width, Signed};
701}
702
703Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) {
704 llvm::Type *DestType = Int8PtrTy;
705 if (ArgValue->getType() != DestType)
706 ArgValue =
707 Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data());
708
709 Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend;
710 return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue);
711}
712
713/// Checks if using the result of __builtin_object_size(p, @p From) in place of
714/// __builtin_object_size(p, @p To) is correct
715static bool areBOSTypesCompatible(int From, int To) {
716 // Note: Our __builtin_object_size implementation currently treats Type=0 and
717 // Type=2 identically. Encoding this implementation detail here may make
718 // improving __builtin_object_size difficult in the future, so it's omitted.
719 return From == To || (From == 0 && To == 1) || (From == 3 && To == 2);
720}
721
722static llvm::Value *
723getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) {
724 return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true);
725}
726
727llvm::Value *
728CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type,
729 llvm::IntegerType *ResType,
730 llvm::Value *EmittedE,
731 bool IsDynamic) {
732 uint64_t ObjectSize;
733 if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type))
734 return emitBuiltinObjectSize(E, Type, ResType, EmittedE, IsDynamic);
735 return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true);
736}
737
738/// Returns a Value corresponding to the size of the given expression.
739/// This Value may be either of the following:
740/// - A llvm::Argument (if E is a param with the pass_object_size attribute on
741/// it)
742/// - A call to the @llvm.objectsize intrinsic
743///
744/// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null
745/// and we wouldn't otherwise try to reference a pass_object_size parameter,
746/// we'll call @llvm.objectsize on EmittedE, rather than emitting E.
747llvm::Value *
748CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type,
749 llvm::IntegerType *ResType,
750 llvm::Value *EmittedE, bool IsDynamic) {
751 // We need to reference an argument if the pointer is a parameter with the
752 // pass_object_size attribute.
753 if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) {
754 auto *Param = dyn_cast<ParmVarDecl>(D->getDecl());
755 auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>();
756 if (Param != nullptr && PS != nullptr &&
757 areBOSTypesCompatible(PS->getType(), Type)) {
758 auto Iter = SizeArguments.find(Param);
759 assert(Iter != SizeArguments.end())(static_cast <bool> (Iter != SizeArguments.end()) ? void
(0) : __assert_fail ("Iter != SizeArguments.end()", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 759, __extension__ __PRETTY_FUNCTION__))
;
760
761 const ImplicitParamDecl *D = Iter->second;
762 auto DIter = LocalDeclMap.find(D);
763 assert(DIter != LocalDeclMap.end())(static_cast <bool> (DIter != LocalDeclMap.end()) ? void
(0) : __assert_fail ("DIter != LocalDeclMap.end()", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 763, __extension__ __PRETTY_FUNCTION__))
;
764
765 return EmitLoadOfScalar(DIter->second, /*Volatile=*/false,
766 getContext().getSizeType(), E->getBeginLoc());
767 }
768 }
769
770 // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't
771 // evaluate E for side-effects. In either case, we shouldn't lower to
772 // @llvm.objectsize.
773 if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext())))
774 return getDefaultBuiltinObjectSizeResult(Type, ResType);
775
776 Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E);
777 assert(Ptr->getType()->isPointerTy() &&(static_cast <bool> (Ptr->getType()->isPointerTy(
) && "Non-pointer passed to __builtin_object_size?") ?
void (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 778, __extension__ __PRETTY_FUNCTION__))
778 "Non-pointer passed to __builtin_object_size?")(static_cast <bool> (Ptr->getType()->isPointerTy(
) && "Non-pointer passed to __builtin_object_size?") ?
void (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 778, __extension__ __PRETTY_FUNCTION__))
;
779
780 Function *F =
781 CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()});
782
783 // LLVM only supports 0 and 2, make sure that we pass along that as a boolean.
784 Value *Min = Builder.getInt1((Type & 2) != 0);
785 // For GCC compatibility, __builtin_object_size treat NULL as unknown size.
786 Value *NullIsUnknown = Builder.getTrue();
787 Value *Dynamic = Builder.getInt1(IsDynamic);
788 return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown, Dynamic});
789}
790
791namespace {
792/// A struct to generically describe a bit test intrinsic.
793struct BitTest {
794 enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set };
795 enum InterlockingKind : uint8_t {
796 Unlocked,
797 Sequential,
798 Acquire,
799 Release,
800 NoFence
801 };
802
803 ActionKind Action;
804 InterlockingKind Interlocking;
805 bool Is64Bit;
806
807 static BitTest decodeBitTestBuiltin(unsigned BuiltinID);
808};
809} // namespace
810
811BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) {
812 switch (BuiltinID) {
813 // Main portable variants.
814 case Builtin::BI_bittest:
815 return {TestOnly, Unlocked, false};
816 case Builtin::BI_bittestandcomplement:
817 return {Complement, Unlocked, false};
818 case Builtin::BI_bittestandreset:
819 return {Reset, Unlocked, false};
820 case Builtin::BI_bittestandset:
821 return {Set, Unlocked, false};
822 case Builtin::BI_interlockedbittestandreset:
823 return {Reset, Sequential, false};
824 case Builtin::BI_interlockedbittestandset:
825 return {Set, Sequential, false};
826
827 // X86-specific 64-bit variants.
828 case Builtin::BI_bittest64:
829 return {TestOnly, Unlocked, true};
830 case Builtin::BI_bittestandcomplement64:
831 return {Complement, Unlocked, true};
832 case Builtin::BI_bittestandreset64:
833 return {Reset, Unlocked, true};
834 case Builtin::BI_bittestandset64:
835 return {Set, Unlocked, true};
836 case Builtin::BI_interlockedbittestandreset64:
837 return {Reset, Sequential, true};
838 case Builtin::BI_interlockedbittestandset64:
839 return {Set, Sequential, true};
840
841 // ARM/AArch64-specific ordering variants.
842 case Builtin::BI_interlockedbittestandset_acq:
843 return {Set, Acquire, false};
844 case Builtin::BI_interlockedbittestandset_rel:
845 return {Set, Release, false};
846 case Builtin::BI_interlockedbittestandset_nf:
847 return {Set, NoFence, false};
848 case Builtin::BI_interlockedbittestandreset_acq:
849 return {Reset, Acquire, false};
850 case Builtin::BI_interlockedbittestandreset_rel:
851 return {Reset, Release, false};
852 case Builtin::BI_interlockedbittestandreset_nf:
853 return {Reset, NoFence, false};
854 }
855 llvm_unreachable("expected only bittest intrinsics")::llvm::llvm_unreachable_internal("expected only bittest intrinsics"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 855)
;
856}
857
858static char bitActionToX86BTCode(BitTest::ActionKind A) {
859 switch (A) {
860 case BitTest::TestOnly: return '\0';
861 case BitTest::Complement: return 'c';
862 case BitTest::Reset: return 'r';
863 case BitTest::Set: return 's';
864 }
865 llvm_unreachable("invalid action")::llvm::llvm_unreachable_internal("invalid action", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 865)
;
866}
867
868static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF,
869 BitTest BT,
870 const CallExpr *E, Value *BitBase,
871 Value *BitPos) {
872 char Action = bitActionToX86BTCode(BT.Action);
873 char SizeSuffix = BT.Is64Bit ? 'q' : 'l';
874
875 // Build the assembly.
876 SmallString<64> Asm;
877 raw_svector_ostream AsmOS(Asm);
878 if (BT.Interlocking != BitTest::Unlocked)
879 AsmOS << "lock ";
880 AsmOS << "bt";
881 if (Action)
882 AsmOS << Action;
883 AsmOS << SizeSuffix << " $2, ($1)";
884
885 // Build the constraints. FIXME: We should support immediates when possible.
886 std::string Constraints = "={@ccc},r,r,~{cc},~{memory}";
887 std::string MachineClobbers = CGF.getTarget().getClobbers();
888 if (!MachineClobbers.empty()) {
889 Constraints += ',';
890 Constraints += MachineClobbers;
891 }
892 llvm::IntegerType *IntType = llvm::IntegerType::get(
893 CGF.getLLVMContext(),
894 CGF.getContext().getTypeSize(E->getArg(1)->getType()));
895 llvm::Type *IntPtrType = IntType->getPointerTo();
896 llvm::FunctionType *FTy =
897 llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false);
898
899 llvm::InlineAsm *IA =
900 llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true);
901 return CGF.Builder.CreateCall(IA, {BitBase, BitPos});
902}
903
904static llvm::AtomicOrdering
905getBitTestAtomicOrdering(BitTest::InterlockingKind I) {
906 switch (I) {
907 case BitTest::Unlocked: return llvm::AtomicOrdering::NotAtomic;
908 case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent;
909 case BitTest::Acquire: return llvm::AtomicOrdering::Acquire;
910 case BitTest::Release: return llvm::AtomicOrdering::Release;
911 case BitTest::NoFence: return llvm::AtomicOrdering::Monotonic;
912 }
913 llvm_unreachable("invalid interlocking")::llvm::llvm_unreachable_internal("invalid interlocking", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 913)
;
914}
915
916/// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of
917/// bits and a bit position and read and optionally modify the bit at that
918/// position. The position index can be arbitrarily large, i.e. it can be larger
919/// than 31 or 63, so we need an indexed load in the general case.
920static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF,
921 unsigned BuiltinID,
922 const CallExpr *E) {
923 Value *BitBase = CGF.EmitScalarExpr(E->getArg(0));
924 Value *BitPos = CGF.EmitScalarExpr(E->getArg(1));
925
926 BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID);
927
928 // X86 has special BT, BTC, BTR, and BTS instructions that handle the array
929 // indexing operation internally. Use them if possible.
930 if (CGF.getTarget().getTriple().isX86())
931 return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos);
932
933 // Otherwise, use generic code to load one byte and test the bit. Use all but
934 // the bottom three bits as the array index, and the bottom three bits to form
935 // a mask.
936 // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0;
937 Value *ByteIndex = CGF.Builder.CreateAShr(
938 BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx");
939 Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy);
940 Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8,
941 ByteIndex, "bittest.byteaddr"),
942 CharUnits::One());
943 Value *PosLow =
944 CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty),
945 llvm::ConstantInt::get(CGF.Int8Ty, 0x7));
946
947 // The updating instructions will need a mask.
948 Value *Mask = nullptr;
949 if (BT.Action != BitTest::TestOnly) {
950 Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow,
951 "bittest.mask");
952 }
953
954 // Check the action and ordering of the interlocked intrinsics.
955 llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking);
956
957 Value *OldByte = nullptr;
958 if (Ordering != llvm::AtomicOrdering::NotAtomic) {
959 // Emit a combined atomicrmw load/store operation for the interlocked
960 // intrinsics.
961 llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or;
962 if (BT.Action == BitTest::Reset) {
963 Mask = CGF.Builder.CreateNot(Mask);
964 RMWOp = llvm::AtomicRMWInst::And;
965 }
966 OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask,
967 Ordering);
968 } else {
969 // Emit a plain load for the non-interlocked intrinsics.
970 OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte");
971 Value *NewByte = nullptr;
972 switch (BT.Action) {
973 case BitTest::TestOnly:
974 // Don't store anything.
975 break;
976 case BitTest::Complement:
977 NewByte = CGF.Builder.CreateXor(OldByte, Mask);
978 break;
979 case BitTest::Reset:
980 NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask));
981 break;
982 case BitTest::Set:
983 NewByte = CGF.Builder.CreateOr(OldByte, Mask);
984 break;
985 }
986 if (NewByte)
987 CGF.Builder.CreateStore(NewByte, ByteAddr);
988 }
989
990 // However we loaded the old byte, either by plain load or atomicrmw, shift
991 // the bit into the low position and mask it to 0 or 1.
992 Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr");
993 return CGF.Builder.CreateAnd(
994 ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res");
995}
996
997static llvm::Value *emitPPCLoadReserveIntrinsic(CodeGenFunction &CGF,
998 unsigned BuiltinID,
999 const CallExpr *E) {
1000 Value *Addr = CGF.EmitScalarExpr(E->getArg(0));
1001
1002 SmallString<64> Asm;
1003 raw_svector_ostream AsmOS(Asm);
1004 llvm::IntegerType *RetType = CGF.Int32Ty;
1005
1006 switch (BuiltinID) {
1007 case clang::PPC::BI__builtin_ppc_ldarx:
1008 AsmOS << "ldarx ";
1009 RetType = CGF.Int64Ty;
1010 break;
1011 case clang::PPC::BI__builtin_ppc_lwarx:
1012 AsmOS << "lwarx ";
1013 RetType = CGF.Int32Ty;
1014 break;
1015 case clang::PPC::BI__builtin_ppc_lharx:
1016 AsmOS << "lharx ";
1017 RetType = CGF.Int16Ty;
1018 break;
1019 case clang::PPC::BI__builtin_ppc_lbarx:
1020 AsmOS << "lbarx ";
1021 RetType = CGF.Int8Ty;
1022 break;
1023 default:
1024 llvm_unreachable("Expected only PowerPC load reserve intrinsics")::llvm::llvm_unreachable_internal("Expected only PowerPC load reserve intrinsics"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1024)
;
1025 }
1026
1027 AsmOS << "$0, ${1:y}";
1028
1029 std::string Constraints = "=r,*Z,~{memory}";
1030 std::string MachineClobbers = CGF.getTarget().getClobbers();
1031 if (!MachineClobbers.empty()) {
1032 Constraints += ',';
1033 Constraints += MachineClobbers;
1034 }
1035
1036 llvm::Type *IntPtrType = RetType->getPointerTo();
1037 llvm::FunctionType *FTy =
1038 llvm::FunctionType::get(RetType, {IntPtrType}, false);
1039
1040 llvm::InlineAsm *IA =
1041 llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true);
1042 return CGF.Builder.CreateCall(IA, {Addr});
1043}
1044
1045namespace {
1046enum class MSVCSetJmpKind {
1047 _setjmpex,
1048 _setjmp3,
1049 _setjmp
1050};
1051}
1052
1053/// MSVC handles setjmp a bit differently on different platforms. On every
1054/// architecture except 32-bit x86, the frame address is passed. On x86, extra
1055/// parameters can be passed as variadic arguments, but we always pass none.
1056static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind,
1057 const CallExpr *E) {
1058 llvm::Value *Arg1 = nullptr;
1059 llvm::Type *Arg1Ty = nullptr;
1060 StringRef Name;
1061 bool IsVarArg = false;
1062 if (SJKind == MSVCSetJmpKind::_setjmp3) {
1063 Name = "_setjmp3";
1064 Arg1Ty = CGF.Int32Ty;
1065 Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0);
1066 IsVarArg = true;
1067 } else {
1068 Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex";
1069 Arg1Ty = CGF.Int8PtrTy;
1070 if (CGF.getTarget().getTriple().getArch() == llvm::Triple::aarch64) {
1071 Arg1 = CGF.Builder.CreateCall(
1072 CGF.CGM.getIntrinsic(Intrinsic::sponentry, CGF.AllocaInt8PtrTy));
1073 } else
1074 Arg1 = CGF.Builder.CreateCall(
1075 CGF.CGM.getIntrinsic(Intrinsic::frameaddress, CGF.AllocaInt8PtrTy),
1076 llvm::ConstantInt::get(CGF.Int32Ty, 0));
1077 }
1078
1079 // Mark the call site and declaration with ReturnsTwice.
1080 llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty};
1081 llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get(
1082 CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex,
1083 llvm::Attribute::ReturnsTwice);
1084 llvm::FunctionCallee SetJmpFn = CGF.CGM.CreateRuntimeFunction(
1085 llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name,
1086 ReturnsTwiceAttr, /*Local=*/true);
1087
1088 llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast(
1089 CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy);
1090 llvm::Value *Args[] = {Buf, Arg1};
1091 llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args);
1092 CB->setAttributes(ReturnsTwiceAttr);
1093 return RValue::get(CB);
1094}
1095
1096// Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code,
1097// we handle them here.
1098enum class CodeGenFunction::MSVCIntrin {
1099 _BitScanForward,
1100 _BitScanReverse,
1101 _InterlockedAnd,
1102 _InterlockedDecrement,
1103 _InterlockedExchange,
1104 _InterlockedExchangeAdd,
1105 _InterlockedExchangeSub,
1106 _InterlockedIncrement,
1107 _InterlockedOr,
1108 _InterlockedXor,
1109 _InterlockedExchangeAdd_acq,
1110 _InterlockedExchangeAdd_rel,
1111 _InterlockedExchangeAdd_nf,
1112 _InterlockedExchange_acq,
1113 _InterlockedExchange_rel,
1114 _InterlockedExchange_nf,
1115 _InterlockedCompareExchange_acq,
1116 _InterlockedCompareExchange_rel,
1117 _InterlockedCompareExchange_nf,
1118 _InterlockedCompareExchange128,
1119 _InterlockedCompareExchange128_acq,
1120 _InterlockedCompareExchange128_rel,
1121 _InterlockedCompareExchange128_nf,
1122 _InterlockedOr_acq,
1123 _InterlockedOr_rel,
1124 _InterlockedOr_nf,
1125 _InterlockedXor_acq,
1126 _InterlockedXor_rel,
1127 _InterlockedXor_nf,
1128 _InterlockedAnd_acq,
1129 _InterlockedAnd_rel,
1130 _InterlockedAnd_nf,
1131 _InterlockedIncrement_acq,
1132 _InterlockedIncrement_rel,
1133 _InterlockedIncrement_nf,
1134 _InterlockedDecrement_acq,
1135 _InterlockedDecrement_rel,
1136 _InterlockedDecrement_nf,
1137 __fastfail,
1138};
1139
1140static Optional<CodeGenFunction::MSVCIntrin>
1141translateArmToMsvcIntrin(unsigned BuiltinID) {
1142 using MSVCIntrin = CodeGenFunction::MSVCIntrin;
1143 switch (BuiltinID) {
1144 default:
1145 return None;
1146 case ARM::BI_BitScanForward:
1147 case ARM::BI_BitScanForward64:
1148 return MSVCIntrin::_BitScanForward;
1149 case ARM::BI_BitScanReverse:
1150 case ARM::BI_BitScanReverse64:
1151 return MSVCIntrin::_BitScanReverse;
1152 case ARM::BI_InterlockedAnd64:
1153 return MSVCIntrin::_InterlockedAnd;
1154 case ARM::BI_InterlockedExchange64:
1155 return MSVCIntrin::_InterlockedExchange;
1156 case ARM::BI_InterlockedExchangeAdd64:
1157 return MSVCIntrin::_InterlockedExchangeAdd;
1158 case ARM::BI_InterlockedExchangeSub64:
1159 return MSVCIntrin::_InterlockedExchangeSub;
1160 case ARM::BI_InterlockedOr64:
1161 return MSVCIntrin::_InterlockedOr;
1162 case ARM::BI_InterlockedXor64:
1163 return MSVCIntrin::_InterlockedXor;
1164 case ARM::BI_InterlockedDecrement64:
1165 return MSVCIntrin::_InterlockedDecrement;
1166 case ARM::BI_InterlockedIncrement64:
1167 return MSVCIntrin::_InterlockedIncrement;
1168 case ARM::BI_InterlockedExchangeAdd8_acq:
1169 case ARM::BI_InterlockedExchangeAdd16_acq:
1170 case ARM::BI_InterlockedExchangeAdd_acq:
1171 case ARM::BI_InterlockedExchangeAdd64_acq:
1172 return MSVCIntrin::_InterlockedExchangeAdd_acq;
1173 case ARM::BI_InterlockedExchangeAdd8_rel:
1174 case ARM::BI_InterlockedExchangeAdd16_rel:
1175 case ARM::BI_InterlockedExchangeAdd_rel:
1176 case ARM::BI_InterlockedExchangeAdd64_rel:
1177 return MSVCIntrin::_InterlockedExchangeAdd_rel;
1178 case ARM::BI_InterlockedExchangeAdd8_nf:
1179 case ARM::BI_InterlockedExchangeAdd16_nf:
1180 case ARM::BI_InterlockedExchangeAdd_nf:
1181 case ARM::BI_InterlockedExchangeAdd64_nf:
1182 return MSVCIntrin::_InterlockedExchangeAdd_nf;
1183 case ARM::BI_InterlockedExchange8_acq:
1184 case ARM::BI_InterlockedExchange16_acq:
1185 case ARM::BI_InterlockedExchange_acq:
1186 case ARM::BI_InterlockedExchange64_acq:
1187 return MSVCIntrin::_InterlockedExchange_acq;
1188 case ARM::BI_InterlockedExchange8_rel:
1189 case ARM::BI_InterlockedExchange16_rel:
1190 case ARM::BI_InterlockedExchange_rel:
1191 case ARM::BI_InterlockedExchange64_rel:
1192 return MSVCIntrin::_InterlockedExchange_rel;
1193 case ARM::BI_InterlockedExchange8_nf:
1194 case ARM::BI_InterlockedExchange16_nf:
1195 case ARM::BI_InterlockedExchange_nf:
1196 case ARM::BI_InterlockedExchange64_nf:
1197 return MSVCIntrin::_InterlockedExchange_nf;
1198 case ARM::BI_InterlockedCompareExchange8_acq:
1199 case ARM::BI_InterlockedCompareExchange16_acq:
1200 case ARM::BI_InterlockedCompareExchange_acq:
1201 case ARM::BI_InterlockedCompareExchange64_acq:
1202 return MSVCIntrin::_InterlockedCompareExchange_acq;
1203 case ARM::BI_InterlockedCompareExchange8_rel:
1204 case ARM::BI_InterlockedCompareExchange16_rel:
1205 case ARM::BI_InterlockedCompareExchange_rel:
1206 case ARM::BI_InterlockedCompareExchange64_rel:
1207 return MSVCIntrin::_InterlockedCompareExchange_rel;
1208 case ARM::BI_InterlockedCompareExchange8_nf:
1209 case ARM::BI_InterlockedCompareExchange16_nf:
1210 case ARM::BI_InterlockedCompareExchange_nf:
1211 case ARM::BI_InterlockedCompareExchange64_nf:
1212 return MSVCIntrin::_InterlockedCompareExchange_nf;
1213 case ARM::BI_InterlockedOr8_acq:
1214 case ARM::BI_InterlockedOr16_acq:
1215 case ARM::BI_InterlockedOr_acq:
1216 case ARM::BI_InterlockedOr64_acq:
1217 return MSVCIntrin::_InterlockedOr_acq;
1218 case ARM::BI_InterlockedOr8_rel:
1219 case ARM::BI_InterlockedOr16_rel:
1220 case ARM::BI_InterlockedOr_rel:
1221 case ARM::BI_InterlockedOr64_rel:
1222 return MSVCIntrin::_InterlockedOr_rel;
1223 case ARM::BI_InterlockedOr8_nf:
1224 case ARM::BI_InterlockedOr16_nf:
1225 case ARM::BI_InterlockedOr_nf:
1226 case ARM::BI_InterlockedOr64_nf:
1227 return MSVCIntrin::_InterlockedOr_nf;
1228 case ARM::BI_InterlockedXor8_acq:
1229 case ARM::BI_InterlockedXor16_acq:
1230 case ARM::BI_InterlockedXor_acq:
1231 case ARM::BI_InterlockedXor64_acq:
1232 return MSVCIntrin::_InterlockedXor_acq;
1233 case ARM::BI_InterlockedXor8_rel:
1234 case ARM::BI_InterlockedXor16_rel:
1235 case ARM::BI_InterlockedXor_rel:
1236 case ARM::BI_InterlockedXor64_rel:
1237 return MSVCIntrin::_InterlockedXor_rel;
1238 case ARM::BI_InterlockedXor8_nf:
1239 case ARM::BI_InterlockedXor16_nf:
1240 case ARM::BI_InterlockedXor_nf:
1241 case ARM::BI_InterlockedXor64_nf:
1242 return MSVCIntrin::_InterlockedXor_nf;
1243 case ARM::BI_InterlockedAnd8_acq:
1244 case ARM::BI_InterlockedAnd16_acq:
1245 case ARM::BI_InterlockedAnd_acq:
1246 case ARM::BI_InterlockedAnd64_acq:
1247 return MSVCIntrin::_InterlockedAnd_acq;
1248 case ARM::BI_InterlockedAnd8_rel:
1249 case ARM::BI_InterlockedAnd16_rel:
1250 case ARM::BI_InterlockedAnd_rel:
1251 case ARM::BI_InterlockedAnd64_rel:
1252 return MSVCIntrin::_InterlockedAnd_rel;
1253 case ARM::BI_InterlockedAnd8_nf:
1254 case ARM::BI_InterlockedAnd16_nf:
1255 case ARM::BI_InterlockedAnd_nf:
1256 case ARM::BI_InterlockedAnd64_nf:
1257 return MSVCIntrin::_InterlockedAnd_nf;
1258 case ARM::BI_InterlockedIncrement16_acq:
1259 case ARM::BI_InterlockedIncrement_acq:
1260 case ARM::BI_InterlockedIncrement64_acq:
1261 return MSVCIntrin::_InterlockedIncrement_acq;
1262 case ARM::BI_InterlockedIncrement16_rel:
1263 case ARM::BI_InterlockedIncrement_rel:
1264 case ARM::BI_InterlockedIncrement64_rel:
1265 return MSVCIntrin::_InterlockedIncrement_rel;
1266 case ARM::BI_InterlockedIncrement16_nf:
1267 case ARM::BI_InterlockedIncrement_nf:
1268 case ARM::BI_InterlockedIncrement64_nf:
1269 return MSVCIntrin::_InterlockedIncrement_nf;
1270 case ARM::BI_InterlockedDecrement16_acq:
1271 case ARM::BI_InterlockedDecrement_acq:
1272 case ARM::BI_InterlockedDecrement64_acq:
1273 return MSVCIntrin::_InterlockedDecrement_acq;
1274 case ARM::BI_InterlockedDecrement16_rel:
1275 case ARM::BI_InterlockedDecrement_rel:
1276 case ARM::BI_InterlockedDecrement64_rel:
1277 return MSVCIntrin::_InterlockedDecrement_rel;
1278 case ARM::BI_InterlockedDecrement16_nf:
1279 case ARM::BI_InterlockedDecrement_nf:
1280 case ARM::BI_InterlockedDecrement64_nf:
1281 return MSVCIntrin::_InterlockedDecrement_nf;
1282 }
1283 llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1283)
;
1284}
1285
1286static Optional<CodeGenFunction::MSVCIntrin>
1287translateAarch64ToMsvcIntrin(unsigned BuiltinID) {
1288 using MSVCIntrin = CodeGenFunction::MSVCIntrin;
1289 switch (BuiltinID) {
1290 default:
1291 return None;
1292 case AArch64::BI_BitScanForward:
1293 case AArch64::BI_BitScanForward64:
1294 return MSVCIntrin::_BitScanForward;
1295 case AArch64::BI_BitScanReverse:
1296 case AArch64::BI_BitScanReverse64:
1297 return MSVCIntrin::_BitScanReverse;
1298 case AArch64::BI_InterlockedAnd64:
1299 return MSVCIntrin::_InterlockedAnd;
1300 case AArch64::BI_InterlockedExchange64:
1301 return MSVCIntrin::_InterlockedExchange;
1302 case AArch64::BI_InterlockedExchangeAdd64:
1303 return MSVCIntrin::_InterlockedExchangeAdd;
1304 case AArch64::BI_InterlockedExchangeSub64:
1305 return MSVCIntrin::_InterlockedExchangeSub;
1306 case AArch64::BI_InterlockedOr64:
1307 return MSVCIntrin::_InterlockedOr;
1308 case AArch64::BI_InterlockedXor64:
1309 return MSVCIntrin::_InterlockedXor;
1310 case AArch64::BI_InterlockedDecrement64:
1311 return MSVCIntrin::_InterlockedDecrement;
1312 case AArch64::BI_InterlockedIncrement64:
1313 return MSVCIntrin::_InterlockedIncrement;
1314 case AArch64::BI_InterlockedExchangeAdd8_acq:
1315 case AArch64::BI_InterlockedExchangeAdd16_acq:
1316 case AArch64::BI_InterlockedExchangeAdd_acq:
1317 case AArch64::BI_InterlockedExchangeAdd64_acq:
1318 return MSVCIntrin::_InterlockedExchangeAdd_acq;
1319 case AArch64::BI_InterlockedExchangeAdd8_rel:
1320 case AArch64::BI_InterlockedExchangeAdd16_rel:
1321 case AArch64::BI_InterlockedExchangeAdd_rel:
1322 case AArch64::BI_InterlockedExchangeAdd64_rel:
1323 return MSVCIntrin::_InterlockedExchangeAdd_rel;
1324 case AArch64::BI_InterlockedExchangeAdd8_nf:
1325 case AArch64::BI_InterlockedExchangeAdd16_nf:
1326 case AArch64::BI_InterlockedExchangeAdd_nf:
1327 case AArch64::BI_InterlockedExchangeAdd64_nf:
1328 return MSVCIntrin::_InterlockedExchangeAdd_nf;
1329 case AArch64::BI_InterlockedExchange8_acq:
1330 case AArch64::BI_InterlockedExchange16_acq:
1331 case AArch64::BI_InterlockedExchange_acq:
1332 case AArch64::BI_InterlockedExchange64_acq:
1333 return MSVCIntrin::_InterlockedExchange_acq;
1334 case AArch64::BI_InterlockedExchange8_rel:
1335 case AArch64::BI_InterlockedExchange16_rel:
1336 case AArch64::BI_InterlockedExchange_rel:
1337 case AArch64::BI_InterlockedExchange64_rel:
1338 return MSVCIntrin::_InterlockedExchange_rel;
1339 case AArch64::BI_InterlockedExchange8_nf:
1340 case AArch64::BI_InterlockedExchange16_nf:
1341 case AArch64::BI_InterlockedExchange_nf:
1342 case AArch64::BI_InterlockedExchange64_nf:
1343 return MSVCIntrin::_InterlockedExchange_nf;
1344 case AArch64::BI_InterlockedCompareExchange8_acq:
1345 case AArch64::BI_InterlockedCompareExchange16_acq:
1346 case AArch64::BI_InterlockedCompareExchange_acq:
1347 case AArch64::BI_InterlockedCompareExchange64_acq:
1348 return MSVCIntrin::_InterlockedCompareExchange_acq;
1349 case AArch64::BI_InterlockedCompareExchange8_rel:
1350 case AArch64::BI_InterlockedCompareExchange16_rel:
1351 case AArch64::BI_InterlockedCompareExchange_rel:
1352 case AArch64::BI_InterlockedCompareExchange64_rel:
1353 return MSVCIntrin::_InterlockedCompareExchange_rel;
1354 case AArch64::BI_InterlockedCompareExchange8_nf:
1355 case AArch64::BI_InterlockedCompareExchange16_nf:
1356 case AArch64::BI_InterlockedCompareExchange_nf:
1357 case AArch64::BI_InterlockedCompareExchange64_nf:
1358 return MSVCIntrin::_InterlockedCompareExchange_nf;
1359 case AArch64::BI_InterlockedCompareExchange128:
1360 return MSVCIntrin::_InterlockedCompareExchange128;
1361 case AArch64::BI_InterlockedCompareExchange128_acq:
1362 return MSVCIntrin::_InterlockedCompareExchange128_acq;
1363 case AArch64::BI_InterlockedCompareExchange128_nf:
1364 return MSVCIntrin::_InterlockedCompareExchange128_nf;
1365 case AArch64::BI_InterlockedCompareExchange128_rel:
1366 return MSVCIntrin::_InterlockedCompareExchange128_rel;
1367 case AArch64::BI_InterlockedOr8_acq:
1368 case AArch64::BI_InterlockedOr16_acq:
1369 case AArch64::BI_InterlockedOr_acq:
1370 case AArch64::BI_InterlockedOr64_acq:
1371 return MSVCIntrin::_InterlockedOr_acq;
1372 case AArch64::BI_InterlockedOr8_rel:
1373 case AArch64::BI_InterlockedOr16_rel:
1374 case AArch64::BI_InterlockedOr_rel:
1375 case AArch64::BI_InterlockedOr64_rel:
1376 return MSVCIntrin::_InterlockedOr_rel;
1377 case AArch64::BI_InterlockedOr8_nf:
1378 case AArch64::BI_InterlockedOr16_nf:
1379 case AArch64::BI_InterlockedOr_nf:
1380 case AArch64::BI_InterlockedOr64_nf:
1381 return MSVCIntrin::_InterlockedOr_nf;
1382 case AArch64::BI_InterlockedXor8_acq:
1383 case AArch64::BI_InterlockedXor16_acq:
1384 case AArch64::BI_InterlockedXor_acq:
1385 case AArch64::BI_InterlockedXor64_acq:
1386 return MSVCIntrin::_InterlockedXor_acq;
1387 case AArch64::BI_InterlockedXor8_rel:
1388 case AArch64::BI_InterlockedXor16_rel:
1389 case AArch64::BI_InterlockedXor_rel:
1390 case AArch64::BI_InterlockedXor64_rel:
1391 return MSVCIntrin::_InterlockedXor_rel;
1392 case AArch64::BI_InterlockedXor8_nf:
1393 case AArch64::BI_InterlockedXor16_nf:
1394 case AArch64::BI_InterlockedXor_nf:
1395 case AArch64::BI_InterlockedXor64_nf:
1396 return MSVCIntrin::_InterlockedXor_nf;
1397 case AArch64::BI_InterlockedAnd8_acq:
1398 case AArch64::BI_InterlockedAnd16_acq:
1399 case AArch64::BI_InterlockedAnd_acq:
1400 case AArch64::BI_InterlockedAnd64_acq:
1401 return MSVCIntrin::_InterlockedAnd_acq;
1402 case AArch64::BI_InterlockedAnd8_rel:
1403 case AArch64::BI_InterlockedAnd16_rel:
1404 case AArch64::BI_InterlockedAnd_rel:
1405 case AArch64::BI_InterlockedAnd64_rel:
1406 return MSVCIntrin::_InterlockedAnd_rel;
1407 case AArch64::BI_InterlockedAnd8_nf:
1408 case AArch64::BI_InterlockedAnd16_nf:
1409 case AArch64::BI_InterlockedAnd_nf:
1410 case AArch64::BI_InterlockedAnd64_nf:
1411 return MSVCIntrin::_InterlockedAnd_nf;
1412 case AArch64::BI_InterlockedIncrement16_acq:
1413 case AArch64::BI_InterlockedIncrement_acq:
1414 case AArch64::BI_InterlockedIncrement64_acq:
1415 return MSVCIntrin::_InterlockedIncrement_acq;
1416 case AArch64::BI_InterlockedIncrement16_rel:
1417 case AArch64::BI_InterlockedIncrement_rel:
1418 case AArch64::BI_InterlockedIncrement64_rel:
1419 return MSVCIntrin::_InterlockedIncrement_rel;
1420 case AArch64::BI_InterlockedIncrement16_nf:
1421 case AArch64::BI_InterlockedIncrement_nf:
1422 case AArch64::BI_InterlockedIncrement64_nf:
1423 return MSVCIntrin::_InterlockedIncrement_nf;
1424 case AArch64::BI_InterlockedDecrement16_acq:
1425 case AArch64::BI_InterlockedDecrement_acq:
1426 case AArch64::BI_InterlockedDecrement64_acq:
1427 return MSVCIntrin::_InterlockedDecrement_acq;
1428 case AArch64::BI_InterlockedDecrement16_rel:
1429 case AArch64::BI_InterlockedDecrement_rel:
1430 case AArch64::BI_InterlockedDecrement64_rel:
1431 return MSVCIntrin::_InterlockedDecrement_rel;
1432 case AArch64::BI_InterlockedDecrement16_nf:
1433 case AArch64::BI_InterlockedDecrement_nf:
1434 case AArch64::BI_InterlockedDecrement64_nf:
1435 return MSVCIntrin::_InterlockedDecrement_nf;
1436 }
1437 llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1437)
;
1438}
1439
1440static Optional<CodeGenFunction::MSVCIntrin>
1441translateX86ToMsvcIntrin(unsigned BuiltinID) {
1442 using MSVCIntrin = CodeGenFunction::MSVCIntrin;
1443 switch (BuiltinID) {
1444 default:
1445 return None;
1446 case clang::X86::BI_BitScanForward:
1447 case clang::X86::BI_BitScanForward64:
1448 return MSVCIntrin::_BitScanForward;
1449 case clang::X86::BI_BitScanReverse:
1450 case clang::X86::BI_BitScanReverse64:
1451 return MSVCIntrin::_BitScanReverse;
1452 case clang::X86::BI_InterlockedAnd64:
1453 return MSVCIntrin::_InterlockedAnd;
1454 case clang::X86::BI_InterlockedCompareExchange128:
1455 return MSVCIntrin::_InterlockedCompareExchange128;
1456 case clang::X86::BI_InterlockedExchange64:
1457 return MSVCIntrin::_InterlockedExchange;
1458 case clang::X86::BI_InterlockedExchangeAdd64:
1459 return MSVCIntrin::_InterlockedExchangeAdd;
1460 case clang::X86::BI_InterlockedExchangeSub64:
1461 return MSVCIntrin::_InterlockedExchangeSub;
1462 case clang::X86::BI_InterlockedOr64:
1463 return MSVCIntrin::_InterlockedOr;
1464 case clang::X86::BI_InterlockedXor64:
1465 return MSVCIntrin::_InterlockedXor;
1466 case clang::X86::BI_InterlockedDecrement64:
1467 return MSVCIntrin::_InterlockedDecrement;
1468 case clang::X86::BI_InterlockedIncrement64:
1469 return MSVCIntrin::_InterlockedIncrement;
1470 }
1471 llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1471)
;
1472}
1473
1474// Emit an MSVC intrinsic. Assumes that arguments have *not* been evaluated.
1475Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID,
1476 const CallExpr *E) {
1477 switch (BuiltinID) {
1478 case MSVCIntrin::_BitScanForward:
1479 case MSVCIntrin::_BitScanReverse: {
1480 Address IndexAddress(EmitPointerWithAlignment(E->getArg(0)));
1481 Value *ArgValue = EmitScalarExpr(E->getArg(1));
1482
1483 llvm::Type *ArgType = ArgValue->getType();
1484 llvm::Type *IndexType =
1485 IndexAddress.getPointer()->getType()->getPointerElementType();
1486 llvm::Type *ResultType = ConvertType(E->getType());
1487
1488 Value *ArgZero = llvm::Constant::getNullValue(ArgType);
1489 Value *ResZero = llvm::Constant::getNullValue(ResultType);
1490 Value *ResOne = llvm::ConstantInt::get(ResultType, 1);
1491
1492 BasicBlock *Begin = Builder.GetInsertBlock();
1493 BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn);
1494 Builder.SetInsertPoint(End);
1495 PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result");
1496
1497 Builder.SetInsertPoint(Begin);
1498 Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero);
1499 BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn);
1500 Builder.CreateCondBr(IsZero, End, NotZero);
1501 Result->addIncoming(ResZero, Begin);
1502
1503 Builder.SetInsertPoint(NotZero);
1504
1505 if (BuiltinID == MSVCIntrin::_BitScanForward) {
1506 Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
1507 Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()});
1508 ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false);
1509 Builder.CreateStore(ZeroCount, IndexAddress, false);
1510 } else {
1511 unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth();
1512 Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1);
1513
1514 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
1515 Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()});
1516 ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false);
1517 Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount);
1518 Builder.CreateStore(Index, IndexAddress, false);
1519 }
1520 Builder.CreateBr(End);
1521 Result->addIncoming(ResOne, NotZero);
1522
1523 Builder.SetInsertPoint(End);
1524 return Result;
1525 }
1526 case MSVCIntrin::_InterlockedAnd:
1527 return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E);
1528 case MSVCIntrin::_InterlockedExchange:
1529 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E);
1530 case MSVCIntrin::_InterlockedExchangeAdd:
1531 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E);
1532 case MSVCIntrin::_InterlockedExchangeSub:
1533 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E);
1534 case MSVCIntrin::_InterlockedOr:
1535 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E);
1536 case MSVCIntrin::_InterlockedXor:
1537 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E);
1538 case MSVCIntrin::_InterlockedExchangeAdd_acq:
1539 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E,
1540 AtomicOrdering::Acquire);
1541 case MSVCIntrin::_InterlockedExchangeAdd_rel:
1542 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E,
1543 AtomicOrdering::Release);
1544 case MSVCIntrin::_InterlockedExchangeAdd_nf:
1545 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E,
1546 AtomicOrdering::Monotonic);
1547 case MSVCIntrin::_InterlockedExchange_acq:
1548 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E,
1549 AtomicOrdering::Acquire);
1550 case MSVCIntrin::_InterlockedExchange_rel:
1551 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E,
1552 AtomicOrdering::Release);
1553 case MSVCIntrin::_InterlockedExchange_nf:
1554 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E,
1555 AtomicOrdering::Monotonic);
1556 case MSVCIntrin::_InterlockedCompareExchange_acq:
1557 return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Acquire);
1558 case MSVCIntrin::_InterlockedCompareExchange_rel:
1559 return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Release);
1560 case MSVCIntrin::_InterlockedCompareExchange_nf:
1561 return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Monotonic);
1562 case MSVCIntrin::_InterlockedCompareExchange128:
1563 return EmitAtomicCmpXchg128ForMSIntrin(
1564 *this, E, AtomicOrdering::SequentiallyConsistent);
1565 case MSVCIntrin::_InterlockedCompareExchange128_acq:
1566 return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Acquire);
1567 case MSVCIntrin::_InterlockedCompareExchange128_rel:
1568 return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Release);
1569 case MSVCIntrin::_InterlockedCompareExchange128_nf:
1570 return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Monotonic);
1571 case MSVCIntrin::_InterlockedOr_acq:
1572 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E,
1573 AtomicOrdering::Acquire);
1574 case MSVCIntrin::_InterlockedOr_rel:
1575 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E,
1576 AtomicOrdering::Release);
1577 case MSVCIntrin::_InterlockedOr_nf:
1578 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E,
1579 AtomicOrdering::Monotonic);
1580 case MSVCIntrin::_InterlockedXor_acq:
1581 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E,
1582 AtomicOrdering::Acquire);
1583 case MSVCIntrin::_InterlockedXor_rel:
1584 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E,
1585 AtomicOrdering::Release);
1586 case MSVCIntrin::_InterlockedXor_nf:
1587 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E,
1588 AtomicOrdering::Monotonic);
1589 case MSVCIntrin::_InterlockedAnd_acq:
1590 return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E,
1591 AtomicOrdering::Acquire);
1592 case MSVCIntrin::_InterlockedAnd_rel:
1593 return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E,
1594 AtomicOrdering::Release);
1595 case MSVCIntrin::_InterlockedAnd_nf:
1596 return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E,
1597 AtomicOrdering::Monotonic);
1598 case MSVCIntrin::_InterlockedIncrement_acq:
1599 return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire);
1600 case MSVCIntrin::_InterlockedIncrement_rel:
1601 return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release);
1602 case MSVCIntrin::_InterlockedIncrement_nf:
1603 return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic);
1604 case MSVCIntrin::_InterlockedDecrement_acq:
1605 return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Acquire);
1606 case MSVCIntrin::_InterlockedDecrement_rel:
1607 return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Release);
1608 case MSVCIntrin::_InterlockedDecrement_nf:
1609 return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Monotonic);
1610
1611 case MSVCIntrin::_InterlockedDecrement:
1612 return EmitAtomicDecrementValue(*this, E);
1613 case MSVCIntrin::_InterlockedIncrement:
1614 return EmitAtomicIncrementValue(*this, E);
1615
1616 case MSVCIntrin::__fastfail: {
1617 // Request immediate process termination from the kernel. The instruction
1618 // sequences to do this are documented on MSDN:
1619 // https://msdn.microsoft.com/en-us/library/dn774154.aspx
1620 llvm::Triple::ArchType ISA = getTarget().getTriple().getArch();
1621 StringRef Asm, Constraints;
1622 switch (ISA) {
1623 default:
1624 ErrorUnsupported(E, "__fastfail call for this architecture");
1625 break;
1626 case llvm::Triple::x86:
1627 case llvm::Triple::x86_64:
1628 Asm = "int $$0x29";
1629 Constraints = "{cx}";
1630 break;
1631 case llvm::Triple::thumb:
1632 Asm = "udf #251";
1633 Constraints = "{r0}";
1634 break;
1635 case llvm::Triple::aarch64:
1636 Asm = "brk #0xF003";
1637 Constraints = "{w0}";
1638 }
1639 llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false);
1640 llvm::InlineAsm *IA =
1641 llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true);
1642 llvm::AttributeList NoReturnAttr = llvm::AttributeList::get(
1643 getLLVMContext(), llvm::AttributeList::FunctionIndex,
1644 llvm::Attribute::NoReturn);
1645 llvm::CallInst *CI = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0)));
1646 CI->setAttributes(NoReturnAttr);
1647 return CI;
1648 }
1649 }
1650 llvm_unreachable("Incorrect MSVC intrinsic!")::llvm::llvm_unreachable_internal("Incorrect MSVC intrinsic!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1650)
;
1651}
1652
1653namespace {
1654// ARC cleanup for __builtin_os_log_format
1655struct CallObjCArcUse final : EHScopeStack::Cleanup {
1656 CallObjCArcUse(llvm::Value *object) : object(object) {}
1657 llvm::Value *object;
1658
1659 void Emit(CodeGenFunction &CGF, Flags flags) override {
1660 CGF.EmitARCIntrinsicUse(object);
1661 }
1662};
1663}
1664
1665Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E,
1666 BuiltinCheckKind Kind) {
1667 assert((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero)(static_cast <bool> ((Kind == BCK_CLZPassedZero || Kind
== BCK_CTZPassedZero) && "Unsupported builtin check kind"
) ? void (0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1668, __extension__ __PRETTY_FUNCTION__))
1668 && "Unsupported builtin check kind")(static_cast <bool> ((Kind == BCK_CLZPassedZero || Kind
== BCK_CTZPassedZero) && "Unsupported builtin check kind"
) ? void (0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1668, __extension__ __PRETTY_FUNCTION__))
;
1669
1670 Value *ArgValue = EmitScalarExpr(E);
1671 if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef())
1672 return ArgValue;
1673
1674 SanitizerScope SanScope(this);
1675 Value *Cond = Builder.CreateICmpNE(
1676 ArgValue, llvm::Constant::getNullValue(ArgValue->getType()));
1677 EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin),
1678 SanitizerHandler::InvalidBuiltin,
1679 {EmitCheckSourceLocation(E->getExprLoc()),
1680 llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)},
1681 None);
1682 return ArgValue;
1683}
1684
1685/// Get the argument type for arguments to os_log_helper.
1686static CanQualType getOSLogArgType(ASTContext &C, int Size) {
1687 QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false);
1688 return C.getCanonicalType(UnsignedTy);
1689}
1690
1691llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction(
1692 const analyze_os_log::OSLogBufferLayout &Layout,
1693 CharUnits BufferAlignment) {
1694 ASTContext &Ctx = getContext();
1695
1696 llvm::SmallString<64> Name;
1697 {
1698 raw_svector_ostream OS(Name);
1699 OS << "__os_log_helper";
1700 OS << "_" << BufferAlignment.getQuantity();
1701 OS << "_" << int(Layout.getSummaryByte());
1702 OS << "_" << int(Layout.getNumArgsByte());
1703 for (const auto &Item : Layout.Items)
1704 OS << "_" << int(Item.getSizeByte()) << "_"
1705 << int(Item.getDescriptorByte());
1706 }
1707
1708 if (llvm::Function *F = CGM.getModule().getFunction(Name))
1709 return F;
1710
1711 llvm::SmallVector<QualType, 4> ArgTys;
1712 FunctionArgList Args;
1713 Args.push_back(ImplicitParamDecl::Create(
1714 Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"), Ctx.VoidPtrTy,
1715 ImplicitParamDecl::Other));
1716 ArgTys.emplace_back(Ctx.VoidPtrTy);
1717
1718 for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) {
1719 char Size = Layout.Items[I].getSizeByte();
1720 if (!Size)
1721 continue;
1722
1723 QualType ArgTy = getOSLogArgType(Ctx, Size);
1724 Args.push_back(ImplicitParamDecl::Create(
1725 Ctx, nullptr, SourceLocation(),
1726 &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)), ArgTy,
1727 ImplicitParamDecl::Other));
1728 ArgTys.emplace_back(ArgTy);
1729 }
1730
1731 QualType ReturnTy = Ctx.VoidTy;
1732
1733 // The helper function has linkonce_odr linkage to enable the linker to merge
1734 // identical functions. To ensure the merging always happens, 'noinline' is
1735 // attached to the function when compiling with -Oz.
1736 const CGFunctionInfo &FI =
1737 CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, Args);
1738 llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI);
1739 llvm::Function *Fn = llvm::Function::Create(
1740 FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule());
1741 Fn->setVisibility(llvm::GlobalValue::HiddenVisibility);
1742 CGM.SetLLVMFunctionAttributes(GlobalDecl(), FI, Fn, /*IsThunk=*/false);
1743 CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn);
1744 Fn->setDoesNotThrow();
1745
1746 // Attach 'noinline' at -Oz.
1747 if (CGM.getCodeGenOpts().OptimizeSize == 2)
1748 Fn->addFnAttr(llvm::Attribute::NoInline);
1749
1750 auto NL = ApplyDebugLocation::CreateEmpty(*this);
1751 StartFunction(GlobalDecl(), ReturnTy, Fn, FI, Args);
1752
1753 // Create a scope with an artificial location for the body of this function.
1754 auto AL = ApplyDebugLocation::CreateArtificial(*this);
1755
1756 CharUnits Offset;
1757 Address BufAddr(Builder.CreateLoad(GetAddrOfLocalVar(Args[0]), "buf"),
1758 BufferAlignment);
1759 Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()),
1760 Builder.CreateConstByteGEP(BufAddr, Offset++, "summary"));
1761 Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()),
1762 Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs"));
1763
1764 unsigned I = 1;
1765 for (const auto &Item : Layout.Items) {
1766 Builder.CreateStore(
1767 Builder.getInt8(Item.getDescriptorByte()),
1768 Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor"));
1769 Builder.CreateStore(
1770 Builder.getInt8(Item.getSizeByte()),
1771 Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize"));
1772
1773 CharUnits Size = Item.size();
1774 if (!Size.getQuantity())
1775 continue;
1776
1777 Address Arg = GetAddrOfLocalVar(Args[I]);
1778 Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData");
1779 Addr = Builder.CreateBitCast(Addr, Arg.getPointer()->getType(),
1780 "argDataCast");
1781 Builder.CreateStore(Builder.CreateLoad(Arg), Addr);
1782 Offset += Size;
1783 ++I;
1784 }
1785
1786 FinishFunction();
1787
1788 return Fn;
1789}
1790
1791RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) {
1792 assert(E.getNumArgs() >= 2 &&(static_cast <bool> (E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments"
) ? void (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1793, __extension__ __PRETTY_FUNCTION__))
1793 "__builtin_os_log_format takes at least 2 arguments")(static_cast <bool> (E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments"
) ? void (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1793, __extension__ __PRETTY_FUNCTION__))
;
1794 ASTContext &Ctx = getContext();
1795 analyze_os_log::OSLogBufferLayout Layout;
1796 analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout);
1797 Address BufAddr = EmitPointerWithAlignment(E.getArg(0));
1798 llvm::SmallVector<llvm::Value *, 4> RetainableOperands;
1799
1800 // Ignore argument 1, the format string. It is not currently used.
1801 CallArgList Args;
1802 Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy);
1803
1804 for (const auto &Item : Layout.Items) {
1805 int Size = Item.getSizeByte();
1806 if (!Size)
1807 continue;
1808
1809 llvm::Value *ArgVal;
1810
1811 if (Item.getKind() == analyze_os_log::OSLogBufferItem::MaskKind) {
1812 uint64_t Val = 0;
1813 for (unsigned I = 0, E = Item.getMaskType().size(); I < E; ++I)
1814 Val |= ((uint64_t)Item.getMaskType()[I]) << I * 8;
1815 ArgVal = llvm::Constant::getIntegerValue(Int64Ty, llvm::APInt(64, Val));
1816 } else if (const Expr *TheExpr = Item.getExpr()) {
1817 ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false);
1818
1819 // If a temporary object that requires destruction after the full
1820 // expression is passed, push a lifetime-extended cleanup to extend its
1821 // lifetime to the end of the enclosing block scope.
1822 auto LifetimeExtendObject = [&](const Expr *E) {
1823 E = E->IgnoreParenCasts();
1824 // Extend lifetimes of objects returned by function calls and message
1825 // sends.
1826
1827 // FIXME: We should do this in other cases in which temporaries are
1828 // created including arguments of non-ARC types (e.g., C++
1829 // temporaries).
1830 if (isa<CallExpr>(E) || isa<ObjCMessageExpr>(E))
1831 return true;
1832 return false;
1833 };
1834
1835 if (TheExpr->getType()->isObjCRetainableType() &&
1836 getLangOpts().ObjCAutoRefCount && LifetimeExtendObject(TheExpr)) {
1837 assert(getEvaluationKind(TheExpr->getType()) == TEK_Scalar &&(static_cast <bool> (getEvaluationKind(TheExpr->getType
()) == TEK_Scalar && "Only scalar can be a ObjC retainable type"
) ? void (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1838, __extension__ __PRETTY_FUNCTION__))
1838 "Only scalar can be a ObjC retainable type")(static_cast <bool> (getEvaluationKind(TheExpr->getType
()) == TEK_Scalar && "Only scalar can be a ObjC retainable type"
) ? void (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1838, __extension__ __PRETTY_FUNCTION__))
;
1839 if (!isa<Constant>(ArgVal)) {
1840 CleanupKind Cleanup = getARCCleanupKind();
1841 QualType Ty = TheExpr->getType();
1842 Address Alloca = Address::invalid();
1843 Address Addr = CreateMemTemp(Ty, "os.log.arg", &Alloca);
1844 ArgVal = EmitARCRetain(Ty, ArgVal);
1845 Builder.CreateStore(ArgVal, Addr);
1846 pushLifetimeExtendedDestroy(Cleanup, Alloca, Ty,
1847 CodeGenFunction::destroyARCStrongPrecise,
1848 Cleanup & EHCleanup);
1849
1850 // Push a clang.arc.use call to ensure ARC optimizer knows that the
1851 // argument has to be alive.
1852 if (CGM.getCodeGenOpts().OptimizationLevel != 0)
1853 pushCleanupAfterFullExpr<CallObjCArcUse>(Cleanup, ArgVal);
1854 }
1855 }
1856 } else {
1857 ArgVal = Builder.getInt32(Item.getConstValue().getQuantity());
1858 }
1859
1860 unsigned ArgValSize =
1861 CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType());
1862 llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(),
1863 ArgValSize);
1864 ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy);
1865 CanQualType ArgTy = getOSLogArgType(Ctx, Size);
1866 // If ArgVal has type x86_fp80, zero-extend ArgVal.
1867 ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy));
1868 Args.add(RValue::get(ArgVal), ArgTy);
1869 }
1870
1871 const CGFunctionInfo &FI =
1872 CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args);
1873 llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction(
1874 Layout, BufAddr.getAlignment());
1875 EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args);
1876 return RValue::get(BufAddr.getPointer());
1877}
1878
1879static bool isSpecialUnsignedMultiplySignedResult(
1880 unsigned BuiltinID, WidthAndSignedness Op1Info, WidthAndSignedness Op2Info,
1881 WidthAndSignedness ResultInfo) {
1882 return BuiltinID == Builtin::BI__builtin_mul_overflow &&
1883 Op1Info.Width == Op2Info.Width && Op2Info.Width == ResultInfo.Width &&
1884 !Op1Info.Signed && !Op2Info.Signed && ResultInfo.Signed;
1885}
1886
1887static RValue EmitCheckedUnsignedMultiplySignedResult(
1888 CodeGenFunction &CGF, const clang::Expr *Op1, WidthAndSignedness Op1Info,
1889 const clang::Expr *Op2, WidthAndSignedness Op2Info,
1890 const clang::Expr *ResultArg, QualType ResultQTy,
1891 WidthAndSignedness ResultInfo) {
1892 assert(isSpecialUnsignedMultiplySignedResult((static_cast <bool> (isSpecialUnsignedMultiplySignedResult
( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo
) && "Cannot specialize this multiply") ? void (0) : __assert_fail
("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1894, __extension__ __PRETTY_FUNCTION__))
1893 Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) &&(static_cast <bool> (isSpecialUnsignedMultiplySignedResult
( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo
) && "Cannot specialize this multiply") ? void (0) : __assert_fail
("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1894, __extension__ __PRETTY_FUNCTION__))
1894 "Cannot specialize this multiply")(static_cast <bool> (isSpecialUnsignedMultiplySignedResult
( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo
) && "Cannot specialize this multiply") ? void (0) : __assert_fail
("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1894, __extension__ __PRETTY_FUNCTION__))
;
1895
1896 llvm::Value *V1 = CGF.EmitScalarExpr(Op1);
1897 llvm::Value *V2 = CGF.EmitScalarExpr(Op2);
1898
1899 llvm::Value *HasOverflow;
1900 llvm::Value *Result = EmitOverflowIntrinsic(
1901 CGF, llvm::Intrinsic::umul_with_overflow, V1, V2, HasOverflow);
1902
1903 // The intrinsic call will detect overflow when the value is > UINT_MAX,
1904 // however, since the original builtin had a signed result, we need to report
1905 // an overflow when the result is greater than INT_MAX.
1906 auto IntMax = llvm::APInt::getSignedMaxValue(ResultInfo.Width);
1907 llvm::Value *IntMaxValue = llvm::ConstantInt::get(Result->getType(), IntMax);
1908
1909 llvm::Value *IntMaxOverflow = CGF.Builder.CreateICmpUGT(Result, IntMaxValue);
1910 HasOverflow = CGF.Builder.CreateOr(HasOverflow, IntMaxOverflow);
1911
1912 bool isVolatile =
1913 ResultArg->getType()->getPointeeType().isVolatileQualified();
1914 Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg);
1915 CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr,
1916 isVolatile);
1917 return RValue::get(HasOverflow);
1918}
1919
1920/// Determine if a binop is a checked mixed-sign multiply we can specialize.
1921static bool isSpecialMixedSignMultiply(unsigned BuiltinID,
1922 WidthAndSignedness Op1Info,
1923 WidthAndSignedness Op2Info,
1924 WidthAndSignedness ResultInfo) {
1925 return BuiltinID == Builtin::BI__builtin_mul_overflow &&
1926 std::max(Op1Info.Width, Op2Info.Width) >= ResultInfo.Width &&
1927 Op1Info.Signed != Op2Info.Signed;
1928}
1929
1930/// Emit a checked mixed-sign multiply. This is a cheaper specialization of
1931/// the generic checked-binop irgen.
1932static RValue
1933EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1,
1934 WidthAndSignedness Op1Info, const clang::Expr *Op2,
1935 WidthAndSignedness Op2Info,
1936 const clang::Expr *ResultArg, QualType ResultQTy,
1937 WidthAndSignedness ResultInfo) {
1938 assert(isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info,(static_cast <bool> (isSpecialMixedSignMultiply(Builtin
::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) &&
"Not a mixed-sign multipliction we can specialize") ? void (
0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1940, __extension__ __PRETTY_FUNCTION__))
1939 Op2Info, ResultInfo) &&(static_cast <bool> (isSpecialMixedSignMultiply(Builtin
::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) &&
"Not a mixed-sign multipliction we can specialize") ? void (
0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1940, __extension__ __PRETTY_FUNCTION__))
1940 "Not a mixed-sign multipliction we can specialize")(static_cast <bool> (isSpecialMixedSignMultiply(Builtin
::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) &&
"Not a mixed-sign multipliction we can specialize") ? void (
0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 1940, __extension__ __PRETTY_FUNCTION__))
;
1941
1942 // Emit the signed and unsigned operands.
1943 const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2;
1944 const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1;
1945 llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp);
1946 llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp);
1947 unsigned SignedOpWidth = Op1Info.Signed ? Op1Info.Width : Op2Info.Width;
1948 unsigned UnsignedOpWidth = Op1Info.Signed ? Op2Info.Width : Op1Info.Width;
1949
1950 // One of the operands may be smaller than the other. If so, [s|z]ext it.
1951 if (SignedOpWidth < UnsignedOpWidth)
1952 Signed = CGF.Builder.CreateSExt(Signed, Unsigned->getType(), "op.sext");
1953 if (UnsignedOpWidth < SignedOpWidth)
1954 Unsigned = CGF.Builder.CreateZExt(Unsigned, Signed->getType(), "op.zext");
1955
1956 llvm::Type *OpTy = Signed->getType();
1957 llvm::Value *Zero = llvm::Constant::getNullValue(OpTy);
1958 Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg);
1959 llvm::Type *ResTy = ResultPtr.getElementType();
1960 unsigned OpWidth = std::max(Op1Info.Width, Op2Info.Width);
1961
1962 // Take the absolute value of the signed operand.
1963 llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero);
1964 llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed);
1965 llvm::Value *AbsSigned =
1966 CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed);
1967
1968 // Perform a checked unsigned multiplication.
1969 llvm::Value *UnsignedOverflow;
1970 llvm::Value *UnsignedResult =
1971 EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned,
1972 Unsigned, UnsignedOverflow);
1973
1974 llvm::Value *Overflow, *Result;
1975 if (ResultInfo.Signed) {
1976 // Signed overflow occurs if the result is greater than INT_MAX or lesser
1977 // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative).
1978 auto IntMax =
1979 llvm::APInt::getSignedMaxValue(ResultInfo.Width).zextOrSelf(OpWidth);
1980 llvm::Value *MaxResult =
1981 CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax),
1982 CGF.Builder.CreateZExt(IsNegative, OpTy));
1983 llvm::Value *SignedOverflow =
1984 CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult);
1985 Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow);
1986
1987 // Prepare the signed result (possibly by negating it).
1988 llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult);
1989 llvm::Value *SignedResult =
1990 CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult);
1991 Result = CGF.Builder.CreateTrunc(SignedResult, ResTy);
1992 } else {
1993 // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX.
1994 llvm::Value *Underflow = CGF.Builder.CreateAnd(
1995 IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult));
1996 Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow);
1997 if (ResultInfo.Width < OpWidth) {
1998 auto IntMax =
1999 llvm::APInt::getMaxValue(ResultInfo.Width).zext(OpWidth);
2000 llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT(
2001 UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax));
2002 Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow);
2003 }
2004
2005 // Negate the product if it would be negative in infinite precision.
2006 Result = CGF.Builder.CreateSelect(
2007 IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult);
2008
2009 Result = CGF.Builder.CreateTrunc(Result, ResTy);
2010 }
2011 assert(Overflow && Result && "Missing overflow or result")(static_cast <bool> (Overflow && Result &&
"Missing overflow or result") ? void (0) : __assert_fail ("Overflow && Result && \"Missing overflow or result\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 2011, __extension__ __PRETTY_FUNCTION__))
;
2012
2013 bool isVolatile =
2014 ResultArg->getType()->getPointeeType().isVolatileQualified();
2015 CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr,
2016 isVolatile);
2017 return RValue::get(Overflow);
2018}
2019
2020static llvm::Value *dumpRecord(CodeGenFunction &CGF, QualType RType,
2021 Value *&RecordPtr, CharUnits Align,
2022 llvm::FunctionCallee Func, int Lvl) {
2023 ASTContext &Context = CGF.getContext();
2024 RecordDecl *RD = RType->castAs<RecordType>()->getDecl()->getDefinition();
2025 std::string Pad = std::string(Lvl * 4, ' ');
2026
2027 Value *GString =
2028 CGF.Builder.CreateGlobalStringPtr(RType.getAsString() + " {\n");
2029 Value *Res = CGF.Builder.CreateCall(Func, {GString});
2030
2031 static llvm::DenseMap<QualType, const char *> Types;
2032 if (Types.empty()) {
2033 Types[Context.CharTy] = "%c";
2034 Types[Context.BoolTy] = "%d";
2035 Types[Context.SignedCharTy] = "%hhd";
2036 Types[Context.UnsignedCharTy] = "%hhu";
2037 Types[Context.IntTy] = "%d";
2038 Types[Context.UnsignedIntTy] = "%u";
2039 Types[Context.LongTy] = "%ld";
2040 Types[Context.UnsignedLongTy] = "%lu";
2041 Types[Context.LongLongTy] = "%lld";
2042 Types[Context.UnsignedLongLongTy] = "%llu";
2043 Types[Context.ShortTy] = "%hd";
2044 Types[Context.UnsignedShortTy] = "%hu";
2045 Types[Context.VoidPtrTy] = "%p";
2046 Types[Context.FloatTy] = "%f";
2047 Types[Context.DoubleTy] = "%f";
2048 Types[Context.LongDoubleTy] = "%Lf";
2049 Types[Context.getPointerType(Context.CharTy)] = "%s";
2050 Types[Context.getPointerType(Context.getConstType(Context.CharTy))] = "%s";
2051 }
2052
2053 for (const auto *FD : RD->fields()) {
2054 Value *FieldPtr = RecordPtr;
2055 if (RD->isUnion())
2056 FieldPtr = CGF.Builder.CreatePointerCast(
2057 FieldPtr, CGF.ConvertType(Context.getPointerType(FD->getType())));
2058 else
2059 FieldPtr = CGF.Builder.CreateStructGEP(CGF.ConvertType(RType), FieldPtr,
2060 FD->getFieldIndex());
2061
2062 GString = CGF.Builder.CreateGlobalStringPtr(
2063 llvm::Twine(Pad)
2064 .concat(FD->getType().getAsString())
2065 .concat(llvm::Twine(' '))
2066 .concat(FD->getNameAsString())
2067 .concat(" : ")
2068 .str());
2069 Value *TmpRes = CGF.Builder.CreateCall(Func, {GString});
2070 Res = CGF.Builder.CreateAdd(Res, TmpRes);
2071
2072 QualType CanonicalType =
2073 FD->getType().getUnqualifiedType().getCanonicalType();
2074
2075 // We check whether we are in a recursive type
2076 if (CanonicalType->isRecordType()) {
2077 TmpRes = dumpRecord(CGF, CanonicalType, FieldPtr, Align, Func, Lvl + 1);
2078 Res = CGF.Builder.CreateAdd(TmpRes, Res);
2079 continue;
2080 }
2081
2082 // We try to determine the best format to print the current field
2083 llvm::Twine Format = Types.find(CanonicalType) == Types.end()
2084 ? Types[Context.VoidPtrTy]
2085 : Types[CanonicalType];
2086
2087 Address FieldAddress = Address(FieldPtr, Align);
2088 FieldPtr = CGF.Builder.CreateLoad(FieldAddress);
2089
2090 // FIXME Need to handle bitfield here
2091 GString = CGF.Builder.CreateGlobalStringPtr(
2092 Format.concat(llvm::Twine('\n')).str());
2093 TmpRes = CGF.Builder.CreateCall(Func, {GString, FieldPtr});
2094 Res = CGF.Builder.CreateAdd(Res, TmpRes);
2095 }
2096
2097 GString = CGF.Builder.CreateGlobalStringPtr(Pad + "}\n");
2098 Value *TmpRes = CGF.Builder.CreateCall(Func, {GString});
2099 Res = CGF.Builder.CreateAdd(Res, TmpRes);
2100 return Res;
2101}
2102
2103static bool
2104TypeRequiresBuiltinLaunderImp(const ASTContext &Ctx, QualType Ty,
2105 llvm::SmallPtrSetImpl<const Decl *> &Seen) {
2106 if (const auto *Arr = Ctx.getAsArrayType(Ty))
2107 Ty = Ctx.getBaseElementType(Arr);
2108
2109 const auto *Record = Ty->getAsCXXRecordDecl();
2110 if (!Record)
2111 return false;
2112
2113 // We've already checked this type, or are in the process of checking it.
2114 if (!Seen.insert(Record).second)
2115 return false;
2116
2117 assert(Record->hasDefinition() &&(static_cast <bool> (Record->hasDefinition() &&
"Incomplete types should already be diagnosed") ? void (0) :
__assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 2118, __extension__ __PRETTY_FUNCTION__))
2118 "Incomplete types should already be diagnosed")(static_cast <bool> (Record->hasDefinition() &&
"Incomplete types should already be diagnosed") ? void (0) :
__assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 2118, __extension__ __PRETTY_FUNCTION__))
;
2119
2120 if (Record->isDynamicClass())
2121 return true;
2122
2123 for (FieldDecl *F : Record->fields()) {
2124 if (TypeRequiresBuiltinLaunderImp(Ctx, F->getType(), Seen))
2125 return true;
2126 }
2127 return false;
2128}
2129
2130/// Determine if the specified type requires laundering by checking if it is a
2131/// dynamic class type or contains a subobject which is a dynamic class type.
2132static bool TypeRequiresBuiltinLaunder(CodeGenModule &CGM, QualType Ty) {
2133 if (!CGM.getCodeGenOpts().StrictVTablePointers)
2134 return false;
2135 llvm::SmallPtrSet<const Decl *, 16> Seen;
2136 return TypeRequiresBuiltinLaunderImp(CGM.getContext(), Ty, Seen);
2137}
2138
2139RValue CodeGenFunction::emitRotate(const CallExpr *E, bool IsRotateRight) {
2140 llvm::Value *Src = EmitScalarExpr(E->getArg(0));
2141 llvm::Value *ShiftAmt = EmitScalarExpr(E->getArg(1));
2142
2143 // The builtin's shift arg may have a different type than the source arg and
2144 // result, but the LLVM intrinsic uses the same type for all values.
2145 llvm::Type *Ty = Src->getType();
2146 ShiftAmt = Builder.CreateIntCast(ShiftAmt, Ty, false);
2147
2148 // Rotate is a special case of LLVM funnel shift - 1st 2 args are the same.
2149 unsigned IID = IsRotateRight ? Intrinsic::fshr : Intrinsic::fshl;
2150 Function *F = CGM.getIntrinsic(IID, Ty);
2151 return RValue::get(Builder.CreateCall(F, { Src, Src, ShiftAmt }));
2152}
2153
2154// Map math builtins for long-double to f128 version.
2155static unsigned mutateLongDoubleBuiltin(unsigned BuiltinID) {
2156 switch (BuiltinID) {
2157#define MUTATE_LDBL(func) \
2158 case Builtin::BI__builtin_##func##l: \
2159 return Builtin::BI__builtin_##func##f128;
2160 MUTATE_LDBL(sqrt)
2161 MUTATE_LDBL(cbrt)
2162 MUTATE_LDBL(fabs)
2163 MUTATE_LDBL(log)
2164 MUTATE_LDBL(log2)
2165 MUTATE_LDBL(log10)
2166 MUTATE_LDBL(log1p)
2167 MUTATE_LDBL(logb)
2168 MUTATE_LDBL(exp)
2169 MUTATE_LDBL(exp2)
2170 MUTATE_LDBL(expm1)
2171 MUTATE_LDBL(fdim)
2172 MUTATE_LDBL(hypot)
2173 MUTATE_LDBL(ilogb)
2174 MUTATE_LDBL(pow)
2175 MUTATE_LDBL(fmin)
2176 MUTATE_LDBL(fmax)
2177 MUTATE_LDBL(ceil)
2178 MUTATE_LDBL(trunc)
2179 MUTATE_LDBL(rint)
2180 MUTATE_LDBL(nearbyint)
2181 MUTATE_LDBL(round)
2182 MUTATE_LDBL(floor)
2183 MUTATE_LDBL(lround)
2184 MUTATE_LDBL(llround)
2185 MUTATE_LDBL(lrint)
2186 MUTATE_LDBL(llrint)
2187 MUTATE_LDBL(fmod)
2188 MUTATE_LDBL(modf)
2189 MUTATE_LDBL(nan)
2190 MUTATE_LDBL(nans)
2191 MUTATE_LDBL(inf)
2192 MUTATE_LDBL(fma)
2193 MUTATE_LDBL(sin)
2194 MUTATE_LDBL(cos)
2195 MUTATE_LDBL(tan)
2196 MUTATE_LDBL(sinh)
2197 MUTATE_LDBL(cosh)
2198 MUTATE_LDBL(tanh)
2199 MUTATE_LDBL(asin)
2200 MUTATE_LDBL(acos)
2201 MUTATE_LDBL(atan)
2202 MUTATE_LDBL(asinh)
2203 MUTATE_LDBL(acosh)
2204 MUTATE_LDBL(atanh)
2205 MUTATE_LDBL(atan2)
2206 MUTATE_LDBL(erf)
2207 MUTATE_LDBL(erfc)
2208 MUTATE_LDBL(ldexp)
2209 MUTATE_LDBL(frexp)
2210 MUTATE_LDBL(huge_val)
2211 MUTATE_LDBL(copysign)
2212 MUTATE_LDBL(nextafter)
2213 MUTATE_LDBL(nexttoward)
2214 MUTATE_LDBL(remainder)
2215 MUTATE_LDBL(remquo)
2216 MUTATE_LDBL(scalbln)
2217 MUTATE_LDBL(scalbn)
2218 MUTATE_LDBL(tgamma)
2219 MUTATE_LDBL(lgamma)
2220#undef MUTATE_LDBL
2221 default:
2222 return BuiltinID;
2223 }
2224}
2225
2226RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
2227 const CallExpr *E,
2228 ReturnValueSlot ReturnValue) {
2229 const FunctionDecl *FD = GD.getDecl()->getAsFunction();
2230 // See if we can constant fold this builtin. If so, don't emit it at all.
2231 Expr::EvalResult Result;
2232 if (E->EvaluateAsRValue(Result, CGM.getContext()) &&
2233 !Result.hasSideEffects()) {
2234 if (Result.Val.isInt())
2235 return RValue::get(llvm::ConstantInt::get(getLLVMContext(),
2236 Result.Val.getInt()));
2237 if (Result.Val.isFloat())
2238 return RValue::get(llvm::ConstantFP::get(getLLVMContext(),
2239 Result.Val.getFloat()));
2240 }
2241
2242 // If current long-double semantics is IEEE 128-bit, replace math builtins
2243 // of long-double with f128 equivalent.
2244 // TODO: This mutation should also be applied to other targets other than PPC,
2245 // after backend supports IEEE 128-bit style libcalls.
2246 if (getTarget().getTriple().isPPC64() &&
2247 &getTarget().getLongDoubleFormat() == &llvm::APFloat::IEEEquad())
2248 BuiltinID = mutateLongDoubleBuiltin(BuiltinID);
2249
2250 // If the builtin has been declared explicitly with an assembler label,
2251 // disable the specialized emitting below. Ideally we should communicate the
2252 // rename in IR, or at least avoid generating the intrinsic calls that are
2253 // likely to get lowered to the renamed library functions.
2254 const unsigned BuiltinIDIfNoAsmLabel =
2255 FD->hasAttr<AsmLabelAttr>() ? 0 : BuiltinID;
2256
2257 // There are LLVM math intrinsics/instructions corresponding to math library
2258 // functions except the LLVM op will never set errno while the math library
2259 // might. Also, math builtins have the same semantics as their math library
2260 // twins. Thus, we can transform math library and builtin calls to their
2261 // LLVM counterparts if the call is marked 'const' (known to never set errno).
2262 if (FD->hasAttr<ConstAttr>()) {
2263 switch (BuiltinIDIfNoAsmLabel) {
2264 case Builtin::BIceil:
2265 case Builtin::BIceilf:
2266 case Builtin::BIceill:
2267 case Builtin::BI__builtin_ceil:
2268 case Builtin::BI__builtin_ceilf:
2269 case Builtin::BI__builtin_ceilf16:
2270 case Builtin::BI__builtin_ceill:
2271 case Builtin::BI__builtin_ceilf128:
2272 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2273 Intrinsic::ceil,
2274 Intrinsic::experimental_constrained_ceil));
2275
2276 case Builtin::BIcopysign:
2277 case Builtin::BIcopysignf:
2278 case Builtin::BIcopysignl:
2279 case Builtin::BI__builtin_copysign:
2280 case Builtin::BI__builtin_copysignf:
2281 case Builtin::BI__builtin_copysignf16:
2282 case Builtin::BI__builtin_copysignl:
2283 case Builtin::BI__builtin_copysignf128:
2284 return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign));
2285
2286 case Builtin::BIcos:
2287 case Builtin::BIcosf:
2288 case Builtin::BIcosl:
2289 case Builtin::BI__builtin_cos:
2290 case Builtin::BI__builtin_cosf:
2291 case Builtin::BI__builtin_cosf16:
2292 case Builtin::BI__builtin_cosl:
2293 case Builtin::BI__builtin_cosf128:
2294 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2295 Intrinsic::cos,
2296 Intrinsic::experimental_constrained_cos));
2297
2298 case Builtin::BIexp:
2299 case Builtin::BIexpf:
2300 case Builtin::BIexpl:
2301 case Builtin::BI__builtin_exp:
2302 case Builtin::BI__builtin_expf:
2303 case Builtin::BI__builtin_expf16:
2304 case Builtin::BI__builtin_expl:
2305 case Builtin::BI__builtin_expf128:
2306 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2307 Intrinsic::exp,
2308 Intrinsic::experimental_constrained_exp));
2309
2310 case Builtin::BIexp2:
2311 case Builtin::BIexp2f:
2312 case Builtin::BIexp2l:
2313 case Builtin::BI__builtin_exp2:
2314 case Builtin::BI__builtin_exp2f:
2315 case Builtin::BI__builtin_exp2f16:
2316 case Builtin::BI__builtin_exp2l:
2317 case Builtin::BI__builtin_exp2f128:
2318 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2319 Intrinsic::exp2,
2320 Intrinsic::experimental_constrained_exp2));
2321
2322 case Builtin::BIfabs:
2323 case Builtin::BIfabsf:
2324 case Builtin::BIfabsl:
2325 case Builtin::BI__builtin_fabs:
2326 case Builtin::BI__builtin_fabsf:
2327 case Builtin::BI__builtin_fabsf16:
2328 case Builtin::BI__builtin_fabsl:
2329 case Builtin::BI__builtin_fabsf128:
2330 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs));
2331
2332 case Builtin::BIfloor:
2333 case Builtin::BIfloorf:
2334 case Builtin::BIfloorl:
2335 case Builtin::BI__builtin_floor:
2336 case Builtin::BI__builtin_floorf:
2337 case Builtin::BI__builtin_floorf16:
2338 case Builtin::BI__builtin_floorl:
2339 case Builtin::BI__builtin_floorf128:
2340 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2341 Intrinsic::floor,
2342 Intrinsic::experimental_constrained_floor));
2343
2344 case Builtin::BIfma:
2345 case Builtin::BIfmaf:
2346 case Builtin::BIfmal:
2347 case Builtin::BI__builtin_fma:
2348 case Builtin::BI__builtin_fmaf:
2349 case Builtin::BI__builtin_fmaf16:
2350 case Builtin::BI__builtin_fmal:
2351 case Builtin::BI__builtin_fmaf128:
2352 return RValue::get(emitTernaryMaybeConstrainedFPBuiltin(*this, E,
2353 Intrinsic::fma,
2354 Intrinsic::experimental_constrained_fma));
2355
2356 case Builtin::BIfmax:
2357 case Builtin::BIfmaxf:
2358 case Builtin::BIfmaxl:
2359 case Builtin::BI__builtin_fmax:
2360 case Builtin::BI__builtin_fmaxf:
2361 case Builtin::BI__builtin_fmaxf16:
2362 case Builtin::BI__builtin_fmaxl:
2363 case Builtin::BI__builtin_fmaxf128:
2364 return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E,
2365 Intrinsic::maxnum,
2366 Intrinsic::experimental_constrained_maxnum));
2367
2368 case Builtin::BIfmin:
2369 case Builtin::BIfminf:
2370 case Builtin::BIfminl:
2371 case Builtin::BI__builtin_fmin:
2372 case Builtin::BI__builtin_fminf:
2373 case Builtin::BI__builtin_fminf16:
2374 case Builtin::BI__builtin_fminl:
2375 case Builtin::BI__builtin_fminf128:
2376 return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E,
2377 Intrinsic::minnum,
2378 Intrinsic::experimental_constrained_minnum));
2379
2380 // fmod() is a special-case. It maps to the frem instruction rather than an
2381 // LLVM intrinsic.
2382 case Builtin::BIfmod:
2383 case Builtin::BIfmodf:
2384 case Builtin::BIfmodl:
2385 case Builtin::BI__builtin_fmod:
2386 case Builtin::BI__builtin_fmodf:
2387 case Builtin::BI__builtin_fmodf16:
2388 case Builtin::BI__builtin_fmodl:
2389 case Builtin::BI__builtin_fmodf128: {
2390 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
2391 Value *Arg1 = EmitScalarExpr(E->getArg(0));
2392 Value *Arg2 = EmitScalarExpr(E->getArg(1));
2393 return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod"));
2394 }
2395
2396 case Builtin::BIlog:
2397 case Builtin::BIlogf:
2398 case Builtin::BIlogl:
2399 case Builtin::BI__builtin_log:
2400 case Builtin::BI__builtin_logf:
2401 case Builtin::BI__builtin_logf16:
2402 case Builtin::BI__builtin_logl:
2403 case Builtin::BI__builtin_logf128:
2404 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2405 Intrinsic::log,
2406 Intrinsic::experimental_constrained_log));
2407
2408 case Builtin::BIlog10:
2409 case Builtin::BIlog10f:
2410 case Builtin::BIlog10l:
2411 case Builtin::BI__builtin_log10:
2412 case Builtin::BI__builtin_log10f:
2413 case Builtin::BI__builtin_log10f16:
2414 case Builtin::BI__builtin_log10l:
2415 case Builtin::BI__builtin_log10f128:
2416 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2417 Intrinsic::log10,
2418 Intrinsic::experimental_constrained_log10));
2419
2420 case Builtin::BIlog2:
2421 case Builtin::BIlog2f:
2422 case Builtin::BIlog2l:
2423 case Builtin::BI__builtin_log2:
2424 case Builtin::BI__builtin_log2f:
2425 case Builtin::BI__builtin_log2f16:
2426 case Builtin::BI__builtin_log2l:
2427 case Builtin::BI__builtin_log2f128:
2428 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2429 Intrinsic::log2,
2430 Intrinsic::experimental_constrained_log2));
2431
2432 case Builtin::BInearbyint:
2433 case Builtin::BInearbyintf:
2434 case Builtin::BInearbyintl:
2435 case Builtin::BI__builtin_nearbyint:
2436 case Builtin::BI__builtin_nearbyintf:
2437 case Builtin::BI__builtin_nearbyintl:
2438 case Builtin::BI__builtin_nearbyintf128:
2439 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2440 Intrinsic::nearbyint,
2441 Intrinsic::experimental_constrained_nearbyint));
2442
2443 case Builtin::BIpow:
2444 case Builtin::BIpowf:
2445 case Builtin::BIpowl:
2446 case Builtin::BI__builtin_pow:
2447 case Builtin::BI__builtin_powf:
2448 case Builtin::BI__builtin_powf16:
2449 case Builtin::BI__builtin_powl:
2450 case Builtin::BI__builtin_powf128:
2451 return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E,
2452 Intrinsic::pow,
2453 Intrinsic::experimental_constrained_pow));
2454
2455 case Builtin::BIrint:
2456 case Builtin::BIrintf:
2457 case Builtin::BIrintl:
2458 case Builtin::BI__builtin_rint:
2459 case Builtin::BI__builtin_rintf:
2460 case Builtin::BI__builtin_rintf16:
2461 case Builtin::BI__builtin_rintl:
2462 case Builtin::BI__builtin_rintf128:
2463 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2464 Intrinsic::rint,
2465 Intrinsic::experimental_constrained_rint));
2466
2467 case Builtin::BIround:
2468 case Builtin::BIroundf:
2469 case Builtin::BIroundl:
2470 case Builtin::BI__builtin_round:
2471 case Builtin::BI__builtin_roundf:
2472 case Builtin::BI__builtin_roundf16:
2473 case Builtin::BI__builtin_roundl:
2474 case Builtin::BI__builtin_roundf128:
2475 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2476 Intrinsic::round,
2477 Intrinsic::experimental_constrained_round));
2478
2479 case Builtin::BIsin:
2480 case Builtin::BIsinf:
2481 case Builtin::BIsinl:
2482 case Builtin::BI__builtin_sin:
2483 case Builtin::BI__builtin_sinf:
2484 case Builtin::BI__builtin_sinf16:
2485 case Builtin::BI__builtin_sinl:
2486 case Builtin::BI__builtin_sinf128:
2487 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2488 Intrinsic::sin,
2489 Intrinsic::experimental_constrained_sin));
2490
2491 case Builtin::BIsqrt:
2492 case Builtin::BIsqrtf:
2493 case Builtin::BIsqrtl:
2494 case Builtin::BI__builtin_sqrt:
2495 case Builtin::BI__builtin_sqrtf:
2496 case Builtin::BI__builtin_sqrtf16:
2497 case Builtin::BI__builtin_sqrtl:
2498 case Builtin::BI__builtin_sqrtf128:
2499 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2500 Intrinsic::sqrt,
2501 Intrinsic::experimental_constrained_sqrt));
2502
2503 case Builtin::BItrunc:
2504 case Builtin::BItruncf:
2505 case Builtin::BItruncl:
2506 case Builtin::BI__builtin_trunc:
2507 case Builtin::BI__builtin_truncf:
2508 case Builtin::BI__builtin_truncf16:
2509 case Builtin::BI__builtin_truncl:
2510 case Builtin::BI__builtin_truncf128:
2511 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E,
2512 Intrinsic::trunc,
2513 Intrinsic::experimental_constrained_trunc));
2514
2515 case Builtin::BIlround:
2516 case Builtin::BIlroundf:
2517 case Builtin::BIlroundl:
2518 case Builtin::BI__builtin_lround:
2519 case Builtin::BI__builtin_lroundf:
2520 case Builtin::BI__builtin_lroundl:
2521 case Builtin::BI__builtin_lroundf128:
2522 return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin(
2523 *this, E, Intrinsic::lround,
2524 Intrinsic::experimental_constrained_lround));
2525
2526 case Builtin::BIllround:
2527 case Builtin::BIllroundf:
2528 case Builtin::BIllroundl:
2529 case Builtin::BI__builtin_llround:
2530 case Builtin::BI__builtin_llroundf:
2531 case Builtin::BI__builtin_llroundl:
2532 case Builtin::BI__builtin_llroundf128:
2533 return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin(
2534 *this, E, Intrinsic::llround,
2535 Intrinsic::experimental_constrained_llround));
2536
2537 case Builtin::BIlrint:
2538 case Builtin::BIlrintf:
2539 case Builtin::BIlrintl:
2540 case Builtin::BI__builtin_lrint:
2541 case Builtin::BI__builtin_lrintf:
2542 case Builtin::BI__builtin_lrintl:
2543 case Builtin::BI__builtin_lrintf128:
2544 return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin(
2545 *this, E, Intrinsic::lrint,
2546 Intrinsic::experimental_constrained_lrint));
2547
2548 case Builtin::BIllrint:
2549 case Builtin::BIllrintf:
2550 case Builtin::BIllrintl:
2551 case Builtin::BI__builtin_llrint:
2552 case Builtin::BI__builtin_llrintf:
2553 case Builtin::BI__builtin_llrintl:
2554 case Builtin::BI__builtin_llrintf128:
2555 return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin(
2556 *this, E, Intrinsic::llrint,
2557 Intrinsic::experimental_constrained_llrint));
2558
2559 default:
2560 break;
2561 }
2562 }
2563
2564 switch (BuiltinIDIfNoAsmLabel) {
2565 default: break;
2566 case Builtin::BI__builtin___CFStringMakeConstantString:
2567 case Builtin::BI__builtin___NSStringMakeConstantString:
2568 return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType()));
2569 case Builtin::BI__builtin_stdarg_start:
2570 case Builtin::BI__builtin_va_start:
2571 case Builtin::BI__va_start:
2572 case Builtin::BI__builtin_va_end:
2573 return RValue::get(
2574 EmitVAStartEnd(BuiltinID == Builtin::BI__va_start
2575 ? EmitScalarExpr(E->getArg(0))
2576 : EmitVAListRef(E->getArg(0)).getPointer(),
2577 BuiltinID != Builtin::BI__builtin_va_end));
2578 case Builtin::BI__builtin_va_copy: {
2579 Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer();
2580 Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer();
2581
2582 llvm::Type *Type = Int8PtrTy;
2583
2584 DstPtr = Builder.CreateBitCast(DstPtr, Type);
2585 SrcPtr = Builder.CreateBitCast(SrcPtr, Type);
2586 return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy),
2587 {DstPtr, SrcPtr}));
2588 }
2589 case Builtin::BI__builtin_abs:
2590 case Builtin::BI__builtin_labs:
2591 case Builtin::BI__builtin_llabs: {
2592 // X < 0 ? -X : X
2593 // The negation has 'nsw' because abs of INT_MIN is undefined.
2594 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2595 Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg");
2596 Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType());
2597 Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond");
2598 Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs");
2599 return RValue::get(Result);
2600 }
2601 case Builtin::BI__builtin_complex: {
2602 Value *Real = EmitScalarExpr(E->getArg(0));
2603 Value *Imag = EmitScalarExpr(E->getArg(1));
2604 return RValue::getComplex({Real, Imag});
2605 }
2606 case Builtin::BI__builtin_conj:
2607 case Builtin::BI__builtin_conjf:
2608 case Builtin::BI__builtin_conjl:
2609 case Builtin::BIconj:
2610 case Builtin::BIconjf:
2611 case Builtin::BIconjl: {
2612 ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
2613 Value *Real = ComplexVal.first;
2614 Value *Imag = ComplexVal.second;
2615 Imag = Builder.CreateFNeg(Imag, "neg");
2616 return RValue::getComplex(std::make_pair(Real, Imag));
2617 }
2618 case Builtin::BI__builtin_creal:
2619 case Builtin::BI__builtin_crealf:
2620 case Builtin::BI__builtin_creall:
2621 case Builtin::BIcreal:
2622 case Builtin::BIcrealf:
2623 case Builtin::BIcreall: {
2624 ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
2625 return RValue::get(ComplexVal.first);
2626 }
2627
2628 case Builtin::BI__builtin_dump_struct: {
2629 llvm::Type *LLVMIntTy = getTypes().ConvertType(getContext().IntTy);
2630 llvm::FunctionType *LLVMFuncType = llvm::FunctionType::get(
2631 LLVMIntTy, {llvm::Type::getInt8PtrTy(getLLVMContext())}, true);
2632
2633 Value *Func = EmitScalarExpr(E->getArg(1)->IgnoreImpCasts());
2634 CharUnits Arg0Align = EmitPointerWithAlignment(E->getArg(0)).getAlignment();
2635
2636 const Expr *Arg0 = E->getArg(0)->IgnoreImpCasts();
2637 QualType Arg0Type = Arg0->getType()->getPointeeType();
2638
2639 Value *RecordPtr = EmitScalarExpr(Arg0);
2640 Value *Res = dumpRecord(*this, Arg0Type, RecordPtr, Arg0Align,
2641 {LLVMFuncType, Func}, 0);
2642 return RValue::get(Res);
2643 }
2644
2645 case Builtin::BI__builtin_preserve_access_index: {
2646 // Only enabled preserved access index region when debuginfo
2647 // is available as debuginfo is needed to preserve user-level
2648 // access pattern.
2649 if (!getDebugInfo()) {
2650 CGM.Error(E->getExprLoc(), "using builtin_preserve_access_index() without -g");
2651 return RValue::get(EmitScalarExpr(E->getArg(0)));
2652 }
2653
2654 // Nested builtin_preserve_access_index() not supported
2655 if (IsInPreservedAIRegion) {
2656 CGM.Error(E->getExprLoc(), "nested builtin_preserve_access_index() not supported");
2657 return RValue::get(EmitScalarExpr(E->getArg(0)));
2658 }
2659
2660 IsInPreservedAIRegion = true;
2661 Value *Res = EmitScalarExpr(E->getArg(0));
2662 IsInPreservedAIRegion = false;
2663 return RValue::get(Res);
2664 }
2665
2666 case Builtin::BI__builtin_cimag:
2667 case Builtin::BI__builtin_cimagf:
2668 case Builtin::BI__builtin_cimagl:
2669 case Builtin::BIcimag:
2670 case Builtin::BIcimagf:
2671 case Builtin::BIcimagl: {
2672 ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
2673 return RValue::get(ComplexVal.second);
2674 }
2675
2676 case Builtin::BI__builtin_clrsb:
2677 case Builtin::BI__builtin_clrsbl:
2678 case Builtin::BI__builtin_clrsbll: {
2679 // clrsb(x) -> clz(x < 0 ? ~x : x) - 1 or
2680 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2681
2682 llvm::Type *ArgType = ArgValue->getType();
2683 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
2684
2685 llvm::Type *ResultType = ConvertType(E->getType());
2686 Value *Zero = llvm::Constant::getNullValue(ArgType);
2687 Value *IsNeg = Builder.CreateICmpSLT(ArgValue, Zero, "isneg");
2688 Value *Inverse = Builder.CreateNot(ArgValue, "not");
2689 Value *Tmp = Builder.CreateSelect(IsNeg, Inverse, ArgValue);
2690 Value *Ctlz = Builder.CreateCall(F, {Tmp, Builder.getFalse()});
2691 Value *Result = Builder.CreateSub(Ctlz, llvm::ConstantInt::get(ArgType, 1));
2692 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
2693 "cast");
2694 return RValue::get(Result);
2695 }
2696 case Builtin::BI__builtin_ctzs:
2697 case Builtin::BI__builtin_ctz:
2698 case Builtin::BI__builtin_ctzl:
2699 case Builtin::BI__builtin_ctzll: {
2700 Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero);
2701
2702 llvm::Type *ArgType = ArgValue->getType();
2703 Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
2704
2705 llvm::Type *ResultType = ConvertType(E->getType());
2706 Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef());
2707 Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef});
2708 if (Result->getType() != ResultType)
2709 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
2710 "cast");
2711 return RValue::get(Result);
2712 }
2713 case Builtin::BI__builtin_clzs:
2714 case Builtin::BI__builtin_clz:
2715 case Builtin::BI__builtin_clzl:
2716 case Builtin::BI__builtin_clzll: {
2717 Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero);
2718
2719 llvm::Type *ArgType = ArgValue->getType();
2720 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
2721
2722 llvm::Type *ResultType = ConvertType(E->getType());
2723 Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef());
2724 Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef});
2725 if (Result->getType() != ResultType)
2726 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
2727 "cast");
2728 return RValue::get(Result);
2729 }
2730 case Builtin::BI__builtin_ffs:
2731 case Builtin::BI__builtin_ffsl:
2732 case Builtin::BI__builtin_ffsll: {
2733 // ffs(x) -> x ? cttz(x) + 1 : 0
2734 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2735
2736 llvm::Type *ArgType = ArgValue->getType();
2737 Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
2738
2739 llvm::Type *ResultType = ConvertType(E->getType());
2740 Value *Tmp =
2741 Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}),
2742 llvm::ConstantInt::get(ArgType, 1));
2743 Value *Zero = llvm::Constant::getNullValue(ArgType);
2744 Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero");
2745 Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs");
2746 if (Result->getType() != ResultType)
2747 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
2748 "cast");
2749 return RValue::get(Result);
2750 }
2751 case Builtin::BI__builtin_parity:
2752 case Builtin::BI__builtin_parityl:
2753 case Builtin::BI__builtin_parityll: {
2754 // parity(x) -> ctpop(x) & 1
2755 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2756
2757 llvm::Type *ArgType = ArgValue->getType();
2758 Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType);
2759
2760 llvm::Type *ResultType = ConvertType(E->getType());
2761 Value *Tmp = Builder.CreateCall(F, ArgValue);
2762 Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1));
2763 if (Result->getType() != ResultType)
2764 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
2765 "cast");
2766 return RValue::get(Result);
2767 }
2768 case Builtin::BI__lzcnt16:
2769 case Builtin::BI__lzcnt:
2770 case Builtin::BI__lzcnt64: {
2771 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2772
2773 llvm::Type *ArgType = ArgValue->getType();
2774 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
2775
2776 llvm::Type *ResultType = ConvertType(E->getType());
2777 Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()});
2778 if (Result->getType() != ResultType)
2779 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
2780 "cast");
2781 return RValue::get(Result);
2782 }
2783 case Builtin::BI__popcnt16:
2784 case Builtin::BI__popcnt:
2785 case Builtin::BI__popcnt64:
2786 case Builtin::BI__builtin_popcount:
2787 case Builtin::BI__builtin_popcountl:
2788 case Builtin::BI__builtin_popcountll: {
2789 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2790
2791 llvm::Type *ArgType = ArgValue->getType();
2792 Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType);
2793
2794 llvm::Type *ResultType = ConvertType(E->getType());
2795 Value *Result = Builder.CreateCall(F, ArgValue);
2796 if (Result->getType() != ResultType)
2797 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
2798 "cast");
2799 return RValue::get(Result);
2800 }
2801 case Builtin::BI__builtin_unpredictable: {
2802 // Always return the argument of __builtin_unpredictable. LLVM does not
2803 // handle this builtin. Metadata for this builtin should be added directly
2804 // to instructions such as branches or switches that use it.
2805 return RValue::get(EmitScalarExpr(E->getArg(0)));
2806 }
2807 case Builtin::BI__builtin_expect: {
2808 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2809 llvm::Type *ArgType = ArgValue->getType();
2810
2811 Value *ExpectedValue = EmitScalarExpr(E->getArg(1));
2812 // Don't generate llvm.expect on -O0 as the backend won't use it for
2813 // anything.
2814 // Note, we still IRGen ExpectedValue because it could have side-effects.
2815 if (CGM.getCodeGenOpts().OptimizationLevel == 0)
2816 return RValue::get(ArgValue);
2817
2818 Function *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType);
2819 Value *Result =
2820 Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval");
2821 return RValue::get(Result);
2822 }
2823 case Builtin::BI__builtin_expect_with_probability: {
2824 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2825 llvm::Type *ArgType = ArgValue->getType();
2826
2827 Value *ExpectedValue = EmitScalarExpr(E->getArg(1));
2828 llvm::APFloat Probability(0.0);
2829 const Expr *ProbArg = E->getArg(2);
2830 bool EvalSucceed = ProbArg->EvaluateAsFloat(Probability, CGM.getContext());
2831 assert(EvalSucceed && "probability should be able to evaluate as float")(static_cast <bool> (EvalSucceed && "probability should be able to evaluate as float"
) ? void (0) : __assert_fail ("EvalSucceed && \"probability should be able to evaluate as float\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 2831, __extension__ __PRETTY_FUNCTION__))
;
2832 (void)EvalSucceed;
2833 bool LoseInfo = false;
2834 Probability.convert(llvm::APFloat::IEEEdouble(),
2835 llvm::RoundingMode::Dynamic, &LoseInfo);
2836 llvm::Type *Ty = ConvertType(ProbArg->getType());
2837 Constant *Confidence = ConstantFP::get(Ty, Probability);
2838 // Don't generate llvm.expect.with.probability on -O0 as the backend
2839 // won't use it for anything.
2840 // Note, we still IRGen ExpectedValue because it could have side-effects.
2841 if (CGM.getCodeGenOpts().OptimizationLevel == 0)
2842 return RValue::get(ArgValue);
2843
2844 Function *FnExpect =
2845 CGM.getIntrinsic(Intrinsic::expect_with_probability, ArgType);
2846 Value *Result = Builder.CreateCall(
2847 FnExpect, {ArgValue, ExpectedValue, Confidence}, "expval");
2848 return RValue::get(Result);
2849 }
2850 case Builtin::BI__builtin_assume_aligned: {
2851 const Expr *Ptr = E->getArg(0);
2852 Value *PtrValue = EmitScalarExpr(Ptr);
2853 Value *OffsetValue =
2854 (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr;
2855
2856 Value *AlignmentValue = EmitScalarExpr(E->getArg(1));
2857 ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue);
2858 if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment))
2859 AlignmentCI = ConstantInt::get(AlignmentCI->getType(),
2860 llvm::Value::MaximumAlignment);
2861
2862 emitAlignmentAssumption(PtrValue, Ptr,
2863 /*The expr loc is sufficient.*/ SourceLocation(),
2864 AlignmentCI, OffsetValue);
2865 return RValue::get(PtrValue);
2866 }
2867 case Builtin::BI__assume:
2868 case Builtin::BI__builtin_assume: {
2869 if (E->getArg(0)->HasSideEffects(getContext()))
2870 return RValue::get(nullptr);
2871
2872 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2873 Function *FnAssume = CGM.getIntrinsic(Intrinsic::assume);
2874 return RValue::get(Builder.CreateCall(FnAssume, ArgValue));
2875 }
2876 case Builtin::BI__arithmetic_fence: {
2877 // Create the builtin call if FastMath is selected, and the target
2878 // supports the builtin, otherwise just return the argument.
2879 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
2880 llvm::FastMathFlags FMF = Builder.getFastMathFlags();
2881 bool isArithmeticFenceEnabled =
2882 FMF.allowReassoc() &&
2883 getContext().getTargetInfo().checkArithmeticFenceSupported();
2884 QualType ArgType = E->getArg(0)->getType();
2885 if (ArgType->isComplexType()) {
2886 if (isArithmeticFenceEnabled) {
2887 QualType ElementType = ArgType->castAs<ComplexType>()->getElementType();
2888 ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
2889 Value *Real = Builder.CreateArithmeticFence(ComplexVal.first,
2890 ConvertType(ElementType));
2891 Value *Imag = Builder.CreateArithmeticFence(ComplexVal.second,
2892 ConvertType(ElementType));
2893 return RValue::getComplex(std::make_pair(Real, Imag));
2894 }
2895 ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
2896 Value *Real = ComplexVal.first;
2897 Value *Imag = ComplexVal.second;
2898 return RValue::getComplex(std::make_pair(Real, Imag));
2899 }
2900 Value *ArgValue = EmitScalarExpr(E->getArg(0));
2901 if (isArithmeticFenceEnabled)
2902 return RValue::get(
2903 Builder.CreateArithmeticFence(ArgValue, ConvertType(ArgType)));
2904 return RValue::get(ArgValue);
2905 }
2906 case Builtin::BI__builtin_bswap16:
2907 case Builtin::BI__builtin_bswap32:
2908 case Builtin::BI__builtin_bswap64: {
2909 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap));
2910 }
2911 case Builtin::BI__builtin_bitreverse8:
2912 case Builtin::BI__builtin_bitreverse16:
2913 case Builtin::BI__builtin_bitreverse32:
2914 case Builtin::BI__builtin_bitreverse64: {
2915 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse));
2916 }
2917 case Builtin::BI__builtin_rotateleft8:
2918 case Builtin::BI__builtin_rotateleft16:
2919 case Builtin::BI__builtin_rotateleft32:
2920 case Builtin::BI__builtin_rotateleft64:
2921 case Builtin::BI_rotl8: // Microsoft variants of rotate left
2922 case Builtin::BI_rotl16:
2923 case Builtin::BI_rotl:
2924 case Builtin::BI_lrotl:
2925 case Builtin::BI_rotl64:
2926 return emitRotate(E, false);
2927
2928 case Builtin::BI__builtin_rotateright8:
2929 case Builtin::BI__builtin_rotateright16:
2930 case Builtin::BI__builtin_rotateright32:
2931 case Builtin::BI__builtin_rotateright64:
2932 case Builtin::BI_rotr8: // Microsoft variants of rotate right
2933 case Builtin::BI_rotr16:
2934 case Builtin::BI_rotr:
2935 case Builtin::BI_lrotr:
2936 case Builtin::BI_rotr64:
2937 return emitRotate(E, true);
2938
2939 case Builtin::BI__builtin_constant_p: {
2940 llvm::Type *ResultType = ConvertType(E->getType());
2941
2942 const Expr *Arg = E->getArg(0);
2943 QualType ArgType = Arg->getType();
2944 // FIXME: The allowance for Obj-C pointers and block pointers is historical
2945 // and likely a mistake.
2946 if (!ArgType->isIntegralOrEnumerationType() && !ArgType->isFloatingType() &&
2947 !ArgType->isObjCObjectPointerType() && !ArgType->isBlockPointerType())
2948 // Per the GCC documentation, only numeric constants are recognized after
2949 // inlining.
2950 return RValue::get(ConstantInt::get(ResultType, 0));
2951
2952 if (Arg->HasSideEffects(getContext()))
2953 // The argument is unevaluated, so be conservative if it might have
2954 // side-effects.
2955 return RValue::get(ConstantInt::get(ResultType, 0));
2956
2957 Value *ArgValue = EmitScalarExpr(Arg);
2958 if (ArgType->isObjCObjectPointerType()) {
2959 // Convert Objective-C objects to id because we cannot distinguish between
2960 // LLVM types for Obj-C classes as they are opaque.
2961 ArgType = CGM.getContext().getObjCIdType();
2962 ArgValue = Builder.CreateBitCast(ArgValue, ConvertType(ArgType));
2963 }
2964 Function *F =
2965 CGM.getIntrinsic(Intrinsic::is_constant, ConvertType(ArgType));
2966 Value *Result = Builder.CreateCall(F, ArgValue);
2967 if (Result->getType() != ResultType)
2968 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/false);
2969 return RValue::get(Result);
2970 }
2971 case Builtin::BI__builtin_dynamic_object_size:
2972 case Builtin::BI__builtin_object_size: {
2973 unsigned Type =
2974 E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue();
2975 auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType()));
2976
2977 // We pass this builtin onto the optimizer so that it can figure out the
2978 // object size in more complex cases.
2979 bool IsDynamic = BuiltinID == Builtin::BI__builtin_dynamic_object_size;
2980 return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType,
2981 /*EmittedE=*/nullptr, IsDynamic));
2982 }
2983 case Builtin::BI__builtin_prefetch: {
2984 Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0));
2985 // FIXME: Technically these constants should of type 'int', yes?
2986 RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) :
2987 llvm::ConstantInt::get(Int32Ty, 0);
2988 Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) :
2989 llvm::ConstantInt::get(Int32Ty, 3);
2990 Value *Data = llvm::ConstantInt::get(Int32Ty, 1);
2991 Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType());
2992 return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data}));
2993 }
2994 case Builtin::BI__builtin_readcyclecounter: {
2995 Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter);
2996 return RValue::get(Builder.CreateCall(F));
2997 }
2998 case Builtin::BI__builtin___clear_cache: {
2999 Value *Begin = EmitScalarExpr(E->getArg(0));
3000 Value *End = EmitScalarExpr(E->getArg(1));
3001 Function *F = CGM.getIntrinsic(Intrinsic::clear_cache);
3002 return RValue::get(Builder.CreateCall(F, {Begin, End}));
3003 }
3004 case Builtin::BI__builtin_trap:
3005 return RValue::get(EmitTrapCall(Intrinsic::trap));
3006 case Builtin::BI__debugbreak:
3007 return RValue::get(EmitTrapCall(Intrinsic::debugtrap));
3008 case Builtin::BI__builtin_unreachable: {
3009 EmitUnreachable(E->getExprLoc());
3010
3011 // We do need to preserve an insertion point.
3012 EmitBlock(createBasicBlock("unreachable.cont"));
3013
3014 return RValue::get(nullptr);
3015 }
3016
3017 case Builtin::BI__builtin_powi:
3018 case Builtin::BI__builtin_powif:
3019 case Builtin::BI__builtin_powil: {
3020 llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
3021 llvm::Value *Src1 = EmitScalarExpr(E->getArg(1));
3022
3023 if (Builder.getIsFPConstrained()) {
3024 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
3025 Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_powi,
3026 Src0->getType());
3027 return RValue::get(Builder.CreateConstrainedFPCall(F, { Src0, Src1 }));
3028 }
3029
3030 Function *F = CGM.getIntrinsic(Intrinsic::powi,
3031 { Src0->getType(), Src1->getType() });
3032 return RValue::get(Builder.CreateCall(F, { Src0, Src1 }));
3033 }
3034 case Builtin::BI__builtin_isgreater:
3035 case Builtin::BI__builtin_isgreaterequal:
3036 case Builtin::BI__builtin_isless:
3037 case Builtin::BI__builtin_islessequal:
3038 case Builtin::BI__builtin_islessgreater:
3039 case Builtin::BI__builtin_isunordered: {
3040 // Ordered comparisons: we know the arguments to these are matching scalar
3041 // floating point values.
3042 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
3043 // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here.
3044 Value *LHS = EmitScalarExpr(E->getArg(0));
3045 Value *RHS = EmitScalarExpr(E->getArg(1));
3046
3047 switch (BuiltinID) {
3048 default: llvm_unreachable("Unknown ordered comparison")::llvm::llvm_unreachable_internal("Unknown ordered comparison"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 3048)
;
3049 case Builtin::BI__builtin_isgreater:
3050 LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp");
3051 break;
3052 case Builtin::BI__builtin_isgreaterequal:
3053 LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp");
3054 break;
3055 case Builtin::BI__builtin_isless:
3056 LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp");
3057 break;
3058 case Builtin::BI__builtin_islessequal:
3059 LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp");
3060 break;
3061 case Builtin::BI__builtin_islessgreater:
3062 LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp");
3063 break;
3064 case Builtin::BI__builtin_isunordered:
3065 LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp");
3066 break;
3067 }
3068 // ZExt bool to int type.
3069 return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType())));
3070 }
3071
3072 case Builtin::BI__builtin_isnan: {
3073 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
3074 Value *V = EmitScalarExpr(E->getArg(0));
3075
3076 if (Value *Result = getTargetHooks().testFPKind(V, BuiltinID, Builder, CGM))
3077 return RValue::get(Result);
3078
3079 Function *F = CGM.getIntrinsic(Intrinsic::isnan, V->getType());
3080 Value *Call = Builder.CreateCall(F, V);
3081 return RValue::get(Builder.CreateZExt(Call, ConvertType(E->getType())));
3082 }
3083
3084 case Builtin::BI__builtin_matrix_transpose: {
3085 const auto *MatrixTy = E->getArg(0)->getType()->getAs<ConstantMatrixType>();
3086 Value *MatValue = EmitScalarExpr(E->getArg(0));
3087 MatrixBuilder<CGBuilderTy> MB(Builder);
3088 Value *Result = MB.CreateMatrixTranspose(MatValue, MatrixTy->getNumRows(),
3089 MatrixTy->getNumColumns());
3090 return RValue::get(Result);
3091 }
3092
3093 case Builtin::BI__builtin_matrix_column_major_load: {
3094 MatrixBuilder<CGBuilderTy> MB(Builder);
3095 // Emit everything that isn't dependent on the first parameter type
3096 Value *Stride = EmitScalarExpr(E->getArg(3));
3097 const auto *ResultTy = E->getType()->getAs<ConstantMatrixType>();
3098 auto *PtrTy = E->getArg(0)->getType()->getAs<PointerType>();
3099 assert(PtrTy && "arg0 must be of pointer type")(static_cast <bool> (PtrTy && "arg0 must be of pointer type"
) ? void (0) : __assert_fail ("PtrTy && \"arg0 must be of pointer type\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 3099, __extension__ __PRETTY_FUNCTION__))
;
3100 bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified();
3101
3102 Address Src = EmitPointerWithAlignment(E->getArg(0));
3103 EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(0)->getType(),
3104 E->getArg(0)->getExprLoc(), FD, 0);
3105 Value *Result = MB.CreateColumnMajorLoad(
3106 Src.getPointer(), Align(Src.getAlignment().getQuantity()), Stride,
3107 IsVolatile, ResultTy->getNumRows(), ResultTy->getNumColumns(),
3108 "matrix");
3109 return RValue::get(Result);
3110 }
3111
3112 case Builtin::BI__builtin_matrix_column_major_store: {
3113 MatrixBuilder<CGBuilderTy> MB(Builder);
3114 Value *Matrix = EmitScalarExpr(E->getArg(0));
3115 Address Dst = EmitPointerWithAlignment(E->getArg(1));
3116 Value *Stride = EmitScalarExpr(E->getArg(2));
3117
3118 const auto *MatrixTy = E->getArg(0)->getType()->getAs<ConstantMatrixType>();
3119 auto *PtrTy = E->getArg(1)->getType()->getAs<PointerType>();
3120 assert(PtrTy && "arg1 must be of pointer type")(static_cast <bool> (PtrTy && "arg1 must be of pointer type"
) ? void (0) : __assert_fail ("PtrTy && \"arg1 must be of pointer type\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 3120, __extension__ __PRETTY_FUNCTION__))
;
3121 bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified();
3122
3123 EmitNonNullArgCheck(RValue::get(Dst.getPointer()), E->getArg(1)->getType(),
3124 E->getArg(1)->getExprLoc(), FD, 0);
3125 Value *Result = MB.CreateColumnMajorStore(
3126 Matrix, Dst.getPointer(), Align(Dst.getAlignment().getQuantity()),
3127 Stride, IsVolatile, MatrixTy->getNumRows(), MatrixTy->getNumColumns());
3128 return RValue::get(Result);
3129 }
3130
3131 case Builtin::BIfinite:
3132 case Builtin::BI__finite:
3133 case Builtin::BIfinitef:
3134 case Builtin::BI__finitef:
3135 case Builtin::BIfinitel:
3136 case Builtin::BI__finitel:
3137 case Builtin::BI__builtin_isinf:
3138 case Builtin::BI__builtin_isfinite: {
3139 // isinf(x) --> fabs(x) == infinity
3140 // isfinite(x) --> fabs(x) != infinity
3141 // x != NaN via the ordered compare in either case.
3142 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
3143 Value *V = EmitScalarExpr(E->getArg(0));
3144 llvm::Type *Ty = V->getType();
3145 if (!Builder.getIsFPConstrained() ||
3146 Builder.getDefaultConstrainedExcept() == fp::ebIgnore ||
3147 !Ty->isIEEE()) {
3148 Value *Fabs = EmitFAbs(*this, V);
3149 Constant *Infinity = ConstantFP::getInfinity(V->getType());
3150 CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf)
3151 ? CmpInst::FCMP_OEQ
3152 : CmpInst::FCMP_ONE;
3153 Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf");
3154 return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType())));
3155 }
3156
3157 if (Value *Result = getTargetHooks().testFPKind(V, BuiltinID, Builder, CGM))
3158 return RValue::get(Result);
3159
3160 // Inf values have all exp bits set and a zero significand. Therefore:
3161 // isinf(V) == ((V << 1) == ((exp mask) << 1))
3162 // isfinite(V) == ((V << 1) < ((exp mask) << 1)) using unsigned comparison
3163 unsigned bitsize = Ty->getScalarSizeInBits();
3164 llvm::IntegerType *IntTy = Builder.getIntNTy(bitsize);
3165 Value *IntV = Builder.CreateBitCast(V, IntTy);
3166 Value *Shl1 = Builder.CreateShl(IntV, 1);
3167 const llvm::fltSemantics &Semantics = Ty->getFltSemantics();
3168 APInt ExpMask = APFloat::getInf(Semantics).bitcastToAPInt();
3169 Value *ExpMaskShl1 = llvm::ConstantInt::get(IntTy, ExpMask.shl(1));
3170 if (BuiltinID == Builtin::BI__builtin_isinf)
3171 V = Builder.CreateICmpEQ(Shl1, ExpMaskShl1);
3172 else
3173 V = Builder.CreateICmpULT(Shl1, ExpMaskShl1);
3174 return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType())));
3175 }
3176
3177 case Builtin::BI__builtin_isinf_sign: {
3178 // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0
3179 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
3180 // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here.
3181 Value *Arg = EmitScalarExpr(E->getArg(0));
3182 Value *AbsArg = EmitFAbs(*this, Arg);
3183 Value *IsInf = Builder.CreateFCmpOEQ(
3184 AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf");
3185 Value *IsNeg = EmitSignBit(*this, Arg);
3186
3187 llvm::Type *IntTy = ConvertType(E->getType());
3188 Value *Zero = Constant::getNullValue(IntTy);
3189 Value *One = ConstantInt::get(IntTy, 1);
3190 Value *NegativeOne = ConstantInt::get(IntTy, -1);
3191 Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One);
3192 Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero);
3193 return RValue::get(Result);
3194 }
3195
3196 case Builtin::BI__builtin_isnormal: {
3197 // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min
3198 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
3199 // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here.
3200 Value *V = EmitScalarExpr(E->getArg(0));
3201 Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq");
3202
3203 Value *Abs = EmitFAbs(*this, V);
3204 Value *IsLessThanInf =
3205 Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf");
3206 APFloat Smallest = APFloat::getSmallestNormalized(
3207 getContext().getFloatTypeSemantics(E->getArg(0)->getType()));
3208 Value *IsNormal =
3209 Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest),
3210 "isnormal");
3211 V = Builder.CreateAnd(Eq, IsLessThanInf, "and");
3212 V = Builder.CreateAnd(V, IsNormal, "and");
3213 return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType())));
3214 }
3215
3216 case Builtin::BI__builtin_flt_rounds: {
3217 Function *F = CGM.getIntrinsic(Intrinsic::flt_rounds);
3218
3219 llvm::Type *ResultType = ConvertType(E->getType());
3220 Value *Result = Builder.CreateCall(F);
3221 if (Result->getType() != ResultType)
3222 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
3223 "cast");
3224 return RValue::get(Result);
3225 }
3226
3227 case Builtin::BI__builtin_fpclassify: {
3228 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
3229 // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here.
3230 Value *V = EmitScalarExpr(E->getArg(5));
3231 llvm::Type *Ty = ConvertType(E->getArg(5)->getType());
3232
3233 // Create Result
3234 BasicBlock *Begin = Builder.GetInsertBlock();
3235 BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn);
3236 Builder.SetInsertPoint(End);
3237 PHINode *Result =
3238 Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4,
3239 "fpclassify_result");
3240
3241 // if (V==0) return FP_ZERO
3242 Builder.SetInsertPoint(Begin);
3243 Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty),
3244 "iszero");
3245 Value *ZeroLiteral = EmitScalarExpr(E->getArg(4));
3246 BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn);
3247 Builder.CreateCondBr(IsZero, End, NotZero);
3248 Result->addIncoming(ZeroLiteral, Begin);
3249
3250 // if (V != V) return FP_NAN
3251 Builder.SetInsertPoint(NotZero);
3252 Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp");
3253 Value *NanLiteral = EmitScalarExpr(E->getArg(0));
3254 BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn);
3255 Builder.CreateCondBr(IsNan, End, NotNan);
3256 Result->addIncoming(NanLiteral, NotZero);
3257
3258 // if (fabs(V) == infinity) return FP_INFINITY
3259 Builder.SetInsertPoint(NotNan);
3260 Value *VAbs = EmitFAbs(*this, V);
3261 Value *IsInf =
3262 Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()),
3263 "isinf");
3264 Value *InfLiteral = EmitScalarExpr(E->getArg(1));
3265 BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn);
3266 Builder.CreateCondBr(IsInf, End, NotInf);
3267 Result->addIncoming(InfLiteral, NotNan);
3268
3269 // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL
3270 Builder.SetInsertPoint(NotInf);
3271 APFloat Smallest = APFloat::getSmallestNormalized(
3272 getContext().getFloatTypeSemantics(E->getArg(5)->getType()));
3273 Value *IsNormal =
3274 Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest),
3275 "isnormal");
3276 Value *NormalResult =
3277 Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)),
3278 EmitScalarExpr(E->getArg(3)));
3279 Builder.CreateBr(End);
3280 Result->addIncoming(NormalResult, NotInf);
3281
3282 // return Result
3283 Builder.SetInsertPoint(End);
3284 return RValue::get(Result);
3285 }
3286
3287 case Builtin::BIalloca:
3288 case Builtin::BI_alloca:
3289 case Builtin::BI__builtin_alloca: {
3290 Value *Size = EmitScalarExpr(E->getArg(0));
3291 const TargetInfo &TI = getContext().getTargetInfo();
3292 // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__.
3293 const Align SuitableAlignmentInBytes =
3294 CGM.getContext()
3295 .toCharUnitsFromBits(TI.getSuitableAlign())
3296 .getAsAlign();
3297 AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size);
3298 AI->setAlignment(SuitableAlignmentInBytes);
3299 initializeAlloca(*this, AI, Size, SuitableAlignmentInBytes);
3300 return RValue::get(AI);
3301 }
3302
3303 case Builtin::BI__builtin_alloca_with_align: {
3304 Value *Size = EmitScalarExpr(E->getArg(0));
3305 Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1));
3306 auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue);
3307 unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue();
3308 const Align AlignmentInBytes =
3309 CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getAsAlign();
3310 AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size);
3311 AI->setAlignment(AlignmentInBytes);
3312 initializeAlloca(*this, AI, Size, AlignmentInBytes);
3313 return RValue::get(AI);
3314 }
3315
3316 case Builtin::BIbzero:
3317 case Builtin::BI__builtin_bzero: {
3318 Address Dest = EmitPointerWithAlignment(E->getArg(0));
3319 Value *SizeVal = EmitScalarExpr(E->getArg(1));
3320 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
3321 E->getArg(0)->getExprLoc(), FD, 0);
3322 Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false);
3323 return RValue::get(nullptr);
3324 }
3325 case Builtin::BImemcpy:
3326 case Builtin::BI__builtin_memcpy:
3327 case Builtin::BImempcpy:
3328 case Builtin::BI__builtin_mempcpy: {
3329 Address Dest = EmitPointerWithAlignment(E->getArg(0));
3330 Address Src = EmitPointerWithAlignment(E->getArg(1));
3331 Value *SizeVal = EmitScalarExpr(E->getArg(2));
3332 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
3333 E->getArg(0)->getExprLoc(), FD, 0);
3334 EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(),
3335 E->getArg(1)->getExprLoc(), FD, 1);
3336 Builder.CreateMemCpy(Dest, Src, SizeVal, false);
3337 if (BuiltinID == Builtin::BImempcpy ||
3338 BuiltinID == Builtin::BI__builtin_mempcpy)
3339 return RValue::get(Builder.CreateInBoundsGEP(Dest.getElementType(),
3340 Dest.getPointer(), SizeVal));
3341 else
3342 return RValue::get(Dest.getPointer());
3343 }
3344
3345 case Builtin::BI__builtin_memcpy_inline: {
3346 Address Dest = EmitPointerWithAlignment(E->getArg(0));
3347 Address Src = EmitPointerWithAlignment(E->getArg(1));
3348 uint64_t Size =
3349 E->getArg(2)->EvaluateKnownConstInt(getContext()).getZExtValue();
3350 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
3351 E->getArg(0)->getExprLoc(), FD, 0);
3352 EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(),
3353 E->getArg(1)->getExprLoc(), FD, 1);
3354 Builder.CreateMemCpyInline(Dest, Src, Size);
3355 return RValue::get(nullptr);
3356 }
3357
3358 case Builtin::BI__builtin_char_memchr:
3359 BuiltinID = Builtin::BI__builtin_memchr;
3360 break;
3361
3362 case Builtin::BI__builtin___memcpy_chk: {
3363 // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2.
3364 Expr::EvalResult SizeResult, DstSizeResult;
3365 if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) ||
3366 !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext()))
3367 break;
3368 llvm::APSInt Size = SizeResult.Val.getInt();
3369 llvm::APSInt DstSize = DstSizeResult.Val.getInt();
3370 if (Size.ugt(DstSize))
3371 break;
3372 Address Dest = EmitPointerWithAlignment(E->getArg(0));
3373 Address Src = EmitPointerWithAlignment(E->getArg(1));
3374 Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
3375 Builder.CreateMemCpy(Dest, Src, SizeVal, false);
3376 return RValue::get(Dest.getPointer());
3377 }
3378
3379 case Builtin::BI__builtin_objc_memmove_collectable: {
3380 Address DestAddr = EmitPointerWithAlignment(E->getArg(0));
3381 Address SrcAddr = EmitPointerWithAlignment(E->getArg(1));
3382 Value *SizeVal = EmitScalarExpr(E->getArg(2));
3383 CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this,
3384 DestAddr, SrcAddr, SizeVal);
3385 return RValue::get(DestAddr.getPointer());
3386 }
3387
3388 case Builtin::BI__builtin___memmove_chk: {
3389 // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2.
3390 Expr::EvalResult SizeResult, DstSizeResult;
3391 if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) ||
3392 !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext()))
3393 break;
3394 llvm::APSInt Size = SizeResult.Val.getInt();
3395 llvm::APSInt DstSize = DstSizeResult.Val.getInt();
3396 if (Size.ugt(DstSize))
3397 break;
3398 Address Dest = EmitPointerWithAlignment(E->getArg(0));
3399 Address Src = EmitPointerWithAlignment(E->getArg(1));
3400 Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
3401 Builder.CreateMemMove(Dest, Src, SizeVal, false);
3402 return RValue::get(Dest.getPointer());
3403 }
3404
3405 case Builtin::BImemmove:
3406 case Builtin::BI__builtin_memmove: {
3407 Address Dest = EmitPointerWithAlignment(E->getArg(0));
3408 Address Src = EmitPointerWithAlignment(E->getArg(1));
3409 Value *SizeVal = EmitScalarExpr(E->getArg(2));
3410 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
3411 E->getArg(0)->getExprLoc(), FD, 0);
3412 EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(),
3413 E->getArg(1)->getExprLoc(), FD, 1);
3414 Builder.CreateMemMove(Dest, Src, SizeVal, false);
3415 return RValue::get(Dest.getPointer());
3416 }
3417 case Builtin::BImemset:
3418 case Builtin::BI__builtin_memset: {
3419 Address Dest = EmitPointerWithAlignment(E->getArg(0));
3420 Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)),
3421 Builder.getInt8Ty());
3422 Value *SizeVal = EmitScalarExpr(E->getArg(2));
3423 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
3424 E->getArg(0)->getExprLoc(), FD, 0);
3425 Builder.CreateMemSet(Dest, ByteVal, SizeVal, false);
3426 return RValue::get(Dest.getPointer());
3427 }
3428 case Builtin::BI__builtin___memset_chk: {
3429 // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2.
3430 Expr::EvalResult SizeResult, DstSizeResult;
3431 if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) ||
3432 !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext()))
3433 break;
3434 llvm::APSInt Size = SizeResult.Val.getInt();
3435 llvm::APSInt DstSize = DstSizeResult.Val.getInt();
3436 if (Size.ugt(DstSize))
3437 break;
3438 Address Dest = EmitPointerWithAlignment(E->getArg(0));
3439 Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)),
3440 Builder.getInt8Ty());
3441 Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
3442 Builder.CreateMemSet(Dest, ByteVal, SizeVal, false);
3443 return RValue::get(Dest.getPointer());
3444 }
3445 case Builtin::BI__builtin_wmemchr: {
3446 // The MSVC runtime library does not provide a definition of wmemchr, so we
3447 // need an inline implementation.
3448 if (!getTarget().getTriple().isOSMSVCRT())
3449 break;
3450
3451 llvm::Type *WCharTy = ConvertType(getContext().WCharTy);
3452 Value *Str = EmitScalarExpr(E->getArg(0));
3453 Value *Chr = EmitScalarExpr(E->getArg(1));
3454 Value *Size = EmitScalarExpr(E->getArg(2));
3455
3456 BasicBlock *Entry = Builder.GetInsertBlock();
3457 BasicBlock *CmpEq = createBasicBlock("wmemchr.eq");
3458 BasicBlock *Next = createBasicBlock("wmemchr.next");
3459 BasicBlock *Exit = createBasicBlock("wmemchr.exit");
3460 Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0));
3461 Builder.CreateCondBr(SizeEq0, Exit, CmpEq);
3462
3463 EmitBlock(CmpEq);
3464 PHINode *StrPhi = Builder.CreatePHI(Str->getType(), 2);
3465 StrPhi->addIncoming(Str, Entry);
3466 PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2);
3467 SizePhi->addIncoming(Size, Entry);
3468 CharUnits WCharAlign =
3469 getContext().getTypeAlignInChars(getContext().WCharTy);
3470 Value *StrCh = Builder.CreateAlignedLoad(WCharTy, StrPhi, WCharAlign);
3471 Value *FoundChr = Builder.CreateConstInBoundsGEP1_32(WCharTy, StrPhi, 0);
3472 Value *StrEqChr = Builder.CreateICmpEQ(StrCh, Chr);
3473 Builder.CreateCondBr(StrEqChr, Exit, Next);
3474
3475 EmitBlock(Next);
3476 Value *NextStr = Builder.CreateConstInBoundsGEP1_32(WCharTy, StrPhi, 1);
3477 Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1));
3478 Value *NextSizeEq0 =
3479 Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0));
3480 Builder.CreateCondBr(NextSizeEq0, Exit, CmpEq);
3481 StrPhi->addIncoming(NextStr, Next);
3482 SizePhi->addIncoming(NextSize, Next);
3483
3484 EmitBlock(Exit);
3485 PHINode *Ret = Builder.CreatePHI(Str->getType(), 3);
3486 Ret->addIncoming(llvm::Constant::getNullValue(Str->getType()), Entry);
3487 Ret->addIncoming(llvm::Constant::getNullValue(Str->getType()), Next);
3488 Ret->addIncoming(FoundChr, CmpEq);
3489 return RValue::get(Ret);
3490 }
3491 case Builtin::BI__builtin_wmemcmp: {
3492 // The MSVC runtime library does not provide a definition of wmemcmp, so we
3493 // need an inline implementation.
3494 if (!getTarget().getTriple().isOSMSVCRT())
3495 break;
3496
3497 llvm::Type *WCharTy = ConvertType(getContext().WCharTy);
3498
3499 Value *Dst = EmitScalarExpr(E->getArg(0));
3500 Value *Src = EmitScalarExpr(E->getArg(1));
3501 Value *Size = EmitScalarExpr(E->getArg(2));
3502
3503 BasicBlock *Entry = Builder.GetInsertBlock();
3504 BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt");
3505 BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt");
3506 BasicBlock *Next = createBasicBlock("wmemcmp.next");
3507 BasicBlock *Exit = createBasicBlock("wmemcmp.exit");
3508 Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0));
3509 Builder.CreateCondBr(SizeEq0, Exit, CmpGT);
3510
3511 EmitBlock(CmpGT);
3512 PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2);
3513 DstPhi->addIncoming(Dst, Entry);
3514 PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2);
3515 SrcPhi->addIncoming(Src, Entry);
3516 PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2);
3517 SizePhi->addIncoming(Size, Entry);
3518 CharUnits WCharAlign =
3519 getContext().getTypeAlignInChars(getContext().WCharTy);
3520 Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign);
3521 Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign);
3522 Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh);
3523 Builder.CreateCondBr(DstGtSrc, Exit, CmpLT);
3524
3525 EmitBlock(CmpLT);
3526 Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh);
3527 Builder.CreateCondBr(DstLtSrc, Exit, Next);
3528
3529 EmitBlock(Next);
3530 Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1);
3531 Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1);
3532 Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1));
3533 Value *NextSizeEq0 =
3534 Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0));
3535 Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT);
3536 DstPhi->addIncoming(NextDst, Next);
3537 SrcPhi->addIncoming(NextSrc, Next);
3538 SizePhi->addIncoming(NextSize, Next);
3539
3540 EmitBlock(Exit);
3541 PHINode *Ret = Builder.CreatePHI(IntTy, 4);
3542 Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry);
3543 Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT);
3544 Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT);
3545 Ret->addIncoming(ConstantInt::get(IntTy, 0), Next);
3546 return RValue::get(Ret);
3547 }
3548 case Builtin::BI__builtin_dwarf_cfa: {
3549 // The offset in bytes from the first argument to the CFA.
3550 //
3551 // Why on earth is this in the frontend? Is there any reason at
3552 // all that the backend can't reasonably determine this while
3553 // lowering llvm.eh.dwarf.cfa()?
3554 //
3555 // TODO: If there's a satisfactory reason, add a target hook for
3556 // this instead of hard-coding 0, which is correct for most targets.
3557 int32_t Offset = 0;
3558
3559 Function *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa);
3560 return RValue::get(Builder.CreateCall(F,
3561 llvm::ConstantInt::get(Int32Ty, Offset)));
3562 }
3563 case Builtin::BI__builtin_return_address: {
3564 Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0),
3565 getContext().UnsignedIntTy);
3566 Function *F = CGM.getIntrinsic(Intrinsic::returnaddress);
3567 return RValue::get(Builder.CreateCall(F, Depth));
3568 }
3569 case Builtin::BI_ReturnAddress: {
3570 Function *F = CGM.getIntrinsic(Intrinsic::returnaddress);
3571 return RValue::get(Builder.CreateCall(F, Builder.getInt32(0)));
3572 }
3573 case Builtin::BI__builtin_frame_address: {
3574 Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0),
3575 getContext().UnsignedIntTy);
3576 Function *F = CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy);
3577 return RValue::get(Builder.CreateCall(F, Depth));
3578 }
3579 case Builtin::BI__builtin_extract_return_addr: {
3580 Value *Address = EmitScalarExpr(E->getArg(0));
3581 Value *Result = getTargetHooks().decodeReturnAddress(*this, Address);
3582 return RValue::get(Result);
3583 }
3584 case Builtin::BI__builtin_frob_return_addr: {
3585 Value *Address = EmitScalarExpr(E->getArg(0));
3586 Value *Result = getTargetHooks().encodeReturnAddress(*this, Address);
3587 return RValue::get(Result);
3588 }
3589 case Builtin::BI__builtin_dwarf_sp_column: {
3590 llvm::IntegerType *Ty
3591 = cast<llvm::IntegerType>(ConvertType(E->getType()));
3592 int Column = getTargetHooks().getDwarfEHStackPointer(CGM);
3593 if (Column == -1) {
3594 CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column");
3595 return RValue::get(llvm::UndefValue::get(Ty));
3596 }
3597 return RValue::get(llvm::ConstantInt::get(Ty, Column, true));
3598 }
3599 case Builtin::BI__builtin_init_dwarf_reg_size_table: {
3600 Value *Address = EmitScalarExpr(E->getArg(0));
3601 if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address))
3602 CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table");
3603 return RValue::get(llvm::UndefValue::get(ConvertType(E->getType())));
3604 }
3605 case Builtin::BI__builtin_eh_return: {
3606 Value *Int = EmitScalarExpr(E->getArg(0));
3607 Value *Ptr = EmitScalarExpr(E->getArg(1));
3608
3609 llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType());
3610 assert((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) &&(static_cast <bool> ((IntTy->getBitWidth() == 32 || IntTy
->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants"
) ? void (0) : __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 3611, __extension__ __PRETTY_FUNCTION__))
3611 "LLVM's __builtin_eh_return only supports 32- and 64-bit variants")(static_cast <bool> ((IntTy->getBitWidth() == 32 || IntTy
->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants"
) ? void (0) : __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 3611, __extension__ __PRETTY_FUNCTION__))
;
3612 Function *F =
3613 CGM.getIntrinsic(IntTy->getBitWidth() == 32 ? Intrinsic::eh_return_i32
3614 : Intrinsic::eh_return_i64);
3615 Builder.CreateCall(F, {Int, Ptr});
3616 Builder.CreateUnreachable();
3617
3618 // We do need to preserve an insertion point.
3619 EmitBlock(createBasicBlock("builtin_eh_return.cont"));
3620
3621 return RValue::get(nullptr);
3622 }
3623 case Builtin::BI__builtin_unwind_init: {
3624 Function *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init);
3625 return RValue::get(Builder.CreateCall(F));
3626 }
3627 case Builtin::BI__builtin_extend_pointer: {
3628 // Extends a pointer to the size of an _Unwind_Word, which is
3629 // uint64_t on all platforms. Generally this gets poked into a
3630 // register and eventually used as an address, so if the
3631 // addressing registers are wider than pointers and the platform
3632 // doesn't implicitly ignore high-order bits when doing
3633 // addressing, we need to make sure we zext / sext based on
3634 // the platform's expectations.
3635 //
3636 // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html
3637
3638 // Cast the pointer to intptr_t.
3639 Value *Ptr = EmitScalarExpr(E->getArg(0));
3640 Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast");
3641
3642 // If that's 64 bits, we're done.
3643 if (IntPtrTy->getBitWidth() == 64)
3644 return RValue::get(Result);
3645
3646 // Otherwise, ask the codegen data what to do.
3647 if (getTargetHooks().extendPointerWithSExt())
3648 return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext"));
3649 else
3650 return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext"));
3651 }
3652 case Builtin::BI__builtin_setjmp: {
3653 // Buffer is a void**.
3654 Address Buf = EmitPointerWithAlignment(E->getArg(0));
3655
3656 // Store the frame pointer to the setjmp buffer.
3657 Value *FrameAddr = Builder.CreateCall(
3658 CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy),
3659 ConstantInt::get(Int32Ty, 0));
3660 Builder.CreateStore(FrameAddr, Buf);
3661
3662 // Store the stack pointer to the setjmp buffer.
3663 Value *StackAddr =
3664 Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave));
3665 Address StackSaveSlot = Builder.CreateConstInBoundsGEP(Buf, 2);
3666 Builder.CreateStore(StackAddr, StackSaveSlot);
3667
3668 // Call LLVM's EH setjmp, which is lightweight.
3669 Function *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp);
3670 Buf = Builder.CreateBitCast(Buf, Int8PtrTy);
3671 return RValue::get(Builder.CreateCall(F, Buf.getPointer()));
3672 }
3673 case Builtin::BI__builtin_longjmp: {
3674 Value *Buf = EmitScalarExpr(E->getArg(0));
3675 Buf = Builder.CreateBitCast(Buf, Int8PtrTy);
3676
3677 // Call LLVM's EH longjmp, which is lightweight.
3678 Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf);
3679
3680 // longjmp doesn't return; mark this as unreachable.
3681 Builder.CreateUnreachable();
3682
3683 // We do need to preserve an insertion point.
3684 EmitBlock(createBasicBlock("longjmp.cont"));
3685
3686 return RValue::get(nullptr);
3687 }
3688 case Builtin::BI__builtin_launder: {
3689 const Expr *Arg = E->getArg(0);
3690 QualType ArgTy = Arg->getType()->getPointeeType();
3691 Value *Ptr = EmitScalarExpr(Arg);
3692 if (TypeRequiresBuiltinLaunder(CGM, ArgTy))
3693 Ptr = Builder.CreateLaunderInvariantGroup(Ptr);
3694
3695 return RValue::get(Ptr);
3696 }
3697 case Builtin::BI__sync_fetch_and_add:
3698 case Builtin::BI__sync_fetch_and_sub:
3699 case Builtin::BI__sync_fetch_and_or:
3700 case Builtin::BI__sync_fetch_and_and:
3701 case Builtin::BI__sync_fetch_and_xor:
3702 case Builtin::BI__sync_fetch_and_nand:
3703 case Builtin::BI__sync_add_and_fetch:
3704 case Builtin::BI__sync_sub_and_fetch:
3705 case Builtin::BI__sync_and_and_fetch:
3706 case Builtin::BI__sync_or_and_fetch:
3707 case Builtin::BI__sync_xor_and_fetch:
3708 case Builtin::BI__sync_nand_and_fetch:
3709 case Builtin::BI__sync_val_compare_and_swap:
3710 case Builtin::BI__sync_bool_compare_and_swap:
3711 case Builtin::BI__sync_lock_test_and_set:
3712 case Builtin::BI__sync_lock_release:
3713 case Builtin::BI__sync_swap:
3714 llvm_unreachable("Shouldn't make it through sema")::llvm::llvm_unreachable_internal("Shouldn't make it through sema"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 3714)
;
3715 case Builtin::BI__sync_fetch_and_add_1:
3716 case Builtin::BI__sync_fetch_and_add_2:
3717 case Builtin::BI__sync_fetch_and_add_4:
3718 case Builtin::BI__sync_fetch_and_add_8:
3719 case Builtin::BI__sync_fetch_and_add_16:
3720 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E);
3721 case Builtin::BI__sync_fetch_and_sub_1:
3722 case Builtin::BI__sync_fetch_and_sub_2:
3723 case Builtin::BI__sync_fetch_and_sub_4:
3724 case Builtin::BI__sync_fetch_and_sub_8:
3725 case Builtin::BI__sync_fetch_and_sub_16:
3726 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E);
3727 case Builtin::BI__sync_fetch_and_or_1:
3728 case Builtin::BI__sync_fetch_and_or_2:
3729 case Builtin::BI__sync_fetch_and_or_4:
3730 case Builtin::BI__sync_fetch_and_or_8:
3731 case Builtin::BI__sync_fetch_and_or_16:
3732 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E);
3733 case Builtin::BI__sync_fetch_and_and_1:
3734 case Builtin::BI__sync_fetch_and_and_2:
3735 case Builtin::BI__sync_fetch_and_and_4:
3736 case Builtin::BI__sync_fetch_and_and_8:
3737 case Builtin::BI__sync_fetch_and_and_16:
3738 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E);
3739 case Builtin::BI__sync_fetch_and_xor_1:
3740 case Builtin::BI__sync_fetch_and_xor_2:
3741 case Builtin::BI__sync_fetch_and_xor_4:
3742 case Builtin::BI__sync_fetch_and_xor_8:
3743 case Builtin::BI__sync_fetch_and_xor_16:
3744 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E);
3745 case Builtin::BI__sync_fetch_and_nand_1:
3746 case Builtin::BI__sync_fetch_and_nand_2:
3747 case Builtin::BI__sync_fetch_and_nand_4:
3748 case Builtin::BI__sync_fetch_and_nand_8:
3749 case Builtin::BI__sync_fetch_and_nand_16:
3750 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E);
3751
3752 // Clang extensions: not overloaded yet.
3753 case Builtin::BI__sync_fetch_and_min:
3754 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E);
3755 case Builtin::BI__sync_fetch_and_max:
3756 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E);
3757 case Builtin::BI__sync_fetch_and_umin:
3758 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E);
3759 case Builtin::BI__sync_fetch_and_umax:
3760 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E);
3761
3762 case Builtin::BI__sync_add_and_fetch_1:
3763 case Builtin::BI__sync_add_and_fetch_2:
3764 case Builtin::BI__sync_add_and_fetch_4:
3765 case Builtin::BI__sync_add_and_fetch_8:
3766 case Builtin::BI__sync_add_and_fetch_16:
3767 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E,
3768 llvm::Instruction::Add);
3769 case Builtin::BI__sync_sub_and_fetch_1:
3770 case Builtin::BI__sync_sub_and_fetch_2:
3771 case Builtin::BI__sync_sub_and_fetch_4:
3772 case Builtin::BI__sync_sub_and_fetch_8:
3773 case Builtin::BI__sync_sub_and_fetch_16:
3774 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E,
3775 llvm::Instruction::Sub);
3776 case Builtin::BI__sync_and_and_fetch_1:
3777 case Builtin::BI__sync_and_and_fetch_2:
3778 case Builtin::BI__sync_and_and_fetch_4:
3779 case Builtin::BI__sync_and_and_fetch_8:
3780 case Builtin::BI__sync_and_and_fetch_16:
3781 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E,
3782 llvm::Instruction::And);
3783 case Builtin::BI__sync_or_and_fetch_1:
3784 case Builtin::BI__sync_or_and_fetch_2:
3785 case Builtin::BI__sync_or_and_fetch_4:
3786 case Builtin::BI__sync_or_and_fetch_8:
3787 case Builtin::BI__sync_or_and_fetch_16:
3788 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E,
3789 llvm::Instruction::Or);
3790 case Builtin::BI__sync_xor_and_fetch_1:
3791 case Builtin::BI__sync_xor_and_fetch_2:
3792 case Builtin::BI__sync_xor_and_fetch_4:
3793 case Builtin::BI__sync_xor_and_fetch_8:
3794 case Builtin::BI__sync_xor_and_fetch_16:
3795 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E,
3796 llvm::Instruction::Xor);
3797 case Builtin::BI__sync_nand_and_fetch_1:
3798 case Builtin::BI__sync_nand_and_fetch_2:
3799 case Builtin::BI__sync_nand_and_fetch_4:
3800 case Builtin::BI__sync_nand_and_fetch_8:
3801 case Builtin::BI__sync_nand_and_fetch_16:
3802 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E,
3803 llvm::Instruction::And, true);
3804
3805 case Builtin::BI__sync_val_compare_and_swap_1:
3806 case Builtin::BI__sync_val_compare_and_swap_2:
3807 case Builtin::BI__sync_val_compare_and_swap_4:
3808 case Builtin::BI__sync_val_compare_and_swap_8:
3809 case Builtin::BI__sync_val_compare_and_swap_16:
3810 return RValue::get(MakeAtomicCmpXchgValue(*this, E, false));
3811
3812 case Builtin::BI__sync_bool_compare_and_swap_1:
3813 case Builtin::BI__sync_bool_compare_and_swap_2:
3814 case Builtin::BI__sync_bool_compare_and_swap_4:
3815 case Builtin::BI__sync_bool_compare_and_swap_8:
3816 case Builtin::BI__sync_bool_compare_and_swap_16:
3817 return RValue::get(MakeAtomicCmpXchgValue(*this, E, true));
3818
3819 case Builtin::BI__sync_swap_1:
3820 case Builtin::BI__sync_swap_2:
3821 case Builtin::BI__sync_swap_4:
3822 case Builtin::BI__sync_swap_8:
3823 case Builtin::BI__sync_swap_16:
3824 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E);
3825
3826 case Builtin::BI__sync_lock_test_and_set_1:
3827 case Builtin::BI__sync_lock_test_and_set_2:
3828 case Builtin::BI__sync_lock_test_and_set_4:
3829 case Builtin::BI__sync_lock_test_and_set_8:
3830 case Builtin::BI__sync_lock_test_and_set_16:
3831 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E);
3832
3833 case Builtin::BI__sync_lock_release_1:
3834 case Builtin::BI__sync_lock_release_2:
3835 case Builtin::BI__sync_lock_release_4:
3836 case Builtin::BI__sync_lock_release_8:
3837 case Builtin::BI__sync_lock_release_16: {
3838 Value *Ptr = EmitScalarExpr(E->getArg(0));
3839 QualType ElTy = E->getArg(0)->getType()->getPointeeType();
3840 CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy);
3841 llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(),
3842 StoreSize.getQuantity() * 8);
3843 Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo());
3844 llvm::StoreInst *Store =
3845 Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr,
3846 StoreSize);
3847 Store->setAtomic(llvm::AtomicOrdering::Release);
3848 return RValue::get(nullptr);
3849 }
3850
3851 case Builtin::BI__sync_synchronize: {
3852 // We assume this is supposed to correspond to a C++0x-style
3853 // sequentially-consistent fence (i.e. this is only usable for
3854 // synchronization, not device I/O or anything like that). This intrinsic
3855 // is really badly designed in the sense that in theory, there isn't
3856 // any way to safely use it... but in practice, it mostly works
3857 // to use it with non-atomic loads and stores to get acquire/release
3858 // semantics.
3859 Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent);
3860 return RValue::get(nullptr);
3861 }
3862
3863 case Builtin::BI__builtin_nontemporal_load:
3864 return RValue::get(EmitNontemporalLoad(*this, E));
3865 case Builtin::BI__builtin_nontemporal_store:
3866 return RValue::get(EmitNontemporalStore(*this, E));
3867 case Builtin::BI__c11_atomic_is_lock_free:
3868 case Builtin::BI__atomic_is_lock_free: {
3869 // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the
3870 // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since
3871 // _Atomic(T) is always properly-aligned.
3872 const char *LibCallName = "__atomic_is_lock_free";
3873 CallArgList Args;
3874 Args.add(RValue::get(EmitScalarExpr(E->getArg(0))),
3875 getContext().getSizeType());
3876 if (BuiltinID == Builtin::BI__atomic_is_lock_free)
3877 Args.add(RValue::get(EmitScalarExpr(E->getArg(1))),
3878 getContext().VoidPtrTy);
3879 else
3880 Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)),
3881 getContext().VoidPtrTy);
3882 const CGFunctionInfo &FuncInfo =
3883 CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args);
3884 llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo);
3885 llvm::FunctionCallee Func = CGM.CreateRuntimeFunction(FTy, LibCallName);
3886 return EmitCall(FuncInfo, CGCallee::forDirect(Func),
3887 ReturnValueSlot(), Args);
3888 }
3889
3890 case Builtin::BI__atomic_test_and_set: {
3891 // Look at the argument type to determine whether this is a volatile
3892 // operation. The parameter type is always volatile.
3893 QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType();
3894 bool Volatile =
3895 PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified();
3896
3897 Value *Ptr = EmitScalarExpr(E->getArg(0));
3898 unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace();
3899 Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace));
3900 Value *NewVal = Builder.getInt8(1);
3901 Value *Order = EmitScalarExpr(E->getArg(1));
3902 if (isa<llvm::ConstantInt>(Order)) {
3903 int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
3904 AtomicRMWInst *Result = nullptr;
3905 switch (ord) {
3906 case 0: // memory_order_relaxed
3907 default: // invalid order
3908 Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
3909 llvm::AtomicOrdering::Monotonic);
3910 break;
3911 case 1: // memory_order_consume
3912 case 2: // memory_order_acquire
3913 Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
3914 llvm::AtomicOrdering::Acquire);
3915 break;
3916 case 3: // memory_order_release
3917 Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
3918 llvm::AtomicOrdering::Release);
3919 break;
3920 case 4: // memory_order_acq_rel
3921
3922 Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
3923 llvm::AtomicOrdering::AcquireRelease);
3924 break;
3925 case 5: // memory_order_seq_cst
3926 Result = Builder.CreateAtomicRMW(
3927 llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
3928 llvm::AtomicOrdering::SequentiallyConsistent);
3929 break;
3930 }
3931 Result->setVolatile(Volatile);
3932 return RValue::get(Builder.CreateIsNotNull(Result, "tobool"));
3933 }
3934
3935 llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
3936
3937 llvm::BasicBlock *BBs[5] = {
3938 createBasicBlock("monotonic", CurFn),
3939 createBasicBlock("acquire", CurFn),
3940 createBasicBlock("release", CurFn),
3941 createBasicBlock("acqrel", CurFn),
3942 createBasicBlock("seqcst", CurFn)
3943 };
3944 llvm::AtomicOrdering Orders[5] = {
3945 llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire,
3946 llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease,
3947 llvm::AtomicOrdering::SequentiallyConsistent};
3948
3949 Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
3950 llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]);
3951
3952 Builder.SetInsertPoint(ContBB);
3953 PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set");
3954
3955 for (unsigned i = 0; i < 5; ++i) {
3956 Builder.SetInsertPoint(BBs[i]);
3957 AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg,
3958 Ptr, NewVal, Orders[i]);
3959 RMW->setVolatile(Volatile);
3960 Result->addIncoming(RMW, BBs[i]);
3961 Builder.CreateBr(ContBB);
3962 }
3963
3964 SI->addCase(Builder.getInt32(0), BBs[0]);
3965 SI->addCase(Builder.getInt32(1), BBs[1]);
3966 SI->addCase(Builder.getInt32(2), BBs[1]);
3967 SI->addCase(Builder.getInt32(3), BBs[2]);
3968 SI->addCase(Builder.getInt32(4), BBs[3]);
3969 SI->addCase(Builder.getInt32(5), BBs[4]);
3970
3971 Builder.SetInsertPoint(ContBB);
3972 return RValue::get(Builder.CreateIsNotNull(Result, "tobool"));
3973 }
3974
3975 case Builtin::BI__atomic_clear: {
3976 QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType();
3977 bool Volatile =
3978 PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified();
3979
3980 Address Ptr = EmitPointerWithAlignment(E->getArg(0));
3981 unsigned AddrSpace = Ptr.getPointer()->getType()->getPointerAddressSpace();
3982 Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace));
3983 Value *NewVal = Builder.getInt8(0);
3984 Value *Order = EmitScalarExpr(E->getArg(1));
3985 if (isa<llvm::ConstantInt>(Order)) {
3986 int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
3987 StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile);
3988 switch (ord) {
3989 case 0: // memory_order_relaxed
3990 default: // invalid order
3991 Store->setOrdering(llvm::AtomicOrdering::Monotonic);
3992 break;
3993 case 3: // memory_order_release
3994 Store->setOrdering(llvm::AtomicOrdering::Release);
3995 break;
3996 case 5: // memory_order_seq_cst
3997 Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent);
3998 break;
3999 }
4000 return RValue::get(nullptr);
4001 }
4002
4003 llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
4004
4005 llvm::BasicBlock *BBs[3] = {
4006 createBasicBlock("monotonic", CurFn),
4007 createBasicBlock("release", CurFn),
4008 createBasicBlock("seqcst", CurFn)
4009 };
4010 llvm::AtomicOrdering Orders[3] = {
4011 llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release,
4012 llvm::AtomicOrdering::SequentiallyConsistent};
4013
4014 Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
4015 llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]);
4016
4017 for (unsigned i = 0; i < 3; ++i) {
4018 Builder.SetInsertPoint(BBs[i]);
4019 StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile);
4020 Store->setOrdering(Orders[i]);
4021 Builder.CreateBr(ContBB);
4022 }
4023
4024 SI->addCase(Builder.getInt32(0), BBs[0]);
4025 SI->addCase(Builder.getInt32(3), BBs[1]);
4026 SI->addCase(Builder.getInt32(5), BBs[2]);
4027
4028 Builder.SetInsertPoint(ContBB);
4029 return RValue::get(nullptr);
4030 }
4031
4032 case Builtin::BI__atomic_thread_fence:
4033 case Builtin::BI__atomic_signal_fence:
4034 case Builtin::BI__c11_atomic_thread_fence:
4035 case Builtin::BI__c11_atomic_signal_fence: {
4036 llvm::SyncScope::ID SSID;
4037 if (BuiltinID == Builtin::BI__atomic_signal_fence ||
4038 BuiltinID == Builtin::BI__c11_atomic_signal_fence)
4039 SSID = llvm::SyncScope::SingleThread;
4040 else
4041 SSID = llvm::SyncScope::System;
4042 Value *Order = EmitScalarExpr(E->getArg(0));
4043 if (isa<llvm::ConstantInt>(Order)) {
4044 int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
4045 switch (ord) {
4046 case 0: // memory_order_relaxed
4047 default: // invalid order
4048 break;
4049 case 1: // memory_order_consume
4050 case 2: // memory_order_acquire
4051 Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID);
4052 break;
4053 case 3: // memory_order_release
4054 Builder.CreateFence(llvm::AtomicOrdering::Release, SSID);
4055 break;
4056 case 4: // memory_order_acq_rel
4057 Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID);
4058 break;
4059 case 5: // memory_order_seq_cst
4060 Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID);
4061 break;
4062 }
4063 return RValue::get(nullptr);
4064 }
4065
4066 llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB;
4067 AcquireBB = createBasicBlock("acquire", CurFn);
4068 ReleaseBB = createBasicBlock("release", CurFn);
4069 AcqRelBB = createBasicBlock("acqrel", CurFn);
4070 SeqCstBB = createBasicBlock("seqcst", CurFn);
4071 llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
4072
4073 Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
4074 llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB);
4075
4076 Builder.SetInsertPoint(AcquireBB);
4077 Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID);
4078 Builder.CreateBr(ContBB);
4079 SI->addCase(Builder.getInt32(1), AcquireBB);
4080 SI->addCase(Builder.getInt32(2), AcquireBB);
4081
4082 Builder.SetInsertPoint(ReleaseBB);
4083 Builder.CreateFence(llvm::AtomicOrdering::Release, SSID);
4084 Builder.CreateBr(ContBB);
4085 SI->addCase(Builder.getInt32(3), ReleaseBB);
4086
4087 Builder.SetInsertPoint(AcqRelBB);
4088 Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID);
4089 Builder.CreateBr(ContBB);
4090 SI->addCase(Builder.getInt32(4), AcqRelBB);
4091
4092 Builder.SetInsertPoint(SeqCstBB);
4093 Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID);
4094 Builder.CreateBr(ContBB);
4095 SI->addCase(Builder.getInt32(5), SeqCstBB);
4096
4097 Builder.SetInsertPoint(ContBB);
4098 return RValue::get(nullptr);
4099 }
4100
4101 case Builtin::BI__builtin_signbit:
4102 case Builtin::BI__builtin_signbitf:
4103 case Builtin::BI__builtin_signbitl: {
4104 return RValue::get(
4105 Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))),
4106 ConvertType(E->getType())));
4107 }
4108 case Builtin::BI__warn_memset_zero_len:
4109 return RValue::getIgnored();
4110 case Builtin::BI__annotation: {
4111 // Re-encode each wide string to UTF8 and make an MDString.
4112 SmallVector<Metadata *, 1> Strings;
4113 for (const Expr *Arg : E->arguments()) {
4114 const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts());
4115 assert(Str->getCharByteWidth() == 2)(static_cast <bool> (Str->getCharByteWidth() == 2) ?
void (0) : __assert_fail ("Str->getCharByteWidth() == 2",
"/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 4115, __extension__ __PRETTY_FUNCTION__))
;
4116 StringRef WideBytes = Str->getBytes();
4117 std::string StrUtf8;
4118 if (!convertUTF16ToUTF8String(
4119 makeArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) {
4120 CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument");
4121 continue;
4122 }
4123 Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8));
4124 }
4125
4126 // Build and MDTuple of MDStrings and emit the intrinsic call.
4127 llvm::Function *F =
4128 CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {});
4129 MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings);
4130 Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple));
4131 return RValue::getIgnored();
4132 }
4133 case Builtin::BI__builtin_annotation: {
4134 llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0));
4135 llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::annotation,
4136 AnnVal->getType());
4137
4138 // Get the annotation string, go through casts. Sema requires this to be a
4139 // non-wide string literal, potentially casted, so the cast<> is safe.
4140 const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts();
4141 StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString();
4142 return RValue::get(
4143 EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc(), nullptr));
4144 }
4145 case Builtin::BI__builtin_addcb:
4146 case Builtin::BI__builtin_addcs:
4147 case Builtin::BI__builtin_addc:
4148 case Builtin::BI__builtin_addcl:
4149 case Builtin::BI__builtin_addcll:
4150 case Builtin::BI__builtin_subcb:
4151 case Builtin::BI__builtin_subcs:
4152 case Builtin::BI__builtin_subc:
4153 case Builtin::BI__builtin_subcl:
4154 case Builtin::BI__builtin_subcll: {
4155
4156 // We translate all of these builtins from expressions of the form:
4157 // int x = ..., y = ..., carryin = ..., carryout, result;
4158 // result = __builtin_addc(x, y, carryin, &carryout);
4159 //
4160 // to LLVM IR of the form:
4161 //
4162 // %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y)
4163 // %tmpsum1 = extractvalue {i32, i1} %tmp1, 0
4164 // %carry1 = extractvalue {i32, i1} %tmp1, 1
4165 // %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1,
4166 // i32 %carryin)
4167 // %result = extractvalue {i32, i1} %tmp2, 0
4168 // %carry2 = extractvalue {i32, i1} %tmp2, 1
4169 // %tmp3 = or i1 %carry1, %carry2
4170 // %tmp4 = zext i1 %tmp3 to i32
4171 // store i32 %tmp4, i32* %carryout
4172
4173 // Scalarize our inputs.
4174 llvm::Value *X = EmitScalarExpr(E->getArg(0));
4175 llvm::Value *Y = EmitScalarExpr(E->getArg(1));
4176 llvm::Value *Carryin = EmitScalarExpr(E->getArg(2));
4177 Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3));
4178
4179 // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow.
4180 llvm::Intrinsic::ID IntrinsicId;
4181 switch (BuiltinID) {
4182 default: llvm_unreachable("Unknown multiprecision builtin id.")::llvm::llvm_unreachable_internal("Unknown multiprecision builtin id."
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 4182)
;
4183 case Builtin::BI__builtin_addcb:
4184 case Builtin::BI__builtin_addcs:
4185 case Builtin::BI__builtin_addc:
4186 case Builtin::BI__builtin_addcl:
4187 case Builtin::BI__builtin_addcll:
4188 IntrinsicId = llvm::Intrinsic::uadd_with_overflow;
4189 break;
4190 case Builtin::BI__builtin_subcb:
4191 case Builtin::BI__builtin_subcs:
4192 case Builtin::BI__builtin_subc:
4193 case Builtin::BI__builtin_subcl:
4194 case Builtin::BI__builtin_subcll:
4195 IntrinsicId = llvm::Intrinsic::usub_with_overflow;
4196 break;
4197 }
4198
4199 // Construct our resulting LLVM IR expression.
4200 llvm::Value *Carry1;
4201 llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId,
4202 X, Y, Carry1);
4203 llvm::Value *Carry2;
4204 llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId,
4205 Sum1, Carryin, Carry2);
4206 llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2),
4207 X->getType());
4208 Builder.CreateStore(CarryOut, CarryOutPtr);
4209 return RValue::get(Sum2);
4210 }
4211
4212 case Builtin::BI__builtin_add_overflow:
4213 case Builtin::BI__builtin_sub_overflow:
4214 case Builtin::BI__builtin_mul_overflow: {
4215 const clang::Expr *LeftArg = E->getArg(0);
4216 const clang::Expr *RightArg = E->getArg(1);
4217 const clang::Expr *ResultArg = E->getArg(2);
4218
4219 clang::QualType ResultQTy =
4220 ResultArg->getType()->castAs<PointerType>()->getPointeeType();
4221
4222 WidthAndSignedness LeftInfo =
4223 getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType());
4224 WidthAndSignedness RightInfo =
4225 getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType());
4226 WidthAndSignedness ResultInfo =
4227 getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy);
4228
4229 // Handle mixed-sign multiplication as a special case, because adding
4230 // runtime or backend support for our generic irgen would be too expensive.
4231 if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo))
4232 return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg,
4233 RightInfo, ResultArg, ResultQTy,
4234 ResultInfo);
4235
4236 if (isSpecialUnsignedMultiplySignedResult(BuiltinID, LeftInfo, RightInfo,
4237 ResultInfo))
4238 return EmitCheckedUnsignedMultiplySignedResult(
4239 *this, LeftArg, LeftInfo, RightArg, RightInfo, ResultArg, ResultQTy,
4240 ResultInfo);
4241
4242 WidthAndSignedness EncompassingInfo =
4243 EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo});
4244
4245 llvm::Type *EncompassingLLVMTy =
4246 llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width);
4247
4248 llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy);
4249
4250 llvm::Intrinsic::ID IntrinsicId;
4251 switch (BuiltinID) {
4252 default:
4253 llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id."
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 4253)
;
4254 case Builtin::BI__builtin_add_overflow:
4255 IntrinsicId = EncompassingInfo.Signed
4256 ? llvm::Intrinsic::sadd_with_overflow
4257 : llvm::Intrinsic::uadd_with_overflow;
4258 break;
4259 case Builtin::BI__builtin_sub_overflow:
4260 IntrinsicId = EncompassingInfo.Signed
4261 ? llvm::Intrinsic::ssub_with_overflow
4262 : llvm::Intrinsic::usub_with_overflow;
4263 break;
4264 case Builtin::BI__builtin_mul_overflow:
4265 IntrinsicId = EncompassingInfo.Signed
4266 ? llvm::Intrinsic::smul_with_overflow
4267 : llvm::Intrinsic::umul_with_overflow;
4268 break;
4269 }
4270
4271 llvm::Value *Left = EmitScalarExpr(LeftArg);
4272 llvm::Value *Right = EmitScalarExpr(RightArg);
4273 Address ResultPtr = EmitPointerWithAlignment(ResultArg);
4274
4275 // Extend each operand to the encompassing type.
4276 Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed);
4277 Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed);
4278
4279 // Perform the operation on the extended values.
4280 llvm::Value *Overflow, *Result;
4281 Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow);
4282
4283 if (EncompassingInfo.Width > ResultInfo.Width) {
4284 // The encompassing type is wider than the result type, so we need to
4285 // truncate it.
4286 llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy);
4287
4288 // To see if the truncation caused an overflow, we will extend
4289 // the result and then compare it to the original result.
4290 llvm::Value *ResultTruncExt = Builder.CreateIntCast(
4291 ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed);
4292 llvm::Value *TruncationOverflow =
4293 Builder.CreateICmpNE(Result, ResultTruncExt);
4294
4295 Overflow = Builder.CreateOr(Overflow, TruncationOverflow);
4296 Result = ResultTrunc;
4297 }
4298
4299 // Finally, store the result using the pointer.
4300 bool isVolatile =
4301 ResultArg->getType()->getPointeeType().isVolatileQualified();
4302 Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile);
4303
4304 return RValue::get(Overflow);
4305 }
4306
4307 case Builtin::BI__builtin_uadd_overflow:
4308 case Builtin::BI__builtin_uaddl_overflow:
4309 case Builtin::BI__builtin_uaddll_overflow:
4310 case Builtin::BI__builtin_usub_overflow:
4311 case Builtin::BI__builtin_usubl_overflow:
4312 case Builtin::BI__builtin_usubll_overflow:
4313 case Builtin::BI__builtin_umul_overflow:
4314 case Builtin::BI__builtin_umull_overflow:
4315 case Builtin::BI__builtin_umulll_overflow:
4316 case Builtin::BI__builtin_sadd_overflow:
4317 case Builtin::BI__builtin_saddl_overflow:
4318 case Builtin::BI__builtin_saddll_overflow:
4319 case Builtin::BI__builtin_ssub_overflow:
4320 case Builtin::BI__builtin_ssubl_overflow:
4321 case Builtin::BI__builtin_ssubll_overflow:
4322 case Builtin::BI__builtin_smul_overflow:
4323 case Builtin::BI__builtin_smull_overflow:
4324 case Builtin::BI__builtin_smulll_overflow: {
4325
4326 // We translate all of these builtins directly to the relevant llvm IR node.
4327
4328 // Scalarize our inputs.
4329 llvm::Value *X = EmitScalarExpr(E->getArg(0));
4330 llvm::Value *Y = EmitScalarExpr(E->getArg(1));
4331 Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2));
4332
4333 // Decide which of the overflow intrinsics we are lowering to:
4334 llvm::Intrinsic::ID IntrinsicId;
4335 switch (BuiltinID) {
4336 default: llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id."
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 4336)
;
4337 case Builtin::BI__builtin_uadd_overflow:
4338 case Builtin::BI__builtin_uaddl_overflow:
4339 case Builtin::BI__builtin_uaddll_overflow:
4340 IntrinsicId = llvm::Intrinsic::uadd_with_overflow;
4341 break;
4342 case Builtin::BI__builtin_usub_overflow:
4343 case Builtin::BI__builtin_usubl_overflow:
4344 case Builtin::BI__builtin_usubll_overflow:
4345 IntrinsicId = llvm::Intrinsic::usub_with_overflow;
4346 break;
4347 case Builtin::BI__builtin_umul_overflow:
4348 case Builtin::BI__builtin_umull_overflow:
4349 case Builtin::BI__builtin_umulll_overflow:
4350 IntrinsicId = llvm::Intrinsic::umul_with_overflow;
4351 break;
4352 case Builtin::BI__builtin_sadd_overflow:
4353 case Builtin::BI__builtin_saddl_overflow:
4354 case Builtin::BI__builtin_saddll_overflow:
4355 IntrinsicId = llvm::Intrinsic::sadd_with_overflow;
4356 break;
4357 case Builtin::BI__builtin_ssub_overflow:
4358 case Builtin::BI__builtin_ssubl_overflow:
4359 case Builtin::BI__builtin_ssubll_overflow:
4360 IntrinsicId = llvm::Intrinsic::ssub_with_overflow;
4361 break;
4362 case Builtin::BI__builtin_smul_overflow:
4363 case Builtin::BI__builtin_smull_overflow:
4364 case Builtin::BI__builtin_smulll_overflow:
4365 IntrinsicId = llvm::Intrinsic::smul_with_overflow;
4366 break;
4367 }
4368
4369
4370 llvm::Value *Carry;
4371 llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry);
4372 Builder.CreateStore(Sum, SumOutPtr);
4373
4374 return RValue::get(Carry);
4375 }
4376 case Builtin::BI__builtin_addressof:
4377 return RValue::get(EmitLValue(E->getArg(0)).getPointer(*this));
4378 case Builtin::BI__builtin_operator_new:
4379 return EmitBuiltinNewDeleteCall(
4380 E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false);
4381 case Builtin::BI__builtin_operator_delete:
4382 return EmitBuiltinNewDeleteCall(
4383 E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true);
4384
4385 case Builtin::BI__builtin_is_aligned:
4386 return EmitBuiltinIsAligned(E);
4387 case Builtin::BI__builtin_align_up:
4388 return EmitBuiltinAlignTo(E, true);
4389 case Builtin::BI__builtin_align_down:
4390 return EmitBuiltinAlignTo(E, false);
4391
4392 case Builtin::BI__noop:
4393 // __noop always evaluates to an integer literal zero.
4394 return RValue::get(ConstantInt::get(IntTy, 0));
4395 case Builtin::BI__builtin_call_with_static_chain: {
4396 const CallExpr *Call = cast<CallExpr>(E->getArg(0));
4397 const Expr *Chain = E->getArg(1);
4398 return EmitCall(Call->getCallee()->getType(),
4399 EmitCallee(Call->getCallee()), Call, ReturnValue,
4400 EmitScalarExpr(Chain));
4401 }
4402 case Builtin::BI_InterlockedExchange8:
4403 case Builtin::BI_InterlockedExchange16:
4404 case Builtin::BI_InterlockedExchange:
4405 case Builtin::BI_InterlockedExchangePointer:
4406 return RValue::get(
4407 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E));
4408 case Builtin::BI_InterlockedCompareExchangePointer:
4409 case Builtin::BI_InterlockedCompareExchangePointer_nf: {
4410 llvm::Type *RTy;
4411 llvm::IntegerType *IntType =
4412 IntegerType::get(getLLVMContext(),
4413 getContext().getTypeSize(E->getType()));
4414 llvm::Type *IntPtrType = IntType->getPointerTo();
4415
4416 llvm::Value *Destination =
4417 Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType);
4418
4419 llvm::Value *Exchange = EmitScalarExpr(E->getArg(1));
4420 RTy = Exchange->getType();
4421 Exchange = Builder.CreatePtrToInt(Exchange, IntType);
4422
4423 llvm::Value *Comparand =
4424 Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType);
4425
4426 auto Ordering =
4427 BuiltinID == Builtin::BI_InterlockedCompareExchangePointer_nf ?
4428 AtomicOrdering::Monotonic : AtomicOrdering::SequentiallyConsistent;
4429
4430 auto Result = Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange,
4431 Ordering, Ordering);
4432 Result->setVolatile(true);
4433
4434 return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result,
4435 0),
4436 RTy));
4437 }
4438 case Builtin::BI_InterlockedCompareExchange8:
4439 case Builtin::BI_InterlockedCompareExchange16:
4440 case Builtin::BI_InterlockedCompareExchange:
4441 case Builtin::BI_InterlockedCompareExchange64:
4442 return RValue::get(EmitAtomicCmpXchgForMSIntrin(*this, E));
4443 case Builtin::BI_InterlockedIncrement16:
4444 case Builtin::BI_InterlockedIncrement:
4445 return RValue::get(
4446 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E));
4447 case Builtin::BI_InterlockedDecrement16:
4448 case Builtin::BI_InterlockedDecrement:
4449 return RValue::get(
4450 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E));
4451 case Builtin::BI_InterlockedAnd8:
4452 case Builtin::BI_InterlockedAnd16:
4453 case Builtin::BI_InterlockedAnd:
4454 return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E));
4455 case Builtin::BI_InterlockedExchangeAdd8:
4456 case Builtin::BI_InterlockedExchangeAdd16:
4457 case Builtin::BI_InterlockedExchangeAdd:
4458 return RValue::get(
4459 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E));
4460 case Builtin::BI_InterlockedExchangeSub8:
4461 case Builtin::BI_InterlockedExchangeSub16:
4462 case Builtin::BI_InterlockedExchangeSub:
4463 return RValue::get(
4464 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E));
4465 case Builtin::BI_InterlockedOr8:
4466 case Builtin::BI_InterlockedOr16:
4467 case Builtin::BI_InterlockedOr:
4468 return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E));
4469 case Builtin::BI_InterlockedXor8:
4470 case Builtin::BI_InterlockedXor16:
4471 case Builtin::BI_InterlockedXor:
4472 return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E));
4473
4474 case Builtin::BI_bittest64:
4475 case Builtin::BI_bittest:
4476 case Builtin::BI_bittestandcomplement64:
4477 case Builtin::BI_bittestandcomplement:
4478 case Builtin::BI_bittestandreset64:
4479 case Builtin::BI_bittestandreset:
4480 case Builtin::BI_bittestandset64:
4481 case Builtin::BI_bittestandset:
4482 case Builtin::BI_interlockedbittestandreset:
4483 case Builtin::BI_interlockedbittestandreset64:
4484 case Builtin::BI_interlockedbittestandset64:
4485 case Builtin::BI_interlockedbittestandset:
4486 case Builtin::BI_interlockedbittestandset_acq:
4487 case Builtin::BI_interlockedbittestandset_rel:
4488 case Builtin::BI_interlockedbittestandset_nf:
4489 case Builtin::BI_interlockedbittestandreset_acq:
4490 case Builtin::BI_interlockedbittestandreset_rel:
4491 case Builtin::BI_interlockedbittestandreset_nf:
4492 return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E));
4493
4494 // These builtins exist to emit regular volatile loads and stores not
4495 // affected by the -fms-volatile setting.
4496 case Builtin::BI__iso_volatile_load8:
4497 case Builtin::BI__iso_volatile_load16:
4498 case Builtin::BI__iso_volatile_load32:
4499 case Builtin::BI__iso_volatile_load64:
4500 return RValue::get(EmitISOVolatileLoad(*this, E));
4501 case Builtin::BI__iso_volatile_store8:
4502 case Builtin::BI__iso_volatile_store16:
4503 case Builtin::BI__iso_volatile_store32:
4504 case Builtin::BI__iso_volatile_store64:
4505 return RValue::get(EmitISOVolatileStore(*this, E));
4506
4507 case Builtin::BI__exception_code:
4508 case Builtin::BI_exception_code:
4509 return RValue::get(EmitSEHExceptionCode());
4510 case Builtin::BI__exception_info:
4511 case Builtin::BI_exception_info:
4512 return RValue::get(EmitSEHExceptionInfo());
4513 case Builtin::BI__abnormal_termination:
4514 case Builtin::BI_abnormal_termination:
4515 return RValue::get(EmitSEHAbnormalTermination());
4516 case Builtin::BI_setjmpex:
4517 if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 &&
4518 E->getArg(0)->getType()->isPointerType())
4519 return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E);
4520 break;
4521 case Builtin::BI_setjmp:
4522 if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 &&
4523 E->getArg(0)->getType()->isPointerType()) {
4524 if (getTarget().getTriple().getArch() == llvm::Triple::x86)
4525 return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E);
4526 else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64)
4527 return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E);
4528 return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E);
4529 }
4530 break;
4531
4532 case Builtin::BI__GetExceptionInfo: {
4533 if (llvm::GlobalVariable *GV =
4534 CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType()))
4535 return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy));
4536 break;
4537 }
4538
4539 case Builtin::BI__fastfail:
4540 return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E));
4541
4542 case Builtin::BI__builtin_coro_size: {
4543 auto & Context = getContext();
4544 auto SizeTy = Context.getSizeType();
4545 auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy));
4546 Function *F = CGM.getIntrinsic(Intrinsic::coro_size, T);
4547 return RValue::get(Builder.CreateCall(F));
4548 }
4549
4550 case Builtin::BI__builtin_coro_id:
4551 return EmitCoroutineIntrinsic(E, Intrinsic::coro_id);
4552 case Builtin::BI__builtin_coro_promise:
4553 return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise);
4554 case Builtin::BI__builtin_coro_resume:
4555 return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume);
4556 case Builtin::BI__builtin_coro_frame:
4557 return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame);
4558 case Builtin::BI__builtin_coro_noop:
4559 return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop);
4560 case Builtin::BI__builtin_coro_free:
4561 return EmitCoroutineIntrinsic(E, Intrinsic::coro_free);
4562 case Builtin::BI__builtin_coro_destroy:
4563 return EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy);
4564 case Builtin::BI__builtin_coro_done:
4565 return EmitCoroutineIntrinsic(E, Intrinsic::coro_done);
4566 case Builtin::BI__builtin_coro_alloc:
4567 return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc);
4568 case Builtin::BI__builtin_coro_begin:
4569 return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin);
4570 case Builtin::BI__builtin_coro_end:
4571 return EmitCoroutineIntrinsic(E, Intrinsic::coro_end);
4572 case Builtin::BI__builtin_coro_suspend:
4573 return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend);
4574 case Builtin::BI__builtin_coro_param:
4575 return EmitCoroutineIntrinsic(E, Intrinsic::coro_param);
4576
4577 // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions
4578 case Builtin::BIread_pipe:
4579 case Builtin::BIwrite_pipe: {
4580 Value *Arg0 = EmitScalarExpr(E->getArg(0)),
4581 *Arg1 = EmitScalarExpr(E->getArg(1));
4582 CGOpenCLRuntime OpenCLRT(CGM);
4583 Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
4584 Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
4585
4586 // Type of the generic packet parameter.
4587 unsigned GenericAS =
4588 getContext().getTargetAddressSpace(LangAS::opencl_generic);
4589 llvm::Type *I8PTy = llvm::PointerType::get(
4590 llvm::Type::getInt8Ty(getLLVMContext()), GenericAS);
4591
4592 // Testing which overloaded version we should generate the call for.
4593 if (2U == E->getNumArgs()) {
4594 const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2"
4595 : "__write_pipe_2";
4596 // Creating a generic function type to be able to call with any builtin or
4597 // user defined type.
4598 llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty};
4599 llvm::FunctionType *FTy = llvm::FunctionType::get(
4600 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
4601 Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy);
4602 return RValue::get(
4603 EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name),
4604 {Arg0, BCast, PacketSize, PacketAlign}));
4605 } else {
4606 assert(4 == E->getNumArgs() &&(static_cast <bool> (4 == E->getNumArgs() &&
"Illegal number of parameters to pipe function") ? void (0) :
__assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 4607, __extension__ __PRETTY_FUNCTION__))
4607 "Illegal number of parameters to pipe function")(static_cast <bool> (4 == E->getNumArgs() &&
"Illegal number of parameters to pipe function") ? void (0) :
__assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 4607, __extension__ __PRETTY_FUNCTION__))
;
4608 const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4"
4609 : "__write_pipe_4";
4610
4611 llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy,
4612 Int32Ty, Int32Ty};
4613 Value *Arg2 = EmitScalarExpr(E->getArg(2)),
4614 *Arg3 = EmitScalarExpr(E->getArg(3));
4615 llvm::FunctionType *FTy = llvm::FunctionType::get(
4616 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
4617 Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy);
4618 // We know the third argument is an integer type, but we may need to cast
4619 // it to i32.
4620 if (Arg2->getType() != Int32Ty)
4621 Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty);
4622 return RValue::get(
4623 EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name),
4624 {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign}));
4625 }
4626 }
4627 // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write
4628 // functions
4629 case Builtin::BIreserve_read_pipe:
4630 case Builtin::BIreserve_write_pipe:
4631 case Builtin::BIwork_group_reserve_read_pipe:
4632 case Builtin::BIwork_group_reserve_write_pipe:
4633 case Builtin::BIsub_group_reserve_read_pipe:
4634 case Builtin::BIsub_group_reserve_write_pipe: {
4635 // Composing the mangled name for the function.
4636 const char *Name;
4637 if (BuiltinID == Builtin::BIreserve_read_pipe)
4638 Name = "__reserve_read_pipe";
4639 else if (BuiltinID == Builtin::BIreserve_write_pipe)
4640 Name = "__reserve_write_pipe";
4641 else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe)
4642 Name = "__work_group_reserve_read_pipe";
4643 else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe)
4644 Name = "__work_group_reserve_write_pipe";
4645 else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe)
4646 Name = "__sub_group_reserve_read_pipe";
4647 else
4648 Name = "__sub_group_reserve_write_pipe";
4649
4650 Value *Arg0 = EmitScalarExpr(E->getArg(0)),
4651 *Arg1 = EmitScalarExpr(E->getArg(1));
4652 llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy);
4653 CGOpenCLRuntime OpenCLRT(CGM);
4654 Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
4655 Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
4656
4657 // Building the generic function prototype.
4658 llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty};
4659 llvm::FunctionType *FTy = llvm::FunctionType::get(
4660 ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
4661 // We know the second argument is an integer type, but we may need to cast
4662 // it to i32.
4663 if (Arg1->getType() != Int32Ty)
4664 Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty);
4665 return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name),
4666 {Arg0, Arg1, PacketSize, PacketAlign}));
4667 }
4668 // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write
4669 // functions
4670 case Builtin::BIcommit_read_pipe:
4671 case Builtin::BIcommit_write_pipe:
4672 case Builtin::BIwork_group_commit_read_pipe:
4673 case Builtin::BIwork_group_commit_write_pipe:
4674 case Builtin::BIsub_group_commit_read_pipe:
4675 case Builtin::BIsub_group_commit_write_pipe: {
4676 const char *Name;
4677 if (BuiltinID == Builtin::BIcommit_read_pipe)
4678 Name = "__commit_read_pipe";
4679 else if (BuiltinID == Builtin::BIcommit_write_pipe)
4680 Name = "__commit_write_pipe";
4681 else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe)
4682 Name = "__work_group_commit_read_pipe";
4683 else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe)
4684 Name = "__work_group_commit_write_pipe";
4685 else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe)
4686 Name = "__sub_group_commit_read_pipe";
4687 else
4688 Name = "__sub_group_commit_write_pipe";
4689
4690 Value *Arg0 = EmitScalarExpr(E->getArg(0)),
4691 *Arg1 = EmitScalarExpr(E->getArg(1));
4692 CGOpenCLRuntime OpenCLRT(CGM);
4693 Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
4694 Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
4695
4696 // Building the generic function prototype.
4697 llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty};
4698 llvm::FunctionType *FTy =
4699 llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()),
4700 llvm::ArrayRef<llvm::Type *>(ArgTys), false);
4701
4702 return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name),
4703 {Arg0, Arg1, PacketSize, PacketAlign}));
4704 }
4705 // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions
4706 case Builtin::BIget_pipe_num_packets:
4707 case Builtin::BIget_pipe_max_packets: {
4708 const char *BaseName;
4709 const auto *PipeTy = E->getArg(0)->getType()->castAs<PipeType>();
4710 if (BuiltinID == Builtin::BIget_pipe_num_packets)
4711 BaseName = "__get_pipe_num_packets";
4712 else
4713 BaseName = "__get_pipe_max_packets";
4714 std::string Name = std::string(BaseName) +
4715 std::string(PipeTy->isReadOnly() ? "_ro" : "_wo");
4716
4717 // Building the generic function prototype.
4718 Value *Arg0 = EmitScalarExpr(E->getArg(0));
4719 CGOpenCLRuntime OpenCLRT(CGM);
4720 Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
4721 Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
4722 llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty};
4723 llvm::FunctionType *FTy = llvm::FunctionType::get(
4724 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
4725
4726 return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name),
4727 {Arg0, PacketSize, PacketAlign}));
4728 }
4729
4730 // OpenCL v2.0 s6.13.9 - Address space qualifier functions.
4731 case Builtin::BIto_global:
4732 case Builtin::BIto_local:
4733 case Builtin::BIto_private: {
4734 auto Arg0 = EmitScalarExpr(E->getArg(0));
4735 auto NewArgT = llvm::PointerType::get(Int8Ty,
4736 CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic));
4737 auto NewRetT = llvm::PointerType::get(Int8Ty,
4738 CGM.getContext().getTargetAddressSpace(
4739 E->getType()->getPointeeType().getAddressSpace()));
4740 auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false);
4741 llvm::Value *NewArg;
4742 if (Arg0->getType()->getPointerAddressSpace() !=
4743 NewArgT->getPointerAddressSpace())
4744 NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT);
4745 else
4746 NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT);
4747 auto NewName = std::string("__") + E->getDirectCallee()->getName().str();
4748 auto NewCall =
4749 EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg});
4750 return RValue::get(Builder.CreateBitOrPointerCast(NewCall,
4751 ConvertType(E->getType())));
4752 }
4753
4754 // OpenCL v2.0, s6.13.17 - Enqueue kernel function.
4755 // It contains four different overload formats specified in Table 6.13.17.1.
4756 case Builtin::BIenqueue_kernel: {
4757 StringRef Name; // Generated function call name
4758 unsigned NumArgs = E->getNumArgs();
4759
4760 llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy);
4761 llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
4762 getContext().getTargetAddressSpace(LangAS::opencl_generic));
4763
4764 llvm::Value *Queue = EmitScalarExpr(E->getArg(0));
4765 llvm::Value *Flags = EmitScalarExpr(E->getArg(1));
4766 LValue NDRangeL = EmitAggExprToLValue(E->getArg(2));
4767 llvm::Value *Range = NDRangeL.getAddress(*this).getPointer();
4768 llvm::Type *RangeTy = NDRangeL.getAddress(*this).getType();
4769
4770 if (NumArgs == 4) {
4771 // The most basic form of the call with parameters:
4772 // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void)
4773 Name = "__enqueue_kernel_basic";
4774 llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy,
4775 GenericVoidPtrTy};
4776 llvm::FunctionType *FTy = llvm::FunctionType::get(
4777 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
4778
4779 auto Info =
4780 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3));
4781 llvm::Value *Kernel =
4782 Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
4783 llvm::Value *Block =
4784 Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
4785
4786 AttrBuilder B;
4787 B.addByValAttr(NDRangeL.getAddress(*this).getElementType());
4788 llvm::AttributeList ByValAttrSet =
4789 llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B);
4790
4791 auto RTCall =
4792 EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet),
4793 {Queue, Flags, Range, Kernel, Block});
4794 RTCall->setAttributes(ByValAttrSet);
4795 return RValue::get(RTCall);
4796 }
4797 assert(NumArgs >= 5 && "Invalid enqueue_kernel signature")(static_cast <bool> (NumArgs >= 5 && "Invalid enqueue_kernel signature"
) ? void (0) : __assert_fail ("NumArgs >= 5 && \"Invalid enqueue_kernel signature\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 4797, __extension__ __PRETTY_FUNCTION__))
;
4798
4799 // Create a temporary array to hold the sizes of local pointer arguments
4800 // for the block. \p First is the position of the first size argument.
4801 auto CreateArrayForSizeVar = [=](unsigned First)
4802 -> std::tuple<llvm::Value *, llvm::Value *, llvm::Value *> {
4803 llvm::APInt ArraySize(32, NumArgs - First);
4804 QualType SizeArrayTy = getContext().getConstantArrayType(
4805 getContext().getSizeType(), ArraySize, nullptr, ArrayType::Normal,
4806 /*IndexTypeQuals=*/0);
4807 auto Tmp = CreateMemTemp(SizeArrayTy, "block_sizes");
4808 llvm::Value *TmpPtr = Tmp.getPointer();
4809 llvm::Value *TmpSize = EmitLifetimeStart(
4810 CGM.getDataLayout().getTypeAllocSize(Tmp.getElementType()), TmpPtr);
4811 llvm::Value *ElemPtr;
4812 // Each of the following arguments specifies the size of the corresponding
4813 // argument passed to the enqueued block.
4814 auto *Zero = llvm::ConstantInt::get(IntTy, 0);
4815 for (unsigned I = First; I < NumArgs; ++I) {
4816 auto *Index = llvm::ConstantInt::get(IntTy, I - First);
4817 auto *GEP = Builder.CreateGEP(Tmp.getElementType(), TmpPtr,
4818 {Zero, Index});
4819 if (I == First)
4820 ElemPtr = GEP;
4821 auto *V =
4822 Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy);
4823 Builder.CreateAlignedStore(
4824 V, GEP, CGM.getDataLayout().getPrefTypeAlign(SizeTy));
4825 }
4826 return std::tie(ElemPtr, TmpSize, TmpPtr);
4827 };
4828
4829 // Could have events and/or varargs.
4830 if (E->getArg(3)->getType()->isBlockPointerType()) {
4831 // No events passed, but has variadic arguments.
4832 Name = "__enqueue_kernel_varargs";
4833 auto Info =
4834 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3));
4835 llvm::Value *Kernel =
4836 Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
4837 auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
4838 llvm::Value *ElemPtr, *TmpSize, *TmpPtr;
4839 std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(4);
4840
4841 // Create a vector of the arguments, as well as a constant value to
4842 // express to the runtime the number of variadic arguments.
4843 llvm::Value *const Args[] = {Queue, Flags,
4844 Range, Kernel,
4845 Block, ConstantInt::get(IntTy, NumArgs - 4),
4846 ElemPtr};
4847 llvm::Type *const ArgTys[] = {
4848 QueueTy, IntTy, RangeTy, GenericVoidPtrTy,
4849 GenericVoidPtrTy, IntTy, ElemPtr->getType()};
4850
4851 llvm::FunctionType *FTy = llvm::FunctionType::get(Int32Ty, ArgTys, false);
4852 auto Call = RValue::get(
4853 EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Args));
4854 if (TmpSize)
4855 EmitLifetimeEnd(TmpSize, TmpPtr);
4856 return Call;
4857 }
4858 // Any calls now have event arguments passed.
4859 if (NumArgs >= 7) {
4860 llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy);
4861 llvm::PointerType *EventPtrTy = EventTy->getPointerTo(
4862 CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic));
4863
4864 llvm::Value *NumEvents =
4865 Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty);
4866
4867 // Since SemaOpenCLBuiltinEnqueueKernel allows fifth and sixth arguments
4868 // to be a null pointer constant (including `0` literal), we can take it
4869 // into account and emit null pointer directly.
4870 llvm::Value *EventWaitList = nullptr;
4871 if (E->getArg(4)->isNullPointerConstant(
4872 getContext(), Expr::NPC_ValueDependentIsNotNull)) {
4873 EventWaitList = llvm::ConstantPointerNull::get(EventPtrTy);
4874 } else {
4875 EventWaitList = E->getArg(4)->getType()->isArrayType()
4876 ? EmitArrayToPointerDecay(E->getArg(4)).getPointer()
4877 : EmitScalarExpr(E->getArg(4));
4878 // Convert to generic address space.
4879 EventWaitList = Builder.CreatePointerCast(EventWaitList, EventPtrTy);
4880 }
4881 llvm::Value *EventRet = nullptr;
4882 if (E->getArg(5)->isNullPointerConstant(
4883 getContext(), Expr::NPC_ValueDependentIsNotNull)) {
4884 EventRet = llvm::ConstantPointerNull::get(EventPtrTy);
4885 } else {
4886 EventRet =
4887 Builder.CreatePointerCast(EmitScalarExpr(E->getArg(5)), EventPtrTy);
4888 }
4889
4890 auto Info =
4891 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6));
4892 llvm::Value *Kernel =
4893 Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
4894 llvm::Value *Block =
4895 Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
4896
4897 std::vector<llvm::Type *> ArgTys = {
4898 QueueTy, Int32Ty, RangeTy, Int32Ty,
4899 EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy};
4900
4901 std::vector<llvm::Value *> Args = {Queue, Flags, Range,
4902 NumEvents, EventWaitList, EventRet,
4903 Kernel, Block};
4904
4905 if (NumArgs == 7) {
4906 // Has events but no variadics.
4907 Name = "__enqueue_kernel_basic_events";
4908 llvm::FunctionType *FTy = llvm::FunctionType::get(
4909 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
4910 return RValue::get(
4911 EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name),
4912 llvm::ArrayRef<llvm::Value *>(Args)));
4913 }
4914 // Has event info and variadics
4915 // Pass the number of variadics to the runtime function too.
4916 Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7));
4917 ArgTys.push_back(Int32Ty);
4918 Name = "__enqueue_kernel_events_varargs";
4919
4920 llvm::Value *ElemPtr, *TmpSize, *TmpPtr;
4921 std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(7);
4922 Args.push_back(ElemPtr);
4923 ArgTys.push_back(ElemPtr->getType());
4924
4925 llvm::FunctionType *FTy = llvm::FunctionType::get(
4926 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
4927 auto Call =
4928 RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name),
4929 llvm::ArrayRef<llvm::Value *>(Args)));
4930 if (TmpSize)
4931 EmitLifetimeEnd(TmpSize, TmpPtr);
4932 return Call;
4933 }
4934 LLVM_FALLTHROUGH[[gnu::fallthrough]];
4935 }
4936 // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block
4937 // parameter.
4938 case Builtin::BIget_kernel_work_group_size: {
4939 llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
4940 getContext().getTargetAddressSpace(LangAS::opencl_generic));
4941 auto Info =
4942 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0));
4943 Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
4944 Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
4945 return RValue::get(EmitRuntimeCall(
4946 CGM.CreateRuntimeFunction(
4947 llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy},
4948 false),
4949 "__get_kernel_work_group_size_impl"),
4950 {Kernel, Arg}));
4951 }
4952 case Builtin::BIget_kernel_preferred_work_group_size_multiple: {
4953 llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
4954 getContext().getTargetAddressSpace(LangAS::opencl_generic));
4955 auto Info =
4956 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0));
4957 Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
4958 Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
4959 return RValue::get(EmitRuntimeCall(
4960 CGM.CreateRuntimeFunction(
4961 llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy},
4962 false),
4963 "__get_kernel_preferred_work_group_size_multiple_impl"),
4964 {Kernel, Arg}));
4965 }
4966 case Builtin::BIget_kernel_max_sub_group_size_for_ndrange:
4967 case Builtin::BIget_kernel_sub_group_count_for_ndrange: {
4968 llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
4969 getContext().getTargetAddressSpace(LangAS::opencl_generic));
4970 LValue NDRangeL = EmitAggExprToLValue(E->getArg(0));
4971 llvm::Value *NDRange = NDRangeL.getAddress(*this).getPointer();
4972 auto Info =
4973 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1));
4974 Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
4975 Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
4976 const char *Name =
4977 BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange
4978 ? "__get_kernel_max_sub_group_size_for_ndrange_impl"
4979 : "__get_kernel_sub_group_count_for_ndrange_impl";
4980 return RValue::get(EmitRuntimeCall(
4981 CGM.CreateRuntimeFunction(
4982 llvm::FunctionType::get(
4983 IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy},
4984 false),
4985 Name),
4986 {NDRange, Kernel, Block}));
4987 }
4988
4989 case Builtin::BI__builtin_store_half:
4990 case Builtin::BI__builtin_store_halff: {
4991 Value *Val = EmitScalarExpr(E->getArg(0));
4992 Address Address = EmitPointerWithAlignment(E->getArg(1));
4993 Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy());
4994 return RValue::get(Builder.CreateStore(HalfVal, Address));
4995 }
4996 case Builtin::BI__builtin_load_half: {
4997 Address Address = EmitPointerWithAlignment(E->getArg(0));
4998 Value *HalfVal = Builder.CreateLoad(Address);
4999 return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy()));
5000 }
5001 case Builtin::BI__builtin_load_halff: {
5002 Address Address = EmitPointerWithAlignment(E->getArg(0));
5003 Value *HalfVal = Builder.CreateLoad(Address);
5004 return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy()));
5005 }
5006 case Builtin::BIprintf:
5007 if (getTarget().getTriple().isNVPTX())
5008 return EmitNVPTXDevicePrintfCallExpr(E, ReturnValue);
5009 if (getTarget().getTriple().getArch() == Triple::amdgcn &&
5010 getLangOpts().HIP)
5011 return EmitAMDGPUDevicePrintfCallExpr(E, ReturnValue);
5012 break;
5013 case Builtin::BI__builtin_canonicalize:
5014 case Builtin::BI__builtin_canonicalizef:
5015 case Builtin::BI__builtin_canonicalizef16:
5016 case Builtin::BI__builtin_canonicalizel:
5017 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize));
5018
5019 case Builtin::BI__builtin_thread_pointer: {
5020 if (!getContext().getTargetInfo().isTLSSupported())
5021 CGM.ErrorUnsupported(E, "__builtin_thread_pointer");
5022 // Fall through - it's already mapped to the intrinsic by GCCBuiltin.
5023 break;
5024 }
5025 case Builtin::BI__builtin_os_log_format:
5026 return emitBuiltinOSLogFormat(*E);
5027
5028 case Builtin::BI__xray_customevent: {
5029 if (!ShouldXRayInstrumentFunction())
5030 return RValue::getIgnored();
5031
5032 if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has(
5033 XRayInstrKind::Custom))
5034 return RValue::getIgnored();
5035
5036 if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>())
5037 if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents())
5038 return RValue::getIgnored();
5039
5040 Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent);
5041 auto FTy = F->getFunctionType();
5042 auto Arg0 = E->getArg(0);
5043 auto Arg0Val = EmitScalarExpr(Arg0);
5044 auto Arg0Ty = Arg0->getType();
5045 auto PTy0 = FTy->getParamType(0);
5046 if (PTy0 != Arg0Val->getType()) {
5047 if (Arg0Ty->isArrayType())
5048 Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer();
5049 else
5050 Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0);
5051 }
5052 auto Arg1 = EmitScalarExpr(E->getArg(1));
5053 auto PTy1 = FTy->getParamType(1);
5054 if (PTy1 != Arg1->getType())
5055 Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1);
5056 return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1}));
5057 }
5058
5059 case Builtin::BI__xray_typedevent: {
5060 // TODO: There should be a way to always emit events even if the current
5061 // function is not instrumented. Losing events in a stream can cripple
5062 // a trace.
5063 if (!ShouldXRayInstrumentFunction())
5064 return RValue::getIgnored();
5065
5066 if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has(
5067 XRayInstrKind::Typed))
5068 return RValue::getIgnored();
5069
5070 if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>())
5071 if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents())
5072 return RValue::getIgnored();
5073
5074 Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent);
5075 auto FTy = F->getFunctionType();
5076 auto Arg0 = EmitScalarExpr(E->getArg(0));
5077 auto PTy0 = FTy->getParamType(0);
5078 if (PTy0 != Arg0->getType())
5079 Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0);
5080 auto Arg1 = E->getArg(1);
5081 auto Arg1Val = EmitScalarExpr(Arg1);
5082 auto Arg1Ty = Arg1->getType();
5083 auto PTy1 = FTy->getParamType(1);
5084 if (PTy1 != Arg1Val->getType()) {
5085 if (Arg1Ty->isArrayType())
5086 Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer();
5087 else
5088 Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1);
5089 }
5090 auto Arg2 = EmitScalarExpr(E->getArg(2));
5091 auto PTy2 = FTy->getParamType(2);
5092 if (PTy2 != Arg2->getType())
5093 Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2);
5094 return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2}));
5095 }
5096
5097 case Builtin::BI__builtin_ms_va_start:
5098 case Builtin::BI__builtin_ms_va_end:
5099 return RValue::get(
5100 EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(),
5101 BuiltinID == Builtin::BI__builtin_ms_va_start));
5102
5103 case Builtin::BI__builtin_ms_va_copy: {
5104 // Lower this manually. We can't reliably determine whether or not any
5105 // given va_copy() is for a Win64 va_list from the calling convention
5106 // alone, because it's legal to do this from a System V ABI function.
5107 // With opaque pointer types, we won't have enough information in LLVM
5108 // IR to determine this from the argument types, either. Best to do it
5109 // now, while we have enough information.
5110 Address DestAddr = EmitMSVAListRef(E->getArg(0));
5111 Address SrcAddr = EmitMSVAListRef(E->getArg(1));
5112
5113 llvm::Type *BPP = Int8PtrPtrTy;
5114
5115 DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"),
5116 DestAddr.getAlignment());
5117 SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"),
5118 SrcAddr.getAlignment());
5119
5120 Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val");
5121 return RValue::get(Builder.CreateStore(ArgPtr, DestAddr));
5122 }
5123
5124 case Builtin::BI__builtin_get_device_side_mangled_name: {
5125 auto Name = CGM.getCUDARuntime().getDeviceSideName(
5126 cast<DeclRefExpr>(E->getArg(0)->IgnoreImpCasts())->getDecl());
5127 auto Str = CGM.GetAddrOfConstantCString(Name, "");
5128 llvm::Constant *Zeros[] = {llvm::ConstantInt::get(SizeTy, 0),
5129 llvm::ConstantInt::get(SizeTy, 0)};
5130 auto *Ptr = llvm::ConstantExpr::getGetElementPtr(Str.getElementType(),
5131 Str.getPointer(), Zeros);
5132 return RValue::get(Ptr);
5133 }
5134 }
5135
5136 // If this is an alias for a lib function (e.g. __builtin_sin), emit
5137 // the call using the normal call path, but using the unmangled
5138 // version of the function name.
5139 if (getContext().BuiltinInfo.isLibFunction(BuiltinID))
5140 return emitLibraryCall(*this, FD, E,
5141 CGM.getBuiltinLibFunction(FD, BuiltinID));
5142
5143 // If this is a predefined lib function (e.g. malloc), emit the call
5144 // using exactly the normal call path.
5145 if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID))
5146 return emitLibraryCall(*this, FD, E,
5147 cast<llvm::Constant>(EmitScalarExpr(E->getCallee())));
5148
5149 // Check that a call to a target specific builtin has the correct target
5150 // features.
5151 // This is down here to avoid non-target specific builtins, however, if
5152 // generic builtins start to require generic target features then we
5153 // can move this up to the beginning of the function.
5154 checkTargetFeatures(E, FD);
5155
5156 if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID))
5157 LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth);
5158
5159 // See if we have a target specific intrinsic.
5160 const char *Name = getContext().BuiltinInfo.getName(BuiltinID);
5161 Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic;
5162 StringRef Prefix =
5163 llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch());
5164 if (!Prefix.empty()) {
5165 IntrinsicID = Intrinsic::getIntrinsicForGCCBuiltin(Prefix.data(), Name);
5166 // NOTE we don't need to perform a compatibility flag check here since the
5167 // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the
5168 // MS builtins via ALL_MS_LANGUAGES and are filtered earlier.
5169 if (IntrinsicID == Intrinsic::not_intrinsic)
5170 IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name);
5171 }
5172
5173 if (IntrinsicID != Intrinsic::not_intrinsic) {
5174 SmallVector<Value*, 16> Args;
5175
5176 // Find out if any arguments are required to be integer constant
5177 // expressions.
5178 unsigned ICEArguments = 0;
5179 ASTContext::GetBuiltinTypeError Error;
5180 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
5181 assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None &&
"Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5181, __extension__ __PRETTY_FUNCTION__))
;
5182
5183 Function *F = CGM.getIntrinsic(IntrinsicID);
5184 llvm::FunctionType *FTy = F->getFunctionType();
5185
5186 for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) {
5187 Value *ArgValue;
5188 // If this is a normal argument, just emit it as a scalar.
5189 if ((ICEArguments & (1 << i)) == 0) {
5190 ArgValue = EmitScalarExpr(E->getArg(i));
5191 } else {
5192 // If this is required to be a constant, constant fold it so that we
5193 // know that the generated intrinsic gets a ConstantInt.
5194 ArgValue = llvm::ConstantInt::get(
5195 getLLVMContext(),
5196 *E->getArg(i)->getIntegerConstantExpr(getContext()));
5197 }
5198
5199 // If the intrinsic arg type is different from the builtin arg type
5200 // we need to do a bit cast.
5201 llvm::Type *PTy = FTy->getParamType(i);
5202 if (PTy != ArgValue->getType()) {
5203 // XXX - vector of pointers?
5204 if (auto *PtrTy = dyn_cast<llvm::PointerType>(PTy)) {
5205 if (PtrTy->getAddressSpace() !=
5206 ArgValue->getType()->getPointerAddressSpace()) {
5207 ArgValue = Builder.CreateAddrSpaceCast(
5208 ArgValue,
5209 ArgValue->getType()->getPointerTo(PtrTy->getAddressSpace()));
5210 }
5211 }
5212
5213 assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&(static_cast <bool> (PTy->canLosslesslyBitCastTo(FTy
->getParamType(i)) && "Must be able to losslessly bit cast to param"
) ? void (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5214, __extension__ __PRETTY_FUNCTION__))
5214 "Must be able to losslessly bit cast to param")(static_cast <bool> (PTy->canLosslesslyBitCastTo(FTy
->getParamType(i)) && "Must be able to losslessly bit cast to param"
) ? void (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5214, __extension__ __PRETTY_FUNCTION__))
;
5215 ArgValue = Builder.CreateBitCast(ArgValue, PTy);
5216 }
5217
5218 Args.push_back(ArgValue);
5219 }
5220
5221 Value *V = Builder.CreateCall(F, Args);
5222 QualType BuiltinRetType = E->getType();
5223
5224 llvm::Type *RetTy = VoidTy;
5225 if (!BuiltinRetType->isVoidType())
5226 RetTy = ConvertType(BuiltinRetType);
5227
5228 if (RetTy != V->getType()) {
5229 // XXX - vector of pointers?
5230 if (auto *PtrTy = dyn_cast<llvm::PointerType>(RetTy)) {
5231 if (PtrTy->getAddressSpace() != V->getType()->getPointerAddressSpace()) {
5232 V = Builder.CreateAddrSpaceCast(
5233 V, V->getType()->getPointerTo(PtrTy->getAddressSpace()));
5234 }
5235 }
5236
5237 assert(V->getType()->canLosslesslyBitCastTo(RetTy) &&(static_cast <bool> (V->getType()->canLosslesslyBitCastTo
(RetTy) && "Must be able to losslessly bit cast result type"
) ? void (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5238, __extension__ __PRETTY_FUNCTION__))
5238 "Must be able to losslessly bit cast result type")(static_cast <bool> (V->getType()->canLosslesslyBitCastTo
(RetTy) && "Must be able to losslessly bit cast result type"
) ? void (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5238, __extension__ __PRETTY_FUNCTION__))
;
5239 V = Builder.CreateBitCast(V, RetTy);
5240 }
5241
5242 return RValue::get(V);
5243 }
5244
5245 // Some target-specific builtins can have aggregate return values, e.g.
5246 // __builtin_arm_mve_vld2q_u32. So if the result is an aggregate, force
5247 // ReturnValue to be non-null, so that the target-specific emission code can
5248 // always just emit into it.
5249 TypeEvaluationKind EvalKind = getEvaluationKind(E->getType());
5250 if (EvalKind == TEK_Aggregate && ReturnValue.isNull()) {
5251 Address DestPtr = CreateMemTemp(E->getType(), "agg.tmp");
5252 ReturnValue = ReturnValueSlot(DestPtr, false);
5253 }
5254
5255 // Now see if we can emit a target-specific builtin.
5256 if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E, ReturnValue)) {
5257 switch (EvalKind) {
5258 case TEK_Scalar:
5259 return RValue::get(V);
5260 case TEK_Aggregate:
5261 return RValue::getAggregate(ReturnValue.getValue(),
5262 ReturnValue.isVolatile());
5263 case TEK_Complex:
5264 llvm_unreachable("No current target builtin returns complex")::llvm::llvm_unreachable_internal("No current target builtin returns complex"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5264)
;
5265 }
5266 llvm_unreachable("Bad evaluation kind in EmitBuiltinExpr")::llvm::llvm_unreachable_internal("Bad evaluation kind in EmitBuiltinExpr"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5266)
;
5267 }
5268
5269 ErrorUnsupported(E, "builtin function");
5270
5271 // Unknown builtin, for now just dump it out and return undef.
5272 return GetUndefRValue(E->getType());
5273}
5274
5275static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF,
5276 unsigned BuiltinID, const CallExpr *E,
5277 ReturnValueSlot ReturnValue,
5278 llvm::Triple::ArchType Arch) {
5279 switch (Arch) {
5280 case llvm::Triple::arm:
5281 case llvm::Triple::armeb:
5282 case llvm::Triple::thumb:
5283 case llvm::Triple::thumbeb:
5284 return CGF->EmitARMBuiltinExpr(BuiltinID, E, ReturnValue, Arch);
5285 case llvm::Triple::aarch64:
5286 case llvm::Triple::aarch64_32:
5287 case llvm::Triple::aarch64_be:
5288 return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch);
5289 case llvm::Triple::bpfeb:
5290 case llvm::Triple::bpfel:
5291 return CGF->EmitBPFBuiltinExpr(BuiltinID, E);
5292 case llvm::Triple::x86:
5293 case llvm::Triple::x86_64:
5294 return CGF->EmitX86BuiltinExpr(BuiltinID, E);
5295 case llvm::Triple::ppc:
5296 case llvm::Triple::ppcle:
5297 case llvm::Triple::ppc64:
5298 case llvm::Triple::ppc64le:
5299 return CGF->EmitPPCBuiltinExpr(BuiltinID, E);
5300 case llvm::Triple::r600:
5301 case llvm::Triple::amdgcn:
5302 return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E);
5303 case llvm::Triple::systemz:
5304 return CGF->EmitSystemZBuiltinExpr(BuiltinID, E);
5305 case llvm::Triple::nvptx:
5306 case llvm::Triple::nvptx64:
5307 return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E);
5308 case llvm::Triple::wasm32:
5309 case llvm::Triple::wasm64:
5310 return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E);
5311 case llvm::Triple::hexagon:
5312 return CGF->EmitHexagonBuiltinExpr(BuiltinID, E);
5313 case llvm::Triple::riscv32:
5314 case llvm::Triple::riscv64:
5315 return CGF->EmitRISCVBuiltinExpr(BuiltinID, E, ReturnValue);
5316 default:
5317 return nullptr;
5318 }
5319}
5320
5321Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID,
5322 const CallExpr *E,
5323 ReturnValueSlot ReturnValue) {
5324 if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) {
5325 assert(getContext().getAuxTargetInfo() && "Missing aux target info")(static_cast <bool> (getContext().getAuxTargetInfo() &&
"Missing aux target info") ? void (0) : __assert_fail ("getContext().getAuxTargetInfo() && \"Missing aux target info\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5325, __extension__ __PRETTY_FUNCTION__))
;
5326 return EmitTargetArchBuiltinExpr(
5327 this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E,
5328 ReturnValue, getContext().getAuxTargetInfo()->getTriple().getArch());
5329 }
5330
5331 return EmitTargetArchBuiltinExpr(this, BuiltinID, E, ReturnValue,
5332 getTarget().getTriple().getArch());
5333}
5334
5335static llvm::FixedVectorType *GetNeonType(CodeGenFunction *CGF,
5336 NeonTypeFlags TypeFlags,
5337 bool HasLegalHalfType = true,
5338 bool V1Ty = false,
5339 bool AllowBFloatArgsAndRet = true) {
5340 int IsQuad = TypeFlags.isQuad();
5341 switch (TypeFlags.getEltType()) {
5342 case NeonTypeFlags::Int8:
5343 case NeonTypeFlags::Poly8:
5344 return llvm::FixedVectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad));
5345 case NeonTypeFlags::Int16:
5346 case NeonTypeFlags::Poly16:
5347 return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad));
5348 case NeonTypeFlags::BFloat16:
5349 if (AllowBFloatArgsAndRet)
5350 return llvm::FixedVectorType::get(CGF->BFloatTy, V1Ty ? 1 : (4 << IsQuad));
5351 else
5352 return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad));
5353 case NeonTypeFlags::Float16:
5354 if (HasLegalHalfType)
5355 return llvm::FixedVectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad));
5356 else
5357 return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad));
5358 case NeonTypeFlags::Int32:
5359 return llvm::FixedVectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad));
5360 case NeonTypeFlags::Int64:
5361 case NeonTypeFlags::Poly64:
5362 return llvm::FixedVectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad));
5363 case NeonTypeFlags::Poly128:
5364 // FIXME: i128 and f128 doesn't get fully support in Clang and llvm.
5365 // There is a lot of i128 and f128 API missing.
5366 // so we use v16i8 to represent poly128 and get pattern matched.
5367 return llvm::FixedVectorType::get(CGF->Int8Ty, 16);
5368 case NeonTypeFlags::Float32:
5369 return llvm::FixedVectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad));
5370 case NeonTypeFlags::Float64:
5371 return llvm::FixedVectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad));
5372 }
5373 llvm_unreachable("Unknown vector element type!")::llvm::llvm_unreachable_internal("Unknown vector element type!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5373)
;
5374}
5375
5376static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF,
5377 NeonTypeFlags IntTypeFlags) {
5378 int IsQuad = IntTypeFlags.isQuad();
5379 switch (IntTypeFlags.getEltType()) {
5380 case NeonTypeFlags::Int16:
5381 return llvm::FixedVectorType::get(CGF->HalfTy, (4 << IsQuad));
5382 case NeonTypeFlags::Int32:
5383 return llvm::FixedVectorType::get(CGF->FloatTy, (2 << IsQuad));
5384 case NeonTypeFlags::Int64:
5385 return llvm::FixedVectorType::get(CGF->DoubleTy, (1 << IsQuad));
5386 default:
5387 llvm_unreachable("Type can't be converted to floating-point!")::llvm::llvm_unreachable_internal("Type can't be converted to floating-point!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 5387)
;
5388 }
5389}
5390
5391Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C,
5392 const ElementCount &Count) {
5393 Value *SV = llvm::ConstantVector::getSplat(Count, C);
5394 return Builder.CreateShuffleVector(V, V, SV, "lane");
5395}
5396
5397Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) {
5398 ElementCount EC = cast<llvm::VectorType>(V->getType())->getElementCount();
5399 return EmitNeonSplat(V, C, EC);
5400}
5401
5402Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops,
5403 const char *name,
5404 unsigned shift, bool rightshift) {
5405 unsigned j = 0;
5406 for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
5407 ai != ae; ++ai, ++j) {
5408 if (F->isConstrainedFPIntrinsic())
5409 if (ai->getType()->isMetadataTy())
5410 continue;
5411 if (shift > 0 && shift == j)
5412 Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift);
5413 else
5414 Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name);
5415 }
5416
5417 if (F->isConstrainedFPIntrinsic())
5418 return Builder.CreateConstrainedFPCall(F, Ops, name);
5419 else
5420 return Builder.CreateCall(F, Ops, name);
5421}
5422
5423Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty,
5424 bool neg) {
5425 int SV = cast<ConstantInt>(V)->getSExtValue();
5426 return ConstantInt::get(Ty, neg ? -SV : SV);
5427}
5428
5429// Right-shift a vector by a constant.
5430Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift,
5431 llvm::Type *Ty, bool usgn,
5432 const char *name) {
5433 llvm::VectorType *VTy = cast<llvm::VectorType>(Ty);
5434
5435 int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue();
5436 int EltSize = VTy->getScalarSizeInBits();
5437
5438 Vec = Builder.CreateBitCast(Vec, Ty);
5439
5440 // lshr/ashr are undefined when the shift amount is equal to the vector
5441 // element size.
5442 if (ShiftAmt == EltSize) {
5443 if (usgn) {
5444 // Right-shifting an unsigned value by its size yields 0.
5445 return llvm::ConstantAggregateZero::get(VTy);
5446 } else {
5447 // Right-shifting a signed value by its size is equivalent
5448 // to a shift of size-1.
5449 --ShiftAmt;
5450 Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt);
5451 }
5452 }
5453
5454 Shift = EmitNeonShiftVector(Shift, Ty, false);
5455 if (usgn)
5456 return Builder.CreateLShr(Vec, Shift, name);
5457 else
5458 return Builder.CreateAShr(Vec, Shift, name);
5459}
5460
5461enum {
5462 AddRetType = (1 << 0),
5463 Add1ArgType = (1 << 1),
5464 Add2ArgTypes = (1 << 2),
5465
5466 VectorizeRetType = (1 << 3),
5467 VectorizeArgTypes = (1 << 4),
5468
5469 InventFloatType = (1 << 5),
5470 UnsignedAlts = (1 << 6),
5471
5472 Use64BitVectors = (1 << 7),
5473 Use128BitVectors = (1 << 8),
5474
5475 Vectorize1ArgType = Add1ArgType | VectorizeArgTypes,
5476 VectorRet = AddRetType | VectorizeRetType,
5477 VectorRetGetArgs01 =
5478 AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes,
5479 FpCmpzModifiers =
5480 AddRetType | VectorizeRetType | Add1ArgType | InventFloatType
5481};
5482
5483namespace {
5484struct ARMVectorIntrinsicInfo {
5485 const char *NameHint;
5486 unsigned BuiltinID;
5487 unsigned LLVMIntrinsic;
5488 unsigned AltLLVMIntrinsic;
5489 uint64_t TypeModifier;
5490
5491 bool operator<(unsigned RHSBuiltinID) const {
5492 return BuiltinID < RHSBuiltinID;
5493 }
5494 bool operator<(const ARMVectorIntrinsicInfo &TE) const {
5495 return BuiltinID < TE.BuiltinID;
5496 }
5497};
5498} // end anonymous namespace
5499
5500#define NEONMAP0(NameBase) \
5501 { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 }
5502
5503#define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \
5504 { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \
5505 Intrinsic::LLVMIntrinsic, 0, TypeModifier }
5506
5507#define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \
5508 { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \
5509 Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \
5510 TypeModifier }
5511
5512static const ARMVectorIntrinsicInfo ARMSIMDIntrinsicMap [] = {
5513 NEONMAP1(__a32_vcvt_bf16_v, arm_neon_vcvtfp2bf, 0),
5514 NEONMAP0(splat_lane_v),
5515 NEONMAP0(splat_laneq_v),
5516 NEONMAP0(splatq_lane_v),
5517 NEONMAP0(splatq_laneq_v),
5518 NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts),
5519 NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts),
5520 NEONMAP1(vabs_v, arm_neon_vabs, 0),
5521 NEONMAP1(vabsq_v, arm_neon_vabs, 0),
5522 NEONMAP0(vadd_v),
5523 NEONMAP0(vaddhn_v),
5524 NEONMAP0(vaddq_v),
5525 NEONMAP1(vaesdq_v, arm_neon_aesd, 0),
5526 NEONMAP1(vaeseq_v, arm_neon_aese, 0),
5527 NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0),
5528 NEONMAP1(vaesmcq_v, arm_neon_aesmc, 0),
5529 NEONMAP1(vbfdot_v, arm_neon_bfdot, 0),
5530 NEONMAP1(vbfdotq_v, arm_neon_bfdot, 0),
5531 NEONMAP1(vbfmlalbq_v, arm_neon_bfmlalb, 0),
5532 NEONMAP1(vbfmlaltq_v, arm_neon_bfmlalt, 0),
5533 NEONMAP1(vbfmmlaq_v, arm_neon_bfmmla, 0),
5534 NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType),
5535 NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType),
5536 NEONMAP1(vcadd_rot270_v, arm_neon_vcadd_rot270, Add1ArgType),
5537 NEONMAP1(vcadd_rot90_v, arm_neon_vcadd_rot90, Add1ArgType),
5538 NEONMAP1(vcaddq_rot270_v, arm_neon_vcadd_rot270, Add1ArgType),
5539 NEONMAP1(vcaddq_rot90_v, arm_neon_vcadd_rot90, Add1ArgType),
5540 NEONMAP1(vcage_v, arm_neon_vacge, 0),
5541 NEONMAP1(vcageq_v, arm_neon_vacge, 0),
5542 NEONMAP1(vcagt_v, arm_neon_vacgt, 0),
5543 NEONMAP1(vcagtq_v, arm_neon_vacgt, 0),
5544 NEONMAP1(vcale_v, arm_neon_vacge, 0),
5545 NEONMAP1(vcaleq_v, arm_neon_vacge, 0),
5546 NEONMAP1(vcalt_v, arm_neon_vacgt, 0),
5547 NEONMAP1(vcaltq_v, arm_neon_vacgt, 0),
5548 NEONMAP0(vceqz_v),
5549 NEONMAP0(vceqzq_v),
5550 NEONMAP0(vcgez_v),
5551 NEONMAP0(vcgezq_v),
5552 NEONMAP0(vcgtz_v),
5553 NEONMAP0(vcgtzq_v),
5554 NEONMAP0(vclez_v),
5555 NEONMAP0(vclezq_v),
5556 NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType),
5557 NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType),
5558 NEONMAP0(vcltz_v),
5559 NEONMAP0(vcltzq_v),
5560 NEONMAP1(vclz_v, ctlz, Add1ArgType),
5561 NEONMAP1(vclzq_v, ctlz, Add1ArgType),
5562 NEONMAP1(vcnt_v, ctpop, Add1ArgType),
5563 NEONMAP1(vcntq_v, ctpop, Add1ArgType),
5564 NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0),
5565 NEONMAP0(vcvt_f16_v),
5566 NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0),
5567 NEONMAP0(vcvt_f32_v),
5568 NEONMAP2(vcvt_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
5569 NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
5570 NEONMAP1(vcvt_n_s16_v, arm_neon_vcvtfp2fxs, 0),
5571 NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0),
5572 NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0),
5573 NEONMAP1(vcvt_n_u16_v, arm_neon_vcvtfp2fxu, 0),
5574 NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0),
5575 NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0),
5576 NEONMAP0(vcvt_s16_v),
5577 NEONMAP0(vcvt_s32_v),
5578 NEONMAP0(vcvt_s64_v),
5579 NEONMAP0(vcvt_u16_v),
5580 NEONMAP0(vcvt_u32_v),
5581 NEONMAP0(vcvt_u64_v),
5582 NEONMAP1(vcvta_s16_v, arm_neon_vcvtas, 0),
5583 NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0),
5584 NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0),
5585 NEONMAP1(vcvta_u16_v, arm_neon_vcvtau, 0),
5586 NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0),
5587 NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0),
5588 NEONMAP1(vcvtaq_s16_v, arm_neon_vcvtas, 0),
5589 NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0),
5590 NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0),
5591 NEONMAP1(vcvtaq_u16_v, arm_neon_vcvtau, 0),
5592 NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0),
5593 NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0),
5594 NEONMAP1(vcvth_bf16_f32, arm_neon_vcvtbfp2bf, 0),
5595 NEONMAP1(vcvtm_s16_v, arm_neon_vcvtms, 0),
5596 NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0),
5597 NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0),
5598 NEONMAP1(vcvtm_u16_v, arm_neon_vcvtmu, 0),
5599 NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0),
5600 NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0),
5601 NEONMAP1(vcvtmq_s16_v, arm_neon_vcvtms, 0),
5602 NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0),
5603 NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0),
5604 NEONMAP1(vcvtmq_u16_v, arm_neon_vcvtmu, 0),
5605 NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0),
5606 NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0),
5607 NEONMAP1(vcvtn_s16_v, arm_neon_vcvtns, 0),
5608 NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0),
5609 NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0),
5610 NEONMAP1(vcvtn_u16_v, arm_neon_vcvtnu, 0),
5611 NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0),
5612 NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0),
5613 NEONMAP1(vcvtnq_s16_v, arm_neon_vcvtns, 0),
5614 NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0),
5615 NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0),
5616 NEONMAP1(vcvtnq_u16_v, arm_neon_vcvtnu, 0),
5617 NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0),
5618 NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0),
5619 NEONMAP1(vcvtp_s16_v, arm_neon_vcvtps, 0),
5620 NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0),
5621 NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0),
5622 NEONMAP1(vcvtp_u16_v, arm_neon_vcvtpu, 0),
5623 NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0),
5624 NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0),
5625 NEONMAP1(vcvtpq_s16_v, arm_neon_vcvtps, 0),
5626 NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0),
5627 NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0),
5628 NEONMAP1(vcvtpq_u16_v, arm_neon_vcvtpu, 0),
5629 NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0),
5630 NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0),
5631 NEONMAP0(vcvtq_f16_v),
5632 NEONMAP0(vcvtq_f32_v),
5633 NEONMAP2(vcvtq_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
5634 NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
5635 NEONMAP1(vcvtq_n_s16_v, arm_neon_vcvtfp2fxs, 0),
5636 NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0),
5637 NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0),
5638 NEONMAP1(vcvtq_n_u16_v, arm_neon_vcvtfp2fxu, 0),
5639 NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0),
5640 NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0),
5641 NEONMAP0(vcvtq_s16_v),
5642 NEONMAP0(vcvtq_s32_v),
5643 NEONMAP0(vcvtq_s64_v),
5644 NEONMAP0(vcvtq_u16_v),
5645 NEONMAP0(vcvtq_u32_v),
5646 NEONMAP0(vcvtq_u64_v),
5647 NEONMAP2(vdot_v, arm_neon_udot, arm_neon_sdot, 0),
5648 NEONMAP2(vdotq_v, arm_neon_udot, arm_neon_sdot, 0),
5649 NEONMAP0(vext_v),
5650 NEONMAP0(vextq_v),
5651 NEONMAP0(vfma_v),
5652 NEONMAP0(vfmaq_v),
5653 NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts),
5654 NEONMAP2(vhaddq_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts),
5655 NEONMAP2(vhsub_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts),
5656 NEONMAP2(vhsubq_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts),
5657 NEONMAP0(vld1_dup_v),
5658 NEONMAP1(vld1_v, arm_neon_vld1, 0),
5659 NEONMAP1(vld1_x2_v, arm_neon_vld1x2, 0),
5660 NEONMAP1(vld1_x3_v, arm_neon_vld1x3, 0),
5661 NEONMAP1(vld1_x4_v, arm_neon_vld1x4, 0),
5662 NEONMAP0(vld1q_dup_v),
5663 NEONMAP1(vld1q_v, arm_neon_vld1, 0),
5664 NEONMAP1(vld1q_x2_v, arm_neon_vld1x2, 0),
5665 NEONMAP1(vld1q_x3_v, arm_neon_vld1x3, 0),
5666 NEONMAP1(vld1q_x4_v, arm_neon_vld1x4, 0),
5667 NEONMAP1(vld2_dup_v, arm_neon_vld2dup, 0),
5668 NEONMAP1(vld2_lane_v, arm_neon_vld2lane, 0),
5669 NEONMAP1(vld2_v, arm_neon_vld2, 0),
5670 NEONMAP1(vld2q_dup_v, arm_neon_vld2dup, 0),
5671 NEONMAP1(vld2q_lane_v, arm_neon_vld2lane, 0),
5672 NEONMAP1(vld2q_v, arm_neon_vld2, 0),
5673 NEONMAP1(vld3_dup_v, arm_neon_vld3dup, 0),
5674 NEONMAP1(vld3_lane_v, arm_neon_vld3lane, 0),
5675 NEONMAP1(vld3_v, arm_neon_vld3, 0),
5676 NEONMAP1(vld3q_dup_v, arm_neon_vld3dup, 0),
5677 NEONMAP1(vld3q_lane_v, arm_neon_vld3lane, 0),
5678 NEONMAP1(vld3q_v, arm_neon_vld3, 0),
5679 NEONMAP1(vld4_dup_v, arm_neon_vld4dup, 0),
5680 NEONMAP1(vld4_lane_v, arm_neon_vld4lane, 0),
5681 NEONMAP1(vld4_v, arm_neon_vld4, 0),
5682 NEONMAP1(vld4q_dup_v, arm_neon_vld4dup, 0),
5683 NEONMAP1(vld4q_lane_v, arm_neon_vld4lane, 0),
5684 NEONMAP1(vld4q_v, arm_neon_vld4, 0),
5685 NEONMAP2(vmax_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts),
5686 NEONMAP1(vmaxnm_v, arm_neon_vmaxnm, Add1ArgType),
5687 NEONMAP1(vmaxnmq_v, arm_neon_vmaxnm, Add1ArgType),
5688 NEONMAP2(vmaxq_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts),
5689 NEONMAP2(vmin_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts),
5690 NEONMAP1(vminnm_v, arm_neon_vminnm, Add1ArgType),
5691 NEONMAP1(vminnmq_v, arm_neon_vminnm, Add1ArgType),
5692 NEONMAP2(vminq_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts),
5693 NEONMAP2(vmmlaq_v, arm_neon_ummla, arm_neon_smmla, 0),
5694 NEONMAP0(vmovl_v),
5695 NEONMAP0(vmovn_v),
5696 NEONMAP1(vmul_v, arm_neon_vmulp, Add1ArgType),
5697 NEONMAP0(vmull_v),
5698 NEONMAP1(vmulq_v, arm_neon_vmulp, Add1ArgType),
5699 NEONMAP2(vpadal_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts),
5700 NEONMAP2(vpadalq_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts),
5701 NEONMAP1(vpadd_v, arm_neon_vpadd, Add1ArgType),
5702 NEONMAP2(vpaddl_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts),
5703 NEONMAP2(vpaddlq_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts),
5704 NEONMAP1(vpaddq_v, arm_neon_vpadd, Add1ArgType),
5705 NEONMAP2(vpmax_v, arm_neon_vpmaxu, arm_neon_vpmaxs, Add1ArgType | UnsignedAlts),
5706 NEONMAP2(vpmin_v, arm_neon_vpminu, arm_neon_vpmins, Add1ArgType | UnsignedAlts),
5707 NEONMAP1(vqabs_v, arm_neon_vqabs, Add1ArgType),
5708 NEONMAP1(vqabsq_v, arm_neon_vqabs, Add1ArgType),
5709 NEONMAP2(vqadd_v, uadd_sat, sadd_sat, Add1ArgType | UnsignedAlts),
5710 NEONMAP2(vqaddq_v, uadd_sat, sadd_sat, Add1ArgType | UnsignedAlts),
5711 NEONMAP2(vqdmlal_v, arm_neon_vqdmull, sadd_sat, 0),
5712 NEONMAP2(vqdmlsl_v, arm_neon_vqdmull, ssub_sat, 0),
5713 NEONMAP1(vqdmulh_v, arm_neon_vqdmulh, Add1ArgType),
5714 NEONMAP1(vqdmulhq_v, arm_neon_vqdmulh, Add1ArgType),
5715 NEONMAP1(vqdmull_v, arm_neon_vqdmull, Add1ArgType),
5716 NEONMAP2(vqmovn_v, arm_neon_vqmovnu, arm_neon_vqmovns, Add1ArgType | UnsignedAlts),
5717 NEONMAP1(vqmovun_v, arm_neon_vqmovnsu, Add1ArgType),
5718 NEONMAP1(vqneg_v, arm_neon_vqneg, Add1ArgType),
5719 NEONMAP1(vqnegq_v, arm_neon_vqneg, Add1ArgType),
5720 NEONMAP1(vqrdmulh_v, arm_neon_vqrdmulh, Add1ArgType),
5721 NEONMAP1(vqrdmulhq_v, arm_neon_vqrdmulh, Add1ArgType),
5722 NEONMAP2(vqrshl_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts),
5723 NEONMAP2(vqrshlq_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts),
5724 NEONMAP2(vqshl_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts),
5725 NEONMAP2(vqshl_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts),
5726 NEONMAP2(vqshlq_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts),
5727 NEONMAP2(vqshlq_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts),
5728 NEONMAP1(vqshlu_n_v, arm_neon_vqshiftsu, 0),
5729 NEONMAP1(vqshluq_n_v, arm_neon_vqshiftsu, 0),
5730 NEONMAP2(vqsub_v, usub_sat, ssub_sat, Add1ArgType | UnsignedAlts),
5731 NEONMAP2(vqsubq_v, usub_sat, ssub_sat, Add1ArgType | UnsignedAlts),
5732 NEONMAP1(vraddhn_v, arm_neon_vraddhn, Add1ArgType),
5733 NEONMAP2(vrecpe_v, arm_neon_vrecpe, arm_neon_vrecpe, 0),
5734 NEONMAP2(vrecpeq_v, arm_neon_vrecpe, arm_neon_vrecpe, 0),
5735 NEONMAP1(vrecps_v, arm_neon_vrecps, Add1ArgType),
5736 NEONMAP1(vrecpsq_v, arm_neon_vrecps, Add1ArgType),
5737 NEONMAP2(vrhadd_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts),
5738 NEONMAP2(vrhaddq_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts),
5739 NEONMAP1(vrnd_v, arm_neon_vrintz, Add1ArgType),
5740 NEONMAP1(vrnda_v, arm_neon_vrinta, Add1ArgType),
5741 NEONMAP1(vrndaq_v, arm_neon_vrinta, Add1ArgType),
5742 NEONMAP0(vrndi_v),
5743 NEONMAP0(vrndiq_v),
5744 NEONMAP1(vrndm_v, arm_neon_vrintm, Add1ArgType),
5745 NEONMAP1(vrndmq_v, arm_neon_vrintm, Add1ArgType),
5746 NEONMAP1(vrndn_v, arm_neon_vrintn, Add1ArgType),
5747 NEONMAP1(vrndnq_v, arm_neon_vrintn, Add1ArgType),
5748 NEONMAP1(vrndp_v, arm_neon_vrintp, Add1ArgType),
5749 NEONMAP1(vrndpq_v, arm_neon_vrintp, Add1ArgType),
5750 NEONMAP1(vrndq_v, arm_neon_vrintz, Add1ArgType),
5751 NEONMAP1(vrndx_v, arm_neon_vrintx, Add1ArgType),
5752 NEONMAP1(vrndxq_v, arm_neon_vrintx, Add1ArgType),
5753 NEONMAP2(vrshl_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts),
5754 NEONMAP2(vrshlq_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts),
5755 NEONMAP2(vrshr_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts),
5756 NEONMAP2(vrshrq_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts),
5757 NEONMAP2(vrsqrte_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0),
5758 NEONMAP2(vrsqrteq_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0),
5759 NEONMAP1(vrsqrts_v, arm_neon_vrsqrts, Add1ArgType),
5760 NEONMAP1(vrsqrtsq_v, arm_neon_vrsqrts, Add1ArgType),
5761 NEONMAP1(vrsubhn_v, arm_neon_vrsubhn, Add1ArgType),
5762 NEONMAP1(vsha1su0q_v, arm_neon_sha1su0, 0),
5763 NEONMAP1(vsha1su1q_v, arm_neon_sha1su1, 0),
5764 NEONMAP1(vsha256h2q_v, arm_neon_sha256h2, 0),
5765 NEONMAP1(vsha256hq_v, arm_neon_sha256h, 0),
5766 NEONMAP1(vsha256su0q_v, arm_neon_sha256su0, 0),
5767 NEONMAP1(vsha256su1q_v, arm_neon_sha256su1, 0),
5768 NEONMAP0(vshl_n_v),
5769 NEONMAP2(vshl_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts),
5770 NEONMAP0(vshll_n_v),
5771 NEONMAP0(vshlq_n_v),
5772 NEONMAP2(vshlq_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts),
5773 NEONMAP0(vshr_n_v),
5774 NEONMAP0(vshrn_n_v),
5775 NEONMAP0(vshrq_n_v),
5776 NEONMAP1(vst1_v, arm_neon_vst1, 0),
5777 NEONMAP1(vst1_x2_v, arm_neon_vst1x2, 0),
5778 NEONMAP1(vst1_x3_v, arm_neon_vst1x3, 0),
5779 NEONMAP1(vst1_x4_v, arm_neon_vst1x4, 0),
5780 NEONMAP1(vst1q_v, arm_neon_vst1, 0),
5781 NEONMAP1(vst1q_x2_v, arm_neon_vst1x2, 0),
5782 NEONMAP1(vst1q_x3_v, arm_neon_vst1x3, 0),
5783 NEONMAP1(vst1q_x4_v, arm_neon_vst1x4, 0),
5784 NEONMAP1(vst2_lane_v, arm_neon_vst2lane, 0),
5785 NEONMAP1(vst2_v, arm_neon_vst2, 0),
5786 NEONMAP1(vst2q_lane_v, arm_neon_vst2lane, 0),
5787 NEONMAP1(vst2q_v, arm_neon_vst2, 0),
5788 NEONMAP1(vst3_lane_v, arm_neon_vst3lane, 0),
5789 NEONMAP1(vst3_v, arm_neon_vst3, 0),
5790 NEONMAP1(vst3q_lane_v, arm_neon_vst3lane, 0),
5791 NEONMAP1(vst3q_v, arm_neon_vst3, 0),
5792 NEONMAP1(vst4_lane_v, arm_neon_vst4lane, 0),
5793 NEONMAP1(vst4_v, arm_neon_vst4, 0),
5794 NEONMAP1(vst4q_lane_v, arm_neon_vst4lane, 0),
5795 NEONMAP1(vst4q_v, arm_neon_vst4, 0),
5796 NEONMAP0(vsubhn_v),
5797 NEONMAP0(vtrn_v),
5798 NEONMAP0(vtrnq_v),
5799 NEONMAP0(vtst_v),
5800 NEONMAP0(vtstq_v),
5801 NEONMAP1(vusdot_v, arm_neon_usdot, 0),
5802 NEONMAP1(vusdotq_v, arm_neon_usdot, 0),
5803 NEONMAP1(vusmmlaq_v, arm_neon_usmmla, 0),
5804 NEONMAP0(vuzp_v),
5805 NEONMAP0(vuzpq_v),
5806 NEONMAP0(vzip_v),
5807 NEONMAP0(vzipq_v)
5808};
5809
5810static const ARMVectorIntrinsicInfo AArch64SIMDIntrinsicMap[] = {
5811 NEONMAP1(__a64_vcvtq_low_bf16_v, aarch64_neon_bfcvtn, 0),
5812 NEONMAP0(splat_lane_v),
5813 NEONMAP0(splat_laneq_v),
5814 NEONMAP0(splatq_lane_v),
5815 NEONMAP0(splatq_laneq_v),
5816 NEONMAP1(vabs_v, aarch64_neon_abs, 0),
5817 NEONMAP1(vabsq_v, aarch64_neon_abs, 0),
5818 NEONMAP0(vadd_v),
5819 NEONMAP0(vaddhn_v),
5820 NEONMAP0(vaddq_p128),
5821 NEONMAP0(vaddq_v),
5822 NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0),
5823 NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0),
5824 NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0),
5825 NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0),
5826 NEONMAP2(vbcaxq_v, aarch64_crypto_bcaxu, aarch64_crypto_bcaxs, Add1ArgType | UnsignedAlts),
5827 NEONMAP1(vbfdot_v, aarch64_neon_bfdot, 0),
5828 NEONMAP1(vbfdotq_v, aarch64_neon_bfdot, 0),
5829 NEONMAP1(vbfmlalbq_v, aarch64_neon_bfmlalb, 0),
5830 NEONMAP1(vbfmlaltq_v, aarch64_neon_bfmlalt, 0),
5831 NEONMAP1(vbfmmlaq_v, aarch64_neon_bfmmla, 0),
5832 NEONMAP1(vcadd_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType),
5833 NEONMAP1(vcadd_rot90_v, aarch64_neon_vcadd_rot90, Add1ArgType),
5834 NEONMAP1(vcaddq_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType),
5835 NEONMAP1(vcaddq_rot90_v, aarch64_neon_vcadd_rot90, Add1ArgType),
5836 NEONMAP1(vcage_v, aarch64_neon_facge, 0),
5837 NEONMAP1(vcageq_v, aarch64_neon_facge, 0),
5838 NEONMAP1(vcagt_v, aarch64_neon_facgt, 0),
5839 NEONMAP1(vcagtq_v, aarch64_neon_facgt, 0),
5840 NEONMAP1(vcale_v, aarch64_neon_facge, 0),
5841 NEONMAP1(vcaleq_v, aarch64_neon_facge, 0),
5842 NEONMAP1(vcalt_v, aarch64_neon_facgt, 0),
5843 NEONMAP1(vcaltq_v, aarch64_neon_facgt, 0),
5844 NEONMAP0(vceqz_v),
5845 NEONMAP0(vceqzq_v),
5846 NEONMAP0(vcgez_v),
5847 NEONMAP0(vcgezq_v),
5848 NEONMAP0(vcgtz_v),
5849 NEONMAP0(vcgtzq_v),
5850 NEONMAP0(vclez_v),
5851 NEONMAP0(vclezq_v),
5852 NEONMAP1(vcls_v, aarch64_neon_cls, Add1ArgType),
5853 NEONMAP1(vclsq_v, aarch64_neon_cls, Add1ArgType),
5854 NEONMAP0(vcltz_v),
5855 NEONMAP0(vcltzq_v),
5856 NEONMAP1(vclz_v, ctlz, Add1ArgType),
5857 NEONMAP1(vclzq_v, ctlz, Add1ArgType),
5858 NEONMAP1(vcmla_rot180_v, aarch64_neon_vcmla_rot180, Add1ArgType),
5859 NEONMAP1(vcmla_rot270_v, aarch64_neon_vcmla_rot270, Add1ArgType),
5860 NEONMAP1(vcmla_rot90_v, aarch64_neon_vcmla_rot90, Add1ArgType),
5861 NEONMAP1(vcmla_v, aarch64_neon_vcmla_rot0, Add1ArgType),
5862 NEONMAP1(vcmlaq_rot180_v, aarch64_neon_vcmla_rot180, Add1ArgType),
5863 NEONMAP1(vcmlaq_rot270_v, aarch64_neon_vcmla_rot270, Add1ArgType),
5864 NEONMAP1(vcmlaq_rot90_v, aarch64_neon_vcmla_rot90, Add1ArgType),
5865 NEONMAP1(vcmlaq_v, aarch64_neon_vcmla_rot0, Add1ArgType),
5866 NEONMAP1(vcnt_v, ctpop, Add1ArgType),
5867 NEONMAP1(vcntq_v, ctpop, Add1ArgType),
5868 NEONMAP1(vcvt_f16_f32, aarch64_neon_vcvtfp2hf, 0),
5869 NEONMAP0(vcvt_f16_v),
5870 NEONMAP1(vcvt_f32_f16, aarch64_neon_vcvthf2fp, 0),
5871 NEONMAP0(vcvt_f32_v),
5872 NEONMAP2(vcvt_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
5873 NEONMAP2(vcvt_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
5874 NEONMAP2(vcvt_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
5875 NEONMAP1(vcvt_n_s16_v, aarch64_neon_vcvtfp2fxs, 0),
5876 NEONMAP1(vcvt_n_s32_v, aarch64_neon_vcvtfp2fxs, 0),
5877 NEONMAP1(vcvt_n_s64_v, aarch64_neon_vcvtfp2fxs, 0),
5878 NEONMAP1(vcvt_n_u16_v, aarch64_neon_vcvtfp2fxu, 0),
5879 NEONMAP1(vcvt_n_u32_v, aarch64_neon_vcvtfp2fxu, 0),
5880 NEONMAP1(vcvt_n_u64_v, aarch64_neon_vcvtfp2fxu, 0),
5881 NEONMAP0(vcvtq_f16_v),
5882 NEONMAP0(vcvtq_f32_v),
5883 NEONMAP1(vcvtq_high_bf16_v, aarch64_neon_bfcvtn2, 0),
5884 NEONMAP2(vcvtq_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
5885 NEONMAP2(vcvtq_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
5886 NEONMAP2(vcvtq_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
5887 NEONMAP1(vcvtq_n_s16_v, aarch64_neon_vcvtfp2fxs, 0),
5888 NEONMAP1(vcvtq_n_s32_v, aarch64_neon_vcvtfp2fxs, 0),
5889 NEONMAP1(vcvtq_n_s64_v, aarch64_neon_vcvtfp2fxs, 0),
5890 NEONMAP1(vcvtq_n_u16_v, aarch64_neon_vcvtfp2fxu, 0),
5891 NEONMAP1(vcvtq_n_u32_v, aarch64_neon_vcvtfp2fxu, 0),
5892 NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0),
5893 NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType),
5894 NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0),
5895 NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0),
5896 NEONMAP2(veor3q_v, aarch64_crypto_eor3u, aarch64_crypto_eor3s, Add1ArgType | UnsignedAlts),
5897 NEONMAP0(vext_v),
5898 NEONMAP0(vextq_v),
5899 NEONMAP0(vfma_v),
5900 NEONMAP0(vfmaq_v),
5901 NEONMAP1(vfmlal_high_v, aarch64_neon_fmlal2, 0),
5902 NEONMAP1(vfmlal_low_v, aarch64_neon_fmlal, 0),
5903 NEONMAP1(vfmlalq_high_v, aarch64_neon_fmlal2, 0),
5904 NEONMAP1(vfmlalq_low_v, aarch64_neon_fmlal, 0),
5905 NEONMAP1(vfmlsl_high_v, aarch64_neon_fmlsl2, 0),
5906 NEONMAP1(vfmlsl_low_v, aarch64_neon_fmlsl, 0),
5907 NEONMAP1(vfmlslq_high_v, aarch64_neon_fmlsl2, 0),
5908 NEONMAP1(vfmlslq_low_v, aarch64_neon_fmlsl, 0),
5909 NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts),
5910 NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts),
5911 NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts),
5912 NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts),
5913 NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0),
5914 NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0),
5915 NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0),
5916 NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0),
5917 NEONMAP1(vld1q_x3_v, aarch64_neon_ld1x3, 0),
5918 NEONMAP1(vld1q_x4_v, aarch64_neon_ld1x4, 0),
5919 NEONMAP2(vmmlaq_v, aarch64_neon_ummla, aarch64_neon_smmla, 0),
5920 NEONMAP0(vmovl_v),
5921 NEONMAP0(vmovn_v),
5922 NEONMAP1(vmul_v, aarch64_neon_pmul, Add1ArgType),
5923 NEONMAP1(vmulq_v, aarch64_neon_pmul, Add1ArgType),
5924 NEONMAP1(vpadd_v, aarch64_neon_addp, Add1ArgType),
5925 NEONMAP2(vpaddl_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts),
5926 NEONMAP2(vpaddlq_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts),
5927 NEONMAP1(vpaddq_v, aarch64_neon_addp, Add1ArgType),
5928 NEONMAP1(vqabs_v, aarch64_neon_sqabs, Add1ArgType),
5929 NEONMAP1(vqabsq_v, aarch64_neon_sqabs, Add1ArgType),
5930 NEONMAP2(vqadd_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts),
5931 NEONMAP2(vqaddq_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts),
5932 NEONMAP2(vqdmlal_v, aarch64_neon_sqdmull, aarch64_neon_sqadd, 0),
5933 NEONMAP2(vqdmlsl_v, aarch64_neon_sqdmull, aarch64_neon_sqsub, 0),
5934 NEONMAP1(vqdmulh_lane_v, aarch64_neon_sqdmulh_lane, 0),
5935 NEONMAP1(vqdmulh_laneq_v, aarch64_neon_sqdmulh_laneq, 0),
5936 NEONMAP1(vqdmulh_v, aarch64_neon_sqdmulh, Add1ArgType),
5937 NEONMAP1(vqdmulhq_lane_v, aarch64_neon_sqdmulh_lane, 0),
5938 NEONMAP1(vqdmulhq_laneq_v, aarch64_neon_sqdmulh_laneq, 0),
5939 NEONMAP1(vqdmulhq_v, aarch64_neon_sqdmulh, Add1ArgType),
5940 NEONMAP1(vqdmull_v, aarch64_neon_sqdmull, Add1ArgType),
5941 NEONMAP2(vqmovn_v, aarch64_neon_uqxtn, aarch64_neon_sqxtn, Add1ArgType | UnsignedAlts),
5942 NEONMAP1(vqmovun_v, aarch64_neon_sqxtun, Add1ArgType),
5943 NEONMAP1(vqneg_v, aarch64_neon_sqneg, Add1ArgType),
5944 NEONMAP1(vqnegq_v, aarch64_neon_sqneg, Add1ArgType),
5945 NEONMAP1(vqrdmulh_lane_v, aarch64_neon_sqrdmulh_lane, 0),
5946 NEONMAP1(vqrdmulh_laneq_v, aarch64_neon_sqrdmulh_laneq, 0),
5947 NEONMAP1(vqrdmulh_v, aarch64_neon_sqrdmulh, Add1ArgType),
5948 NEONMAP1(vqrdmulhq_lane_v, aarch64_neon_sqrdmulh_lane, 0),
5949 NEONMAP1(vqrdmulhq_laneq_v, aarch64_neon_sqrdmulh_laneq, 0),
5950 NEONMAP1(vqrdmulhq_v, aarch64_neon_sqrdmulh, Add1ArgType),
5951 NEONMAP2(vqrshl_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts),
5952 NEONMAP2(vqrshlq_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts),
5953 NEONMAP2(vqshl_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl, UnsignedAlts),
5954 NEONMAP2(vqshl_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts),
5955 NEONMAP2(vqshlq_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl,UnsignedAlts),
5956 NEONMAP2(vqshlq_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts),
5957 NEONMAP1(vqshlu_n_v, aarch64_neon_sqshlu, 0),
5958 NEONMAP1(vqshluq_n_v, aarch64_neon_sqshlu, 0),
5959 NEONMAP2(vqsub_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts),
5960 NEONMAP2(vqsubq_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts),
5961 NEONMAP1(vraddhn_v, aarch64_neon_raddhn, Add1ArgType),
5962 NEONMAP1(vrax1q_v, aarch64_crypto_rax1, 0),
5963 NEONMAP2(vrecpe_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0),
5964 NEONMAP2(vrecpeq_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0),
5965 NEONMAP1(vrecps_v, aarch64_neon_frecps, Add1ArgType),
5966 NEONMAP1(vrecpsq_v, aarch64_neon_frecps, Add1ArgType),
5967 NEONMAP2(vrhadd_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts),
5968 NEONMAP2(vrhaddq_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts),
5969 NEONMAP1(vrnd32x_v, aarch64_neon_frint32x, Add1ArgType),
5970 NEONMAP1(vrnd32xq_v, aarch64_neon_frint32x, Add1ArgType),
5971 NEONMAP1(vrnd32z_v, aarch64_neon_frint32z, Add1ArgType),
5972 NEONMAP1(vrnd32zq_v, aarch64_neon_frint32z, Add1ArgType),
5973 NEONMAP1(vrnd64x_v, aarch64_neon_frint64x, Add1ArgType),
5974 NEONMAP1(vrnd64xq_v, aarch64_neon_frint64x, Add1ArgType),
5975 NEONMAP1(vrnd64z_v, aarch64_neon_frint64z, Add1ArgType),
5976 NEONMAP1(vrnd64zq_v, aarch64_neon_frint64z, Add1ArgType),
5977 NEONMAP0(vrndi_v),
5978 NEONMAP0(vrndiq_v),
5979 NEONMAP2(vrshl_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts),
5980 NEONMAP2(vrshlq_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts),
5981 NEONMAP2(vrshr_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts),
5982 NEONMAP2(vrshrq_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts),
5983 NEONMAP2(vrsqrte_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0),
5984 NEONMAP2(vrsqrteq_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0),
5985 NEONMAP1(vrsqrts_v, aarch64_neon_frsqrts, Add1ArgType),
5986 NEONMAP1(vrsqrtsq_v, aarch64_neon_frsqrts, Add1ArgType),
5987 NEONMAP1(vrsubhn_v, aarch64_neon_rsubhn, Add1ArgType),
5988 NEONMAP1(vsha1su0q_v, aarch64_crypto_sha1su0, 0),
5989 NEONMAP1(vsha1su1q_v, aarch64_crypto_sha1su1, 0),
5990 NEONMAP1(vsha256h2q_v, aarch64_crypto_sha256h2, 0),
5991 NEONMAP1(vsha256hq_v, aarch64_crypto_sha256h, 0),
5992 NEONMAP1(vsha256su0q_v, aarch64_crypto_sha256su0, 0),
5993 NEONMAP1(vsha256su1q_v, aarch64_crypto_sha256su1, 0),
5994 NEONMAP1(vsha512h2q_v, aarch64_crypto_sha512h2, 0),
5995 NEONMAP1(vsha512hq_v, aarch64_crypto_sha512h, 0),
5996 NEONMAP1(vsha512su0q_v, aarch64_crypto_sha512su0, 0),
5997 NEONMAP1(vsha512su1q_v, aarch64_crypto_sha512su1, 0),
5998 NEONMAP0(vshl_n_v),
5999 NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts),
6000 NEONMAP0(vshll_n_v),
6001 NEONMAP0(vshlq_n_v),
6002 NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts),
6003 NEONMAP0(vshr_n_v),
6004 NEONMAP0(vshrn_n_v),
6005 NEONMAP0(vshrq_n_v),
6006 NEONMAP1(vsm3partw1q_v, aarch64_crypto_sm3partw1, 0),
6007 NEONMAP1(vsm3partw2q_v, aarch64_crypto_sm3partw2, 0),
6008 NEONMAP1(vsm3ss1q_v, aarch64_crypto_sm3ss1, 0),
6009 NEONMAP1(vsm3tt1aq_v, aarch64_crypto_sm3tt1a, 0),
6010 NEONMAP1(vsm3tt1bq_v, aarch64_crypto_sm3tt1b, 0),
6011 NEONMAP1(vsm3tt2aq_v, aarch64_crypto_sm3tt2a, 0),
6012 NEONMAP1(vsm3tt2bq_v, aarch64_crypto_sm3tt2b, 0),
6013 NEONMAP1(vsm4ekeyq_v, aarch64_crypto_sm4ekey, 0),
6014 NEONMAP1(vsm4eq_v, aarch64_crypto_sm4e, 0),
6015 NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0),
6016 NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0),
6017 NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0),
6018 NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0),
6019 NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0),
6020 NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0),
6021 NEONMAP0(vsubhn_v),
6022 NEONMAP0(vtst_v),
6023 NEONMAP0(vtstq_v),
6024 NEONMAP1(vusdot_v, aarch64_neon_usdot, 0),
6025 NEONMAP1(vusdotq_v, aarch64_neon_usdot, 0),
6026 NEONMAP1(vusmmlaq_v, aarch64_neon_usmmla, 0),
6027 NEONMAP1(vxarq_v, aarch64_crypto_xar, 0),
6028};
6029
6030static const ARMVectorIntrinsicInfo AArch64SISDIntrinsicMap[] = {
6031 NEONMAP1(vabdd_f64, aarch64_sisd_fabd, Add1ArgType),
6032 NEONMAP1(vabds_f32, aarch64_sisd_fabd, Add1ArgType),
6033 NEONMAP1(vabsd_s64, aarch64_neon_abs, Add1ArgType),
6034 NEONMAP1(vaddlv_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType),
6035 NEONMAP1(vaddlv_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType),
6036 NEONMAP1(vaddlvq_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType),
6037 NEONMAP1(vaddlvq_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType),
6038 NEONMAP1(vaddv_f32, aarch64_neon_faddv, AddRetType | Add1ArgType),
6039 NEONMAP1(vaddv_s32, aarch64_neon_saddv, AddRetType | Add1ArgType),
6040 NEONMAP1(vaddv_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType),
6041 NEONMAP1(vaddvq_f32, aarch64_neon_faddv, AddRetType | Add1ArgType),
6042 NEONMAP1(vaddvq_f64, aarch64_neon_faddv, AddRetType | Add1ArgType),
6043 NEONMAP1(vaddvq_s32, aarch64_neon_saddv, AddRetType | Add1ArgType),
6044 NEONMAP1(vaddvq_s64, aarch64_neon_saddv, AddRetType | Add1ArgType),
6045 NEONMAP1(vaddvq_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType),
6046 NEONMAP1(vaddvq_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
6047 NEONMAP1(vcaged_f64, aarch64_neon_facge, AddRetType | Add1ArgType),
6048 NEONMAP1(vcages_f32, aarch64_neon_facge, AddRetType | Add1ArgType),
6049 NEONMAP1(vcagtd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType),
6050 NEONMAP1(vcagts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType),
6051 NEONMAP1(vcaled_f64, aarch64_neon_facge, AddRetType | Add1ArgType),
6052 NEONMAP1(vcales_f32, aarch64_neon_facge, AddRetType | Add1ArgType),
6053 NEONMAP1(vcaltd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType),
6054 NEONMAP1(vcalts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType),
6055 NEONMAP1(vcvtad_s64_f64, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
6056 NEONMAP1(vcvtad_u64_f64, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
6057 NEONMAP1(vcvtas_s32_f32, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
6058 NEONMAP1(vcvtas_u32_f32, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
6059 NEONMAP1(vcvtd_n_f64_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
6060 NEONMAP1(vcvtd_n_f64_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
6061 NEONMAP1(vcvtd_n_s64_f64, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
6062 NEONMAP1(vcvtd_n_u64_f64, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
6063 NEONMAP1(vcvtd_s64_f64, aarch64_neon_fcvtzs, AddRetType | Add1ArgType),
6064 NEONMAP1(vcvtd_u64_f64, aarch64_neon_fcvtzu, AddRetType | Add1ArgType),
6065 NEONMAP1(vcvth_bf16_f32, aarch64_neon_bfcvt, 0),
6066 NEONMAP1(vcvtmd_s64_f64, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
6067 NEONMAP1(vcvtmd_u64_f64, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
6068 NEONMAP1(vcvtms_s32_f32, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
6069 NEONMAP1(vcvtms_u32_f32, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
6070 NEONMAP1(vcvtnd_s64_f64, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
6071 NEONMAP1(vcvtnd_u64_f64, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
6072 NEONMAP1(vcvtns_s32_f32, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
6073 NEONMAP1(vcvtns_u32_f32, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
6074 NEONMAP1(vcvtpd_s64_f64, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
6075 NEONMAP1(vcvtpd_u64_f64, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
6076 NEONMAP1(vcvtps_s32_f32, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
6077 NEONMAP1(vcvtps_u32_f32, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
6078 NEONMAP1(vcvts_n_f32_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
6079 NEONMAP1(vcvts_n_f32_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
6080 NEONMAP1(vcvts_n_s32_f32, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
6081 NEONMAP1(vcvts_n_u32_f32, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
6082 NEONMAP1(vcvts_s32_f32, aarch64_neon_fcvtzs, AddRetType | Add1ArgType),
6083 NEONMAP1(vcvts_u32_f32, aarch64_neon_fcvtzu, AddRetType | Add1ArgType),
6084 NEONMAP1(vcvtxd_f32_f64, aarch64_sisd_fcvtxn, 0),
6085 NEONMAP1(vmaxnmv_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
6086 NEONMAP1(vmaxnmvq_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
6087 NEONMAP1(vmaxnmvq_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
6088 NEONMAP1(vmaxv_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
6089 NEONMAP1(vmaxv_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType),
6090 NEONMAP1(vmaxv_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType),
6091 NEONMAP1(vmaxvq_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
6092 NEONMAP1(vmaxvq_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
6093 NEONMAP1(vmaxvq_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType),
6094 NEONMAP1(vmaxvq_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType),
6095 NEONMAP1(vminnmv_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
6096 NEONMAP1(vminnmvq_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
6097 NEONMAP1(vminnmvq_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
6098 NEONMAP1(vminv_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
6099 NEONMAP1(vminv_s32, aarch64_neon_sminv, AddRetType | Add1ArgType),
6100 NEONMAP1(vminv_u32, aarch64_neon_uminv, AddRetType | Add1ArgType),
6101 NEONMAP1(vminvq_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
6102 NEONMAP1(vminvq_f64, aarch64_neon_fminv, AddRetType | Add1ArgType),
6103 NEONMAP1(vminvq_s32, aarch64_neon_sminv, AddRetType | Add1ArgType),
6104 NEONMAP1(vminvq_u32, aarch64_neon_uminv, AddRetType | Add1ArgType),
6105 NEONMAP1(vmull_p64, aarch64_neon_pmull64, 0),
6106 NEONMAP1(vmulxd_f64, aarch64_neon_fmulx, Add1ArgType),
6107 NEONMAP1(vmulxs_f32, aarch64_neon_fmulx, Add1ArgType),
6108 NEONMAP1(vpaddd_s64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
6109 NEONMAP1(vpaddd_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
6110 NEONMAP1(vpmaxnmqd_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
6111 NEONMAP1(vpmaxnms_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
6112 NEONMAP1(vpmaxqd_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
6113 NEONMAP1(vpmaxs_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
6114 NEONMAP1(vpminnmqd_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
6115 NEONMAP1(vpminnms_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
6116 NEONMAP1(vpminqd_f64, aarch64_neon_fminv, AddRetType | Add1ArgType),
6117 NEONMAP1(vpmins_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
6118 NEONMAP1(vqabsb_s8, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors),
6119 NEONMAP1(vqabsd_s64, aarch64_neon_sqabs, Add1ArgType),
6120 NEONMAP1(vqabsh_s16, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors),
6121 NEONMAP1(vqabss_s32, aarch64_neon_sqabs, Add1ArgType),
6122 NEONMAP1(vqaddb_s8, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors),
6123 NEONMAP1(vqaddb_u8, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors),
6124 NEONMAP1(vqaddd_s64, aarch64_neon_sqadd, Add1ArgType),
6125 NEONMAP1(vqaddd_u64, aarch64_neon_uqadd, Add1ArgType),
6126 NEONMAP1(vqaddh_s16, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors),
6127 NEONMAP1(vqaddh_u16, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors),
6128 NEONMAP1(vqadds_s32, aarch64_neon_sqadd, Add1ArgType),
6129 NEONMAP1(vqadds_u32, aarch64_neon_uqadd, Add1ArgType),
6130 NEONMAP1(vqdmulhh_s16, aarch64_neon_sqdmulh, Vectorize1ArgType | Use64BitVectors),
6131 NEONMAP1(vqdmulhs_s32, aarch64_neon_sqdmulh, Add1ArgType),
6132 NEONMAP1(vqdmullh_s16, aarch64_neon_sqdmull, VectorRet | Use128BitVectors),
6133 NEONMAP1(vqdmulls_s32, aarch64_neon_sqdmulls_scalar, 0),
6134 NEONMAP1(vqmovnd_s64, aarch64_neon_scalar_sqxtn, AddRetType | Add1ArgType),
6135 NEONMAP1(vqmovnd_u64, aarch64_neon_scalar_uqxtn, AddRetType | Add1ArgType),
6136 NEONMAP1(vqmovnh_s16, aarch64_neon_sqxtn, VectorRet | Use64BitVectors),
6137 NEONMAP1(vqmovnh_u16, aarch64_neon_uqxtn, VectorRet | Use64BitVectors),
6138 NEONMAP1(vqmovns_s32, aarch64_neon_sqxtn, VectorRet | Use64BitVectors),
6139 NEONMAP1(vqmovns_u32, aarch64_neon_uqxtn, VectorRet | Use64BitVectors),
6140 NEONMAP1(vqmovund_s64, aarch64_neon_scalar_sqxtun, AddRetType | Add1ArgType),
6141 NEONMAP1(vqmovunh_s16, aarch64_neon_sqxtun, VectorRet | Use64BitVectors),
6142 NEONMAP1(vqmovuns_s32, aarch64_neon_sqxtun, VectorRet | Use64BitVectors),
6143 NEONMAP1(vqnegb_s8, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors),
6144 NEONMAP1(vqnegd_s64, aarch64_neon_sqneg, Add1ArgType),
6145 NEONMAP1(vqnegh_s16, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors),
6146 NEONMAP1(vqnegs_s32, aarch64_neon_sqneg, Add1ArgType),
6147 NEONMAP1(vqrdmulhh_s16, aarch64_neon_sqrdmulh, Vectorize1ArgType | Use64BitVectors),
6148 NEONMAP1(vqrdmulhs_s32, aarch64_neon_sqrdmulh, Add1ArgType),
6149 NEONMAP1(vqrshlb_s8, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors),
6150 NEONMAP1(vqrshlb_u8, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors),
6151 NEONMAP1(vqrshld_s64, aarch64_neon_sqrshl, Add1ArgType),
6152 NEONMAP1(vqrshld_u64, aarch64_neon_uqrshl, Add1ArgType),
6153 NEONMAP1(vqrshlh_s16, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors),
6154 NEONMAP1(vqrshlh_u16, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors),
6155 NEONMAP1(vqrshls_s32, aarch64_neon_sqrshl, Add1ArgType),
6156 NEONMAP1(vqrshls_u32, aarch64_neon_uqrshl, Add1ArgType),
6157 NEONMAP1(vqrshrnd_n_s64, aarch64_neon_sqrshrn, AddRetType),
6158 NEONMAP1(vqrshrnd_n_u64, aarch64_neon_uqrshrn, AddRetType),
6159 NEONMAP1(vqrshrnh_n_s16, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors),
6160 NEONMAP1(vqrshrnh_n_u16, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors),
6161 NEONMAP1(vqrshrns_n_s32, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors),
6162 NEONMAP1(vqrshrns_n_u32, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors),
6163 NEONMAP1(vqrshrund_n_s64, aarch64_neon_sqrshrun, AddRetType),
6164 NEONMAP1(vqrshrunh_n_s16, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors),
6165 NEONMAP1(vqrshruns_n_s32, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors),
6166 NEONMAP1(vqshlb_n_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
6167 NEONMAP1(vqshlb_n_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
6168 NEONMAP1(vqshlb_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
6169 NEONMAP1(vqshlb_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
6170 NEONMAP1(vqshld_s64, aarch64_neon_sqshl, Add1ArgType),
6171 NEONMAP1(vqshld_u64, aarch64_neon_uqshl, Add1ArgType),
6172 NEONMAP1(vqshlh_n_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
6173 NEONMAP1(vqshlh_n_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
6174 NEONMAP1(vqshlh_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
6175 NEONMAP1(vqshlh_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
6176 NEONMAP1(vqshls_n_s32, aarch64_neon_sqshl, Add1ArgType),
6177 NEONMAP1(vqshls_n_u32, aarch64_neon_uqshl, Add1ArgType),
6178 NEONMAP1(vqshls_s32, aarch64_neon_sqshl, Add1ArgType),
6179 NEONMAP1(vqshls_u32, aarch64_neon_uqshl, Add1ArgType),
6180 NEONMAP1(vqshlub_n_s8, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors),
6181 NEONMAP1(vqshluh_n_s16, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors),
6182 NEONMAP1(vqshlus_n_s32, aarch64_neon_sqshlu, Add1ArgType),
6183 NEONMAP1(vqshrnd_n_s64, aarch64_neon_sqshrn, AddRetType),
6184 NEONMAP1(vqshrnd_n_u64, aarch64_neon_uqshrn, AddRetType),
6185 NEONMAP1(vqshrnh_n_s16, aarch64_neon_sqshrn, VectorRet | Use64BitVectors),
6186 NEONMAP1(vqshrnh_n_u16, aarch64_neon_uqshrn, VectorRet | Use64BitVectors),
6187 NEONMAP1(vqshrns_n_s32, aarch64_neon_sqshrn, VectorRet | Use64BitVectors),
6188 NEONMAP1(vqshrns_n_u32, aarch64_neon_uqshrn, VectorRet | Use64BitVectors),
6189 NEONMAP1(vqshrund_n_s64, aarch64_neon_sqshrun, AddRetType),
6190 NEONMAP1(vqshrunh_n_s16, aarch64_neon_sqshrun, VectorRet | Use64BitVectors),
6191 NEONMAP1(vqshruns_n_s32, aarch64_neon_sqshrun, VectorRet | Use64BitVectors),
6192 NEONMAP1(vqsubb_s8, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors),
6193 NEONMAP1(vqsubb_u8, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors),
6194 NEONMAP1(vqsubd_s64, aarch64_neon_sqsub, Add1ArgType),
6195 NEONMAP1(vqsubd_u64, aarch64_neon_uqsub, Add1ArgType),
6196 NEONMAP1(vqsubh_s16, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors),
6197 NEONMAP1(vqsubh_u16, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors),
6198 NEONMAP1(vqsubs_s32, aarch64_neon_sqsub, Add1ArgType),
6199 NEONMAP1(vqsubs_u32, aarch64_neon_uqsub, Add1ArgType),
6200 NEONMAP1(vrecped_f64, aarch64_neon_frecpe, Add1ArgType),
6201 NEONMAP1(vrecpes_f32, aarch64_neon_frecpe, Add1ArgType),
6202 NEONMAP1(vrecpxd_f64, aarch64_neon_frecpx, Add1ArgType),
6203 NEONMAP1(vrecpxs_f32, aarch64_neon_frecpx, Add1ArgType),
6204 NEONMAP1(vrshld_s64, aarch64_neon_srshl, Add1ArgType),
6205 NEONMAP1(vrshld_u64, aarch64_neon_urshl, Add1ArgType),
6206 NEONMAP1(vrsqrted_f64, aarch64_neon_frsqrte, Add1ArgType),
6207 NEONMAP1(vrsqrtes_f32, aarch64_neon_frsqrte, Add1ArgType),
6208 NEONMAP1(vrsqrtsd_f64, aarch64_neon_frsqrts, Add1ArgType),
6209 NEONMAP1(vrsqrtss_f32, aarch64_neon_frsqrts, Add1ArgType),
6210 NEONMAP1(vsha1cq_u32, aarch64_crypto_sha1c, 0),
6211 NEONMAP1(vsha1h_u32, aarch64_crypto_sha1h, 0),
6212 NEONMAP1(vsha1mq_u32, aarch64_crypto_sha1m, 0),
6213 NEONMAP1(vsha1pq_u32, aarch64_crypto_sha1p, 0),
6214 NEONMAP1(vshld_s64, aarch64_neon_sshl, Add1ArgType),
6215 NEONMAP1(vshld_u64, aarch64_neon_ushl, Add1ArgType),
6216 NEONMAP1(vslid_n_s64, aarch64_neon_vsli, Vectorize1ArgType),
6217 NEONMAP1(vslid_n_u64, aarch64_neon_vsli, Vectorize1ArgType),
6218 NEONMAP1(vsqaddb_u8, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors),
6219 NEONMAP1(vsqaddd_u64, aarch64_neon_usqadd, Add1ArgType),
6220 NEONMAP1(vsqaddh_u16, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors),
6221 NEONMAP1(vsqadds_u32, aarch64_neon_usqadd, Add1ArgType),
6222 NEONMAP1(vsrid_n_s64, aarch64_neon_vsri, Vectorize1ArgType),
6223 NEONMAP1(vsrid_n_u64, aarch64_neon_vsri, Vectorize1ArgType),
6224 NEONMAP1(vuqaddb_s8, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors),
6225 NEONMAP1(vuqaddd_s64, aarch64_neon_suqadd, Add1ArgType),
6226 NEONMAP1(vuqaddh_s16, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors),
6227 NEONMAP1(vuqadds_s32, aarch64_neon_suqadd, Add1ArgType),
6228 // FP16 scalar intrinisics go here.
6229 NEONMAP1(vabdh_f16, aarch64_sisd_fabd, Add1ArgType),
6230 NEONMAP1(vcvtah_s32_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
6231 NEONMAP1(vcvtah_s64_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
6232 NEONMAP1(vcvtah_u32_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
6233 NEONMAP1(vcvtah_u64_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
6234 NEONMAP1(vcvth_n_f16_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
6235 NEONMAP1(vcvth_n_f16_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
6236 NEONMAP1(vcvth_n_f16_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
6237 NEONMAP1(vcvth_n_f16_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
6238 NEONMAP1(vcvth_n_s32_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
6239 NEONMAP1(vcvth_n_s64_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
6240 NEONMAP1(vcvth_n_u32_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
6241 NEONMAP1(vcvth_n_u64_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
6242 NEONMAP1(vcvth_s32_f16, aarch64_neon_fcvtzs, AddRetType | Add1ArgType),
6243 NEONMAP1(vcvth_s64_f16, aarch64_neon_fcvtzs, AddRetType | Add1ArgType),
6244 NEONMAP1(vcvth_u32_f16, aarch64_neon_fcvtzu, AddRetType | Add1ArgType),
6245 NEONMAP1(vcvth_u64_f16, aarch64_neon_fcvtzu, AddRetType | Add1ArgType),
6246 NEONMAP1(vcvtmh_s32_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
6247 NEONMAP1(vcvtmh_s64_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
6248 NEONMAP1(vcvtmh_u32_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
6249 NEONMAP1(vcvtmh_u64_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
6250 NEONMAP1(vcvtnh_s32_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
6251 NEONMAP1(vcvtnh_s64_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
6252 NEONMAP1(vcvtnh_u32_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
6253 NEONMAP1(vcvtnh_u64_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
6254 NEONMAP1(vcvtph_s32_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
6255 NEONMAP1(vcvtph_s64_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
6256 NEONMAP1(vcvtph_u32_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
6257 NEONMAP1(vcvtph_u64_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
6258 NEONMAP1(vmulxh_f16, aarch64_neon_fmulx, Add1ArgType),
6259 NEONMAP1(vrecpeh_f16, aarch64_neon_frecpe, Add1ArgType),
6260 NEONMAP1(vrecpxh_f16, aarch64_neon_frecpx, Add1ArgType),
6261 NEONMAP1(vrsqrteh_f16, aarch64_neon_frsqrte, Add1ArgType),
6262 NEONMAP1(vrsqrtsh_f16, aarch64_neon_frsqrts, Add1ArgType),
6263};
6264
6265#undef NEONMAP0
6266#undef NEONMAP1
6267#undef NEONMAP2
6268
6269#define SVEMAP1(NameBase, LLVMIntrinsic, TypeModifier) \
6270 { \
6271 #NameBase, SVE::BI__builtin_sve_##NameBase, Intrinsic::LLVMIntrinsic, 0, \
6272 TypeModifier \
6273 }
6274
6275#define SVEMAP2(NameBase, TypeModifier) \
6276 { #NameBase, SVE::BI__builtin_sve_##NameBase, 0, 0, TypeModifier }
6277static const ARMVectorIntrinsicInfo AArch64SVEIntrinsicMap[] = {
6278#define GET_SVE_LLVM_INTRINSIC_MAP
6279#include "clang/Basic/arm_sve_builtin_cg.inc"
6280#undef GET_SVE_LLVM_INTRINSIC_MAP
6281};
6282
6283#undef SVEMAP1
6284#undef SVEMAP2
6285
6286static bool NEONSIMDIntrinsicsProvenSorted = false;
6287
6288static bool AArch64SIMDIntrinsicsProvenSorted = false;
6289static bool AArch64SISDIntrinsicsProvenSorted = false;
6290static bool AArch64SVEIntrinsicsProvenSorted = false;
6291
6292static const ARMVectorIntrinsicInfo *
6293findARMVectorIntrinsicInMap(ArrayRef<ARMVectorIntrinsicInfo> IntrinsicMap,
6294 unsigned BuiltinID, bool &MapProvenSorted) {
6295
6296#ifndef NDEBUG
6297 if (!MapProvenSorted) {
6298 assert(llvm::is_sorted(IntrinsicMap))(static_cast <bool> (llvm::is_sorted(IntrinsicMap)) ? void
(0) : __assert_fail ("llvm::is_sorted(IntrinsicMap)", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 6298, __extension__ __PRETTY_FUNCTION__))
;
6299 MapProvenSorted = true;
6300 }
6301#endif
6302
6303 const ARMVectorIntrinsicInfo *Builtin =
6304 llvm::lower_bound(IntrinsicMap, BuiltinID);
6305
6306 if (Builtin != IntrinsicMap.end() && Builtin->BuiltinID == BuiltinID)
6307 return Builtin;
6308
6309 return nullptr;
6310}
6311
6312Function *CodeGenFunction::LookupNeonLLVMIntrinsic(unsigned IntrinsicID,
6313 unsigned Modifier,
6314 llvm::Type *ArgType,
6315 const CallExpr *E) {
6316 int VectorSize = 0;
6317 if (Modifier & Use64BitVectors)
6318 VectorSize = 64;
6319 else if (Modifier & Use128BitVectors)
6320 VectorSize = 128;
6321
6322 // Return type.
6323 SmallVector<llvm::Type *, 3> Tys;
6324 if (Modifier & AddRetType) {
6325 llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext()));
6326 if (Modifier & VectorizeRetType)
6327 Ty = llvm::FixedVectorType::get(
6328 Ty, VectorSize ? VectorSize / Ty->getPrimitiveSizeInBits() : 1);
6329
6330 Tys.push_back(Ty);
6331 }
6332
6333 // Arguments.
6334 if (Modifier & VectorizeArgTypes) {
6335 int Elts = VectorSize ? VectorSize / ArgType->getPrimitiveSizeInBits() : 1;
6336 ArgType = llvm::FixedVectorType::get(ArgType, Elts);
6337 }
6338
6339 if (Modifier & (Add1ArgType | Add2ArgTypes))
6340 Tys.push_back(ArgType);
6341
6342 if (Modifier & Add2ArgTypes)
6343 Tys.push_back(ArgType);
6344
6345 if (Modifier & InventFloatType)
6346 Tys.push_back(FloatTy);
6347
6348 return CGM.getIntrinsic(IntrinsicID, Tys);
6349}
6350
6351static Value *EmitCommonNeonSISDBuiltinExpr(
6352 CodeGenFunction &CGF, const ARMVectorIntrinsicInfo &SISDInfo,
6353 SmallVectorImpl<Value *> &Ops, const CallExpr *E) {
6354 unsigned BuiltinID = SISDInfo.BuiltinID;
6355 unsigned int Int = SISDInfo.LLVMIntrinsic;
6356 unsigned Modifier = SISDInfo.TypeModifier;
6357 const char *s = SISDInfo.NameHint;
6358
6359 switch (BuiltinID) {
6360 case NEON::BI__builtin_neon_vcled_s64:
6361 case NEON::BI__builtin_neon_vcled_u64:
6362 case NEON::BI__builtin_neon_vcles_f32:
6363 case NEON::BI__builtin_neon_vcled_f64:
6364 case NEON::BI__builtin_neon_vcltd_s64:
6365 case NEON::BI__builtin_neon_vcltd_u64:
6366 case NEON::BI__builtin_neon_vclts_f32:
6367 case NEON::BI__builtin_neon_vcltd_f64:
6368 case NEON::BI__builtin_neon_vcales_f32:
6369 case NEON::BI__builtin_neon_vcaled_f64:
6370 case NEON::BI__builtin_neon_vcalts_f32:
6371 case NEON::BI__builtin_neon_vcaltd_f64:
6372 // Only one direction of comparisons actually exist, cmle is actually a cmge
6373 // with swapped operands. The table gives us the right intrinsic but we
6374 // still need to do the swap.
6375 std::swap(Ops[0], Ops[1]);
6376 break;
6377 }
6378
6379 assert(Int && "Generic code assumes a valid intrinsic")(static_cast <bool> (Int && "Generic code assumes a valid intrinsic"
) ? void (0) : __assert_fail ("Int && \"Generic code assumes a valid intrinsic\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 6379, __extension__ __PRETTY_FUNCTION__))
;
6380
6381 // Determine the type(s) of this overloaded AArch64 intrinsic.
6382 const Expr *Arg = E->getArg(0);
6383 llvm::Type *ArgTy = CGF.ConvertType(Arg->getType());
6384 Function *F = CGF.LookupNeonLLVMIntrinsic(Int, Modifier, ArgTy, E);
6385
6386 int j = 0;
6387 ConstantInt *C0 = ConstantInt::get(CGF.SizeTy, 0);
6388 for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
6389 ai != ae; ++ai, ++j) {
6390 llvm::Type *ArgTy = ai->getType();
6391 if (Ops[j]->getType()->getPrimitiveSizeInBits() ==
6392 ArgTy->getPrimitiveSizeInBits())
6393 continue;
6394
6395 assert(ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy())(static_cast <bool> (ArgTy->isVectorTy() && !
Ops[j]->getType()->isVectorTy()) ? void (0) : __assert_fail
("ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy()"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 6395, __extension__ __PRETTY_FUNCTION__))
;
6396 // The constant argument to an _n_ intrinsic always has Int32Ty, so truncate
6397 // it before inserting.
6398 Ops[j] = CGF.Builder.CreateTruncOrBitCast(
6399 Ops[j], cast<llvm::VectorType>(ArgTy)->getElementType());
6400 Ops[j] =
6401 CGF.Builder.CreateInsertElement(UndefValue::get(ArgTy), Ops[j], C0);
6402 }
6403
6404 Value *Result = CGF.EmitNeonCall(F, Ops, s);
6405 llvm::Type *ResultType = CGF.ConvertType(E->getType());
6406 if (ResultType->getPrimitiveSizeInBits().getFixedSize() <
6407 Result->getType()->getPrimitiveSizeInBits().getFixedSize())
6408 return CGF.Builder.CreateExtractElement(Result, C0);
6409
6410 return CGF.Builder.CreateBitCast(Result, ResultType, s);
6411}
6412
6413Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
6414 unsigned BuiltinID, unsigned LLVMIntrinsic, unsigned AltLLVMIntrinsic,
6415 const char *NameHint, unsigned Modifier, const CallExpr *E,
6416 SmallVectorImpl<llvm::Value *> &Ops, Address PtrOp0, Address PtrOp1,
6417 llvm::Triple::ArchType Arch) {
6418 // Get the last argument, which specifies the vector type.
6419 const Expr *Arg = E->getArg(E->getNumArgs() - 1);
6420 Optional<llvm::APSInt> NeonTypeConst =
6421 Arg->getIntegerConstantExpr(getContext());
6422 if (!NeonTypeConst)
6423 return nullptr;
6424
6425 // Determine the type of this overloaded NEON intrinsic.
6426 NeonTypeFlags Type(NeonTypeConst->getZExtValue());
6427 bool Usgn = Type.isUnsigned();
6428 bool Quad = Type.isQuad();
6429 const bool HasLegalHalfType = getTarget().hasLegalHalfType();
6430 const bool AllowBFloatArgsAndRet =
6431 getTargetHooks().getABIInfo().allowBFloatArgsAndRet();
6432
6433 llvm::FixedVectorType *VTy =
6434 GetNeonType(this, Type, HasLegalHalfType, false, AllowBFloatArgsAndRet);
6435 llvm::Type *Ty = VTy;
6436 if (!Ty)
6437 return nullptr;
6438
6439 auto getAlignmentValue32 = [&](Address addr) -> Value* {
6440 return Builder.getInt32(addr.getAlignment().getQuantity());
6441 };
6442
6443 unsigned Int = LLVMIntrinsic;
6444 if ((Modifier & UnsignedAlts) && !Usgn)
6445 Int = AltLLVMIntrinsic;
6446
6447 switch (BuiltinID) {
6448 default: break;
6449 case NEON::BI__builtin_neon_splat_lane_v:
6450 case NEON::BI__builtin_neon_splat_laneq_v:
6451 case NEON::BI__builtin_neon_splatq_lane_v:
6452 case NEON::BI__builtin_neon_splatq_laneq_v: {
6453 auto NumElements = VTy->getElementCount();
6454 if (BuiltinID == NEON::BI__builtin_neon_splatq_lane_v)
6455 NumElements = NumElements * 2;
6456 if (BuiltinID == NEON::BI__builtin_neon_splat_laneq_v)
6457 NumElements = NumElements.divideCoefficientBy(2);
6458
6459 Ops[0] = Builder.CreateBitCast(Ops[0], VTy);
6460 return EmitNeonSplat(Ops[0], cast<ConstantInt>(Ops[1]), NumElements);
6461 }
6462 case NEON::BI__builtin_neon_vpadd_v:
6463 case NEON::BI__builtin_neon_vpaddq_v:
6464 // We don't allow fp/int overloading of intrinsics.
6465 if (VTy->getElementType()->isFloatingPointTy() &&
6466 Int == Intrinsic::aarch64_neon_addp)
6467 Int = Intrinsic::aarch64_neon_faddp;
6468 break;
6469 case NEON::BI__builtin_neon_vabs_v:
6470 case NEON::BI__builtin_neon_vabsq_v:
6471 if (VTy->getElementType()->isFloatingPointTy())
6472 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, Ty), Ops, "vabs");
6473 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), Ops, "vabs");
6474 case NEON::BI__builtin_neon_vadd_v:
6475 case NEON::BI__builtin_neon_vaddq_v: {
6476 llvm::Type *VTy = llvm::FixedVectorType::get(Int8Ty, Quad ? 16 : 8);
6477 Ops[0] = Builder.CreateBitCast(Ops[0], VTy);
6478 Ops[1] = Builder.CreateBitCast(Ops[1], VTy);
6479 Ops[0] = Builder.CreateXor(Ops[0], Ops[1]);
6480 return Builder.CreateBitCast(Ops[0], Ty);
6481 }
6482 case NEON::BI__builtin_neon_vaddhn_v: {
6483 llvm::FixedVectorType *SrcTy =
6484 llvm::FixedVectorType::getExtendedElementVectorType(VTy);
6485
6486 // %sum = add <4 x i32> %lhs, %rhs
6487 Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
6488 Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy);
6489 Ops[0] = Builder.CreateAdd(Ops[0], Ops[1], "vaddhn");
6490
6491 // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16>
6492 Constant *ShiftAmt =
6493 ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2);
6494 Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vaddhn");
6495
6496 // %res = trunc <4 x i32> %high to <4 x i16>
6497 return Builder.CreateTrunc(Ops[0], VTy, "vaddhn");
6498 }
6499 case NEON::BI__builtin_neon_vcale_v:
6500 case NEON::BI__builtin_neon_vcaleq_v:
6501 case NEON::BI__builtin_neon_vcalt_v:
6502 case NEON::BI__builtin_neon_vcaltq_v:
6503 std::swap(Ops[0], Ops[1]);
6504 LLVM_FALLTHROUGH[[gnu::fallthrough]];
6505 case NEON::BI__builtin_neon_vcage_v:
6506 case NEON::BI__builtin_neon_vcageq_v:
6507 case NEON::BI__builtin_neon_vcagt_v:
6508 case NEON::BI__builtin_neon_vcagtq_v: {
6509 llvm::Type *Ty;
6510 switch (VTy->getScalarSizeInBits()) {
6511 default: llvm_unreachable("unexpected type")::llvm::llvm_unreachable_internal("unexpected type", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 6511)
;
6512 case 32:
6513 Ty = FloatTy;
6514 break;
6515 case 64:
6516 Ty = DoubleTy;
6517 break;
6518 case 16:
6519 Ty = HalfTy;
6520 break;
6521 }
6522 auto *VecFlt = llvm::FixedVectorType::get(Ty, VTy->getNumElements());
6523 llvm::Type *Tys[] = { VTy, VecFlt };
6524 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
6525 return EmitNeonCall(F, Ops, NameHint);
6526 }
6527 case NEON::BI__builtin_neon_vceqz_v:
6528 case NEON::BI__builtin_neon_vceqzq_v:
6529 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OEQ,
6530 ICmpInst::ICMP_EQ, "vceqz");
6531 case NEON::BI__builtin_neon_vcgez_v:
6532 case NEON::BI__builtin_neon_vcgezq_v:
6533 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGE,
6534 ICmpInst::ICMP_SGE, "vcgez");
6535 case NEON::BI__builtin_neon_vclez_v:
6536 case NEON::BI__builtin_neon_vclezq_v:
6537 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLE,
6538 ICmpInst::ICMP_SLE, "vclez");
6539 case NEON::BI__builtin_neon_vcgtz_v:
6540 case NEON::BI__builtin_neon_vcgtzq_v:
6541 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGT,
6542 ICmpInst::ICMP_SGT, "vcgtz");
6543 case NEON::BI__builtin_neon_vcltz_v:
6544 case NEON::BI__builtin_neon_vcltzq_v:
6545 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLT,
6546 ICmpInst::ICMP_SLT, "vcltz");
6547 case NEON::BI__builtin_neon_vclz_v:
6548 case NEON::BI__builtin_neon_vclzq_v:
6549 // We generate target-independent intrinsic, which needs a second argument
6550 // for whether or not clz of zero is undefined; on ARM it isn't.
6551 Ops.push_back(Builder.getInt1(getTarget().isCLZForZeroUndef()));
6552 break;
6553 case NEON::BI__builtin_neon_vcvt_f32_v:
6554 case NEON::BI__builtin_neon_vcvtq_f32_v:
6555 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6556 Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float32, false, Quad),
6557 HasLegalHalfType);
6558 return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
6559 : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
6560 case NEON::BI__builtin_neon_vcvt_f16_v:
6561 case NEON::BI__builtin_neon_vcvtq_f16_v:
6562 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6563 Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float16, false, Quad),
6564 HasLegalHalfType);
6565 return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
6566 : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
6567 case NEON::BI__builtin_neon_vcvt_n_f16_v:
6568 case NEON::BI__builtin_neon_vcvt_n_f32_v:
6569 case NEON::BI__builtin_neon_vcvt_n_f64_v:
6570 case NEON::BI__builtin_neon_vcvtq_n_f16_v:
6571 case NEON::BI__builtin_neon_vcvtq_n_f32_v:
6572 case NEON::BI__builtin_neon_vcvtq_n_f64_v: {
6573 llvm::Type *Tys[2] = { GetFloatNeonType(this, Type), Ty };
6574 Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic;
6575 Function *F = CGM.getIntrinsic(Int, Tys);
6576 return EmitNeonCall(F, Ops, "vcvt_n");
6577 }
6578 case NEON::BI__builtin_neon_vcvt_n_s16_v:
6579 case NEON::BI__builtin_neon_vcvt_n_s32_v:
6580 case NEON::BI__builtin_neon_vcvt_n_u16_v:
6581 case NEON::BI__builtin_neon_vcvt_n_u32_v:
6582 case NEON::BI__builtin_neon_vcvt_n_s64_v:
6583 case NEON::BI__builtin_neon_vcvt_n_u64_v:
6584 case NEON::BI__builtin_neon_vcvtq_n_s16_v:
6585 case NEON::BI__builtin_neon_vcvtq_n_s32_v:
6586 case NEON::BI__builtin_neon_vcvtq_n_u16_v:
6587 case NEON::BI__builtin_neon_vcvtq_n_u32_v:
6588 case NEON::BI__builtin_neon_vcvtq_n_s64_v:
6589 case NEON::BI__builtin_neon_vcvtq_n_u64_v: {
6590 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
6591 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
6592 return EmitNeonCall(F, Ops, "vcvt_n");
6593 }
6594 case NEON::BI__builtin_neon_vcvt_s32_v:
6595 case NEON::BI__builtin_neon_vcvt_u32_v:
6596 case NEON::BI__builtin_neon_vcvt_s64_v:
6597 case NEON::BI__builtin_neon_vcvt_u64_v:
6598 case NEON::BI__builtin_neon_vcvt_s16_v:
6599 case NEON::BI__builtin_neon_vcvt_u16_v:
6600 case NEON::BI__builtin_neon_vcvtq_s32_v:
6601 case NEON::BI__builtin_neon_vcvtq_u32_v:
6602 case NEON::BI__builtin_neon_vcvtq_s64_v:
6603 case NEON::BI__builtin_neon_vcvtq_u64_v:
6604 case NEON::BI__builtin_neon_vcvtq_s16_v:
6605 case NEON::BI__builtin_neon_vcvtq_u16_v: {
6606 Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type));
6607 return Usgn ? Builder.CreateFPToUI(Ops[0], Ty, "vcvt")
6608 : Builder.CreateFPToSI(Ops[0], Ty, "vcvt");
6609 }
6610 case NEON::BI__builtin_neon_vcvta_s16_v:
6611 case NEON::BI__builtin_neon_vcvta_s32_v:
6612 case NEON::BI__builtin_neon_vcvta_s64_v:
6613 case NEON::BI__builtin_neon_vcvta_u16_v:
6614 case NEON::BI__builtin_neon_vcvta_u32_v:
6615 case NEON::BI__builtin_neon_vcvta_u64_v:
6616 case NEON::BI__builtin_neon_vcvtaq_s16_v:
6617 case NEON::BI__builtin_neon_vcvtaq_s32_v:
6618 case NEON::BI__builtin_neon_vcvtaq_s64_v:
6619 case NEON::BI__builtin_neon_vcvtaq_u16_v:
6620 case NEON::BI__builtin_neon_vcvtaq_u32_v:
6621 case NEON::BI__builtin_neon_vcvtaq_u64_v:
6622 case NEON::BI__builtin_neon_vcvtn_s16_v:
6623 case NEON::BI__builtin_neon_vcvtn_s32_v:
6624 case NEON::BI__builtin_neon_vcvtn_s64_v:
6625 case NEON::BI__builtin_neon_vcvtn_u16_v:
6626 case NEON::BI__builtin_neon_vcvtn_u32_v:
6627 case NEON::BI__builtin_neon_vcvtn_u64_v:
6628 case NEON::BI__builtin_neon_vcvtnq_s16_v:
6629 case NEON::BI__builtin_neon_vcvtnq_s32_v:
6630 case NEON::BI__builtin_neon_vcvtnq_s64_v:
6631 case NEON::BI__builtin_neon_vcvtnq_u16_v:
6632 case NEON::BI__builtin_neon_vcvtnq_u32_v:
6633 case NEON::BI__builtin_neon_vcvtnq_u64_v:
6634 case NEON::BI__builtin_neon_vcvtp_s16_v:
6635 case NEON::BI__builtin_neon_vcvtp_s32_v:
6636 case NEON::BI__builtin_neon_vcvtp_s64_v:
6637 case NEON::BI__builtin_neon_vcvtp_u16_v:
6638 case NEON::BI__builtin_neon_vcvtp_u32_v:
6639 case NEON::BI__builtin_neon_vcvtp_u64_v:
6640 case NEON::BI__builtin_neon_vcvtpq_s16_v:
6641 case NEON::BI__builtin_neon_vcvtpq_s32_v:
6642 case NEON::BI__builtin_neon_vcvtpq_s64_v:
6643 case NEON::BI__builtin_neon_vcvtpq_u16_v:
6644 case NEON::BI__builtin_neon_vcvtpq_u32_v:
6645 case NEON::BI__builtin_neon_vcvtpq_u64_v:
6646 case NEON::BI__builtin_neon_vcvtm_s16_v:
6647 case NEON::BI__builtin_neon_vcvtm_s32_v:
6648 case NEON::BI__builtin_neon_vcvtm_s64_v:
6649 case NEON::BI__builtin_neon_vcvtm_u16_v:
6650 case NEON::BI__builtin_neon_vcvtm_u32_v:
6651 case NEON::BI__builtin_neon_vcvtm_u64_v:
6652 case NEON::BI__builtin_neon_vcvtmq_s16_v:
6653 case NEON::BI__builtin_neon_vcvtmq_s32_v:
6654 case NEON::BI__builtin_neon_vcvtmq_s64_v:
6655 case NEON::BI__builtin_neon_vcvtmq_u16_v:
6656 case NEON::BI__builtin_neon_vcvtmq_u32_v:
6657 case NEON::BI__builtin_neon_vcvtmq_u64_v: {
6658 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
6659 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint);
6660 }
6661 case NEON::BI__builtin_neon_vcvtx_f32_v: {
6662 llvm::Type *Tys[2] = { VTy->getTruncatedElementVectorType(VTy), Ty};
6663 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint);
6664
6665 }
6666 case NEON::BI__builtin_neon_vext_v:
6667 case NEON::BI__builtin_neon_vextq_v: {
6668 int CV = cast<ConstantInt>(Ops[2])->getSExtValue();
6669 SmallVector<int, 16> Indices;
6670 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
6671 Indices.push_back(i+CV);
6672
6673 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6674 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6675 return Builder.CreateShuffleVector(Ops[0], Ops[1], Indices, "vext");
6676 }
6677 case NEON::BI__builtin_neon_vfma_v:
6678 case NEON::BI__builtin_neon_vfmaq_v: {
6679 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6680 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6681 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
6682
6683 // NEON intrinsic puts accumulator first, unlike the LLVM fma.
6684 return emitCallMaybeConstrainedFPBuiltin(
6685 *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, Ty,
6686 {Ops[1], Ops[2], Ops[0]});
6687 }
6688 case NEON::BI__builtin_neon_vld1_v:
6689 case NEON::BI__builtin_neon_vld1q_v: {
6690 llvm::Type *Tys[] = {Ty, Int8PtrTy};
6691 Ops.push_back(getAlignmentValue32(PtrOp0));
6692 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "vld1");
6693 }
6694 case NEON::BI__builtin_neon_vld1_x2_v:
6695 case NEON::BI__builtin_neon_vld1q_x2_v:
6696 case NEON::BI__builtin_neon_vld1_x3_v:
6697 case NEON::BI__builtin_neon_vld1q_x3_v:
6698 case NEON::BI__builtin_neon_vld1_x4_v:
6699 case NEON::BI__builtin_neon_vld1q_x4_v: {
6700 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getElementType());
6701 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
6702 llvm::Type *Tys[2] = { VTy, PTy };
6703 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
6704 Ops[1] = Builder.CreateCall(F, Ops[1], "vld1xN");
6705 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
6706 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6707 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
6708 }
6709 case NEON::BI__builtin_neon_vld2_v:
6710 case NEON::BI__builtin_neon_vld2q_v:
6711 case NEON::BI__builtin_neon_vld3_v:
6712 case NEON::BI__builtin_neon_vld3q_v:
6713 case NEON::BI__builtin_neon_vld4_v:
6714 case NEON::BI__builtin_neon_vld4q_v:
6715 case NEON::BI__builtin_neon_vld2_dup_v:
6716 case NEON::BI__builtin_neon_vld2q_dup_v:
6717 case NEON::BI__builtin_neon_vld3_dup_v:
6718 case NEON::BI__builtin_neon_vld3q_dup_v:
6719 case NEON::BI__builtin_neon_vld4_dup_v:
6720 case NEON::BI__builtin_neon_vld4q_dup_v: {
6721 llvm::Type *Tys[] = {Ty, Int8PtrTy};
6722 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
6723 Value *Align = getAlignmentValue32(PtrOp1);
6724 Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, NameHint);
6725 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
6726 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6727 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
6728 }
6729 case NEON::BI__builtin_neon_vld1_dup_v:
6730 case NEON::BI__builtin_neon_vld1q_dup_v: {
6731 Value *V = UndefValue::get(Ty);
6732 Ty = llvm::PointerType::getUnqual(VTy->getElementType());
6733 PtrOp0 = Builder.CreateBitCast(PtrOp0, Ty);
6734 LoadInst *Ld = Builder.CreateLoad(PtrOp0);
6735 llvm::Constant *CI = ConstantInt::get(SizeTy, 0);
6736 Ops[0] = Builder.CreateInsertElement(V, Ld, CI);
6737 return EmitNeonSplat(Ops[0], CI);
6738 }
6739 case NEON::BI__builtin_neon_vld2_lane_v:
6740 case NEON::BI__builtin_neon_vld2q_lane_v:
6741 case NEON::BI__builtin_neon_vld3_lane_v:
6742 case NEON::BI__builtin_neon_vld3q_lane_v:
6743 case NEON::BI__builtin_neon_vld4_lane_v:
6744 case NEON::BI__builtin_neon_vld4q_lane_v: {
6745 llvm::Type *Tys[] = {Ty, Int8PtrTy};
6746 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
6747 for (unsigned I = 2; I < Ops.size() - 1; ++I)
6748 Ops[I] = Builder.CreateBitCast(Ops[I], Ty);
6749 Ops.push_back(getAlignmentValue32(PtrOp1));
6750 Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), NameHint);
6751 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
6752 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6753 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
6754 }
6755 case NEON::BI__builtin_neon_vmovl_v: {
6756 llvm::FixedVectorType *DTy =
6757 llvm::FixedVectorType::getTruncatedElementVectorType(VTy);
6758 Ops[0] = Builder.CreateBitCast(Ops[0], DTy);
6759 if (Usgn)
6760 return Builder.CreateZExt(Ops[0], Ty, "vmovl");
6761 return Builder.CreateSExt(Ops[0], Ty, "vmovl");
6762 }
6763 case NEON::BI__builtin_neon_vmovn_v: {
6764 llvm::FixedVectorType *QTy =
6765 llvm::FixedVectorType::getExtendedElementVectorType(VTy);
6766 Ops[0] = Builder.CreateBitCast(Ops[0], QTy);
6767 return Builder.CreateTrunc(Ops[0], Ty, "vmovn");
6768 }
6769 case NEON::BI__builtin_neon_vmull_v:
6770 // FIXME: the integer vmull operations could be emitted in terms of pure
6771 // LLVM IR (2 exts followed by a mul). Unfortunately LLVM has a habit of
6772 // hoisting the exts outside loops. Until global ISel comes along that can
6773 // see through such movement this leads to bad CodeGen. So we need an
6774 // intrinsic for now.
6775 Int = Usgn ? Intrinsic::arm_neon_vmullu : Intrinsic::arm_neon_vmulls;
6776 Int = Type.isPoly() ? (unsigned)Intrinsic::arm_neon_vmullp : Int;
6777 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull");
6778 case NEON::BI__builtin_neon_vpadal_v:
6779 case NEON::BI__builtin_neon_vpadalq_v: {
6780 // The source operand type has twice as many elements of half the size.
6781 unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits();
6782 llvm::Type *EltTy =
6783 llvm::IntegerType::get(getLLVMContext(), EltBits / 2);
6784 auto *NarrowTy =
6785 llvm::FixedVectorType::get(EltTy, VTy->getNumElements() * 2);
6786 llvm::Type *Tys[2] = { Ty, NarrowTy };
6787 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint);
6788 }
6789 case NEON::BI__builtin_neon_vpaddl_v:
6790 case NEON::BI__builtin_neon_vpaddlq_v: {
6791 // The source operand type has twice as many elements of half the size.
6792 unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits();
6793 llvm::Type *EltTy = llvm::IntegerType::get(getLLVMContext(), EltBits / 2);
6794 auto *NarrowTy =
6795 llvm::FixedVectorType::get(EltTy, VTy->getNumElements() * 2);
6796 llvm::Type *Tys[2] = { Ty, NarrowTy };
6797 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vpaddl");
6798 }
6799 case NEON::BI__builtin_neon_vqdmlal_v:
6800 case NEON::BI__builtin_neon_vqdmlsl_v: {
6801 SmallVector<Value *, 2> MulOps(Ops.begin() + 1, Ops.end());
6802 Ops[1] =
6803 EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), MulOps, "vqdmlal");
6804 Ops.resize(2);
6805 return EmitNeonCall(CGM.getIntrinsic(AltLLVMIntrinsic, Ty), Ops, NameHint);
6806 }
6807 case NEON::BI__builtin_neon_vqdmulhq_lane_v:
6808 case NEON::BI__builtin_neon_vqdmulh_lane_v:
6809 case NEON::BI__builtin_neon_vqrdmulhq_lane_v:
6810 case NEON::BI__builtin_neon_vqrdmulh_lane_v: {
6811 auto *RTy = cast<llvm::FixedVectorType>(Ty);
6812 if (BuiltinID == NEON::BI__builtin_neon_vqdmulhq_lane_v ||
6813 BuiltinID == NEON::BI__builtin_neon_vqrdmulhq_lane_v)
6814 RTy = llvm::FixedVectorType::get(RTy->getElementType(),
6815 RTy->getNumElements() * 2);
6816 llvm::Type *Tys[2] = {
6817 RTy, GetNeonType(this, NeonTypeFlags(Type.getEltType(), false,
6818 /*isQuad*/ false))};
6819 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint);
6820 }
6821 case NEON::BI__builtin_neon_vqdmulhq_laneq_v:
6822 case NEON::BI__builtin_neon_vqdmulh_laneq_v:
6823 case NEON::BI__builtin_neon_vqrdmulhq_laneq_v:
6824 case NEON::BI__builtin_neon_vqrdmulh_laneq_v: {
6825 llvm::Type *Tys[2] = {
6826 Ty, GetNeonType(this, NeonTypeFlags(Type.getEltType(), false,
6827 /*isQuad*/ true))};
6828 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint);
6829 }
6830 case NEON::BI__builtin_neon_vqshl_n_v:
6831 case NEON::BI__builtin_neon_vqshlq_n_v:
6832 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshl_n",
6833 1, false);
6834 case NEON::BI__builtin_neon_vqshlu_n_v:
6835 case NEON::BI__builtin_neon_vqshluq_n_v:
6836 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshlu_n",
6837 1, false);
6838 case NEON::BI__builtin_neon_vrecpe_v:
6839 case NEON::BI__builtin_neon_vrecpeq_v:
6840 case NEON::BI__builtin_neon_vrsqrte_v:
6841 case NEON::BI__builtin_neon_vrsqrteq_v:
6842 Int = Ty->isFPOrFPVectorTy() ? LLVMIntrinsic : AltLLVMIntrinsic;
6843 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint);
6844 case NEON::BI__builtin_neon_vrndi_v:
6845 case NEON::BI__builtin_neon_vrndiq_v:
6846 Int = Builder.getIsFPConstrained()
6847 ? Intrinsic::experimental_constrained_nearbyint
6848 : Intrinsic::nearbyint;
6849 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint);
6850 case NEON::BI__builtin_neon_vrshr_n_v:
6851 case NEON::BI__builtin_neon_vrshrq_n_v:
6852 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshr_n",
6853 1, true);
6854 case NEON::BI__builtin_neon_vsha512hq_v:
6855 case NEON::BI__builtin_neon_vsha512h2q_v:
6856 case NEON::BI__builtin_neon_vsha512su0q_v:
6857 case NEON::BI__builtin_neon_vsha512su1q_v: {
6858 Function *F = CGM.getIntrinsic(Int);
6859 return EmitNeonCall(F, Ops, "");
6860 }
6861 case NEON::BI__builtin_neon_vshl_n_v:
6862 case NEON::BI__builtin_neon_vshlq_n_v:
6863 Ops[1] = EmitNeonShiftVector(Ops[1], Ty, false);
6864 return Builder.CreateShl(Builder.CreateBitCast(Ops[0],Ty), Ops[1],
6865 "vshl_n");
6866 case NEON::BI__builtin_neon_vshll_n_v: {
6867 llvm::FixedVectorType *SrcTy =
6868 llvm::FixedVectorType::getTruncatedElementVectorType(VTy);
6869 Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
6870 if (Usgn)
6871 Ops[0] = Builder.CreateZExt(Ops[0], VTy);
6872 else
6873 Ops[0] = Builder.CreateSExt(Ops[0], VTy);
6874 Ops[1] = EmitNeonShiftVector(Ops[1], VTy, false);
6875 return Builder.CreateShl(Ops[0], Ops[1], "vshll_n");
6876 }
6877 case NEON::BI__builtin_neon_vshrn_n_v: {
6878 llvm::FixedVectorType *SrcTy =
6879 llvm::FixedVectorType::getExtendedElementVectorType(VTy);
6880 Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
6881 Ops[1] = EmitNeonShiftVector(Ops[1], SrcTy, false);
6882 if (Usgn)
6883 Ops[0] = Builder.CreateLShr(Ops[0], Ops[1]);
6884 else
6885 Ops[0] = Builder.CreateAShr(Ops[0], Ops[1]);
6886 return Builder.CreateTrunc(Ops[0], Ty, "vshrn_n");
6887 }
6888 case NEON::BI__builtin_neon_vshr_n_v:
6889 case NEON::BI__builtin_neon_vshrq_n_v:
6890 return EmitNeonRShiftImm(Ops[0], Ops[1], Ty, Usgn, "vshr_n");
6891 case NEON::BI__builtin_neon_vst1_v:
6892 case NEON::BI__builtin_neon_vst1q_v:
6893 case NEON::BI__builtin_neon_vst2_v:
6894 case NEON::BI__builtin_neon_vst2q_v:
6895 case NEON::BI__builtin_neon_vst3_v:
6896 case NEON::BI__builtin_neon_vst3q_v:
6897 case NEON::BI__builtin_neon_vst4_v:
6898 case NEON::BI__builtin_neon_vst4q_v:
6899 case NEON::BI__builtin_neon_vst2_lane_v:
6900 case NEON::BI__builtin_neon_vst2q_lane_v:
6901 case NEON::BI__builtin_neon_vst3_lane_v:
6902 case NEON::BI__builtin_neon_vst3q_lane_v:
6903 case NEON::BI__builtin_neon_vst4_lane_v:
6904 case NEON::BI__builtin_neon_vst4q_lane_v: {
6905 llvm::Type *Tys[] = {Int8PtrTy, Ty};
6906 Ops.push_back(getAlignmentValue32(PtrOp0));
6907 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "");
6908 }
6909 case NEON::BI__builtin_neon_vsm3partw1q_v:
6910 case NEON::BI__builtin_neon_vsm3partw2q_v:
6911 case NEON::BI__builtin_neon_vsm3ss1q_v:
6912 case NEON::BI__builtin_neon_vsm4ekeyq_v:
6913 case NEON::BI__builtin_neon_vsm4eq_v: {
6914 Function *F = CGM.getIntrinsic(Int);
6915 return EmitNeonCall(F, Ops, "");
6916 }
6917 case NEON::BI__builtin_neon_vsm3tt1aq_v:
6918 case NEON::BI__builtin_neon_vsm3tt1bq_v:
6919 case NEON::BI__builtin_neon_vsm3tt2aq_v:
6920 case NEON::BI__builtin_neon_vsm3tt2bq_v: {
6921 Function *F = CGM.getIntrinsic(Int);
6922 Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty);
6923 return EmitNeonCall(F, Ops, "");
6924 }
6925 case NEON::BI__builtin_neon_vst1_x2_v:
6926 case NEON::BI__builtin_neon_vst1q_x2_v:
6927 case NEON::BI__builtin_neon_vst1_x3_v:
6928 case NEON::BI__builtin_neon_vst1q_x3_v:
6929 case NEON::BI__builtin_neon_vst1_x4_v:
6930 case NEON::BI__builtin_neon_vst1q_x4_v: {
6931 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getElementType());
6932 // TODO: Currently in AArch32 mode the pointer operand comes first, whereas
6933 // in AArch64 it comes last. We may want to stick to one or another.
6934 if (Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_be ||
6935 Arch == llvm::Triple::aarch64_32) {
6936 llvm::Type *Tys[2] = { VTy, PTy };
6937 std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
6938 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "");
6939 }
6940 llvm::Type *Tys[2] = { PTy, VTy };
6941 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "");
6942 }
6943 case NEON::BI__builtin_neon_vsubhn_v: {
6944 llvm::FixedVectorType *SrcTy =
6945 llvm::FixedVectorType::getExtendedElementVectorType(VTy);
6946
6947 // %sum = add <4 x i32> %lhs, %rhs
6948 Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
6949 Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy);
6950 Ops[0] = Builder.CreateSub(Ops[0], Ops[1], "vsubhn");
6951
6952 // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16>
6953 Constant *ShiftAmt =
6954 ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2);
6955 Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vsubhn");
6956
6957 // %res = trunc <4 x i32> %high to <4 x i16>
6958 return Builder.CreateTrunc(Ops[0], VTy, "vsubhn");
6959 }
6960 case NEON::BI__builtin_neon_vtrn_v:
6961 case NEON::BI__builtin_neon_vtrnq_v: {
6962 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
6963 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6964 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
6965 Value *SV = nullptr;
6966
6967 for (unsigned vi = 0; vi != 2; ++vi) {
6968 SmallVector<int, 16> Indices;
6969 for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
6970 Indices.push_back(i+vi);
6971 Indices.push_back(i+e+vi);
6972 }
6973 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
6974 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn");
6975 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
6976 }
6977 return SV;
6978 }
6979 case NEON::BI__builtin_neon_vtst_v:
6980 case NEON::BI__builtin_neon_vtstq_v: {
6981 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6982 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6983 Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]);
6984 Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0],
6985 ConstantAggregateZero::get(Ty));
6986 return Builder.CreateSExt(Ops[0], Ty, "vtst");
6987 }
6988 case NEON::BI__builtin_neon_vuzp_v:
6989 case NEON::BI__builtin_neon_vuzpq_v: {
6990 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
6991 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6992 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
6993 Value *SV = nullptr;
6994
6995 for (unsigned vi = 0; vi != 2; ++vi) {
6996 SmallVector<int, 16> Indices;
6997 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
6998 Indices.push_back(2*i+vi);
6999
7000 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
7001 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp");
7002 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
7003 }
7004 return SV;
7005 }
7006 case NEON::BI__builtin_neon_vxarq_v: {
7007 Function *F = CGM.getIntrinsic(Int);
7008 Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty);
7009 return EmitNeonCall(F, Ops, "");
7010 }
7011 case NEON::BI__builtin_neon_vzip_v:
7012 case NEON::BI__builtin_neon_vzipq_v: {
7013 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
7014 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7015 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
7016 Value *SV = nullptr;
7017
7018 for (unsigned vi = 0; vi != 2; ++vi) {
7019 SmallVector<int, 16> Indices;
7020 for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
7021 Indices.push_back((i + vi*e) >> 1);
7022 Indices.push_back(((i + vi*e) >> 1)+e);
7023 }
7024 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
7025 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip");
7026 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
7027 }
7028 return SV;
7029 }
7030 case NEON::BI__builtin_neon_vdot_v:
7031 case NEON::BI__builtin_neon_vdotq_v: {
7032 auto *InputTy =
7033 llvm::FixedVectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8);
7034 llvm::Type *Tys[2] = { Ty, InputTy };
7035 Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic;
7036 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vdot");
7037 }
7038 case NEON::BI__builtin_neon_vfmlal_low_v:
7039 case NEON::BI__builtin_neon_vfmlalq_low_v: {
7040 auto *InputTy =
7041 llvm::FixedVectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
7042 llvm::Type *Tys[2] = { Ty, InputTy };
7043 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_low");
7044 }
7045 case NEON::BI__builtin_neon_vfmlsl_low_v:
7046 case NEON::BI__builtin_neon_vfmlslq_low_v: {
7047 auto *InputTy =
7048 llvm::FixedVectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
7049 llvm::Type *Tys[2] = { Ty, InputTy };
7050 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_low");
7051 }
7052 case NEON::BI__builtin_neon_vfmlal_high_v:
7053 case NEON::BI__builtin_neon_vfmlalq_high_v: {
7054 auto *InputTy =
7055 llvm::FixedVectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
7056 llvm::Type *Tys[2] = { Ty, InputTy };
7057 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_high");
7058 }
7059 case NEON::BI__builtin_neon_vfmlsl_high_v:
7060 case NEON::BI__builtin_neon_vfmlslq_high_v: {
7061 auto *InputTy =
7062 llvm::FixedVectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
7063 llvm::Type *Tys[2] = { Ty, InputTy };
7064 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_high");
7065 }
7066 case NEON::BI__builtin_neon_vmmlaq_v: {
7067 auto *InputTy =
7068 llvm::FixedVectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8);
7069 llvm::Type *Tys[2] = { Ty, InputTy };
7070 Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic;
7071 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmmla");
7072 }
7073 case NEON::BI__builtin_neon_vusmmlaq_v: {
7074 auto *InputTy =
7075 llvm::FixedVectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8);
7076 llvm::Type *Tys[2] = { Ty, InputTy };
7077 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vusmmla");
7078 }
7079 case NEON::BI__builtin_neon_vusdot_v:
7080 case NEON::BI__builtin_neon_vusdotq_v: {
7081 auto *InputTy =
7082 llvm::FixedVectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8);
7083 llvm::Type *Tys[2] = { Ty, InputTy };
7084 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vusdot");
7085 }
7086 case NEON::BI__builtin_neon_vbfdot_v:
7087 case NEON::BI__builtin_neon_vbfdotq_v: {
7088 llvm::Type *InputTy =
7089 llvm::FixedVectorType::get(BFloatTy, Ty->getPrimitiveSizeInBits() / 16);
7090 llvm::Type *Tys[2] = { Ty, InputTy };
7091 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vbfdot");
7092 }
7093 case NEON::BI__builtin_neon___a32_vcvt_bf16_v: {
7094 llvm::Type *Tys[1] = { Ty };
7095 Function *F = CGM.getIntrinsic(Int, Tys);
7096 return EmitNeonCall(F, Ops, "vcvtfp2bf");
7097 }
7098
7099 }
7100
7101 assert(Int && "Expected valid intrinsic number")(static_cast <bool> (Int && "Expected valid intrinsic number"
) ? void (0) : __assert_fail ("Int && \"Expected valid intrinsic number\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7101, __extension__ __PRETTY_FUNCTION__))
;
7102
7103 // Determine the type(s) of this overloaded AArch64 intrinsic.
7104 Function *F = LookupNeonLLVMIntrinsic(Int, Modifier, Ty, E);
7105
7106 Value *Result = EmitNeonCall(F, Ops, NameHint);
7107 llvm::Type *ResultType = ConvertType(E->getType());
7108 // AArch64 intrinsic one-element vector type cast to
7109 // scalar type expected by the builtin
7110 return Builder.CreateBitCast(Result, ResultType, NameHint);
7111}
7112
7113Value *CodeGenFunction::EmitAArch64CompareBuiltinExpr(
7114 Value *Op, llvm::Type *Ty, const CmpInst::Predicate Fp,
7115 const CmpInst::Predicate Ip, const Twine &Name) {
7116 llvm::Type *OTy = Op->getType();
7117
7118 // FIXME: this is utterly horrific. We should not be looking at previous
7119 // codegen context to find out what needs doing. Unfortunately TableGen
7120 // currently gives us exactly the same calls for vceqz_f32 and vceqz_s32
7121 // (etc).
7122 if (BitCastInst *BI = dyn_cast<BitCastInst>(Op))
7123 OTy = BI->getOperand(0)->getType();
7124
7125 Op = Builder.CreateBitCast(Op, OTy);
7126 if (OTy->getScalarType()->isFloatingPointTy()) {
7127 Op = Builder.CreateFCmp(Fp, Op, Constant::getNullValue(OTy));
7128 } else {
7129 Op = Builder.CreateICmp(Ip, Op, Constant::getNullValue(OTy));
7130 }
7131 return Builder.CreateSExt(Op, Ty, Name);
7132}
7133
7134static Value *packTBLDVectorList(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
7135 Value *ExtOp, Value *IndexOp,
7136 llvm::Type *ResTy, unsigned IntID,
7137 const char *Name) {
7138 SmallVector<Value *, 2> TblOps;
7139 if (ExtOp)
7140 TblOps.push_back(ExtOp);
7141
7142 // Build a vector containing sequential number like (0, 1, 2, ..., 15)
7143 SmallVector<int, 16> Indices;
7144 auto *TblTy = cast<llvm::FixedVectorType>(Ops[0]->getType());
7145 for (unsigned i = 0, e = TblTy->getNumElements(); i != e; ++i) {
7146 Indices.push_back(2*i);
7147 Indices.push_back(2*i+1);
7148 }
7149
7150 int PairPos = 0, End = Ops.size() - 1;
7151 while (PairPos < End) {
7152 TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos],
7153 Ops[PairPos+1], Indices,
7154 Name));
7155 PairPos += 2;
7156 }
7157
7158 // If there's an odd number of 64-bit lookup table, fill the high 64-bit
7159 // of the 128-bit lookup table with zero.
7160 if (PairPos == End) {
7161 Value *ZeroTbl = ConstantAggregateZero::get(TblTy);
7162 TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos],
7163 ZeroTbl, Indices, Name));
7164 }
7165
7166 Function *TblF;
7167 TblOps.push_back(IndexOp);
7168 TblF = CGF.CGM.getIntrinsic(IntID, ResTy);
7169
7170 return CGF.EmitNeonCall(TblF, TblOps, Name);
7171}
7172
7173Value *CodeGenFunction::GetValueForARMHint(unsigned BuiltinID) {
7174 unsigned Value;
7175 switch (BuiltinID) {
7176 default:
7177 return nullptr;
7178 case ARM::BI__builtin_arm_nop:
7179 Value = 0;
7180 break;
7181 case ARM::BI__builtin_arm_yield:
7182 case ARM::BI__yield:
7183 Value = 1;
7184 break;
7185 case ARM::BI__builtin_arm_wfe:
7186 case ARM::BI__wfe:
7187 Value = 2;
7188 break;
7189 case ARM::BI__builtin_arm_wfi:
7190 case ARM::BI__wfi:
7191 Value = 3;
7192 break;
7193 case ARM::BI__builtin_arm_sev:
7194 case ARM::BI__sev:
7195 Value = 4;
7196 break;
7197 case ARM::BI__builtin_arm_sevl:
7198 case ARM::BI__sevl:
7199 Value = 5;
7200 break;
7201 }
7202
7203 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_hint),
7204 llvm::ConstantInt::get(Int32Ty, Value));
7205}
7206
7207enum SpecialRegisterAccessKind {
7208 NormalRead,
7209 VolatileRead,
7210 Write,
7211};
7212
7213// Generates the IR for the read/write special register builtin,
7214// ValueType is the type of the value that is to be written or read,
7215// RegisterType is the type of the register being written to or read from.
7216static Value *EmitSpecialRegisterBuiltin(CodeGenFunction &CGF,
7217 const CallExpr *E,
7218 llvm::Type *RegisterType,
7219 llvm::Type *ValueType,
7220 SpecialRegisterAccessKind AccessKind,
7221 StringRef SysReg = "") {
7222 // write and register intrinsics only support 32 and 64 bit operations.
7223 assert((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64))(static_cast <bool> ((RegisterType->isIntegerTy(32) ||
RegisterType->isIntegerTy(64)) && "Unsupported size for register."
) ? void (0) : __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7224, __extension__ __PRETTY_FUNCTION__))
7224 && "Unsupported size for register.")(static_cast <bool> ((RegisterType->isIntegerTy(32) ||
RegisterType->isIntegerTy(64)) && "Unsupported size for register."
) ? void (0) : __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7224, __extension__ __PRETTY_FUNCTION__))
;
7225
7226 CodeGen::CGBuilderTy &Builder = CGF.Builder;
7227 CodeGen::CodeGenModule &CGM = CGF.CGM;
7228 LLVMContext &Context = CGM.getLLVMContext();
7229
7230 if (SysReg.empty()) {
7231 const Expr *SysRegStrExpr = E->getArg(0)->IgnoreParenCasts();
7232 SysReg = cast<clang::StringLiteral>(SysRegStrExpr)->getString();
7233 }
7234
7235 llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysReg) };
7236 llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops);
7237 llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName);
7238
7239 llvm::Type *Types[] = { RegisterType };
7240
7241 bool MixedTypes = RegisterType->isIntegerTy(64) && ValueType->isIntegerTy(32);
7242 assert(!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64))(static_cast <bool> (!(RegisterType->isIntegerTy(32)
&& ValueType->isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register"
) ? void (0) : __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7243, __extension__ __PRETTY_FUNCTION__))
7243 && "Can't fit 64-bit value in 32-bit register")(static_cast <bool> (!(RegisterType->isIntegerTy(32)
&& ValueType->isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register"
) ? void (0) : __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7243, __extension__ __PRETTY_FUNCTION__))
;
7244
7245 if (AccessKind != Write) {
7246 assert(AccessKind == NormalRead || AccessKind == VolatileRead)(static_cast <bool> (AccessKind == NormalRead || AccessKind
== VolatileRead) ? void (0) : __assert_fail ("AccessKind == NormalRead || AccessKind == VolatileRead"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7246, __extension__ __PRETTY_FUNCTION__))
;
7247 llvm::Function *F = CGM.getIntrinsic(
7248 AccessKind == VolatileRead ? llvm::Intrinsic::read_volatile_register
7249 : llvm::Intrinsic::read_register,
7250 Types);
7251 llvm::Value *Call = Builder.CreateCall(F, Metadata);
7252
7253 if (MixedTypes)
7254 // Read into 64 bit register and then truncate result to 32 bit.
7255 return Builder.CreateTrunc(Call, ValueType);
7256
7257 if (ValueType->isPointerTy())
7258 // Have i32/i64 result (Call) but want to return a VoidPtrTy (i8*).
7259 return Builder.CreateIntToPtr(Call, ValueType);
7260
7261 return Call;
7262 }
7263
7264 llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types);
7265 llvm::Value *ArgValue = CGF.EmitScalarExpr(E->getArg(1));
7266 if (MixedTypes) {
7267 // Extend 32 bit write value to 64 bit to pass to write.
7268 ArgValue = Builder.CreateZExt(ArgValue, RegisterType);
7269 return Builder.CreateCall(F, { Metadata, ArgValue });
7270 }
7271
7272 if (ValueType->isPointerTy()) {
7273 // Have VoidPtrTy ArgValue but want to return an i32/i64.
7274 ArgValue = Builder.CreatePtrToInt(ArgValue, RegisterType);
7275 return Builder.CreateCall(F, { Metadata, ArgValue });
7276 }
7277
7278 return Builder.CreateCall(F, { Metadata, ArgValue });
7279}
7280
7281/// Return true if BuiltinID is an overloaded Neon intrinsic with an extra
7282/// argument that specifies the vector type.
7283static bool HasExtraNeonArgument(unsigned BuiltinID) {
7284 switch (BuiltinID) {
7285 default: break;
7286 case NEON::BI__builtin_neon_vget_lane_i8:
7287 case NEON::BI__builtin_neon_vget_lane_i16:
7288 case NEON::BI__builtin_neon_vget_lane_bf16:
7289 case NEON::BI__builtin_neon_vget_lane_i32:
7290 case NEON::BI__builtin_neon_vget_lane_i64:
7291 case NEON::BI__builtin_neon_vget_lane_f32:
7292 case NEON::BI__builtin_neon_vgetq_lane_i8:
7293 case NEON::BI__builtin_neon_vgetq_lane_i16:
7294 case NEON::BI__builtin_neon_vgetq_lane_bf16:
7295 case NEON::BI__builtin_neon_vgetq_lane_i32:
7296 case NEON::BI__builtin_neon_vgetq_lane_i64:
7297 case NEON::BI__builtin_neon_vgetq_lane_f32:
7298 case NEON::BI__builtin_neon_vduph_lane_bf16:
7299 case NEON::BI__builtin_neon_vduph_laneq_bf16:
7300 case NEON::BI__builtin_neon_vset_lane_i8:
7301 case NEON::BI__builtin_neon_vset_lane_i16:
7302 case NEON::BI__builtin_neon_vset_lane_bf16:
7303 case NEON::BI__builtin_neon_vset_lane_i32:
7304 case NEON::BI__builtin_neon_vset_lane_i64:
7305 case NEON::BI__builtin_neon_vset_lane_f32:
7306 case NEON::BI__builtin_neon_vsetq_lane_i8:
7307 case NEON::BI__builtin_neon_vsetq_lane_i16:
7308 case NEON::BI__builtin_neon_vsetq_lane_bf16:
7309 case NEON::BI__builtin_neon_vsetq_lane_i32:
7310 case NEON::BI__builtin_neon_vsetq_lane_i64:
7311 case NEON::BI__builtin_neon_vsetq_lane_f32:
7312 case NEON::BI__builtin_neon_vsha1h_u32:
7313 case NEON::BI__builtin_neon_vsha1cq_u32:
7314 case NEON::BI__builtin_neon_vsha1pq_u32:
7315 case NEON::BI__builtin_neon_vsha1mq_u32:
7316 case NEON::BI__builtin_neon_vcvth_bf16_f32:
7317 case clang::ARM::BI_MoveToCoprocessor:
7318 case clang::ARM::BI_MoveToCoprocessor2:
7319 return false;
7320 }
7321 return true;
7322}
7323
7324Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
7325 const CallExpr *E,
7326 ReturnValueSlot ReturnValue,
7327 llvm::Triple::ArchType Arch) {
7328 if (auto Hint = GetValueForARMHint(BuiltinID))
7329 return Hint;
7330
7331 if (BuiltinID == ARM::BI__emit) {
7332 bool IsThumb = getTarget().getTriple().getArch() == llvm::Triple::thumb;
7333 llvm::FunctionType *FTy =
7334 llvm::FunctionType::get(VoidTy, /*Variadic=*/false);
7335
7336 Expr::EvalResult Result;
7337 if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext()))
7338 llvm_unreachable("Sema will ensure that the parameter is constant")::llvm::llvm_unreachable_internal("Sema will ensure that the parameter is constant"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7338)
;
7339
7340 llvm::APSInt Value = Result.Val.getInt();
7341 uint64_t ZExtValue = Value.zextOrTrunc(IsThumb ? 16 : 32).getZExtValue();
7342
7343 llvm::InlineAsm *Emit =
7344 IsThumb ? InlineAsm::get(FTy, ".inst.n 0x" + utohexstr(ZExtValue), "",
7345 /*hasSideEffects=*/true)
7346 : InlineAsm::get(FTy, ".inst 0x" + utohexstr(ZExtValue), "",
7347 /*hasSideEffects=*/true);
7348
7349 return Builder.CreateCall(Emit);
7350 }
7351
7352 if (BuiltinID == ARM::BI__builtin_arm_dbg) {
7353 Value *Option = EmitScalarExpr(E->getArg(0));
7354 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_dbg), Option);
7355 }
7356
7357 if (BuiltinID == ARM::BI__builtin_arm_prefetch) {
7358 Value *Address = EmitScalarExpr(E->getArg(0));
7359 Value *RW = EmitScalarExpr(E->getArg(1));
7360 Value *IsData = EmitScalarExpr(E->getArg(2));
7361
7362 // Locality is not supported on ARM target
7363 Value *Locality = llvm::ConstantInt::get(Int32Ty, 3);
7364
7365 Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType());
7366 return Builder.CreateCall(F, {Address, RW, Locality, IsData});
7367 }
7368
7369 if (BuiltinID == ARM::BI__builtin_arm_rbit) {
7370 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
7371 return Builder.CreateCall(
7372 CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
7373 }
7374
7375 if (BuiltinID == ARM::BI__builtin_arm_cls) {
7376 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
7377 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_cls), Arg, "cls");
7378 }
7379 if (BuiltinID == ARM::BI__builtin_arm_cls64) {
7380 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
7381 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_cls64), Arg,
7382 "cls");
7383 }
7384
7385 if (BuiltinID == ARM::BI__clear_cache) {
7386 assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments")(static_cast <bool> (E->getNumArgs() == 2 &&
"__clear_cache takes 2 arguments") ? void (0) : __assert_fail
("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7386, __extension__ __PRETTY_FUNCTION__))
;
7387 const FunctionDecl *FD = E->getDirectCallee();
7388 Value *Ops[2];
7389 for (unsigned i = 0; i < 2; i++)
7390 Ops[i] = EmitScalarExpr(E->getArg(i));
7391 llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType());
7392 llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty);
7393 StringRef Name = FD->getName();
7394 return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops);
7395 }
7396
7397 if (BuiltinID == ARM::BI__builtin_arm_mcrr ||
7398 BuiltinID == ARM::BI__builtin_arm_mcrr2) {
7399 Function *F;
7400
7401 switch (BuiltinID) {
7402 default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7402)
;
7403 case ARM::BI__builtin_arm_mcrr:
7404 F = CGM.getIntrinsic(Intrinsic::arm_mcrr);
7405 break;
7406 case ARM::BI__builtin_arm_mcrr2:
7407 F = CGM.getIntrinsic(Intrinsic::arm_mcrr2);
7408 break;
7409 }
7410
7411 // MCRR{2} instruction has 5 operands but
7412 // the intrinsic has 4 because Rt and Rt2
7413 // are represented as a single unsigned 64
7414 // bit integer in the intrinsic definition
7415 // but internally it's represented as 2 32
7416 // bit integers.
7417
7418 Value *Coproc = EmitScalarExpr(E->getArg(0));
7419 Value *Opc1 = EmitScalarExpr(E->getArg(1));
7420 Value *RtAndRt2 = EmitScalarExpr(E->getArg(2));
7421 Value *CRm = EmitScalarExpr(E->getArg(3));
7422
7423 Value *C1 = llvm::ConstantInt::get(Int64Ty, 32);
7424 Value *Rt = Builder.CreateTruncOrBitCast(RtAndRt2, Int32Ty);
7425 Value *Rt2 = Builder.CreateLShr(RtAndRt2, C1);
7426 Rt2 = Builder.CreateTruncOrBitCast(Rt2, Int32Ty);
7427
7428 return Builder.CreateCall(F, {Coproc, Opc1, Rt, Rt2, CRm});
7429 }
7430
7431 if (BuiltinID == ARM::BI__builtin_arm_mrrc ||
7432 BuiltinID == ARM::BI__builtin_arm_mrrc2) {
7433 Function *F;
7434
7435 switch (BuiltinID) {
7436 default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7436)
;
7437 case ARM::BI__builtin_arm_mrrc:
7438 F = CGM.getIntrinsic(Intrinsic::arm_mrrc);
7439 break;
7440 case ARM::BI__builtin_arm_mrrc2:
7441 F = CGM.getIntrinsic(Intrinsic::arm_mrrc2);
7442 break;
7443 }
7444
7445 Value *Coproc = EmitScalarExpr(E->getArg(0));
7446 Value *Opc1 = EmitScalarExpr(E->getArg(1));
7447 Value *CRm = EmitScalarExpr(E->getArg(2));
7448 Value *RtAndRt2 = Builder.CreateCall(F, {Coproc, Opc1, CRm});
7449
7450 // Returns an unsigned 64 bit integer, represented
7451 // as two 32 bit integers.
7452
7453 Value *Rt = Builder.CreateExtractValue(RtAndRt2, 1);
7454 Value *Rt1 = Builder.CreateExtractValue(RtAndRt2, 0);
7455 Rt = Builder.CreateZExt(Rt, Int64Ty);
7456 Rt1 = Builder.CreateZExt(Rt1, Int64Ty);
7457
7458 Value *ShiftCast = llvm::ConstantInt::get(Int64Ty, 32);
7459 RtAndRt2 = Builder.CreateShl(Rt, ShiftCast, "shl", true);
7460 RtAndRt2 = Builder.CreateOr(RtAndRt2, Rt1);
7461
7462 return Builder.CreateBitCast(RtAndRt2, ConvertType(E->getType()));
7463 }
7464
7465 if (BuiltinID == ARM::BI__builtin_arm_ldrexd ||
7466 ((BuiltinID == ARM::BI__builtin_arm_ldrex ||
7467 BuiltinID == ARM::BI__builtin_arm_ldaex) &&
7468 getContext().getTypeSize(E->getType()) == 64) ||
7469 BuiltinID == ARM::BI__ldrexd) {
7470 Function *F;
7471
7472 switch (BuiltinID) {
7473 default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7473)
;
7474 case ARM::BI__builtin_arm_ldaex:
7475 F = CGM.getIntrinsic(Intrinsic::arm_ldaexd);
7476 break;
7477 case ARM::BI__builtin_arm_ldrexd:
7478 case ARM::BI__builtin_arm_ldrex:
7479 case ARM::BI__ldrexd:
7480 F = CGM.getIntrinsic(Intrinsic::arm_ldrexd);
7481 break;
7482 }
7483
7484 Value *LdPtr = EmitScalarExpr(E->getArg(0));
7485 Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy),
7486 "ldrexd");
7487
7488 Value *Val0 = Builder.CreateExtractValue(Val, 1);
7489 Value *Val1 = Builder.CreateExtractValue(Val, 0);
7490 Val0 = Builder.CreateZExt(Val0, Int64Ty);
7491 Val1 = Builder.CreateZExt(Val1, Int64Ty);
7492
7493 Value *ShiftCst = llvm::ConstantInt::get(Int64Ty, 32);
7494 Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */);
7495 Val = Builder.CreateOr(Val, Val1);
7496 return Builder.CreateBitCast(Val, ConvertType(E->getType()));
7497 }
7498
7499 if (BuiltinID == ARM::BI__builtin_arm_ldrex ||
7500 BuiltinID == ARM::BI__builtin_arm_ldaex) {
7501 Value *LoadAddr = EmitScalarExpr(E->getArg(0));
7502
7503 QualType Ty = E->getType();
7504 llvm::Type *RealResTy = ConvertType(Ty);
7505 llvm::Type *PtrTy = llvm::IntegerType::get(
7506 getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo();
7507 LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy);
7508
7509 Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_ldaex
7510 ? Intrinsic::arm_ldaex
7511 : Intrinsic::arm_ldrex,
7512 PtrTy);
7513 Value *Val = Builder.CreateCall(F, LoadAddr, "ldrex");
7514
7515 if (RealResTy->isPointerTy())
7516 return Builder.CreateIntToPtr(Val, RealResTy);
7517 else {
7518 llvm::Type *IntResTy = llvm::IntegerType::get(
7519 getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy));
7520 Val = Builder.CreateTruncOrBitCast(Val, IntResTy);
7521 return Builder.CreateBitCast(Val, RealResTy);
7522 }
7523 }
7524
7525 if (BuiltinID == ARM::BI__builtin_arm_strexd ||
7526 ((BuiltinID == ARM::BI__builtin_arm_stlex ||
7527 BuiltinID == ARM::BI__builtin_arm_strex) &&
7528 getContext().getTypeSize(E->getArg(0)->getType()) == 64)) {
7529 Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex
7530 ? Intrinsic::arm_stlexd
7531 : Intrinsic::arm_strexd);
7532 llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty);
7533
7534 Address Tmp = CreateMemTemp(E->getArg(0)->getType());
7535 Value *Val = EmitScalarExpr(E->getArg(0));
7536 Builder.CreateStore(Val, Tmp);
7537
7538 Address LdPtr = Builder.CreateBitCast(Tmp,llvm::PointerType::getUnqual(STy));
7539 Val = Builder.CreateLoad(LdPtr);
7540
7541 Value *Arg0 = Builder.CreateExtractValue(Val, 0);
7542 Value *Arg1 = Builder.CreateExtractValue(Val, 1);
7543 Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), Int8PtrTy);
7544 return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "strexd");
7545 }
7546
7547 if (BuiltinID == ARM::BI__builtin_arm_strex ||
7548 BuiltinID == ARM::BI__builtin_arm_stlex) {
7549 Value *StoreVal = EmitScalarExpr(E->getArg(0));
7550 Value *StoreAddr = EmitScalarExpr(E->getArg(1));
7551
7552 QualType Ty = E->getArg(0)->getType();
7553 llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(),
7554 getContext().getTypeSize(Ty));
7555 StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo());
7556
7557 if (StoreVal->getType()->isPointerTy())
7558 StoreVal = Builder.CreatePtrToInt(StoreVal, Int32Ty);
7559 else {
7560 llvm::Type *IntTy = llvm::IntegerType::get(
7561 getLLVMContext(),
7562 CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType()));
7563 StoreVal = Builder.CreateBitCast(StoreVal, IntTy);
7564 StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int32Ty);
7565 }
7566
7567 Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex
7568 ? Intrinsic::arm_stlex
7569 : Intrinsic::arm_strex,
7570 StoreAddr->getType());
7571 return Builder.CreateCall(F, {StoreVal, StoreAddr}, "strex");
7572 }
7573
7574 if (BuiltinID == ARM::BI__builtin_arm_clrex) {
7575 Function *F = CGM.getIntrinsic(Intrinsic::arm_clrex);
7576 return Builder.CreateCall(F);
7577 }
7578
7579 // CRC32
7580 Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic;
7581 switch (BuiltinID) {
7582 case ARM::BI__builtin_arm_crc32b:
7583 CRCIntrinsicID = Intrinsic::arm_crc32b; break;
7584 case ARM::BI__builtin_arm_crc32cb:
7585 CRCIntrinsicID = Intrinsic::arm_crc32cb; break;
7586 case ARM::BI__builtin_arm_crc32h:
7587 CRCIntrinsicID = Intrinsic::arm_crc32h; break;
7588 case ARM::BI__builtin_arm_crc32ch:
7589 CRCIntrinsicID = Intrinsic::arm_crc32ch; break;
7590 case ARM::BI__builtin_arm_crc32w:
7591 case ARM::BI__builtin_arm_crc32d:
7592 CRCIntrinsicID = Intrinsic::arm_crc32w; break;
7593 case ARM::BI__builtin_arm_crc32cw:
7594 case ARM::BI__builtin_arm_crc32cd:
7595 CRCIntrinsicID = Intrinsic::arm_crc32cw; break;
7596 }
7597
7598 if (CRCIntrinsicID != Intrinsic::not_intrinsic) {
7599 Value *Arg0 = EmitScalarExpr(E->getArg(0));
7600 Value *Arg1 = EmitScalarExpr(E->getArg(1));
7601
7602 // crc32{c,}d intrinsics are implemnted as two calls to crc32{c,}w
7603 // intrinsics, hence we need different codegen for these cases.
7604 if (BuiltinID == ARM::BI__builtin_arm_crc32d ||
7605 BuiltinID == ARM::BI__builtin_arm_crc32cd) {
7606 Value *C1 = llvm::ConstantInt::get(Int64Ty, 32);
7607 Value *Arg1a = Builder.CreateTruncOrBitCast(Arg1, Int32Ty);
7608 Value *Arg1b = Builder.CreateLShr(Arg1, C1);
7609 Arg1b = Builder.CreateTruncOrBitCast(Arg1b, Int32Ty);
7610
7611 Function *F = CGM.getIntrinsic(CRCIntrinsicID);
7612 Value *Res = Builder.CreateCall(F, {Arg0, Arg1a});
7613 return Builder.CreateCall(F, {Res, Arg1b});
7614 } else {
7615 Arg1 = Builder.CreateZExtOrBitCast(Arg1, Int32Ty);
7616
7617 Function *F = CGM.getIntrinsic(CRCIntrinsicID);
7618 return Builder.CreateCall(F, {Arg0, Arg1});
7619 }
7620 }
7621
7622 if (BuiltinID == ARM::BI__builtin_arm_rsr ||
7623 BuiltinID == ARM::BI__builtin_arm_rsr64 ||
7624 BuiltinID == ARM::BI__builtin_arm_rsrp ||
7625 BuiltinID == ARM::BI__builtin_arm_wsr ||
7626 BuiltinID == ARM::BI__builtin_arm_wsr64 ||
7627 BuiltinID == ARM::BI__builtin_arm_wsrp) {
7628
7629 SpecialRegisterAccessKind AccessKind = Write;
7630 if (BuiltinID == ARM::BI__builtin_arm_rsr ||
7631 BuiltinID == ARM::BI__builtin_arm_rsr64 ||
7632 BuiltinID == ARM::BI__builtin_arm_rsrp)
7633 AccessKind = VolatileRead;
7634
7635 bool IsPointerBuiltin = BuiltinID == ARM::BI__builtin_arm_rsrp ||
7636 BuiltinID == ARM::BI__builtin_arm_wsrp;
7637
7638 bool Is64Bit = BuiltinID == ARM::BI__builtin_arm_rsr64 ||
7639 BuiltinID == ARM::BI__builtin_arm_wsr64;
7640
7641 llvm::Type *ValueType;
7642 llvm::Type *RegisterType;
7643 if (IsPointerBuiltin) {
7644 ValueType = VoidPtrTy;
7645 RegisterType = Int32Ty;
7646 } else if (Is64Bit) {
7647 ValueType = RegisterType = Int64Ty;
7648 } else {
7649 ValueType = RegisterType = Int32Ty;
7650 }
7651
7652 return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType,
7653 AccessKind);
7654 }
7655
7656 // Handle MSVC intrinsics before argument evaluation to prevent double
7657 // evaluation.
7658 if (Optional<MSVCIntrin> MsvcIntId = translateArmToMsvcIntrin(BuiltinID))
7659 return EmitMSVCBuiltinExpr(*MsvcIntId, E);
7660
7661 // Deal with MVE builtins
7662 if (Value *Result = EmitARMMVEBuiltinExpr(BuiltinID, E, ReturnValue, Arch))
7663 return Result;
7664 // Handle CDE builtins
7665 if (Value *Result = EmitARMCDEBuiltinExpr(BuiltinID, E, ReturnValue, Arch))
7666 return Result;
7667
7668 // Find out if any arguments are required to be integer constant
7669 // expressions.
7670 unsigned ICEArguments = 0;
7671 ASTContext::GetBuiltinTypeError Error;
7672 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
7673 assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None &&
"Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7673, __extension__ __PRETTY_FUNCTION__))
;
7674
7675 auto getAlignmentValue32 = [&](Address addr) -> Value* {
7676 return Builder.getInt32(addr.getAlignment().getQuantity());
7677 };
7678
7679 Address PtrOp0 = Address::invalid();
7680 Address PtrOp1 = Address::invalid();
7681 SmallVector<Value*, 4> Ops;
7682 bool HasExtraArg = HasExtraNeonArgument(BuiltinID);
7683 unsigned NumArgs = E->getNumArgs() - (HasExtraArg ? 1 : 0);
7684 for (unsigned i = 0, e = NumArgs; i != e; i++) {
7685 if (i == 0) {
7686 switch (BuiltinID) {
7687 case NEON::BI__builtin_neon_vld1_v:
7688 case NEON::BI__builtin_neon_vld1q_v:
7689 case NEON::BI__builtin_neon_vld1q_lane_v:
7690 case NEON::BI__builtin_neon_vld1_lane_v:
7691 case NEON::BI__builtin_neon_vld1_dup_v:
7692 case NEON::BI__builtin_neon_vld1q_dup_v:
7693 case NEON::BI__builtin_neon_vst1_v:
7694 case NEON::BI__builtin_neon_vst1q_v:
7695 case NEON::BI__builtin_neon_vst1q_lane_v:
7696 case NEON::BI__builtin_neon_vst1_lane_v:
7697 case NEON::BI__builtin_neon_vst2_v:
7698 case NEON::BI__builtin_neon_vst2q_v:
7699 case NEON::BI__builtin_neon_vst2_lane_v:
7700 case NEON::BI__builtin_neon_vst2q_lane_v:
7701 case NEON::BI__builtin_neon_vst3_v:
7702 case NEON::BI__builtin_neon_vst3q_v:
7703 case NEON::BI__builtin_neon_vst3_lane_v:
7704 case NEON::BI__builtin_neon_vst3q_lane_v:
7705 case NEON::BI__builtin_neon_vst4_v:
7706 case NEON::BI__builtin_neon_vst4q_v:
7707 case NEON::BI__builtin_neon_vst4_lane_v:
7708 case NEON::BI__builtin_neon_vst4q_lane_v:
7709 // Get the alignment for the argument in addition to the value;
7710 // we'll use it later.
7711 PtrOp0 = EmitPointerWithAlignment(E->getArg(0));
7712 Ops.push_back(PtrOp0.getPointer());
7713 continue;
7714 }
7715 }
7716 if (i == 1) {
7717 switch (BuiltinID) {
7718 case NEON::BI__builtin_neon_vld2_v:
7719 case NEON::BI__builtin_neon_vld2q_v:
7720 case NEON::BI__builtin_neon_vld3_v:
7721 case NEON::BI__builtin_neon_vld3q_v:
7722 case NEON::BI__builtin_neon_vld4_v:
7723 case NEON::BI__builtin_neon_vld4q_v:
7724 case NEON::BI__builtin_neon_vld2_lane_v:
7725 case NEON::BI__builtin_neon_vld2q_lane_v:
7726 case NEON::BI__builtin_neon_vld3_lane_v:
7727 case NEON::BI__builtin_neon_vld3q_lane_v:
7728 case NEON::BI__builtin_neon_vld4_lane_v:
7729 case NEON::BI__builtin_neon_vld4q_lane_v:
7730 case NEON::BI__builtin_neon_vld2_dup_v:
7731 case NEON::BI__builtin_neon_vld2q_dup_v:
7732 case NEON::BI__builtin_neon_vld3_dup_v:
7733 case NEON::BI__builtin_neon_vld3q_dup_v:
7734 case NEON::BI__builtin_neon_vld4_dup_v:
7735 case NEON::BI__builtin_neon_vld4q_dup_v:
7736 // Get the alignment for the argument in addition to the value;
7737 // we'll use it later.
7738 PtrOp1 = EmitPointerWithAlignment(E->getArg(1));
7739 Ops.push_back(PtrOp1.getPointer());
7740 continue;
7741 }
7742 }
7743
7744 if ((ICEArguments & (1 << i)) == 0) {
7745 Ops.push_back(EmitScalarExpr(E->getArg(i)));
7746 } else {
7747 // If this is required to be a constant, constant fold it so that we know
7748 // that the generated intrinsic gets a ConstantInt.
7749 Ops.push_back(llvm::ConstantInt::get(
7750 getLLVMContext(),
7751 *E->getArg(i)->getIntegerConstantExpr(getContext())));
7752 }
7753 }
7754
7755 switch (BuiltinID) {
7756 default: break;
7757
7758 case NEON::BI__builtin_neon_vget_lane_i8:
7759 case NEON::BI__builtin_neon_vget_lane_i16:
7760 case NEON::BI__builtin_neon_vget_lane_i32:
7761 case NEON::BI__builtin_neon_vget_lane_i64:
7762 case NEON::BI__builtin_neon_vget_lane_bf16:
7763 case NEON::BI__builtin_neon_vget_lane_f32:
7764 case NEON::BI__builtin_neon_vgetq_lane_i8:
7765 case NEON::BI__builtin_neon_vgetq_lane_i16:
7766 case NEON::BI__builtin_neon_vgetq_lane_i32:
7767 case NEON::BI__builtin_neon_vgetq_lane_i64:
7768 case NEON::BI__builtin_neon_vgetq_lane_bf16:
7769 case NEON::BI__builtin_neon_vgetq_lane_f32:
7770 case NEON::BI__builtin_neon_vduph_lane_bf16:
7771 case NEON::BI__builtin_neon_vduph_laneq_bf16:
7772 return Builder.CreateExtractElement(Ops[0], Ops[1], "vget_lane");
7773
7774 case NEON::BI__builtin_neon_vrndns_f32: {
7775 Value *Arg = EmitScalarExpr(E->getArg(0));
7776 llvm::Type *Tys[] = {Arg->getType()};
7777 Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vrintn, Tys);
7778 return Builder.CreateCall(F, {Arg}, "vrndn"); }
7779
7780 case NEON::BI__builtin_neon_vset_lane_i8:
7781 case NEON::BI__builtin_neon_vset_lane_i16:
7782 case NEON::BI__builtin_neon_vset_lane_i32:
7783 case NEON::BI__builtin_neon_vset_lane_i64:
7784 case NEON::BI__builtin_neon_vset_lane_bf16:
7785 case NEON::BI__builtin_neon_vset_lane_f32:
7786 case NEON::BI__builtin_neon_vsetq_lane_i8:
7787 case NEON::BI__builtin_neon_vsetq_lane_i16:
7788 case NEON::BI__builtin_neon_vsetq_lane_i32:
7789 case NEON::BI__builtin_neon_vsetq_lane_i64:
7790 case NEON::BI__builtin_neon_vsetq_lane_bf16:
7791 case NEON::BI__builtin_neon_vsetq_lane_f32:
7792 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
7793
7794 case NEON::BI__builtin_neon_vsha1h_u32:
7795 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1h), Ops,
7796 "vsha1h");
7797 case NEON::BI__builtin_neon_vsha1cq_u32:
7798 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1c), Ops,
7799 "vsha1h");
7800 case NEON::BI__builtin_neon_vsha1pq_u32:
7801 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1p), Ops,
7802 "vsha1h");
7803 case NEON::BI__builtin_neon_vsha1mq_u32:
7804 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1m), Ops,
7805 "vsha1h");
7806
7807 case NEON::BI__builtin_neon_vcvth_bf16_f32: {
7808 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vcvtbfp2bf), Ops,
7809 "vcvtbfp2bf");
7810 }
7811
7812 // The ARM _MoveToCoprocessor builtins put the input register value as
7813 // the first argument, but the LLVM intrinsic expects it as the third one.
7814 case ARM::BI_MoveToCoprocessor:
7815 case ARM::BI_MoveToCoprocessor2: {
7816 Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI_MoveToCoprocessor ?
7817 Intrinsic::arm_mcr : Intrinsic::arm_mcr2);
7818 return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0],
7819 Ops[3], Ops[4], Ops[5]});
7820 }
7821 }
7822
7823 // Get the last argument, which specifies the vector type.
7824 assert(HasExtraArg)(static_cast <bool> (HasExtraArg) ? void (0) : __assert_fail
("HasExtraArg", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 7824, __extension__ __PRETTY_FUNCTION__))
;
7825 const Expr *Arg = E->getArg(E->getNumArgs()-1);
7826 Optional<llvm::APSInt> Result = Arg->getIntegerConstantExpr(getContext());
7827 if (!Result)
7828 return nullptr;
7829
7830 if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f ||
7831 BuiltinID == ARM::BI__builtin_arm_vcvtr_d) {
7832 // Determine the overloaded type of this builtin.
7833 llvm::Type *Ty;
7834 if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f)
7835 Ty = FloatTy;
7836 else
7837 Ty = DoubleTy;
7838
7839 // Determine whether this is an unsigned conversion or not.
7840 bool usgn = Result->getZExtValue() == 1;
7841 unsigned Int = usgn ? Intrinsic::arm_vcvtru : Intrinsic::arm_vcvtr;
7842
7843 // Call the appropriate intrinsic.
7844 Function *F = CGM.getIntrinsic(Int, Ty);
7845 return Builder.CreateCall(F, Ops, "vcvtr");
7846 }
7847
7848 // Determine the type of this overloaded NEON intrinsic.
7849 NeonTypeFlags Type = Result->getZExtValue();
7850 bool usgn = Type.isUnsigned();
7851 bool rightShift = false;
7852
7853 llvm::FixedVectorType *VTy =
7854 GetNeonType(this, Type, getTarget().hasLegalHalfType(), false,
7855 getTarget().hasBFloat16Type());
7856 llvm::Type *Ty = VTy;
7857 if (!Ty)
7858 return nullptr;
7859
7860 // Many NEON builtins have identical semantics and uses in ARM and
7861 // AArch64. Emit these in a single function.
7862 auto IntrinsicMap = makeArrayRef(ARMSIMDIntrinsicMap);
7863 const ARMVectorIntrinsicInfo *Builtin = findARMVectorIntrinsicInMap(
7864 IntrinsicMap, BuiltinID, NEONSIMDIntrinsicsProvenSorted);
7865 if (Builtin)
7866 return EmitCommonNeonBuiltinExpr(
7867 Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic,
7868 Builtin->NameHint, Builtin->TypeModifier, E, Ops, PtrOp0, PtrOp1, Arch);
7869
7870 unsigned Int;
7871 switch (BuiltinID) {
7872 default: return nullptr;
7873 case NEON::BI__builtin_neon_vld1q_lane_v:
7874 // Handle 64-bit integer elements as a special case. Use shuffles of
7875 // one-element vectors to avoid poor code for i64 in the backend.
7876 if (VTy->getElementType()->isIntegerTy(64)) {
7877 // Extract the other lane.
7878 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7879 int Lane = cast<ConstantInt>(Ops[2])->getZExtValue();
7880 Value *SV = llvm::ConstantVector::get(ConstantInt::get(Int32Ty, 1-Lane));
7881 Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV);
7882 // Load the value as a one-element vector.
7883 Ty = llvm::FixedVectorType::get(VTy->getElementType(), 1);
7884 llvm::Type *Tys[] = {Ty, Int8PtrTy};
7885 Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vld1, Tys);
7886 Value *Align = getAlignmentValue32(PtrOp0);
7887 Value *Ld = Builder.CreateCall(F, {Ops[0], Align});
7888 // Combine them.
7889 int Indices[] = {1 - Lane, Lane};
7890 return Builder.CreateShuffleVector(Ops[1], Ld, Indices, "vld1q_lane");
7891 }
7892 LLVM_FALLTHROUGH[[gnu::fallthrough]];
7893 case NEON::BI__builtin_neon_vld1_lane_v: {
7894 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7895 PtrOp0 = Builder.CreateElementBitCast(PtrOp0, VTy->getElementType());
7896 Value *Ld = Builder.CreateLoad(PtrOp0);
7897 return Builder.CreateInsertElement(Ops[1], Ld, Ops[2], "vld1_lane");
7898 }
7899 case NEON::BI__builtin_neon_vqrshrn_n_v:
7900 Int =
7901 usgn ? Intrinsic::arm_neon_vqrshiftnu : Intrinsic::arm_neon_vqrshiftns;
7902 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n",
7903 1, true);
7904 case NEON::BI__builtin_neon_vqrshrun_n_v:
7905 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqrshiftnsu, Ty),
7906 Ops, "vqrshrun_n", 1, true);
7907 case NEON::BI__builtin_neon_vqshrn_n_v:
7908 Int = usgn ? Intrinsic::arm_neon_vqshiftnu : Intrinsic::arm_neon_vqshiftns;
7909 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n",
7910 1, true);
7911 case NEON::BI__builtin_neon_vqshrun_n_v:
7912 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqshiftnsu, Ty),
7913 Ops, "vqshrun_n", 1, true);
7914 case NEON::BI__builtin_neon_vrecpe_v:
7915 case NEON::BI__builtin_neon_vrecpeq_v:
7916 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrecpe, Ty),
7917 Ops, "vrecpe");
7918 case NEON::BI__builtin_neon_vrshrn_n_v:
7919 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrshiftn, Ty),
7920 Ops, "vrshrn_n", 1, true);
7921 case NEON::BI__builtin_neon_vrsra_n_v:
7922 case NEON::BI__builtin_neon_vrsraq_n_v:
7923 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
7924 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7925 Ops[2] = EmitNeonShiftVector(Ops[2], Ty, true);
7926 Int = usgn ? Intrinsic::arm_neon_vrshiftu : Intrinsic::arm_neon_vrshifts;
7927 Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Ty), {Ops[1], Ops[2]});
7928 return Builder.CreateAdd(Ops[0], Ops[1], "vrsra_n");
7929 case NEON::BI__builtin_neon_vsri_n_v:
7930 case NEON::BI__builtin_neon_vsriq_n_v:
7931 rightShift = true;
7932 LLVM_FALLTHROUGH[[gnu::fallthrough]];
7933 case NEON::BI__builtin_neon_vsli_n_v:
7934 case NEON::BI__builtin_neon_vsliq_n_v:
7935 Ops[2] = EmitNeonShiftVector(Ops[2], Ty, rightShift);
7936 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vshiftins, Ty),
7937 Ops, "vsli_n");
7938 case NEON::BI__builtin_neon_vsra_n_v:
7939 case NEON::BI__builtin_neon_vsraq_n_v:
7940 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
7941 Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n");
7942 return Builder.CreateAdd(Ops[0], Ops[1]);
7943 case NEON::BI__builtin_neon_vst1q_lane_v:
7944 // Handle 64-bit integer elements as a special case. Use a shuffle to get
7945 // a one-element vector and avoid poor code for i64 in the backend.
7946 if (VTy->getElementType()->isIntegerTy(64)) {
7947 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7948 Value *SV = llvm::ConstantVector::get(cast<llvm::Constant>(Ops[2]));
7949 Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV);
7950 Ops[2] = getAlignmentValue32(PtrOp0);
7951 llvm::Type *Tys[] = {Int8PtrTy, Ops[1]->getType()};
7952 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_neon_vst1,
7953 Tys), Ops);
7954 }
7955 LLVM_FALLTHROUGH[[gnu::fallthrough]];
7956 case NEON::BI__builtin_neon_vst1_lane_v: {
7957 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7958 Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]);
7959 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
7960 auto St = Builder.CreateStore(Ops[1], Builder.CreateBitCast(PtrOp0, Ty));
7961 return St;
7962 }
7963 case NEON::BI__builtin_neon_vtbl1_v:
7964 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl1),
7965 Ops, "vtbl1");
7966 case NEON::BI__builtin_neon_vtbl2_v:
7967 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl2),
7968 Ops, "vtbl2");
7969 case NEON::BI__builtin_neon_vtbl3_v:
7970 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl3),
7971 Ops, "vtbl3");
7972 case NEON::BI__builtin_neon_vtbl4_v:
7973 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl4),
7974 Ops, "vtbl4");
7975 case NEON::BI__builtin_neon_vtbx1_v:
7976 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx1),
7977 Ops, "vtbx1");
7978 case NEON::BI__builtin_neon_vtbx2_v:
7979 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx2),
7980 Ops, "vtbx2");
7981 case NEON::BI__builtin_neon_vtbx3_v:
7982 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx3),
7983 Ops, "vtbx3");
7984 case NEON::BI__builtin_neon_vtbx4_v:
7985 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx4),
7986 Ops, "vtbx4");
7987 }
7988}
7989
7990template<typename Integer>
7991static Integer GetIntegerConstantValue(const Expr *E, ASTContext &Context) {
7992 return E->getIntegerConstantExpr(Context)->getExtValue();
7993}
7994
7995static llvm::Value *SignOrZeroExtend(CGBuilderTy &Builder, llvm::Value *V,
7996 llvm::Type *T, bool Unsigned) {
7997 // Helper function called by Tablegen-constructed ARM MVE builtin codegen,
7998 // which finds it convenient to specify signed/unsigned as a boolean flag.
7999 return Unsigned ? Builder.CreateZExt(V, T) : Builder.CreateSExt(V, T);
8000}
8001
8002static llvm::Value *MVEImmediateShr(CGBuilderTy &Builder, llvm::Value *V,
8003 uint32_t Shift, bool Unsigned) {
8004 // MVE helper function for integer shift right. This must handle signed vs
8005 // unsigned, and also deal specially with the case where the shift count is
8006 // equal to the lane size. In LLVM IR, an LShr with that parameter would be
8007 // undefined behavior, but in MVE it's legal, so we must convert it to code
8008 // that is not undefined in IR.
8009 unsigned LaneBits = cast<llvm::VectorType>(V->getType())
8010 ->getElementType()
8011 ->getPrimitiveSizeInBits();
8012 if (Shift == LaneBits) {
8013 // An unsigned shift of the full lane size always generates zero, so we can
8014 // simply emit a zero vector. A signed shift of the full lane size does the
8015 // same thing as shifting by one bit fewer.
8016 if (Unsigned)
8017 return llvm::Constant::getNullValue(V->getType());
8018 else
8019 --Shift;
8020 }
8021 return Unsigned ? Builder.CreateLShr(V, Shift) : Builder.CreateAShr(V, Shift);
8022}
8023
8024static llvm::Value *ARMMVEVectorSplat(CGBuilderTy &Builder, llvm::Value *V) {
8025 // MVE-specific helper function for a vector splat, which infers the element
8026 // count of the output vector by knowing that MVE vectors are all 128 bits
8027 // wide.
8028 unsigned Elements = 128 / V->getType()->getPrimitiveSizeInBits();
8029 return Builder.CreateVectorSplat(Elements, V);
8030}
8031
8032static llvm::Value *ARMMVEVectorReinterpret(CGBuilderTy &Builder,
8033 CodeGenFunction *CGF,
8034 llvm::Value *V,
8035 llvm::Type *DestType) {
8036 // Convert one MVE vector type into another by reinterpreting its in-register
8037 // format.
8038 //
8039 // Little-endian, this is identical to a bitcast (which reinterprets the
8040 // memory format). But big-endian, they're not necessarily the same, because
8041 // the register and memory formats map to each other differently depending on
8042 // the lane size.
8043 //
8044 // We generate a bitcast whenever we can (if we're little-endian, or if the
8045 // lane sizes are the same anyway). Otherwise we fall back to an IR intrinsic
8046 // that performs the different kind of reinterpretation.
8047 if (CGF->getTarget().isBigEndian() &&
8048 V->getType()->getScalarSizeInBits() != DestType->getScalarSizeInBits()) {
8049 return Builder.CreateCall(
8050 CGF->CGM.getIntrinsic(Intrinsic::arm_mve_vreinterpretq,
8051 {DestType, V->getType()}),
8052 V);
8053 } else {
8054 return Builder.CreateBitCast(V, DestType);
8055 }
8056}
8057
8058static llvm::Value *VectorUnzip(CGBuilderTy &Builder, llvm::Value *V, bool Odd) {
8059 // Make a shufflevector that extracts every other element of a vector (evens
8060 // or odds, as desired).
8061 SmallVector<int, 16> Indices;
8062 unsigned InputElements =
8063 cast<llvm::FixedVectorType>(V->getType())->getNumElements();
8064 for (unsigned i = 0; i < InputElements; i += 2)
8065 Indices.push_back(i + Odd);
8066 return Builder.CreateShuffleVector(V, Indices);
8067}
8068
8069static llvm::Value *VectorZip(CGBuilderTy &Builder, llvm::Value *V0,
8070 llvm::Value *V1) {
8071 // Make a shufflevector that interleaves two vectors element by element.
8072 assert(V0->getType() == V1->getType() && "Can't zip different vector types")(static_cast <bool> (V0->getType() == V1->getType
() && "Can't zip different vector types") ? void (0) :
__assert_fail ("V0->getType() == V1->getType() && \"Can't zip different vector types\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8072, __extension__ __PRETTY_FUNCTION__))
;
8073 SmallVector<int, 16> Indices;
8074 unsigned InputElements =
8075 cast<llvm::FixedVectorType>(V0->getType())->getNumElements();
8076 for (unsigned i = 0; i < InputElements; i++) {
8077 Indices.push_back(i);
8078 Indices.push_back(i + InputElements);
8079 }
8080 return Builder.CreateShuffleVector(V0, V1, Indices);
8081}
8082
8083template<unsigned HighBit, unsigned OtherBits>
8084static llvm::Value *ARMMVEConstantSplat(CGBuilderTy &Builder, llvm::Type *VT) {
8085 // MVE-specific helper function to make a vector splat of a constant such as
8086 // UINT_MAX or INT_MIN, in which all bits below the highest one are equal.
8087 llvm::Type *T = cast<llvm::VectorType>(VT)->getElementType();
8088 unsigned LaneBits = T->getPrimitiveSizeInBits();
8089 uint32_t Value = HighBit << (LaneBits - 1);
8090 if (OtherBits)
8091 Value |= (1UL << (LaneBits - 1)) - 1;
8092 llvm::Value *Lane = llvm::ConstantInt::get(T, Value);
8093 return ARMMVEVectorSplat(Builder, Lane);
8094}
8095
8096static llvm::Value *ARMMVEVectorElementReverse(CGBuilderTy &Builder,
8097 llvm::Value *V,
8098 unsigned ReverseWidth) {
8099 // MVE-specific helper function which reverses the elements of a
8100 // vector within every (ReverseWidth)-bit collection of lanes.
8101 SmallVector<int, 16> Indices;
8102 unsigned LaneSize = V->getType()->getScalarSizeInBits();
8103 unsigned Elements = 128 / LaneSize;
8104 unsigned Mask = ReverseWidth / LaneSize - 1;
8105 for (unsigned i = 0; i < Elements; i++)
8106 Indices.push_back(i ^ Mask);
8107 return Builder.CreateShuffleVector(V, Indices);
8108}
8109
8110Value *CodeGenFunction::EmitARMMVEBuiltinExpr(unsigned BuiltinID,
8111 const CallExpr *E,
8112 ReturnValueSlot ReturnValue,
8113 llvm::Triple::ArchType Arch) {
8114 enum class CustomCodeGen { VLD24, VST24 } CustomCodeGenType;
8115 Intrinsic::ID IRIntr;
8116 unsigned NumVectors;
8117
8118 // Code autogenerated by Tablegen will handle all the simple builtins.
8119 switch (BuiltinID) {
8120 #include "clang/Basic/arm_mve_builtin_cg.inc"
8121
8122 // If we didn't match an MVE builtin id at all, go back to the
8123 // main EmitARMBuiltinExpr.
8124 default:
8125 return nullptr;
8126 }
8127
8128 // Anything that breaks from that switch is an MVE builtin that
8129 // needs handwritten code to generate.
8130
8131 switch (CustomCodeGenType) {
8132
8133 case CustomCodeGen::VLD24: {
8134 llvm::SmallVector<Value *, 4> Ops;
8135 llvm::SmallVector<llvm::Type *, 4> Tys;
8136
8137 auto MvecCType = E->getType();
8138 auto MvecLType = ConvertType(MvecCType);
8139 assert(MvecLType->isStructTy() &&(static_cast <bool> (MvecLType->isStructTy() &&
"Return type for vld[24]q should be a struct") ? void (0) : __assert_fail
("MvecLType->isStructTy() && \"Return type for vld[24]q should be a struct\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8140, __extension__ __PRETTY_FUNCTION__))
8140 "Return type for vld[24]q should be a struct")(static_cast <bool> (MvecLType->isStructTy() &&
"Return type for vld[24]q should be a struct") ? void (0) : __assert_fail
("MvecLType->isStructTy() && \"Return type for vld[24]q should be a struct\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8140, __extension__ __PRETTY_FUNCTION__))
;
8141 assert(MvecLType->getStructNumElements() == 1 &&(static_cast <bool> (MvecLType->getStructNumElements
() == 1 && "Return-type struct for vld[24]q should have one element"
) ? void (0) : __assert_fail ("MvecLType->getStructNumElements() == 1 && \"Return-type struct for vld[24]q should have one element\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8142, __extension__ __PRETTY_FUNCTION__))
8142 "Return-type struct for vld[24]q should have one element")(static_cast <bool> (MvecLType->getStructNumElements
() == 1 && "Return-type struct for vld[24]q should have one element"
) ? void (0) : __assert_fail ("MvecLType->getStructNumElements() == 1 && \"Return-type struct for vld[24]q should have one element\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8142, __extension__ __PRETTY_FUNCTION__))
;
8143 auto MvecLTypeInner = MvecLType->getStructElementType(0);
8144 assert(MvecLTypeInner->isArrayTy() &&(static_cast <bool> (MvecLTypeInner->isArrayTy() &&
"Return-type struct for vld[24]q should contain an array") ?
void (0) : __assert_fail ("MvecLTypeInner->isArrayTy() && \"Return-type struct for vld[24]q should contain an array\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8145, __extension__ __PRETTY_FUNCTION__))
8145 "Return-type struct for vld[24]q should contain an array")(static_cast <bool> (MvecLTypeInner->isArrayTy() &&
"Return-type struct for vld[24]q should contain an array") ?
void (0) : __assert_fail ("MvecLTypeInner->isArrayTy() && \"Return-type struct for vld[24]q should contain an array\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8145, __extension__ __PRETTY_FUNCTION__))
;
8146 assert(MvecLTypeInner->getArrayNumElements() == NumVectors &&(static_cast <bool> (MvecLTypeInner->getArrayNumElements
() == NumVectors && "Array member of return-type struct vld[24]q has wrong length"
) ? void (0) : __assert_fail ("MvecLTypeInner->getArrayNumElements() == NumVectors && \"Array member of return-type struct vld[24]q has wrong length\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8147, __extension__ __PRETTY_FUNCTION__))
8147 "Array member of return-type struct vld[24]q has wrong length")(static_cast <bool> (MvecLTypeInner->getArrayNumElements
() == NumVectors && "Array member of return-type struct vld[24]q has wrong length"
) ? void (0) : __assert_fail ("MvecLTypeInner->getArrayNumElements() == NumVectors && \"Array member of return-type struct vld[24]q has wrong length\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8147, __extension__ __PRETTY_FUNCTION__))
;
8148 auto VecLType = MvecLTypeInner->getArrayElementType();
8149
8150 Tys.push_back(VecLType);
8151
8152 auto Addr = E->getArg(0);
8153 Ops.push_back(EmitScalarExpr(Addr));
8154 Tys.push_back(ConvertType(Addr->getType()));
8155
8156 Function *F = CGM.getIntrinsic(IRIntr, makeArrayRef(Tys));
8157 Value *LoadResult = Builder.CreateCall(F, Ops);
8158 Value *MvecOut = UndefValue::get(MvecLType);
8159 for (unsigned i = 0; i < NumVectors; ++i) {
8160 Value *Vec = Builder.CreateExtractValue(LoadResult, i);
8161 MvecOut = Builder.CreateInsertValue(MvecOut, Vec, {0, i});
8162 }
8163
8164 if (ReturnValue.isNull())
8165 return MvecOut;
8166 else
8167 return Builder.CreateStore(MvecOut, ReturnValue.getValue());
8168 }
8169
8170 case CustomCodeGen::VST24: {
8171 llvm::SmallVector<Value *, 4> Ops;
8172 llvm::SmallVector<llvm::Type *, 4> Tys;
8173
8174 auto Addr = E->getArg(0);
8175 Ops.push_back(EmitScalarExpr(Addr));
8176 Tys.push_back(ConvertType(Addr->getType()));
8177
8178 auto MvecCType = E->getArg(1)->getType();
8179 auto MvecLType = ConvertType(MvecCType);
8180 assert(MvecLType->isStructTy() && "Data type for vst2q should be a struct")(static_cast <bool> (MvecLType->isStructTy() &&
"Data type for vst2q should be a struct") ? void (0) : __assert_fail
("MvecLType->isStructTy() && \"Data type for vst2q should be a struct\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8180, __extension__ __PRETTY_FUNCTION__))
;
8181 assert(MvecLType->getStructNumElements() == 1 &&(static_cast <bool> (MvecLType->getStructNumElements
() == 1 && "Data-type struct for vst2q should have one element"
) ? void (0) : __assert_fail ("MvecLType->getStructNumElements() == 1 && \"Data-type struct for vst2q should have one element\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8182, __extension__ __PRETTY_FUNCTION__))
8182 "Data-type struct for vst2q should have one element")(static_cast <bool> (MvecLType->getStructNumElements
() == 1 && "Data-type struct for vst2q should have one element"
) ? void (0) : __assert_fail ("MvecLType->getStructNumElements() == 1 && \"Data-type struct for vst2q should have one element\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8182, __extension__ __PRETTY_FUNCTION__))
;
8183 auto MvecLTypeInner = MvecLType->getStructElementType(0);
8184 assert(MvecLTypeInner->isArrayTy() &&(static_cast <bool> (MvecLTypeInner->isArrayTy() &&
"Data-type struct for vst2q should contain an array") ? void
(0) : __assert_fail ("MvecLTypeInner->isArrayTy() && \"Data-type struct for vst2q should contain an array\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8185, __extension__ __PRETTY_FUNCTION__))
8185 "Data-type struct for vst2q should contain an array")(static_cast <bool> (MvecLTypeInner->isArrayTy() &&
"Data-type struct for vst2q should contain an array") ? void
(0) : __assert_fail ("MvecLTypeInner->isArrayTy() && \"Data-type struct for vst2q should contain an array\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8185, __extension__ __PRETTY_FUNCTION__))
;
8186 assert(MvecLTypeInner->getArrayNumElements() == NumVectors &&(static_cast <bool> (MvecLTypeInner->getArrayNumElements
() == NumVectors && "Array member of return-type struct vld[24]q has wrong length"
) ? void (0) : __assert_fail ("MvecLTypeInner->getArrayNumElements() == NumVectors && \"Array member of return-type struct vld[24]q has wrong length\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8187, __extension__ __PRETTY_FUNCTION__))
8187 "Array member of return-type struct vld[24]q has wrong length")(static_cast <bool> (MvecLTypeInner->getArrayNumElements
() == NumVectors && "Array member of return-type struct vld[24]q has wrong length"
) ? void (0) : __assert_fail ("MvecLTypeInner->getArrayNumElements() == NumVectors && \"Array member of return-type struct vld[24]q has wrong length\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8187, __extension__ __PRETTY_FUNCTION__))
;
8188 auto VecLType = MvecLTypeInner->getArrayElementType();
8189
8190 Tys.push_back(VecLType);
8191
8192 AggValueSlot MvecSlot = CreateAggTemp(MvecCType);
8193 EmitAggExpr(E->getArg(1), MvecSlot);
8194 auto Mvec = Builder.CreateLoad(MvecSlot.getAddress());
8195 for (unsigned i = 0; i < NumVectors; i++)
8196 Ops.push_back(Builder.CreateExtractValue(Mvec, {0, i}));
8197
8198 Function *F = CGM.getIntrinsic(IRIntr, makeArrayRef(Tys));
8199 Value *ToReturn = nullptr;
8200 for (unsigned i = 0; i < NumVectors; i++) {
8201 Ops.push_back(llvm::ConstantInt::get(Int32Ty, i));
8202 ToReturn = Builder.CreateCall(F, Ops);
8203 Ops.pop_back();
8204 }
8205 return ToReturn;
8206 }
8207 }
8208 llvm_unreachable("unknown custom codegen type.")::llvm::llvm_unreachable_internal("unknown custom codegen type."
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8208)
;
8209}
8210
8211Value *CodeGenFunction::EmitARMCDEBuiltinExpr(unsigned BuiltinID,
8212 const CallExpr *E,
8213 ReturnValueSlot ReturnValue,
8214 llvm::Triple::ArchType Arch) {
8215 switch (BuiltinID) {
8216 default:
8217 return nullptr;
8218#include "clang/Basic/arm_cde_builtin_cg.inc"
8219 }
8220}
8221
8222static Value *EmitAArch64TblBuiltinExpr(CodeGenFunction &CGF, unsigned BuiltinID,
8223 const CallExpr *E,
8224 SmallVectorImpl<Value *> &Ops,
8225 llvm::Triple::ArchType Arch) {
8226 unsigned int Int = 0;
8227 const char *s = nullptr;
8228
8229 switch (BuiltinID) {
8230 default:
8231 return nullptr;
8232 case NEON::BI__builtin_neon_vtbl1_v:
8233 case NEON::BI__builtin_neon_vqtbl1_v:
8234 case NEON::BI__builtin_neon_vqtbl1q_v:
8235 case NEON::BI__builtin_neon_vtbl2_v:
8236 case NEON::BI__builtin_neon_vqtbl2_v:
8237 case NEON::BI__builtin_neon_vqtbl2q_v:
8238 case NEON::BI__builtin_neon_vtbl3_v:
8239 case NEON::BI__builtin_neon_vqtbl3_v:
8240 case NEON::BI__builtin_neon_vqtbl3q_v:
8241 case NEON::BI__builtin_neon_vtbl4_v:
8242 case NEON::BI__builtin_neon_vqtbl4_v:
8243 case NEON::BI__builtin_neon_vqtbl4q_v:
8244 break;
8245 case NEON::BI__builtin_neon_vtbx1_v:
8246 case NEON::BI__builtin_neon_vqtbx1_v:
8247 case NEON::BI__builtin_neon_vqtbx1q_v:
8248 case NEON::BI__builtin_neon_vtbx2_v:
8249 case NEON::BI__builtin_neon_vqtbx2_v:
8250 case NEON::BI__builtin_neon_vqtbx2q_v:
8251 case NEON::BI__builtin_neon_vtbx3_v:
8252 case NEON::BI__builtin_neon_vqtbx3_v:
8253 case NEON::BI__builtin_neon_vqtbx3q_v:
8254 case NEON::BI__builtin_neon_vtbx4_v:
8255 case NEON::BI__builtin_neon_vqtbx4_v:
8256 case NEON::BI__builtin_neon_vqtbx4q_v:
8257 break;
8258 }
8259
8260 assert(E->getNumArgs() >= 3)(static_cast <bool> (E->getNumArgs() >= 3) ? void
(0) : __assert_fail ("E->getNumArgs() >= 3", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8260, __extension__ __PRETTY_FUNCTION__))
;
8261
8262 // Get the last argument, which specifies the vector type.
8263 const Expr *Arg = E->getArg(E->getNumArgs() - 1);
8264 Optional<llvm::APSInt> Result = Arg->getIntegerConstantExpr(CGF.getContext());
8265 if (!Result)
8266 return nullptr;
8267
8268 // Determine the type of this overloaded NEON intrinsic.
8269 NeonTypeFlags Type = Result->getZExtValue();
8270 llvm::FixedVectorType *Ty = GetNeonType(&CGF, Type);
8271 if (!Ty)
8272 return nullptr;
8273
8274 CodeGen::CGBuilderTy &Builder = CGF.Builder;
8275
8276 // AArch64 scalar builtins are not overloaded, they do not have an extra
8277 // argument that specifies the vector type, need to handle each case.
8278 switch (BuiltinID) {
8279 case NEON::BI__builtin_neon_vtbl1_v: {
8280 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 1), nullptr,
8281 Ops[1], Ty, Intrinsic::aarch64_neon_tbl1,
8282 "vtbl1");
8283 }
8284 case NEON::BI__builtin_neon_vtbl2_v: {
8285 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 2), nullptr,
8286 Ops[2], Ty, Intrinsic::aarch64_neon_tbl1,
8287 "vtbl1");
8288 }
8289 case NEON::BI__builtin_neon_vtbl3_v: {
8290 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 3), nullptr,
8291 Ops[3], Ty, Intrinsic::aarch64_neon_tbl2,
8292 "vtbl2");
8293 }
8294 case NEON::BI__builtin_neon_vtbl4_v: {
8295 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 4), nullptr,
8296 Ops[4], Ty, Intrinsic::aarch64_neon_tbl2,
8297 "vtbl2");
8298 }
8299 case NEON::BI__builtin_neon_vtbx1_v: {
8300 Value *TblRes =
8301 packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 1), nullptr, Ops[2],
8302 Ty, Intrinsic::aarch64_neon_tbl1, "vtbl1");
8303
8304 llvm::Constant *EightV = ConstantInt::get(Ty, 8);
8305 Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[2], EightV);
8306 CmpRes = Builder.CreateSExt(CmpRes, Ty);
8307
8308 Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]);
8309 Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes);
8310 return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx");
8311 }
8312 case NEON::BI__builtin_neon_vtbx2_v: {
8313 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 2), Ops[0],
8314 Ops[3], Ty, Intrinsic::aarch64_neon_tbx1,
8315 "vtbx1");
8316 }
8317 case NEON::BI__builtin_neon_vtbx3_v: {
8318 Value *TblRes =
8319 packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 3), nullptr, Ops[4],
8320 Ty, Intrinsic::aarch64_neon_tbl2, "vtbl2");
8321
8322 llvm::Constant *TwentyFourV = ConstantInt::get(Ty, 24);
8323 Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[4],
8324 TwentyFourV);
8325 CmpRes = Builder.CreateSExt(CmpRes, Ty);
8326
8327 Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]);
8328 Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes);
8329 return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx");
8330 }
8331 case NEON::BI__builtin_neon_vtbx4_v: {
8332 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 4), Ops[0],
8333 Ops[5], Ty, Intrinsic::aarch64_neon_tbx2,
8334 "vtbx2");
8335 }
8336 case NEON::BI__builtin_neon_vqtbl1_v:
8337 case NEON::BI__builtin_neon_vqtbl1q_v:
8338 Int = Intrinsic::aarch64_neon_tbl1; s = "vtbl1"; break;
8339 case NEON::BI__builtin_neon_vqtbl2_v:
8340 case NEON::BI__builtin_neon_vqtbl2q_v: {
8341 Int = Intrinsic::aarch64_neon_tbl2; s = "vtbl2"; break;
8342 case NEON::BI__builtin_neon_vqtbl3_v:
8343 case NEON::BI__builtin_neon_vqtbl3q_v:
8344 Int = Intrinsic::aarch64_neon_tbl3; s = "vtbl3"; break;
8345 case NEON::BI__builtin_neon_vqtbl4_v:
8346 case NEON::BI__builtin_neon_vqtbl4q_v:
8347 Int = Intrinsic::aarch64_neon_tbl4; s = "vtbl4"; break;
8348 case NEON::BI__builtin_neon_vqtbx1_v:
8349 case NEON::BI__builtin_neon_vqtbx1q_v:
8350 Int = Intrinsic::aarch64_neon_tbx1; s = "vtbx1"; break;
8351 case NEON::BI__builtin_neon_vqtbx2_v:
8352 case NEON::BI__builtin_neon_vqtbx2q_v:
8353 Int = Intrinsic::aarch64_neon_tbx2; s = "vtbx2"; break;
8354 case NEON::BI__builtin_neon_vqtbx3_v:
8355 case NEON::BI__builtin_neon_vqtbx3q_v:
8356 Int = Intrinsic::aarch64_neon_tbx3; s = "vtbx3"; break;
8357 case NEON::BI__builtin_neon_vqtbx4_v:
8358 case NEON::BI__builtin_neon_vqtbx4q_v:
8359 Int = Intrinsic::aarch64_neon_tbx4; s = "vtbx4"; break;
8360 }
8361 }
8362
8363 if (!Int)
8364 return nullptr;
8365
8366 Function *F = CGF.CGM.getIntrinsic(Int, Ty);
8367 return CGF.EmitNeonCall(F, Ops, s);
8368}
8369
8370Value *CodeGenFunction::vectorWrapScalar16(Value *Op) {
8371 auto *VTy = llvm::FixedVectorType::get(Int16Ty, 4);
8372 Op = Builder.CreateBitCast(Op, Int16Ty);
8373 Value *V = UndefValue::get(VTy);
8374 llvm::Constant *CI = ConstantInt::get(SizeTy, 0);
8375 Op = Builder.CreateInsertElement(V, Op, CI);
8376 return Op;
8377}
8378
8379/// SVEBuiltinMemEltTy - Returns the memory element type for this memory
8380/// access builtin. Only required if it can't be inferred from the base pointer
8381/// operand.
8382llvm::Type *CodeGenFunction::SVEBuiltinMemEltTy(const SVETypeFlags &TypeFlags) {
8383 switch (TypeFlags.getMemEltType()) {
8384 case SVETypeFlags::MemEltTyDefault:
8385 return getEltType(TypeFlags);
8386 case SVETypeFlags::MemEltTyInt8:
8387 return Builder.getInt8Ty();
8388 case SVETypeFlags::MemEltTyInt16:
8389 return Builder.getInt16Ty();
8390 case SVETypeFlags::MemEltTyInt32:
8391 return Builder.getInt32Ty();
8392 case SVETypeFlags::MemEltTyInt64:
8393 return Builder.getInt64Ty();
8394 }
8395 llvm_unreachable("Unknown MemEltType")::llvm::llvm_unreachable_internal("Unknown MemEltType", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8395)
;
8396}
8397
8398llvm::Type *CodeGenFunction::getEltType(const SVETypeFlags &TypeFlags) {
8399 switch (TypeFlags.getEltType()) {
8400 default:
8401 llvm_unreachable("Invalid SVETypeFlag!")::llvm::llvm_unreachable_internal("Invalid SVETypeFlag!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8401)
;
8402
8403 case SVETypeFlags::EltTyInt8:
8404 return Builder.getInt8Ty();
8405 case SVETypeFlags::EltTyInt16:
8406 return Builder.getInt16Ty();
8407 case SVETypeFlags::EltTyInt32:
8408 return Builder.getInt32Ty();
8409 case SVETypeFlags::EltTyInt64:
8410 return Builder.getInt64Ty();
8411
8412 case SVETypeFlags::EltTyFloat16:
8413 return Builder.getHalfTy();
8414 case SVETypeFlags::EltTyFloat32:
8415 return Builder.getFloatTy();
8416 case SVETypeFlags::EltTyFloat64:
8417 return Builder.getDoubleTy();
8418
8419 case SVETypeFlags::EltTyBFloat16:
8420 return Builder.getBFloatTy();
8421
8422 case SVETypeFlags::EltTyBool8:
8423 case SVETypeFlags::EltTyBool16:
8424 case SVETypeFlags::EltTyBool32:
8425 case SVETypeFlags::EltTyBool64:
8426 return Builder.getInt1Ty();
8427 }
8428}
8429
8430// Return the llvm predicate vector type corresponding to the specified element
8431// TypeFlags.
8432llvm::ScalableVectorType *
8433CodeGenFunction::getSVEPredType(const SVETypeFlags &TypeFlags) {
8434 switch (TypeFlags.getEltType()) {
8435 default: llvm_unreachable("Unhandled SVETypeFlag!")::llvm::llvm_unreachable_internal("Unhandled SVETypeFlag!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8435)
;
8436
8437 case SVETypeFlags::EltTyInt8:
8438 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 16);
8439 case SVETypeFlags::EltTyInt16:
8440 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8);
8441 case SVETypeFlags::EltTyInt32:
8442 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4);
8443 case SVETypeFlags::EltTyInt64:
8444 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2);
8445
8446 case SVETypeFlags::EltTyBFloat16:
8447 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8);
8448 case SVETypeFlags::EltTyFloat16:
8449 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8);
8450 case SVETypeFlags::EltTyFloat32:
8451 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4);
8452 case SVETypeFlags::EltTyFloat64:
8453 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2);
8454
8455 case SVETypeFlags::EltTyBool8:
8456 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 16);
8457 case SVETypeFlags::EltTyBool16:
8458 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8);
8459 case SVETypeFlags::EltTyBool32:
8460 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4);
8461 case SVETypeFlags::EltTyBool64:
8462 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2);
8463 }
8464}
8465
8466// Return the llvm vector type corresponding to the specified element TypeFlags.
8467llvm::ScalableVectorType *
8468CodeGenFunction::getSVEType(const SVETypeFlags &TypeFlags) {
8469 switch (TypeFlags.getEltType()) {
8470 default:
8471 llvm_unreachable("Invalid SVETypeFlag!")::llvm::llvm_unreachable_internal("Invalid SVETypeFlag!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8471)
;
8472
8473 case SVETypeFlags::EltTyInt8:
8474 return llvm::ScalableVectorType::get(Builder.getInt8Ty(), 16);
8475 case SVETypeFlags::EltTyInt16:
8476 return llvm::ScalableVectorType::get(Builder.getInt16Ty(), 8);
8477 case SVETypeFlags::EltTyInt32:
8478 return llvm::ScalableVectorType::get(Builder.getInt32Ty(), 4);
8479 case SVETypeFlags::EltTyInt64:
8480 return llvm::ScalableVectorType::get(Builder.getInt64Ty(), 2);
8481
8482 case SVETypeFlags::EltTyFloat16:
8483 return llvm::ScalableVectorType::get(Builder.getHalfTy(), 8);
8484 case SVETypeFlags::EltTyBFloat16:
8485 return llvm::ScalableVectorType::get(Builder.getBFloatTy(), 8);
8486 case SVETypeFlags::EltTyFloat32:
8487 return llvm::ScalableVectorType::get(Builder.getFloatTy(), 4);
8488 case SVETypeFlags::EltTyFloat64:
8489 return llvm::ScalableVectorType::get(Builder.getDoubleTy(), 2);
8490
8491 case SVETypeFlags::EltTyBool8:
8492 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 16);
8493 case SVETypeFlags::EltTyBool16:
8494 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8);
8495 case SVETypeFlags::EltTyBool32:
8496 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4);
8497 case SVETypeFlags::EltTyBool64:
8498 return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2);
8499 }
8500}
8501
8502llvm::Value *
8503CodeGenFunction::EmitSVEAllTruePred(const SVETypeFlags &TypeFlags) {
8504 Function *Ptrue =
8505 CGM.getIntrinsic(Intrinsic::aarch64_sve_ptrue, getSVEPredType(TypeFlags));
8506 return Builder.CreateCall(Ptrue, {Builder.getInt32(/*SV_ALL*/ 31)});
8507}
8508
8509constexpr unsigned SVEBitsPerBlock = 128;
8510
8511static llvm::ScalableVectorType *getSVEVectorForElementType(llvm::Type *EltTy) {
8512 unsigned NumElts = SVEBitsPerBlock / EltTy->getScalarSizeInBits();
8513 return llvm::ScalableVectorType::get(EltTy, NumElts);
8514}
8515
8516// Reinterpret the input predicate so that it can be used to correctly isolate
8517// the elements of the specified datatype.
8518Value *CodeGenFunction::EmitSVEPredicateCast(Value *Pred,
8519 llvm::ScalableVectorType *VTy) {
8520 auto *RTy = llvm::VectorType::get(IntegerType::get(getLLVMContext(), 1), VTy);
8521 if (Pred->getType() == RTy)
8522 return Pred;
8523
8524 unsigned IntID;
8525 llvm::Type *IntrinsicTy;
8526 switch (VTy->getMinNumElements()) {
8527 default:
8528 llvm_unreachable("unsupported element count!")::llvm::llvm_unreachable_internal("unsupported element count!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8528)
;
8529 case 2:
8530 case 4:
8531 case 8:
8532 IntID = Intrinsic::aarch64_sve_convert_from_svbool;
8533 IntrinsicTy = RTy;
8534 break;
8535 case 16:
8536 IntID = Intrinsic::aarch64_sve_convert_to_svbool;
8537 IntrinsicTy = Pred->getType();
8538 break;
8539 }
8540
8541 Function *F = CGM.getIntrinsic(IntID, IntrinsicTy);
8542 Value *C = Builder.CreateCall(F, Pred);
8543 assert(C->getType() == RTy && "Unexpected return type!")(static_cast <bool> (C->getType() == RTy && "Unexpected return type!"
) ? void (0) : __assert_fail ("C->getType() == RTy && \"Unexpected return type!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8543, __extension__ __PRETTY_FUNCTION__))
;
8544 return C;
8545}
8546
8547Value *CodeGenFunction::EmitSVEGatherLoad(const SVETypeFlags &TypeFlags,
8548 SmallVectorImpl<Value *> &Ops,
8549 unsigned IntID) {
8550 auto *ResultTy = getSVEType(TypeFlags);
8551 auto *OverloadedTy =
8552 llvm::ScalableVectorType::get(SVEBuiltinMemEltTy(TypeFlags), ResultTy);
8553
8554 // At the ACLE level there's only one predicate type, svbool_t, which is
8555 // mapped to <n x 16 x i1>. However, this might be incompatible with the
8556 // actual type being loaded. For example, when loading doubles (i64) the
8557 // predicated should be <n x 2 x i1> instead. At the IR level the type of
8558 // the predicate and the data being loaded must match. Cast accordingly.
8559 Ops[0] = EmitSVEPredicateCast(Ops[0], OverloadedTy);
8560
8561 Function *F = nullptr;
8562 if (Ops[1]->getType()->isVectorTy())
8563 // This is the "vector base, scalar offset" case. In order to uniquely
8564 // map this built-in to an LLVM IR intrinsic, we need both the return type
8565 // and the type of the vector base.
8566 F = CGM.getIntrinsic(IntID, {OverloadedTy, Ops[1]->getType()});
8567 else
8568 // This is the "scalar base, vector offset case". The type of the offset
8569 // is encoded in the name of the intrinsic. We only need to specify the
8570 // return type in order to uniquely map this built-in to an LLVM IR
8571 // intrinsic.
8572 F = CGM.getIntrinsic(IntID, OverloadedTy);
8573
8574 // Pass 0 when the offset is missing. This can only be applied when using
8575 // the "vector base" addressing mode for which ACLE allows no offset. The
8576 // corresponding LLVM IR always requires an offset.
8577 if (Ops.size() == 2) {
8578 assert(Ops[1]->getType()->isVectorTy() && "Scalar base requires an offset")(static_cast <bool> (Ops[1]->getType()->isVectorTy
() && "Scalar base requires an offset") ? void (0) : __assert_fail
("Ops[1]->getType()->isVectorTy() && \"Scalar base requires an offset\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8578, __extension__ __PRETTY_FUNCTION__))
;
8579 Ops.push_back(ConstantInt::get(Int64Ty, 0));
8580 }
8581
8582 // For "vector base, scalar index" scale the index so that it becomes a
8583 // scalar offset.
8584 if (!TypeFlags.isByteIndexed() && Ops[1]->getType()->isVectorTy()) {
8585 unsigned BytesPerElt =
8586 OverloadedTy->getElementType()->getScalarSizeInBits() / 8;
8587 Value *Scale = ConstantInt::get(Int64Ty, BytesPerElt);
8588 Ops[2] = Builder.CreateMul(Ops[2], Scale);
8589 }
8590
8591 Value *Call = Builder.CreateCall(F, Ops);
8592
8593 // The following sext/zext is only needed when ResultTy != OverloadedTy. In
8594 // other cases it's folded into a nop.
8595 return TypeFlags.isZExtReturn() ? Builder.CreateZExt(Call, ResultTy)
8596 : Builder.CreateSExt(Call, ResultTy);
8597}
8598
8599Value *CodeGenFunction::EmitSVEScatterStore(const SVETypeFlags &TypeFlags,
8600 SmallVectorImpl<Value *> &Ops,
8601 unsigned IntID) {
8602 auto *SrcDataTy = getSVEType(TypeFlags);
8603 auto *OverloadedTy =
8604 llvm::ScalableVectorType::get(SVEBuiltinMemEltTy(TypeFlags), SrcDataTy);
8605
8606 // In ACLE the source data is passed in the last argument, whereas in LLVM IR
8607 // it's the first argument. Move it accordingly.
8608 Ops.insert(Ops.begin(), Ops.pop_back_val());
8609
8610 Function *F = nullptr;
8611 if (Ops[2]->getType()->isVectorTy())
8612 // This is the "vector base, scalar offset" case. In order to uniquely
8613 // map this built-in to an LLVM IR intrinsic, we need both the return type
8614 // and the type of the vector base.
8615 F = CGM.getIntrinsic(IntID, {OverloadedTy, Ops[2]->getType()});
8616 else
8617 // This is the "scalar base, vector offset case". The type of the offset
8618 // is encoded in the name of the intrinsic. We only need to specify the
8619 // return type in order to uniquely map this built-in to an LLVM IR
8620 // intrinsic.
8621 F = CGM.getIntrinsic(IntID, OverloadedTy);
8622
8623 // Pass 0 when the offset is missing. This can only be applied when using
8624 // the "vector base" addressing mode for which ACLE allows no offset. The
8625 // corresponding LLVM IR always requires an offset.
8626 if (Ops.size() == 3) {
8627 assert(Ops[1]->getType()->isVectorTy() && "Scalar base requires an offset")(static_cast <bool> (Ops[1]->getType()->isVectorTy
() && "Scalar base requires an offset") ? void (0) : __assert_fail
("Ops[1]->getType()->isVectorTy() && \"Scalar base requires an offset\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8627, __extension__ __PRETTY_FUNCTION__))
;
8628 Ops.push_back(ConstantInt::get(Int64Ty, 0));
8629 }
8630
8631 // Truncation is needed when SrcDataTy != OverloadedTy. In other cases it's
8632 // folded into a nop.
8633 Ops[0] = Builder.CreateTrunc(Ops[0], OverloadedTy);
8634
8635 // At the ACLE level there's only one predicate type, svbool_t, which is
8636 // mapped to <n x 16 x i1>. However, this might be incompatible with the
8637 // actual type being stored. For example, when storing doubles (i64) the
8638 // predicated should be <n x 2 x i1> instead. At the IR level the type of
8639 // the predicate and the data being stored must match. Cast accordingly.
8640 Ops[1] = EmitSVEPredicateCast(Ops[1], OverloadedTy);
8641
8642 // For "vector base, scalar index" scale the index so that it becomes a
8643 // scalar offset.
8644 if (!TypeFlags.isByteIndexed() && Ops[2]->getType()->isVectorTy()) {
8645 unsigned BytesPerElt =
8646 OverloadedTy->getElementType()->getScalarSizeInBits() / 8;
8647 Value *Scale = ConstantInt::get(Int64Ty, BytesPerElt);
8648 Ops[3] = Builder.CreateMul(Ops[3], Scale);
8649 }
8650
8651 return Builder.CreateCall(F, Ops);
8652}
8653
8654Value *CodeGenFunction::EmitSVEGatherPrefetch(const SVETypeFlags &TypeFlags,
8655 SmallVectorImpl<Value *> &Ops,
8656 unsigned IntID) {
8657 // The gather prefetches are overloaded on the vector input - this can either
8658 // be the vector of base addresses or vector of offsets.
8659 auto *OverloadedTy = dyn_cast<llvm::ScalableVectorType>(Ops[1]->getType());
8660 if (!OverloadedTy)
8661 OverloadedTy = cast<llvm::ScalableVectorType>(Ops[2]->getType());
8662
8663 // Cast the predicate from svbool_t to the right number of elements.
8664 Ops[0] = EmitSVEPredicateCast(Ops[0], OverloadedTy);
8665
8666 // vector + imm addressing modes
8667 if (Ops[1]->getType()->isVectorTy()) {
8668 if (Ops.size() == 3) {
8669 // Pass 0 for 'vector+imm' when the index is omitted.
8670 Ops.push_back(ConstantInt::get(Int64Ty, 0));
8671
8672 // The sv_prfop is the last operand in the builtin and IR intrinsic.
8673 std::swap(Ops[2], Ops[3]);
8674 } else {
8675 // Index needs to be passed as scaled offset.
8676 llvm::Type *MemEltTy = SVEBuiltinMemEltTy(TypeFlags);
8677 unsigned BytesPerElt = MemEltTy->getPrimitiveSizeInBits() / 8;
8678 Value *Scale = ConstantInt::get(Int64Ty, BytesPerElt);
8679 Ops[2] = Builder.CreateMul(Ops[2], Scale);
8680 }
8681 }
8682
8683 Function *F = CGM.getIntrinsic(IntID, OverloadedTy);
8684 return Builder.CreateCall(F, Ops);
8685}
8686
8687Value *CodeGenFunction::EmitSVEStructLoad(const SVETypeFlags &TypeFlags,
8688 SmallVectorImpl<Value*> &Ops,
8689 unsigned IntID) {
8690 llvm::ScalableVectorType *VTy = getSVEType(TypeFlags);
8691 auto VecPtrTy = llvm::PointerType::getUnqual(VTy);
8692 auto EltPtrTy = llvm::PointerType::getUnqual(VTy->getElementType());
8693
8694 unsigned N;
8695 switch (IntID) {
8696 case Intrinsic::aarch64_sve_ld2:
8697 N = 2;
8698 break;
8699 case Intrinsic::aarch64_sve_ld3:
8700 N = 3;
8701 break;
8702 case Intrinsic::aarch64_sve_ld4:
8703 N = 4;
8704 break;
8705 default:
8706 llvm_unreachable("unknown intrinsic!")::llvm::llvm_unreachable_internal("unknown intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8706)
;
8707 }
8708 auto RetTy = llvm::VectorType::get(VTy->getElementType(),
8709 VTy->getElementCount() * N);
8710
8711 Value *Predicate = EmitSVEPredicateCast(Ops[0], VTy);
8712 Value *BasePtr= Builder.CreateBitCast(Ops[1], VecPtrTy);
8713 Value *Offset = Ops.size() > 2 ? Ops[2] : Builder.getInt32(0);
8714 BasePtr = Builder.CreateGEP(VTy, BasePtr, Offset);
8715 BasePtr = Builder.CreateBitCast(BasePtr, EltPtrTy);
8716
8717 Function *F = CGM.getIntrinsic(IntID, {RetTy, Predicate->getType()});
8718 return Builder.CreateCall(F, { Predicate, BasePtr });
8719}
8720
8721Value *CodeGenFunction::EmitSVEStructStore(const SVETypeFlags &TypeFlags,
8722 SmallVectorImpl<Value*> &Ops,
8723 unsigned IntID) {
8724 llvm::ScalableVectorType *VTy = getSVEType(TypeFlags);
8725 auto VecPtrTy = llvm::PointerType::getUnqual(VTy);
8726 auto EltPtrTy = llvm::PointerType::getUnqual(VTy->getElementType());
8727
8728 unsigned N;
8729 switch (IntID) {
8730 case Intrinsic::aarch64_sve_st2:
8731 N = 2;
8732 break;
8733 case Intrinsic::aarch64_sve_st3:
8734 N = 3;
8735 break;
8736 case Intrinsic::aarch64_sve_st4:
8737 N = 4;
8738 break;
8739 default:
8740 llvm_unreachable("unknown intrinsic!")::llvm::llvm_unreachable_internal("unknown intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8740)
;
8741 }
8742 auto TupleTy =
8743 llvm::VectorType::get(VTy->getElementType(), VTy->getElementCount() * N);
8744
8745 Value *Predicate = EmitSVEPredicateCast(Ops[0], VTy);
8746 Value *BasePtr = Builder.CreateBitCast(Ops[1], VecPtrTy);
8747 Value *Offset = Ops.size() > 3 ? Ops[2] : Builder.getInt32(0);
8748 Value *Val = Ops.back();
8749 BasePtr = Builder.CreateGEP(VTy, BasePtr, Offset);
8750 BasePtr = Builder.CreateBitCast(BasePtr, EltPtrTy);
8751
8752 // The llvm.aarch64.sve.st2/3/4 intrinsics take legal part vectors, so we
8753 // need to break up the tuple vector.
8754 SmallVector<llvm::Value*, 5> Operands;
8755 Function *FExtr =
8756 CGM.getIntrinsic(Intrinsic::aarch64_sve_tuple_get, {VTy, TupleTy});
8757 for (unsigned I = 0; I < N; ++I)
8758 Operands.push_back(Builder.CreateCall(FExtr, {Val, Builder.getInt32(I)}));
8759 Operands.append({Predicate, BasePtr});
8760
8761 Function *F = CGM.getIntrinsic(IntID, { VTy });
8762 return Builder.CreateCall(F, Operands);
8763}
8764
8765// SVE2's svpmullb and svpmullt builtins are similar to the svpmullb_pair and
8766// svpmullt_pair intrinsics, with the exception that their results are bitcast
8767// to a wider type.
8768Value *CodeGenFunction::EmitSVEPMull(const SVETypeFlags &TypeFlags,
8769 SmallVectorImpl<Value *> &Ops,
8770 unsigned BuiltinID) {
8771 // Splat scalar operand to vector (intrinsics with _n infix)
8772 if (TypeFlags.hasSplatOperand()) {
8773 unsigned OpNo = TypeFlags.getSplatOperand();
8774 Ops[OpNo] = EmitSVEDupX(Ops[OpNo]);
8775 }
8776
8777 // The pair-wise function has a narrower overloaded type.
8778 Function *F = CGM.getIntrinsic(BuiltinID, Ops[0]->getType());
8779 Value *Call = Builder.CreateCall(F, {Ops[0], Ops[1]});
8780
8781 // Now bitcast to the wider result type.
8782 llvm::ScalableVectorType *Ty = getSVEType(TypeFlags);
8783 return EmitSVEReinterpret(Call, Ty);
8784}
8785
8786Value *CodeGenFunction::EmitSVEMovl(const SVETypeFlags &TypeFlags,
8787 ArrayRef<Value *> Ops, unsigned BuiltinID) {
8788 llvm::Type *OverloadedTy = getSVEType(TypeFlags);
8789 Function *F = CGM.getIntrinsic(BuiltinID, OverloadedTy);
8790 return Builder.CreateCall(F, {Ops[0], Builder.getInt32(0)});
8791}
8792
8793Value *CodeGenFunction::EmitSVEPrefetchLoad(const SVETypeFlags &TypeFlags,
8794 SmallVectorImpl<Value *> &Ops,
8795 unsigned BuiltinID) {
8796 auto *MemEltTy = SVEBuiltinMemEltTy(TypeFlags);
8797 auto *VectorTy = getSVEVectorForElementType(MemEltTy);
8798 auto *MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy);
8799
8800 Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy);
8801 Value *BasePtr = Ops[1];
8802
8803 // Implement the index operand if not omitted.
8804 if (Ops.size() > 3) {
8805 BasePtr = Builder.CreateBitCast(BasePtr, MemoryTy->getPointerTo());
8806 BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Ops[2]);
8807 }
8808
8809 // Prefetch intriniscs always expect an i8*
8810 BasePtr = Builder.CreateBitCast(BasePtr, llvm::PointerType::getUnqual(Int8Ty));
8811 Value *PrfOp = Ops.back();
8812
8813 Function *F = CGM.getIntrinsic(BuiltinID, Predicate->getType());
8814 return Builder.CreateCall(F, {Predicate, BasePtr, PrfOp});
8815}
8816
8817Value *CodeGenFunction::EmitSVEMaskedLoad(const CallExpr *E,
8818 llvm::Type *ReturnTy,
8819 SmallVectorImpl<Value *> &Ops,
8820 unsigned BuiltinID,
8821 bool IsZExtReturn) {
8822 QualType LangPTy = E->getArg(1)->getType();
8823 llvm::Type *MemEltTy = CGM.getTypes().ConvertType(
8824 LangPTy->castAs<PointerType>()->getPointeeType());
8825
8826 // The vector type that is returned may be different from the
8827 // eventual type loaded from memory.
8828 auto VectorTy = cast<llvm::ScalableVectorType>(ReturnTy);
8829 auto MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy);
8830
8831 Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy);
8832 Value *BasePtr = Builder.CreateBitCast(Ops[1], MemoryTy->getPointerTo());
8833 Value *Offset = Ops.size() > 2 ? Ops[2] : Builder.getInt32(0);
8834 BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Offset);
8835
8836 BasePtr = Builder.CreateBitCast(BasePtr, MemEltTy->getPointerTo());
8837 Function *F = CGM.getIntrinsic(BuiltinID, MemoryTy);
8838 Value *Load = Builder.CreateCall(F, {Predicate, BasePtr});
8839
8840 return IsZExtReturn ? Builder.CreateZExt(Load, VectorTy)
8841 : Builder.CreateSExt(Load, VectorTy);
8842}
8843
8844Value *CodeGenFunction::EmitSVEMaskedStore(const CallExpr *E,
8845 SmallVectorImpl<Value *> &Ops,
8846 unsigned BuiltinID) {
8847 QualType LangPTy = E->getArg(1)->getType();
8848 llvm::Type *MemEltTy = CGM.getTypes().ConvertType(
8849 LangPTy->castAs<PointerType>()->getPointeeType());
8850
8851 // The vector type that is stored may be different from the
8852 // eventual type stored to memory.
8853 auto VectorTy = cast<llvm::ScalableVectorType>(Ops.back()->getType());
8854 auto MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy);
8855
8856 Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy);
8857 Value *BasePtr = Builder.CreateBitCast(Ops[1], MemoryTy->getPointerTo());
8858 Value *Offset = Ops.size() == 4 ? Ops[2] : Builder.getInt32(0);
8859 BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Offset);
8860
8861 // Last value is always the data
8862 llvm::Value *Val = Builder.CreateTrunc(Ops.back(), MemoryTy);
8863
8864 BasePtr = Builder.CreateBitCast(BasePtr, MemEltTy->getPointerTo());
8865 Function *F = CGM.getIntrinsic(BuiltinID, MemoryTy);
8866 return Builder.CreateCall(F, {Val, Predicate, BasePtr});
8867}
8868
8869// Limit the usage of scalable llvm IR generated by the ACLE by using the
8870// sve dup.x intrinsic instead of IRBuilder::CreateVectorSplat.
8871Value *CodeGenFunction::EmitSVEDupX(Value *Scalar, llvm::Type *Ty) {
8872 auto F = CGM.getIntrinsic(Intrinsic::aarch64_sve_dup_x, Ty);
8873 return Builder.CreateCall(F, Scalar);
8874}
8875
8876Value *CodeGenFunction::EmitSVEDupX(Value* Scalar) {
8877 return EmitSVEDupX(Scalar, getSVEVectorForElementType(Scalar->getType()));
8878}
8879
8880Value *CodeGenFunction::EmitSVEReinterpret(Value *Val, llvm::Type *Ty) {
8881 // FIXME: For big endian this needs an additional REV, or needs a separate
8882 // intrinsic that is code-generated as a no-op, because the LLVM bitcast
8883 // instruction is defined as 'bitwise' equivalent from memory point of
8884 // view (when storing/reloading), whereas the svreinterpret builtin
8885 // implements bitwise equivalent cast from register point of view.
8886 // LLVM CodeGen for a bitcast must add an explicit REV for big-endian.
8887 return Builder.CreateBitCast(Val, Ty);
8888}
8889
8890static void InsertExplicitZeroOperand(CGBuilderTy &Builder, llvm::Type *Ty,
8891 SmallVectorImpl<Value *> &Ops) {
8892 auto *SplatZero = Constant::getNullValue(Ty);
8893 Ops.insert(Ops.begin(), SplatZero);
8894}
8895
8896static void InsertExplicitUndefOperand(CGBuilderTy &Builder, llvm::Type *Ty,
8897 SmallVectorImpl<Value *> &Ops) {
8898 auto *SplatUndef = UndefValue::get(Ty);
8899 Ops.insert(Ops.begin(), SplatUndef);
8900}
8901
8902SmallVector<llvm::Type *, 2>
8903CodeGenFunction::getSVEOverloadTypes(const SVETypeFlags &TypeFlags,
8904 llvm::Type *ResultType,
8905 ArrayRef<Value *> Ops) {
8906 if (TypeFlags.isOverloadNone())
8907 return {};
8908
8909 llvm::Type *DefaultType = getSVEType(TypeFlags);
8910
8911 if (TypeFlags.isOverloadWhile())
8912 return {DefaultType, Ops[1]->getType()};
8913
8914 if (TypeFlags.isOverloadWhileRW())
8915 return {getSVEPredType(TypeFlags), Ops[0]->getType()};
8916
8917 if (TypeFlags.isOverloadCvt() || TypeFlags.isTupleSet())
8918 return {Ops[0]->getType(), Ops.back()->getType()};
8919
8920 if (TypeFlags.isTupleCreate() || TypeFlags.isTupleGet())
8921 return {ResultType, Ops[0]->getType()};
8922
8923 assert(TypeFlags.isOverloadDefault() && "Unexpected value for overloads")(static_cast <bool> (TypeFlags.isOverloadDefault() &&
"Unexpected value for overloads") ? void (0) : __assert_fail
("TypeFlags.isOverloadDefault() && \"Unexpected value for overloads\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8923, __extension__ __PRETTY_FUNCTION__))
;
8924 return {DefaultType};
8925}
8926
8927Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
8928 const CallExpr *E) {
8929 // Find out if any arguments are required to be integer constant expressions.
8930 unsigned ICEArguments = 0;
8931 ASTContext::GetBuiltinTypeError Error;
8932 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
8933 assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None &&
"Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8933, __extension__ __PRETTY_FUNCTION__))
;
8934
8935 llvm::Type *Ty = ConvertType(E->getType());
8936 if (BuiltinID >= SVE::BI__builtin_sve_reinterpret_s8_s8 &&
8937 BuiltinID <= SVE::BI__builtin_sve_reinterpret_f64_f64) {
8938 Value *Val = EmitScalarExpr(E->getArg(0));
8939 return EmitSVEReinterpret(Val, Ty);
8940 }
8941
8942 llvm::SmallVector<Value *, 4> Ops;
8943 for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) {
8944 if ((ICEArguments & (1 << i)) == 0)
8945 Ops.push_back(EmitScalarExpr(E->getArg(i)));
8946 else {
8947 // If this is required to be a constant, constant fold it so that we know
8948 // that the generated intrinsic gets a ConstantInt.
8949 Optional<llvm::APSInt> Result =
8950 E->getArg(i)->getIntegerConstantExpr(getContext());
8951 assert(Result && "Expected argument to be a constant")(static_cast <bool> (Result && "Expected argument to be a constant"
) ? void (0) : __assert_fail ("Result && \"Expected argument to be a constant\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 8951, __extension__ __PRETTY_FUNCTION__))
;
8952
8953 // Immediates for SVE llvm intrinsics are always 32bit. We can safely
8954 // truncate because the immediate has been range checked and no valid
8955 // immediate requires more than a handful of bits.
8956 *Result = Result->extOrTrunc(32);
8957 Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), *Result));
8958 }
8959 }
8960
8961 auto *Builtin = findARMVectorIntrinsicInMap(AArch64SVEIntrinsicMap, BuiltinID,
8962 AArch64SVEIntrinsicsProvenSorted);
8963 SVETypeFlags TypeFlags(Builtin->TypeModifier);
8964 if (TypeFlags.isLoad())
8965 return EmitSVEMaskedLoad(E, Ty, Ops, Builtin->LLVMIntrinsic,
8966 TypeFlags.isZExtReturn());
8967 else if (TypeFlags.isStore())
8968 return EmitSVEMaskedStore(E, Ops, Builtin->LLVMIntrinsic);
8969 else if (TypeFlags.isGatherLoad())
8970 return EmitSVEGatherLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic);
8971 else if (TypeFlags.isScatterStore())
8972 return EmitSVEScatterStore(TypeFlags, Ops, Builtin->LLVMIntrinsic);
8973 else if (TypeFlags.isPrefetch())
8974 return EmitSVEPrefetchLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic);
8975 else if (TypeFlags.isGatherPrefetch())
8976 return EmitSVEGatherPrefetch(TypeFlags, Ops, Builtin->LLVMIntrinsic);
8977 else if (TypeFlags.isStructLoad())
8978 return EmitSVEStructLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic);
8979 else if (TypeFlags.isStructStore())
8980 return EmitSVEStructStore(TypeFlags, Ops, Builtin->LLVMIntrinsic);
8981 else if (TypeFlags.isUndef())
8982 return UndefValue::get(Ty);
8983 else if (Builtin->LLVMIntrinsic != 0) {
8984 if (TypeFlags.getMergeType() == SVETypeFlags::MergeZeroExp)
8985 InsertExplicitZeroOperand(Builder, Ty, Ops);
8986
8987 if (TypeFlags.getMergeType() == SVETypeFlags::MergeAnyExp)
8988 InsertExplicitUndefOperand(Builder, Ty, Ops);
8989
8990 // Some ACLE builtins leave out the argument to specify the predicate
8991 // pattern, which is expected to be expanded to an SV_ALL pattern.
8992 if (TypeFlags.isAppendSVALL())
8993 Ops.push_back(Builder.getInt32(/*SV_ALL*/ 31));
8994 if (TypeFlags.isInsertOp1SVALL())
8995 Ops.insert(&Ops[1], Builder.getInt32(/*SV_ALL*/ 31));
8996
8997 // Predicates must match the main datatype.
8998 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
8999 if (auto PredTy = dyn_cast<llvm::VectorType>(Ops[i]->getType()))
9000 if (PredTy->getElementType()->isIntegerTy(1))
9001 Ops[i] = EmitSVEPredicateCast(Ops[i], getSVEType(TypeFlags));
9002
9003 // Splat scalar operand to vector (intrinsics with _n infix)
9004 if (TypeFlags.hasSplatOperand()) {
9005 unsigned OpNo = TypeFlags.getSplatOperand();
9006 Ops[OpNo] = EmitSVEDupX(Ops[OpNo]);
9007 }
9008
9009 if (TypeFlags.isReverseCompare())
9010 std::swap(Ops[1], Ops[2]);
9011
9012 if (TypeFlags.isReverseUSDOT())
9013 std::swap(Ops[1], Ops[2]);
9014
9015 // Predicated intrinsics with _z suffix need a select w/ zeroinitializer.
9016 if (TypeFlags.getMergeType() == SVETypeFlags::MergeZero) {
9017 llvm::Type *OpndTy = Ops[1]->getType();
9018 auto *SplatZero = Constant::getNullValue(OpndTy);
9019 Function *Sel = CGM.getIntrinsic(Intrinsic::aarch64_sve_sel, OpndTy);
9020 Ops[1] = Builder.CreateCall(Sel, {Ops[0], Ops[1], SplatZero});
9021 }
9022
9023 Function *F = CGM.getIntrinsic(Builtin->LLVMIntrinsic,
9024 getSVEOverloadTypes(TypeFlags, Ty, Ops));
9025 Value *Call = Builder.CreateCall(F, Ops);
9026
9027 // Predicate results must be converted to svbool_t.
9028 if (auto PredTy = dyn_cast<llvm::VectorType>(Call->getType()))
9029 if (PredTy->getScalarType()->isIntegerTy(1))
9030 Call = EmitSVEPredicateCast(Call, cast<llvm::ScalableVectorType>(Ty));
9031
9032 return Call;
9033 }
9034
9035 switch (BuiltinID) {
9036 default:
9037 return nullptr;
9038
9039 case SVE::BI__builtin_sve_svmov_b_z: {
9040 // svmov_b_z(pg, op) <=> svand_b_z(pg, op, op)
9041 SVETypeFlags TypeFlags(Builtin->TypeModifier);
9042 llvm::Type* OverloadedTy = getSVEType(TypeFlags);
9043 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_sve_and_z, OverloadedTy);
9044 return Builder.CreateCall(F, {Ops[0], Ops[1], Ops[1]});
9045 }
9046
9047 case SVE::BI__builtin_sve_svnot_b_z: {
9048 // svnot_b_z(pg, op) <=> sveor_b_z(pg, op, pg)
9049 SVETypeFlags TypeFlags(Builtin->TypeModifier);
9050 llvm::Type* OverloadedTy = getSVEType(TypeFlags);
9051 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_sve_eor_z, OverloadedTy);
9052 return Builder.CreateCall(F, {Ops[0], Ops[1], Ops[0]});
9053 }
9054
9055 case SVE::BI__builtin_sve_svmovlb_u16:
9056 case SVE::BI__builtin_sve_svmovlb_u32:
9057 case SVE::BI__builtin_sve_svmovlb_u64:
9058 return EmitSVEMovl(TypeFlags, Ops, Intrinsic::aarch64_sve_ushllb);
9059
9060 case SVE::BI__builtin_sve_svmovlb_s16:
9061 case SVE::BI__builtin_sve_svmovlb_s32:
9062 case SVE::BI__builtin_sve_svmovlb_s64:
9063 return EmitSVEMovl(TypeFlags, Ops, Intrinsic::aarch64_sve_sshllb);
9064
9065 case SVE::BI__builtin_sve_svmovlt_u16:
9066 case SVE::BI__builtin_sve_svmovlt_u32:
9067 case SVE::BI__builtin_sve_svmovlt_u64:
9068 return EmitSVEMovl(TypeFlags, Ops, Intrinsic::aarch64_sve_ushllt);
9069
9070 case SVE::BI__builtin_sve_svmovlt_s16:
9071 case SVE::BI__builtin_sve_svmovlt_s32:
9072 case SVE::BI__builtin_sve_svmovlt_s64:
9073 return EmitSVEMovl(TypeFlags, Ops, Intrinsic::aarch64_sve_sshllt);
9074
9075 case SVE::BI__builtin_sve_svpmullt_u16:
9076 case SVE::BI__builtin_sve_svpmullt_u64:
9077 case SVE::BI__builtin_sve_svpmullt_n_u16:
9078 case SVE::BI__builtin_sve_svpmullt_n_u64:
9079 return EmitSVEPMull(TypeFlags, Ops, Intrinsic::aarch64_sve_pmullt_pair);
9080
9081 case SVE::BI__builtin_sve_svpmullb_u16:
9082 case SVE::BI__builtin_sve_svpmullb_u64:
9083 case SVE::BI__builtin_sve_svpmullb_n_u16:
9084 case SVE::BI__builtin_sve_svpmullb_n_u64:
9085 return EmitSVEPMull(TypeFlags, Ops, Intrinsic::aarch64_sve_pmullb_pair);
9086
9087 case SVE::BI__builtin_sve_svdup_n_b8:
9088 case SVE::BI__builtin_sve_svdup_n_b16:
9089 case SVE::BI__builtin_sve_svdup_n_b32:
9090 case SVE::BI__builtin_sve_svdup_n_b64: {
9091 Value *CmpNE =
9092 Builder.CreateICmpNE(Ops[0], Constant::getNullValue(Ops[0]->getType()));
9093 llvm::ScalableVectorType *OverloadedTy = getSVEType(TypeFlags);
9094 Value *Dup = EmitSVEDupX(CmpNE, OverloadedTy);
9095 return EmitSVEPredicateCast(Dup, cast<llvm::ScalableVectorType>(Ty));
9096 }
9097
9098 case SVE::BI__builtin_sve_svdupq_n_b8:
9099 case SVE::BI__builtin_sve_svdupq_n_b16:
9100 case SVE::BI__builtin_sve_svdupq_n_b32:
9101 case SVE::BI__builtin_sve_svdupq_n_b64:
9102 case SVE::BI__builtin_sve_svdupq_n_u8:
9103 case SVE::BI__builtin_sve_svdupq_n_s8:
9104 case SVE::BI__builtin_sve_svdupq_n_u64:
9105 case SVE::BI__builtin_sve_svdupq_n_f64:
9106 case SVE::BI__builtin_sve_svdupq_n_s64:
9107 case SVE::BI__builtin_sve_svdupq_n_u16:
9108 case SVE::BI__builtin_sve_svdupq_n_f16:
9109 case SVE::BI__builtin_sve_svdupq_n_bf16:
9110 case SVE::BI__builtin_sve_svdupq_n_s16:
9111 case SVE::BI__builtin_sve_svdupq_n_u32:
9112 case SVE::BI__builtin_sve_svdupq_n_f32:
9113 case SVE::BI__builtin_sve_svdupq_n_s32: {
9114 // These builtins are implemented by storing each element to an array and using
9115 // ld1rq to materialize a vector.
9116 unsigned NumOpnds = Ops.size();
9117
9118 bool IsBoolTy =
9119 cast<llvm::VectorType>(Ty)->getElementType()->isIntegerTy(1);
9120
9121 // For svdupq_n_b* the element type of is an integer of type 128/numelts,
9122 // so that the compare can use the width that is natural for the expected
9123 // number of predicate lanes.
9124 llvm::Type *EltTy = Ops[0]->getType();
9125 if (IsBoolTy)
9126 EltTy = IntegerType::get(getLLVMContext(), SVEBitsPerBlock / NumOpnds);
9127
9128 SmallVector<llvm::Value *, 16> VecOps;
9129 for (unsigned I = 0; I < NumOpnds; ++I)
9130 VecOps.push_back(Builder.CreateZExt(Ops[I], EltTy));
9131 Value *Vec = BuildVector(VecOps);
9132
9133 SVETypeFlags TypeFlags(Builtin->TypeModifier);
9134 Value *Pred = EmitSVEAllTruePred(TypeFlags);
9135
9136 llvm::Type *OverloadedTy = getSVEVectorForElementType(EltTy);
9137 Value *InsertSubVec = Builder.CreateInsertVector(
9138 OverloadedTy, UndefValue::get(OverloadedTy), Vec, Builder.getInt64(0));
9139
9140 Function *F =
9141 CGM.getIntrinsic(Intrinsic::aarch64_sve_dupq_lane, OverloadedTy);
9142 Value *DupQLane =
9143 Builder.CreateCall(F, {InsertSubVec, Builder.getInt64(0)});
9144
9145 if (!IsBoolTy)
9146 return DupQLane;
9147
9148 // For svdupq_n_b* we need to add an additional 'cmpne' with '0'.
9149 F = CGM.getIntrinsic(NumOpnds == 2 ? Intrinsic::aarch64_sve_cmpne
9150 : Intrinsic::aarch64_sve_cmpne_wide,
9151 OverloadedTy);
9152 Value *Call = Builder.CreateCall(
9153 F, {Pred, DupQLane, EmitSVEDupX(Builder.getInt64(0))});
9154 return EmitSVEPredicateCast(Call, cast<llvm::ScalableVectorType>(Ty));
9155 }
9156
9157 case SVE::BI__builtin_sve_svpfalse_b:
9158 return ConstantInt::getFalse(Ty);
9159
9160 case SVE::BI__builtin_sve_svlen_bf16:
9161 case SVE::BI__builtin_sve_svlen_f16:
9162 case SVE::BI__builtin_sve_svlen_f32:
9163 case SVE::BI__builtin_sve_svlen_f64:
9164 case SVE::BI__builtin_sve_svlen_s8:
9165 case SVE::BI__builtin_sve_svlen_s16:
9166 case SVE::BI__builtin_sve_svlen_s32:
9167 case SVE::BI__builtin_sve_svlen_s64:
9168 case SVE::BI__builtin_sve_svlen_u8:
9169 case SVE::BI__builtin_sve_svlen_u16:
9170 case SVE::BI__builtin_sve_svlen_u32:
9171 case SVE::BI__builtin_sve_svlen_u64: {
9172 SVETypeFlags TF(Builtin->TypeModifier);
9173 auto VTy = cast<llvm::VectorType>(getSVEType(TF));
9174 auto *NumEls =
9175 llvm::ConstantInt::get(Ty, VTy->getElementCount().getKnownMinValue());
9176
9177 Function *F = CGM.getIntrinsic(Intrinsic::vscale, Ty);
9178 return Builder.CreateMul(NumEls, Builder.CreateCall(F));
9179 }
9180
9181 case SVE::BI__builtin_sve_svtbl2_u8:
9182 case SVE::BI__builtin_sve_svtbl2_s8:
9183 case SVE::BI__builtin_sve_svtbl2_u16:
9184 case SVE::BI__builtin_sve_svtbl2_s16:
9185 case SVE::BI__builtin_sve_svtbl2_u32:
9186 case SVE::BI__builtin_sve_svtbl2_s32:
9187 case SVE::BI__builtin_sve_svtbl2_u64:
9188 case SVE::BI__builtin_sve_svtbl2_s64:
9189 case SVE::BI__builtin_sve_svtbl2_f16:
9190 case SVE::BI__builtin_sve_svtbl2_bf16:
9191 case SVE::BI__builtin_sve_svtbl2_f32:
9192 case SVE::BI__builtin_sve_svtbl2_f64: {
9193 SVETypeFlags TF(Builtin->TypeModifier);
9194 auto VTy = cast<llvm::VectorType>(getSVEType(TF));
9195 auto TupleTy = llvm::VectorType::getDoubleElementsVectorType(VTy);
9196 Function *FExtr =
9197 CGM.getIntrinsic(Intrinsic::aarch64_sve_tuple_get, {VTy, TupleTy});
9198 Value *V0 = Builder.CreateCall(FExtr, {Ops[0], Builder.getInt32(0)});
9199 Value *V1 = Builder.CreateCall(FExtr, {Ops[0], Builder.getInt32(1)});
9200 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_sve_tbl2, VTy);
9201 return Builder.CreateCall(F, {V0, V1, Ops[1]});
9202 }
9203 }
9204
9205 /// Should not happen
9206 return nullptr;
9207}
9208
9209Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
9210 const CallExpr *E,
9211 llvm::Triple::ArchType Arch) {
9212 if (BuiltinID >= AArch64::FirstSVEBuiltin &&
9213 BuiltinID <= AArch64::LastSVEBuiltin)
9214 return EmitAArch64SVEBuiltinExpr(BuiltinID, E);
9215
9216 unsigned HintID = static_cast<unsigned>(-1);
9217 switch (BuiltinID) {
9218 default: break;
9219 case AArch64::BI__builtin_arm_nop:
9220 HintID = 0;
9221 break;
9222 case AArch64::BI__builtin_arm_yield:
9223 case AArch64::BI__yield:
9224 HintID = 1;
9225 break;
9226 case AArch64::BI__builtin_arm_wfe:
9227 case AArch64::BI__wfe:
9228 HintID = 2;
9229 break;
9230 case AArch64::BI__builtin_arm_wfi:
9231 case AArch64::BI__wfi:
9232 HintID = 3;
9233 break;
9234 case AArch64::BI__builtin_arm_sev:
9235 case AArch64::BI__sev:
9236 HintID = 4;
9237 break;
9238 case AArch64::BI__builtin_arm_sevl:
9239 case AArch64::BI__sevl:
9240 HintID = 5;
9241 break;
9242 }
9243
9244 if (HintID != static_cast<unsigned>(-1)) {
9245 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_hint);
9246 return Builder.CreateCall(F, llvm::ConstantInt::get(Int32Ty, HintID));
9247 }
9248
9249 if (BuiltinID == AArch64::BI__builtin_arm_prefetch) {
9250 Value *Address = EmitScalarExpr(E->getArg(0));
9251 Value *RW = EmitScalarExpr(E->getArg(1));
9252 Value *CacheLevel = EmitScalarExpr(E->getArg(2));
9253 Value *RetentionPolicy = EmitScalarExpr(E->getArg(3));
9254 Value *IsData = EmitScalarExpr(E->getArg(4));
9255
9256 Value *Locality = nullptr;
9257 if (cast<llvm::ConstantInt>(RetentionPolicy)->isZero()) {
9258 // Temporal fetch, needs to convert cache level to locality.
9259 Locality = llvm::ConstantInt::get(Int32Ty,
9260 -cast<llvm::ConstantInt>(CacheLevel)->getValue() + 3);
9261 } else {
9262 // Streaming fetch.
9263 Locality = llvm::ConstantInt::get(Int32Ty, 0);
9264 }
9265
9266 // FIXME: We need AArch64 specific LLVM intrinsic if we want to specify
9267 // PLDL3STRM or PLDL2STRM.
9268 Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType());
9269 return Builder.CreateCall(F, {Address, RW, Locality, IsData});
9270 }
9271
9272 if (BuiltinID == AArch64::BI__builtin_arm_rbit) {
9273 assert((getContext().getTypeSize(E->getType()) == 32) &&(static_cast <bool> ((getContext().getTypeSize(E->getType
()) == 32) && "rbit of unusual size!") ? void (0) : __assert_fail
("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9274, __extension__ __PRETTY_FUNCTION__))
9274 "rbit of unusual size!")(static_cast <bool> ((getContext().getTypeSize(E->getType
()) == 32) && "rbit of unusual size!") ? void (0) : __assert_fail
("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9274, __extension__ __PRETTY_FUNCTION__))
;
9275 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9276 return Builder.CreateCall(
9277 CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
9278 }
9279 if (BuiltinID == AArch64::BI__builtin_arm_rbit64) {
9280 assert((getContext().getTypeSize(E->getType()) == 64) &&(static_cast <bool> ((getContext().getTypeSize(E->getType
()) == 64) && "rbit of unusual size!") ? void (0) : __assert_fail
("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9281, __extension__ __PRETTY_FUNCTION__))
9281 "rbit of unusual size!")(static_cast <bool> ((getContext().getTypeSize(E->getType
()) == 64) && "rbit of unusual size!") ? void (0) : __assert_fail
("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9281, __extension__ __PRETTY_FUNCTION__))
;
9282 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9283 return Builder.CreateCall(
9284 CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
9285 }
9286
9287 if (BuiltinID == AArch64::BI__builtin_arm_cls) {
9288 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9289 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_cls), Arg,
9290 "cls");
9291 }
9292 if (BuiltinID == AArch64::BI__builtin_arm_cls64) {
9293 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9294 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_cls64), Arg,
9295 "cls");
9296 }
9297
9298 if (BuiltinID == AArch64::BI__builtin_arm_frint32zf ||
9299 BuiltinID == AArch64::BI__builtin_arm_frint32z) {
9300 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9301 llvm::Type *Ty = Arg->getType();
9302 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_frint32z, Ty),
9303 Arg, "frint32z");
9304 }
9305
9306 if (BuiltinID == AArch64::BI__builtin_arm_frint64zf ||
9307 BuiltinID == AArch64::BI__builtin_arm_frint64z) {
9308 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9309 llvm::Type *Ty = Arg->getType();
9310 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_frint64z, Ty),
9311 Arg, "frint64z");
9312 }
9313
9314 if (BuiltinID == AArch64::BI__builtin_arm_frint32xf ||
9315 BuiltinID == AArch64::BI__builtin_arm_frint32x) {
9316 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9317 llvm::Type *Ty = Arg->getType();
9318 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_frint32x, Ty),
9319 Arg, "frint32x");
9320 }
9321
9322 if (BuiltinID == AArch64::BI__builtin_arm_frint64xf ||
9323 BuiltinID == AArch64::BI__builtin_arm_frint64x) {
9324 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9325 llvm::Type *Ty = Arg->getType();
9326 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_frint64x, Ty),
9327 Arg, "frint64x");
9328 }
9329
9330 if (BuiltinID == AArch64::BI__builtin_arm_jcvt) {
9331 assert((getContext().getTypeSize(E->getType()) == 32) &&(static_cast <bool> ((getContext().getTypeSize(E->getType
()) == 32) && "__jcvt of unusual size!") ? void (0) :
__assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"__jcvt of unusual size!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9332, __extension__ __PRETTY_FUNCTION__))
9332 "__jcvt of unusual size!")(static_cast <bool> ((getContext().getTypeSize(E->getType
()) == 32) && "__jcvt of unusual size!") ? void (0) :
__assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"__jcvt of unusual size!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9332, __extension__ __PRETTY_FUNCTION__))
;
9333 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
9334 return Builder.CreateCall(
9335 CGM.getIntrinsic(Intrinsic::aarch64_fjcvtzs), Arg);
9336 }
9337
9338 if (BuiltinID == AArch64::BI__builtin_arm_ld64b ||
9339 BuiltinID == AArch64::BI__builtin_arm_st64b ||
9340 BuiltinID == AArch64::BI__builtin_arm_st64bv ||
9341 BuiltinID == AArch64::BI__builtin_arm_st64bv0) {
9342 llvm::Value *MemAddr = EmitScalarExpr(E->getArg(0));
9343 llvm::Value *ValPtr = EmitScalarExpr(E->getArg(1));
9344
9345 if (BuiltinID == AArch64::BI__builtin_arm_ld64b) {
9346 // Load from the address via an LLVM intrinsic, receiving a
9347 // tuple of 8 i64 words, and store each one to ValPtr.
9348 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_ld64b);
9349 llvm::Value *Val = Builder.CreateCall(F, MemAddr);
9350 llvm::Value *ToRet;
9351 for (size_t i = 0; i < 8; i++) {
9352 llvm::Value *ValOffsetPtr =
9353 Builder.CreateGEP(Int64Ty, ValPtr, Builder.getInt32(i));
9354 Address Addr(ValOffsetPtr, CharUnits::fromQuantity(8));
9355 ToRet = Builder.CreateStore(Builder.CreateExtractValue(Val, i), Addr);
9356 }
9357 return ToRet;
9358 } else {
9359 // Load 8 i64 words from ValPtr, and store them to the address
9360 // via an LLVM intrinsic.
9361 SmallVector<llvm::Value *, 9> Args;
9362 Args.push_back(MemAddr);
9363 for (size_t i = 0; i < 8; i++) {
9364 llvm::Value *ValOffsetPtr =
9365 Builder.CreateGEP(Int64Ty, ValPtr, Builder.getInt32(i));
9366 Address Addr(ValOffsetPtr, CharUnits::fromQuantity(8));
9367 Args.push_back(Builder.CreateLoad(Addr));
9368 }
9369
9370 auto Intr = (BuiltinID == AArch64::BI__builtin_arm_st64b
9371 ? Intrinsic::aarch64_st64b
9372 : BuiltinID == AArch64::BI__builtin_arm_st64bv
9373 ? Intrinsic::aarch64_st64bv
9374 : Intrinsic::aarch64_st64bv0);
9375 Function *F = CGM.getIntrinsic(Intr);
9376 return Builder.CreateCall(F, Args);
9377 }
9378 }
9379
9380 if (BuiltinID == AArch64::BI__builtin_arm_rndr ||
9381 BuiltinID == AArch64::BI__builtin_arm_rndrrs) {
9382
9383 auto Intr = (BuiltinID == AArch64::BI__builtin_arm_rndr
9384 ? Intrinsic::aarch64_rndr
9385 : Intrinsic::aarch64_rndrrs);
9386 Function *F = CGM.getIntrinsic(Intr);
9387 llvm::Value *Val = Builder.CreateCall(F);
9388 Value *RandomValue = Builder.CreateExtractValue(Val, 0);
9389 Value *Status = Builder.CreateExtractValue(Val, 1);
9390
9391 Address MemAddress = EmitPointerWithAlignment(E->getArg(0));
9392 Builder.CreateStore(RandomValue, MemAddress);
9393 Status = Builder.CreateZExt(Status, Int32Ty);
9394 return Status;
9395 }
9396
9397 if (BuiltinID == AArch64::BI__clear_cache) {
9398 assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments")(static_cast <bool> (E->getNumArgs() == 2 &&
"__clear_cache takes 2 arguments") ? void (0) : __assert_fail
("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9398, __extension__ __PRETTY_FUNCTION__))
;
9399 const FunctionDecl *FD = E->getDirectCallee();
9400 Value *Ops[2];
9401 for (unsigned i = 0; i < 2; i++)
9402 Ops[i] = EmitScalarExpr(E->getArg(i));
9403 llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType());
9404 llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty);
9405 StringRef Name = FD->getName();
9406 return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops);
9407 }
9408
9409 if ((BuiltinID == AArch64::BI__builtin_arm_ldrex ||
9410 BuiltinID == AArch64::BI__builtin_arm_ldaex) &&
9411 getContext().getTypeSize(E->getType()) == 128) {
9412 Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex
9413 ? Intrinsic::aarch64_ldaxp
9414 : Intrinsic::aarch64_ldxp);
9415
9416 Value *LdPtr = EmitScalarExpr(E->getArg(0));
9417 Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy),
9418 "ldxp");
9419
9420 Value *Val0 = Builder.CreateExtractValue(Val, 1);
9421 Value *Val1 = Builder.CreateExtractValue(Val, 0);
9422 llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128);
9423 Val0 = Builder.CreateZExt(Val0, Int128Ty);
9424 Val1 = Builder.CreateZExt(Val1, Int128Ty);
9425
9426 Value *ShiftCst = llvm::ConstantInt::get(Int128Ty, 64);
9427 Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */);
9428 Val = Builder.CreateOr(Val, Val1);
9429 return Builder.CreateBitCast(Val, ConvertType(E->getType()));
9430 } else if (BuiltinID == AArch64::BI__builtin_arm_ldrex ||
9431 BuiltinID == AArch64::BI__builtin_arm_ldaex) {
9432 Value *LoadAddr = EmitScalarExpr(E->getArg(0));
9433
9434 QualType Ty = E->getType();
9435 llvm::Type *RealResTy = ConvertType(Ty);
9436 llvm::Type *PtrTy = llvm::IntegerType::get(
9437 getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo();
9438 LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy);
9439
9440 Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex
9441 ? Intrinsic::aarch64_ldaxr
9442 : Intrinsic::aarch64_ldxr,
9443 PtrTy);
9444 Value *Val = Builder.CreateCall(F, LoadAddr, "ldxr");
9445
9446 if (RealResTy->isPointerTy())
9447 return Builder.CreateIntToPtr(Val, RealResTy);
9448
9449 llvm::Type *IntResTy = llvm::IntegerType::get(
9450 getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy));
9451 Val = Builder.CreateTruncOrBitCast(Val, IntResTy);
9452 return Builder.CreateBitCast(Val, RealResTy);
9453 }
9454
9455 if ((BuiltinID == AArch64::BI__builtin_arm_strex ||
9456 BuiltinID == AArch64::BI__builtin_arm_stlex) &&
9457 getContext().getTypeSize(E->getArg(0)->getType()) == 128) {
9458 Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex
9459 ? Intrinsic::aarch64_stlxp
9460 : Intrinsic::aarch64_stxp);
9461 llvm::Type *STy = llvm::StructType::get(Int64Ty, Int64Ty);
9462
9463 Address Tmp = CreateMemTemp(E->getArg(0)->getType());
9464 EmitAnyExprToMem(E->getArg(0), Tmp, Qualifiers(), /*init*/ true);
9465
9466 Tmp = Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(STy));
9467 llvm::Value *Val = Builder.CreateLoad(Tmp);
9468
9469 Value *Arg0 = Builder.CreateExtractValue(Val, 0);
9470 Value *Arg1 = Builder.CreateExtractValue(Val, 1);
9471 Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)),
9472 Int8PtrTy);
9473 return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "stxp");
9474 }
9475
9476 if (BuiltinID == AArch64::BI__builtin_arm_strex ||
9477 BuiltinID == AArch64::BI__builtin_arm_stlex) {
9478 Value *StoreVal = EmitScalarExpr(E->getArg(0));
9479 Value *StoreAddr = EmitScalarExpr(E->getArg(1));
9480
9481 QualType Ty = E->getArg(0)->getType();
9482 llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(),
9483 getContext().getTypeSize(Ty));
9484 StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo());
9485
9486 if (StoreVal->getType()->isPointerTy())
9487 StoreVal = Builder.CreatePtrToInt(StoreVal, Int64Ty);
9488 else {
9489 llvm::Type *IntTy = llvm::IntegerType::get(
9490 getLLVMContext(),
9491 CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType()));
9492 StoreVal = Builder.CreateBitCast(StoreVal, IntTy);
9493 StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int64Ty);
9494 }
9495
9496 Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex
9497 ? Intrinsic::aarch64_stlxr
9498 : Intrinsic::aarch64_stxr,
9499 StoreAddr->getType());
9500 return Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr");
9501 }
9502
9503 if (BuiltinID == AArch64::BI__getReg) {
9504 Expr::EvalResult Result;
9505 if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext()))
9506 llvm_unreachable("Sema will ensure that the parameter is constant")::llvm::llvm_unreachable_internal("Sema will ensure that the parameter is constant"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9506)
;
9507
9508 llvm::APSInt Value = Result.Val.getInt();
9509 LLVMContext &Context = CGM.getLLVMContext();
9510 std::string Reg = Value == 31 ? "sp" : "x" + toString(Value, 10);
9511
9512 llvm::Metadata *Ops[] = {llvm::MDString::get(Context, Reg)};
9513 llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops);
9514 llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName);
9515
9516 llvm::Function *F =
9517 CGM.getIntrinsic(llvm::Intrinsic::read_register, {Int64Ty});
9518 return Builder.CreateCall(F, Metadata);
9519 }
9520
9521 if (BuiltinID == AArch64::BI__builtin_arm_clrex) {
9522 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_clrex);
9523 return Builder.CreateCall(F);
9524 }
9525
9526 if (BuiltinID == AArch64::BI_ReadWriteBarrier)
9527 return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent,
9528 llvm::SyncScope::SingleThread);
9529
9530 // CRC32
9531 Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic;
9532 switch (BuiltinID) {
9533 case AArch64::BI__builtin_arm_crc32b:
9534 CRCIntrinsicID = Intrinsic::aarch64_crc32b; break;
9535 case AArch64::BI__builtin_arm_crc32cb:
9536 CRCIntrinsicID = Intrinsic::aarch64_crc32cb; break;
9537 case AArch64::BI__builtin_arm_crc32h:
9538 CRCIntrinsicID = Intrinsic::aarch64_crc32h; break;
9539 case AArch64::BI__builtin_arm_crc32ch:
9540 CRCIntrinsicID = Intrinsic::aarch64_crc32ch; break;
9541 case AArch64::BI__builtin_arm_crc32w:
9542 CRCIntrinsicID = Intrinsic::aarch64_crc32w; break;
9543 case AArch64::BI__builtin_arm_crc32cw:
9544 CRCIntrinsicID = Intrinsic::aarch64_crc32cw; break;
9545 case AArch64::BI__builtin_arm_crc32d:
9546 CRCIntrinsicID = Intrinsic::aarch64_crc32x; break;
9547 case AArch64::BI__builtin_arm_crc32cd:
9548 CRCIntrinsicID = Intrinsic::aarch64_crc32cx; break;
9549 }
9550
9551 if (CRCIntrinsicID != Intrinsic::not_intrinsic) {
9552 Value *Arg0 = EmitScalarExpr(E->getArg(0));
9553 Value *Arg1 = EmitScalarExpr(E->getArg(1));
9554 Function *F = CGM.getIntrinsic(CRCIntrinsicID);
9555
9556 llvm::Type *DataTy = F->getFunctionType()->getParamType(1);
9557 Arg1 = Builder.CreateZExtOrBitCast(Arg1, DataTy);
9558
9559 return Builder.CreateCall(F, {Arg0, Arg1});
9560 }
9561
9562 // Memory Tagging Extensions (MTE) Intrinsics
9563 Intrinsic::ID MTEIntrinsicID = Intrinsic::not_intrinsic;
9564 switch (BuiltinID) {
9565 case AArch64::BI__builtin_arm_irg:
9566 MTEIntrinsicID = Intrinsic::aarch64_irg; break;
9567 case AArch64::BI__builtin_arm_addg:
9568 MTEIntrinsicID = Intrinsic::aarch64_addg; break;
9569 case AArch64::BI__builtin_arm_gmi:
9570 MTEIntrinsicID = Intrinsic::aarch64_gmi; break;
9571 case AArch64::BI__builtin_arm_ldg:
9572 MTEIntrinsicID = Intrinsic::aarch64_ldg; break;
9573 case AArch64::BI__builtin_arm_stg:
9574 MTEIntrinsicID = Intrinsic::aarch64_stg; break;
9575 case AArch64::BI__builtin_arm_subp:
9576 MTEIntrinsicID = Intrinsic::aarch64_subp; break;
9577 }
9578
9579 if (MTEIntrinsicID != Intrinsic::not_intrinsic) {
9580 llvm::Type *T = ConvertType(E->getType());
9581
9582 if (MTEIntrinsicID == Intrinsic::aarch64_irg) {
9583 Value *Pointer = EmitScalarExpr(E->getArg(0));
9584 Value *Mask = EmitScalarExpr(E->getArg(1));
9585
9586 Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy);
9587 Mask = Builder.CreateZExt(Mask, Int64Ty);
9588 Value *RV = Builder.CreateCall(
9589 CGM.getIntrinsic(MTEIntrinsicID), {Pointer, Mask});
9590 return Builder.CreatePointerCast(RV, T);
9591 }
9592 if (MTEIntrinsicID == Intrinsic::aarch64_addg) {
9593 Value *Pointer = EmitScalarExpr(E->getArg(0));
9594 Value *TagOffset = EmitScalarExpr(E->getArg(1));
9595
9596 Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy);
9597 TagOffset = Builder.CreateZExt(TagOffset, Int64Ty);
9598 Value *RV = Builder.CreateCall(
9599 CGM.getIntrinsic(MTEIntrinsicID), {Pointer, TagOffset});
9600 return Builder.CreatePointerCast(RV, T);
9601 }
9602 if (MTEIntrinsicID == Intrinsic::aarch64_gmi) {
9603 Value *Pointer = EmitScalarExpr(E->getArg(0));
9604 Value *ExcludedMask = EmitScalarExpr(E->getArg(1));
9605
9606 ExcludedMask = Builder.CreateZExt(ExcludedMask, Int64Ty);
9607 Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy);
9608 return Builder.CreateCall(
9609 CGM.getIntrinsic(MTEIntrinsicID), {Pointer, ExcludedMask});
9610 }
9611 // Although it is possible to supply a different return
9612 // address (first arg) to this intrinsic, for now we set
9613 // return address same as input address.
9614 if (MTEIntrinsicID == Intrinsic::aarch64_ldg) {
9615 Value *TagAddress = EmitScalarExpr(E->getArg(0));
9616 TagAddress = Builder.CreatePointerCast(TagAddress, Int8PtrTy);
9617 Value *RV = Builder.CreateCall(
9618 CGM.getIntrinsic(MTEIntrinsicID), {TagAddress, TagAddress});
9619 return Builder.CreatePointerCast(RV, T);
9620 }
9621 // Although it is possible to supply a different tag (to set)
9622 // to this intrinsic (as first arg), for now we supply
9623 // the tag that is in input address arg (common use case).
9624 if (MTEIntrinsicID == Intrinsic::aarch64_stg) {
9625 Value *TagAddress = EmitScalarExpr(E->getArg(0));
9626 TagAddress = Builder.CreatePointerCast(TagAddress, Int8PtrTy);
9627 return Builder.CreateCall(
9628 CGM.getIntrinsic(MTEIntrinsicID), {TagAddress, TagAddress});
9629 }
9630 if (MTEIntrinsicID == Intrinsic::aarch64_subp) {
9631 Value *PointerA = EmitScalarExpr(E->getArg(0));
9632 Value *PointerB = EmitScalarExpr(E->getArg(1));
9633 PointerA = Builder.CreatePointerCast(PointerA, Int8PtrTy);
9634 PointerB = Builder.CreatePointerCast(PointerB, Int8PtrTy);
9635 return Builder.CreateCall(
9636 CGM.getIntrinsic(MTEIntrinsicID), {PointerA, PointerB});
9637 }
9638 }
9639
9640 if (BuiltinID == AArch64::BI__builtin_arm_rsr ||
9641 BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
9642 BuiltinID == AArch64::BI__builtin_arm_rsrp ||
9643 BuiltinID == AArch64::BI__builtin_arm_wsr ||
9644 BuiltinID == AArch64::BI__builtin_arm_wsr64 ||
9645 BuiltinID == AArch64::BI__builtin_arm_wsrp) {
9646
9647 SpecialRegisterAccessKind AccessKind = Write;
9648 if (BuiltinID == AArch64::BI__builtin_arm_rsr ||
9649 BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
9650 BuiltinID == AArch64::BI__builtin_arm_rsrp)
9651 AccessKind = VolatileRead;
9652
9653 bool IsPointerBuiltin = BuiltinID == AArch64::BI__builtin_arm_rsrp ||
9654 BuiltinID == AArch64::BI__builtin_arm_wsrp;
9655
9656 bool Is64Bit = BuiltinID != AArch64::BI__builtin_arm_rsr &&
9657 BuiltinID != AArch64::BI__builtin_arm_wsr;
9658
9659 llvm::Type *ValueType;
9660 llvm::Type *RegisterType = Int64Ty;
9661 if (IsPointerBuiltin) {
9662 ValueType = VoidPtrTy;
9663 } else if (Is64Bit) {
9664 ValueType = Int64Ty;
9665 } else {
9666 ValueType = Int32Ty;
9667 }
9668
9669 return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType,
9670 AccessKind);
9671 }
9672
9673 if (BuiltinID == AArch64::BI_ReadStatusReg ||
9674 BuiltinID == AArch64::BI_WriteStatusReg) {
9675 LLVMContext &Context = CGM.getLLVMContext();
9676
9677 unsigned SysReg =
9678 E->getArg(0)->EvaluateKnownConstInt(getContext()).getZExtValue();
9679
9680 std::string SysRegStr;
9681 llvm::raw_string_ostream(SysRegStr) <<
9682 ((1 << 1) | ((SysReg >> 14) & 1)) << ":" <<
9683 ((SysReg >> 11) & 7) << ":" <<
9684 ((SysReg >> 7) & 15) << ":" <<
9685 ((SysReg >> 3) & 15) << ":" <<
9686 ( SysReg & 7);
9687
9688 llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysRegStr) };
9689 llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops);
9690 llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName);
9691
9692 llvm::Type *RegisterType = Int64Ty;
9693 llvm::Type *Types[] = { RegisterType };
9694
9695 if (BuiltinID == AArch64::BI_ReadStatusReg) {
9696 llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types);
9697
9698 return Builder.CreateCall(F, Metadata);
9699 }
9700
9701 llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types);
9702 llvm::Value *ArgValue = EmitScalarExpr(E->getArg(1));
9703
9704 return Builder.CreateCall(F, { Metadata, ArgValue });
9705 }
9706
9707 if (BuiltinID == AArch64::BI_AddressOfReturnAddress) {
9708 llvm::Function *F =
9709 CGM.getIntrinsic(Intrinsic::addressofreturnaddress, AllocaInt8PtrTy);
9710 return Builder.CreateCall(F);
9711 }
9712
9713 if (BuiltinID == AArch64::BI__builtin_sponentry) {
9714 llvm::Function *F = CGM.getIntrinsic(Intrinsic::sponentry, AllocaInt8PtrTy);
9715 return Builder.CreateCall(F);
9716 }
9717
9718 if (BuiltinID == AArch64::BI__mulh || BuiltinID == AArch64::BI__umulh) {
9719 llvm::Type *ResType = ConvertType(E->getType());
9720 llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128);
9721
9722 bool IsSigned = BuiltinID == AArch64::BI__mulh;
9723 Value *LHS =
9724 Builder.CreateIntCast(EmitScalarExpr(E->getArg(0)), Int128Ty, IsSigned);
9725 Value *RHS =
9726 Builder.CreateIntCast(EmitScalarExpr(E->getArg(1)), Int128Ty, IsSigned);
9727
9728 Value *MulResult, *HigherBits;
9729 if (IsSigned) {
9730 MulResult = Builder.CreateNSWMul(LHS, RHS);
9731 HigherBits = Builder.CreateAShr(MulResult, 64);
9732 } else {
9733 MulResult = Builder.CreateNUWMul(LHS, RHS);
9734 HigherBits = Builder.CreateLShr(MulResult, 64);
9735 }
9736 HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned);
9737
9738 return HigherBits;
9739 }
9740
9741 // Handle MSVC intrinsics before argument evaluation to prevent double
9742 // evaluation.
9743 if (Optional<MSVCIntrin> MsvcIntId = translateAarch64ToMsvcIntrin(BuiltinID))
9744 return EmitMSVCBuiltinExpr(*MsvcIntId, E);
9745
9746 // Find out if any arguments are required to be integer constant
9747 // expressions.
9748 unsigned ICEArguments = 0;
9749 ASTContext::GetBuiltinTypeError Error;
9750 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
9751 assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None &&
"Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9751, __extension__ __PRETTY_FUNCTION__))
;
9752
9753 llvm::SmallVector<Value*, 4> Ops;
9754 Address PtrOp0 = Address::invalid();
9755 for (unsigned i = 0, e = E->getNumArgs() - 1; i != e; i++) {
9756 if (i == 0) {
9757 switch (BuiltinID) {
9758 case NEON::BI__builtin_neon_vld1_v:
9759 case NEON::BI__builtin_neon_vld1q_v:
9760 case NEON::BI__builtin_neon_vld1_dup_v:
9761 case NEON::BI__builtin_neon_vld1q_dup_v:
9762 case NEON::BI__builtin_neon_vld1_lane_v:
9763 case NEON::BI__builtin_neon_vld1q_lane_v:
9764 case NEON::BI__builtin_neon_vst1_v:
9765 case NEON::BI__builtin_neon_vst1q_v:
9766 case NEON::BI__builtin_neon_vst1_lane_v:
9767 case NEON::BI__builtin_neon_vst1q_lane_v:
9768 // Get the alignment for the argument in addition to the value;
9769 // we'll use it later.
9770 PtrOp0 = EmitPointerWithAlignment(E->getArg(0));
9771 Ops.push_back(PtrOp0.getPointer());
9772 continue;
9773 }
9774 }
9775 if ((ICEArguments & (1 << i)) == 0) {
9776 Ops.push_back(EmitScalarExpr(E->getArg(i)));
9777 } else {
9778 // If this is required to be a constant, constant fold it so that we know
9779 // that the generated intrinsic gets a ConstantInt.
9780 Ops.push_back(llvm::ConstantInt::get(
9781 getLLVMContext(),
9782 *E->getArg(i)->getIntegerConstantExpr(getContext())));
9783 }
9784 }
9785
9786 auto SISDMap = makeArrayRef(AArch64SISDIntrinsicMap);
9787 const ARMVectorIntrinsicInfo *Builtin = findARMVectorIntrinsicInMap(
9788 SISDMap, BuiltinID, AArch64SISDIntrinsicsProvenSorted);
9789
9790 if (Builtin) {
9791 Ops.push_back(EmitScalarExpr(E->getArg(E->getNumArgs() - 1)));
9792 Value *Result = EmitCommonNeonSISDBuiltinExpr(*this, *Builtin, Ops, E);
9793 assert(Result && "SISD intrinsic should have been handled")(static_cast <bool> (Result && "SISD intrinsic should have been handled"
) ? void (0) : __assert_fail ("Result && \"SISD intrinsic should have been handled\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9793, __extension__ __PRETTY_FUNCTION__))
;
9794 return Result;
9795 }
9796
9797 const Expr *Arg = E->getArg(E->getNumArgs()-1);
9798 NeonTypeFlags Type(0);
9799 if (Optional<llvm::APSInt> Result = Arg->getIntegerConstantExpr(getContext()))
9800 // Determine the type of this overloaded NEON intrinsic.
9801 Type = NeonTypeFlags(Result->getZExtValue());
9802
9803 bool usgn = Type.isUnsigned();
9804 bool quad = Type.isQuad();
9805
9806 // Handle non-overloaded intrinsics first.
9807 switch (BuiltinID) {
9808 default: break;
9809 case NEON::BI__builtin_neon_vabsh_f16:
9810 Ops.push_back(EmitScalarExpr(E->getArg(0)));
9811 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, HalfTy), Ops, "vabs");
9812 case NEON::BI__builtin_neon_vaddq_p128: {
9813 llvm::Type *Ty = GetNeonType(this, NeonTypeFlags::Poly128);
9814 Ops.push_back(EmitScalarExpr(E->getArg(1)));
9815 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
9816 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
9817 Ops[0] = Builder.CreateXor(Ops[0], Ops[1]);
9818 llvm::Type *Int128Ty = llvm::Type::getIntNTy(getLLVMContext(), 128);
9819 return Builder.CreateBitCast(Ops[0], Int128Ty);
9820 }
9821 case NEON::BI__builtin_neon_vldrq_p128: {
9822 llvm::Type *Int128Ty = llvm::Type::getIntNTy(getLLVMContext(), 128);
9823 llvm::Type *Int128PTy = llvm::PointerType::get(Int128Ty, 0);
9824 Value *Ptr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int128PTy);
9825 return Builder.CreateAlignedLoad(Int128Ty, Ptr,
9826 CharUnits::fromQuantity(16));
9827 }
9828 case NEON::BI__builtin_neon_vstrq_p128: {
9829 llvm::Type *Int128PTy = llvm::Type::getIntNPtrTy(getLLVMContext(), 128);
9830 Value *Ptr = Builder.CreateBitCast(Ops[0], Int128PTy);
9831 return Builder.CreateDefaultAlignedStore(EmitScalarExpr(E->getArg(1)), Ptr);
9832 }
9833 case NEON::BI__builtin_neon_vcvts_f32_u32:
9834 case NEON::BI__builtin_neon_vcvtd_f64_u64:
9835 usgn = true;
9836 LLVM_FALLTHROUGH[[gnu::fallthrough]];
9837 case NEON::BI__builtin_neon_vcvts_f32_s32:
9838 case NEON::BI__builtin_neon_vcvtd_f64_s64: {
9839 Ops.push_back(EmitScalarExpr(E->getArg(0)));
9840 bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64;
9841 llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty;
9842 llvm::Type *FTy = Is64 ? DoubleTy : FloatTy;
9843 Ops[0] = Builder.CreateBitCast(Ops[0], InTy);
9844 if (usgn)
9845 return Builder.CreateUIToFP(Ops[0], FTy);
9846 return Builder.CreateSIToFP(Ops[0], FTy);
9847 }
9848 case NEON::BI__builtin_neon_vcvth_f16_u16:
9849 case NEON::BI__builtin_neon_vcvth_f16_u32:
9850 case NEON::BI__builtin_neon_vcvth_f16_u64:
9851 usgn = true;
9852 LLVM_FALLTHROUGH[[gnu::fallthrough]];
9853 case NEON::BI__builtin_neon_vcvth_f16_s16:
9854 case NEON::BI__builtin_neon_vcvth_f16_s32:
9855 case NEON::BI__builtin_neon_vcvth_f16_s64: {
9856 Ops.push_back(EmitScalarExpr(E->getArg(0)));
9857 llvm::Type *FTy = HalfTy;
9858 llvm::Type *InTy;
9859 if (Ops[0]->getType()->getPrimitiveSizeInBits() == 64)
9860 InTy = Int64Ty;
9861 else if (Ops[0]->getType()->getPrimitiveSizeInBits() == 32)
9862 InTy = Int32Ty;
9863 else
9864 InTy = Int16Ty;
9865 Ops[0] = Builder.CreateBitCast(Ops[0], InTy);
9866 if (usgn)
9867 return Builder.CreateUIToFP(Ops[0], FTy);
9868 return Builder.CreateSIToFP(Ops[0], FTy);
9869 }
9870 case NEON::BI__builtin_neon_vcvtah_u16_f16:
9871 case NEON::BI__builtin_neon_vcvtmh_u16_f16:
9872 case NEON::BI__builtin_neon_vcvtnh_u16_f16:
9873 case NEON::BI__builtin_neon_vcvtph_u16_f16:
9874 case NEON::BI__builtin_neon_vcvth_u16_f16:
9875 case NEON::BI__builtin_neon_vcvtah_s16_f16:
9876 case NEON::BI__builtin_neon_vcvtmh_s16_f16:
9877 case NEON::BI__builtin_neon_vcvtnh_s16_f16:
9878 case NEON::BI__builtin_neon_vcvtph_s16_f16:
9879 case NEON::BI__builtin_neon_vcvth_s16_f16: {
9880 unsigned Int;
9881 llvm::Type* InTy = Int32Ty;
9882 llvm::Type* FTy = HalfTy;
9883 llvm::Type *Tys[2] = {InTy, FTy};
9884 Ops.push_back(EmitScalarExpr(E->getArg(0)));
9885 switch (BuiltinID) {
9886 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9886)
;
9887 case NEON::BI__builtin_neon_vcvtah_u16_f16:
9888 Int = Intrinsic::aarch64_neon_fcvtau; break;
9889 case NEON::BI__builtin_neon_vcvtmh_u16_f16:
9890 Int = Intrinsic::aarch64_neon_fcvtmu; break;
9891 case NEON::BI__builtin_neon_vcvtnh_u16_f16:
9892 Int = Intrinsic::aarch64_neon_fcvtnu; break;
9893 case NEON::BI__builtin_neon_vcvtph_u16_f16:
9894 Int = Intrinsic::aarch64_neon_fcvtpu; break;
9895 case NEON::BI__builtin_neon_vcvth_u16_f16:
9896 Int = Intrinsic::aarch64_neon_fcvtzu; break;
9897 case NEON::BI__builtin_neon_vcvtah_s16_f16:
9898 Int = Intrinsic::aarch64_neon_fcvtas; break;
9899 case NEON::BI__builtin_neon_vcvtmh_s16_f16:
9900 Int = Intrinsic::aarch64_neon_fcvtms; break;
9901 case NEON::BI__builtin_neon_vcvtnh_s16_f16:
9902 Int = Intrinsic::aarch64_neon_fcvtns; break;
9903 case NEON::BI__builtin_neon_vcvtph_s16_f16:
9904 Int = Intrinsic::aarch64_neon_fcvtps; break;
9905 case NEON::BI__builtin_neon_vcvth_s16_f16:
9906 Int = Intrinsic::aarch64_neon_fcvtzs; break;
9907 }
9908 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvt");
9909 return Builder.CreateTrunc(Ops[0], Int16Ty);
9910 }
9911 case NEON::BI__builtin_neon_vcaleh_f16:
9912 case NEON::BI__builtin_neon_vcalth_f16:
9913 case NEON::BI__builtin_neon_vcageh_f16:
9914 case NEON::BI__builtin_neon_vcagth_f16: {
9915 unsigned Int;
9916 llvm::Type* InTy = Int32Ty;
9917 llvm::Type* FTy = HalfTy;
9918 llvm::Type *Tys[2] = {InTy, FTy};
9919 Ops.push_back(EmitScalarExpr(E->getArg(1)));
9920 switch (BuiltinID) {
9921 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9921)
;
9922 case NEON::BI__builtin_neon_vcageh_f16:
9923 Int = Intrinsic::aarch64_neon_facge; break;
9924 case NEON::BI__builtin_neon_vcagth_f16:
9925 Int = Intrinsic::aarch64_neon_facgt; break;
9926 case NEON::BI__builtin_neon_vcaleh_f16:
9927 Int = Intrinsic::aarch64_neon_facge; std::swap(Ops[0], Ops[1]); break;
9928 case NEON::BI__builtin_neon_vcalth_f16:
9929 Int = Intrinsic::aarch64_neon_facgt; std::swap(Ops[0], Ops[1]); break;
9930 }
9931 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "facg");
9932 return Builder.CreateTrunc(Ops[0], Int16Ty);
9933 }
9934 case NEON::BI__builtin_neon_vcvth_n_s16_f16:
9935 case NEON::BI__builtin_neon_vcvth_n_u16_f16: {
9936 unsigned Int;
9937 llvm::Type* InTy = Int32Ty;
9938 llvm::Type* FTy = HalfTy;
9939 llvm::Type *Tys[2] = {InTy, FTy};
9940 Ops.push_back(EmitScalarExpr(E->getArg(1)));
9941 switch (BuiltinID) {
9942 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9942)
;
9943 case NEON::BI__builtin_neon_vcvth_n_s16_f16:
9944 Int = Intrinsic::aarch64_neon_vcvtfp2fxs; break;
9945 case NEON::BI__builtin_neon_vcvth_n_u16_f16:
9946 Int = Intrinsic::aarch64_neon_vcvtfp2fxu; break;
9947 }
9948 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n");
9949 return Builder.CreateTrunc(Ops[0], Int16Ty);
9950 }
9951 case NEON::BI__builtin_neon_vcvth_n_f16_s16:
9952 case NEON::BI__builtin_neon_vcvth_n_f16_u16: {
9953 unsigned Int;
9954 llvm::Type* FTy = HalfTy;
9955 llvm::Type* InTy = Int32Ty;
9956 llvm::Type *Tys[2] = {FTy, InTy};
9957 Ops.push_back(EmitScalarExpr(E->getArg(1)));
9958 switch (BuiltinID) {
9959 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 9959)
;
9960 case NEON::BI__builtin_neon_vcvth_n_f16_s16:
9961 Int = Intrinsic::aarch64_neon_vcvtfxs2fp;
9962 Ops[0] = Builder.CreateSExt(Ops[0], InTy, "sext");
9963 break;
9964 case NEON::BI__builtin_neon_vcvth_n_f16_u16:
9965 Int = Intrinsic::aarch64_neon_vcvtfxu2fp;
9966 Ops[0] = Builder.CreateZExt(Ops[0], InTy);
9967 break;
9968 }
9969 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n");
9970 }
9971 case NEON::BI__builtin_neon_vpaddd_s64: {
9972 auto *Ty = llvm::FixedVectorType::get(Int64Ty, 2);
9973 Value *Vec = EmitScalarExpr(E->getArg(0));
9974 // The vector is v2f64, so make sure it's bitcast to that.
9975 Vec = Builder.CreateBitCast(Vec, Ty, "v2i64");
9976 llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
9977 llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
9978 Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
9979 Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
9980 // Pairwise addition of a v2f64 into a scalar f64.
9981 return Builder.CreateAdd(Op0, Op1, "vpaddd");
9982 }
9983 case NEON::BI__builtin_neon_vpaddd_f64: {
9984 auto *Ty = llvm::FixedVectorType::get(DoubleTy, 2);
9985 Value *Vec = EmitScalarExpr(E->getArg(0));
9986 // The vector is v2f64, so make sure it's bitcast to that.
9987 Vec = Builder.CreateBitCast(Vec, Ty, "v2f64");
9988 llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
9989 llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
9990 Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
9991 Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
9992 // Pairwise addition of a v2f64 into a scalar f64.
9993 return Builder.CreateFAdd(Op0, Op1, "vpaddd");
9994 }
9995 case NEON::BI__builtin_neon_vpadds_f32: {
9996 auto *Ty = llvm::FixedVectorType::get(FloatTy, 2);
9997 Value *Vec = EmitScalarExpr(E->getArg(0));
9998 // The vector is v2f32, so make sure it's bitcast to that.
9999 Vec = Builder.CreateBitCast(Vec, Ty, "v2f32");
10000 llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
10001 llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
10002 Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
10003 Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
10004 // Pairwise addition of a v2f32 into a scalar f32.
10005 return Builder.CreateFAdd(Op0, Op1, "vpaddd");
10006 }
10007 case NEON::BI__builtin_neon_vceqzd_s64:
10008 case NEON::BI__builtin_neon_vceqzd_f64:
10009 case NEON::BI__builtin_neon_vceqzs_f32:
10010 case NEON::BI__builtin_neon_vceqzh_f16:
10011 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10012 return EmitAArch64CompareBuiltinExpr(
10013 Ops[0], ConvertType(E->getCallReturnType(getContext())),
10014 ICmpInst::FCMP_OEQ, ICmpInst::ICMP_EQ, "vceqz");
10015 case NEON::BI__builtin_neon_vcgezd_s64:
10016 case NEON::BI__builtin_neon_vcgezd_f64:
10017 case NEON::BI__builtin_neon_vcgezs_f32:
10018 case NEON::BI__builtin_neon_vcgezh_f16:
10019 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10020 return EmitAArch64CompareBuiltinExpr(
10021 Ops[0], ConvertType(E->getCallReturnType(getContext())),
10022 ICmpInst::FCMP_OGE, ICmpInst::ICMP_SGE, "vcgez");
10023 case NEON::BI__builtin_neon_vclezd_s64:
10024 case NEON::BI__builtin_neon_vclezd_f64:
10025 case NEON::BI__builtin_neon_vclezs_f32:
10026 case NEON::BI__builtin_neon_vclezh_f16:
10027 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10028 return EmitAArch64CompareBuiltinExpr(
10029 Ops[0], ConvertType(E->getCallReturnType(getContext())),
10030 ICmpInst::FCMP_OLE, ICmpInst::ICMP_SLE, "vclez");
10031 case NEON::BI__builtin_neon_vcgtzd_s64:
10032 case NEON::BI__builtin_neon_vcgtzd_f64:
10033 case NEON::BI__builtin_neon_vcgtzs_f32:
10034 case NEON::BI__builtin_neon_vcgtzh_f16:
10035 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10036 return EmitAArch64CompareBuiltinExpr(
10037 Ops[0], ConvertType(E->getCallReturnType(getContext())),
10038 ICmpInst::FCMP_OGT, ICmpInst::ICMP_SGT, "vcgtz");
10039 case NEON::BI__builtin_neon_vcltzd_s64:
10040 case NEON::BI__builtin_neon_vcltzd_f64:
10041 case NEON::BI__builtin_neon_vcltzs_f32:
10042 case NEON::BI__builtin_neon_vcltzh_f16:
10043 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10044 return EmitAArch64CompareBuiltinExpr(
10045 Ops[0], ConvertType(E->getCallReturnType(getContext())),
10046 ICmpInst::FCMP_OLT, ICmpInst::ICMP_SLT, "vcltz");
10047
10048 case NEON::BI__builtin_neon_vceqzd_u64: {
10049 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10050 Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
10051 Ops[0] =
10052 Builder.CreateICmpEQ(Ops[0], llvm::Constant::getNullValue(Int64Ty));
10053 return Builder.CreateSExt(Ops[0], Int64Ty, "vceqzd");
10054 }
10055 case NEON::BI__builtin_neon_vceqd_f64:
10056 case NEON::BI__builtin_neon_vcled_f64:
10057 case NEON::BI__builtin_neon_vcltd_f64:
10058 case NEON::BI__builtin_neon_vcged_f64:
10059 case NEON::BI__builtin_neon_vcgtd_f64: {
10060 llvm::CmpInst::Predicate P;
10061 switch (BuiltinID) {
10062 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 10062)
;
10063 case NEON::BI__builtin_neon_vceqd_f64: P = llvm::FCmpInst::FCMP_OEQ; break;
10064 case NEON::BI__builtin_neon_vcled_f64: P = llvm::FCmpInst::FCMP_OLE; break;
10065 case NEON::BI__builtin_neon_vcltd_f64: P = llvm::FCmpInst::FCMP_OLT; break;
10066 case NEON::BI__builtin_neon_vcged_f64: P = llvm::FCmpInst::FCMP_OGE; break;
10067 case NEON::BI__builtin_neon_vcgtd_f64: P = llvm::FCmpInst::FCMP_OGT; break;
10068 }
10069 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10070 Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
10071 Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy);
10072 Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
10073 return Builder.CreateSExt(Ops[0], Int64Ty, "vcmpd");
10074 }
10075 case NEON::BI__builtin_neon_vceqs_f32:
10076 case NEON::BI__builtin_neon_vcles_f32:
10077 case NEON::BI__builtin_neon_vclts_f32:
10078 case NEON::BI__builtin_neon_vcges_f32:
10079 case NEON::BI__builtin_neon_vcgts_f32: {
10080 llvm::CmpInst::Predicate P;
10081 switch (BuiltinID) {
10082 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 10082)
;
10083 case NEON::BI__builtin_neon_vceqs_f32: P = llvm::FCmpInst::FCMP_OEQ; break;
10084 case NEON::BI__builtin_neon_vcles_f32: P = llvm::FCmpInst::FCMP_OLE; break;
10085 case NEON::BI__builtin_neon_vclts_f32: P = llvm::FCmpInst::FCMP_OLT; break;
10086 case NEON::BI__builtin_neon_vcges_f32: P = llvm::FCmpInst::FCMP_OGE; break;
10087 case NEON::BI__builtin_neon_vcgts_f32: P = llvm::FCmpInst::FCMP_OGT; break;
10088 }
10089 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10090 Ops[0] = Builder.CreateBitCast(Ops[0], FloatTy);
10091 Ops[1] = Builder.CreateBitCast(Ops[1], FloatTy);
10092 Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
10093 return Builder.CreateSExt(Ops[0], Int32Ty, "vcmpd");
10094 }
10095 case NEON::BI__builtin_neon_vceqh_f16:
10096 case NEON::BI__builtin_neon_vcleh_f16:
10097 case NEON::BI__builtin_neon_vclth_f16:
10098 case NEON::BI__builtin_neon_vcgeh_f16:
10099 case NEON::BI__builtin_neon_vcgth_f16: {
10100 llvm::CmpInst::Predicate P;
10101 switch (BuiltinID) {
10102 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 10102)
;
10103 case NEON::BI__builtin_neon_vceqh_f16: P = llvm::FCmpInst::FCMP_OEQ; break;
10104 case NEON::BI__builtin_neon_vcleh_f16: P = llvm::FCmpInst::FCMP_OLE; break;
10105 case NEON::BI__builtin_neon_vclth_f16: P = llvm::FCmpInst::FCMP_OLT; break;
10106 case NEON::BI__builtin_neon_vcgeh_f16: P = llvm::FCmpInst::FCMP_OGE; break;
10107 case NEON::BI__builtin_neon_vcgth_f16: P = llvm::FCmpInst::FCMP_OGT; break;
10108 }
10109 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10110 Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
10111 Ops[1] = Builder.CreateBitCast(Ops[1], HalfTy);
10112 Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
10113 return Builder.CreateSExt(Ops[0], Int16Ty, "vcmpd");
10114 }
10115 case NEON::BI__builtin_neon_vceqd_s64:
10116 case NEON::BI__builtin_neon_vceqd_u64:
10117 case NEON::BI__builtin_neon_vcgtd_s64:
10118 case NEON::BI__builtin_neon_vcgtd_u64:
10119 case NEON::BI__builtin_neon_vcltd_s64:
10120 case NEON::BI__builtin_neon_vcltd_u64:
10121 case NEON::BI__builtin_neon_vcged_u64:
10122 case NEON::BI__builtin_neon_vcged_s64:
10123 case NEON::BI__builtin_neon_vcled_u64:
10124 case NEON::BI__builtin_neon_vcled_s64: {
10125 llvm::CmpInst::Predicate P;
10126 switch (BuiltinID) {
10127 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 10127)
;
10128 case NEON::BI__builtin_neon_vceqd_s64:
10129 case NEON::BI__builtin_neon_vceqd_u64:P = llvm::ICmpInst::ICMP_EQ;break;
10130 case NEON::BI__builtin_neon_vcgtd_s64:P = llvm::ICmpInst::ICMP_SGT;break;
10131 case NEON::BI__builtin_neon_vcgtd_u64:P = llvm::ICmpInst::ICMP_UGT;break;
10132 case NEON::BI__builtin_neon_vcltd_s64:P = llvm::ICmpInst::ICMP_SLT;break;
10133 case NEON::BI__builtin_neon_vcltd_u64:P = llvm::ICmpInst::ICMP_ULT;break;
10134 case NEON::BI__builtin_neon_vcged_u64:P = llvm::ICmpInst::ICMP_UGE;break;
10135 case NEON::BI__builtin_neon_vcged_s64:P = llvm::ICmpInst::ICMP_SGE;break;
10136 case NEON::BI__builtin_neon_vcled_u64:P = llvm::ICmpInst::ICMP_ULE;break;
10137 case NEON::BI__builtin_neon_vcled_s64:P = llvm::ICmpInst::ICMP_SLE;break;
10138 }
10139 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10140 Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
10141 Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
10142 Ops[0] = Builder.CreateICmp(P, Ops[0], Ops[1]);
10143 return Builder.CreateSExt(Ops[0], Int64Ty, "vceqd");
10144 }
10145 case NEON::BI__builtin_neon_vtstd_s64:
10146 case NEON::BI__builtin_neon_vtstd_u64: {
10147 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10148 Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
10149 Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
10150 Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]);
10151 Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0],
10152 llvm::Constant::getNullValue(Int64Ty));
10153 return Builder.CreateSExt(Ops[0], Int64Ty, "vtstd");
10154 }
10155 case NEON::BI__builtin_neon_vset_lane_i8:
10156 case NEON::BI__builtin_neon_vset_lane_i16:
10157 case NEON::BI__builtin_neon_vset_lane_i32:
10158 case NEON::BI__builtin_neon_vset_lane_i64:
10159 case NEON::BI__builtin_neon_vset_lane_bf16:
10160 case NEON::BI__builtin_neon_vset_lane_f32:
10161 case NEON::BI__builtin_neon_vsetq_lane_i8:
10162 case NEON::BI__builtin_neon_vsetq_lane_i16:
10163 case NEON::BI__builtin_neon_vsetq_lane_i32:
10164 case NEON::BI__builtin_neon_vsetq_lane_i64:
10165 case NEON::BI__builtin_neon_vsetq_lane_bf16:
10166 case NEON::BI__builtin_neon_vsetq_lane_f32:
10167 Ops.push_back(EmitScalarExpr(E->getArg(2)));
10168 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
10169 case NEON::BI__builtin_neon_vset_lane_f64:
10170 // The vector type needs a cast for the v1f64 variant.
10171 Ops[1] =
10172 Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(DoubleTy, 1));
10173 Ops.push_back(EmitScalarExpr(E->getArg(2)));
10174 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
10175 case NEON::BI__builtin_neon_vsetq_lane_f64:
10176 // The vector type needs a cast for the v2f64 variant.
10177 Ops[1] =
10178 Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(DoubleTy, 2));
10179 Ops.push_back(EmitScalarExpr(E->getArg(2)));
10180 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
10181
10182 case NEON::BI__builtin_neon_vget_lane_i8:
10183 case NEON::BI__builtin_neon_vdupb_lane_i8:
10184 Ops[0] =
10185 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int8Ty, 8));
10186 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10187 "vget_lane");
10188 case NEON::BI__builtin_neon_vgetq_lane_i8:
10189 case NEON::BI__builtin_neon_vdupb_laneq_i8:
10190 Ops[0] =
10191 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int8Ty, 16));
10192 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10193 "vgetq_lane");
10194 case NEON::BI__builtin_neon_vget_lane_i16:
10195 case NEON::BI__builtin_neon_vduph_lane_i16:
10196 Ops[0] =
10197 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int16Ty, 4));
10198 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10199 "vget_lane");
10200 case NEON::BI__builtin_neon_vgetq_lane_i16:
10201 case NEON::BI__builtin_neon_vduph_laneq_i16:
10202 Ops[0] =
10203 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int16Ty, 8));
10204 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10205 "vgetq_lane");
10206 case NEON::BI__builtin_neon_vget_lane_i32:
10207 case NEON::BI__builtin_neon_vdups_lane_i32:
10208 Ops[0] =
10209 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int32Ty, 2));
10210 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10211 "vget_lane");
10212 case NEON::BI__builtin_neon_vdups_lane_f32:
10213 Ops[0] =
10214 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(FloatTy, 2));
10215 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10216 "vdups_lane");
10217 case NEON::BI__builtin_neon_vgetq_lane_i32:
10218 case NEON::BI__builtin_neon_vdups_laneq_i32:
10219 Ops[0] =
10220 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int32Ty, 4));
10221 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10222 "vgetq_lane");
10223 case NEON::BI__builtin_neon_vget_lane_i64:
10224 case NEON::BI__builtin_neon_vdupd_lane_i64:
10225 Ops[0] =
10226 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 1));
10227 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10228 "vget_lane");
10229 case NEON::BI__builtin_neon_vdupd_lane_f64:
10230 Ops[0] =
10231 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(DoubleTy, 1));
10232 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10233 "vdupd_lane");
10234 case NEON::BI__builtin_neon_vgetq_lane_i64:
10235 case NEON::BI__builtin_neon_vdupd_laneq_i64:
10236 Ops[0] =
10237 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 2));
10238 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10239 "vgetq_lane");
10240 case NEON::BI__builtin_neon_vget_lane_f32:
10241 Ops[0] =
10242 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(FloatTy, 2));
10243 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10244 "vget_lane");
10245 case NEON::BI__builtin_neon_vget_lane_f64:
10246 Ops[0] =
10247 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(DoubleTy, 1));
10248 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10249 "vget_lane");
10250 case NEON::BI__builtin_neon_vgetq_lane_f32:
10251 case NEON::BI__builtin_neon_vdups_laneq_f32:
10252 Ops[0] =
10253 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(FloatTy, 4));
10254 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10255 "vgetq_lane");
10256 case NEON::BI__builtin_neon_vgetq_lane_f64:
10257 case NEON::BI__builtin_neon_vdupd_laneq_f64:
10258 Ops[0] =
10259 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(DoubleTy, 2));
10260 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10261 "vgetq_lane");
10262 case NEON::BI__builtin_neon_vaddh_f16:
10263 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10264 return Builder.CreateFAdd(Ops[0], Ops[1], "vaddh");
10265 case NEON::BI__builtin_neon_vsubh_f16:
10266 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10267 return Builder.CreateFSub(Ops[0], Ops[1], "vsubh");
10268 case NEON::BI__builtin_neon_vmulh_f16:
10269 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10270 return Builder.CreateFMul(Ops[0], Ops[1], "vmulh");
10271 case NEON::BI__builtin_neon_vdivh_f16:
10272 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10273 return Builder.CreateFDiv(Ops[0], Ops[1], "vdivh");
10274 case NEON::BI__builtin_neon_vfmah_f16:
10275 // NEON intrinsic puts accumulator first, unlike the LLVM fma.
10276 return emitCallMaybeConstrainedFPBuiltin(
10277 *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, HalfTy,
10278 {EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), Ops[0]});
10279 case NEON::BI__builtin_neon_vfmsh_f16: {
10280 // FIXME: This should be an fneg instruction:
10281 Value *Zero = llvm::ConstantFP::getZeroValueForNegation(HalfTy);
10282 Value* Sub = Builder.CreateFSub(Zero, EmitScalarExpr(E->getArg(1)), "vsubh");
10283
10284 // NEON intrinsic puts accumulator first, unlike the LLVM fma.
10285 return emitCallMaybeConstrainedFPBuiltin(
10286 *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, HalfTy,
10287 {Sub, EmitScalarExpr(E->getArg(2)), Ops[0]});
10288 }
10289 case NEON::BI__builtin_neon_vaddd_s64:
10290 case NEON::BI__builtin_neon_vaddd_u64:
10291 return Builder.CreateAdd(Ops[0], EmitScalarExpr(E->getArg(1)), "vaddd");
10292 case NEON::BI__builtin_neon_vsubd_s64:
10293 case NEON::BI__builtin_neon_vsubd_u64:
10294 return Builder.CreateSub(Ops[0], EmitScalarExpr(E->getArg(1)), "vsubd");
10295 case NEON::BI__builtin_neon_vqdmlalh_s16:
10296 case NEON::BI__builtin_neon_vqdmlslh_s16: {
10297 SmallVector<Value *, 2> ProductOps;
10298 ProductOps.push_back(vectorWrapScalar16(Ops[1]));
10299 ProductOps.push_back(vectorWrapScalar16(EmitScalarExpr(E->getArg(2))));
10300 auto *VTy = llvm::FixedVectorType::get(Int32Ty, 4);
10301 Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy),
10302 ProductOps, "vqdmlXl");
10303 Constant *CI = ConstantInt::get(SizeTy, 0);
10304 Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0");
10305
10306 unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlalh_s16
10307 ? Intrinsic::aarch64_neon_sqadd
10308 : Intrinsic::aarch64_neon_sqsub;
10309 return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int32Ty), Ops, "vqdmlXl");
10310 }
10311 case NEON::BI__builtin_neon_vqshlud_n_s64: {
10312 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10313 Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty);
10314 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqshlu, Int64Ty),
10315 Ops, "vqshlu_n");
10316 }
10317 case NEON::BI__builtin_neon_vqshld_n_u64:
10318 case NEON::BI__builtin_neon_vqshld_n_s64: {
10319 unsigned Int = BuiltinID == NEON::BI__builtin_neon_vqshld_n_u64
10320 ? Intrinsic::aarch64_neon_uqshl
10321 : Intrinsic::aarch64_neon_sqshl;
10322 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10323 Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty);
10324 return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vqshl_n");
10325 }
10326 case NEON::BI__builtin_neon_vrshrd_n_u64:
10327 case NEON::BI__builtin_neon_vrshrd_n_s64: {
10328 unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrshrd_n_u64
10329 ? Intrinsic::aarch64_neon_urshl
10330 : Intrinsic::aarch64_neon_srshl;
10331 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10332 int SV = cast<ConstantInt>(Ops[1])->getSExtValue();
10333 Ops[1] = ConstantInt::get(Int64Ty, -SV);
10334 return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vrshr_n");
10335 }
10336 case NEON::BI__builtin_neon_vrsrad_n_u64:
10337 case NEON::BI__builtin_neon_vrsrad_n_s64: {
10338 unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrsrad_n_u64
10339 ? Intrinsic::aarch64_neon_urshl
10340 : Intrinsic::aarch64_neon_srshl;
10341 Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
10342 Ops.push_back(Builder.CreateNeg(EmitScalarExpr(E->getArg(2))));
10343 Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Int64Ty),
10344 {Ops[1], Builder.CreateSExt(Ops[2], Int64Ty)});
10345 return Builder.CreateAdd(Ops[0], Builder.CreateBitCast(Ops[1], Int64Ty));
10346 }
10347 case NEON::BI__builtin_neon_vshld_n_s64:
10348 case NEON::BI__builtin_neon_vshld_n_u64: {
10349 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
10350 return Builder.CreateShl(
10351 Ops[0], ConstantInt::get(Int64Ty, Amt->getZExtValue()), "shld_n");
10352 }
10353 case NEON::BI__builtin_neon_vshrd_n_s64: {
10354 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
10355 return Builder.CreateAShr(
10356 Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
10357 Amt->getZExtValue())),
10358 "shrd_n");
10359 }
10360 case NEON::BI__builtin_neon_vshrd_n_u64: {
10361 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
10362 uint64_t ShiftAmt = Amt->getZExtValue();
10363 // Right-shifting an unsigned value by its size yields 0.
10364 if (ShiftAmt == 64)
10365 return ConstantInt::get(Int64Ty, 0);
10366 return Builder.CreateLShr(Ops[0], ConstantInt::get(Int64Ty, ShiftAmt),
10367 "shrd_n");
10368 }
10369 case NEON::BI__builtin_neon_vsrad_n_s64: {
10370 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2)));
10371 Ops[1] = Builder.CreateAShr(
10372 Ops[1], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
10373 Amt->getZExtValue())),
10374 "shrd_n");
10375 return Builder.CreateAdd(Ops[0], Ops[1]);
10376 }
10377 case NEON::BI__builtin_neon_vsrad_n_u64: {
10378 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2)));
10379 uint64_t ShiftAmt = Amt->getZExtValue();
10380 // Right-shifting an unsigned value by its size yields 0.
10381 // As Op + 0 = Op, return Ops[0] directly.
10382 if (ShiftAmt == 64)
10383 return Ops[0];
10384 Ops[1] = Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, ShiftAmt),
10385 "shrd_n");
10386 return Builder.CreateAdd(Ops[0], Ops[1]);
10387 }
10388 case NEON::BI__builtin_neon_vqdmlalh_lane_s16:
10389 case NEON::BI__builtin_neon_vqdmlalh_laneq_s16:
10390 case NEON::BI__builtin_neon_vqdmlslh_lane_s16:
10391 case NEON::BI__builtin_neon_vqdmlslh_laneq_s16: {
10392 Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)),
10393 "lane");
10394 SmallVector<Value *, 2> ProductOps;
10395 ProductOps.push_back(vectorWrapScalar16(Ops[1]));
10396 ProductOps.push_back(vectorWrapScalar16(Ops[2]));
10397 auto *VTy = llvm::FixedVectorType::get(Int32Ty, 4);
10398 Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy),
10399 ProductOps, "vqdmlXl");
10400 Constant *CI = ConstantInt::get(SizeTy, 0);
10401 Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0");
10402 Ops.pop_back();
10403
10404 unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlalh_lane_s16 ||
10405 BuiltinID == NEON::BI__builtin_neon_vqdmlalh_laneq_s16)
10406 ? Intrinsic::aarch64_neon_sqadd
10407 : Intrinsic::aarch64_neon_sqsub;
10408 return EmitNeonCall(CGM.getIntrinsic(AccInt, Int32Ty), Ops, "vqdmlXl");
10409 }
10410 case NEON::BI__builtin_neon_vqdmlals_s32:
10411 case NEON::BI__builtin_neon_vqdmlsls_s32: {
10412 SmallVector<Value *, 2> ProductOps;
10413 ProductOps.push_back(Ops[1]);
10414 ProductOps.push_back(EmitScalarExpr(E->getArg(2)));
10415 Ops[1] =
10416 EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar),
10417 ProductOps, "vqdmlXl");
10418
10419 unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlals_s32
10420 ? Intrinsic::aarch64_neon_sqadd
10421 : Intrinsic::aarch64_neon_sqsub;
10422 return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int64Ty), Ops, "vqdmlXl");
10423 }
10424 case NEON::BI__builtin_neon_vqdmlals_lane_s32:
10425 case NEON::BI__builtin_neon_vqdmlals_laneq_s32:
10426 case NEON::BI__builtin_neon_vqdmlsls_lane_s32:
10427 case NEON::BI__builtin_neon_vqdmlsls_laneq_s32: {
10428 Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)),
10429 "lane");
10430 SmallVector<Value *, 2> ProductOps;
10431 ProductOps.push_back(Ops[1]);
10432 ProductOps.push_back(Ops[2]);
10433 Ops[1] =
10434 EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar),
10435 ProductOps, "vqdmlXl");
10436 Ops.pop_back();
10437
10438 unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlals_lane_s32 ||
10439 BuiltinID == NEON::BI__builtin_neon_vqdmlals_laneq_s32)
10440 ? Intrinsic::aarch64_neon_sqadd
10441 : Intrinsic::aarch64_neon_sqsub;
10442 return EmitNeonCall(CGM.getIntrinsic(AccInt, Int64Ty), Ops, "vqdmlXl");
10443 }
10444 case NEON::BI__builtin_neon_vget_lane_bf16:
10445 case NEON::BI__builtin_neon_vduph_lane_bf16:
10446 case NEON::BI__builtin_neon_vduph_lane_f16: {
10447 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10448 "vget_lane");
10449 }
10450 case NEON::BI__builtin_neon_vgetq_lane_bf16:
10451 case NEON::BI__builtin_neon_vduph_laneq_bf16:
10452 case NEON::BI__builtin_neon_vduph_laneq_f16: {
10453 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
10454 "vgetq_lane");
10455 }
10456
10457 case AArch64::BI_InterlockedAdd: {
10458 Value *Arg0 = EmitScalarExpr(E->getArg(0));
10459 Value *Arg1 = EmitScalarExpr(E->getArg(1));
10460 AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
10461 AtomicRMWInst::Add, Arg0, Arg1,
10462 llvm::AtomicOrdering::SequentiallyConsistent);
10463 return Builder.CreateAdd(RMWI, Arg1);
10464 }
10465 }
10466
10467 llvm::FixedVectorType *VTy = GetNeonType(this, Type);
10468 llvm::Type *Ty = VTy;
10469 if (!Ty)
10470 return nullptr;
10471
10472 // Not all intrinsics handled by the common case work for AArch64 yet, so only
10473 // defer to common code if it's been added to our special map.
10474 Builtin = findARMVectorIntrinsicInMap(AArch64SIMDIntrinsicMap, BuiltinID,
10475 AArch64SIMDIntrinsicsProvenSorted);
10476
10477 if (Builtin)
10478 return EmitCommonNeonBuiltinExpr(
10479 Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic,
10480 Builtin->NameHint, Builtin->TypeModifier, E, Ops,
10481 /*never use addresses*/ Address::invalid(), Address::invalid(), Arch);
10482
10483 if (Value *V = EmitAArch64TblBuiltinExpr(*this, BuiltinID, E, Ops, Arch))
10484 return V;
10485
10486 unsigned Int;
10487 switch (BuiltinID) {
10488 default: return nullptr;
10489 case NEON::BI__builtin_neon_vbsl_v:
10490 case NEON::BI__builtin_neon_vbslq_v: {
10491 llvm::Type *BitTy = llvm::VectorType::getInteger(VTy);
10492 Ops[0] = Builder.CreateBitCast(Ops[0], BitTy, "vbsl");
10493 Ops[1] = Builder.CreateBitCast(Ops[1], BitTy, "vbsl");
10494 Ops[2] = Builder.CreateBitCast(Ops[2], BitTy, "vbsl");
10495
10496 Ops[1] = Builder.CreateAnd(Ops[0], Ops[1], "vbsl");
10497 Ops[2] = Builder.CreateAnd(Builder.CreateNot(Ops[0]), Ops[2], "vbsl");
10498 Ops[0] = Builder.CreateOr(Ops[1], Ops[2], "vbsl");
10499 return Builder.CreateBitCast(Ops[0], Ty);
10500 }
10501 case NEON::BI__builtin_neon_vfma_lane_v:
10502 case NEON::BI__builtin_neon_vfmaq_lane_v: { // Only used for FP types
10503 // The ARM builtins (and instructions) have the addend as the first
10504 // operand, but the 'fma' intrinsics have it last. Swap it around here.
10505 Value *Addend = Ops[0];
10506 Value *Multiplicand = Ops[1];
10507 Value *LaneSource = Ops[2];
10508 Ops[0] = Multiplicand;
10509 Ops[1] = LaneSource;
10510 Ops[2] = Addend;
10511
10512 // Now adjust things to handle the lane access.
10513 auto *SourceTy = BuiltinID == NEON::BI__builtin_neon_vfmaq_lane_v
10514 ? llvm::FixedVectorType::get(VTy->getElementType(),
10515 VTy->getNumElements() / 2)
10516 : VTy;
10517 llvm::Constant *cst = cast<Constant>(Ops[3]);
10518 Value *SV = llvm::ConstantVector::getSplat(VTy->getElementCount(), cst);
10519 Ops[1] = Builder.CreateBitCast(Ops[1], SourceTy);
10520 Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV, "lane");
10521
10522 Ops.pop_back();
10523 Int = Builder.getIsFPConstrained() ? Intrinsic::experimental_constrained_fma
10524 : Intrinsic::fma;
10525 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "fmla");
10526 }
10527 case NEON::BI__builtin_neon_vfma_laneq_v: {
10528 auto *VTy = cast<llvm::FixedVectorType>(Ty);
10529 // v1f64 fma should be mapped to Neon scalar f64 fma
10530 if (VTy && VTy->getElementType() == DoubleTy) {
10531 Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
10532 Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy);
10533 llvm::FixedVectorType *VTy =
10534 GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, true));
10535 Ops[2] = Builder.CreateBitCast(Ops[2], VTy);
10536 Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract");
10537 Value *Result;
10538 Result = emitCallMaybeConstrainedFPBuiltin(
10539 *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma,
10540 DoubleTy, {Ops[1], Ops[2], Ops[0]});
10541 return Builder.CreateBitCast(Result, Ty);
10542 }
10543 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
10544 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
10545
10546 auto *STy = llvm::FixedVectorType::get(VTy->getElementType(),
10547 VTy->getNumElements() * 2);
10548 Ops[2] = Builder.CreateBitCast(Ops[2], STy);
10549 Value *SV = llvm::ConstantVector::getSplat(VTy->getElementCount(),
10550 cast<ConstantInt>(Ops[3]));
10551 Ops[2] = Builder.CreateShuffleVector(Ops[2], Ops[2], SV, "lane");
10552
10553 return emitCallMaybeConstrainedFPBuiltin(
10554 *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, Ty,
10555 {Ops[2], Ops[1], Ops[0]});
10556 }
10557 case NEON::BI__builtin_neon_vfmaq_laneq_v: {
10558 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
10559 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
10560
10561 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
10562 Ops[2] = EmitNeonSplat(Ops[2], cast<ConstantInt>(Ops[3]));
10563 return emitCallMaybeConstrainedFPBuiltin(
10564 *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, Ty,
10565 {Ops[2], Ops[1], Ops[0]});
10566 }
10567 case NEON::BI__builtin_neon_vfmah_lane_f16:
10568 case NEON::BI__builtin_neon_vfmas_lane_f32:
10569 case NEON::BI__builtin_neon_vfmah_laneq_f16:
10570 case NEON::BI__builtin_neon_vfmas_laneq_f32:
10571 case NEON::BI__builtin_neon_vfmad_lane_f64:
10572 case NEON::BI__builtin_neon_vfmad_laneq_f64: {
10573 Ops.push_back(EmitScalarExpr(E->getArg(3)));
10574 llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext()));
10575 Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract");
10576 return emitCallMaybeConstrainedFPBuiltin(
10577 *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, Ty,
10578 {Ops[1], Ops[2], Ops[0]});
10579 }
10580 case NEON::BI__builtin_neon_vmull_v:
10581 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
10582 Int = usgn ? Intrinsic::aarch64_neon_umull : Intrinsic::aarch64_neon_smull;
10583 if (Type.isPoly()) Int = Intrinsic::aarch64_neon_pmull;
10584 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull");
10585 case NEON::BI__builtin_neon_vmax_v:
10586 case NEON::BI__builtin_neon_vmaxq_v:
10587 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
10588 Int = usgn ? Intrinsic::aarch64_neon_umax : Intrinsic::aarch64_neon_smax;
10589 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmax;
10590 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmax");
10591 case NEON::BI__builtin_neon_vmaxh_f16: {
10592 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10593 Int = Intrinsic::aarch64_neon_fmax;
10594 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmax");
10595 }
10596 case NEON::BI__builtin_neon_vmin_v:
10597 case NEON::BI__builtin_neon_vminq_v:
10598 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
10599 Int = usgn ? Intrinsic::aarch64_neon_umin : Intrinsic::aarch64_neon_smin;
10600 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmin;
10601 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmin");
10602 case NEON::BI__builtin_neon_vminh_f16: {
10603 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10604 Int = Intrinsic::aarch64_neon_fmin;
10605 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmin");
10606 }
10607 case NEON::BI__builtin_neon_vabd_v:
10608 case NEON::BI__builtin_neon_vabdq_v:
10609 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
10610 Int = usgn ? Intrinsic::aarch64_neon_uabd : Intrinsic::aarch64_neon_sabd;
10611 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fabd;
10612 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vabd");
10613 case NEON::BI__builtin_neon_vpadal_v:
10614 case NEON::BI__builtin_neon_vpadalq_v: {
10615 unsigned ArgElts = VTy->getNumElements();
10616 llvm::IntegerType *EltTy = cast<IntegerType>(VTy->getElementType());
10617 unsigned BitWidth = EltTy->getBitWidth();
10618 auto *ArgTy = llvm::FixedVectorType::get(
10619 llvm::IntegerType::get(getLLVMContext(), BitWidth / 2), 2 * ArgElts);
10620 llvm::Type* Tys[2] = { VTy, ArgTy };
10621 Int = usgn ? Intrinsic::aarch64_neon_uaddlp : Intrinsic::aarch64_neon_saddlp;
10622 SmallVector<llvm::Value*, 1> TmpOps;
10623 TmpOps.push_back(Ops[1]);
10624 Function *F = CGM.getIntrinsic(Int, Tys);
10625 llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vpadal");
10626 llvm::Value *addend = Builder.CreateBitCast(Ops[0], tmp->getType());
10627 return Builder.CreateAdd(tmp, addend);
10628 }
10629 case NEON::BI__builtin_neon_vpmin_v:
10630 case NEON::BI__builtin_neon_vpminq_v:
10631 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
10632 Int = usgn ? Intrinsic::aarch64_neon_uminp : Intrinsic::aarch64_neon_sminp;
10633 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fminp;
10634 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmin");
10635 case NEON::BI__builtin_neon_vpmax_v:
10636 case NEON::BI__builtin_neon_vpmaxq_v:
10637 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
10638 Int = usgn ? Intrinsic::aarch64_neon_umaxp : Intrinsic::aarch64_neon_smaxp;
10639 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmaxp;
10640 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmax");
10641 case NEON::BI__builtin_neon_vminnm_v:
10642 case NEON::BI__builtin_neon_vminnmq_v:
10643 Int = Intrinsic::aarch64_neon_fminnm;
10644 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vminnm");
10645 case NEON::BI__builtin_neon_vminnmh_f16:
10646 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10647 Int = Intrinsic::aarch64_neon_fminnm;
10648 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vminnm");
10649 case NEON::BI__builtin_neon_vmaxnm_v:
10650 case NEON::BI__builtin_neon_vmaxnmq_v:
10651 Int = Intrinsic::aarch64_neon_fmaxnm;
10652 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmaxnm");
10653 case NEON::BI__builtin_neon_vmaxnmh_f16:
10654 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10655 Int = Intrinsic::aarch64_neon_fmaxnm;
10656 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmaxnm");
10657 case NEON::BI__builtin_neon_vrecpss_f32: {
10658 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10659 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, FloatTy),
10660 Ops, "vrecps");
10661 }
10662 case NEON::BI__builtin_neon_vrecpsd_f64:
10663 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10664 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, DoubleTy),
10665 Ops, "vrecps");
10666 case NEON::BI__builtin_neon_vrecpsh_f16:
10667 Ops.push_back(EmitScalarExpr(E->getArg(1)));
10668 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, HalfTy),
10669 Ops, "vrecps");
10670 case NEON::BI__builtin_neon_vqshrun_n_v:
10671 Int = Intrinsic::aarch64_neon_sqshrun;
10672 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrun_n");
10673 case NEON::BI__builtin_neon_vqrshrun_n_v:
10674 Int = Intrinsic::aarch64_neon_sqrshrun;
10675 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrun_n");
10676 case NEON::BI__builtin_neon_vqshrn_n_v:
10677 Int = usgn ? Intrinsic::aarch64_neon_uqshrn : Intrinsic::aarch64_neon_sqshrn;
10678 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n");
10679 case NEON::BI__builtin_neon_vrshrn_n_v:
10680 Int = Intrinsic::aarch64_neon_rshrn;
10681 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshrn_n");
10682 case NEON::BI__builtin_neon_vqrshrn_n_v:
10683 Int = usgn ? Intrinsic::aarch64_neon_uqrshrn : Intrinsic::aarch64_neon_sqrshrn;
10684 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n");
10685 case NEON::BI__builtin_neon_vrndah_f16: {
10686 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10687 Int = Builder.getIsFPConstrained()
10688 ? Intrinsic::experimental_constrained_round
10689 : Intrinsic::round;
10690 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrnda");
10691 }
10692 case NEON::BI__builtin_neon_vrnda_v:
10693 case NEON::BI__builtin_neon_vrndaq_v: {
10694 Int = Builder.getIsFPConstrained()
10695 ? Intrinsic::experimental_constrained_round
10696 : Intrinsic::round;
10697 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnda");
10698 }
10699 case NEON::BI__builtin_neon_vrndih_f16: {
10700 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10701 Int = Builder.getIsFPConstrained()
10702 ? Intrinsic::experimental_constrained_nearbyint
10703 : Intrinsic::nearbyint;
10704 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndi");
10705 }
10706 case NEON::BI__builtin_neon_vrndmh_f16: {
10707 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10708 Int = Builder.getIsFPConstrained()
10709 ? Intrinsic::experimental_constrained_floor
10710 : Intrinsic::floor;
10711 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndm");
10712 }
10713 case NEON::BI__builtin_neon_vrndm_v:
10714 case NEON::BI__builtin_neon_vrndmq_v: {
10715 Int = Builder.getIsFPConstrained()
10716 ? Intrinsic::experimental_constrained_floor
10717 : Intrinsic::floor;
10718 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndm");
10719 }
10720 case NEON::BI__builtin_neon_vrndnh_f16: {
10721 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10722 Int = Builder.getIsFPConstrained()
10723 ? Intrinsic::experimental_constrained_roundeven
10724 : Intrinsic::roundeven;
10725 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndn");
10726 }
10727 case NEON::BI__builtin_neon_vrndn_v:
10728 case NEON::BI__builtin_neon_vrndnq_v: {
10729 Int = Builder.getIsFPConstrained()
10730 ? Intrinsic::experimental_constrained_roundeven
10731 : Intrinsic::roundeven;
10732 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndn");
10733 }
10734 case NEON::BI__builtin_neon_vrndns_f32: {
10735 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10736 Int = Builder.getIsFPConstrained()
10737 ? Intrinsic::experimental_constrained_roundeven
10738 : Intrinsic::roundeven;
10739 return EmitNeonCall(CGM.getIntrinsic(Int, FloatTy), Ops, "vrndn");
10740 }
10741 case NEON::BI__builtin_neon_vrndph_f16: {
10742 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10743 Int = Builder.getIsFPConstrained()
10744 ? Intrinsic::experimental_constrained_ceil
10745 : Intrinsic::ceil;
10746 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndp");
10747 }
10748 case NEON::BI__builtin_neon_vrndp_v:
10749 case NEON::BI__builtin_neon_vrndpq_v: {
10750 Int = Builder.getIsFPConstrained()
10751 ? Intrinsic::experimental_constrained_ceil
10752 : Intrinsic::ceil;
10753 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndp");
10754 }
10755 case NEON::BI__builtin_neon_vrndxh_f16: {
10756 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10757 Int = Builder.getIsFPConstrained()
10758 ? Intrinsic::experimental_constrained_rint
10759 : Intrinsic::rint;
10760 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndx");
10761 }
10762 case NEON::BI__builtin_neon_vrndx_v:
10763 case NEON::BI__builtin_neon_vrndxq_v: {
10764 Int = Builder.getIsFPConstrained()
10765 ? Intrinsic::experimental_constrained_rint
10766 : Intrinsic::rint;
10767 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndx");
10768 }
10769 case NEON::BI__builtin_neon_vrndh_f16: {
10770 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10771 Int = Builder.getIsFPConstrained()
10772 ? Intrinsic::experimental_constrained_trunc
10773 : Intrinsic::trunc;
10774 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndz");
10775 }
10776 case NEON::BI__builtin_neon_vrnd32x_v:
10777 case NEON::BI__builtin_neon_vrnd32xq_v: {
10778 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10779 Int = Intrinsic::aarch64_neon_frint32x;
10780 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnd32x");
10781 }
10782 case NEON::BI__builtin_neon_vrnd32z_v:
10783 case NEON::BI__builtin_neon_vrnd32zq_v: {
10784 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10785 Int = Intrinsic::aarch64_neon_frint32z;
10786 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnd32z");
10787 }
10788 case NEON::BI__builtin_neon_vrnd64x_v:
10789 case NEON::BI__builtin_neon_vrnd64xq_v: {
10790 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10791 Int = Intrinsic::aarch64_neon_frint64x;
10792 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnd64x");
10793 }
10794 case NEON::BI__builtin_neon_vrnd64z_v:
10795 case NEON::BI__builtin_neon_vrnd64zq_v: {
10796 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10797 Int = Intrinsic::aarch64_neon_frint64z;
10798 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnd64z");
10799 }
10800 case NEON::BI__builtin_neon_vrnd_v:
10801 case NEON::BI__builtin_neon_vrndq_v: {
10802 Int = Builder.getIsFPConstrained()
10803 ? Intrinsic::experimental_constrained_trunc
10804 : Intrinsic::trunc;
10805 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndz");
10806 }
10807 case NEON::BI__builtin_neon_vcvt_f64_v:
10808 case NEON::BI__builtin_neon_vcvtq_f64_v:
10809 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
10810 Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, quad));
10811 return usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
10812 : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
10813 case NEON::BI__builtin_neon_vcvt_f64_f32: {
10814 assert(Type.getEltType() == NeonTypeFlags::Float64 && quad &&(static_cast <bool> (Type.getEltType() == NeonTypeFlags
::Float64 && quad && "unexpected vcvt_f64_f32 builtin"
) ? void (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float64 && quad && \"unexpected vcvt_f64_f32 builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 10815, __extension__ __PRETTY_FUNCTION__))
10815 "unexpected vcvt_f64_f32 builtin")(static_cast <bool> (Type.getEltType() == NeonTypeFlags
::Float64 && quad && "unexpected vcvt_f64_f32 builtin"
) ? void (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float64 && quad && \"unexpected vcvt_f64_f32 builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 10815, __extension__ __PRETTY_FUNCTION__))
;
10816 NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float32, false, false);
10817 Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag));
10818
10819 return Builder.CreateFPExt(Ops[0], Ty, "vcvt");
10820 }
10821 case NEON::BI__builtin_neon_vcvt_f32_f64: {
10822 assert(Type.getEltType() == NeonTypeFlags::Float32 &&(static_cast <bool> (Type.getEltType() == NeonTypeFlags
::Float32 && "unexpected vcvt_f32_f64 builtin") ? void
(0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float32 && \"unexpected vcvt_f32_f64 builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 10823, __extension__ __PRETTY_FUNCTION__))
10823 "unexpected vcvt_f32_f64 builtin")(static_cast <bool> (Type.getEltType() == NeonTypeFlags
::Float32 && "unexpected vcvt_f32_f64 builtin") ? void
(0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float32 && \"unexpected vcvt_f32_f64 builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 10823, __extension__ __PRETTY_FUNCTION__))
;
10824 NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float64, false, true);
10825 Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag));
10826
10827 return Builder.CreateFPTrunc(Ops[0], Ty, "vcvt");
10828 }
10829 case NEON::BI__builtin_neon_vcvt_s32_v:
10830 case NEON::BI__builtin_neon_vcvt_u32_v:
10831 case NEON::BI__builtin_neon_vcvt_s64_v:
10832 case NEON::BI__builtin_neon_vcvt_u64_v:
10833 case NEON::BI__builtin_neon_vcvt_s16_v:
10834 case NEON::BI__builtin_neon_vcvt_u16_v:
10835 case NEON::BI__builtin_neon_vcvtq_s32_v:
10836 case NEON::BI__builtin_neon_vcvtq_u32_v:
10837 case NEON::BI__builtin_neon_vcvtq_s64_v:
10838 case NEON::BI__builtin_neon_vcvtq_u64_v:
10839 case NEON::BI__builtin_neon_vcvtq_s16_v:
10840 case NEON::BI__builtin_neon_vcvtq_u16_v: {
10841 Int =
10842 usgn ? Intrinsic::aarch64_neon_fcvtzu : Intrinsic::aarch64_neon_fcvtzs;
10843 llvm::Type *Tys[2] = {Ty, GetFloatNeonType(this, Type)};
10844 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtz");
10845 }
10846 case NEON::BI__builtin_neon_vcvta_s16_v:
10847 case NEON::BI__builtin_neon_vcvta_u16_v:
10848 case NEON::BI__builtin_neon_vcvta_s32_v:
10849 case NEON::BI__builtin_neon_vcvtaq_s16_v:
10850 case NEON::BI__builtin_neon_vcvtaq_s32_v:
10851 case NEON::BI__builtin_neon_vcvta_u32_v:
10852 case NEON::BI__builtin_neon_vcvtaq_u16_v:
10853 case NEON::BI__builtin_neon_vcvtaq_u32_v:
10854 case NEON::BI__builtin_neon_vcvta_s64_v:
10855 case NEON::BI__builtin_neon_vcvtaq_s64_v:
10856 case NEON::BI__builtin_neon_vcvta_u64_v:
10857 case NEON::BI__builtin_neon_vcvtaq_u64_v: {
10858 Int = usgn ? Intrinsic::aarch64_neon_fcvtau : Intrinsic::aarch64_neon_fcvtas;
10859 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
10860 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvta");
10861 }
10862 case NEON::BI__builtin_neon_vcvtm_s16_v:
10863 case NEON::BI__builtin_neon_vcvtm_s32_v:
10864 case NEON::BI__builtin_neon_vcvtmq_s16_v:
10865 case NEON::BI__builtin_neon_vcvtmq_s32_v:
10866 case NEON::BI__builtin_neon_vcvtm_u16_v:
10867 case NEON::BI__builtin_neon_vcvtm_u32_v:
10868 case NEON::BI__builtin_neon_vcvtmq_u16_v:
10869 case NEON::BI__builtin_neon_vcvtmq_u32_v:
10870 case NEON::BI__builtin_neon_vcvtm_s64_v:
10871 case NEON::BI__builtin_neon_vcvtmq_s64_v:
10872 case NEON::BI__builtin_neon_vcvtm_u64_v:
10873 case NEON::BI__builtin_neon_vcvtmq_u64_v: {
10874 Int = usgn ? Intrinsic::aarch64_neon_fcvtmu : Intrinsic::aarch64_neon_fcvtms;
10875 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
10876 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtm");
10877 }
10878 case NEON::BI__builtin_neon_vcvtn_s16_v:
10879 case NEON::BI__builtin_neon_vcvtn_s32_v:
10880 case NEON::BI__builtin_neon_vcvtnq_s16_v:
10881 case NEON::BI__builtin_neon_vcvtnq_s32_v:
10882 case NEON::BI__builtin_neon_vcvtn_u16_v:
10883 case NEON::BI__builtin_neon_vcvtn_u32_v:
10884 case NEON::BI__builtin_neon_vcvtnq_u16_v:
10885 case NEON::BI__builtin_neon_vcvtnq_u32_v:
10886 case NEON::BI__builtin_neon_vcvtn_s64_v:
10887 case NEON::BI__builtin_neon_vcvtnq_s64_v:
10888 case NEON::BI__builtin_neon_vcvtn_u64_v:
10889 case NEON::BI__builtin_neon_vcvtnq_u64_v: {
10890 Int = usgn ? Intrinsic::aarch64_neon_fcvtnu : Intrinsic::aarch64_neon_fcvtns;
10891 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
10892 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtn");
10893 }
10894 case NEON::BI__builtin_neon_vcvtp_s16_v:
10895 case NEON::BI__builtin_neon_vcvtp_s32_v:
10896 case NEON::BI__builtin_neon_vcvtpq_s16_v:
10897 case NEON::BI__builtin_neon_vcvtpq_s32_v:
10898 case NEON::BI__builtin_neon_vcvtp_u16_v:
10899 case NEON::BI__builtin_neon_vcvtp_u32_v:
10900 case NEON::BI__builtin_neon_vcvtpq_u16_v:
10901 case NEON::BI__builtin_neon_vcvtpq_u32_v:
10902 case NEON::BI__builtin_neon_vcvtp_s64_v:
10903 case NEON::BI__builtin_neon_vcvtpq_s64_v:
10904 case NEON::BI__builtin_neon_vcvtp_u64_v:
10905 case NEON::BI__builtin_neon_vcvtpq_u64_v: {
10906 Int = usgn ? Intrinsic::aarch64_neon_fcvtpu : Intrinsic::aarch64_neon_fcvtps;
10907 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
10908 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtp");
10909 }
10910 case NEON::BI__builtin_neon_vmulx_v:
10911 case NEON::BI__builtin_neon_vmulxq_v: {
10912 Int = Intrinsic::aarch64_neon_fmulx;
10913 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmulx");
10914 }
10915 case NEON::BI__builtin_neon_vmulxh_lane_f16:
10916 case NEON::BI__builtin_neon_vmulxh_laneq_f16: {
10917 // vmulx_lane should be mapped to Neon scalar mulx after
10918 // extracting the scalar element
10919 Ops.push_back(EmitScalarExpr(E->getArg(2)));
10920 Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract");
10921 Ops.pop_back();
10922 Int = Intrinsic::aarch64_neon_fmulx;
10923 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmulx");
10924 }
10925 case NEON::BI__builtin_neon_vmul_lane_v:
10926 case NEON::BI__builtin_neon_vmul_laneq_v: {
10927 // v1f64 vmul_lane should be mapped to Neon scalar mul lane
10928 bool Quad = false;
10929 if (BuiltinID == NEON::BI__builtin_neon_vmul_laneq_v)
10930 Quad = true;
10931 Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
10932 llvm::FixedVectorType *VTy =
10933 GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, Quad));
10934 Ops[1] = Builder.CreateBitCast(Ops[1], VTy);
10935 Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract");
10936 Value *Result = Builder.CreateFMul(Ops[0], Ops[1]);
10937 return Builder.CreateBitCast(Result, Ty);
10938 }
10939 case NEON::BI__builtin_neon_vnegd_s64:
10940 return Builder.CreateNeg(EmitScalarExpr(E->getArg(0)), "vnegd");
10941 case NEON::BI__builtin_neon_vnegh_f16:
10942 return Builder.CreateFNeg(EmitScalarExpr(E->getArg(0)), "vnegh");
10943 case NEON::BI__builtin_neon_vpmaxnm_v:
10944 case NEON::BI__builtin_neon_vpmaxnmq_v: {
10945 Int = Intrinsic::aarch64_neon_fmaxnmp;
10946 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmaxnm");
10947 }
10948 case NEON::BI__builtin_neon_vpminnm_v:
10949 case NEON::BI__builtin_neon_vpminnmq_v: {
10950 Int = Intrinsic::aarch64_neon_fminnmp;
10951 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpminnm");
10952 }
10953 case NEON::BI__builtin_neon_vsqrth_f16: {
10954 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10955 Int = Builder.getIsFPConstrained()
10956 ? Intrinsic::experimental_constrained_sqrt
10957 : Intrinsic::sqrt;
10958 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vsqrt");
10959 }
10960 case NEON::BI__builtin_neon_vsqrt_v:
10961 case NEON::BI__builtin_neon_vsqrtq_v: {
10962 Int = Builder.getIsFPConstrained()
10963 ? Intrinsic::experimental_constrained_sqrt
10964 : Intrinsic::sqrt;
10965 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
10966 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqrt");
10967 }
10968 case NEON::BI__builtin_neon_vrbit_v:
10969 case NEON::BI__builtin_neon_vrbitq_v: {
10970 Int = Intrinsic::bitreverse;
10971 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrbit");
10972 }
10973 case NEON::BI__builtin_neon_vaddv_u8:
10974 // FIXME: These are handled by the AArch64 scalar code.
10975 usgn = true;
10976 LLVM_FALLTHROUGH[[gnu::fallthrough]];
10977 case NEON::BI__builtin_neon_vaddv_s8: {
10978 Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
10979 Ty = Int32Ty;
10980 VTy = llvm::FixedVectorType::get(Int8Ty, 8);
10981 llvm::Type *Tys[2] = { Ty, VTy };
10982 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10983 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
10984 return Builder.CreateTrunc(Ops[0], Int8Ty);
10985 }
10986 case NEON::BI__builtin_neon_vaddv_u16:
10987 usgn = true;
10988 LLVM_FALLTHROUGH[[gnu::fallthrough]];
10989 case NEON::BI__builtin_neon_vaddv_s16: {
10990 Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
10991 Ty = Int32Ty;
10992 VTy = llvm::FixedVectorType::get(Int16Ty, 4);
10993 llvm::Type *Tys[2] = { Ty, VTy };
10994 Ops.push_back(EmitScalarExpr(E->getArg(0)));
10995 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
10996 return Builder.CreateTrunc(Ops[0], Int16Ty);
10997 }
10998 case NEON::BI__builtin_neon_vaddvq_u8:
10999 usgn = true;
11000 LLVM_FALLTHROUGH[[gnu::fallthrough]];
11001 case NEON::BI__builtin_neon_vaddvq_s8: {
11002 Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
11003 Ty = Int32Ty;
11004 VTy = llvm::FixedVectorType::get(Int8Ty, 16);
11005 llvm::Type *Tys[2] = { Ty, VTy };
11006 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11007 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
11008 return Builder.CreateTrunc(Ops[0], Int8Ty);
11009 }
11010 case NEON::BI__builtin_neon_vaddvq_u16:
11011 usgn = true;
11012 LLVM_FALLTHROUGH[[gnu::fallthrough]];
11013 case NEON::BI__builtin_neon_vaddvq_s16: {
11014 Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
11015 Ty = Int32Ty;
11016 VTy = llvm::FixedVectorType::get(Int16Ty, 8);
11017 llvm::Type *Tys[2] = { Ty, VTy };
11018 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11019 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
11020 return Builder.CreateTrunc(Ops[0], Int16Ty);
11021 }
11022 case NEON::BI__builtin_neon_vmaxv_u8: {
11023 Int = Intrinsic::aarch64_neon_umaxv;
11024 Ty = Int32Ty;
11025 VTy = llvm::FixedVectorType::get(Int8Ty, 8);
11026 llvm::Type *Tys[2] = { Ty, VTy };
11027 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11028 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11029 return Builder.CreateTrunc(Ops[0], Int8Ty);
11030 }
11031 case NEON::BI__builtin_neon_vmaxv_u16: {
11032 Int = Intrinsic::aarch64_neon_umaxv;
11033 Ty = Int32Ty;
11034 VTy = llvm::FixedVectorType::get(Int16Ty, 4);
11035 llvm::Type *Tys[2] = { Ty, VTy };
11036 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11037 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11038 return Builder.CreateTrunc(Ops[0], Int16Ty);
11039 }
11040 case NEON::BI__builtin_neon_vmaxvq_u8: {
11041 Int = Intrinsic::aarch64_neon_umaxv;
11042 Ty = Int32Ty;
11043 VTy = llvm::FixedVectorType::get(Int8Ty, 16);
11044 llvm::Type *Tys[2] = { Ty, VTy };
11045 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11046 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11047 return Builder.CreateTrunc(Ops[0], Int8Ty);
11048 }
11049 case NEON::BI__builtin_neon_vmaxvq_u16: {
11050 Int = Intrinsic::aarch64_neon_umaxv;
11051 Ty = Int32Ty;
11052 VTy = llvm::FixedVectorType::get(Int16Ty, 8);
11053 llvm::Type *Tys[2] = { Ty, VTy };
11054 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11055 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11056 return Builder.CreateTrunc(Ops[0], Int16Ty);
11057 }
11058 case NEON::BI__builtin_neon_vmaxv_s8: {
11059 Int = Intrinsic::aarch64_neon_smaxv;
11060 Ty = Int32Ty;
11061 VTy = llvm::FixedVectorType::get(Int8Ty, 8);
11062 llvm::Type *Tys[2] = { Ty, VTy };
11063 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11064 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11065 return Builder.CreateTrunc(Ops[0], Int8Ty);
11066 }
11067 case NEON::BI__builtin_neon_vmaxv_s16: {
11068 Int = Intrinsic::aarch64_neon_smaxv;
11069 Ty = Int32Ty;
11070 VTy = llvm::FixedVectorType::get(Int16Ty, 4);
11071 llvm::Type *Tys[2] = { Ty, VTy };
11072 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11073 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11074 return Builder.CreateTrunc(Ops[0], Int16Ty);
11075 }
11076 case NEON::BI__builtin_neon_vmaxvq_s8: {
11077 Int = Intrinsic::aarch64_neon_smaxv;
11078 Ty = Int32Ty;
11079 VTy = llvm::FixedVectorType::get(Int8Ty, 16);
11080 llvm::Type *Tys[2] = { Ty, VTy };
11081 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11082 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11083 return Builder.CreateTrunc(Ops[0], Int8Ty);
11084 }
11085 case NEON::BI__builtin_neon_vmaxvq_s16: {
11086 Int = Intrinsic::aarch64_neon_smaxv;
11087 Ty = Int32Ty;
11088 VTy = llvm::FixedVectorType::get(Int16Ty, 8);
11089 llvm::Type *Tys[2] = { Ty, VTy };
11090 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11091 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11092 return Builder.CreateTrunc(Ops[0], Int16Ty);
11093 }
11094 case NEON::BI__builtin_neon_vmaxv_f16: {
11095 Int = Intrinsic::aarch64_neon_fmaxv;
11096 Ty = HalfTy;
11097 VTy = llvm::FixedVectorType::get(HalfTy, 4);
11098 llvm::Type *Tys[2] = { Ty, VTy };
11099 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11100 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11101 return Builder.CreateTrunc(Ops[0], HalfTy);
11102 }
11103 case NEON::BI__builtin_neon_vmaxvq_f16: {
11104 Int = Intrinsic::aarch64_neon_fmaxv;
11105 Ty = HalfTy;
11106 VTy = llvm::FixedVectorType::get(HalfTy, 8);
11107 llvm::Type *Tys[2] = { Ty, VTy };
11108 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11109 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
11110 return Builder.CreateTrunc(Ops[0], HalfTy);
11111 }
11112 case NEON::BI__builtin_neon_vminv_u8: {
11113 Int = Intrinsic::aarch64_neon_uminv;
11114 Ty = Int32Ty;
11115 VTy = llvm::FixedVectorType::get(Int8Ty, 8);
11116 llvm::Type *Tys[2] = { Ty, VTy };
11117 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11118 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11119 return Builder.CreateTrunc(Ops[0], Int8Ty);
11120 }
11121 case NEON::BI__builtin_neon_vminv_u16: {
11122 Int = Intrinsic::aarch64_neon_uminv;
11123 Ty = Int32Ty;
11124 VTy = llvm::FixedVectorType::get(Int16Ty, 4);
11125 llvm::Type *Tys[2] = { Ty, VTy };
11126 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11127 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11128 return Builder.CreateTrunc(Ops[0], Int16Ty);
11129 }
11130 case NEON::BI__builtin_neon_vminvq_u8: {
11131 Int = Intrinsic::aarch64_neon_uminv;
11132 Ty = Int32Ty;
11133 VTy = llvm::FixedVectorType::get(Int8Ty, 16);
11134 llvm::Type *Tys[2] = { Ty, VTy };
11135 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11136 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11137 return Builder.CreateTrunc(Ops[0], Int8Ty);
11138 }
11139 case NEON::BI__builtin_neon_vminvq_u16: {
11140 Int = Intrinsic::aarch64_neon_uminv;
11141 Ty = Int32Ty;
11142 VTy = llvm::FixedVectorType::get(Int16Ty, 8);
11143 llvm::Type *Tys[2] = { Ty, VTy };
11144 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11145 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11146 return Builder.CreateTrunc(Ops[0], Int16Ty);
11147 }
11148 case NEON::BI__builtin_neon_vminv_s8: {
11149 Int = Intrinsic::aarch64_neon_sminv;
11150 Ty = Int32Ty;
11151 VTy = llvm::FixedVectorType::get(Int8Ty, 8);
11152 llvm::Type *Tys[2] = { Ty, VTy };
11153 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11154 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11155 return Builder.CreateTrunc(Ops[0], Int8Ty);
11156 }
11157 case NEON::BI__builtin_neon_vminv_s16: {
11158 Int = Intrinsic::aarch64_neon_sminv;
11159 Ty = Int32Ty;
11160 VTy = llvm::FixedVectorType::get(Int16Ty, 4);
11161 llvm::Type *Tys[2] = { Ty, VTy };
11162 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11163 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11164 return Builder.CreateTrunc(Ops[0], Int16Ty);
11165 }
11166 case NEON::BI__builtin_neon_vminvq_s8: {
11167 Int = Intrinsic::aarch64_neon_sminv;
11168 Ty = Int32Ty;
11169 VTy = llvm::FixedVectorType::get(Int8Ty, 16);
11170 llvm::Type *Tys[2] = { Ty, VTy };
11171 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11172 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11173 return Builder.CreateTrunc(Ops[0], Int8Ty);
11174 }
11175 case NEON::BI__builtin_neon_vminvq_s16: {
11176 Int = Intrinsic::aarch64_neon_sminv;
11177 Ty = Int32Ty;
11178 VTy = llvm::FixedVectorType::get(Int16Ty, 8);
11179 llvm::Type *Tys[2] = { Ty, VTy };
11180 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11181 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11182 return Builder.CreateTrunc(Ops[0], Int16Ty);
11183 }
11184 case NEON::BI__builtin_neon_vminv_f16: {
11185 Int = Intrinsic::aarch64_neon_fminv;
11186 Ty = HalfTy;
11187 VTy = llvm::FixedVectorType::get(HalfTy, 4);
11188 llvm::Type *Tys[2] = { Ty, VTy };
11189 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11190 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11191 return Builder.CreateTrunc(Ops[0], HalfTy);
11192 }
11193 case NEON::BI__builtin_neon_vminvq_f16: {
11194 Int = Intrinsic::aarch64_neon_fminv;
11195 Ty = HalfTy;
11196 VTy = llvm::FixedVectorType::get(HalfTy, 8);
11197 llvm::Type *Tys[2] = { Ty, VTy };
11198 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11199 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
11200 return Builder.CreateTrunc(Ops[0], HalfTy);
11201 }
11202 case NEON::BI__builtin_neon_vmaxnmv_f16: {
11203 Int = Intrinsic::aarch64_neon_fmaxnmv;
11204 Ty = HalfTy;
11205 VTy = llvm::FixedVectorType::get(HalfTy, 4);
11206 llvm::Type *Tys[2] = { Ty, VTy };
11207 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11208 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv");
11209 return Builder.CreateTrunc(Ops[0], HalfTy);
11210 }
11211 case NEON::BI__builtin_neon_vmaxnmvq_f16: {
11212 Int = Intrinsic::aarch64_neon_fmaxnmv;
11213 Ty = HalfTy;
11214 VTy = llvm::FixedVectorType::get(HalfTy, 8);
11215 llvm::Type *Tys[2] = { Ty, VTy };
11216 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11217 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv");
11218 return Builder.CreateTrunc(Ops[0], HalfTy);
11219 }
11220 case NEON::BI__builtin_neon_vminnmv_f16: {
11221 Int = Intrinsic::aarch64_neon_fminnmv;
11222 Ty = HalfTy;
11223 VTy = llvm::FixedVectorType::get(HalfTy, 4);
11224 llvm::Type *Tys[2] = { Ty, VTy };
11225 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11226 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv");
11227 return Builder.CreateTrunc(Ops[0], HalfTy);
11228 }
11229 case NEON::BI__builtin_neon_vminnmvq_f16: {
11230 Int = Intrinsic::aarch64_neon_fminnmv;
11231 Ty = HalfTy;
11232 VTy = llvm::FixedVectorType::get(HalfTy, 8);
11233 llvm::Type *Tys[2] = { Ty, VTy };
11234 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11235 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv");
11236 return Builder.CreateTrunc(Ops[0], HalfTy);
11237 }
11238 case NEON::BI__builtin_neon_vmul_n_f64: {
11239 Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
11240 Value *RHS = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), DoubleTy);
11241 return Builder.CreateFMul(Ops[0], RHS);
11242 }
11243 case NEON::BI__builtin_neon_vaddlv_u8: {
11244 Int = Intrinsic::aarch64_neon_uaddlv;
11245 Ty = Int32Ty;
11246 VTy = llvm::FixedVectorType::get(Int8Ty, 8);
11247 llvm::Type *Tys[2] = { Ty, VTy };
11248 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11249 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
11250 return Builder.CreateTrunc(Ops[0], Int16Ty);
11251 }
11252 case NEON::BI__builtin_neon_vaddlv_u16: {
11253 Int = Intrinsic::aarch64_neon_uaddlv;
11254 Ty = Int32Ty;
11255 VTy = llvm::FixedVectorType::get(Int16Ty, 4);
11256 llvm::Type *Tys[2] = { Ty, VTy };
11257 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11258 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
11259 }
11260 case NEON::BI__builtin_neon_vaddlvq_u8: {
11261 Int = Intrinsic::aarch64_neon_uaddlv;
11262 Ty = Int32Ty;
11263 VTy = llvm::FixedVectorType::get(Int8Ty, 16);
11264 llvm::Type *Tys[2] = { Ty, VTy };
11265 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11266 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
11267 return Builder.CreateTrunc(Ops[0], Int16Ty);
11268 }
11269 case NEON::BI__builtin_neon_vaddlvq_u16: {
11270 Int = Intrinsic::aarch64_neon_uaddlv;
11271 Ty = Int32Ty;
11272 VTy = llvm::FixedVectorType::get(Int16Ty, 8);
11273 llvm::Type *Tys[2] = { Ty, VTy };
11274 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11275 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
11276 }
11277 case NEON::BI__builtin_neon_vaddlv_s8: {
11278 Int = Intrinsic::aarch64_neon_saddlv;
11279 Ty = Int32Ty;
11280 VTy = llvm::FixedVectorType::get(Int8Ty, 8);
11281 llvm::Type *Tys[2] = { Ty, VTy };
11282 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11283 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
11284 return Builder.CreateTrunc(Ops[0], Int16Ty);
11285 }
11286 case NEON::BI__builtin_neon_vaddlv_s16: {
11287 Int = Intrinsic::aarch64_neon_saddlv;
11288 Ty = Int32Ty;
11289 VTy = llvm::FixedVectorType::get(Int16Ty, 4);
11290 llvm::Type *Tys[2] = { Ty, VTy };
11291 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11292 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
11293 }
11294 case NEON::BI__builtin_neon_vaddlvq_s8: {
11295 Int = Intrinsic::aarch64_neon_saddlv;
11296 Ty = Int32Ty;
11297 VTy = llvm::FixedVectorType::get(Int8Ty, 16);
11298 llvm::Type *Tys[2] = { Ty, VTy };
11299 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11300 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
11301 return Builder.CreateTrunc(Ops[0], Int16Ty);
11302 }
11303 case NEON::BI__builtin_neon_vaddlvq_s16: {
11304 Int = Intrinsic::aarch64_neon_saddlv;
11305 Ty = Int32Ty;
11306 VTy = llvm::FixedVectorType::get(Int16Ty, 8);
11307 llvm::Type *Tys[2] = { Ty, VTy };
11308 Ops.push_back(EmitScalarExpr(E->getArg(0)));
11309 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
11310 }
11311 case NEON::BI__builtin_neon_vsri_n_v:
11312 case NEON::BI__builtin_neon_vsriq_n_v: {
11313 Int = Intrinsic::aarch64_neon_vsri;
11314 llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty);
11315 return EmitNeonCall(Intrin, Ops, "vsri_n");
11316 }
11317 case NEON::BI__builtin_neon_vsli_n_v:
11318 case NEON::BI__builtin_neon_vsliq_n_v: {
11319 Int = Intrinsic::aarch64_neon_vsli;
11320 llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty);
11321 return EmitNeonCall(Intrin, Ops, "vsli_n");
11322 }
11323 case NEON::BI__builtin_neon_vsra_n_v:
11324 case NEON::BI__builtin_neon_vsraq_n_v:
11325 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
11326 Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n");
11327 return Builder.CreateAdd(Ops[0], Ops[1]);
11328 case NEON::BI__builtin_neon_vrsra_n_v:
11329 case NEON::BI__builtin_neon_vrsraq_n_v: {
11330 Int = usgn ? Intrinsic::aarch64_neon_urshl : Intrinsic::aarch64_neon_srshl;
11331 SmallVector<llvm::Value*,2> TmpOps;
11332 TmpOps.push_back(Ops[1]);
11333 TmpOps.push_back(Ops[2]);
11334 Function* F = CGM.getIntrinsic(Int, Ty);
11335 llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vrshr_n", 1, true);
11336 Ops[0] = Builder.CreateBitCast(Ops[0], VTy);
11337 return Builder.CreateAdd(Ops[0], tmp);
11338 }
11339 case NEON::BI__builtin_neon_vld1_v:
11340 case NEON::BI__builtin_neon_vld1q_v: {
11341 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy));
11342 return Builder.CreateAlignedLoad(VTy, Ops[0], PtrOp0.getAlignment());
11343 }
11344 case NEON::BI__builtin_neon_vst1_v:
11345 case NEON::BI__builtin_neon_vst1q_v:
11346 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy));
11347 Ops[1] = Builder.CreateBitCast(Ops[1], VTy);
11348 return Builder.CreateAlignedStore(Ops[1], Ops[0], PtrOp0.getAlignment());
11349 case NEON::BI__builtin_neon_vld1_lane_v:
11350 case NEON::BI__builtin_neon_vld1q_lane_v: {
11351 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
11352 Ty = llvm::PointerType::getUnqual(VTy->getElementType());
11353 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
11354 Ops[0] = Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0],
11355 PtrOp0.getAlignment());
11356 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vld1_lane");
11357 }
11358 case NEON::BI__builtin_neon_vld1_dup_v:
11359 case NEON::BI__builtin_neon_vld1q_dup_v: {
11360 Value *V = UndefValue::get(Ty);
11361 Ty = llvm::PointerType::getUnqual(VTy->getElementType());
11362 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
11363 Ops[0] = Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0],
11364 PtrOp0.getAlignment());
11365 llvm::Constant *CI = ConstantInt::get(Int32Ty, 0);
11366 Ops[0] = Builder.CreateInsertElement(V, Ops[0], CI);
11367 return EmitNeonSplat(Ops[0], CI);
11368 }
11369 case NEON::BI__builtin_neon_vst1_lane_v:
11370 case NEON::BI__builtin_neon_vst1q_lane_v:
11371 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
11372 Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]);
11373 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
11374 return Builder.CreateAlignedStore(Ops[1], Builder.CreateBitCast(Ops[0], Ty),
11375 PtrOp0.getAlignment());
11376 case NEON::BI__builtin_neon_vld2_v:
11377 case NEON::BI__builtin_neon_vld2q_v: {
11378 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
11379 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
11380 llvm::Type *Tys[2] = { VTy, PTy };
11381 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2, Tys);
11382 Ops[1] = Builder.CreateCall(F, Ops[1], "vld2");
11383 Ops[0] = Builder.CreateBitCast(Ops[0],
11384 llvm::PointerType::getUnqual(Ops[1]->getType()));
11385 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11386 }
11387 case NEON::BI__builtin_neon_vld3_v:
11388 case NEON::BI__builtin_neon_vld3q_v: {
11389 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
11390 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
11391 llvm::Type *Tys[2] = { VTy, PTy };
11392 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3, Tys);
11393 Ops[1] = Builder.CreateCall(F, Ops[1], "vld3");
11394 Ops[0] = Builder.CreateBitCast(Ops[0],
11395 llvm::PointerType::getUnqual(Ops[1]->getType()));
11396 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11397 }
11398 case NEON::BI__builtin_neon_vld4_v:
11399 case NEON::BI__builtin_neon_vld4q_v: {
11400 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
11401 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
11402 llvm::Type *Tys[2] = { VTy, PTy };
11403 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4, Tys);
11404 Ops[1] = Builder.CreateCall(F, Ops[1], "vld4");
11405 Ops[0] = Builder.CreateBitCast(Ops[0],
11406 llvm::PointerType::getUnqual(Ops[1]->getType()));
11407 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11408 }
11409 case NEON::BI__builtin_neon_vld2_dup_v:
11410 case NEON::BI__builtin_neon_vld2q_dup_v: {
11411 llvm::Type *PTy =
11412 llvm::PointerType::getUnqual(VTy->getElementType());
11413 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
11414 llvm::Type *Tys[2] = { VTy, PTy };
11415 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2r, Tys);
11416 Ops[1] = Builder.CreateCall(F, Ops[1], "vld2");
11417 Ops[0] = Builder.CreateBitCast(Ops[0],
11418 llvm::PointerType::getUnqual(Ops[1]->getType()));
11419 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11420 }
11421 case NEON::BI__builtin_neon_vld3_dup_v:
11422 case NEON::BI__builtin_neon_vld3q_dup_v: {
11423 llvm::Type *PTy =
11424 llvm::PointerType::getUnqual(VTy->getElementType());
11425 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
11426 llvm::Type *Tys[2] = { VTy, PTy };
11427 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3r, Tys);
11428 Ops[1] = Builder.CreateCall(F, Ops[1], "vld3");
11429 Ops[0] = Builder.CreateBitCast(Ops[0],
11430 llvm::PointerType::getUnqual(Ops[1]->getType()));
11431 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11432 }
11433 case NEON::BI__builtin_neon_vld4_dup_v:
11434 case NEON::BI__builtin_neon_vld4q_dup_v: {
11435 llvm::Type *PTy =
11436 llvm::PointerType::getUnqual(VTy->getElementType());
11437 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
11438 llvm::Type *Tys[2] = { VTy, PTy };
11439 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4r, Tys);
11440 Ops[1] = Builder.CreateCall(F, Ops[1], "vld4");
11441 Ops[0] = Builder.CreateBitCast(Ops[0],
11442 llvm::PointerType::getUnqual(Ops[1]->getType()));
11443 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11444 }
11445 case NEON::BI__builtin_neon_vld2_lane_v:
11446 case NEON::BI__builtin_neon_vld2q_lane_v: {
11447 llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
11448 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2lane, Tys);
11449 std::rotate(Ops.begin() + 1, Ops.begin() + 2, Ops.end());
11450 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
11451 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
11452 Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty);
11453 Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld2_lane");
11454 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
11455 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
11456 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11457 }
11458 case NEON::BI__builtin_neon_vld3_lane_v:
11459 case NEON::BI__builtin_neon_vld3q_lane_v: {
11460 llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
11461 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3lane, Tys);
11462 std::rotate(Ops.begin() + 1, Ops.begin() + 2, Ops.end());
11463 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
11464 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
11465 Ops[3] = Builder.CreateBitCast(Ops[3], Ty);
11466 Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty);
11467 Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld3_lane");
11468 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
11469 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
11470 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11471 }
11472 case NEON::BI__builtin_neon_vld4_lane_v:
11473 case NEON::BI__builtin_neon_vld4q_lane_v: {
11474 llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
11475 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4lane, Tys);
11476 std::rotate(Ops.begin() + 1, Ops.begin() + 2, Ops.end());
11477 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
11478 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
11479 Ops[3] = Builder.CreateBitCast(Ops[3], Ty);
11480 Ops[4] = Builder.CreateBitCast(Ops[4], Ty);
11481 Ops[5] = Builder.CreateZExt(Ops[5], Int64Ty);
11482 Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld4_lane");
11483 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
11484 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
11485 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
11486 }
11487 case NEON::BI__builtin_neon_vst2_v:
11488 case NEON::BI__builtin_neon_vst2q_v: {
11489 std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
11490 llvm::Type *Tys[2] = { VTy, Ops[2]->getType() };
11491 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2, Tys),
11492 Ops, "");
11493 }
11494 case NEON::BI__builtin_neon_vst2_lane_v:
11495 case NEON::BI__builtin_neon_vst2q_lane_v: {
11496 std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
11497 Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty);
11498 llvm::Type *Tys[2] = { VTy, Ops[3]->getType() };
11499 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2lane, Tys),
11500 Ops, "");
11501 }
11502 case NEON::BI__builtin_neon_vst3_v:
11503 case NEON::BI__builtin_neon_vst3q_v: {
11504 std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
11505 llvm::Type *Tys[2] = { VTy, Ops[3]->getType() };
11506 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3, Tys),
11507 Ops, "");
11508 }
11509 case NEON::BI__builtin_neon_vst3_lane_v:
11510 case NEON::BI__builtin_neon_vst3q_lane_v: {
11511 std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
11512 Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty);
11513 llvm::Type *Tys[2] = { VTy, Ops[4]->getType() };
11514 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3lane, Tys),
11515 Ops, "");
11516 }
11517 case NEON::BI__builtin_neon_vst4_v:
11518 case NEON::BI__builtin_neon_vst4q_v: {
11519 std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
11520 llvm::Type *Tys[2] = { VTy, Ops[4]->getType() };
11521 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4, Tys),
11522 Ops, "");
11523 }
11524 case NEON::BI__builtin_neon_vst4_lane_v:
11525 case NEON::BI__builtin_neon_vst4q_lane_v: {
11526 std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
11527 Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty);
11528 llvm::Type *Tys[2] = { VTy, Ops[5]->getType() };
11529 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4lane, Tys),
11530 Ops, "");
11531 }
11532 case NEON::BI__builtin_neon_vtrn_v:
11533 case NEON::BI__builtin_neon_vtrnq_v: {
11534 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
11535 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
11536 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
11537 Value *SV = nullptr;
11538
11539 for (unsigned vi = 0; vi != 2; ++vi) {
11540 SmallVector<int, 16> Indices;
11541 for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
11542 Indices.push_back(i+vi);
11543 Indices.push_back(i+e+vi);
11544 }
11545 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
11546 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn");
11547 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
11548 }
11549 return SV;
11550 }
11551 case NEON::BI__builtin_neon_vuzp_v:
11552 case NEON::BI__builtin_neon_vuzpq_v: {
11553 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
11554 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
11555 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
11556 Value *SV = nullptr;
11557
11558 for (unsigned vi = 0; vi != 2; ++vi) {
11559 SmallVector<int, 16> Indices;
11560 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
11561 Indices.push_back(2*i+vi);
11562
11563 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
11564 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp");
11565 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
11566 }
11567 return SV;
11568 }
11569 case NEON::BI__builtin_neon_vzip_v:
11570 case NEON::BI__builtin_neon_vzipq_v: {
11571 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
11572 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
11573 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
11574 Value *SV = nullptr;
11575
11576 for (unsigned vi = 0; vi != 2; ++vi) {
11577 SmallVector<int, 16> Indices;
11578 for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
11579 Indices.push_back((i + vi*e) >> 1);
11580 Indices.push_back(((i + vi*e) >> 1)+e);
11581 }
11582 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
11583 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip");
11584 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
11585 }
11586 return SV;
11587 }
11588 case NEON::BI__builtin_neon_vqtbl1q_v: {
11589 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl1, Ty),
11590 Ops, "vtbl1");
11591 }
11592 case NEON::BI__builtin_neon_vqtbl2q_v: {
11593 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl2, Ty),
11594 Ops, "vtbl2");
11595 }
11596 case NEON::BI__builtin_neon_vqtbl3q_v: {
11597 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl3, Ty),
11598 Ops, "vtbl3");
11599 }
11600 case NEON::BI__builtin_neon_vqtbl4q_v: {
11601 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl4, Ty),
11602 Ops, "vtbl4");
11603 }
11604 case NEON::BI__builtin_neon_vqtbx1q_v: {
11605 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx1, Ty),
11606 Ops, "vtbx1");
11607 }
11608 case NEON::BI__builtin_neon_vqtbx2q_v: {
11609 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx2, Ty),
11610 Ops, "vtbx2");
11611 }
11612 case NEON::BI__builtin_neon_vqtbx3q_v: {
11613 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx3, Ty),
11614 Ops, "vtbx3");
11615 }
11616 case NEON::BI__builtin_neon_vqtbx4q_v: {
11617 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx4, Ty),
11618 Ops, "vtbx4");
11619 }
11620 case NEON::BI__builtin_neon_vsqadd_v:
11621 case NEON::BI__builtin_neon_vsqaddq_v: {
11622 Int = Intrinsic::aarch64_neon_usqadd;
11623 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqadd");
11624 }
11625 case NEON::BI__builtin_neon_vuqadd_v:
11626 case NEON::BI__builtin_neon_vuqaddq_v: {
11627 Int = Intrinsic::aarch64_neon_suqadd;
11628 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vuqadd");
11629 }
11630 }
11631}
11632
11633Value *CodeGenFunction::EmitBPFBuiltinExpr(unsigned BuiltinID,
11634 const CallExpr *E) {
11635 assert((BuiltinID == BPF::BI__builtin_preserve_field_info ||(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info
|| BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID ==
BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value
) && "unexpected BPF builtin") ? void (0) : __assert_fail
("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11639, __extension__ __PRETTY_FUNCTION__))
11636 BuiltinID == BPF::BI__builtin_btf_type_id ||(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info
|| BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID ==
BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value
) && "unexpected BPF builtin") ? void (0) : __assert_fail
("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11639, __extension__ __PRETTY_FUNCTION__))
11637 BuiltinID == BPF::BI__builtin_preserve_type_info ||(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info
|| BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID ==
BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value
) && "unexpected BPF builtin") ? void (0) : __assert_fail
("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11639, __extension__ __PRETTY_FUNCTION__))
11638 BuiltinID == BPF::BI__builtin_preserve_enum_value) &&(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info
|| BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID ==
BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value
) && "unexpected BPF builtin") ? void (0) : __assert_fail
("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11639, __extension__ __PRETTY_FUNCTION__))
11639 "unexpected BPF builtin")(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info
|| BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID ==
BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value
) && "unexpected BPF builtin") ? void (0) : __assert_fail
("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11639, __extension__ __PRETTY_FUNCTION__))
;
11640
11641 // A sequence number, injected into IR builtin functions, to
11642 // prevent CSE given the only difference of the funciton
11643 // may just be the debuginfo metadata.
11644 static uint32_t BuiltinSeqNum;
11645
11646 switch (BuiltinID) {
11647 default:
11648 llvm_unreachable("Unexpected BPF builtin")::llvm::llvm_unreachable_internal("Unexpected BPF builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11648)
;
11649 case BPF::BI__builtin_preserve_field_info: {
11650 const Expr *Arg = E->getArg(0);
11651 bool IsBitField = Arg->IgnoreParens()->getObjectKind() == OK_BitField;
11652
11653 if (!getDebugInfo()) {
11654 CGM.Error(E->getExprLoc(),
11655 "using __builtin_preserve_field_info() without -g");
11656 return IsBitField ? EmitLValue(Arg).getBitFieldPointer()
11657 : EmitLValue(Arg).getPointer(*this);
11658 }
11659
11660 // Enable underlying preserve_*_access_index() generation.
11661 bool OldIsInPreservedAIRegion = IsInPreservedAIRegion;
11662 IsInPreservedAIRegion = true;
11663 Value *FieldAddr = IsBitField ? EmitLValue(Arg).getBitFieldPointer()
11664 : EmitLValue(Arg).getPointer(*this);
11665 IsInPreservedAIRegion = OldIsInPreservedAIRegion;
11666
11667 ConstantInt *C = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
11668 Value *InfoKind = ConstantInt::get(Int64Ty, C->getSExtValue());
11669
11670 // Built the IR for the preserve_field_info intrinsic.
11671 llvm::Function *FnGetFieldInfo = llvm::Intrinsic::getDeclaration(
11672 &CGM.getModule(), llvm::Intrinsic::bpf_preserve_field_info,
11673 {FieldAddr->getType()});
11674 return Builder.CreateCall(FnGetFieldInfo, {FieldAddr, InfoKind});
11675 }
11676 case BPF::BI__builtin_btf_type_id:
11677 case BPF::BI__builtin_preserve_type_info: {
11678 if (!getDebugInfo()) {
11679 CGM.Error(E->getExprLoc(), "using builtin function without -g");
11680 return nullptr;
11681 }
11682
11683 const Expr *Arg0 = E->getArg(0);
11684 llvm::DIType *DbgInfo = getDebugInfo()->getOrCreateStandaloneType(
11685 Arg0->getType(), Arg0->getExprLoc());
11686
11687 ConstantInt *Flag = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
11688 Value *FlagValue = ConstantInt::get(Int64Ty, Flag->getSExtValue());
11689 Value *SeqNumVal = ConstantInt::get(Int32Ty, BuiltinSeqNum++);
11690
11691 llvm::Function *FnDecl;
11692 if (BuiltinID == BPF::BI__builtin_btf_type_id)
11693 FnDecl = llvm::Intrinsic::getDeclaration(
11694 &CGM.getModule(), llvm::Intrinsic::bpf_btf_type_id, {});
11695 else
11696 FnDecl = llvm::Intrinsic::getDeclaration(
11697 &CGM.getModule(), llvm::Intrinsic::bpf_preserve_type_info, {});
11698 CallInst *Fn = Builder.CreateCall(FnDecl, {SeqNumVal, FlagValue});
11699 Fn->setMetadata(LLVMContext::MD_preserve_access_index, DbgInfo);
11700 return Fn;
11701 }
11702 case BPF::BI__builtin_preserve_enum_value: {
11703 if (!getDebugInfo()) {
11704 CGM.Error(E->getExprLoc(), "using builtin function without -g");
11705 return nullptr;
11706 }
11707
11708 const Expr *Arg0 = E->getArg(0);
11709 llvm::DIType *DbgInfo = getDebugInfo()->getOrCreateStandaloneType(
11710 Arg0->getType(), Arg0->getExprLoc());
11711
11712 // Find enumerator
11713 const auto *UO = cast<UnaryOperator>(Arg0->IgnoreParens());
11714 const auto *CE = cast<CStyleCastExpr>(UO->getSubExpr());
11715 const auto *DR = cast<DeclRefExpr>(CE->getSubExpr());
11716 const auto *Enumerator = cast<EnumConstantDecl>(DR->getDecl());
11717
11718 auto &InitVal = Enumerator->getInitVal();
11719 std::string InitValStr;
11720 if (InitVal.isNegative() || InitVal > uint64_t(INT64_MAX(9223372036854775807L)))
11721 InitValStr = std::to_string(InitVal.getSExtValue());
11722 else
11723 InitValStr = std::to_string(InitVal.getZExtValue());
11724 std::string EnumStr = Enumerator->getNameAsString() + ":" + InitValStr;
11725 Value *EnumStrVal = Builder.CreateGlobalStringPtr(EnumStr);
11726
11727 ConstantInt *Flag = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
11728 Value *FlagValue = ConstantInt::get(Int64Ty, Flag->getSExtValue());
11729 Value *SeqNumVal = ConstantInt::get(Int32Ty, BuiltinSeqNum++);
11730
11731 llvm::Function *IntrinsicFn = llvm::Intrinsic::getDeclaration(
11732 &CGM.getModule(), llvm::Intrinsic::bpf_preserve_enum_value, {});
11733 CallInst *Fn =
11734 Builder.CreateCall(IntrinsicFn, {SeqNumVal, EnumStrVal, FlagValue});
11735 Fn->setMetadata(LLVMContext::MD_preserve_access_index, DbgInfo);
11736 return Fn;
11737 }
11738 }
11739}
11740
11741llvm::Value *CodeGenFunction::
11742BuildVector(ArrayRef<llvm::Value*> Ops) {
11743 assert((Ops.size() & (Ops.size() - 1)) == 0 &&(static_cast <bool> ((Ops.size() & (Ops.size() - 1)
) == 0 && "Not a power-of-two sized vector!") ? void (
0) : __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11744, __extension__ __PRETTY_FUNCTION__))
11744 "Not a power-of-two sized vector!")(static_cast <bool> ((Ops.size() & (Ops.size() - 1)
) == 0 && "Not a power-of-two sized vector!") ? void (
0) : __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11744, __extension__ __PRETTY_FUNCTION__))
;
11745 bool AllConstants = true;
11746 for (unsigned i = 0, e = Ops.size(); i != e && AllConstants; ++i)
11747 AllConstants &= isa<Constant>(Ops[i]);
11748
11749 // If this is a constant vector, create a ConstantVector.
11750 if (AllConstants) {
11751 SmallVector<llvm::Constant*, 16> CstOps;
11752 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
11753 CstOps.push_back(cast<Constant>(Ops[i]));
11754 return llvm::ConstantVector::get(CstOps);
11755 }
11756
11757 // Otherwise, insertelement the values to build the vector.
11758 Value *Result = llvm::UndefValue::get(
11759 llvm::FixedVectorType::get(Ops[0]->getType(), Ops.size()));
11760
11761 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
11762 Result = Builder.CreateInsertElement(Result, Ops[i], Builder.getInt32(i));
11763
11764 return Result;
11765}
11766
11767// Convert the mask from an integer type to a vector of i1.
11768static Value *getMaskVecValue(CodeGenFunction &CGF, Value *Mask,
11769 unsigned NumElts) {
11770
11771 auto *MaskTy = llvm::FixedVectorType::get(
11772 CGF.Builder.getInt1Ty(),
11773 cast<IntegerType>(Mask->getType())->getBitWidth());
11774 Value *MaskVec = CGF.Builder.CreateBitCast(Mask, MaskTy);
11775
11776 // If we have less than 8 elements, then the starting mask was an i8 and
11777 // we need to extract down to the right number of elements.
11778 if (NumElts < 8) {
11779 int Indices[4];
11780 for (unsigned i = 0; i != NumElts; ++i)
11781 Indices[i] = i;
11782 MaskVec = CGF.Builder.CreateShuffleVector(MaskVec, MaskVec,
11783 makeArrayRef(Indices, NumElts),
11784 "extract");
11785 }
11786 return MaskVec;
11787}
11788
11789static Value *EmitX86MaskedStore(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
11790 Align Alignment) {
11791 // Cast the pointer to right type.
11792 Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
11793 llvm::PointerType::getUnqual(Ops[1]->getType()));
11794
11795 Value *MaskVec = getMaskVecValue(
11796 CGF, Ops[2],
11797 cast<llvm::FixedVectorType>(Ops[1]->getType())->getNumElements());
11798
11799 return CGF.Builder.CreateMaskedStore(Ops[1], Ptr, Alignment, MaskVec);
11800}
11801
11802static Value *EmitX86MaskedLoad(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
11803 Align Alignment) {
11804 // Cast the pointer to right type.
11805 llvm::Type *Ty = Ops[1]->getType();
11806 Value *Ptr =
11807 CGF.Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
11808
11809 Value *MaskVec = getMaskVecValue(
11810 CGF, Ops[2], cast<llvm::FixedVectorType>(Ty)->getNumElements());
11811
11812 return CGF.Builder.CreateMaskedLoad(Ty, Ptr, Alignment, MaskVec, Ops[1]);
11813}
11814
11815static Value *EmitX86ExpandLoad(CodeGenFunction &CGF,
11816 ArrayRef<Value *> Ops) {
11817 auto *ResultTy = cast<llvm::VectorType>(Ops[1]->getType());
11818 llvm::Type *PtrTy = ResultTy->getElementType();
11819
11820 // Cast the pointer to element type.
11821 Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
11822 llvm::PointerType::getUnqual(PtrTy));
11823
11824 Value *MaskVec = getMaskVecValue(
11825 CGF, Ops[2], cast<FixedVectorType>(ResultTy)->getNumElements());
11826
11827 llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_expandload,
11828 ResultTy);
11829 return CGF.Builder.CreateCall(F, { Ptr, MaskVec, Ops[1] });
11830}
11831
11832static Value *EmitX86CompressExpand(CodeGenFunction &CGF,
11833 ArrayRef<Value *> Ops,
11834 bool IsCompress) {
11835 auto *ResultTy = cast<llvm::FixedVectorType>(Ops[1]->getType());
11836
11837 Value *MaskVec = getMaskVecValue(CGF, Ops[2], ResultTy->getNumElements());
11838
11839 Intrinsic::ID IID = IsCompress ? Intrinsic::x86_avx512_mask_compress
11840 : Intrinsic::x86_avx512_mask_expand;
11841 llvm::Function *F = CGF.CGM.getIntrinsic(IID, ResultTy);
11842 return CGF.Builder.CreateCall(F, { Ops[0], Ops[1], MaskVec });
11843}
11844
11845static Value *EmitX86CompressStore(CodeGenFunction &CGF,
11846 ArrayRef<Value *> Ops) {
11847 auto *ResultTy = cast<llvm::FixedVectorType>(Ops[1]->getType());
11848 llvm::Type *PtrTy = ResultTy->getElementType();
11849
11850 // Cast the pointer to element type.
11851 Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
11852 llvm::PointerType::getUnqual(PtrTy));
11853
11854 Value *MaskVec = getMaskVecValue(CGF, Ops[2], ResultTy->getNumElements());
11855
11856 llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_compressstore,
11857 ResultTy);
11858 return CGF.Builder.CreateCall(F, { Ops[1], Ptr, MaskVec });
11859}
11860
11861static Value *EmitX86MaskLogic(CodeGenFunction &CGF, Instruction::BinaryOps Opc,
11862 ArrayRef<Value *> Ops,
11863 bool InvertLHS = false) {
11864 unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
11865 Value *LHS = getMaskVecValue(CGF, Ops[0], NumElts);
11866 Value *RHS = getMaskVecValue(CGF, Ops[1], NumElts);
11867
11868 if (InvertLHS)
11869 LHS = CGF.Builder.CreateNot(LHS);
11870
11871 return CGF.Builder.CreateBitCast(CGF.Builder.CreateBinOp(Opc, LHS, RHS),
11872 Ops[0]->getType());
11873}
11874
11875static Value *EmitX86FunnelShift(CodeGenFunction &CGF, Value *Op0, Value *Op1,
11876 Value *Amt, bool IsRight) {
11877 llvm::Type *Ty = Op0->getType();
11878
11879 // Amount may be scalar immediate, in which case create a splat vector.
11880 // Funnel shifts amounts are treated as modulo and types are all power-of-2 so
11881 // we only care about the lowest log2 bits anyway.
11882 if (Amt->getType() != Ty) {
11883 unsigned NumElts = cast<llvm::FixedVectorType>(Ty)->getNumElements();
11884 Amt = CGF.Builder.CreateIntCast(Amt, Ty->getScalarType(), false);
11885 Amt = CGF.Builder.CreateVectorSplat(NumElts, Amt);
11886 }
11887
11888 unsigned IID = IsRight ? Intrinsic::fshr : Intrinsic::fshl;
11889 Function *F = CGF.CGM.getIntrinsic(IID, Ty);
11890 return CGF.Builder.CreateCall(F, {Op0, Op1, Amt});
11891}
11892
11893static Value *EmitX86vpcom(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
11894 bool IsSigned) {
11895 Value *Op0 = Ops[0];
11896 Value *Op1 = Ops[1];
11897 llvm::Type *Ty = Op0->getType();
11898 uint64_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
11899
11900 CmpInst::Predicate Pred;
11901 switch (Imm) {
11902 case 0x0:
11903 Pred = IsSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
11904 break;
11905 case 0x1:
11906 Pred = IsSigned ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE;
11907 break;
11908 case 0x2:
11909 Pred = IsSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
11910 break;
11911 case 0x3:
11912 Pred = IsSigned ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE;
11913 break;
11914 case 0x4:
11915 Pred = ICmpInst::ICMP_EQ;
11916 break;
11917 case 0x5:
11918 Pred = ICmpInst::ICMP_NE;
11919 break;
11920 case 0x6:
11921 return llvm::Constant::getNullValue(Ty); // FALSE
11922 case 0x7:
11923 return llvm::Constant::getAllOnesValue(Ty); // TRUE
11924 default:
11925 llvm_unreachable("Unexpected XOP vpcom/vpcomu predicate")::llvm::llvm_unreachable_internal("Unexpected XOP vpcom/vpcomu predicate"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11925)
;
11926 }
11927
11928 Value *Cmp = CGF.Builder.CreateICmp(Pred, Op0, Op1);
11929 Value *Res = CGF.Builder.CreateSExt(Cmp, Ty);
11930 return Res;
11931}
11932
11933static Value *EmitX86Select(CodeGenFunction &CGF,
11934 Value *Mask, Value *Op0, Value *Op1) {
11935
11936 // If the mask is all ones just return first argument.
11937 if (const auto *C = dyn_cast<Constant>(Mask))
11938 if (C->isAllOnesValue())
11939 return Op0;
11940
11941 Mask = getMaskVecValue(
11942 CGF, Mask, cast<llvm::FixedVectorType>(Op0->getType())->getNumElements());
11943
11944 return CGF.Builder.CreateSelect(Mask, Op0, Op1);
11945}
11946
11947static Value *EmitX86ScalarSelect(CodeGenFunction &CGF,
11948 Value *Mask, Value *Op0, Value *Op1) {
11949 // If the mask is all ones just return first argument.
11950 if (const auto *C = dyn_cast<Constant>(Mask))
11951 if (C->isAllOnesValue())
11952 return Op0;
11953
11954 auto *MaskTy = llvm::FixedVectorType::get(
11955 CGF.Builder.getInt1Ty(), Mask->getType()->getIntegerBitWidth());
11956 Mask = CGF.Builder.CreateBitCast(Mask, MaskTy);
11957 Mask = CGF.Builder.CreateExtractElement(Mask, (uint64_t)0);
11958 return CGF.Builder.CreateSelect(Mask, Op0, Op1);
11959}
11960
11961static Value *EmitX86MaskedCompareResult(CodeGenFunction &CGF, Value *Cmp,
11962 unsigned NumElts, Value *MaskIn) {
11963 if (MaskIn
28.1
'MaskIn' is null
) {
29
Taking false branch
11964 const auto *C = dyn_cast<Constant>(MaskIn);
11965 if (!C || !C->isAllOnesValue())
11966 Cmp = CGF.Builder.CreateAnd(Cmp, getMaskVecValue(CGF, MaskIn, NumElts));
11967 }
11968
11969 if (NumElts < 8) {
30
Assuming 'NumElts' is < 8
31
Taking true branch
11970 int Indices[8];
11971 for (unsigned i = 0; i != NumElts; ++i)
32
Assuming 'i' is equal to 'NumElts'
33
Loop condition is false. Execution continues on line 11973
11972 Indices[i] = i;
11973 for (unsigned i = NumElts; i != 8; ++i)
34
Loop condition is true. Entering loop body
11974 Indices[i] = i % NumElts + NumElts;
35
Division by zero
11975 Cmp = CGF.Builder.CreateShuffleVector(
11976 Cmp, llvm::Constant::getNullValue(Cmp->getType()), Indices);
11977 }
11978
11979 return CGF.Builder.CreateBitCast(Cmp,
11980 IntegerType::get(CGF.getLLVMContext(),
11981 std::max(NumElts, 8U)));
11982}
11983
11984static Value *EmitX86MaskedCompare(CodeGenFunction &CGF, unsigned CC,
11985 bool Signed, ArrayRef<Value *> Ops) {
11986 assert((Ops.size() == 2 || Ops.size() == 4) &&(static_cast <bool> ((Ops.size() == 2 || Ops.size() == 4
) && "Unexpected number of arguments") ? void (0) : __assert_fail
("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11987, __extension__ __PRETTY_FUNCTION__))
15
Assuming the condition is false
16
Assuming the condition is true
17
'?' condition is true
11987 "Unexpected number of arguments")(static_cast <bool> ((Ops.size() == 2 || Ops.size() == 4
) && "Unexpected number of arguments") ? void (0) : __assert_fail
("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 11987, __extension__ __PRETTY_FUNCTION__))
;
11988 unsigned NumElts =
19
'NumElts' initialized here
11989 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
18
The object is a 'FixedVectorType'
11990 Value *Cmp;
11991
11992 if (CC
19.1
'CC' is not equal to 3
== 3) {
20
Taking false branch
11993 Cmp = Constant::getNullValue(
11994 llvm::FixedVectorType::get(CGF.Builder.getInt1Ty(), NumElts));
11995 } else if (CC
20.1
'CC' is not equal to 7
== 7) {
21
Taking false branch
11996 Cmp = Constant::getAllOnesValue(
11997 llvm::FixedVectorType::get(CGF.Builder.getInt1Ty(), NumElts));
11998 } else {
11999 ICmpInst::Predicate Pred;
12000 switch (CC) {
22
Control jumps to 'case 1:' at line 12003
12001 default: llvm_unreachable("Unknown condition code")::llvm::llvm_unreachable_internal("Unknown condition code", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12001)
;
12002 case 0: Pred = ICmpInst::ICMP_EQ; break;
12003 case 1: Pred = Signed
22.1
'Signed' is true
? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; break;
23
'?' condition is true
24
Execution continues on line 12009
12004 case 2: Pred = Signed ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; break;
12005 case 4: Pred = ICmpInst::ICMP_NE; break;
12006 case 5: Pred = Signed ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; break;
12007 case 6: Pred = Signed ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; break;
12008 }
12009 Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]);
12010 }
12011
12012 Value *MaskIn = nullptr;
12013 if (Ops.size() == 4)
25
Assuming the condition is false
26
Taking false branch
12014 MaskIn = Ops[3];
12015
12016 return EmitX86MaskedCompareResult(CGF, Cmp, NumElts, MaskIn);
27
Passing value via 3rd parameter 'NumElts'
28
Calling 'EmitX86MaskedCompareResult'
12017}
12018
12019static Value *EmitX86ConvertToMask(CodeGenFunction &CGF, Value *In) {
12020 Value *Zero = Constant::getNullValue(In->getType());
12021 return EmitX86MaskedCompare(CGF, 1, true, { In, Zero });
14
Calling 'EmitX86MaskedCompare'
12022}
12023
12024static Value *EmitX86ConvertIntToFp(CodeGenFunction &CGF, const CallExpr *E,
12025 ArrayRef<Value *> Ops, bool IsSigned) {
12026 unsigned Rnd = cast<llvm::ConstantInt>(Ops[3])->getZExtValue();
12027 llvm::Type *Ty = Ops[1]->getType();
12028
12029 Value *Res;
12030 if (Rnd != 4) {
12031 Intrinsic::ID IID = IsSigned ? Intrinsic::x86_avx512_sitofp_round
12032 : Intrinsic::x86_avx512_uitofp_round;
12033 Function *F = CGF.CGM.getIntrinsic(IID, { Ty, Ops[0]->getType() });
12034 Res = CGF.Builder.CreateCall(F, { Ops[0], Ops[3] });
12035 } else {
12036 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
12037 Res = IsSigned ? CGF.Builder.CreateSIToFP(Ops[0], Ty)
12038 : CGF.Builder.CreateUIToFP(Ops[0], Ty);
12039 }
12040
12041 return EmitX86Select(CGF, Ops[2], Res, Ops[1]);
12042}
12043
12044// Lowers X86 FMA intrinsics to IR.
12045static Value *EmitX86FMAExpr(CodeGenFunction &CGF, const CallExpr *E,
12046 ArrayRef<Value *> Ops, unsigned BuiltinID,
12047 bool IsAddSub) {
12048
12049 bool Subtract = false;
12050 Intrinsic::ID IID = Intrinsic::not_intrinsic;
12051 switch (BuiltinID) {
12052 default: break;
12053 case clang::X86::BI__builtin_ia32_vfmsubph512_mask3:
12054 Subtract = true;
12055 LLVM_FALLTHROUGH[[gnu::fallthrough]];
12056 case clang::X86::BI__builtin_ia32_vfmaddph512_mask:
12057 case clang::X86::BI__builtin_ia32_vfmaddph512_maskz:
12058 case clang::X86::BI__builtin_ia32_vfmaddph512_mask3:
12059 IID = llvm::Intrinsic::x86_avx512fp16_vfmadd_ph_512;
12060 break;
12061 case clang::X86::BI__builtin_ia32_vfmsubaddph512_mask3:
12062 Subtract = true;
12063 LLVM_FALLTHROUGH[[gnu::fallthrough]];
12064 case clang::X86::BI__builtin_ia32_vfmaddsubph512_mask:
12065 case clang::X86::BI__builtin_ia32_vfmaddsubph512_maskz:
12066 case clang::X86::BI__builtin_ia32_vfmaddsubph512_mask3:
12067 IID = llvm::Intrinsic::x86_avx512fp16_vfmaddsub_ph_512;
12068 break;
12069 case clang::X86::BI__builtin_ia32_vfmsubps512_mask3:
12070 Subtract = true;
12071 LLVM_FALLTHROUGH[[gnu::fallthrough]];
12072 case clang::X86::BI__builtin_ia32_vfmaddps512_mask:
12073 case clang::X86::BI__builtin_ia32_vfmaddps512_maskz:
12074 case clang::X86::BI__builtin_ia32_vfmaddps512_mask3:
12075 IID = llvm::Intrinsic::x86_avx512_vfmadd_ps_512; break;
12076 case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3:
12077 Subtract = true;
12078 LLVM_FALLTHROUGH[[gnu::fallthrough]];
12079 case clang::X86::BI__builtin_ia32_vfmaddpd512_mask:
12080 case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz:
12081 case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3:
12082 IID = llvm::Intrinsic::x86_avx512_vfmadd_pd_512; break;
12083 case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3:
12084 Subtract = true;
12085 LLVM_FALLTHROUGH[[gnu::fallthrough]];
12086 case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask:
12087 case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz:
12088 case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3:
12089 IID = llvm::Intrinsic::x86_avx512_vfmaddsub_ps_512;
12090 break;
12091 case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
12092 Subtract = true;
12093 LLVM_FALLTHROUGH[[gnu::fallthrough]];
12094 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask:
12095 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
12096 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
12097 IID = llvm::Intrinsic::x86_avx512_vfmaddsub_pd_512;
12098 break;
12099 }
12100
12101 Value *A = Ops[0];
12102 Value *B = Ops[1];
12103 Value *C = Ops[2];
12104
12105 if (Subtract)
12106 C = CGF.Builder.CreateFNeg(C);
12107
12108 Value *Res;
12109
12110 // Only handle in case of _MM_FROUND_CUR_DIRECTION/4 (no rounding).
12111 if (IID != Intrinsic::not_intrinsic &&
12112 (cast<llvm::ConstantInt>(Ops.back())->getZExtValue() != (uint64_t)4 ||
12113 IsAddSub)) {
12114 Function *Intr = CGF.CGM.getIntrinsic(IID);
12115 Res = CGF.Builder.CreateCall(Intr, {A, B, C, Ops.back() });
12116 } else {
12117 llvm::Type *Ty = A->getType();
12118 Function *FMA;
12119 if (CGF.Builder.getIsFPConstrained()) {
12120 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
12121 FMA = CGF.CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, Ty);
12122 Res = CGF.Builder.CreateConstrainedFPCall(FMA, {A, B, C});
12123 } else {
12124 FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ty);
12125 Res = CGF.Builder.CreateCall(FMA, {A, B, C});
12126 }
12127 }
12128
12129 // Handle any required masking.
12130 Value *MaskFalseVal = nullptr;
12131 switch (BuiltinID) {
12132 case clang::X86::BI__builtin_ia32_vfmaddph512_mask:
12133 case clang::X86::BI__builtin_ia32_vfmaddps512_mask:
12134 case clang::X86::BI__builtin_ia32_vfmaddpd512_mask:
12135 case clang::X86::BI__builtin_ia32_vfmaddsubph512_mask:
12136 case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask:
12137 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask:
12138 MaskFalseVal = Ops[0];
12139 break;
12140 case clang::X86::BI__builtin_ia32_vfmaddph512_maskz:
12141 case clang::X86::BI__builtin_ia32_vfmaddps512_maskz:
12142 case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz:
12143 case clang::X86::BI__builtin_ia32_vfmaddsubph512_maskz:
12144 case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz:
12145 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
12146 MaskFalseVal = Constant::getNullValue(Ops[0]->getType());
12147 break;
12148 case clang::X86::BI__builtin_ia32_vfmsubph512_mask3:
12149 case clang::X86::BI__builtin_ia32_vfmaddph512_mask3:
12150 case clang::X86::BI__builtin_ia32_vfmsubps512_mask3:
12151 case clang::X86::BI__builtin_ia32_vfmaddps512_mask3:
12152 case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3:
12153 case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3:
12154 case clang::X86::BI__builtin_ia32_vfmsubaddph512_mask3:
12155 case clang::X86::BI__builtin_ia32_vfmaddsubph512_mask3:
12156 case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3:
12157 case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3:
12158 case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
12159 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
12160 MaskFalseVal = Ops[2];
12161 break;
12162 }
12163
12164 if (MaskFalseVal)
12165 return EmitX86Select(CGF, Ops[3], Res, MaskFalseVal);
12166
12167 return Res;
12168}
12169
12170static Value *EmitScalarFMAExpr(CodeGenFunction &CGF, const CallExpr *E,
12171 MutableArrayRef<Value *> Ops, Value *Upper,
12172 bool ZeroMask = false, unsigned PTIdx = 0,
12173 bool NegAcc = false) {
12174 unsigned Rnd = 4;
12175 if (Ops.size() > 4)
12176 Rnd = cast<llvm::ConstantInt>(Ops[4])->getZExtValue();
12177
12178 if (NegAcc)
12179 Ops[2] = CGF.Builder.CreateFNeg(Ops[2]);
12180
12181 Ops[0] = CGF.Builder.CreateExtractElement(Ops[0], (uint64_t)0);
12182 Ops[1] = CGF.Builder.CreateExtractElement(Ops[1], (uint64_t)0);
12183 Ops[2] = CGF.Builder.CreateExtractElement(Ops[2], (uint64_t)0);
12184 Value *Res;
12185 if (Rnd != 4) {
12186 Intrinsic::ID IID;
12187
12188 switch (Ops[0]->getType()->getPrimitiveSizeInBits()) {
12189 case 16:
12190 IID = Intrinsic::x86_avx512fp16_vfmadd_f16;
12191 break;
12192 case 32:
12193 IID = Intrinsic::x86_avx512_vfmadd_f32;
12194 break;
12195 case 64:
12196 IID = Intrinsic::x86_avx512_vfmadd_f64;
12197 break;
12198 default:
12199 llvm_unreachable("Unexpected size")::llvm::llvm_unreachable_internal("Unexpected size", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12199)
;
12200 }
12201 Res = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID),
12202 {Ops[0], Ops[1], Ops[2], Ops[4]});
12203 } else if (CGF.Builder.getIsFPConstrained()) {
12204 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
12205 Function *FMA = CGF.CGM.getIntrinsic(
12206 Intrinsic::experimental_constrained_fma, Ops[0]->getType());
12207 Res = CGF.Builder.CreateConstrainedFPCall(FMA, Ops.slice(0, 3));
12208 } else {
12209 Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ops[0]->getType());
12210 Res = CGF.Builder.CreateCall(FMA, Ops.slice(0, 3));
12211 }
12212 // If we have more than 3 arguments, we need to do masking.
12213 if (Ops.size() > 3) {
12214 Value *PassThru = ZeroMask ? Constant::getNullValue(Res->getType())
12215 : Ops[PTIdx];
12216
12217 // If we negated the accumulator and the its the PassThru value we need to
12218 // bypass the negate. Conveniently Upper should be the same thing in this
12219 // case.
12220 if (NegAcc && PTIdx == 2)
12221 PassThru = CGF.Builder.CreateExtractElement(Upper, (uint64_t)0);
12222
12223 Res = EmitX86ScalarSelect(CGF, Ops[3], Res, PassThru);
12224 }
12225 return CGF.Builder.CreateInsertElement(Upper, Res, (uint64_t)0);
12226}
12227
12228static Value *EmitX86Muldq(CodeGenFunction &CGF, bool IsSigned,
12229 ArrayRef<Value *> Ops) {
12230 llvm::Type *Ty = Ops[0]->getType();
12231 // Arguments have a vXi32 type so cast to vXi64.
12232 Ty = llvm::FixedVectorType::get(CGF.Int64Ty,
12233 Ty->getPrimitiveSizeInBits() / 64);
12234 Value *LHS = CGF.Builder.CreateBitCast(Ops[0], Ty);
12235 Value *RHS = CGF.Builder.CreateBitCast(Ops[1], Ty);
12236
12237 if (IsSigned) {
12238 // Shift left then arithmetic shift right.
12239 Constant *ShiftAmt = ConstantInt::get(Ty, 32);
12240 LHS = CGF.Builder.CreateShl(LHS, ShiftAmt);
12241 LHS = CGF.Builder.CreateAShr(LHS, ShiftAmt);
12242 RHS = CGF.Builder.CreateShl(RHS, ShiftAmt);
12243 RHS = CGF.Builder.CreateAShr(RHS, ShiftAmt);
12244 } else {
12245 // Clear the upper bits.
12246 Constant *Mask = ConstantInt::get(Ty, 0xffffffff);
12247 LHS = CGF.Builder.CreateAnd(LHS, Mask);
12248 RHS = CGF.Builder.CreateAnd(RHS, Mask);
12249 }
12250
12251 return CGF.Builder.CreateMul(LHS, RHS);
12252}
12253
12254// Emit a masked pternlog intrinsic. This only exists because the header has to
12255// use a macro and we aren't able to pass the input argument to a pternlog
12256// builtin and a select builtin without evaluating it twice.
12257static Value *EmitX86Ternlog(CodeGenFunction &CGF, bool ZeroMask,
12258 ArrayRef<Value *> Ops) {
12259 llvm::Type *Ty = Ops[0]->getType();
12260
12261 unsigned VecWidth = Ty->getPrimitiveSizeInBits();
12262 unsigned EltWidth = Ty->getScalarSizeInBits();
12263 Intrinsic::ID IID;
12264 if (VecWidth == 128 && EltWidth == 32)
12265 IID = Intrinsic::x86_avx512_pternlog_d_128;
12266 else if (VecWidth == 256 && EltWidth == 32)
12267 IID = Intrinsic::x86_avx512_pternlog_d_256;
12268 else if (VecWidth == 512 && EltWidth == 32)
12269 IID = Intrinsic::x86_avx512_pternlog_d_512;
12270 else if (VecWidth == 128 && EltWidth == 64)
12271 IID = Intrinsic::x86_avx512_pternlog_q_128;
12272 else if (VecWidth == 256 && EltWidth == 64)
12273 IID = Intrinsic::x86_avx512_pternlog_q_256;
12274 else if (VecWidth == 512 && EltWidth == 64)
12275 IID = Intrinsic::x86_avx512_pternlog_q_512;
12276 else
12277 llvm_unreachable("Unexpected intrinsic")::llvm::llvm_unreachable_internal("Unexpected intrinsic", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12277)
;
12278
12279 Value *Ternlog = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID),
12280 Ops.drop_back());
12281 Value *PassThru = ZeroMask ? ConstantAggregateZero::get(Ty) : Ops[0];
12282 return EmitX86Select(CGF, Ops[4], Ternlog, PassThru);
12283}
12284
12285static Value *EmitX86SExtMask(CodeGenFunction &CGF, Value *Op,
12286 llvm::Type *DstTy) {
12287 unsigned NumberOfElements =
12288 cast<llvm::FixedVectorType>(DstTy)->getNumElements();
12289 Value *Mask = getMaskVecValue(CGF, Op, NumberOfElements);
12290 return CGF.Builder.CreateSExt(Mask, DstTy, "vpmovm2");
12291}
12292
12293// Emit binary intrinsic with the same type used in result/args.
12294static Value *EmitX86BinaryIntrinsic(CodeGenFunction &CGF,
12295 ArrayRef<Value *> Ops, Intrinsic::ID IID) {
12296 llvm::Function *F = CGF.CGM.getIntrinsic(IID, Ops[0]->getType());
12297 return CGF.Builder.CreateCall(F, {Ops[0], Ops[1]});
12298}
12299
12300Value *CodeGenFunction::EmitX86CpuIs(const CallExpr *E) {
12301 const Expr *CPUExpr = E->getArg(0)->IgnoreParenCasts();
12302 StringRef CPUStr = cast<clang::StringLiteral>(CPUExpr)->getString();
12303 return EmitX86CpuIs(CPUStr);
12304}
12305
12306// Convert F16 halfs to floats.
12307static Value *EmitX86CvtF16ToFloatExpr(CodeGenFunction &CGF,
12308 ArrayRef<Value *> Ops,
12309 llvm::Type *DstTy) {
12310 assert((Ops.size() == 1 || Ops.size() == 3 || Ops.size() == 4) &&(static_cast <bool> ((Ops.size() == 1 || Ops.size() == 3
|| Ops.size() == 4) && "Unknown cvtph2ps intrinsic")
? void (0) : __assert_fail ("(Ops.size() == 1 || Ops.size() == 3 || Ops.size() == 4) && \"Unknown cvtph2ps intrinsic\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12311, __extension__ __PRETTY_FUNCTION__))
12311 "Unknown cvtph2ps intrinsic")(static_cast <bool> ((Ops.size() == 1 || Ops.size() == 3
|| Ops.size() == 4) && "Unknown cvtph2ps intrinsic")
? void (0) : __assert_fail ("(Ops.size() == 1 || Ops.size() == 3 || Ops.size() == 4) && \"Unknown cvtph2ps intrinsic\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12311, __extension__ __PRETTY_FUNCTION__))
;
12312
12313 // If the SAE intrinsic doesn't use default rounding then we can't upgrade.
12314 if (Ops.size() == 4 && cast<llvm::ConstantInt>(Ops[3])->getZExtValue() != 4) {
12315 Function *F =
12316 CGF.CGM.getIntrinsic(Intrinsic::x86_avx512_mask_vcvtph2ps_512);
12317 return CGF.Builder.CreateCall(F, {Ops[0], Ops[1], Ops[2], Ops[3]});
12318 }
12319
12320 unsigned NumDstElts = cast<llvm::FixedVectorType>(DstTy)->getNumElements();
12321 Value *Src = Ops[0];
12322
12323 // Extract the subvector.
12324 if (NumDstElts !=
12325 cast<llvm::FixedVectorType>(Src->getType())->getNumElements()) {
12326 assert(NumDstElts == 4 && "Unexpected vector size")(static_cast <bool> (NumDstElts == 4 && "Unexpected vector size"
) ? void (0) : __assert_fail ("NumDstElts == 4 && \"Unexpected vector size\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12326, __extension__ __PRETTY_FUNCTION__))
;
12327 Src = CGF.Builder.CreateShuffleVector(Src, ArrayRef<int>{0, 1, 2, 3});
12328 }
12329
12330 // Bitcast from vXi16 to vXf16.
12331 auto *HalfTy = llvm::FixedVectorType::get(
12332 llvm::Type::getHalfTy(CGF.getLLVMContext()), NumDstElts);
12333 Src = CGF.Builder.CreateBitCast(Src, HalfTy);
12334
12335 // Perform the fp-extension.
12336 Value *Res = CGF.Builder.CreateFPExt(Src, DstTy, "cvtph2ps");
12337
12338 if (Ops.size() >= 3)
12339 Res = EmitX86Select(CGF, Ops[2], Res, Ops[1]);
12340 return Res;
12341}
12342
12343// Convert a BF16 to a float.
12344static Value *EmitX86CvtBF16ToFloatExpr(CodeGenFunction &CGF,
12345 const CallExpr *E,
12346 ArrayRef<Value *> Ops) {
12347 llvm::Type *Int32Ty = CGF.Builder.getInt32Ty();
12348 Value *ZeroExt = CGF.Builder.CreateZExt(Ops[0], Int32Ty);
12349 Value *Shl = CGF.Builder.CreateShl(ZeroExt, 16);
12350 llvm::Type *ResultType = CGF.ConvertType(E->getType());
12351 Value *BitCast = CGF.Builder.CreateBitCast(Shl, ResultType);
12352 return BitCast;
12353}
12354
12355Value *CodeGenFunction::EmitX86CpuIs(StringRef CPUStr) {
12356
12357 llvm::Type *Int32Ty = Builder.getInt32Ty();
12358
12359 // Matching the struct layout from the compiler-rt/libgcc structure that is
12360 // filled in:
12361 // unsigned int __cpu_vendor;
12362 // unsigned int __cpu_type;
12363 // unsigned int __cpu_subtype;
12364 // unsigned int __cpu_features[1];
12365 llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty,
12366 llvm::ArrayType::get(Int32Ty, 1));
12367
12368 // Grab the global __cpu_model.
12369 llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model");
12370 cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true);
12371
12372 // Calculate the index needed to access the correct field based on the
12373 // range. Also adjust the expected value.
12374 unsigned Index;
12375 unsigned Value;
12376 std::tie(Index, Value) = StringSwitch<std::pair<unsigned, unsigned>>(CPUStr)
12377#define X86_VENDOR(ENUM, STRING) \
12378 .Case(STRING, {0u, static_cast<unsigned>(llvm::X86::ENUM)})
12379#define X86_CPU_TYPE_ALIAS(ENUM, ALIAS) \
12380 .Case(ALIAS, {1u, static_cast<unsigned>(llvm::X86::ENUM)})
12381#define X86_CPU_TYPE(ENUM, STR) \
12382 .Case(STR, {1u, static_cast<unsigned>(llvm::X86::ENUM)})
12383#define X86_CPU_SUBTYPE(ENUM, STR) \
12384 .Case(STR, {2u, static_cast<unsigned>(llvm::X86::ENUM)})
12385#include "llvm/Support/X86TargetParser.def"
12386 .Default({0, 0});
12387 assert(Value != 0 && "Invalid CPUStr passed to CpuIs")(static_cast <bool> (Value != 0 && "Invalid CPUStr passed to CpuIs"
) ? void (0) : __assert_fail ("Value != 0 && \"Invalid CPUStr passed to CpuIs\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12387, __extension__ __PRETTY_FUNCTION__))
;
12388
12389 // Grab the appropriate field from __cpu_model.
12390 llvm::Value *Idxs[] = {ConstantInt::get(Int32Ty, 0),
12391 ConstantInt::get(Int32Ty, Index)};
12392 llvm::Value *CpuValue = Builder.CreateGEP(STy, CpuModel, Idxs);
12393 CpuValue = Builder.CreateAlignedLoad(Int32Ty, CpuValue,
12394 CharUnits::fromQuantity(4));
12395
12396 // Check the value of the field against the requested value.
12397 return Builder.CreateICmpEQ(CpuValue,
12398 llvm::ConstantInt::get(Int32Ty, Value));
12399}
12400
12401Value *CodeGenFunction::EmitX86CpuSupports(const CallExpr *E) {
12402 const Expr *FeatureExpr = E->getArg(0)->IgnoreParenCasts();
12403 StringRef FeatureStr = cast<StringLiteral>(FeatureExpr)->getString();
12404 return EmitX86CpuSupports(FeatureStr);
12405}
12406
12407uint64_t
12408CodeGenFunction::GetX86CpuSupportsMask(ArrayRef<StringRef> FeatureStrs) {
12409 // Processor features and mapping to processor feature value.
12410 uint64_t FeaturesMask = 0;
12411 for (const StringRef &FeatureStr : FeatureStrs) {
12412 unsigned Feature =
12413 StringSwitch<unsigned>(FeatureStr)
12414#define X86_FEATURE_COMPAT(ENUM, STR, PRIORITY) \
12415 .Case(STR, llvm::X86::FEATURE_##ENUM)
12416#include "llvm/Support/X86TargetParser.def"
12417 ;
12418 FeaturesMask |= (1ULL << Feature);
12419 }
12420 return FeaturesMask;
12421}
12422
12423Value *CodeGenFunction::EmitX86CpuSupports(ArrayRef<StringRef> FeatureStrs) {
12424 return EmitX86CpuSupports(GetX86CpuSupportsMask(FeatureStrs));
12425}
12426
12427llvm::Value *CodeGenFunction::EmitX86CpuSupports(uint64_t FeaturesMask) {
12428 uint32_t Features1 = Lo_32(FeaturesMask);
12429 uint32_t Features2 = Hi_32(FeaturesMask);
12430
12431 Value *Result = Builder.getTrue();
12432
12433 if (Features1 != 0) {
12434 // Matching the struct layout from the compiler-rt/libgcc structure that is
12435 // filled in:
12436 // unsigned int __cpu_vendor;
12437 // unsigned int __cpu_type;
12438 // unsigned int __cpu_subtype;
12439 // unsigned int __cpu_features[1];
12440 llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty,
12441 llvm::ArrayType::get(Int32Ty, 1));
12442
12443 // Grab the global __cpu_model.
12444 llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model");
12445 cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true);
12446
12447 // Grab the first (0th) element from the field __cpu_features off of the
12448 // global in the struct STy.
12449 Value *Idxs[] = {Builder.getInt32(0), Builder.getInt32(3),
12450 Builder.getInt32(0)};
12451 Value *CpuFeatures = Builder.CreateGEP(STy, CpuModel, Idxs);
12452 Value *Features = Builder.CreateAlignedLoad(Int32Ty, CpuFeatures,
12453 CharUnits::fromQuantity(4));
12454
12455 // Check the value of the bit corresponding to the feature requested.
12456 Value *Mask = Builder.getInt32(Features1);
12457 Value *Bitset = Builder.CreateAnd(Features, Mask);
12458 Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask);
12459 Result = Builder.CreateAnd(Result, Cmp);
12460 }
12461
12462 if (Features2 != 0) {
12463 llvm::Constant *CpuFeatures2 = CGM.CreateRuntimeVariable(Int32Ty,
12464 "__cpu_features2");
12465 cast<llvm::GlobalValue>(CpuFeatures2)->setDSOLocal(true);
12466
12467 Value *Features = Builder.CreateAlignedLoad(Int32Ty, CpuFeatures2,
12468 CharUnits::fromQuantity(4));
12469
12470 // Check the value of the bit corresponding to the feature requested.
12471 Value *Mask = Builder.getInt32(Features2);
12472 Value *Bitset = Builder.CreateAnd(Features, Mask);
12473 Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask);
12474 Result = Builder.CreateAnd(Result, Cmp);
12475 }
12476
12477 return Result;
12478}
12479
12480Value *CodeGenFunction::EmitX86CpuInit() {
12481 llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy,
12482 /*Variadic*/ false);
12483 llvm::FunctionCallee Func =
12484 CGM.CreateRuntimeFunction(FTy, "__cpu_indicator_init");
12485 cast<llvm::GlobalValue>(Func.getCallee())->setDSOLocal(true);
12486 cast<llvm::GlobalValue>(Func.getCallee())
12487 ->setDLLStorageClass(llvm::GlobalValue::DefaultStorageClass);
12488 return Builder.CreateCall(Func);
12489}
12490
12491Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
12492 const CallExpr *E) {
12493 if (BuiltinID == X86::BI__builtin_cpu_is)
1
Assuming 'BuiltinID' is not equal to BI__builtin_cpu_is
2
Taking false branch
12494 return EmitX86CpuIs(E);
12495 if (BuiltinID == X86::BI__builtin_cpu_supports)
3
Assuming 'BuiltinID' is not equal to BI__builtin_cpu_supports
4
Taking false branch
12496 return EmitX86CpuSupports(E);
12497 if (BuiltinID == X86::BI__builtin_cpu_init)
5
Assuming 'BuiltinID' is not equal to BI__builtin_cpu_init
6
Taking false branch
12498 return EmitX86CpuInit();
12499
12500 // Handle MSVC intrinsics before argument evaluation to prevent double
12501 // evaluation.
12502 if (Optional<MSVCIntrin> MsvcIntId = translateX86ToMsvcIntrin(BuiltinID))
7
Taking false branch
12503 return EmitMSVCBuiltinExpr(*MsvcIntId, E);
12504
12505 SmallVector<Value*, 4> Ops;
12506 bool IsMaskFCmp = false;
12507
12508 // Find out if any arguments are required to be integer constant expressions.
12509 unsigned ICEArguments = 0;
12510 ASTContext::GetBuiltinTypeError Error;
12511 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
12512 assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None &&
"Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12512, __extension__ __PRETTY_FUNCTION__))
;
8
Assuming 'Error' is equal to GE_None
9
'?' condition is true
12513
12514 for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) {
10
Assuming 'i' is equal to 'e'
11
Loop condition is false. Execution continues on line 12533
12515 // If this is a normal argument, just emit it as a scalar.
12516 if ((ICEArguments & (1 << i)) == 0) {
12517 Ops.push_back(EmitScalarExpr(E->getArg(i)));
12518 continue;
12519 }
12520
12521 // If this is required to be a constant, constant fold it so that we know
12522 // that the generated intrinsic gets a ConstantInt.
12523 Ops.push_back(llvm::ConstantInt::get(
12524 getLLVMContext(), *E->getArg(i)->getIntegerConstantExpr(getContext())));
12525 }
12526
12527 // These exist so that the builtin that takes an immediate can be bounds
12528 // checked by clang to avoid passing bad immediates to the backend. Since
12529 // AVX has a larger immediate than SSE we would need separate builtins to
12530 // do the different bounds checking. Rather than create a clang specific
12531 // SSE only builtin, this implements eight separate builtins to match gcc
12532 // implementation.
12533 auto getCmpIntrinsicCall = [this, &Ops](Intrinsic::ID ID, unsigned Imm) {
12534 Ops.push_back(llvm::ConstantInt::get(Int8Ty, Imm));
12535 llvm::Function *F = CGM.getIntrinsic(ID);
12536 return Builder.CreateCall(F, Ops);
12537 };
12538
12539 // For the vector forms of FP comparisons, translate the builtins directly to
12540 // IR.
12541 // TODO: The builtins could be removed if the SSE header files used vector
12542 // extension comparisons directly (vector ordered/unordered may need
12543 // additional support via __builtin_isnan()).
12544 auto getVectorFCmpIR = [this, &Ops, E](CmpInst::Predicate Pred,
12545 bool IsSignaling) {
12546 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
12547 Value *Cmp;
12548 if (IsSignaling)
12549 Cmp = Builder.CreateFCmpS(Pred, Ops[0], Ops[1]);
12550 else
12551 Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]);
12552 llvm::VectorType *FPVecTy = cast<llvm::VectorType>(Ops[0]->getType());
12553 llvm::VectorType *IntVecTy = llvm::VectorType::getInteger(FPVecTy);
12554 Value *Sext = Builder.CreateSExt(Cmp, IntVecTy);
12555 return Builder.CreateBitCast(Sext, FPVecTy);
12556 };
12557
12558 switch (BuiltinID) {
12
Control jumps to 'case BI__builtin_ia32_cvtq2mask512:' at line 12783
12559 default: return nullptr;
12560 case X86::BI_mm_prefetch: {
12561 Value *Address = Ops[0];
12562 ConstantInt *C = cast<ConstantInt>(Ops[1]);
12563 Value *RW = ConstantInt::get(Int32Ty, (C->getZExtValue() >> 2) & 0x1);
12564 Value *Locality = ConstantInt::get(Int32Ty, C->getZExtValue() & 0x3);
12565 Value *Data = ConstantInt::get(Int32Ty, 1);
12566 Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType());
12567 return Builder.CreateCall(F, {Address, RW, Locality, Data});
12568 }
12569 case X86::BI_mm_clflush: {
12570 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_clflush),
12571 Ops[0]);
12572 }
12573 case X86::BI_mm_lfence: {
12574 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_lfence));
12575 }
12576 case X86::BI_mm_mfence: {
12577 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_mfence));
12578 }
12579 case X86::BI_mm_sfence: {
12580 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_sfence));
12581 }
12582 case X86::BI_mm_pause: {
12583 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_pause));
12584 }
12585 case X86::BI__rdtsc: {
12586 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtsc));
12587 }
12588 case X86::BI__builtin_ia32_rdtscp: {
12589 Value *Call = Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtscp));
12590 Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1),
12591 Ops[0]);
12592 return Builder.CreateExtractValue(Call, 0);
12593 }
12594 case X86::BI__builtin_ia32_lzcnt_u16:
12595 case X86::BI__builtin_ia32_lzcnt_u32:
12596 case X86::BI__builtin_ia32_lzcnt_u64: {
12597 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType());
12598 return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)});
12599 }
12600 case X86::BI__builtin_ia32_tzcnt_u16:
12601 case X86::BI__builtin_ia32_tzcnt_u32:
12602 case X86::BI__builtin_ia32_tzcnt_u64: {
12603 Function *F = CGM.getIntrinsic(Intrinsic::cttz, Ops[0]->getType());
12604 return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)});
12605 }
12606 case X86::BI__builtin_ia32_undef128:
12607 case X86::BI__builtin_ia32_undef256:
12608 case X86::BI__builtin_ia32_undef512:
12609 // The x86 definition of "undef" is not the same as the LLVM definition
12610 // (PR32176). We leave optimizing away an unnecessary zero constant to the
12611 // IR optimizer and backend.
12612 // TODO: If we had a "freeze" IR instruction to generate a fixed undef
12613 // value, we should use that here instead of a zero.
12614 return llvm::Constant::getNullValue(ConvertType(E->getType()));
12615 case X86::BI__builtin_ia32_vec_init_v8qi:
12616 case X86::BI__builtin_ia32_vec_init_v4hi:
12617 case X86::BI__builtin_ia32_vec_init_v2si:
12618 return Builder.CreateBitCast(BuildVector(Ops),
12619 llvm::Type::getX86_MMXTy(getLLVMContext()));
12620 case X86::BI__builtin_ia32_vec_ext_v2si:
12621 case X86::BI__builtin_ia32_vec_ext_v16qi:
12622 case X86::BI__builtin_ia32_vec_ext_v8hi:
12623 case X86::BI__builtin_ia32_vec_ext_v4si:
12624 case X86::BI__builtin_ia32_vec_ext_v4sf:
12625 case X86::BI__builtin_ia32_vec_ext_v2di:
12626 case X86::BI__builtin_ia32_vec_ext_v32qi:
12627 case X86::BI__builtin_ia32_vec_ext_v16hi:
12628 case X86::BI__builtin_ia32_vec_ext_v8si:
12629 case X86::BI__builtin_ia32_vec_ext_v4di: {
12630 unsigned NumElts =
12631 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
12632 uint64_t Index = cast<ConstantInt>(Ops[1])->getZExtValue();
12633 Index &= NumElts - 1;
12634 // These builtins exist so we can ensure the index is an ICE and in range.
12635 // Otherwise we could just do this in the header file.
12636 return Builder.CreateExtractElement(Ops[0], Index);
12637 }
12638 case X86::BI__builtin_ia32_vec_set_v16qi:
12639 case X86::BI__builtin_ia32_vec_set_v8hi:
12640 case X86::BI__builtin_ia32_vec_set_v4si:
12641 case X86::BI__builtin_ia32_vec_set_v2di:
12642 case X86::BI__builtin_ia32_vec_set_v32qi:
12643 case X86::BI__builtin_ia32_vec_set_v16hi:
12644 case X86::BI__builtin_ia32_vec_set_v8si:
12645 case X86::BI__builtin_ia32_vec_set_v4di: {
12646 unsigned NumElts =
12647 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
12648 unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue();
12649 Index &= NumElts - 1;
12650 // These builtins exist so we can ensure the index is an ICE and in range.
12651 // Otherwise we could just do this in the header file.
12652 return Builder.CreateInsertElement(Ops[0], Ops[1], Index);
12653 }
12654 case X86::BI_mm_setcsr:
12655 case X86::BI__builtin_ia32_ldmxcsr: {
12656 Address Tmp = CreateMemTemp(E->getArg(0)->getType());
12657 Builder.CreateStore(Ops[0], Tmp);
12658 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_ldmxcsr),
12659 Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy));
12660 }
12661 case X86::BI_mm_getcsr:
12662 case X86::BI__builtin_ia32_stmxcsr: {
12663 Address Tmp = CreateMemTemp(E->getType());
12664 Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_stmxcsr),
12665 Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy));
12666 return Builder.CreateLoad(Tmp, "stmxcsr");
12667 }
12668 case X86::BI__builtin_ia32_xsave:
12669 case X86::BI__builtin_ia32_xsave64:
12670 case X86::BI__builtin_ia32_xrstor:
12671 case X86::BI__builtin_ia32_xrstor64:
12672 case X86::BI__builtin_ia32_xsaveopt:
12673 case X86::BI__builtin_ia32_xsaveopt64:
12674 case X86::BI__builtin_ia32_xrstors:
12675 case X86::BI__builtin_ia32_xrstors64:
12676 case X86::BI__builtin_ia32_xsavec:
12677 case X86::BI__builtin_ia32_xsavec64:
12678 case X86::BI__builtin_ia32_xsaves:
12679 case X86::BI__builtin_ia32_xsaves64:
12680 case X86::BI__builtin_ia32_xsetbv:
12681 case X86::BI_xsetbv: {
12682 Intrinsic::ID ID;
12683#define INTRINSIC_X86_XSAVE_ID(NAME) \
12684 case X86::BI__builtin_ia32_##NAME: \
12685 ID = Intrinsic::x86_##NAME; \
12686 break
12687 switch (BuiltinID) {
12688 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 12688)
;
12689 INTRINSIC_X86_XSAVE_ID(xsave);
12690 INTRINSIC_X86_XSAVE_ID(xsave64);
12691 INTRINSIC_X86_XSAVE_ID(xrstor);
12692 INTRINSIC_X86_XSAVE_ID(xrstor64);
12693 INTRINSIC_X86_XSAVE_ID(xsaveopt);
12694 INTRINSIC_X86_XSAVE_ID(xsaveopt64);
12695 INTRINSIC_X86_XSAVE_ID(xrstors);
12696 INTRINSIC_X86_XSAVE_ID(xrstors64);
12697 INTRINSIC_X86_XSAVE_ID(xsavec);
12698 INTRINSIC_X86_XSAVE_ID(xsavec64);
12699 INTRINSIC_X86_XSAVE_ID(xsaves);
12700 INTRINSIC_X86_XSAVE_ID(xsaves64);
12701 INTRINSIC_X86_XSAVE_ID(xsetbv);
12702 case X86::BI_xsetbv:
12703 ID = Intrinsic::x86_xsetbv;
12704 break;
12705 }
12706#undef INTRINSIC_X86_XSAVE_ID
12707 Value *Mhi = Builder.CreateTrunc(
12708 Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, 32)), Int32Ty);
12709 Value *Mlo = Builder.CreateTrunc(Ops[1], Int32Ty);
12710 Ops[1] = Mhi;
12711 Ops.push_back(Mlo);
12712 return Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
12713 }
12714 case X86::BI__builtin_ia32_xgetbv:
12715 case X86::BI_xgetbv:
12716 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_xgetbv), Ops);
12717 case X86::BI__builtin_ia32_storedqudi128_mask:
12718 case X86::BI__builtin_ia32_storedqusi128_mask:
12719 case X86::BI__builtin_ia32_storedquhi128_mask:
12720 case X86::BI__builtin_ia32_storedquqi128_mask:
12721 case X86::BI__builtin_ia32_storeupd128_mask:
12722 case X86::BI__builtin_ia32_storeups128_mask:
12723 case X86::BI__builtin_ia32_storedqudi256_mask:
12724 case X86::BI__builtin_ia32_storedqusi256_mask:
12725 case X86::BI__builtin_ia32_storedquhi256_mask:
12726 case X86::BI__builtin_ia32_storedquqi256_mask:
12727 case X86::BI__builtin_ia32_storeupd256_mask:
12728 case X86::BI__builtin_ia32_storeups256_mask:
12729 case X86::BI__builtin_ia32_storedqudi512_mask:
12730 case X86::BI__builtin_ia32_storedqusi512_mask:
12731 case X86::BI__builtin_ia32_storedquhi512_mask:
12732 case X86::BI__builtin_ia32_storedquqi512_mask:
12733 case X86::BI__builtin_ia32_storeupd512_mask:
12734 case X86::BI__builtin_ia32_storeups512_mask:
12735 return EmitX86MaskedStore(*this, Ops, Align(1));
12736
12737 case X86::BI__builtin_ia32_storesh128_mask:
12738 case X86::BI__builtin_ia32_storess128_mask:
12739 case X86::BI__builtin_ia32_storesd128_mask:
12740 return EmitX86MaskedStore(*this, Ops, Align(1));
12741
12742 case X86::BI__builtin_ia32_vpopcntb_128:
12743 case X86::BI__builtin_ia32_vpopcntd_128:
12744 case X86::BI__builtin_ia32_vpopcntq_128:
12745 case X86::BI__builtin_ia32_vpopcntw_128:
12746 case X86::BI__builtin_ia32_vpopcntb_256:
12747 case X86::BI__builtin_ia32_vpopcntd_256:
12748 case X86::BI__builtin_ia32_vpopcntq_256:
12749 case X86::BI__builtin_ia32_vpopcntw_256:
12750 case X86::BI__builtin_ia32_vpopcntb_512:
12751 case X86::BI__builtin_ia32_vpopcntd_512:
12752 case X86::BI__builtin_ia32_vpopcntq_512:
12753 case X86::BI__builtin_ia32_vpopcntw_512: {
12754 llvm::Type *ResultType = ConvertType(E->getType());
12755 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
12756 return Builder.CreateCall(F, Ops);
12757 }
12758 case X86::BI__builtin_ia32_cvtmask2b128:
12759 case X86::BI__builtin_ia32_cvtmask2b256:
12760 case X86::BI__builtin_ia32_cvtmask2b512:
12761 case X86::BI__builtin_ia32_cvtmask2w128:
12762 case X86::BI__builtin_ia32_cvtmask2w256:
12763 case X86::BI__builtin_ia32_cvtmask2w512:
12764 case X86::BI__builtin_ia32_cvtmask2d128:
12765 case X86::BI__builtin_ia32_cvtmask2d256:
12766 case X86::BI__builtin_ia32_cvtmask2d512:
12767 case X86::BI__builtin_ia32_cvtmask2q128:
12768 case X86::BI__builtin_ia32_cvtmask2q256:
12769 case X86::BI__builtin_ia32_cvtmask2q512:
12770 return EmitX86SExtMask(*this, Ops[0], ConvertType(E->getType()));
12771
12772 case X86::BI__builtin_ia32_cvtb2mask128:
12773 case X86::BI__builtin_ia32_cvtb2mask256:
12774 case X86::BI__builtin_ia32_cvtb2mask512:
12775 case X86::BI__builtin_ia32_cvtw2mask128:
12776 case X86::BI__builtin_ia32_cvtw2mask256:
12777 case X86::BI__builtin_ia32_cvtw2mask512:
12778 case X86::BI__builtin_ia32_cvtd2mask128:
12779 case X86::BI__builtin_ia32_cvtd2mask256:
12780 case X86::BI__builtin_ia32_cvtd2mask512:
12781 case X86::BI__builtin_ia32_cvtq2mask128:
12782 case X86::BI__builtin_ia32_cvtq2mask256:
12783 case X86::BI__builtin_ia32_cvtq2mask512:
12784 return EmitX86ConvertToMask(*this, Ops[0]);
13
Calling 'EmitX86ConvertToMask'
12785
12786 case X86::BI__builtin_ia32_cvtdq2ps512_mask:
12787 case X86::BI__builtin_ia32_cvtqq2ps512_mask:
12788 case X86::BI__builtin_ia32_cvtqq2pd512_mask:
12789 case X86::BI__builtin_ia32_vcvtw2ph512_mask:
12790 case X86::BI__builtin_ia32_vcvtdq2ph512_mask:
12791 case X86::BI__builtin_ia32_vcvtqq2ph512_mask:
12792 return EmitX86ConvertIntToFp(*this, E, Ops, /*IsSigned*/ true);
12793 case X86::BI__builtin_ia32_cvtudq2ps512_mask:
12794 case X86::BI__builtin_ia32_cvtuqq2ps512_mask:
12795 case X86::BI__builtin_ia32_cvtuqq2pd512_mask:
12796 case X86::BI__builtin_ia32_vcvtuw2ph512_mask:
12797 case X86::BI__builtin_ia32_vcvtudq2ph512_mask:
12798 case X86::BI__builtin_ia32_vcvtuqq2ph512_mask:
12799 return EmitX86ConvertIntToFp(*this, E, Ops, /*IsSigned*/ false);
12800
12801 case X86::BI__builtin_ia32_vfmaddss3:
12802 case X86::BI__builtin_ia32_vfmaddsd3:
12803 case X86::BI__builtin_ia32_vfmaddsh3_mask:
12804 case X86::BI__builtin_ia32_vfmaddss3_mask:
12805 case X86::BI__builtin_ia32_vfmaddsd3_mask:
12806 return EmitScalarFMAExpr(*this, E, Ops, Ops[0]);
12807 case X86::BI__builtin_ia32_vfmaddss:
12808 case X86::BI__builtin_ia32_vfmaddsd:
12809 return EmitScalarFMAExpr(*this, E, Ops,
12810 Constant::getNullValue(Ops[0]->getType()));
12811 case X86::BI__builtin_ia32_vfmaddsh3_maskz:
12812 case X86::BI__builtin_ia32_vfmaddss3_maskz:
12813 case X86::BI__builtin_ia32_vfmaddsd3_maskz:
12814 return EmitScalarFMAExpr(*this, E, Ops, Ops[0], /*ZeroMask*/ true);
12815 case X86::BI__builtin_ia32_vfmaddsh3_mask3:
12816 case X86::BI__builtin_ia32_vfmaddss3_mask3:
12817 case X86::BI__builtin_ia32_vfmaddsd3_mask3:
12818 return EmitScalarFMAExpr(*this, E, Ops, Ops[2], /*ZeroMask*/ false, 2);
12819 case X86::BI__builtin_ia32_vfmsubsh3_mask3:
12820 case X86::BI__builtin_ia32_vfmsubss3_mask3:
12821 case X86::BI__builtin_ia32_vfmsubsd3_mask3:
12822 return EmitScalarFMAExpr(*this, E, Ops, Ops[2], /*ZeroMask*/ false, 2,
12823 /*NegAcc*/ true);
12824 case X86::BI__builtin_ia32_vfmaddph:
12825 case X86::BI__builtin_ia32_vfmaddps:
12826 case X86::BI__builtin_ia32_vfmaddpd:
12827 case X86::BI__builtin_ia32_vfmaddph256:
12828 case X86::BI__builtin_ia32_vfmaddps256:
12829 case X86::BI__builtin_ia32_vfmaddpd256:
12830 case X86::BI__builtin_ia32_vfmaddph512_mask:
12831 case X86::BI__builtin_ia32_vfmaddph512_maskz:
12832 case X86::BI__builtin_ia32_vfmaddph512_mask3:
12833 case X86::BI__builtin_ia32_vfmaddps512_mask:
12834 case X86::BI__builtin_ia32_vfmaddps512_maskz:
12835 case X86::BI__builtin_ia32_vfmaddps512_mask3:
12836 case X86::BI__builtin_ia32_vfmsubps512_mask3:
12837 case X86::BI__builtin_ia32_vfmaddpd512_mask:
12838 case X86::BI__builtin_ia32_vfmaddpd512_maskz:
12839 case X86::BI__builtin_ia32_vfmaddpd512_mask3:
12840 case X86::BI__builtin_ia32_vfmsubpd512_mask3:
12841 case X86::BI__builtin_ia32_vfmsubph512_mask3:
12842 return EmitX86FMAExpr(*this, E, Ops, BuiltinID, /*IsAddSub*/ false);
12843 case X86::BI__builtin_ia32_vfmaddsubph512_mask:
12844 case X86::BI__builtin_ia32_vfmaddsubph512_maskz:
12845 case X86::BI__builtin_ia32_vfmaddsubph512_mask3:
12846 case X86::BI__builtin_ia32_vfmsubaddph512_mask3:
12847 case X86::BI__builtin_ia32_vfmaddsubps512_mask:
12848 case X86::BI__builtin_ia32_vfmaddsubps512_maskz:
12849 case X86::BI__builtin_ia32_vfmaddsubps512_mask3:
12850 case X86::BI__builtin_ia32_vfmsubaddps512_mask3:
12851 case X86::BI__builtin_ia32_vfmaddsubpd512_mask:
12852 case X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
12853 case X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
12854 case X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
12855 return EmitX86FMAExpr(*this, E, Ops, BuiltinID, /*IsAddSub*/ true);
12856
12857 case X86::BI__builtin_ia32_movdqa32store128_mask:
12858 case X86::BI__builtin_ia32_movdqa64store128_mask:
12859 case X86::BI__builtin_ia32_storeaps128_mask:
12860 case X86::BI__builtin_ia32_storeapd128_mask:
12861 case X86::BI__builtin_ia32_movdqa32store256_mask:
12862 case X86::BI__builtin_ia32_movdqa64store256_mask:
12863 case X86::BI__builtin_ia32_storeaps256_mask:
12864 case X86::BI__builtin_ia32_storeapd256_mask:
12865 case X86::BI__builtin_ia32_movdqa32store512_mask:
12866 case X86::BI__builtin_ia32_movdqa64store512_mask:
12867 case X86::BI__builtin_ia32_storeaps512_mask:
12868 case X86::BI__builtin_ia32_storeapd512_mask:
12869 return EmitX86MaskedStore(
12870 *this, Ops,
12871 getContext().getTypeAlignInChars(E->getArg(1)->getType()).getAsAlign());
12872
12873 case X86::BI__builtin_ia32_loadups128_mask:
12874 case X86::BI__builtin_ia32_loadups256_mask:
12875 case X86::BI__builtin_ia32_loadups512_mask:
12876 case X86::BI__builtin_ia32_loadupd128_mask:
12877 case X86::BI__builtin_ia32_loadupd256_mask:
12878 case X86::BI__builtin_ia32_loadupd512_mask:
12879 case X86::BI__builtin_ia32_loaddquqi128_mask:
12880 case X86::BI__builtin_ia32_loaddquqi256_mask:
12881 case X86::BI__builtin_ia32_loaddquqi512_mask:
12882 case X86::BI__builtin_ia32_loaddquhi128_mask:
12883 case X86::BI__builtin_ia32_loaddquhi256_mask:
12884 case X86::BI__builtin_ia32_loaddquhi512_mask:
12885 case X86::BI__builtin_ia32_loaddqusi128_mask:
12886 case X86::BI__builtin_ia32_loaddqusi256_mask:
12887 case X86::BI__builtin_ia32_loaddqusi512_mask:
12888 case X86::BI__builtin_ia32_loaddqudi128_mask:
12889 case X86::BI__builtin_ia32_loaddqudi256_mask:
12890 case X86::BI__builtin_ia32_loaddqudi512_mask:
12891 return EmitX86MaskedLoad(*this, Ops, Align(1));
12892
12893 case X86::BI__builtin_ia32_loadsh128_mask:
12894 case X86::BI__builtin_ia32_loadss128_mask:
12895 case X86::BI__builtin_ia32_loadsd128_mask:
12896 return EmitX86MaskedLoad(*this, Ops, Align(1));
12897
12898 case X86::BI__builtin_ia32_loadaps128_mask:
12899 case X86::BI__builtin_ia32_loadaps256_mask:
12900 case X86::BI__builtin_ia32_loadaps512_mask:
12901 case X86::BI__builtin_ia32_loadapd128_mask:
12902 case X86::BI__builtin_ia32_loadapd256_mask:
12903 case X86::BI__builtin_ia32_loadapd512_mask:
12904 case X86::BI__builtin_ia32_movdqa32load128_mask:
12905 case X86::BI__builtin_ia32_movdqa32load256_mask:
12906 case X86::BI__builtin_ia32_movdqa32load512_mask:
12907 case X86::BI__builtin_ia32_movdqa64load128_mask:
12908 case X86::BI__builtin_ia32_movdqa64load256_mask:
12909 case X86::BI__builtin_ia32_movdqa64load512_mask:
12910 return EmitX86MaskedLoad(
12911 *this, Ops,
12912 getContext().getTypeAlignInChars(E->getArg(1)->getType()).getAsAlign());
12913
12914 case X86::BI__builtin_ia32_expandloaddf128_mask:
12915 case X86::BI__builtin_ia32_expandloaddf256_mask:
12916 case X86::BI__builtin_ia32_expandloaddf512_mask:
12917 case X86::BI__builtin_ia32_expandloadsf128_mask:
12918 case X86::BI__builtin_ia32_expandloadsf256_mask:
12919 case X86::BI__builtin_ia32_expandloadsf512_mask:
12920 case X86::BI__builtin_ia32_expandloaddi128_mask:
12921 case X86::BI__builtin_ia32_expandloaddi256_mask:
12922 case X86::BI__builtin_ia32_expandloaddi512_mask:
12923 case X86::BI__builtin_ia32_expandloadsi128_mask:
12924 case X86::BI__builtin_ia32_expandloadsi256_mask:
12925 case X86::BI__builtin_ia32_expandloadsi512_mask:
12926 case X86::BI__builtin_ia32_expandloadhi128_mask:
12927 case X86::BI__builtin_ia32_expandloadhi256_mask:
12928 case X86::BI__builtin_ia32_expandloadhi512_mask:
12929 case X86::BI__builtin_ia32_expandloadqi128_mask:
12930 case X86::BI__builtin_ia32_expandloadqi256_mask:
12931 case X86::BI__builtin_ia32_expandloadqi512_mask:
12932 return EmitX86ExpandLoad(*this, Ops);
12933
12934 case X86::BI__builtin_ia32_compressstoredf128_mask:
12935 case X86::BI__builtin_ia32_compressstoredf256_mask:
12936 case X86::BI__builtin_ia32_compressstoredf512_mask:
12937 case X86::BI__builtin_ia32_compressstoresf128_mask:
12938 case X86::BI__builtin_ia32_compressstoresf256_mask:
12939 case X86::BI__builtin_ia32_compressstoresf512_mask:
12940 case X86::BI__builtin_ia32_compressstoredi128_mask:
12941 case X86::BI__builtin_ia32_compressstoredi256_mask:
12942 case X86::BI__builtin_ia32_compressstoredi512_mask:
12943 case X86::BI__builtin_ia32_compressstoresi128_mask:
12944 case X86::BI__builtin_ia32_compressstoresi256_mask:
12945 case X86::BI__builtin_ia32_compressstoresi512_mask:
12946 case X86::BI__builtin_ia32_compressstorehi128_mask:
12947 case X86::BI__builtin_ia32_compressstorehi256_mask:
12948 case X86::BI__builtin_ia32_compressstorehi512_mask:
12949 case X86::BI__builtin_ia32_compressstoreqi128_mask:
12950 case X86::BI__builtin_ia32_compressstoreqi256_mask:
12951 case X86::BI__builtin_ia32_compressstoreqi512_mask:
12952 return EmitX86CompressStore(*this, Ops);
12953
12954 case X86::BI__builtin_ia32_expanddf128_mask:
12955 case X86::BI__builtin_ia32_expanddf256_mask:
12956 case X86::BI__builtin_ia32_expanddf512_mask:
12957 case X86::BI__builtin_ia32_expandsf128_mask:
12958 case X86::BI__builtin_ia32_expandsf256_mask:
12959 case X86::BI__builtin_ia32_expandsf512_mask:
12960 case X86::BI__builtin_ia32_expanddi128_mask:
12961 case X86::BI__builtin_ia32_expanddi256_mask:
12962 case X86::BI__builtin_ia32_expanddi512_mask:
12963 case X86::BI__builtin_ia32_expandsi128_mask:
12964 case X86::BI__builtin_ia32_expandsi256_mask:
12965 case X86::BI__builtin_ia32_expandsi512_mask:
12966 case X86::BI__builtin_ia32_expandhi128_mask:
12967 case X86::BI__builtin_ia32_expandhi256_mask:
12968 case X86::BI__builtin_ia32_expandhi512_mask:
12969 case X86::BI__builtin_ia32_expandqi128_mask:
12970 case X86::BI__builtin_ia32_expandqi256_mask:
12971 case X86::BI__builtin_ia32_expandqi512_mask:
12972 return EmitX86CompressExpand(*this, Ops, /*IsCompress*/false);
12973
12974 case X86::BI__builtin_ia32_compressdf128_mask:
12975 case X86::BI__builtin_ia32_compressdf256_mask:
12976 case X86::BI__builtin_ia32_compressdf512_mask:
12977 case X86::BI__builtin_ia32_compresssf128_mask:
12978 case X86::BI__builtin_ia32_compresssf256_mask:
12979 case X86::BI__builtin_ia32_compresssf512_mask:
12980 case X86::BI__builtin_ia32_compressdi128_mask:
12981 case X86::BI__builtin_ia32_compressdi256_mask:
12982 case X86::BI__builtin_ia32_compressdi512_mask:
12983 case X86::BI__builtin_ia32_compresssi128_mask:
12984 case X86::BI__builtin_ia32_compresssi256_mask:
12985 case X86::BI__builtin_ia32_compresssi512_mask:
12986 case X86::BI__builtin_ia32_compresshi128_mask:
12987 case X86::BI__builtin_ia32_compresshi256_mask:
12988 case X86::BI__builtin_ia32_compresshi512_mask:
12989 case X86::BI__builtin_ia32_compressqi128_mask:
12990 case X86::BI__builtin_ia32_compressqi256_mask:
12991 case X86::BI__builtin_ia32_compressqi512_mask:
12992 return EmitX86CompressExpand(*this, Ops, /*IsCompress*/true);
12993
12994 case X86::BI__builtin_ia32_gather3div2df:
12995 case X86::BI__builtin_ia32_gather3div2di:
12996 case X86::BI__builtin_ia32_gather3div4df:
12997 case X86::BI__builtin_ia32_gather3div4di:
12998 case X86::BI__builtin_ia32_gather3div4sf:
12999 case X86::BI__builtin_ia32_gather3div4si:
13000 case X86::BI__builtin_ia32_gather3div8sf:
13001 case X86::BI__builtin_ia32_gather3div8si:
13002 case X86::BI__builtin_ia32_gather3siv2df:
13003 case X86::BI__builtin_ia32_gather3siv2di:
13004 case X86::BI__builtin_ia32_gather3siv4df:
13005 case X86::BI__builtin_ia32_gather3siv4di:
13006 case X86::BI__builtin_ia32_gather3siv4sf:
13007 case X86::BI__builtin_ia32_gather3siv4si:
13008 case X86::BI__builtin_ia32_gather3siv8sf:
13009 case X86::BI__builtin_ia32_gather3siv8si:
13010 case X86::BI__builtin_ia32_gathersiv8df:
13011 case X86::BI__builtin_ia32_gathersiv16sf:
13012 case X86::BI__builtin_ia32_gatherdiv8df:
13013 case X86::BI__builtin_ia32_gatherdiv16sf:
13014 case X86::BI__builtin_ia32_gathersiv8di:
13015 case X86::BI__builtin_ia32_gathersiv16si:
13016 case X86::BI__builtin_ia32_gatherdiv8di:
13017 case X86::BI__builtin_ia32_gatherdiv16si: {
13018 Intrinsic::ID IID;
13019 switch (BuiltinID) {
13020 default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 13020)
;
13021 case X86::BI__builtin_ia32_gather3div2df:
13022 IID = Intrinsic::x86_avx512_mask_gather3div2_df;
13023 break;
13024 case X86::BI__builtin_ia32_gather3div2di:
13025 IID = Intrinsic::x86_avx512_mask_gather3div2_di;
13026 break;
13027 case X86::BI__builtin_ia32_gather3div4df:
13028 IID = Intrinsic::x86_avx512_mask_gather3div4_df;
13029 break;
13030 case X86::BI__builtin_ia32_gather3div4di:
13031 IID = Intrinsic::x86_avx512_mask_gather3div4_di;
13032 break;
13033 case X86::BI__builtin_ia32_gather3div4sf:
13034 IID = Intrinsic::x86_avx512_mask_gather3div4_sf;
13035 break;
13036 case X86::BI__builtin_ia32_gather3div4si:
13037 IID = Intrinsic::x86_avx512_mask_gather3div4_si;
13038 break;
13039 case X86::BI__builtin_ia32_gather3div8sf:
13040 IID = Intrinsic::x86_avx512_mask_gather3div8_sf;
13041 break;
13042 case X86::BI__builtin_ia32_gather3div8si:
13043 IID = Intrinsic::x86_avx512_mask_gather3div8_si;
13044 break;
13045 case X86::BI__builtin_ia32_gather3siv2df:
13046 IID = Intrinsic::x86_avx512_mask_gather3siv2_df;
13047 break;
13048 case X86::BI__builtin_ia32_gather3siv2di:
13049 IID = Intrinsic::x86_avx512_mask_gather3siv2_di;
13050 break;
13051 case X86::BI__builtin_ia32_gather3siv4df:
13052 IID = Intrinsic::x86_avx512_mask_gather3siv4_df;
13053 break;
13054 case X86::BI__builtin_ia32_gather3siv4di:
13055 IID = Intrinsic::x86_avx512_mask_gather3siv4_di;
13056 break;
13057 case X86::BI__builtin_ia32_gather3siv4sf:
13058 IID = Intrinsic::x86_avx512_mask_gather3siv4_sf;
13059 break;
13060 case X86::BI__builtin_ia32_gather3siv4si:
13061 IID = Intrinsic::x86_avx512_mask_gather3siv4_si;
13062 break;
13063 case X86::BI__builtin_ia32_gather3siv8sf:
13064 IID = Intrinsic::x86_avx512_mask_gather3siv8_sf;
13065 break;
13066 case X86::BI__builtin_ia32_gather3siv8si:
13067 IID = Intrinsic::x86_avx512_mask_gather3siv8_si;
13068 break;
13069 case X86::BI__builtin_ia32_gathersiv8df:
13070 IID = Intrinsic::x86_avx512_mask_gather_dpd_512;
13071 break;
13072 case X86::BI__builtin_ia32_gathersiv16sf:
13073 IID = Intrinsic::x86_avx512_mask_gather_dps_512;
13074 break;
13075 case X86::BI__builtin_ia32_gatherdiv8df:
13076 IID = Intrinsic::x86_avx512_mask_gather_qpd_512;
13077 break;
13078 case X86::BI__builtin_ia32_gatherdiv16sf:
13079 IID = Intrinsic::x86_avx512_mask_gather_qps_512;
13080 break;
13081 case X86::BI__builtin_ia32_gathersiv8di:
13082 IID = Intrinsic::x86_avx512_mask_gather_dpq_512;
13083 break;
13084 case X86::BI__builtin_ia32_gathersiv16si:
13085 IID = Intrinsic::x86_avx512_mask_gather_dpi_512;
13086 break;
13087 case X86::BI__builtin_ia32_gatherdiv8di:
13088 IID = Intrinsic::x86_avx512_mask_gather_qpq_512;
13089 break;
13090 case X86::BI__builtin_ia32_gatherdiv16si:
13091 IID = Intrinsic::x86_avx512_mask_gather_qpi_512;
13092 break;
13093 }
13094
13095 unsigned MinElts = std::min(
13096 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(),
13097 cast<llvm::FixedVectorType>(Ops[2]->getType())->getNumElements());
13098 Ops[3] = getMaskVecValue(*this, Ops[3], MinElts);
13099 Function *Intr = CGM.getIntrinsic(IID);
13100 return Builder.CreateCall(Intr, Ops);
13101 }
13102
13103 case X86::BI__builtin_ia32_scattersiv8df:
13104 case X86::BI__builtin_ia32_scattersiv16sf:
13105 case X86::BI__builtin_ia32_scatterdiv8df:
13106 case X86::BI__builtin_ia32_scatterdiv16sf:
13107 case X86::BI__builtin_ia32_scattersiv8di:
13108 case X86::BI__builtin_ia32_scattersiv16si:
13109 case X86::BI__builtin_ia32_scatterdiv8di:
13110 case X86::BI__builtin_ia32_scatterdiv16si:
13111 case X86::BI__builtin_ia32_scatterdiv2df:
13112 case X86::BI__builtin_ia32_scatterdiv2di:
13113 case X86::BI__builtin_ia32_scatterdiv4df:
13114 case X86::BI__builtin_ia32_scatterdiv4di:
13115 case X86::BI__builtin_ia32_scatterdiv4sf:
13116 case X86::BI__builtin_ia32_scatterdiv4si:
13117 case X86::BI__builtin_ia32_scatterdiv8sf:
13118 case X86::BI__builtin_ia32_scatterdiv8si:
13119 case X86::BI__builtin_ia32_scattersiv2df:
13120 case X86::BI__builtin_ia32_scattersiv2di:
13121 case X86::BI__builtin_ia32_scattersiv4df:
13122 case X86::BI__builtin_ia32_scattersiv4di:
13123 case X86::BI__builtin_ia32_scattersiv4sf:
13124 case X86::BI__builtin_ia32_scattersiv4si:
13125 case X86::BI__builtin_ia32_scattersiv8sf:
13126 case X86::BI__builtin_ia32_scattersiv8si: {
13127 Intrinsic::ID IID;
13128 switch (BuiltinID) {
13129 default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 13129)
;
13130 case X86::BI__builtin_ia32_scattersiv8df:
13131 IID = Intrinsic::x86_avx512_mask_scatter_dpd_512;
13132 break;
13133 case X86::BI__builtin_ia32_scattersiv16sf:
13134 IID = Intrinsic::x86_avx512_mask_scatter_dps_512;
13135 break;
13136 case X86::BI__builtin_ia32_scatterdiv8df:
13137 IID = Intrinsic::x86_avx512_mask_scatter_qpd_512;
13138 break;
13139 case X86::BI__builtin_ia32_scatterdiv16sf:
13140 IID = Intrinsic::x86_avx512_mask_scatter_qps_512;
13141 break;
13142 case X86::BI__builtin_ia32_scattersiv8di:
13143 IID = Intrinsic::x86_avx512_mask_scatter_dpq_512;
13144 break;
13145 case X86::BI__builtin_ia32_scattersiv16si:
13146 IID = Intrinsic::x86_avx512_mask_scatter_dpi_512;
13147 break;
13148 case X86::BI__builtin_ia32_scatterdiv8di:
13149 IID = Intrinsic::x86_avx512_mask_scatter_qpq_512;
13150 break;
13151 case X86::BI__builtin_ia32_scatterdiv16si:
13152 IID = Intrinsic::x86_avx512_mask_scatter_qpi_512;
13153 break;
13154 case X86::BI__builtin_ia32_scatterdiv2df:
13155 IID = Intrinsic::x86_avx512_mask_scatterdiv2_df;
13156 break;
13157 case X86::BI__builtin_ia32_scatterdiv2di:
13158 IID = Intrinsic::x86_avx512_mask_scatterdiv2_di;
13159 break;
13160 case X86::BI__builtin_ia32_scatterdiv4df:
13161 IID = Intrinsic::x86_avx512_mask_scatterdiv4_df;
13162 break;
13163 case X86::BI__builtin_ia32_scatterdiv4di:
13164 IID = Intrinsic::x86_avx512_mask_scatterdiv4_di;
13165 break;
13166 case X86::BI__builtin_ia32_scatterdiv4sf:
13167 IID = Intrinsic::x86_avx512_mask_scatterdiv4_sf;
13168 break;
13169 case X86::BI__builtin_ia32_scatterdiv4si:
13170 IID = Intrinsic::x86_avx512_mask_scatterdiv4_si;
13171 break;
13172 case X86::BI__builtin_ia32_scatterdiv8sf:
13173 IID = Intrinsic::x86_avx512_mask_scatterdiv8_sf;
13174 break;
13175 case X86::BI__builtin_ia32_scatterdiv8si:
13176 IID = Intrinsic::x86_avx512_mask_scatterdiv8_si;
13177 break;
13178 case X86::BI__builtin_ia32_scattersiv2df:
13179 IID = Intrinsic::x86_avx512_mask_scattersiv2_df;
13180 break;
13181 case X86::BI__builtin_ia32_scattersiv2di:
13182 IID = Intrinsic::x86_avx512_mask_scattersiv2_di;
13183 break;
13184 case X86::BI__builtin_ia32_scattersiv4df:
13185 IID = Intrinsic::x86_avx512_mask_scattersiv4_df;
13186 break;
13187 case X86::BI__builtin_ia32_scattersiv4di:
13188 IID = Intrinsic::x86_avx512_mask_scattersiv4_di;
13189 break;
13190 case X86::BI__builtin_ia32_scattersiv4sf:
13191 IID = Intrinsic::x86_avx512_mask_scattersiv4_sf;
13192 break;
13193 case X86::BI__builtin_ia32_scattersiv4si:
13194 IID = Intrinsic::x86_avx512_mask_scattersiv4_si;
13195 break;
13196 case X86::BI__builtin_ia32_scattersiv8sf:
13197 IID = Intrinsic::x86_avx512_mask_scattersiv8_sf;
13198 break;
13199 case X86::BI__builtin_ia32_scattersiv8si:
13200 IID = Intrinsic::x86_avx512_mask_scattersiv8_si;
13201 break;
13202 }
13203
13204 unsigned MinElts = std::min(
13205 cast<llvm::FixedVectorType>(Ops[2]->getType())->getNumElements(),
13206 cast<llvm::FixedVectorType>(Ops[3]->getType())->getNumElements());
13207 Ops[1] = getMaskVecValue(*this, Ops[1], MinElts);
13208 Function *Intr = CGM.getIntrinsic(IID);
13209 return Builder.CreateCall(Intr, Ops);
13210 }
13211
13212 case X86::BI__builtin_ia32_vextractf128_pd256:
13213 case X86::BI__builtin_ia32_vextractf128_ps256:
13214 case X86::BI__builtin_ia32_vextractf128_si256:
13215 case X86::BI__builtin_ia32_extract128i256:
13216 case X86::BI__builtin_ia32_extractf64x4_mask:
13217 case X86::BI__builtin_ia32_extractf32x4_mask:
13218 case X86::BI__builtin_ia32_extracti64x4_mask:
13219 case X86::BI__builtin_ia32_extracti32x4_mask:
13220 case X86::BI__builtin_ia32_extractf32x8_mask:
13221 case X86::BI__builtin_ia32_extracti32x8_mask:
13222 case X86::BI__builtin_ia32_extractf32x4_256_mask:
13223 case X86::BI__builtin_ia32_extracti32x4_256_mask:
13224 case X86::BI__builtin_ia32_extractf64x2_256_mask:
13225 case X86::BI__builtin_ia32_extracti64x2_256_mask:
13226 case X86::BI__builtin_ia32_extractf64x2_512_mask:
13227 case X86::BI__builtin_ia32_extracti64x2_512_mask: {
13228 auto *DstTy = cast<llvm::FixedVectorType>(ConvertType(E->getType()));
13229 unsigned NumElts = DstTy->getNumElements();
13230 unsigned SrcNumElts =
13231 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
13232 unsigned SubVectors = SrcNumElts / NumElts;
13233 unsigned Index = cast<ConstantInt>(Ops[1])->getZExtValue();
13234 assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors")(static_cast <bool> (llvm::isPowerOf2_32(SubVectors) &&
"Expected power of 2 subvectors") ? void (0) : __assert_fail
("llvm::isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 13234, __extension__ __PRETTY_FUNCTION__))
;
13235 Index &= SubVectors - 1; // Remove any extra bits.
13236 Index *= NumElts;
13237
13238 int Indices[16];
13239 for (unsigned i = 0; i != NumElts; ++i)
13240 Indices[i] = i + Index;
13241
13242 Value *Res = Builder.CreateShuffleVector(Ops[0],
13243 makeArrayRef(Indices, NumElts),
13244 "extract");
13245
13246 if (Ops.size() == 4)
13247 Res = EmitX86Select(*this, Ops[3], Res, Ops[2]);
13248
13249 return Res;
13250 }
13251 case X86::BI__builtin_ia32_vinsertf128_pd256:
13252 case X86::BI__builtin_ia32_vinsertf128_ps256:
13253 case X86::BI__builtin_ia32_vinsertf128_si256:
13254 case X86::BI__builtin_ia32_insert128i256:
13255 case X86::BI__builtin_ia32_insertf64x4:
13256 case X86::BI__builtin_ia32_insertf32x4:
13257 case X86::BI__builtin_ia32_inserti64x4:
13258 case X86::BI__builtin_ia32_inserti32x4:
13259 case X86::BI__builtin_ia32_insertf32x8:
13260 case X86::BI__builtin_ia32_inserti32x8:
13261 case X86::BI__builtin_ia32_insertf32x4_256:
13262 case X86::BI__builtin_ia32_inserti32x4_256:
13263 case X86::BI__builtin_ia32_insertf64x2_256:
13264 case X86::BI__builtin_ia32_inserti64x2_256:
13265 case X86::BI__builtin_ia32_insertf64x2_512:
13266 case X86::BI__builtin_ia32_inserti64x2_512: {
13267 unsigned DstNumElts =
13268 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
13269 unsigned SrcNumElts =
13270 cast<llvm::FixedVectorType>(Ops[1]->getType())->getNumElements();
13271 unsigned SubVectors = DstNumElts / SrcNumElts;
13272 unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue();
13273 assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors")(static_cast <bool> (llvm::isPowerOf2_32(SubVectors) &&
"Expected power of 2 subvectors") ? void (0) : __assert_fail
("llvm::isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 13273, __extension__ __PRETTY_FUNCTION__))
;
13274 Index &= SubVectors - 1; // Remove any extra bits.
13275 Index *= SrcNumElts;
13276
13277 int Indices[16];
13278 for (unsigned i = 0; i != DstNumElts; ++i)
13279 Indices[i] = (i >= SrcNumElts) ? SrcNumElts + (i % SrcNumElts) : i;
13280
13281 Value *Op1 = Builder.CreateShuffleVector(Ops[1],
13282 makeArrayRef(Indices, DstNumElts),
13283 "widen");
13284
13285 for (unsigned i = 0; i != DstNumElts; ++i) {
13286 if (i >= Index && i < (Index + SrcNumElts))
13287 Indices[i] = (i - Index) + DstNumElts;
13288 else
13289 Indices[i] = i;
13290 }
13291
13292 return Builder.CreateShuffleVector(Ops[0], Op1,
13293 makeArrayRef(Indices, DstNumElts),
13294 "insert");
13295 }
13296 case X86::BI__builtin_ia32_pmovqd512_mask:
13297 case X86::BI__builtin_ia32_pmovwb512_mask: {
13298 Value *Res = Builder.CreateTrunc(Ops[0], Ops[1]->getType());
13299 return EmitX86Select(*this, Ops[2], Res, Ops[1]);
13300 }
13301 case X86::BI__builtin_ia32_pmovdb512_mask:
13302 case X86::BI__builtin_ia32_pmovdw512_mask:
13303 case X86::BI__builtin_ia32_pmovqw512_mask: {
13304 if (const auto *C = dyn_cast<Constant>(Ops[2]))
13305 if (C->isAllOnesValue())
13306 return Builder.CreateTrunc(Ops[0], Ops[1]->getType());
13307
13308 Intrinsic::ID IID;
13309 switch (BuiltinID) {
13310 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 13310)
;
13311 case X86::BI__builtin_ia32_pmovdb512_mask:
13312 IID = Intrinsic::x86_avx512_mask_pmov_db_512;
13313 break;
13314 case X86::BI__builtin_ia32_pmovdw512_mask:
13315 IID = Intrinsic::x86_avx512_mask_pmov_dw_512;
13316 break;
13317 case X86::BI__builtin_ia32_pmovqw512_mask:
13318 IID = Intrinsic::x86_avx512_mask_pmov_qw_512;
13319 break;
13320 }
13321
13322 Function *Intr = CGM.getIntrinsic(IID);
13323 return Builder.CreateCall(Intr, Ops);
13324 }
13325 case X86::BI__builtin_ia32_pblendw128:
13326 case X86::BI__builtin_ia32_blendpd:
13327 case X86::BI__builtin_ia32_blendps:
13328 case X86::BI__builtin_ia32_blendpd256:
13329 case X86::BI__builtin_ia32_blendps256:
13330 case X86::BI__builtin_ia32_pblendw256:
13331 case X86::BI__builtin_ia32_pblendd128:
13332 case X86::BI__builtin_ia32_pblendd256: {
13333 unsigned NumElts =
13334 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
13335 unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
13336
13337 int Indices[16];
13338 // If there are more than 8 elements, the immediate is used twice so make
13339 // sure we handle that.
13340 for (unsigned i = 0; i != NumElts; ++i)
13341 Indices[i] = ((Imm >> (i % 8)) & 0x1) ? NumElts + i : i;
13342
13343 return Builder.CreateShuffleVector(Ops[0], Ops[1],
13344 makeArrayRef(Indices, NumElts),
13345 "blend");
13346 }
13347 case X86::BI__builtin_ia32_pshuflw:
13348 case X86::BI__builtin_ia32_pshuflw256:
13349 case X86::BI__builtin_ia32_pshuflw512: {
13350 uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
13351 auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType());
13352 unsigned NumElts = Ty->getNumElements();
13353
13354 // Splat the 8-bits of immediate 4 times to help the loop wrap around.
13355 Imm = (Imm & 0xff) * 0x01010101;
13356
13357 int Indices[32];
13358 for (unsigned l = 0; l != NumElts; l += 8) {
13359 for (unsigned i = 0; i != 4; ++i) {
13360 Indices[l + i] = l + (Imm & 3);
13361 Imm >>= 2;
13362 }
13363 for (unsigned i = 4; i != 8; ++i)
13364 Indices[l + i] = l + i;
13365 }
13366
13367 return Builder.CreateShuffleVector(Ops[0], makeArrayRef(Indices, NumElts),
13368 "pshuflw");
13369 }
13370 case X86::BI__builtin_ia32_pshufhw:
13371 case X86::BI__builtin_ia32_pshufhw256:
13372 case X86::BI__builtin_ia32_pshufhw512: {
13373 uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
13374 auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType());
13375 unsigned NumElts = Ty->getNumElements();
13376
13377 // Splat the 8-bits of immediate 4 times to help the loop wrap around.
13378 Imm = (Imm & 0xff) * 0x01010101;
13379
13380 int Indices[32];
13381 for (unsigned l = 0; l != NumElts; l += 8) {
13382 for (unsigned i = 0; i != 4; ++i)
13383 Indices[l + i] = l + i;
13384 for (unsigned i = 4; i != 8; ++i) {
13385 Indices[l + i] = l + 4 + (Imm & 3);
13386 Imm >>= 2;
13387 }
13388 }
13389
13390 return Builder.CreateShuffleVector(Ops[0], makeArrayRef(Indices, NumElts),
13391 "pshufhw");
13392 }
13393 case X86::BI__builtin_ia32_pshufd:
13394 case X86::BI__builtin_ia32_pshufd256:
13395 case X86::BI__builtin_ia32_pshufd512:
13396 case X86::BI__builtin_ia32_vpermilpd:
13397 case X86::BI__builtin_ia32_vpermilps:
13398 case X86::BI__builtin_ia32_vpermilpd256:
13399 case X86::BI__builtin_ia32_vpermilps256:
13400 case X86::BI__builtin_ia32_vpermilpd512:
13401 case X86::BI__builtin_ia32_vpermilps512: {
13402 uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
13403 auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType());
13404 unsigned NumElts = Ty->getNumElements();
13405 unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128;
13406 unsigned NumLaneElts = NumElts / NumLanes;
13407
13408 // Splat the 8-bits of immediate 4 times to help the loop wrap around.
13409 Imm = (Imm & 0xff) * 0x01010101;
13410
13411 int Indices[16];
13412 for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
13413 for (unsigned i = 0; i != NumLaneElts; ++i) {
13414 Indices[i + l] = (Imm % NumLaneElts) + l;
13415 Imm /= NumLaneElts;
13416 }
13417 }
13418
13419 return Builder.CreateShuffleVector(Ops[0], makeArrayRef(Indices, NumElts),
13420 "permil");
13421 }
13422 case X86::BI__builtin_ia32_shufpd:
13423 case X86::BI__builtin_ia32_shufpd256:
13424 case X86::BI__builtin_ia32_shufpd512:
13425 case X86::BI__builtin_ia32_shufps:
13426 case X86::BI__builtin_ia32_shufps256:
13427 case X86::BI__builtin_ia32_shufps512: {
13428 uint32_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
13429 auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType());
13430 unsigned NumElts = Ty->getNumElements();
13431 unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128;
13432 unsigned NumLaneElts = NumElts / NumLanes;
13433
13434 // Splat the 8-bits of immediate 4 times to help the loop wrap around.
13435 Imm = (Imm & 0xff) * 0x01010101;
13436
13437 int Indices[16];
13438 for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
13439 for (unsigned i = 0; i != NumLaneElts; ++i) {
13440 unsigned Index = Imm % NumLaneElts;
13441 Imm /= NumLaneElts;
13442 if (i >= (NumLaneElts / 2))
13443 Index += NumElts;
13444 Indices[l + i] = l + Index;
13445 }
13446 }
13447
13448 return Builder.CreateShuffleVector(Ops[0], Ops[1],
13449 makeArrayRef(Indices, NumElts),
13450 "shufp");
13451 }
13452 case X86::BI__builtin_ia32_permdi256:
13453 case X86::BI__builtin_ia32_permdf256:
13454 case X86::BI__builtin_ia32_permdi512:
13455 case X86::BI__builtin_ia32_permdf512: {
13456 unsigned Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
13457 auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType());
13458 unsigned NumElts = Ty->getNumElements();
13459
13460 // These intrinsics operate on 256-bit lanes of four 64-bit elements.
13461 int Indices[8];
13462 for (unsigned l = 0; l != NumElts; l += 4)
13463 for (unsigned i = 0; i != 4; ++i)
13464 Indices[l + i] = l + ((Imm >> (2 * i)) & 0x3);
13465
13466 return Builder.CreateShuffleVector(Ops[0], makeArrayRef(Indices, NumElts),
13467 "perm");
13468 }
13469 case X86::BI__builtin_ia32_palignr128:
13470 case X86::BI__builtin_ia32_palignr256:
13471 case X86::BI__builtin_ia32_palignr512: {
13472 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff;
13473
13474 unsigned NumElts =
13475 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
13476 assert(NumElts % 16 == 0)(static_cast <bool> (NumElts % 16 == 0) ? void (0) : __assert_fail
("NumElts % 16 == 0", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 13476, __extension__ __PRETTY_FUNCTION__))
;
13477
13478 // If palignr is shifting the pair of vectors more than the size of two
13479 // lanes, emit zero.
13480 if (ShiftVal >= 32)
13481 return llvm::Constant::getNullValue(ConvertType(E->getType()));
13482
13483 // If palignr is shifting the pair of input vectors more than one lane,
13484 // but less than two lanes, convert to shifting in zeroes.
13485 if (ShiftVal > 16) {
13486 ShiftVal -= 16;
13487 Ops[1] = Ops[0];
13488 Ops[0] = llvm::Constant::getNullValue(Ops[0]->getType());
13489 }
13490
13491 int Indices[64];
13492 // 256-bit palignr operates on 128-bit lanes so we need to handle that
13493 for (unsigned l = 0; l != NumElts; l += 16) {
13494 for (unsigned i = 0; i != 16; ++i) {
13495 unsigned Idx = ShiftVal + i;
13496 if (Idx >= 16)
13497 Idx += NumElts - 16; // End of lane, switch operand.
13498 Indices[l + i] = Idx + l;
13499 }
13500 }
13501
13502 return Builder.CreateShuffleVector(Ops[1], Ops[0],
13503 makeArrayRef(Indices, NumElts),
13504 "palignr");
13505 }
13506 case X86::BI__builtin_ia32_alignd128:
13507 case X86::BI__builtin_ia32_alignd256:
13508 case X86::BI__builtin_ia32_alignd512:
13509 case X86::BI__builtin_ia32_alignq128:
13510 case X86::BI__builtin_ia32_alignq256:
13511 case X86::BI__builtin_ia32_alignq512: {
13512 unsigned NumElts =
13513 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
13514 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff;
13515
13516 // Mask the shift amount to width of a vector.
13517 ShiftVal &= NumElts - 1;
13518
13519 int Indices[16];
13520 for (unsigned i = 0; i != NumElts; ++i)
13521 Indices[i] = i + ShiftVal;
13522
13523 return Builder.CreateShuffleVector(Ops[1], Ops[0],
13524 makeArrayRef(Indices, NumElts),
13525 "valign");
13526 }
13527 case X86::BI__builtin_ia32_shuf_f32x4_256:
13528 case X86::BI__builtin_ia32_shuf_f64x2_256:
13529 case X86::BI__builtin_ia32_shuf_i32x4_256:
13530 case X86::BI__builtin_ia32_shuf_i64x2_256:
13531 case X86::BI__builtin_ia32_shuf_f32x4:
13532 case X86::BI__builtin_ia32_shuf_f64x2:
13533 case X86::BI__builtin_ia32_shuf_i32x4:
13534 case X86::BI__builtin_ia32_shuf_i64x2: {
13535 unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
13536 auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType());
13537 unsigned NumElts = Ty->getNumElements();
13538 unsigned NumLanes = Ty->getPrimitiveSizeInBits() == 512 ? 4 : 2;
13539 unsigned NumLaneElts = NumElts / NumLanes;
13540
13541 int Indices[16];
13542 for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
13543 unsigned Index = (Imm % NumLanes) * NumLaneElts;
13544 Imm /= NumLanes; // Discard the bits we just used.
13545 if (l >= (NumElts / 2))
13546 Index += NumElts; // Switch to other source.
13547 for (unsigned i = 0; i != NumLaneElts; ++i) {
13548 Indices[l + i] = Index + i;
13549 }
13550 }
13551
13552 return Builder.CreateShuffleVector(Ops[0], Ops[1],
13553 makeArrayRef(Indices, NumElts),
13554 "shuf");
13555 }
13556
13557 case X86::BI__builtin_ia32_vperm2f128_pd256:
13558 case X86::BI__builtin_ia32_vperm2f128_ps256:
13559 case X86::BI__builtin_ia32_vperm2f128_si256:
13560 case X86::BI__builtin_ia32_permti256: {
13561 unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
13562 unsigned NumElts =
13563 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
13564
13565 // This takes a very simple approach since there are two lanes and a
13566 // shuffle can have 2 inputs. So we reserve the first input for the first
13567 // lane and the second input for the second lane. This may result in
13568 // duplicate sources, but this can be dealt with in the backend.
13569
13570 Value *OutOps[2];
13571 int Indices[8];
13572 for (unsigned l = 0; l != 2; ++l) {
13573 // Determine the source for this lane.
13574 if (Imm & (1 << ((l * 4) + 3)))
13575 OutOps[l] = llvm::ConstantAggregateZero::get(Ops[0]->getType());
13576 else if (Imm & (1 << ((l * 4) + 1)))
13577 OutOps[l] = Ops[1];
13578 else
13579 OutOps[l] = Ops[0];
13580
13581 for (unsigned i = 0; i != NumElts/2; ++i) {
13582 // Start with ith element of the source for this lane.
13583 unsigned Idx = (l * NumElts) + i;
13584 // If bit 0 of the immediate half is set, switch to the high half of
13585 // the source.
13586 if (Imm & (1 << (l * 4)))
13587 Idx += NumElts/2;
13588 Indices[(l * (NumElts/2)) + i] = Idx;
13589 }
13590 }
13591
13592 return Builder.CreateShuffleVector(OutOps[0], OutOps[1],
13593 makeArrayRef(Indices, NumElts),
13594 "vperm");
13595 }
13596
13597 case X86::BI__builtin_ia32_pslldqi128_byteshift:
13598 case X86::BI__builtin_ia32_pslldqi256_byteshift:
13599 case X86::BI__builtin_ia32_pslldqi512_byteshift: {
13600 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
13601 auto *ResultType = cast<llvm::FixedVectorType>(Ops[0]->getType());
13602 // Builtin type is vXi64 so multiply by 8 to get bytes.
13603 unsigned NumElts = ResultType->getNumElements() * 8;
13604
13605 // If pslldq is shifting the vector more than 15 bytes, emit zero.
13606 if (ShiftVal >= 16)
13607 return llvm::Constant::getNullValue(ResultType);
13608
13609 int Indices[64];
13610 // 256/512-bit pslldq operates on 128-bit lanes so we need to handle that
13611 for (unsigned l = 0; l != NumElts; l += 16) {
13612 for (unsigned i = 0; i != 16; ++i) {
13613 unsigned Idx = NumElts + i - ShiftVal;
13614 if (Idx < NumElts) Idx -= NumElts - 16; // end of lane, switch operand.
13615 Indices[l + i] = Idx + l;
13616 }
13617 }
13618
13619 auto *VecTy = llvm::FixedVectorType::get(Int8Ty, NumElts);
13620 Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast");
13621 Value *Zero = llvm::Constant::getNullValue(VecTy);
13622 Value *SV = Builder.CreateShuffleVector(Zero, Cast,
13623 makeArrayRef(Indices, NumElts),
13624 "pslldq");
13625 return Builder.CreateBitCast(SV, Ops[0]->getType(), "cast");
13626 }
13627 case X86::BI__builtin_ia32_psrldqi128_byteshift:
13628 case X86::BI__builtin_ia32_psrldqi256_byteshift:
13629 case X86::BI__builtin_ia32_psrldqi512_byteshift: {
13630 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
13631 auto *ResultType = cast<llvm::FixedVectorType>(Ops[0]->getType());
13632 // Builtin type is vXi64 so multiply by 8 to get bytes.
13633 unsigned NumElts = ResultType->getNumElements() * 8;
13634
13635 // If psrldq is shifting the vector more than 15 bytes, emit zero.
13636 if (ShiftVal >= 16)
13637 return llvm::Constant::getNullValue(ResultType);
13638
13639 int Indices[64];
13640 // 256/512-bit psrldq operates on 128-bit lanes so we need to handle that
13641 for (unsigned l = 0; l != NumElts; l += 16) {
13642 for (unsigned i = 0; i != 16; ++i) {
13643 unsigned Idx = i + ShiftVal;
13644 if (Idx >= 16) Idx += NumElts - 16; // end of lane, switch operand.
13645 Indices[l + i] = Idx + l;
13646 }
13647 }
13648
13649 auto *VecTy = llvm::FixedVectorType::get(Int8Ty, NumElts);
13650 Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast");
13651 Value *Zero = llvm::Constant::getNullValue(VecTy);
13652 Value *SV = Builder.CreateShuffleVector(Cast, Zero,
13653 makeArrayRef(Indices, NumElts),
13654 "psrldq");
13655 return Builder.CreateBitCast(SV, ResultType, "cast");
13656 }
13657 case X86::BI__builtin_ia32_kshiftliqi:
13658 case X86::BI__builtin_ia32_kshiftlihi:
13659 case X86::BI__builtin_ia32_kshiftlisi:
13660 case X86::BI__builtin_ia32_kshiftlidi: {
13661 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
13662 unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
13663
13664 if (ShiftVal >= NumElts)
13665 return llvm::Constant::getNullValue(Ops[0]->getType());
13666
13667 Value *In = getMaskVecValue(*this, Ops[0], NumElts);
13668
13669 int Indices[64];
13670 for (unsigned i = 0; i != NumElts; ++i)
13671 Indices[i] = NumElts + i - ShiftVal;
13672
13673 Value *Zero = llvm::Constant::getNullValue(In->getType());
13674 Value *SV = Builder.CreateShuffleVector(Zero, In,
13675 makeArrayRef(Indices, NumElts),
13676 "kshiftl");
13677 return Builder.CreateBitCast(SV, Ops[0]->getType());
13678 }
13679 case X86::BI__builtin_ia32_kshiftriqi:
13680 case X86::BI__builtin_ia32_kshiftrihi:
13681 case X86::BI__builtin_ia32_kshiftrisi:
13682 case X86::BI__builtin_ia32_kshiftridi: {
13683 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
13684 unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
13685
13686 if (ShiftVal >= NumElts)
13687 return llvm::Constant::getNullValue(Ops[0]->getType());
13688
13689 Value *In = getMaskVecValue(*this, Ops[0], NumElts);
13690
13691 int Indices[64];
13692 for (unsigned i = 0; i != NumElts; ++i)
13693 Indices[i] = i + ShiftVal;
13694
13695 Value *Zero = llvm::Constant::getNullValue(In->getType());
13696 Value *SV = Builder.CreateShuffleVector(In, Zero,
13697 makeArrayRef(Indices, NumElts),
13698 "kshiftr");
13699 return Builder.CreateBitCast(SV, Ops[0]->getType());
13700 }
13701 case X86::BI__builtin_ia32_movnti:
13702 case X86::BI__builtin_ia32_movnti64:
13703 case X86::BI__builtin_ia32_movntsd:
13704 case X86::BI__builtin_ia32_movntss: {
13705 llvm::MDNode *Node = llvm::MDNode::get(
13706 getLLVMContext(), llvm::ConstantAsMetadata::get(Builder.getInt32(1)));
13707
13708 Value *Ptr = Ops[0];
13709 Value *Src = Ops[1];
13710
13711 // Extract the 0'th element of the source vector.
13712 if (BuiltinID == X86::BI__builtin_ia32_movntsd ||
13713 BuiltinID == X86::BI__builtin_ia32_movntss)
13714 Src = Builder.CreateExtractElement(Src, (uint64_t)0, "extract");
13715
13716 // Convert the type of the pointer to a pointer to the stored type.
13717 Value *BC = Builder.CreateBitCast(
13718 Ptr, llvm::PointerType::getUnqual(Src->getType()), "cast");
13719
13720 // Unaligned nontemporal store of the scalar value.
13721 StoreInst *SI = Builder.CreateDefaultAlignedStore(Src, BC);
13722 SI->setMetadata(CGM.getModule().getMDKindID("nontemporal"), Node);
13723 SI->setAlignment(llvm::Align(1));
13724 return SI;
13725 }
13726 // Rotate is a special case of funnel shift - 1st 2 args are the same.
13727 case X86::BI__builtin_ia32_vprotb:
13728 case X86::BI__builtin_ia32_vprotw:
13729 case X86::BI__builtin_ia32_vprotd:
13730 case X86::BI__builtin_ia32_vprotq:
13731 case X86::BI__builtin_ia32_vprotbi:
13732 case X86::BI__builtin_ia32_vprotwi:
13733 case X86::BI__builtin_ia32_vprotdi:
13734 case X86::BI__builtin_ia32_vprotqi:
13735 case X86::BI__builtin_ia32_prold128:
13736 case X86::BI__builtin_ia32_prold256:
13737 case X86::BI__builtin_ia32_prold512:
13738 case X86::BI__builtin_ia32_prolq128:
13739 case X86::BI__builtin_ia32_prolq256:
13740 case X86::BI__builtin_ia32_prolq512:
13741 case X86::BI__builtin_ia32_prolvd128:
13742 case X86::BI__builtin_ia32_prolvd256:
13743 case X86::BI__builtin_ia32_prolvd512:
13744 case X86::BI__builtin_ia32_prolvq128:
13745 case X86::BI__builtin_ia32_prolvq256:
13746 case X86::BI__builtin_ia32_prolvq512:
13747 return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], false);
13748 case X86::BI__builtin_ia32_prord128:
13749 case X86::BI__builtin_ia32_prord256:
13750 case X86::BI__builtin_ia32_prord512:
13751 case X86::BI__builtin_ia32_prorq128:
13752 case X86::BI__builtin_ia32_prorq256:
13753 case X86::BI__builtin_ia32_prorq512:
13754 case X86::BI__builtin_ia32_prorvd128:
13755 case X86::BI__builtin_ia32_prorvd256:
13756 case X86::BI__builtin_ia32_prorvd512:
13757 case X86::BI__builtin_ia32_prorvq128:
13758 case X86::BI__builtin_ia32_prorvq256:
13759 case X86::BI__builtin_ia32_prorvq512:
13760 return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], true);
13761 case X86::BI__builtin_ia32_selectb_128:
13762 case X86::BI__builtin_ia32_selectb_256:
13763 case X86::BI__builtin_ia32_selectb_512:
13764 case X86::BI__builtin_ia32_selectw_128:
13765 case X86::BI__builtin_ia32_selectw_256:
13766 case X86::BI__builtin_ia32_selectw_512:
13767 case X86::BI__builtin_ia32_selectd_128:
13768 case X86::BI__builtin_ia32_selectd_256:
13769 case X86::BI__builtin_ia32_selectd_512:
13770 case X86::BI__builtin_ia32_selectq_128:
13771 case X86::BI__builtin_ia32_selectq_256:
13772 case X86::BI__builtin_ia32_selectq_512:
13773 case X86::BI__builtin_ia32_selectph_128:
13774 case X86::BI__builtin_ia32_selectph_256:
13775 case X86::BI__builtin_ia32_selectph_512:
13776 case X86::BI__builtin_ia32_selectps_128:
13777 case X86::BI__builtin_ia32_selectps_256:
13778 case X86::BI__builtin_ia32_selectps_512:
13779 case X86::BI__builtin_ia32_selectpd_128:
13780 case X86::BI__builtin_ia32_selectpd_256:
13781 case X86::BI__builtin_ia32_selectpd_512:
13782 return EmitX86Select(*this, Ops[0], Ops[1], Ops[2]);
13783 case X86::BI__builtin_ia32_selectsh_128:
13784 case X86::BI__builtin_ia32_selectss_128:
13785 case X86::BI__builtin_ia32_selectsd_128: {
13786 Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0);
13787 Value *B = Builder.CreateExtractElement(Ops[2], (uint64_t)0);
13788 A = EmitX86ScalarSelect(*this, Ops[0], A, B);
13789 return Builder.CreateInsertElement(Ops[1], A, (uint64_t)0);
13790 }
13791 case X86::BI__builtin_ia32_cmpb128_mask:
13792 case X86::BI__builtin_ia32_cmpb256_mask:
13793 case X86::BI__builtin_ia32_cmpb512_mask:
13794 case X86::BI__builtin_ia32_cmpw128_mask:
13795 case X86::BI__builtin_ia32_cmpw256_mask:
13796 case X86::BI__builtin_ia32_cmpw512_mask:
13797 case X86::BI__builtin_ia32_cmpd128_mask:
13798 case X86::BI__builtin_ia32_cmpd256_mask:
13799 case X86::BI__builtin_ia32_cmpd512_mask:
13800 case X86::BI__builtin_ia32_cmpq128_mask:
13801 case X86::BI__builtin_ia32_cmpq256_mask:
13802 case X86::BI__builtin_ia32_cmpq512_mask: {
13803 unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
13804 return EmitX86MaskedCompare(*this, CC, true, Ops);
13805 }
13806 case X86::BI__builtin_ia32_ucmpb128_mask:
13807 case X86::BI__builtin_ia32_ucmpb256_mask:
13808 case X86::BI__builtin_ia32_ucmpb512_mask:
13809 case X86::BI__builtin_ia32_ucmpw128_mask:
13810 case X86::BI__builtin_ia32_ucmpw256_mask:
13811 case X86::BI__builtin_ia32_ucmpw512_mask:
13812 case X86::BI__builtin_ia32_ucmpd128_mask:
13813 case X86::BI__builtin_ia32_ucmpd256_mask:
13814 case X86::BI__builtin_ia32_ucmpd512_mask:
13815 case X86::BI__builtin_ia32_ucmpq128_mask:
13816 case X86::BI__builtin_ia32_ucmpq256_mask:
13817 case X86::BI__builtin_ia32_ucmpq512_mask: {
13818 unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
13819 return EmitX86MaskedCompare(*this, CC, false, Ops);
13820 }
13821 case X86::BI__builtin_ia32_vpcomb:
13822 case X86::BI__builtin_ia32_vpcomw:
13823 case X86::BI__builtin_ia32_vpcomd:
13824 case X86::BI__builtin_ia32_vpcomq:
13825 return EmitX86vpcom(*this, Ops, true);
13826 case X86::BI__builtin_ia32_vpcomub:
13827 case X86::BI__builtin_ia32_vpcomuw:
13828 case X86::BI__builtin_ia32_vpcomud:
13829 case X86::BI__builtin_ia32_vpcomuq:
13830 return EmitX86vpcom(*this, Ops, false);
13831
13832 case X86::BI__builtin_ia32_kortestcqi:
13833 case X86::BI__builtin_ia32_kortestchi:
13834 case X86::BI__builtin_ia32_kortestcsi:
13835 case X86::BI__builtin_ia32_kortestcdi: {
13836 Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops);
13837 Value *C = llvm::Constant::getAllOnesValue(Ops[0]->getType());
13838 Value *Cmp = Builder.CreateICmpEQ(Or, C);
13839 return Builder.CreateZExt(Cmp, ConvertType(E->getType()));
13840 }
13841 case X86::BI__builtin_ia32_kortestzqi:
13842 case X86::BI__builtin_ia32_kortestzhi:
13843 case X86::BI__builtin_ia32_kortestzsi:
13844 case X86::BI__builtin_ia32_kortestzdi: {
13845 Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops);
13846 Value *C = llvm::Constant::getNullValue(Ops[0]->getType());
13847 Value *Cmp = Builder.CreateICmpEQ(Or, C);
13848 return Builder.CreateZExt(Cmp, ConvertType(E->getType()));
13849 }
13850
13851 case X86::BI__builtin_ia32_ktestcqi:
13852 case X86::BI__builtin_ia32_ktestzqi:
13853 case X86::BI__builtin_ia32_ktestchi:
13854 case X86::BI__builtin_ia32_ktestzhi:
13855 case X86::BI__builtin_ia32_ktestcsi:
13856 case X86::BI__builtin_ia32_ktestzsi:
13857 case X86::BI__builtin_ia32_ktestcdi:
13858 case X86::BI__builtin_ia32_ktestzdi: {
13859 Intrinsic::ID IID;
13860 switch (BuiltinID) {
13861 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 13861)
;
13862 case X86::BI__builtin_ia32_ktestcqi:
13863 IID = Intrinsic::x86_avx512_ktestc_b;
13864 break;
13865 case X86::BI__builtin_ia32_ktestzqi:
13866 IID = Intrinsic::x86_avx512_ktestz_b;
13867 break;
13868 case X86::BI__builtin_ia32_ktestchi:
13869 IID = Intrinsic::x86_avx512_ktestc_w;
13870 break;
13871 case X86::BI__builtin_ia32_ktestzhi:
13872 IID = Intrinsic::x86_avx512_ktestz_w;
13873 break;
13874 case X86::BI__builtin_ia32_ktestcsi:
13875 IID = Intrinsic::x86_avx512_ktestc_d;
13876 break;
13877 case X86::BI__builtin_ia32_ktestzsi:
13878 IID = Intrinsic::x86_avx512_ktestz_d;
13879 break;
13880 case X86::BI__builtin_ia32_ktestcdi:
13881 IID = Intrinsic::x86_avx512_ktestc_q;
13882 break;
13883 case X86::BI__builtin_ia32_ktestzdi:
13884 IID = Intrinsic::x86_avx512_ktestz_q;
13885 break;
13886 }
13887
13888 unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
13889 Value *LHS = getMaskVecValue(*this, Ops[0], NumElts);
13890 Value *RHS = getMaskVecValue(*this, Ops[1], NumElts);
13891 Function *Intr = CGM.getIntrinsic(IID);
13892 return Builder.CreateCall(Intr, {LHS, RHS});
13893 }
13894
13895 case X86::BI__builtin_ia32_kaddqi:
13896 case X86::BI__builtin_ia32_kaddhi:
13897 case X86::BI__builtin_ia32_kaddsi:
13898 case X86::BI__builtin_ia32_kadddi: {
13899 Intrinsic::ID IID;
13900 switch (BuiltinID) {
13901 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 13901)
;
13902 case X86::BI__builtin_ia32_kaddqi:
13903 IID = Intrinsic::x86_avx512_kadd_b;
13904 break;
13905 case X86::BI__builtin_ia32_kaddhi:
13906 IID = Intrinsic::x86_avx512_kadd_w;
13907 break;
13908 case X86::BI__builtin_ia32_kaddsi:
13909 IID = Intrinsic::x86_avx512_kadd_d;
13910 break;
13911 case X86::BI__builtin_ia32_kadddi:
13912 IID = Intrinsic::x86_avx512_kadd_q;
13913 break;
13914 }
13915
13916 unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
13917 Value *LHS = getMaskVecValue(*this, Ops[0], NumElts);
13918 Value *RHS = getMaskVecValue(*this, Ops[1], NumElts);
13919 Function *Intr = CGM.getIntrinsic(IID);
13920 Value *Res = Builder.CreateCall(Intr, {LHS, RHS});
13921 return Builder.CreateBitCast(Res, Ops[0]->getType());
13922 }
13923 case X86::BI__builtin_ia32_kandqi:
13924 case X86::BI__builtin_ia32_kandhi:
13925 case X86::BI__builtin_ia32_kandsi:
13926 case X86::BI__builtin_ia32_kanddi:
13927 return EmitX86MaskLogic(*this, Instruction::And, Ops);
13928 case X86::BI__builtin_ia32_kandnqi:
13929 case X86::BI__builtin_ia32_kandnhi:
13930 case X86::BI__builtin_ia32_kandnsi:
13931 case X86::BI__builtin_ia32_kandndi:
13932 return EmitX86MaskLogic(*this, Instruction::And, Ops, true);
13933 case X86::BI__builtin_ia32_korqi:
13934 case X86::BI__builtin_ia32_korhi:
13935 case X86::BI__builtin_ia32_korsi:
13936 case X86::BI__builtin_ia32_kordi:
13937 return EmitX86MaskLogic(*this, Instruction::Or, Ops);
13938 case X86::BI__builtin_ia32_kxnorqi:
13939 case X86::BI__builtin_ia32_kxnorhi:
13940 case X86::BI__builtin_ia32_kxnorsi:
13941 case X86::BI__builtin_ia32_kxnordi:
13942 return EmitX86MaskLogic(*this, Instruction::Xor, Ops, true);
13943 case X86::BI__builtin_ia32_kxorqi:
13944 case X86::BI__builtin_ia32_kxorhi:
13945 case X86::BI__builtin_ia32_kxorsi:
13946 case X86::BI__builtin_ia32_kxordi:
13947 return EmitX86MaskLogic(*this, Instruction::Xor, Ops);
13948 case X86::BI__builtin_ia32_knotqi:
13949 case X86::BI__builtin_ia32_knothi:
13950 case X86::BI__builtin_ia32_knotsi:
13951 case X86::BI__builtin_ia32_knotdi: {
13952 unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
13953 Value *Res = getMaskVecValue(*this, Ops[0], NumElts);
13954 return Builder.CreateBitCast(Builder.CreateNot(Res),
13955 Ops[0]->getType());
13956 }
13957 case X86::BI__builtin_ia32_kmovb:
13958 case X86::BI__builtin_ia32_kmovw:
13959 case X86::BI__builtin_ia32_kmovd:
13960 case X86::BI__builtin_ia32_kmovq: {
13961 // Bitcast to vXi1 type and then back to integer. This gets the mask
13962 // register type into the IR, but might be optimized out depending on
13963 // what's around it.
13964 unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
13965 Value *Res = getMaskVecValue(*this, Ops[0], NumElts);
13966 return Builder.CreateBitCast(Res, Ops[0]->getType());
13967 }
13968
13969 case X86::BI__builtin_ia32_kunpckdi:
13970 case X86::BI__builtin_ia32_kunpcksi:
13971 case X86::BI__builtin_ia32_kunpckhi: {
13972 unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
13973 Value *LHS = getMaskVecValue(*this, Ops[0], NumElts);
13974 Value *RHS = getMaskVecValue(*this, Ops[1], NumElts);
13975 int Indices[64];
13976 for (unsigned i = 0; i != NumElts; ++i)
13977 Indices[i] = i;
13978
13979 // First extract half of each vector. This gives better codegen than
13980 // doing it in a single shuffle.
13981 LHS = Builder.CreateShuffleVector(LHS, LHS,
13982 makeArrayRef(Indices, NumElts / 2));
13983 RHS = Builder.CreateShuffleVector(RHS, RHS,
13984 makeArrayRef(Indices, NumElts / 2));
13985 // Concat the vectors.
13986 // NOTE: Operands are swapped to match the intrinsic definition.
13987 Value *Res = Builder.CreateShuffleVector(RHS, LHS,
13988 makeArrayRef(Indices, NumElts));
13989 return Builder.CreateBitCast(Res, Ops[0]->getType());
13990 }
13991
13992 case X86::BI__builtin_ia32_vplzcntd_128:
13993 case X86::BI__builtin_ia32_vplzcntd_256:
13994 case X86::BI__builtin_ia32_vplzcntd_512:
13995 case X86::BI__builtin_ia32_vplzcntq_128:
13996 case X86::BI__builtin_ia32_vplzcntq_256:
13997 case X86::BI__builtin_ia32_vplzcntq_512: {
13998 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType());
13999 return Builder.CreateCall(F, {Ops[0],Builder.getInt1(false)});
14000 }
14001 case X86::BI__builtin_ia32_sqrtss:
14002 case X86::BI__builtin_ia32_sqrtsd: {
14003 Value *A = Builder.CreateExtractElement(Ops[0], (uint64_t)0);
14004 Function *F;
14005 if (Builder.getIsFPConstrained()) {
14006 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
14007 F = CGM.getIntrinsic(Intrinsic::experimental_constrained_sqrt,
14008 A->getType());
14009 A = Builder.CreateConstrainedFPCall(F, {A});
14010 } else {
14011 F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType());
14012 A = Builder.CreateCall(F, {A});
14013 }
14014 return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0);
14015 }
14016 case X86::BI__builtin_ia32_sqrtsh_round_mask:
14017 case X86::BI__builtin_ia32_sqrtsd_round_mask:
14018 case X86::BI__builtin_ia32_sqrtss_round_mask: {
14019 unsigned CC = cast<llvm::ConstantInt>(Ops[4])->getZExtValue();
14020 // Support only if the rounding mode is 4 (AKA CUR_DIRECTION),
14021 // otherwise keep the intrinsic.
14022 if (CC != 4) {
14023 Intrinsic::ID IID;
14024
14025 switch (BuiltinID) {
14026 default:
14027 llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14027)
;
14028 case X86::BI__builtin_ia32_sqrtsh_round_mask:
14029 IID = Intrinsic::x86_avx512fp16_mask_sqrt_sh;
14030 break;
14031 case X86::BI__builtin_ia32_sqrtsd_round_mask:
14032 IID = Intrinsic::x86_avx512_mask_sqrt_sd;
14033 break;
14034 case X86::BI__builtin_ia32_sqrtss_round_mask:
14035 IID = Intrinsic::x86_avx512_mask_sqrt_ss;
14036 break;
14037 }
14038 return Builder.CreateCall(CGM.getIntrinsic(IID), Ops);
14039 }
14040 Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0);
14041 Function *F;
14042 if (Builder.getIsFPConstrained()) {
14043 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
14044 F = CGM.getIntrinsic(Intrinsic::experimental_constrained_sqrt,
14045 A->getType());
14046 A = Builder.CreateConstrainedFPCall(F, A);
14047 } else {
14048 F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType());
14049 A = Builder.CreateCall(F, A);
14050 }
14051 Value *Src = Builder.CreateExtractElement(Ops[2], (uint64_t)0);
14052 A = EmitX86ScalarSelect(*this, Ops[3], A, Src);
14053 return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0);
14054 }
14055 case X86::BI__builtin_ia32_sqrtpd256:
14056 case X86::BI__builtin_ia32_sqrtpd:
14057 case X86::BI__builtin_ia32_sqrtps256:
14058 case X86::BI__builtin_ia32_sqrtps:
14059 case X86::BI__builtin_ia32_sqrtph256:
14060 case X86::BI__builtin_ia32_sqrtph:
14061 case X86::BI__builtin_ia32_sqrtph512:
14062 case X86::BI__builtin_ia32_sqrtps512:
14063 case X86::BI__builtin_ia32_sqrtpd512: {
14064 if (Ops.size() == 2) {
14065 unsigned CC = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
14066 // Support only if the rounding mode is 4 (AKA CUR_DIRECTION),
14067 // otherwise keep the intrinsic.
14068 if (CC != 4) {
14069 Intrinsic::ID IID;
14070
14071 switch (BuiltinID) {
14072 default:
14073 llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14073)
;
14074 case X86::BI__builtin_ia32_sqrtph512:
14075 IID = Intrinsic::x86_avx512fp16_sqrt_ph_512;
14076 break;
14077 case X86::BI__builtin_ia32_sqrtps512:
14078 IID = Intrinsic::x86_avx512_sqrt_ps_512;
14079 break;
14080 case X86::BI__builtin_ia32_sqrtpd512:
14081 IID = Intrinsic::x86_avx512_sqrt_pd_512;
14082 break;
14083 }
14084 return Builder.CreateCall(CGM.getIntrinsic(IID), Ops);
14085 }
14086 }
14087 if (Builder.getIsFPConstrained()) {
14088 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
14089 Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_sqrt,
14090 Ops[0]->getType());
14091 return Builder.CreateConstrainedFPCall(F, Ops[0]);
14092 } else {
14093 Function *F = CGM.getIntrinsic(Intrinsic::sqrt, Ops[0]->getType());
14094 return Builder.CreateCall(F, Ops[0]);
14095 }
14096 }
14097 case X86::BI__builtin_ia32_pabsb128:
14098 case X86::BI__builtin_ia32_pabsw128:
14099 case X86::BI__builtin_ia32_pabsd128:
14100 case X86::BI__builtin_ia32_pabsb256:
14101 case X86::BI__builtin_ia32_pabsw256:
14102 case X86::BI__builtin_ia32_pabsd256:
14103 case X86::BI__builtin_ia32_pabsq128:
14104 case X86::BI__builtin_ia32_pabsq256:
14105 case X86::BI__builtin_ia32_pabsb512:
14106 case X86::BI__builtin_ia32_pabsw512:
14107 case X86::BI__builtin_ia32_pabsd512:
14108 case X86::BI__builtin_ia32_pabsq512: {
14109 Function *F = CGM.getIntrinsic(Intrinsic::abs, Ops[0]->getType());
14110 return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)});
14111 }
14112 case X86::BI__builtin_ia32_pmaxsb128:
14113 case X86::BI__builtin_ia32_pmaxsw128:
14114 case X86::BI__builtin_ia32_pmaxsd128:
14115 case X86::BI__builtin_ia32_pmaxsq128:
14116 case X86::BI__builtin_ia32_pmaxsb256:
14117 case X86::BI__builtin_ia32_pmaxsw256:
14118 case X86::BI__builtin_ia32_pmaxsd256:
14119 case X86::BI__builtin_ia32_pmaxsq256:
14120 case X86::BI__builtin_ia32_pmaxsb512:
14121 case X86::BI__builtin_ia32_pmaxsw512:
14122 case X86::BI__builtin_ia32_pmaxsd512:
14123 case X86::BI__builtin_ia32_pmaxsq512:
14124 return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::smax);
14125 case X86::BI__builtin_ia32_pmaxub128:
14126 case X86::BI__builtin_ia32_pmaxuw128:
14127 case X86::BI__builtin_ia32_pmaxud128:
14128 case X86::BI__builtin_ia32_pmaxuq128:
14129 case X86::BI__builtin_ia32_pmaxub256:
14130 case X86::BI__builtin_ia32_pmaxuw256:
14131 case X86::BI__builtin_ia32_pmaxud256:
14132 case X86::BI__builtin_ia32_pmaxuq256:
14133 case X86::BI__builtin_ia32_pmaxub512:
14134 case X86::BI__builtin_ia32_pmaxuw512:
14135 case X86::BI__builtin_ia32_pmaxud512:
14136 case X86::BI__builtin_ia32_pmaxuq512:
14137 return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::umax);
14138 case X86::BI__builtin_ia32_pminsb128:
14139 case X86::BI__builtin_ia32_pminsw128:
14140 case X86::BI__builtin_ia32_pminsd128:
14141 case X86::BI__builtin_ia32_pminsq128:
14142 case X86::BI__builtin_ia32_pminsb256:
14143 case X86::BI__builtin_ia32_pminsw256:
14144 case X86::BI__builtin_ia32_pminsd256:
14145 case X86::BI__builtin_ia32_pminsq256:
14146 case X86::BI__builtin_ia32_pminsb512:
14147 case X86::BI__builtin_ia32_pminsw512:
14148 case X86::BI__builtin_ia32_pminsd512:
14149 case X86::BI__builtin_ia32_pminsq512:
14150 return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::smin);
14151 case X86::BI__builtin_ia32_pminub128:
14152 case X86::BI__builtin_ia32_pminuw128:
14153 case X86::BI__builtin_ia32_pminud128:
14154 case X86::BI__builtin_ia32_pminuq128:
14155 case X86::BI__builtin_ia32_pminub256:
14156 case X86::BI__builtin_ia32_pminuw256:
14157 case X86::BI__builtin_ia32_pminud256:
14158 case X86::BI__builtin_ia32_pminuq256:
14159 case X86::BI__builtin_ia32_pminub512:
14160 case X86::BI__builtin_ia32_pminuw512:
14161 case X86::BI__builtin_ia32_pminud512:
14162 case X86::BI__builtin_ia32_pminuq512:
14163 return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::umin);
14164
14165 case X86::BI__builtin_ia32_pmuludq128:
14166 case X86::BI__builtin_ia32_pmuludq256:
14167 case X86::BI__builtin_ia32_pmuludq512:
14168 return EmitX86Muldq(*this, /*IsSigned*/false, Ops);
14169
14170 case X86::BI__builtin_ia32_pmuldq128:
14171 case X86::BI__builtin_ia32_pmuldq256:
14172 case X86::BI__builtin_ia32_pmuldq512:
14173 return EmitX86Muldq(*this, /*IsSigned*/true, Ops);
14174
14175 case X86::BI__builtin_ia32_pternlogd512_mask:
14176 case X86::BI__builtin_ia32_pternlogq512_mask:
14177 case X86::BI__builtin_ia32_pternlogd128_mask:
14178 case X86::BI__builtin_ia32_pternlogd256_mask:
14179 case X86::BI__builtin_ia32_pternlogq128_mask:
14180 case X86::BI__builtin_ia32_pternlogq256_mask:
14181 return EmitX86Ternlog(*this, /*ZeroMask*/false, Ops);
14182
14183 case X86::BI__builtin_ia32_pternlogd512_maskz:
14184 case X86::BI__builtin_ia32_pternlogq512_maskz:
14185 case X86::BI__builtin_ia32_pternlogd128_maskz:
14186 case X86::BI__builtin_ia32_pternlogd256_maskz:
14187 case X86::BI__builtin_ia32_pternlogq128_maskz:
14188 case X86::BI__builtin_ia32_pternlogq256_maskz:
14189 return EmitX86Ternlog(*this, /*ZeroMask*/true, Ops);
14190
14191 case X86::BI__builtin_ia32_vpshldd128:
14192 case X86::BI__builtin_ia32_vpshldd256:
14193 case X86::BI__builtin_ia32_vpshldd512:
14194 case X86::BI__builtin_ia32_vpshldq128:
14195 case X86::BI__builtin_ia32_vpshldq256:
14196 case X86::BI__builtin_ia32_vpshldq512:
14197 case X86::BI__builtin_ia32_vpshldw128:
14198 case X86::BI__builtin_ia32_vpshldw256:
14199 case X86::BI__builtin_ia32_vpshldw512:
14200 return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false);
14201
14202 case X86::BI__builtin_ia32_vpshrdd128:
14203 case X86::BI__builtin_ia32_vpshrdd256:
14204 case X86::BI__builtin_ia32_vpshrdd512:
14205 case X86::BI__builtin_ia32_vpshrdq128:
14206 case X86::BI__builtin_ia32_vpshrdq256:
14207 case X86::BI__builtin_ia32_vpshrdq512:
14208 case X86::BI__builtin_ia32_vpshrdw128:
14209 case X86::BI__builtin_ia32_vpshrdw256:
14210 case X86::BI__builtin_ia32_vpshrdw512:
14211 // Ops 0 and 1 are swapped.
14212 return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true);
14213
14214 case X86::BI__builtin_ia32_vpshldvd128:
14215 case X86::BI__builtin_ia32_vpshldvd256:
14216 case X86::BI__builtin_ia32_vpshldvd512:
14217 case X86::BI__builtin_ia32_vpshldvq128:
14218 case X86::BI__builtin_ia32_vpshldvq256:
14219 case X86::BI__builtin_ia32_vpshldvq512:
14220 case X86::BI__builtin_ia32_vpshldvw128:
14221 case X86::BI__builtin_ia32_vpshldvw256:
14222 case X86::BI__builtin_ia32_vpshldvw512:
14223 return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false);
14224
14225 case X86::BI__builtin_ia32_vpshrdvd128:
14226 case X86::BI__builtin_ia32_vpshrdvd256:
14227 case X86::BI__builtin_ia32_vpshrdvd512:
14228 case X86::BI__builtin_ia32_vpshrdvq128:
14229 case X86::BI__builtin_ia32_vpshrdvq256:
14230 case X86::BI__builtin_ia32_vpshrdvq512:
14231 case X86::BI__builtin_ia32_vpshrdvw128:
14232 case X86::BI__builtin_ia32_vpshrdvw256:
14233 case X86::BI__builtin_ia32_vpshrdvw512:
14234 // Ops 0 and 1 are swapped.
14235 return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true);
14236
14237 // Reductions
14238 case X86::BI__builtin_ia32_reduce_add_d512:
14239 case X86::BI__builtin_ia32_reduce_add_q512: {
14240 Function *F =
14241 CGM.getIntrinsic(Intrinsic::vector_reduce_add, Ops[0]->getType());
14242 return Builder.CreateCall(F, {Ops[0]});
14243 }
14244 case X86::BI__builtin_ia32_reduce_and_d512:
14245 case X86::BI__builtin_ia32_reduce_and_q512: {
14246 Function *F =
14247 CGM.getIntrinsic(Intrinsic::vector_reduce_and, Ops[0]->getType());
14248 return Builder.CreateCall(F, {Ops[0]});
14249 }
14250 case X86::BI__builtin_ia32_reduce_fadd_pd512:
14251 case X86::BI__builtin_ia32_reduce_fadd_ps512:
14252 case X86::BI__builtin_ia32_reduce_fadd_ph512:
14253 case X86::BI__builtin_ia32_reduce_fadd_ph256:
14254 case X86::BI__builtin_ia32_reduce_fadd_ph128: {
14255 Function *F =
14256 CGM.getIntrinsic(Intrinsic::vector_reduce_fadd, Ops[1]->getType());
14257 Builder.getFastMathFlags().setAllowReassoc();
14258 return Builder.CreateCall(F, {Ops[0], Ops[1]});
14259 }
14260 case X86::BI__builtin_ia32_reduce_fmul_pd512:
14261 case X86::BI__builtin_ia32_reduce_fmul_ps512:
14262 case X86::BI__builtin_ia32_reduce_fmul_ph512:
14263 case X86::BI__builtin_ia32_reduce_fmul_ph256:
14264 case X86::BI__builtin_ia32_reduce_fmul_ph128: {
14265 Function *F =
14266 CGM.getIntrinsic(Intrinsic::vector_reduce_fmul, Ops[1]->getType());
14267 Builder.getFastMathFlags().setAllowReassoc();
14268 return Builder.CreateCall(F, {Ops[0], Ops[1]});
14269 }
14270 case X86::BI__builtin_ia32_reduce_fmax_pd512:
14271 case X86::BI__builtin_ia32_reduce_fmax_ps512:
14272 case X86::BI__builtin_ia32_reduce_fmax_ph512:
14273 case X86::BI__builtin_ia32_reduce_fmax_ph256:
14274 case X86::BI__builtin_ia32_reduce_fmax_ph128: {
14275 Function *F =
14276 CGM.getIntrinsic(Intrinsic::vector_reduce_fmax, Ops[0]->getType());
14277 Builder.getFastMathFlags().setNoNaNs();
14278 return Builder.CreateCall(F, {Ops[0]});
14279 }
14280 case X86::BI__builtin_ia32_reduce_fmin_pd512:
14281 case X86::BI__builtin_ia32_reduce_fmin_ps512:
14282 case X86::BI__builtin_ia32_reduce_fmin_ph512:
14283 case X86::BI__builtin_ia32_reduce_fmin_ph256:
14284 case X86::BI__builtin_ia32_reduce_fmin_ph128: {
14285 Function *F =
14286 CGM.getIntrinsic(Intrinsic::vector_reduce_fmin, Ops[0]->getType());
14287 Builder.getFastMathFlags().setNoNaNs();
14288 return Builder.CreateCall(F, {Ops[0]});
14289 }
14290 case X86::BI__builtin_ia32_reduce_mul_d512:
14291 case X86::BI__builtin_ia32_reduce_mul_q512: {
14292 Function *F =
14293 CGM.getIntrinsic(Intrinsic::vector_reduce_mul, Ops[0]->getType());
14294 return Builder.CreateCall(F, {Ops[0]});
14295 }
14296 case X86::BI__builtin_ia32_reduce_or_d512:
14297 case X86::BI__builtin_ia32_reduce_or_q512: {
14298 Function *F =
14299 CGM.getIntrinsic(Intrinsic::vector_reduce_or, Ops[0]->getType());
14300 return Builder.CreateCall(F, {Ops[0]});
14301 }
14302 case X86::BI__builtin_ia32_reduce_smax_d512:
14303 case X86::BI__builtin_ia32_reduce_smax_q512: {
14304 Function *F =
14305 CGM.getIntrinsic(Intrinsic::vector_reduce_smax, Ops[0]->getType());
14306 return Builder.CreateCall(F, {Ops[0]});
14307 }
14308 case X86::BI__builtin_ia32_reduce_smin_d512:
14309 case X86::BI__builtin_ia32_reduce_smin_q512: {
14310 Function *F =
14311 CGM.getIntrinsic(Intrinsic::vector_reduce_smin, Ops[0]->getType());
14312 return Builder.CreateCall(F, {Ops[0]});
14313 }
14314 case X86::BI__builtin_ia32_reduce_umax_d512:
14315 case X86::BI__builtin_ia32_reduce_umax_q512: {
14316 Function *F =
14317 CGM.getIntrinsic(Intrinsic::vector_reduce_umax, Ops[0]->getType());
14318 return Builder.CreateCall(F, {Ops[0]});
14319 }
14320 case X86::BI__builtin_ia32_reduce_umin_d512:
14321 case X86::BI__builtin_ia32_reduce_umin_q512: {
14322 Function *F =
14323 CGM.getIntrinsic(Intrinsic::vector_reduce_umin, Ops[0]->getType());
14324 return Builder.CreateCall(F, {Ops[0]});
14325 }
14326
14327 // 3DNow!
14328 case X86::BI__builtin_ia32_pswapdsf:
14329 case X86::BI__builtin_ia32_pswapdsi: {
14330 llvm::Type *MMXTy = llvm::Type::getX86_MMXTy(getLLVMContext());
14331 Ops[0] = Builder.CreateBitCast(Ops[0], MMXTy, "cast");
14332 llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_3dnowa_pswapd);
14333 return Builder.CreateCall(F, Ops, "pswapd");
14334 }
14335 case X86::BI__builtin_ia32_rdrand16_step:
14336 case X86::BI__builtin_ia32_rdrand32_step:
14337 case X86::BI__builtin_ia32_rdrand64_step:
14338 case X86::BI__builtin_ia32_rdseed16_step:
14339 case X86::BI__builtin_ia32_rdseed32_step:
14340 case X86::BI__builtin_ia32_rdseed64_step: {
14341 Intrinsic::ID ID;
14342 switch (BuiltinID) {
14343 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14343)
;
14344 case X86::BI__builtin_ia32_rdrand16_step:
14345 ID = Intrinsic::x86_rdrand_16;
14346 break;
14347 case X86::BI__builtin_ia32_rdrand32_step:
14348 ID = Intrinsic::x86_rdrand_32;
14349 break;
14350 case X86::BI__builtin_ia32_rdrand64_step:
14351 ID = Intrinsic::x86_rdrand_64;
14352 break;
14353 case X86::BI__builtin_ia32_rdseed16_step:
14354 ID = Intrinsic::x86_rdseed_16;
14355 break;
14356 case X86::BI__builtin_ia32_rdseed32_step:
14357 ID = Intrinsic::x86_rdseed_32;
14358 break;
14359 case X86::BI__builtin_ia32_rdseed64_step:
14360 ID = Intrinsic::x86_rdseed_64;
14361 break;
14362 }
14363
14364 Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID));
14365 Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 0),
14366 Ops[0]);
14367 return Builder.CreateExtractValue(Call, 1);
14368 }
14369 case X86::BI__builtin_ia32_addcarryx_u32:
14370 case X86::BI__builtin_ia32_addcarryx_u64:
14371 case X86::BI__builtin_ia32_subborrow_u32:
14372 case X86::BI__builtin_ia32_subborrow_u64: {
14373 Intrinsic::ID IID;
14374 switch (BuiltinID) {
14375 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14375)
;
14376 case X86::BI__builtin_ia32_addcarryx_u32:
14377 IID = Intrinsic::x86_addcarry_32;
14378 break;
14379 case X86::BI__builtin_ia32_addcarryx_u64:
14380 IID = Intrinsic::x86_addcarry_64;
14381 break;
14382 case X86::BI__builtin_ia32_subborrow_u32:
14383 IID = Intrinsic::x86_subborrow_32;
14384 break;
14385 case X86::BI__builtin_ia32_subborrow_u64:
14386 IID = Intrinsic::x86_subborrow_64;
14387 break;
14388 }
14389
14390 Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID),
14391 { Ops[0], Ops[1], Ops[2] });
14392 Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1),
14393 Ops[3]);
14394 return Builder.CreateExtractValue(Call, 0);
14395 }
14396
14397 case X86::BI__builtin_ia32_fpclassps128_mask:
14398 case X86::BI__builtin_ia32_fpclassps256_mask:
14399 case X86::BI__builtin_ia32_fpclassps512_mask:
14400 case X86::BI__builtin_ia32_fpclassph128_mask:
14401 case X86::BI__builtin_ia32_fpclassph256_mask:
14402 case X86::BI__builtin_ia32_fpclassph512_mask:
14403 case X86::BI__builtin_ia32_fpclasspd128_mask:
14404 case X86::BI__builtin_ia32_fpclasspd256_mask:
14405 case X86::BI__builtin_ia32_fpclasspd512_mask: {
14406 unsigned NumElts =
14407 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
14408 Value *MaskIn = Ops[2];
14409 Ops.erase(&Ops[2]);
14410
14411 Intrinsic::ID ID;
14412 switch (BuiltinID) {
14413 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14413)
;
14414 case X86::BI__builtin_ia32_fpclassph128_mask:
14415 ID = Intrinsic::x86_avx512fp16_fpclass_ph_128;
14416 break;
14417 case X86::BI__builtin_ia32_fpclassph256_mask:
14418 ID = Intrinsic::x86_avx512fp16_fpclass_ph_256;
14419 break;
14420 case X86::BI__builtin_ia32_fpclassph512_mask:
14421 ID = Intrinsic::x86_avx512fp16_fpclass_ph_512;
14422 break;
14423 case X86::BI__builtin_ia32_fpclassps128_mask:
14424 ID = Intrinsic::x86_avx512_fpclass_ps_128;
14425 break;
14426 case X86::BI__builtin_ia32_fpclassps256_mask:
14427 ID = Intrinsic::x86_avx512_fpclass_ps_256;
14428 break;
14429 case X86::BI__builtin_ia32_fpclassps512_mask:
14430 ID = Intrinsic::x86_avx512_fpclass_ps_512;
14431 break;
14432 case X86::BI__builtin_ia32_fpclasspd128_mask:
14433 ID = Intrinsic::x86_avx512_fpclass_pd_128;
14434 break;
14435 case X86::BI__builtin_ia32_fpclasspd256_mask:
14436 ID = Intrinsic::x86_avx512_fpclass_pd_256;
14437 break;
14438 case X86::BI__builtin_ia32_fpclasspd512_mask:
14439 ID = Intrinsic::x86_avx512_fpclass_pd_512;
14440 break;
14441 }
14442
14443 Value *Fpclass = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
14444 return EmitX86MaskedCompareResult(*this, Fpclass, NumElts, MaskIn);
14445 }
14446
14447 case X86::BI__builtin_ia32_vp2intersect_q_512:
14448 case X86::BI__builtin_ia32_vp2intersect_q_256:
14449 case X86::BI__builtin_ia32_vp2intersect_q_128:
14450 case X86::BI__builtin_ia32_vp2intersect_d_512:
14451 case X86::BI__builtin_ia32_vp2intersect_d_256:
14452 case X86::BI__builtin_ia32_vp2intersect_d_128: {
14453 unsigned NumElts =
14454 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
14455 Intrinsic::ID ID;
14456
14457 switch (BuiltinID) {
14458 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14458)
;
14459 case X86::BI__builtin_ia32_vp2intersect_q_512:
14460 ID = Intrinsic::x86_avx512_vp2intersect_q_512;
14461 break;
14462 case X86::BI__builtin_ia32_vp2intersect_q_256:
14463 ID = Intrinsic::x86_avx512_vp2intersect_q_256;
14464 break;
14465 case X86::BI__builtin_ia32_vp2intersect_q_128:
14466 ID = Intrinsic::x86_avx512_vp2intersect_q_128;
14467 break;
14468 case X86::BI__builtin_ia32_vp2intersect_d_512:
14469 ID = Intrinsic::x86_avx512_vp2intersect_d_512;
14470 break;
14471 case X86::BI__builtin_ia32_vp2intersect_d_256:
14472 ID = Intrinsic::x86_avx512_vp2intersect_d_256;
14473 break;
14474 case X86::BI__builtin_ia32_vp2intersect_d_128:
14475 ID = Intrinsic::x86_avx512_vp2intersect_d_128;
14476 break;
14477 }
14478
14479 Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID), {Ops[0], Ops[1]});
14480 Value *Result = Builder.CreateExtractValue(Call, 0);
14481 Result = EmitX86MaskedCompareResult(*this, Result, NumElts, nullptr);
14482 Builder.CreateDefaultAlignedStore(Result, Ops[2]);
14483
14484 Result = Builder.CreateExtractValue(Call, 1);
14485 Result = EmitX86MaskedCompareResult(*this, Result, NumElts, nullptr);
14486 return Builder.CreateDefaultAlignedStore(Result, Ops[3]);
14487 }
14488
14489 case X86::BI__builtin_ia32_vpmultishiftqb128:
14490 case X86::BI__builtin_ia32_vpmultishiftqb256:
14491 case X86::BI__builtin_ia32_vpmultishiftqb512: {
14492 Intrinsic::ID ID;
14493 switch (BuiltinID) {
14494 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14494)
;
14495 case X86::BI__builtin_ia32_vpmultishiftqb128:
14496 ID = Intrinsic::x86_avx512_pmultishift_qb_128;
14497 break;
14498 case X86::BI__builtin_ia32_vpmultishiftqb256:
14499 ID = Intrinsic::x86_avx512_pmultishift_qb_256;
14500 break;
14501 case X86::BI__builtin_ia32_vpmultishiftqb512:
14502 ID = Intrinsic::x86_avx512_pmultishift_qb_512;
14503 break;
14504 }
14505
14506 return Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
14507 }
14508
14509 case X86::BI__builtin_ia32_vpshufbitqmb128_mask:
14510 case X86::BI__builtin_ia32_vpshufbitqmb256_mask:
14511 case X86::BI__builtin_ia32_vpshufbitqmb512_mask: {
14512 unsigned NumElts =
14513 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
14514 Value *MaskIn = Ops[2];
14515 Ops.erase(&Ops[2]);
14516
14517 Intrinsic::ID ID;
14518 switch (BuiltinID) {
14519 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14519)
;
14520 case X86::BI__builtin_ia32_vpshufbitqmb128_mask:
14521 ID = Intrinsic::x86_avx512_vpshufbitqmb_128;
14522 break;
14523 case X86::BI__builtin_ia32_vpshufbitqmb256_mask:
14524 ID = Intrinsic::x86_avx512_vpshufbitqmb_256;
14525 break;
14526 case X86::BI__builtin_ia32_vpshufbitqmb512_mask:
14527 ID = Intrinsic::x86_avx512_vpshufbitqmb_512;
14528 break;
14529 }
14530
14531 Value *Shufbit = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
14532 return EmitX86MaskedCompareResult(*this, Shufbit, NumElts, MaskIn);
14533 }
14534
14535 // packed comparison intrinsics
14536 case X86::BI__builtin_ia32_cmpeqps:
14537 case X86::BI__builtin_ia32_cmpeqpd:
14538 return getVectorFCmpIR(CmpInst::FCMP_OEQ, /*IsSignaling*/false);
14539 case X86::BI__builtin_ia32_cmpltps:
14540 case X86::BI__builtin_ia32_cmpltpd:
14541 return getVectorFCmpIR(CmpInst::FCMP_OLT, /*IsSignaling*/true);
14542 case X86::BI__builtin_ia32_cmpleps:
14543 case X86::BI__builtin_ia32_cmplepd:
14544 return getVectorFCmpIR(CmpInst::FCMP_OLE, /*IsSignaling*/true);
14545 case X86::BI__builtin_ia32_cmpunordps:
14546 case X86::BI__builtin_ia32_cmpunordpd:
14547 return getVectorFCmpIR(CmpInst::FCMP_UNO, /*IsSignaling*/false);
14548 case X86::BI__builtin_ia32_cmpneqps:
14549 case X86::BI__builtin_ia32_cmpneqpd:
14550 return getVectorFCmpIR(CmpInst::FCMP_UNE, /*IsSignaling*/false);
14551 case X86::BI__builtin_ia32_cmpnltps:
14552 case X86::BI__builtin_ia32_cmpnltpd:
14553 return getVectorFCmpIR(CmpInst::FCMP_UGE, /*IsSignaling*/true);
14554 case X86::BI__builtin_ia32_cmpnleps:
14555 case X86::BI__builtin_ia32_cmpnlepd:
14556 return getVectorFCmpIR(CmpInst::FCMP_UGT, /*IsSignaling*/true);
14557 case X86::BI__builtin_ia32_cmpordps:
14558 case X86::BI__builtin_ia32_cmpordpd:
14559 return getVectorFCmpIR(CmpInst::FCMP_ORD, /*IsSignaling*/false);
14560 case X86::BI__builtin_ia32_cmpph128_mask:
14561 case X86::BI__builtin_ia32_cmpph256_mask:
14562 case X86::BI__builtin_ia32_cmpph512_mask:
14563 case X86::BI__builtin_ia32_cmpps128_mask:
14564 case X86::BI__builtin_ia32_cmpps256_mask:
14565 case X86::BI__builtin_ia32_cmpps512_mask:
14566 case X86::BI__builtin_ia32_cmppd128_mask:
14567 case X86::BI__builtin_ia32_cmppd256_mask:
14568 case X86::BI__builtin_ia32_cmppd512_mask:
14569 IsMaskFCmp = true;
14570 LLVM_FALLTHROUGH[[gnu::fallthrough]];
14571 case X86::BI__builtin_ia32_cmpps:
14572 case X86::BI__builtin_ia32_cmpps256:
14573 case X86::BI__builtin_ia32_cmppd:
14574 case X86::BI__builtin_ia32_cmppd256: {
14575 // Lowering vector comparisons to fcmp instructions, while
14576 // ignoring signalling behaviour requested
14577 // ignoring rounding mode requested
14578 // This is only possible if fp-model is not strict and FENV_ACCESS is off.
14579
14580 // The third argument is the comparison condition, and integer in the
14581 // range [0, 31]
14582 unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x1f;
14583
14584 // Lowering to IR fcmp instruction.
14585 // Ignoring requested signaling behaviour,
14586 // e.g. both _CMP_GT_OS & _CMP_GT_OQ are translated to FCMP_OGT.
14587 FCmpInst::Predicate Pred;
14588 bool IsSignaling;
14589 // Predicates for 16-31 repeat the 0-15 predicates. Only the signalling
14590 // behavior is inverted. We'll handle that after the switch.
14591 switch (CC & 0xf) {
14592 case 0x00: Pred = FCmpInst::FCMP_OEQ; IsSignaling = false; break;
14593 case 0x01: Pred = FCmpInst::FCMP_OLT; IsSignaling = true; break;
14594 case 0x02: Pred = FCmpInst::FCMP_OLE; IsSignaling = true; break;
14595 case 0x03: Pred = FCmpInst::FCMP_UNO; IsSignaling = false; break;
14596 case 0x04: Pred = FCmpInst::FCMP_UNE; IsSignaling = false; break;
14597 case 0x05: Pred = FCmpInst::FCMP_UGE; IsSignaling = true; break;
14598 case 0x06: Pred = FCmpInst::FCMP_UGT; IsSignaling = true; break;
14599 case 0x07: Pred = FCmpInst::FCMP_ORD; IsSignaling = false; break;
14600 case 0x08: Pred = FCmpInst::FCMP_UEQ; IsSignaling = false; break;
14601 case 0x09: Pred = FCmpInst::FCMP_ULT; IsSignaling = true; break;
14602 case 0x0a: Pred = FCmpInst::FCMP_ULE; IsSignaling = true; break;
14603 case 0x0b: Pred = FCmpInst::FCMP_FALSE; IsSignaling = false; break;
14604 case 0x0c: Pred = FCmpInst::FCMP_ONE; IsSignaling = false; break;
14605 case 0x0d: Pred = FCmpInst::FCMP_OGE; IsSignaling = true; break;
14606 case 0x0e: Pred = FCmpInst::FCMP_OGT; IsSignaling = true; break;
14607 case 0x0f: Pred = FCmpInst::FCMP_TRUE; IsSignaling = false; break;
14608 default: llvm_unreachable("Unhandled CC")::llvm::llvm_unreachable_internal("Unhandled CC", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14608)
;
14609 }
14610
14611 // Invert the signalling behavior for 16-31.
14612 if (CC & 0x10)
14613 IsSignaling = !IsSignaling;
14614
14615 // If the predicate is true or false and we're using constrained intrinsics,
14616 // we don't have a compare intrinsic we can use. Just use the legacy X86
14617 // specific intrinsic.
14618 // If the intrinsic is mask enabled and we're using constrained intrinsics,
14619 // use the legacy X86 specific intrinsic.
14620 if (Builder.getIsFPConstrained() &&
14621 (Pred == FCmpInst::FCMP_TRUE || Pred == FCmpInst::FCMP_FALSE ||
14622 IsMaskFCmp)) {
14623
14624 Intrinsic::ID IID;
14625 switch (BuiltinID) {
14626 default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14626)
;
14627 case X86::BI__builtin_ia32_cmpps:
14628 IID = Intrinsic::x86_sse_cmp_ps;
14629 break;
14630 case X86::BI__builtin_ia32_cmpps256:
14631 IID = Intrinsic::x86_avx_cmp_ps_256;
14632 break;
14633 case X86::BI__builtin_ia32_cmppd:
14634 IID = Intrinsic::x86_sse2_cmp_pd;
14635 break;
14636 case X86::BI__builtin_ia32_cmppd256:
14637 IID = Intrinsic::x86_avx_cmp_pd_256;
14638 break;
14639 case X86::BI__builtin_ia32_cmpps512_mask:
14640 IID = Intrinsic::x86_avx512_mask_cmp_ps_512;
14641 break;
14642 case X86::BI__builtin_ia32_cmppd512_mask:
14643 IID = Intrinsic::x86_avx512_mask_cmp_pd_512;
14644 break;
14645 case X86::BI__builtin_ia32_cmpps128_mask:
14646 IID = Intrinsic::x86_avx512_mask_cmp_ps_128;
14647 break;
14648 case X86::BI__builtin_ia32_cmpps256_mask:
14649 IID = Intrinsic::x86_avx512_mask_cmp_ps_256;
14650 break;
14651 case X86::BI__builtin_ia32_cmppd128_mask:
14652 IID = Intrinsic::x86_avx512_mask_cmp_pd_128;
14653 break;
14654 case X86::BI__builtin_ia32_cmppd256_mask:
14655 IID = Intrinsic::x86_avx512_mask_cmp_pd_256;
14656 break;
14657 }
14658
14659 Function *Intr = CGM.getIntrinsic(IID);
14660 if (IsMaskFCmp) {
14661 unsigned NumElts =
14662 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
14663 Ops[3] = getMaskVecValue(*this, Ops[3], NumElts);
14664 Value *Cmp = Builder.CreateCall(Intr, Ops);
14665 return EmitX86MaskedCompareResult(*this, Cmp, NumElts, nullptr);
14666 }
14667
14668 return Builder.CreateCall(Intr, Ops);
14669 }
14670
14671 // Builtins without the _mask suffix return a vector of integers
14672 // of the same width as the input vectors
14673 if (IsMaskFCmp) {
14674 // We ignore SAE if strict FP is disabled. We only keep precise
14675 // exception behavior under strict FP.
14676 // NOTE: If strict FP does ever go through here a CGFPOptionsRAII
14677 // object will be required.
14678 unsigned NumElts =
14679 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements();
14680 Value *Cmp;
14681 if (IsSignaling)
14682 Cmp = Builder.CreateFCmpS(Pred, Ops[0], Ops[1]);
14683 else
14684 Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]);
14685 return EmitX86MaskedCompareResult(*this, Cmp, NumElts, Ops[3]);
14686 }
14687
14688 return getVectorFCmpIR(Pred, IsSignaling);
14689 }
14690
14691 // SSE scalar comparison intrinsics
14692 case X86::BI__builtin_ia32_cmpeqss:
14693 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 0);
14694 case X86::BI__builtin_ia32_cmpltss:
14695 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 1);
14696 case X86::BI__builtin_ia32_cmpless:
14697 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 2);
14698 case X86::BI__builtin_ia32_cmpunordss:
14699 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 3);
14700 case X86::BI__builtin_ia32_cmpneqss:
14701 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 4);
14702 case X86::BI__builtin_ia32_cmpnltss:
14703 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 5);
14704 case X86::BI__builtin_ia32_cmpnless:
14705 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 6);
14706 case X86::BI__builtin_ia32_cmpordss:
14707 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 7);
14708 case X86::BI__builtin_ia32_cmpeqsd:
14709 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 0);
14710 case X86::BI__builtin_ia32_cmpltsd:
14711 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 1);
14712 case X86::BI__builtin_ia32_cmplesd:
14713 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 2);
14714 case X86::BI__builtin_ia32_cmpunordsd:
14715 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 3);
14716 case X86::BI__builtin_ia32_cmpneqsd:
14717 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 4);
14718 case X86::BI__builtin_ia32_cmpnltsd:
14719 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 5);
14720 case X86::BI__builtin_ia32_cmpnlesd:
14721 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 6);
14722 case X86::BI__builtin_ia32_cmpordsd:
14723 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 7);
14724
14725 // f16c half2float intrinsics
14726 case X86::BI__builtin_ia32_vcvtph2ps:
14727 case X86::BI__builtin_ia32_vcvtph2ps256:
14728 case X86::BI__builtin_ia32_vcvtph2ps_mask:
14729 case X86::BI__builtin_ia32_vcvtph2ps256_mask:
14730 case X86::BI__builtin_ia32_vcvtph2ps512_mask: {
14731 CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
14732 return EmitX86CvtF16ToFloatExpr(*this, Ops, ConvertType(E->getType()));
14733 }
14734
14735// AVX512 bf16 intrinsics
14736 case X86::BI__builtin_ia32_cvtneps2bf16_128_mask: {
14737 Ops[2] = getMaskVecValue(
14738 *this, Ops[2],
14739 cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements());
14740 Intrinsic::ID IID = Intrinsic::x86_avx512bf16_mask_cvtneps2bf16_128;
14741 return Builder.CreateCall(CGM.getIntrinsic(IID), Ops);
14742 }
14743 case X86::BI__builtin_ia32_cvtsbf162ss_32:
14744 return EmitX86CvtBF16ToFloatExpr(*this, E, Ops);
14745
14746 case X86::BI__builtin_ia32_cvtneps2bf16_256_mask:
14747 case X86::BI__builtin_ia32_cvtneps2bf16_512_mask: {
14748 Intrinsic::ID IID;
14749 switch (BuiltinID) {
14750 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14750)
;
14751 case X86::BI__builtin_ia32_cvtneps2bf16_256_mask:
14752 IID = Intrinsic::x86_avx512bf16_cvtneps2bf16_256;
14753 break;
14754 case X86::BI__builtin_ia32_cvtneps2bf16_512_mask:
14755 IID = Intrinsic::x86_avx512bf16_cvtneps2bf16_512;
14756 break;
14757 }
14758 Value *Res = Builder.CreateCall(CGM.getIntrinsic(IID), Ops[0]);
14759 return EmitX86Select(*this, Ops[2], Res, Ops[1]);
14760 }
14761
14762 case X86::BI__emul:
14763 case X86::BI__emulu: {
14764 llvm::Type *Int64Ty = llvm::IntegerType::get(getLLVMContext(), 64);
14765 bool isSigned = (BuiltinID == X86::BI__emul);
14766 Value *LHS = Builder.CreateIntCast(Ops[0], Int64Ty, isSigned);
14767 Value *RHS = Builder.CreateIntCast(Ops[1], Int64Ty, isSigned);
14768 return Builder.CreateMul(LHS, RHS, "", !isSigned, isSigned);
14769 }
14770 case X86::BI__mulh:
14771 case X86::BI__umulh:
14772 case X86::BI_mul128:
14773 case X86::BI_umul128: {
14774 llvm::Type *ResType = ConvertType(E->getType());
14775 llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128);
14776
14777 bool IsSigned = (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI_mul128);
14778 Value *LHS = Builder.CreateIntCast(Ops[0], Int128Ty, IsSigned);
14779 Value *RHS = Builder.CreateIntCast(Ops[1], Int128Ty, IsSigned);
14780
14781 Value *MulResult, *HigherBits;
14782 if (IsSigned) {
14783 MulResult = Builder.CreateNSWMul(LHS, RHS);
14784 HigherBits = Builder.CreateAShr(MulResult, 64);
14785 } else {
14786 MulResult = Builder.CreateNUWMul(LHS, RHS);
14787 HigherBits = Builder.CreateLShr(MulResult, 64);
14788 }
14789 HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned);
14790
14791 if (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI__umulh)
14792 return HigherBits;
14793
14794 Address HighBitsAddress = EmitPointerWithAlignment(E->getArg(2));
14795 Builder.CreateStore(HigherBits, HighBitsAddress);
14796 return Builder.CreateIntCast(MulResult, ResType, IsSigned);
14797 }
14798
14799 case X86::BI__faststorefence: {
14800 return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent,
14801 llvm::SyncScope::System);
14802 }
14803 case X86::BI__shiftleft128:
14804 case X86::BI__shiftright128: {
14805 llvm::Function *F = CGM.getIntrinsic(
14806 BuiltinID == X86::BI__shiftleft128 ? Intrinsic::fshl : Intrinsic::fshr,
14807 Int64Ty);
14808 // Flip low/high ops and zero-extend amount to matching type.
14809 // shiftleft128(Low, High, Amt) -> fshl(High, Low, Amt)
14810 // shiftright128(Low, High, Amt) -> fshr(High, Low, Amt)
14811 std::swap(Ops[0], Ops[1]);
14812 Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty);
14813 return Builder.CreateCall(F, Ops);
14814 }
14815 case X86::BI_ReadWriteBarrier:
14816 case X86::BI_ReadBarrier:
14817 case X86::BI_WriteBarrier: {
14818 return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent,
14819 llvm::SyncScope::SingleThread);
14820 }
14821
14822 case X86::BI_AddressOfReturnAddress: {
14823 Function *F =
14824 CGM.getIntrinsic(Intrinsic::addressofreturnaddress, AllocaInt8PtrTy);
14825 return Builder.CreateCall(F);
14826 }
14827 case X86::BI__stosb: {
14828 // We treat __stosb as a volatile memset - it may not generate "rep stosb"
14829 // instruction, but it will create a memset that won't be optimized away.
14830 return Builder.CreateMemSet(Ops[0], Ops[1], Ops[2], Align(1), true);
14831 }
14832 case X86::BI__ud2:
14833 // llvm.trap makes a ud2a instruction on x86.
14834 return EmitTrapCall(Intrinsic::trap);
14835 case X86::BI__int2c: {
14836 // This syscall signals a driver assertion failure in x86 NT kernels.
14837 llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false);
14838 llvm::InlineAsm *IA =
14839 llvm::InlineAsm::get(FTy, "int $$0x2c", "", /*hasSideEffects=*/true);
14840 llvm::AttributeList NoReturnAttr = llvm::AttributeList::get(
14841 getLLVMContext(), llvm::AttributeList::FunctionIndex,
14842 llvm::Attribute::NoReturn);
14843 llvm::CallInst *CI = Builder.CreateCall(IA);
14844 CI->setAttributes(NoReturnAttr);
14845 return CI;
14846 }
14847 case X86::BI__readfsbyte:
14848 case X86::BI__readfsword:
14849 case X86::BI__readfsdword:
14850 case X86::BI__readfsqword: {
14851 llvm::Type *IntTy = ConvertType(E->getType());
14852 Value *Ptr =
14853 Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 257));
14854 LoadInst *Load = Builder.CreateAlignedLoad(
14855 IntTy, Ptr, getContext().getTypeAlignInChars(E->getType()));
14856 Load->setVolatile(true);
14857 return Load;
14858 }
14859 case X86::BI__readgsbyte:
14860 case X86::BI__readgsword:
14861 case X86::BI__readgsdword:
14862 case X86::BI__readgsqword: {
14863 llvm::Type *IntTy = ConvertType(E->getType());
14864 Value *Ptr =
14865 Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 256));
14866 LoadInst *Load = Builder.CreateAlignedLoad(
14867 IntTy, Ptr, getContext().getTypeAlignInChars(E->getType()));
14868 Load->setVolatile(true);
14869 return Load;
14870 }
14871 case X86::BI__builtin_ia32_paddsb512:
14872 case X86::BI__builtin_ia32_paddsw512:
14873 case X86::BI__builtin_ia32_paddsb256:
14874 case X86::BI__builtin_ia32_paddsw256:
14875 case X86::BI__builtin_ia32_paddsb128:
14876 case X86::BI__builtin_ia32_paddsw128:
14877 return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::sadd_sat);
14878 case X86::BI__builtin_ia32_paddusb512:
14879 case X86::BI__builtin_ia32_paddusw512:
14880 case X86::BI__builtin_ia32_paddusb256:
14881 case X86::BI__builtin_ia32_paddusw256:
14882 case X86::BI__builtin_ia32_paddusb128:
14883 case X86::BI__builtin_ia32_paddusw128:
14884 return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::uadd_sat);
14885 case X86::BI__builtin_ia32_psubsb512:
14886 case X86::BI__builtin_ia32_psubsw512:
14887 case X86::BI__builtin_ia32_psubsb256:
14888 case X86::BI__builtin_ia32_psubsw256:
14889 case X86::BI__builtin_ia32_psubsb128:
14890 case X86::BI__builtin_ia32_psubsw128:
14891 return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::ssub_sat);
14892 case X86::BI__builtin_ia32_psubusb512:
14893 case X86::BI__builtin_ia32_psubusw512:
14894 case X86::BI__builtin_ia32_psubusb256:
14895 case X86::BI__builtin_ia32_psubusw256:
14896 case X86::BI__builtin_ia32_psubusb128:
14897 case X86::BI__builtin_ia32_psubusw128:
14898 return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::usub_sat);
14899 case X86::BI__builtin_ia32_encodekey128_u32: {
14900 Intrinsic::ID IID = Intrinsic::x86_encodekey128;
14901
14902 Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), {Ops[0], Ops[1]});
14903
14904 for (int i = 0; i < 6; ++i) {
14905 Value *Extract = Builder.CreateExtractValue(Call, i + 1);
14906 Value *Ptr = Builder.CreateConstGEP1_32(Int8Ty, Ops[2], i * 16);
14907 Ptr = Builder.CreateBitCast(
14908 Ptr, llvm::PointerType::getUnqual(Extract->getType()));
14909 Builder.CreateAlignedStore(Extract, Ptr, Align(1));
14910 }
14911
14912 return Builder.CreateExtractValue(Call, 0);
14913 }
14914 case X86::BI__builtin_ia32_encodekey256_u32: {
14915 Intrinsic::ID IID = Intrinsic::x86_encodekey256;
14916
14917 Value *Call =
14918 Builder.CreateCall(CGM.getIntrinsic(IID), {Ops[0], Ops[1], Ops[2]});
14919
14920 for (int i = 0; i < 7; ++i) {
14921 Value *Extract = Builder.CreateExtractValue(Call, i + 1);
14922 Value *Ptr = Builder.CreateConstGEP1_32(Int8Ty, Ops[3], i * 16);
14923 Ptr = Builder.CreateBitCast(
14924 Ptr, llvm::PointerType::getUnqual(Extract->getType()));
14925 Builder.CreateAlignedStore(Extract, Ptr, Align(1));
14926 }
14927
14928 return Builder.CreateExtractValue(Call, 0);
14929 }
14930 case X86::BI__builtin_ia32_aesenc128kl_u8:
14931 case X86::BI__builtin_ia32_aesdec128kl_u8:
14932 case X86::BI__builtin_ia32_aesenc256kl_u8:
14933 case X86::BI__builtin_ia32_aesdec256kl_u8: {
14934 Intrinsic::ID IID;
14935 StringRef BlockName;
14936 switch (BuiltinID) {
14937 default:
14938 llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 14938)
;
14939 case X86::BI__builtin_ia32_aesenc128kl_u8:
14940 IID = Intrinsic::x86_aesenc128kl;
14941 BlockName = "aesenc128kl";
14942 break;
14943 case X86::BI__builtin_ia32_aesdec128kl_u8:
14944 IID = Intrinsic::x86_aesdec128kl;
14945 BlockName = "aesdec128kl";
14946 break;
14947 case X86::BI__builtin_ia32_aesenc256kl_u8:
14948 IID = Intrinsic::x86_aesenc256kl;
14949 BlockName = "aesenc256kl";
14950 break;
14951 case X86::BI__builtin_ia32_aesdec256kl_u8:
14952 IID = Intrinsic::x86_aesdec256kl;
14953 BlockName = "aesdec256kl";
14954 break;
14955 }
14956
14957 Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), {Ops[1], Ops[2]});
14958
14959 BasicBlock *NoError =
14960 createBasicBlock(BlockName + "_no_error", this->CurFn);
14961 BasicBlock *Error = createBasicBlock(BlockName + "_error", this->CurFn);
14962 BasicBlock *End = createBasicBlock(BlockName + "_end", this->CurFn);
14963
14964 Value *Ret = Builder.CreateExtractValue(Call, 0);
14965 Value *Succ = Builder.CreateTrunc(Ret, Builder.getInt1Ty());
14966 Value *Out = Builder.CreateExtractValue(Call, 1);
14967 Builder.CreateCondBr(Succ, NoError, Error);
14968
14969 Builder.SetInsertPoint(NoError);
14970 Builder.CreateDefaultAlignedStore(Out, Ops[0]);
14971 Builder.CreateBr(End);
14972
14973 Builder.SetInsertPoint(Error);
14974 Constant *Zero = llvm::Constant::getNullValue(Out->getType());
14975 Builder.CreateDefaultAlignedStore(Zero, Ops[0]);
14976 Builder.CreateBr(End);
14977
14978 Builder.SetInsertPoint(End);
14979 return Builder.CreateExtractValue(Call, 0);
14980 }
14981 case X86::BI__builtin_ia32_aesencwide128kl_u8:
14982 case X86::BI__builtin_ia32_aesdecwide128kl_u8:
14983 case X86::BI__builtin_ia32_aesencwide256kl_u8:
14984 case X86::BI__builtin_ia32_aesdecwide256kl_u8: {
14985 Intrinsic::ID IID;
14986 StringRef BlockName;
14987 switch (BuiltinID) {
14988 case X86::BI__builtin_ia32_aesencwide128kl_u8:
14989 IID = Intrinsic::x86_aesencwide128kl;
14990 BlockName = "aesencwide128kl";
14991 break;
14992 case X86::BI__builtin_ia32_aesdecwide128kl_u8:
14993 IID = Intrinsic::x86_aesdecwide128kl;
14994 BlockName = "aesdecwide128kl";
14995 break;
14996 case X86::BI__builtin_ia32_aesencwide256kl_u8:
14997 IID = Intrinsic::x86_aesencwide256kl;
14998 BlockName = "aesencwide256kl";
14999 break;
15000 case X86::BI__builtin_ia32_aesdecwide256kl_u8:
15001 IID = Intrinsic::x86_aesdecwide256kl;
15002 BlockName = "aesdecwide256kl";
15003 break;
15004 }
15005
15006 llvm::Type *Ty = FixedVectorType::get(Builder.getInt64Ty(), 2);
15007 Value *InOps[9];
15008 InOps[0] = Ops[2];
15009 for (int i = 0; i != 8; ++i) {
15010 Value *Ptr = Builder.CreateConstGEP1_32(Ty, Ops[1], i);
15011 InOps[i + 1] = Builder.CreateAlignedLoad(Ty, Ptr, Align(16));
15012 }
15013
15014 Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), InOps);
15015
15016 BasicBlock *NoError =
15017 createBasicBlock(BlockName + "_no_error", this->CurFn);
15018 BasicBlock *Error = createBasicBlock(BlockName + "_error", this->CurFn);
15019 BasicBlock *End = createBasicBlock(BlockName + "_end", this->CurFn);
15020
15021 Value *Ret = Builder.CreateExtractValue(Call, 0);
15022 Value *Succ = Builder.CreateTrunc(Ret, Builder.getInt1Ty());
15023 Builder.CreateCondBr(Succ, NoError, Error);
15024
15025 Builder.SetInsertPoint(NoError);
15026 for (int i = 0; i != 8; ++i) {
15027 Value *Extract = Builder.CreateExtractValue(Call, i + 1);
15028 Value *Ptr = Builder.CreateConstGEP1_32(Extract->getType(), Ops[0], i);
15029 Builder.CreateAlignedStore(Extract, Ptr, Align(16));
15030 }
15031 Builder.CreateBr(End);
15032
15033 Builder.SetInsertPoint(Error);
15034 for (int i = 0; i != 8; ++i) {
15035 Value *Out = Builder.CreateExtractValue(Call, i + 1);
15036 Constant *Zero = llvm::Constant::getNullValue(Out->getType());
15037 Value *Ptr = Builder.CreateConstGEP1_32(Out->getType(), Ops[0], i);
15038 Builder.CreateAlignedStore(Zero, Ptr, Align(16));
15039 }
15040 Builder.CreateBr(End);
15041
15042 Builder.SetInsertPoint(End);
15043 return Builder.CreateExtractValue(Call, 0);
15044 }
15045 }
15046}
15047
15048Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID,
15049 const CallExpr *E) {
15050 SmallVector<Value*, 4> Ops;
15051
15052 for (unsigned i = 0, e = E->getNumArgs(); i != e; i++)
15053 Ops.push_back(EmitScalarExpr(E->getArg(i)));
15054
15055 Intrinsic::ID ID = Intrinsic::not_intrinsic;
15056
15057 switch (BuiltinID) {
15058 default: return nullptr;
15059
15060 // __builtin_ppc_get_timebase is GCC 4.8+'s PowerPC-specific name for what we
15061 // call __builtin_readcyclecounter.
15062 case PPC::BI__builtin_ppc_get_timebase:
15063 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::readcyclecounter));
15064
15065 // vec_ld, vec_xl_be, vec_lvsl, vec_lvsr
15066 case PPC::BI__builtin_altivec_lvx:
15067 case PPC::BI__builtin_altivec_lvxl:
15068 case PPC::BI__builtin_altivec_lvebx:
15069 case PPC::BI__builtin_altivec_lvehx:
15070 case PPC::BI__builtin_altivec_lvewx:
15071 case PPC::BI__builtin_altivec_lvsl:
15072 case PPC::BI__builtin_altivec_lvsr:
15073 case PPC::BI__builtin_vsx_lxvd2x:
15074 case PPC::BI__builtin_vsx_lxvw4x:
15075 case PPC::BI__builtin_vsx_lxvd2x_be:
15076 case PPC::BI__builtin_vsx_lxvw4x_be:
15077 case PPC::BI__builtin_vsx_lxvl:
15078 case PPC::BI__builtin_vsx_lxvll:
15079 {
15080 if(BuiltinID == PPC::BI__builtin_vsx_lxvl ||
15081 BuiltinID == PPC::BI__builtin_vsx_lxvll){
15082 Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy);
15083 }else {
15084 Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy);
15085 Ops[0] = Builder.CreateGEP(Int8Ty, Ops[1], Ops[0]);
15086 Ops.pop_back();
15087 }
15088
15089 switch (BuiltinID) {
15090 default: llvm_unreachable("Unsupported ld/lvsl/lvsr intrinsic!")::llvm::llvm_unreachable_internal("Unsupported ld/lvsl/lvsr intrinsic!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15090)
;
15091 case PPC::BI__builtin_altivec_lvx:
15092 ID = Intrinsic::ppc_altivec_lvx;
15093 break;
15094 case PPC::BI__builtin_altivec_lvxl:
15095 ID = Intrinsic::ppc_altivec_lvxl;
15096 break;
15097 case PPC::BI__builtin_altivec_lvebx:
15098 ID = Intrinsic::ppc_altivec_lvebx;
15099 break;
15100 case PPC::BI__builtin_altivec_lvehx:
15101 ID = Intrinsic::ppc_altivec_lvehx;
15102 break;
15103 case PPC::BI__builtin_altivec_lvewx:
15104 ID = Intrinsic::ppc_altivec_lvewx;
15105 break;
15106 case PPC::BI__builtin_altivec_lvsl:
15107 ID = Intrinsic::ppc_altivec_lvsl;
15108 break;
15109 case PPC::BI__builtin_altivec_lvsr:
15110 ID = Intrinsic::ppc_altivec_lvsr;
15111 break;
15112 case PPC::BI__builtin_vsx_lxvd2x:
15113 ID = Intrinsic::ppc_vsx_lxvd2x;
15114 break;
15115 case PPC::BI__builtin_vsx_lxvw4x:
15116 ID = Intrinsic::ppc_vsx_lxvw4x;
15117 break;
15118 case PPC::BI__builtin_vsx_lxvd2x_be:
15119 ID = Intrinsic::ppc_vsx_lxvd2x_be;
15120 break;
15121 case PPC::BI__builtin_vsx_lxvw4x_be:
15122 ID = Intrinsic::ppc_vsx_lxvw4x_be;
15123 break;
15124 case PPC::BI__builtin_vsx_lxvl:
15125 ID = Intrinsic::ppc_vsx_lxvl;
15126 break;
15127 case PPC::BI__builtin_vsx_lxvll:
15128 ID = Intrinsic::ppc_vsx_lxvll;
15129 break;
15130 }
15131 llvm::Function *F = CGM.getIntrinsic(ID);
15132 return Builder.CreateCall(F, Ops, "");
15133 }
15134
15135 // vec_st, vec_xst_be
15136 case PPC::BI__builtin_altivec_stvx:
15137 case PPC::BI__builtin_altivec_stvxl:
15138 case PPC::BI__builtin_altivec_stvebx:
15139 case PPC::BI__builtin_altivec_stvehx:
15140 case PPC::BI__builtin_altivec_stvewx:
15141 case PPC::BI__builtin_vsx_stxvd2x:
15142 case PPC::BI__builtin_vsx_stxvw4x:
15143 case PPC::BI__builtin_vsx_stxvd2x_be:
15144 case PPC::BI__builtin_vsx_stxvw4x_be:
15145 case PPC::BI__builtin_vsx_stxvl:
15146 case PPC::BI__builtin_vsx_stxvll:
15147 {
15148 if(BuiltinID == PPC::BI__builtin_vsx_stxvl ||
15149 BuiltinID == PPC::BI__builtin_vsx_stxvll ){
15150 Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy);
15151 }else {
15152 Ops[2] = Builder.CreateBitCast(Ops[2], Int8PtrTy);
15153 Ops[1] = Builder.CreateGEP(Int8Ty, Ops[2], Ops[1]);
15154 Ops.pop_back();
15155 }
15156
15157 switch (BuiltinID) {
15158 default: llvm_unreachable("Unsupported st intrinsic!")::llvm::llvm_unreachable_internal("Unsupported st intrinsic!"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15158)
;
15159 case PPC::BI__builtin_altivec_stvx:
15160 ID = Intrinsic::ppc_altivec_stvx;
15161 break;
15162 case PPC::BI__builtin_altivec_stvxl:
15163 ID = Intrinsic::ppc_altivec_stvxl;
15164 break;
15165 case PPC::BI__builtin_altivec_stvebx:
15166 ID = Intrinsic::ppc_altivec_stvebx;
15167 break;
15168 case PPC::BI__builtin_altivec_stvehx:
15169 ID = Intrinsic::ppc_altivec_stvehx;
15170 break;
15171 case PPC::BI__builtin_altivec_stvewx:
15172 ID = Intrinsic::ppc_altivec_stvewx;
15173 break;
15174 case PPC::BI__builtin_vsx_stxvd2x:
15175 ID = Intrinsic::ppc_vsx_stxvd2x;
15176 break;
15177 case PPC::BI__builtin_vsx_stxvw4x:
15178 ID = Intrinsic::ppc_vsx_stxvw4x;
15179 break;
15180 case PPC::BI__builtin_vsx_stxvd2x_be:
15181 ID = Intrinsic::ppc_vsx_stxvd2x_be;
15182 break;
15183 case PPC::BI__builtin_vsx_stxvw4x_be:
15184 ID = Intrinsic::ppc_vsx_stxvw4x_be;
15185 break;
15186 case PPC::BI__builtin_vsx_stxvl:
15187 ID = Intrinsic::ppc_vsx_stxvl;
15188 break;
15189 case PPC::BI__builtin_vsx_stxvll:
15190 ID = Intrinsic::ppc_vsx_stxvll;
15191 break;
15192 }
15193 llvm::Function *F = CGM.getIntrinsic(ID);
15194 return Builder.CreateCall(F, Ops, "");
15195 }
15196 case PPC::BI__builtin_vsx_ldrmb: {
15197 // Essentially boils down to performing an unaligned VMX load sequence so
15198 // as to avoid crossing a page boundary and then shuffling the elements
15199 // into the right side of the vector register.
15200 int64_t NumBytes = cast<ConstantInt>(Ops[1])->getZExtValue();
15201 llvm::Type *ResTy = ConvertType(E->getType());
15202 bool IsLE = getTarget().isLittleEndian();
15203
15204 // If the user wants the entire vector, just load the entire vector.
15205 if (NumBytes == 16) {
15206 Value *BC = Builder.CreateBitCast(Ops[0], ResTy->getPointerTo());
15207 Value *LD = Builder.CreateLoad(Address(BC, CharUnits::fromQuantity(1)));
15208 if (!IsLE)
15209 return LD;
15210
15211 // Reverse the bytes on LE.
15212 SmallVector<int, 16> RevMask;
15213 for (int Idx = 0; Idx < 16; Idx++)
15214 RevMask.push_back(15 - Idx);
15215 return Builder.CreateShuffleVector(LD, LD, RevMask);
15216 }
15217
15218 llvm::Function *Lvx = CGM.getIntrinsic(Intrinsic::ppc_altivec_lvx);
15219 llvm::Function *Lvs = CGM.getIntrinsic(IsLE ? Intrinsic::ppc_altivec_lvsr
15220 : Intrinsic::ppc_altivec_lvsl);
15221 llvm::Function *Vperm = CGM.getIntrinsic(Intrinsic::ppc_altivec_vperm);
15222 Value *HiMem = Builder.CreateGEP(
15223 Int8Ty, Ops[0], ConstantInt::get(Ops[1]->getType(), NumBytes - 1));
15224 Value *LoLd = Builder.CreateCall(Lvx, Ops[0], "ld.lo");
15225 Value *HiLd = Builder.CreateCall(Lvx, HiMem, "ld.hi");
15226 Value *Mask1 = Builder.CreateCall(Lvs, Ops[0], "mask1");
15227
15228 Ops.clear();
15229 Ops.push_back(IsLE ? HiLd : LoLd);
15230 Ops.push_back(IsLE ? LoLd : HiLd);
15231 Ops.push_back(Mask1);
15232 Value *AllElts = Builder.CreateCall(Vperm, Ops, "shuffle1");
15233 Constant *Zero = llvm::Constant::getNullValue(IsLE ? ResTy : AllElts->getType());
15234
15235 if (IsLE) {
15236 SmallVector<int, 16> Consts;
15237 for (int Idx = 0; Idx < 16; Idx++) {
15238 int Val = (NumBytes - Idx - 1 >= 0) ? (NumBytes - Idx - 1)
15239 : 16 - (NumBytes - Idx);
15240 Consts.push_back(Val);
15241 }
15242 return Builder.CreateShuffleVector(Builder.CreateBitCast(AllElts, ResTy),
15243 Zero, Consts);
15244 }
15245 SmallVector<Constant *, 16> Consts;
15246 for (int Idx = 0; Idx < 16; Idx++)
15247 Consts.push_back(Builder.getInt8(NumBytes + Idx));
15248 Value *Mask2 = ConstantVector::get(Consts);
15249 return Builder.CreateBitCast(
15250 Builder.CreateCall(Vperm, {Zero, AllElts, Mask2}, "shuffle2"), ResTy);
15251 }
15252 case PPC::BI__builtin_vsx_strmb: {
15253 int64_t NumBytes = cast<ConstantInt>(Ops[1])->getZExtValue();
15254 bool IsLE = getTarget().isLittleEndian();
15255 auto StoreSubVec = [&](unsigned Width, unsigned Offset, unsigned EltNo) {
15256 // Storing the whole vector, simply store it on BE and reverse bytes and
15257 // store on LE.
15258 if (Width == 16) {
15259 Value *BC =
15260 Builder.CreateBitCast(Ops[0], Ops[2]->getType()->getPointerTo());
15261 Value *StVec = Ops[2];
15262 if (IsLE) {
15263 SmallVector<int, 16> RevMask;
15264 for (int Idx = 0; Idx < 16; Idx++)
15265 RevMask.push_back(15 - Idx);
15266 StVec = Builder.CreateShuffleVector(Ops[2], Ops[2], RevMask);
15267 }
15268 return Builder.CreateStore(StVec,
15269 Address(BC, CharUnits::fromQuantity(1)));
15270 }
15271 auto *ConvTy = Int64Ty;
15272 unsigned NumElts = 0;
15273 switch (Width) {
15274 default:
15275 llvm_unreachable("width for stores must be a power of 2")::llvm::llvm_unreachable_internal("width for stores must be a power of 2"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15275)
;
15276 case 8:
15277 ConvTy = Int64Ty;
15278 NumElts = 2;
15279 break;
15280 case 4:
15281 ConvTy = Int32Ty;
15282 NumElts = 4;
15283 break;
15284 case 2:
15285 ConvTy = Int16Ty;
15286 NumElts = 8;
15287 break;
15288 case 1:
15289 ConvTy = Int8Ty;
15290 NumElts = 16;
15291 break;
15292 }
15293 Value *Vec = Builder.CreateBitCast(
15294 Ops[2], llvm::FixedVectorType::get(ConvTy, NumElts));
15295 Value *Ptr = Builder.CreateGEP(Int8Ty, Ops[0],
15296 ConstantInt::get(Int64Ty, Offset));
15297 Value *PtrBC = Builder.CreateBitCast(Ptr, ConvTy->getPointerTo());
15298 Value *Elt = Builder.CreateExtractElement(Vec, EltNo);
15299 if (IsLE && Width > 1) {
15300 Function *F = CGM.getIntrinsic(Intrinsic::bswap, ConvTy);
15301 Elt = Builder.CreateCall(F, Elt);
15302 }
15303 return Builder.CreateStore(Elt,
15304 Address(PtrBC, CharUnits::fromQuantity(1)));
15305 };
15306 unsigned Stored = 0;
15307 unsigned RemainingBytes = NumBytes;
15308 Value *Result;
15309 if (NumBytes == 16)
15310 return StoreSubVec(16, 0, 0);
15311 if (NumBytes >= 8) {
15312 Result = StoreSubVec(8, NumBytes - 8, IsLE ? 0 : 1);
15313 RemainingBytes -= 8;
15314 Stored += 8;
15315 }
15316 if (RemainingBytes >= 4) {
15317 Result = StoreSubVec(4, NumBytes - Stored - 4,
15318 IsLE ? (Stored >> 2) : 3 - (Stored >> 2));
15319 RemainingBytes -= 4;
15320 Stored += 4;
15321 }
15322 if (RemainingBytes >= 2) {
15323 Result = StoreSubVec(2, NumBytes - Stored - 2,
15324 IsLE ? (Stored >> 1) : 7 - (Stored >> 1));
15325 RemainingBytes -= 2;
15326 Stored += 2;
15327 }
15328 if (RemainingBytes)
15329 Result =
15330 StoreSubVec(1, NumBytes - Stored - 1, IsLE ? Stored : 15 - Stored);
15331 return Result;
15332 }
15333 // Square root
15334 case PPC::BI__builtin_vsx_xvsqrtsp:
15335 case PPC::BI__builtin_vsx_xvsqrtdp: {
15336 llvm::Type *ResultType = ConvertType(E->getType());
15337 Value *X = EmitScalarExpr(E->getArg(0));
15338 if (Builder.getIsFPConstrained()) {
15339 llvm::Function *F = CGM.getIntrinsic(
15340 Intrinsic::experimental_constrained_sqrt, ResultType);
15341 return Builder.CreateConstrainedFPCall(F, X);
15342 } else {
15343 llvm::Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType);
15344 return Builder.CreateCall(F, X);
15345 }
15346 }
15347 // Count leading zeros
15348 case PPC::BI__builtin_altivec_vclzb:
15349 case PPC::BI__builtin_altivec_vclzh:
15350 case PPC::BI__builtin_altivec_vclzw:
15351 case PPC::BI__builtin_altivec_vclzd: {
15352 llvm::Type *ResultType = ConvertType(E->getType());
15353 Value *X = EmitScalarExpr(E->getArg(0));
15354 Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
15355 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType);
15356 return Builder.CreateCall(F, {X, Undef});
15357 }
15358 case PPC::BI__builtin_altivec_vctzb:
15359 case PPC::BI__builtin_altivec_vctzh:
15360 case PPC::BI__builtin_altivec_vctzw:
15361 case PPC::BI__builtin_altivec_vctzd: {
15362 llvm::Type *ResultType = ConvertType(E->getType());
15363 Value *X = EmitScalarExpr(E->getArg(0));
15364 Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
15365 Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType);
15366 return Builder.CreateCall(F, {X, Undef});
15367 }
15368 case PPC::BI__builtin_altivec_vec_replace_elt:
15369 case PPC::BI__builtin_altivec_vec_replace_unaligned: {
15370 // The third argument of vec_replace_elt and vec_replace_unaligned must
15371 // be a compile time constant and will be emitted either to the vinsw
15372 // or vinsd instruction.
15373 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
15374 assert(ArgCI &&(static_cast <bool> (ArgCI && "Third Arg to vinsw/vinsd intrinsic must be a constant integer!"
) ? void (0) : __assert_fail ("ArgCI && \"Third Arg to vinsw/vinsd intrinsic must be a constant integer!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15375, __extension__ __PRETTY_FUNCTION__))
15375 "Third Arg to vinsw/vinsd intrinsic must be a constant integer!")(static_cast <bool> (ArgCI && "Third Arg to vinsw/vinsd intrinsic must be a constant integer!"
) ? void (0) : __assert_fail ("ArgCI && \"Third Arg to vinsw/vinsd intrinsic must be a constant integer!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15375, __extension__ __PRETTY_FUNCTION__))
;
15376 llvm::Type *ResultType = ConvertType(E->getType());
15377 llvm::Function *F = nullptr;
15378 Value *Call = nullptr;
15379 int64_t ConstArg = ArgCI->getSExtValue();
15380 unsigned ArgWidth = Ops[1]->getType()->getPrimitiveSizeInBits();
15381 bool Is32Bit = false;
15382 assert((ArgWidth == 32 || ArgWidth == 64) && "Invalid argument width")(static_cast <bool> ((ArgWidth == 32 || ArgWidth == 64)
&& "Invalid argument width") ? void (0) : __assert_fail
("(ArgWidth == 32 || ArgWidth == 64) && \"Invalid argument width\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15382, __extension__ __PRETTY_FUNCTION__))
;
15383 // The input to vec_replace_elt is an element index, not a byte index.
15384 if (BuiltinID == PPC::BI__builtin_altivec_vec_replace_elt)
15385 ConstArg *= ArgWidth / 8;
15386 if (ArgWidth == 32) {
15387 Is32Bit = true;
15388 // When the second argument is 32 bits, it can either be an integer or
15389 // a float. The vinsw intrinsic is used in this case.
15390 F = CGM.getIntrinsic(Intrinsic::ppc_altivec_vinsw);
15391 // Fix the constant according to endianess.
15392 if (getTarget().isLittleEndian())
15393 ConstArg = 12 - ConstArg;
15394 } else {
15395 // When the second argument is 64 bits, it can either be a long long or
15396 // a double. The vinsd intrinsic is used in this case.
15397 F = CGM.getIntrinsic(Intrinsic::ppc_altivec_vinsd);
15398 // Fix the constant for little endian.
15399 if (getTarget().isLittleEndian())
15400 ConstArg = 8 - ConstArg;
15401 }
15402 Ops[2] = ConstantInt::getSigned(Int32Ty, ConstArg);
15403 // Depending on ArgWidth, the input vector could be a float or a double.
15404 // If the input vector is a float type, bitcast the inputs to integers. Or,
15405 // if the input vector is a double, bitcast the inputs to 64-bit integers.
15406 if (!Ops[1]->getType()->isIntegerTy(ArgWidth)) {
15407 Ops[0] = Builder.CreateBitCast(
15408 Ops[0], Is32Bit ? llvm::FixedVectorType::get(Int32Ty, 4)
15409 : llvm::FixedVectorType::get(Int64Ty, 2));
15410 Ops[1] = Builder.CreateBitCast(Ops[1], Is32Bit ? Int32Ty : Int64Ty);
15411 }
15412 // Emit the call to vinsw or vinsd.
15413 Call = Builder.CreateCall(F, Ops);
15414 // Depending on the builtin, bitcast to the approriate result type.
15415 if (BuiltinID == PPC::BI__builtin_altivec_vec_replace_elt &&
15416 !Ops[1]->getType()->isIntegerTy())
15417 return Builder.CreateBitCast(Call, ResultType);
15418 else if (BuiltinID == PPC::BI__builtin_altivec_vec_replace_elt &&
15419 Ops[1]->getType()->isIntegerTy())
15420 return Call;
15421 else
15422 return Builder.CreateBitCast(Call,
15423 llvm::FixedVectorType::get(Int8Ty, 16));
15424 }
15425 case PPC::BI__builtin_altivec_vpopcntb:
15426 case PPC::BI__builtin_altivec_vpopcnth:
15427 case PPC::BI__builtin_altivec_vpopcntw:
15428 case PPC::BI__builtin_altivec_vpopcntd: {
15429 llvm::Type *ResultType = ConvertType(E->getType());
15430 Value *X = EmitScalarExpr(E->getArg(0));
15431 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
15432 return Builder.CreateCall(F, X);
15433 }
15434 case PPC::BI__builtin_altivec_vadduqm:
15435 case PPC::BI__builtin_altivec_vsubuqm: {
15436 llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128);
15437 Ops[0] =
15438 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int128Ty, 1));
15439 Ops[1] =
15440 Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(Int128Ty, 1));
15441 if (BuiltinID == PPC::BI__builtin_altivec_vadduqm)
15442 return Builder.CreateAdd(Ops[0], Ops[1], "vadduqm");
15443 else
15444 return Builder.CreateSub(Ops[0], Ops[1], "vsubuqm");
15445 }
15446 // Rotate and insert under mask operation.
15447 // __rldimi(rs, is, shift, mask)
15448 // (rotl64(rs, shift) & mask) | (is & ~mask)
15449 // __rlwimi(rs, is, shift, mask)
15450 // (rotl(rs, shift) & mask) | (is & ~mask)
15451 case PPC::BI__builtin_ppc_rldimi:
15452 case PPC::BI__builtin_ppc_rlwimi: {
15453 llvm::Type *Ty = Ops[0]->getType();
15454 Function *F = CGM.getIntrinsic(Intrinsic::fshl, Ty);
15455 if (BuiltinID == PPC::BI__builtin_ppc_rldimi)
15456 Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty);
15457 Value *Shift = Builder.CreateCall(F, {Ops[0], Ops[0], Ops[2]});
15458 Value *X = Builder.CreateAnd(Shift, Ops[3]);
15459 Value *Y = Builder.CreateAnd(Ops[1], Builder.CreateNot(Ops[3]));
15460 return Builder.CreateOr(X, Y);
15461 }
15462 // Rotate and insert under mask operation.
15463 // __rlwnm(rs, shift, mask)
15464 // rotl(rs, shift) & mask
15465 case PPC::BI__builtin_ppc_rlwnm: {
15466 llvm::Type *Ty = Ops[0]->getType();
15467 Function *F = CGM.getIntrinsic(Intrinsic::fshl, Ty);
15468 Value *Shift = Builder.CreateCall(F, {Ops[0], Ops[0], Ops[1]});
15469 return Builder.CreateAnd(Shift, Ops[2]);
15470 }
15471 case PPC::BI__builtin_ppc_poppar4:
15472 case PPC::BI__builtin_ppc_poppar8: {
15473 llvm::Type *ArgType = Ops[0]->getType();
15474 Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType);
15475 Value *Tmp = Builder.CreateCall(F, Ops[0]);
15476
15477 llvm::Type *ResultType = ConvertType(E->getType());
15478 Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1));
15479 if (Result->getType() != ResultType)
15480 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
15481 "cast");
15482 return Result;
15483 }
15484 case PPC::BI__builtin_ppc_cmpb: {
15485 if (getTarget().getTriple().isPPC64()) {
15486 Function *F =
15487 CGM.getIntrinsic(Intrinsic::ppc_cmpb, {Int64Ty, Int64Ty, Int64Ty});
15488 return Builder.CreateCall(F, Ops, "cmpb");
15489 }
15490 // For 32 bit, emit the code as below:
15491 // %conv = trunc i64 %a to i32
15492 // %conv1 = trunc i64 %b to i32
15493 // %shr = lshr i64 %a, 32
15494 // %conv2 = trunc i64 %shr to i32
15495 // %shr3 = lshr i64 %b, 32
15496 // %conv4 = trunc i64 %shr3 to i32
15497 // %0 = tail call i32 @llvm.ppc.cmpb32(i32 %conv, i32 %conv1)
15498 // %conv5 = zext i32 %0 to i64
15499 // %1 = tail call i32 @llvm.ppc.cmpb32(i32 %conv2, i32 %conv4)
15500 // %conv614 = zext i32 %1 to i64
15501 // %shl = shl nuw i64 %conv614, 32
15502 // %or = or i64 %shl, %conv5
15503 // ret i64 %or
15504 Function *F =
15505 CGM.getIntrinsic(Intrinsic::ppc_cmpb, {Int32Ty, Int32Ty, Int32Ty});
15506 Value *ArgOneLo = Builder.CreateTrunc(Ops[0], Int32Ty);
15507 Value *ArgTwoLo = Builder.CreateTrunc(Ops[1], Int32Ty);
15508 Constant *ShiftAmt = ConstantInt::get(Int64Ty, 32);
15509 Value *ArgOneHi =
15510 Builder.CreateTrunc(Builder.CreateLShr(Ops[0], ShiftAmt), Int32Ty);
15511 Value *ArgTwoHi =
15512 Builder.CreateTrunc(Builder.CreateLShr(Ops[1], ShiftAmt), Int32Ty);
15513 Value *ResLo = Builder.CreateZExt(
15514 Builder.CreateCall(F, {ArgOneLo, ArgTwoLo}, "cmpb"), Int64Ty);
15515 Value *ResHiShift = Builder.CreateZExt(
15516 Builder.CreateCall(F, {ArgOneHi, ArgTwoHi}, "cmpb"), Int64Ty);
15517 Value *ResHi = Builder.CreateShl(ResHiShift, ShiftAmt);
15518 return Builder.CreateOr(ResLo, ResHi);
15519 }
15520 // Copy sign
15521 case PPC::BI__builtin_vsx_xvcpsgnsp:
15522 case PPC::BI__builtin_vsx_xvcpsgndp: {
15523 llvm::Type *ResultType = ConvertType(E->getType());
15524 Value *X = EmitScalarExpr(E->getArg(0));
15525 Value *Y = EmitScalarExpr(E->getArg(1));
15526 ID = Intrinsic::copysign;
15527 llvm::Function *F = CGM.getIntrinsic(ID, ResultType);
15528 return Builder.CreateCall(F, {X, Y});
15529 }
15530 // Rounding/truncation
15531 case PPC::BI__builtin_vsx_xvrspip:
15532 case PPC::BI__builtin_vsx_xvrdpip:
15533 case PPC::BI__builtin_vsx_xvrdpim:
15534 case PPC::BI__builtin_vsx_xvrspim:
15535 case PPC::BI__builtin_vsx_xvrdpi:
15536 case PPC::BI__builtin_vsx_xvrspi:
15537 case PPC::BI__builtin_vsx_xvrdpic:
15538 case PPC::BI__builtin_vsx_xvrspic:
15539 case PPC::BI__builtin_vsx_xvrdpiz:
15540 case PPC::BI__builtin_vsx_xvrspiz: {
15541 llvm::Type *ResultType = ConvertType(E->getType());
15542 Value *X = EmitScalarExpr(E->getArg(0));
15543 if (BuiltinID == PPC::BI__builtin_vsx_xvrdpim ||
15544 BuiltinID == PPC::BI__builtin_vsx_xvrspim)
15545 ID = Builder.getIsFPConstrained()
15546 ? Intrinsic::experimental_constrained_floor
15547 : Intrinsic::floor;
15548 else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpi ||
15549 BuiltinID == PPC::BI__builtin_vsx_xvrspi)
15550 ID = Builder.getIsFPConstrained()
15551 ? Intrinsic::experimental_constrained_round
15552 : Intrinsic::round;
15553 else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpic ||
15554 BuiltinID == PPC::BI__builtin_vsx_xvrspic)
15555 ID = Builder.getIsFPConstrained()
15556 ? Intrinsic::experimental_constrained_rint
15557 : Intrinsic::rint;
15558 else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpip ||
15559 BuiltinID == PPC::BI__builtin_vsx_xvrspip)
15560 ID = Builder.getIsFPConstrained()
15561 ? Intrinsic::experimental_constrained_ceil
15562 : Intrinsic::ceil;
15563 else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpiz ||
15564 BuiltinID == PPC::BI__builtin_vsx_xvrspiz)
15565 ID = Builder.getIsFPConstrained()
15566 ? Intrinsic::experimental_constrained_trunc
15567 : Intrinsic::trunc;
15568 llvm::Function *F = CGM.getIntrinsic(ID, ResultType);
15569 return Builder.getIsFPConstrained() ? Builder.CreateConstrainedFPCall(F, X)
15570 : Builder.CreateCall(F, X);
15571 }
15572
15573 // Absolute value
15574 case PPC::BI__builtin_vsx_xvabsdp:
15575 case PPC::BI__builtin_vsx_xvabssp: {
15576 llvm::Type *ResultType = ConvertType(E->getType());
15577 Value *X = EmitScalarExpr(E->getArg(0));
15578 llvm::Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
15579 return Builder.CreateCall(F, X);
15580 }
15581
15582 // Fastmath by default
15583 case PPC::BI__builtin_ppc_recipdivf:
15584 case PPC::BI__builtin_ppc_recipdivd:
15585 case PPC::BI__builtin_ppc_rsqrtf:
15586 case PPC::BI__builtin_ppc_rsqrtd: {
15587 FastMathFlags FMF = Builder.getFastMathFlags();
15588 Builder.getFastMathFlags().setFast();
15589 llvm::Type *ResultType = ConvertType(E->getType());
15590 Value *X = EmitScalarExpr(E->getArg(0));
15591
15592 if (BuiltinID == PPC::BI__builtin_ppc_recipdivf ||
15593 BuiltinID == PPC::BI__builtin_ppc_recipdivd) {
15594 Value *Y = EmitScalarExpr(E->getArg(1));
15595 Value *FDiv = Builder.CreateFDiv(X, Y, "recipdiv");
15596 Builder.getFastMathFlags() &= (FMF);
15597 return FDiv;
15598 }
15599 auto *One = ConstantFP::get(ResultType, 1.0);
15600 llvm::Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType);
15601 Value *FDiv = Builder.CreateFDiv(One, Builder.CreateCall(F, X), "rsqrt");
15602 Builder.getFastMathFlags() &= (FMF);
15603 return FDiv;
15604 }
15605 case PPC::BI__builtin_ppc_alignx: {
15606 ConstantInt *AlignmentCI = cast<ConstantInt>(Ops[0]);
15607 if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment))
15608 AlignmentCI = ConstantInt::get(AlignmentCI->getType(),
15609 llvm::Value::MaximumAlignment);
15610
15611 emitAlignmentAssumption(Ops[1], E->getArg(1),
15612 /*The expr loc is sufficient.*/ SourceLocation(),
15613 AlignmentCI, nullptr);
15614 return Ops[1];
15615 }
15616 case PPC::BI__builtin_ppc_rdlam: {
15617 llvm::Type *Ty = Ops[0]->getType();
15618 Value *ShiftAmt = Builder.CreateIntCast(Ops[1], Ty, false);
15619 Function *F = CGM.getIntrinsic(Intrinsic::fshl, Ty);
15620 Value *Rotate = Builder.CreateCall(F, {Ops[0], Ops[0], ShiftAmt});
15621 return Builder.CreateAnd(Rotate, Ops[2]);
15622 }
15623 // FMA variations
15624 case PPC::BI__builtin_vsx_xvmaddadp:
15625 case PPC::BI__builtin_vsx_xvmaddasp:
15626 case PPC::BI__builtin_vsx_xvnmaddadp:
15627 case PPC::BI__builtin_vsx_xvnmaddasp:
15628 case PPC::BI__builtin_vsx_xvmsubadp:
15629 case PPC::BI__builtin_vsx_xvmsubasp:
15630 case PPC::BI__builtin_vsx_xvnmsubadp:
15631 case PPC::BI__builtin_vsx_xvnmsubasp: {
15632 llvm::Type *ResultType = ConvertType(E->getType());
15633 Value *X = EmitScalarExpr(E->getArg(0));
15634 Value *Y = EmitScalarExpr(E->getArg(1));
15635 Value *Z = EmitScalarExpr(E->getArg(2));
15636 llvm::Function *F;
15637 if (Builder.getIsFPConstrained())
15638 F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType);
15639 else
15640 F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
15641 switch (BuiltinID) {
15642 case PPC::BI__builtin_vsx_xvmaddadp:
15643 case PPC::BI__builtin_vsx_xvmaddasp:
15644 if (Builder.getIsFPConstrained())
15645 return Builder.CreateConstrainedFPCall(F, {X, Y, Z});
15646 else
15647 return Builder.CreateCall(F, {X, Y, Z});
15648 case PPC::BI__builtin_vsx_xvnmaddadp:
15649 case PPC::BI__builtin_vsx_xvnmaddasp:
15650 if (Builder.getIsFPConstrained())
15651 return Builder.CreateFNeg(
15652 Builder.CreateConstrainedFPCall(F, {X, Y, Z}), "neg");
15653 else
15654 return Builder.CreateFNeg(Builder.CreateCall(F, {X, Y, Z}), "neg");
15655 case PPC::BI__builtin_vsx_xvmsubadp:
15656 case PPC::BI__builtin_vsx_xvmsubasp:
15657 if (Builder.getIsFPConstrained())
15658 return Builder.CreateConstrainedFPCall(
15659 F, {X, Y, Builder.CreateFNeg(Z, "neg")});
15660 else
15661 return Builder.CreateCall(F, {X, Y, Builder.CreateFNeg(Z, "neg")});
15662 case PPC::BI__builtin_vsx_xvnmsubadp:
15663 case PPC::BI__builtin_vsx_xvnmsubasp:
15664 if (Builder.getIsFPConstrained())
15665 return Builder.CreateFNeg(
15666 Builder.CreateConstrainedFPCall(
15667 F, {X, Y, Builder.CreateFNeg(Z, "neg")}),
15668 "neg");
15669 else
15670 return Builder.CreateFNeg(
15671 Builder.CreateCall(F, {X, Y, Builder.CreateFNeg(Z, "neg")}),
15672 "neg");
15673 }
15674 llvm_unreachable("Unknown FMA operation")::llvm::llvm_unreachable_internal("Unknown FMA operation", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15674)
;
15675 return nullptr; // Suppress no-return warning
15676 }
15677
15678 case PPC::BI__builtin_vsx_insertword: {
15679 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxinsertw);
15680
15681 // Third argument is a compile time constant int. It must be clamped to
15682 // to the range [0, 12].
15683 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
15684 assert(ArgCI &&(static_cast <bool> (ArgCI && "Third arg to xxinsertw intrinsic must be constant integer"
) ? void (0) : __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15685, __extension__ __PRETTY_FUNCTION__))
15685 "Third arg to xxinsertw intrinsic must be constant integer")(static_cast <bool> (ArgCI && "Third arg to xxinsertw intrinsic must be constant integer"
) ? void (0) : __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15685, __extension__ __PRETTY_FUNCTION__))
;
15686 const int64_t MaxIndex = 12;
15687 int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex);
15688
15689 // The builtin semantics don't exactly match the xxinsertw instructions
15690 // semantics (which ppc_vsx_xxinsertw follows). The builtin extracts the
15691 // word from the first argument, and inserts it in the second argument. The
15692 // instruction extracts the word from its second input register and inserts
15693 // it into its first input register, so swap the first and second arguments.
15694 std::swap(Ops[0], Ops[1]);
15695
15696 // Need to cast the second argument from a vector of unsigned int to a
15697 // vector of long long.
15698 Ops[1] =
15699 Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(Int64Ty, 2));
15700
15701 if (getTarget().isLittleEndian()) {
15702 // Reverse the double words in the vector we will extract from.
15703 Ops[0] =
15704 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 2));
15705 Ops[0] = Builder.CreateShuffleVector(Ops[0], Ops[0], ArrayRef<int>{1, 0});
15706
15707 // Reverse the index.
15708 Index = MaxIndex - Index;
15709 }
15710
15711 // Intrinsic expects the first arg to be a vector of int.
15712 Ops[0] =
15713 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int32Ty, 4));
15714 Ops[2] = ConstantInt::getSigned(Int32Ty, Index);
15715 return Builder.CreateCall(F, Ops);
15716 }
15717
15718 case PPC::BI__builtin_vsx_extractuword: {
15719 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxextractuw);
15720
15721 // Intrinsic expects the first argument to be a vector of doublewords.
15722 Ops[0] =
15723 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 2));
15724
15725 // The second argument is a compile time constant int that needs to
15726 // be clamped to the range [0, 12].
15727 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[1]);
15728 assert(ArgCI &&(static_cast <bool> (ArgCI && "Second Arg to xxextractuw intrinsic must be a constant integer!"
) ? void (0) : __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15729, __extension__ __PRETTY_FUNCTION__))
15729 "Second Arg to xxextractuw intrinsic must be a constant integer!")(static_cast <bool> (ArgCI && "Second Arg to xxextractuw intrinsic must be a constant integer!"
) ? void (0) : __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15729, __extension__ __PRETTY_FUNCTION__))
;
15730 const int64_t MaxIndex = 12;
15731 int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex);
15732
15733 if (getTarget().isLittleEndian()) {
15734 // Reverse the index.
15735 Index = MaxIndex - Index;
15736 Ops[1] = ConstantInt::getSigned(Int32Ty, Index);
15737
15738 // Emit the call, then reverse the double words of the results vector.
15739 Value *Call = Builder.CreateCall(F, Ops);
15740
15741 Value *ShuffleCall =
15742 Builder.CreateShuffleVector(Call, Call, ArrayRef<int>{1, 0});
15743 return ShuffleCall;
15744 } else {
15745 Ops[1] = ConstantInt::getSigned(Int32Ty, Index);
15746 return Builder.CreateCall(F, Ops);
15747 }
15748 }
15749
15750 case PPC::BI__builtin_vsx_xxpermdi: {
15751 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
15752 assert(ArgCI && "Third arg must be constant integer!")(static_cast <bool> (ArgCI && "Third arg must be constant integer!"
) ? void (0) : __assert_fail ("ArgCI && \"Third arg must be constant integer!\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15752, __extension__ __PRETTY_FUNCTION__))
;
15753
15754 unsigned Index = ArgCI->getZExtValue();
15755 Ops[0] =
15756 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 2));
15757 Ops[1] =
15758 Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(Int64Ty, 2));
15759
15760 // Account for endianness by treating this as just a shuffle. So we use the
15761 // same indices for both LE and BE in order to produce expected results in
15762 // both cases.
15763 int ElemIdx0 = (Index & 2) >> 1;
15764 int ElemIdx1 = 2 + (Index & 1);
15765
15766 int ShuffleElts[2] = {ElemIdx0, ElemIdx1};
15767 Value *ShuffleCall =
15768 Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleElts);
15769 QualType BIRetType = E->getType();
15770 auto RetTy = ConvertType(BIRetType);
15771 return Builder.CreateBitCast(ShuffleCall, RetTy);
15772 }
15773
15774 case PPC::BI__builtin_vsx_xxsldwi: {
15775 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
15776 assert(ArgCI && "Third argument must be a compile time constant")(static_cast <bool> (ArgCI && "Third argument must be a compile time constant"
) ? void (0) : __assert_fail ("ArgCI && \"Third argument must be a compile time constant\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 15776, __extension__ __PRETTY_FUNCTION__))
;
15777 unsigned Index = ArgCI->getZExtValue() & 0x3;
15778 Ops[0] =
15779 Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int32Ty, 4));
15780 Ops[1] =
15781 Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(Int32Ty, 4));
15782
15783 // Create a shuffle mask
15784 int ElemIdx0;
15785 int ElemIdx1;
15786 int ElemIdx2;
15787 int ElemIdx3;
15788 if (getTarget().isLittleEndian()) {
15789 // Little endian element N comes from element 8+N-Index of the
15790 // concatenated wide vector (of course, using modulo arithmetic on
15791 // the total number of elements).
15792 ElemIdx0 = (8 - Index) % 8;
15793 ElemIdx1 = (9 - Index) % 8;
15794 ElemIdx2 = (10 - Index) % 8;
15795 ElemIdx3 = (11 - Index) % 8;
15796 } else {
15797 // Big endian ElemIdx<N> = Index + N
15798 ElemIdx0 = Index;
15799 ElemIdx1 = Index + 1;
15800 ElemIdx2 = Index + 2;
15801 ElemIdx3 = Index + 3;
15802 }
15803
15804 int ShuffleElts[4] = {ElemIdx0, ElemIdx1, ElemIdx2, ElemIdx3};
15805 Value *ShuffleCall =
15806 Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleElts);
15807 QualType BIRetType = E->getType();
15808 auto RetTy = ConvertType(BIRetType);
15809 return Builder.CreateBitCast(ShuffleCall, RetTy);
15810 }
15811
15812 case PPC::BI__builtin_pack_vector_int128: {
15813 bool isLittleEndian = getTarget().isLittleEndian();
15814 Value *UndefValue =
15815 llvm::UndefValue::get(llvm::FixedVectorType::get(Ops[0]->getType(), 2));
15816 Value *Res = Builder.CreateInsertElement(
15817 UndefValue, Ops[0], (uint64_t)(isLittleEndian ? 1 : 0));
15818 Res = Builder.CreateInsertElement(Res, Ops[1],
15819 (uint64_t)(isLittleEndian ? 0 : 1));
15820 return Builder.CreateBitCast(Res, ConvertType(E->getType()));
15821 }
15822
15823 case PPC::BI__builtin_unpack_vector_int128: {
15824 ConstantInt *Index = cast<ConstantInt>(Ops[1]);
15825 Value *Unpacked = Builder.CreateBitCast(
15826 Ops[0], llvm::FixedVectorType::get(ConvertType(E->getType()), 2));
15827
15828 if (getTarget().isLittleEndian())
15829 Index = ConstantInt::get(Index->getType(), 1 - Index->getZExtValue());
15830
15831 return Builder.CreateExtractElement(Unpacked, Index);
15832 }
15833
15834 case PPC::BI__builtin_ppc_sthcx: {
15835 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_sthcx);
15836 Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy);
15837 Ops[1] = Builder.CreateSExt(Ops[1], Int32Ty);
15838 return Builder.CreateCall(F, Ops);
15839 }
15840
15841 // The PPC MMA builtins take a pointer to a __vector_quad as an argument.
15842 // Some of the MMA instructions accumulate their result into an existing
15843 // accumulator whereas the others generate a new accumulator. So we need to
15844 // use custom code generation to expand a builtin call with a pointer to a
15845 // load (if the corresponding instruction accumulates its result) followed by
15846 // the call to the intrinsic and a store of the result.
15847#define CUSTOM_BUILTIN(Name, Intr, Types, Accumulate) \
15848 case PPC::BI__builtin_##Name:
15849#include "clang/Basic/BuiltinsPPC.def"
15850 {
15851 // The first argument of these two builtins is a pointer used to store their
15852 // result. However, the llvm intrinsics return their result in multiple
15853 // return values. So, here we emit code extracting these values from the
15854 // intrinsic results and storing them using that pointer.
15855 if (BuiltinID == PPC::BI__builtin_mma_disassemble_acc ||
15856 BuiltinID == PPC::BI__builtin_vsx_disassemble_pair ||
15857 BuiltinID == PPC::BI__builtin_mma_disassemble_pair) {
15858 unsigned NumVecs = 2;
15859 auto Intrinsic = Intrinsic::ppc_vsx_disassemble_pair;
15860 if (BuiltinID == PPC::BI__builtin_mma_disassemble_acc) {
15861 NumVecs = 4;
15862 Intrinsic = Intrinsic::ppc_mma_disassemble_acc;
15863 }
15864 llvm::Function *F = CGM.getIntrinsic(Intrinsic);
15865 Address Addr = EmitPointerWithAlignment(E->getArg(1));
15866 Value *Vec = Builder.CreateLoad(Addr);
15867 Value *Call = Builder.CreateCall(F, {Vec});
15868 llvm::Type *VTy = llvm::FixedVectorType::get(Int8Ty, 16);
15869 Value *Ptr = Builder.CreateBitCast(Ops[0], VTy->getPointerTo());
15870 for (unsigned i=0; i<NumVecs; i++) {
15871 Value *Vec = Builder.CreateExtractValue(Call, i);
15872 llvm::ConstantInt* Index = llvm::ConstantInt::get(IntTy, i);
15873 Value *GEP = Builder.CreateInBoundsGEP(VTy, Ptr, Index);
15874 Builder.CreateAlignedStore(Vec, GEP, MaybeAlign(16));
15875 }
15876 return Call;
15877 }
15878 bool Accumulate;
15879 switch (BuiltinID) {
15880 #define CUSTOM_BUILTIN(Name, Intr, Types, Acc) \
15881 case PPC::BI__builtin_##Name: \
15882 ID = Intrinsic::ppc_##Intr; \
15883 Accumulate = Acc; \
15884 break;
15885 #include "clang/Basic/BuiltinsPPC.def"
15886 }
15887 if (BuiltinID == PPC::BI__builtin_vsx_lxvp ||
15888 BuiltinID == PPC::BI__builtin_vsx_stxvp ||
15889 BuiltinID == PPC::BI__builtin_mma_lxvp ||
15890 BuiltinID == PPC::BI__builtin_mma_stxvp) {
15891 if (BuiltinID == PPC::BI__builtin_vsx_lxvp ||
15892 BuiltinID == PPC::BI__builtin_mma_lxvp) {
15893 Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy);
15894 Ops[0] = Builder.CreateGEP(Int8Ty, Ops[1], Ops[0]);
15895 } else {
15896 Ops[2] = Builder.CreateBitCast(Ops[2], Int8PtrTy);
15897 Ops[1] = Builder.CreateGEP(Int8Ty, Ops[2], Ops[1]);
15898 }
15899 Ops.pop_back();
15900 llvm::Function *F = CGM.getIntrinsic(ID);
15901 return Builder.CreateCall(F, Ops, "");
15902 }
15903 SmallVector<Value*, 4> CallOps;
15904 if (Accumulate) {
15905 Address Addr = EmitPointerWithAlignment(E->getArg(0));
15906 Value *Acc = Builder.CreateLoad(Addr);
15907 CallOps.push_back(Acc);
15908 }
15909 for (unsigned i=1; i<Ops.size(); i++)
15910 CallOps.push_back(Ops[i]);
15911 llvm::Function *F = CGM.getIntrinsic(ID);
15912 Value *Call = Builder.CreateCall(F, CallOps);
15913 return Builder.CreateAlignedStore(Call, Ops[0], MaybeAlign(64));
15914 }
15915
15916 case PPC::BI__builtin_ppc_compare_and_swap:
15917 case PPC::BI__builtin_ppc_compare_and_swaplp: {
15918 Address Addr = EmitPointerWithAlignment(E->getArg(0));
15919 Address OldValAddr = EmitPointerWithAlignment(E->getArg(1));
15920 Value *OldVal = Builder.CreateLoad(OldValAddr);
15921 QualType AtomicTy = E->getArg(0)->getType()->getPointeeType();
15922 LValue LV = MakeAddrLValue(Addr, AtomicTy);
15923 auto Pair = EmitAtomicCompareExchange(
15924 LV, RValue::get(OldVal), RValue::get(Ops[2]), E->getExprLoc(),
15925 llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Monotonic, true);
15926 // Unlike c11's atomic_compare_exchange, accroding to
15927 // https://www.ibm.com/docs/en/xl-c-and-cpp-aix/16.1?topic=functions-compare-swap-compare-swaplp
15928 // > In either case, the contents of the memory location specified by addr
15929 // > are copied into the memory location specified by old_val_addr.
15930 // But it hasn't specified storing to OldValAddr is atomic or not and
15931 // which order to use. Now following XL's codegen, treat it as a normal
15932 // store.
15933 Value *LoadedVal = Pair.first.getScalarVal();
15934 Builder.CreateStore(LoadedVal, OldValAddr);
15935 return Builder.CreateZExt(Pair.second, Builder.getInt32Ty());
15936 }
15937 case PPC::BI__builtin_ppc_fetch_and_add:
15938 case PPC::BI__builtin_ppc_fetch_and_addlp: {
15939 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E,
15940 llvm::AtomicOrdering::Monotonic);
15941 }
15942 case PPC::BI__builtin_ppc_fetch_and_and:
15943 case PPC::BI__builtin_ppc_fetch_and_andlp: {
15944 return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E,
15945 llvm::AtomicOrdering::Monotonic);
15946 }
15947
15948 case PPC::BI__builtin_ppc_fetch_and_or:
15949 case PPC::BI__builtin_ppc_fetch_and_orlp: {
15950 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E,
15951 llvm::AtomicOrdering::Monotonic);
15952 }
15953 case PPC::BI__builtin_ppc_fetch_and_swap:
15954 case PPC::BI__builtin_ppc_fetch_and_swaplp: {
15955 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E,
15956 llvm::AtomicOrdering::Monotonic);
15957 }
15958 case PPC::BI__builtin_ppc_ldarx:
15959 case PPC::BI__builtin_ppc_lwarx:
15960 case PPC::BI__builtin_ppc_lharx:
15961 case PPC::BI__builtin_ppc_lbarx:
15962 return emitPPCLoadReserveIntrinsic(*this, BuiltinID, E);
15963 case PPC::BI__builtin_ppc_mfspr: {
15964 llvm::Type *RetType = CGM.getDataLayout().getTypeSizeInBits(VoidPtrTy) == 32
15965 ? Int32Ty
15966 : Int64Ty;
15967 Function *F = CGM.getIntrinsic(Intrinsic::ppc_mfspr, RetType);
15968 return Builder.CreateCall(F, Ops);
15969 }
15970 case PPC::BI__builtin_ppc_mtspr: {
15971 llvm::Type *RetType = CGM.getDataLayout().getTypeSizeInBits(VoidPtrTy) == 32
15972 ? Int32Ty
15973 : Int64Ty;
15974 Function *F = CGM.getIntrinsic(Intrinsic::ppc_mtspr, RetType);
15975 return Builder.CreateCall(F, Ops);
15976 }
15977 case PPC::BI__builtin_ppc_popcntb: {
15978 Value *ArgValue = EmitScalarExpr(E->getArg(0));
15979 llvm::Type *ArgType = ArgValue->getType();
15980 Function *F = CGM.getIntrinsic(Intrinsic::ppc_popcntb, {ArgType, ArgType});
15981 return Builder.CreateCall(F, Ops, "popcntb");
15982 }
15983 case PPC::BI__builtin_ppc_mtfsf: {
15984 // The builtin takes a uint32 that needs to be cast to an
15985 // f64 to be passed to the intrinsic.
15986 Value *Cast = Builder.CreateUIToFP(Ops[1], DoubleTy);
15987 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_mtfsf);
15988 return Builder.CreateCall(F, {Ops[0], Cast}, "");
15989 }
15990
15991 case PPC::BI__builtin_ppc_swdiv_nochk:
15992 case PPC::BI__builtin_ppc_swdivs_nochk: {
15993 FastMathFlags FMF = Builder.getFastMathFlags();
15994 Builder.getFastMathFlags().setFast();
15995 Value *FDiv = Builder.CreateFDiv(Ops[0], Ops[1], "swdiv_nochk");
15996 Builder.getFastMathFlags() &= (FMF);
15997 return FDiv;
15998 }
15999 case PPC::BI__builtin_ppc_fric:
16000 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
16001 *this, E, Intrinsic::rint,
16002 Intrinsic::experimental_constrained_rint))
16003 .getScalarVal();
16004 case PPC::BI__builtin_ppc_frim:
16005 case PPC::BI__builtin_ppc_frims:
16006 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
16007 *this, E, Intrinsic::floor,
16008 Intrinsic::experimental_constrained_floor))
16009 .getScalarVal();
16010 case PPC::BI__builtin_ppc_frin:
16011 case PPC::BI__builtin_ppc_frins:
16012 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
16013 *this, E, Intrinsic::round,
16014 Intrinsic::experimental_constrained_round))
16015 .getScalarVal();
16016 case PPC::BI__builtin_ppc_frip:
16017 case PPC::BI__builtin_ppc_frips:
16018 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
16019 *this, E, Intrinsic::ceil,
16020 Intrinsic::experimental_constrained_ceil))
16021 .getScalarVal();
16022 case PPC::BI__builtin_ppc_friz:
16023 case PPC::BI__builtin_ppc_frizs:
16024 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
16025 *this, E, Intrinsic::trunc,
16026 Intrinsic::experimental_constrained_trunc))
16027 .getScalarVal();
16028 case PPC::BI__builtin_ppc_fsqrt:
16029 case PPC::BI__builtin_ppc_fsqrts:
16030 return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
16031 *this, E, Intrinsic::sqrt,
16032 Intrinsic::experimental_constrained_sqrt))
16033 .getScalarVal();
16034 }
16035}
16036
16037namespace {
16038// If \p E is not null pointer, insert address space cast to match return
16039// type of \p E if necessary.
16040Value *EmitAMDGPUDispatchPtr(CodeGenFunction &CGF,
16041 const CallExpr *E = nullptr) {
16042 auto *F = CGF.CGM.getIntrinsic(Intrinsic::amdgcn_dispatch_ptr);
16043 auto *Call = CGF.Builder.CreateCall(F);
16044 Call->addRetAttr(
16045 Attribute::getWithDereferenceableBytes(Call->getContext(), 64));
16046 Call->addRetAttr(Attribute::getWithAlignment(Call->getContext(), Align(4)));
16047 if (!E)
16048 return Call;
16049 QualType BuiltinRetType = E->getType();
16050 auto *RetTy = cast<llvm::PointerType>(CGF.ConvertType(BuiltinRetType));
16051 if (RetTy == Call->getType())
16052 return Call;
16053 return CGF.Builder.CreateAddrSpaceCast(Call, RetTy);
16054}
16055
16056// \p Index is 0, 1, and 2 for x, y, and z dimension, respectively.
16057Value *EmitAMDGPUWorkGroupSize(CodeGenFunction &CGF, unsigned Index) {
16058 const unsigned XOffset = 4;
16059 auto *DP = EmitAMDGPUDispatchPtr(CGF);
16060 // Indexing the HSA kernel_dispatch_packet struct.
16061 auto *Offset = llvm::ConstantInt::get(CGF.Int32Ty, XOffset + Index * 2);
16062 auto *GEP = CGF.Builder.CreateGEP(CGF.Int8Ty, DP, Offset);
16063 auto *DstTy =
16064 CGF.Int16Ty->getPointerTo(GEP->getType()->getPointerAddressSpace());
16065 auto *Cast = CGF.Builder.CreateBitCast(GEP, DstTy);
16066 auto *LD = CGF.Builder.CreateLoad(Address(Cast, CharUnits::fromQuantity(2)));
16067 llvm::MDBuilder MDHelper(CGF.getLLVMContext());
16068 llvm::MDNode *RNode = MDHelper.createRange(APInt(16, 1),
16069 APInt(16, CGF.getTarget().getMaxOpenCLWorkGroupSize() + 1));
16070 LD->setMetadata(llvm::LLVMContext::MD_range, RNode);
16071 LD->setMetadata(llvm::LLVMContext::MD_invariant_load,
16072 llvm::MDNode::get(CGF.getLLVMContext(), None));
16073 return LD;
16074}
16075
16076// \p Index is 0, 1, and 2 for x, y, and z dimension, respectively.
16077Value *EmitAMDGPUGridSize(CodeGenFunction &CGF, unsigned Index) {
16078 const unsigned XOffset = 12;
16079 auto *DP = EmitAMDGPUDispatchPtr(CGF);
16080 // Indexing the HSA kernel_dispatch_packet struct.
16081 auto *Offset = llvm::ConstantInt::get(CGF.Int32Ty, XOffset + Index * 4);
16082 auto *GEP = CGF.Builder.CreateGEP(CGF.Int8Ty, DP, Offset);
16083 auto *DstTy =
16084 CGF.Int32Ty->getPointerTo(GEP->getType()->getPointerAddressSpace());
16085 auto *Cast = CGF.Builder.CreateBitCast(GEP, DstTy);
16086 auto *LD = CGF.Builder.CreateLoad(Address(Cast, CharUnits::fromQuantity(4)));
16087 LD->setMetadata(llvm::LLVMContext::MD_invariant_load,
16088 llvm::MDNode::get(CGF.getLLVMContext(), None));
16089 return LD;
16090}
16091} // namespace
16092
16093// For processing memory ordering and memory scope arguments of various
16094// amdgcn builtins.
16095// \p Order takes a C++11 comptabile memory-ordering specifier and converts
16096// it into LLVM's memory ordering specifier using atomic C ABI, and writes
16097// to \p AO. \p Scope takes a const char * and converts it into AMDGCN
16098// specific SyncScopeID and writes it to \p SSID.
16099bool CodeGenFunction::ProcessOrderScopeAMDGCN(Value *Order, Value *Scope,
16100 llvm::AtomicOrdering &AO,
16101 llvm::SyncScope::ID &SSID) {
16102 if (isa<llvm::ConstantInt>(Order)) {
16103 int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
16104
16105 // Map C11/C++11 memory ordering to LLVM memory ordering
16106 assert(llvm::isValidAtomicOrderingCABI(ord))(static_cast <bool> (llvm::isValidAtomicOrderingCABI(ord
)) ? void (0) : __assert_fail ("llvm::isValidAtomicOrderingCABI(ord)"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 16106, __extension__ __PRETTY_FUNCTION__))
;
16107 switch (static_cast<llvm::AtomicOrderingCABI>(ord)) {
16108 case llvm::AtomicOrderingCABI::acquire:
16109 case llvm::AtomicOrderingCABI::consume:
16110 AO = llvm::AtomicOrdering::Acquire;
16111 break;
16112 case llvm::AtomicOrderingCABI::release:
16113 AO = llvm::AtomicOrdering::Release;
16114 break;
16115 case llvm::AtomicOrderingCABI::acq_rel:
16116 AO = llvm::AtomicOrdering::AcquireRelease;
16117 break;
16118 case llvm::AtomicOrderingCABI::seq_cst:
16119 AO = llvm::AtomicOrdering::SequentiallyConsistent;
16120 break;
16121 case llvm::AtomicOrderingCABI::relaxed:
16122 AO = llvm::AtomicOrdering::Monotonic;
16123 break;
16124 }
16125
16126 StringRef scp;
16127 llvm::getConstantStringInfo(Scope, scp);
16128 SSID = getLLVMContext().getOrInsertSyncScopeID(scp);
16129 return true;
16130 }
16131 return false;
16132}
16133
16134Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
16135 const CallExpr *E) {
16136 llvm::AtomicOrdering AO = llvm::AtomicOrdering::SequentiallyConsistent;
16137 llvm::SyncScope::ID SSID;
16138 switch (BuiltinID) {
16139 case AMDGPU::BI__builtin_amdgcn_div_scale:
16140 case AMDGPU::BI__builtin_amdgcn_div_scalef: {
16141 // Translate from the intrinsics's struct return to the builtin's out
16142 // argument.
16143
16144 Address FlagOutPtr = EmitPointerWithAlignment(E->getArg(3));
16145
16146 llvm::Value *X = EmitScalarExpr(E->getArg(0));
16147 llvm::Value *Y = EmitScalarExpr(E->getArg(1));
16148 llvm::Value *Z = EmitScalarExpr(E->getArg(2));
16149
16150 llvm::Function *Callee = CGM.getIntrinsic(Intrinsic::amdgcn_div_scale,
16151 X->getType());
16152
16153 llvm::Value *Tmp = Builder.CreateCall(Callee, {X, Y, Z});
16154
16155 llvm::Value *Result = Builder.CreateExtractValue(Tmp, 0);
16156 llvm::Value *Flag = Builder.CreateExtractValue(Tmp, 1);
16157
16158 llvm::Type *RealFlagType
16159 = FlagOutPtr.getPointer()->getType()->getPointerElementType();
16160
16161 llvm::Value *FlagExt = Builder.CreateZExt(Flag, RealFlagType);
16162 Builder.CreateStore(FlagExt, FlagOutPtr);
16163 return Result;
16164 }
16165 case AMDGPU::BI__builtin_amdgcn_div_fmas:
16166 case AMDGPU::BI__builtin_amdgcn_div_fmasf: {
16167 llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
16168 llvm::Value *Src1 = EmitScalarExpr(E->getArg(1));
16169 llvm::Value *Src2 = EmitScalarExpr(E->getArg(2));
16170 llvm::Value *Src3 = EmitScalarExpr(E->getArg(3));
16171
16172 llvm::Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_div_fmas,
16173 Src0->getType());
16174 llvm::Value *Src3ToBool = Builder.CreateIsNotNull(Src3);
16175 return Builder.CreateCall(F, {Src0, Src1, Src2, Src3ToBool});
16176 }
16177
16178 case AMDGPU::BI__builtin_amdgcn_ds_swizzle:
16179 return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_ds_swizzle);
16180 case AMDGPU::BI__builtin_amdgcn_mov_dpp8:
16181 return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_mov_dpp8);
16182 case AMDGPU::BI__builtin_amdgcn_mov_dpp:
16183 case AMDGPU::BI__builtin_amdgcn_update_dpp: {
16184 llvm::SmallVector<llvm::Value *, 6> Args;
16185 for (unsigned I = 0; I != E->getNumArgs(); ++I)
16186 Args.push_back(EmitScalarExpr(E->getArg(I)));
16187 assert(Args.size() == 5 || Args.size() == 6)(static_cast <bool> (Args.size() == 5 || Args.size() ==
6) ? void (0) : __assert_fail ("Args.size() == 5 || Args.size() == 6"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 16187, __extension__ __PRETTY_FUNCTION__))
;
16188 if (Args.size() == 5)
16189 Args.insert(Args.begin(), llvm::UndefValue::get(Args[0]->getType()));
16190 Function *F =
16191 CGM.getIntrinsic(Intrinsic::amdgcn_update_dpp, Args[0]->getType());
16192 return Builder.CreateCall(F, Args);
16193 }
16194 case AMDGPU::BI__builtin_amdgcn_div_fixup:
16195 case AMDGPU::BI__builtin_amdgcn_div_fixupf:
16196 case AMDGPU::BI__builtin_amdgcn_div_fixuph:
16197 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_div_fixup);
16198 case AMDGPU::BI__builtin_amdgcn_trig_preop:
16199 case AMDGPU::BI__builtin_amdgcn_trig_preopf:
16200 return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_trig_preop);
16201 case AMDGPU::BI__builtin_amdgcn_rcp:
16202 case AMDGPU::BI__builtin_amdgcn_rcpf:
16203 case AMDGPU::BI__builtin_amdgcn_rcph:
16204 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rcp);
16205 case AMDGPU::BI__builtin_amdgcn_sqrt:
16206 case AMDGPU::BI__builtin_amdgcn_sqrtf:
16207 case AMDGPU::BI__builtin_amdgcn_sqrth:
16208 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_sqrt);
16209 case AMDGPU::BI__builtin_amdgcn_rsq:
16210 case AMDGPU::BI__builtin_amdgcn_rsqf:
16211 case AMDGPU::BI__builtin_amdgcn_rsqh:
16212 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq);
16213 case AMDGPU::BI__builtin_amdgcn_rsq_clamp:
16214 case AMDGPU::BI__builtin_amdgcn_rsq_clampf:
16215 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq_clamp);
16216 case AMDGPU::BI__builtin_amdgcn_sinf:
16217 case AMDGPU::BI__builtin_amdgcn_sinh:
16218 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_sin);
16219 case AMDGPU::BI__builtin_amdgcn_cosf:
16220 case AMDGPU::BI__builtin_amdgcn_cosh:
16221 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_cos);
16222 case AMDGPU::BI__builtin_amdgcn_dispatch_ptr:
16223 return EmitAMDGPUDispatchPtr(*this, E);
16224 case AMDGPU::BI__builtin_amdgcn_log_clampf:
16225 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_log_clamp);
16226 case AMDGPU::BI__builtin_amdgcn_ldexp:
16227 case AMDGPU::BI__builtin_amdgcn_ldexpf:
16228 case AMDGPU::BI__builtin_amdgcn_ldexph:
16229 return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_ldexp);
16230 case AMDGPU::BI__builtin_amdgcn_frexp_mant:
16231 case AMDGPU::BI__builtin_amdgcn_frexp_mantf:
16232 case AMDGPU::BI__builtin_amdgcn_frexp_manth:
16233 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_frexp_mant);
16234 case AMDGPU::BI__builtin_amdgcn_frexp_exp:
16235 case AMDGPU::BI__builtin_amdgcn_frexp_expf: {
16236 Value *Src0 = EmitScalarExpr(E->getArg(0));
16237 Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp,
16238 { Builder.getInt32Ty(), Src0->getType() });
16239 return Builder.CreateCall(F, Src0);
16240 }
16241 case AMDGPU::BI__builtin_amdgcn_frexp_exph: {
16242 Value *Src0 = EmitScalarExpr(E->getArg(0));
16243 Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp,
16244 { Builder.getInt16Ty(), Src0->getType() });
16245 return Builder.CreateCall(F, Src0);
16246 }
16247 case AMDGPU::BI__builtin_amdgcn_fract:
16248 case AMDGPU::BI__builtin_amdgcn_fractf:
16249 case AMDGPU::BI__builtin_amdgcn_fracth:
16250 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_fract);
16251 case AMDGPU::BI__builtin_amdgcn_lerp:
16252 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_lerp);
16253 case AMDGPU::BI__builtin_amdgcn_ubfe:
16254 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_ubfe);
16255 case AMDGPU::BI__builtin_amdgcn_sbfe:
16256 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_sbfe);
16257 case AMDGPU::BI__builtin_amdgcn_uicmp:
16258 case AMDGPU::BI__builtin_amdgcn_uicmpl:
16259 case AMDGPU::BI__builtin_amdgcn_sicmp:
16260 case AMDGPU::BI__builtin_amdgcn_sicmpl: {
16261 llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
16262 llvm::Value *Src1 = EmitScalarExpr(E->getArg(1));
16263 llvm::Value *Src2 = EmitScalarExpr(E->getArg(2));
16264
16265 // FIXME-GFX10: How should 32 bit mask be handled?
16266 Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_icmp,
16267 { Builder.getInt64Ty(), Src0->getType() });
16268 return Builder.CreateCall(F, { Src0, Src1, Src2 });
16269 }
16270 case AMDGPU::BI__builtin_amdgcn_fcmp:
16271 case AMDGPU::BI__builtin_amdgcn_fcmpf: {
16272 llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
16273 llvm::Value *Src1 = EmitScalarExpr(E->getArg(1));
16274 llvm::Value *Src2 = EmitScalarExpr(E->getArg(2));
16275
16276 // FIXME-GFX10: How should 32 bit mask be handled?
16277 Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_fcmp,
16278 { Builder.getInt64Ty(), Src0->getType() });
16279 return Builder.CreateCall(F, { Src0, Src1, Src2 });
16280 }
16281 case AMDGPU::BI__builtin_amdgcn_class:
16282 case AMDGPU::BI__builtin_amdgcn_classf:
16283 case AMDGPU::BI__builtin_amdgcn_classh:
16284 return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_class);
16285 case AMDGPU::BI__builtin_amdgcn_fmed3f:
16286 case AMDGPU::BI__builtin_amdgcn_fmed3h:
16287 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fmed3);
16288 case AMDGPU::BI__builtin_amdgcn_ds_append:
16289 case AMDGPU::BI__builtin_amdgcn_ds_consume: {
16290 Intrinsic::ID Intrin = BuiltinID == AMDGPU::BI__builtin_amdgcn_ds_append ?
16291 Intrinsic::amdgcn_ds_append : Intrinsic::amdgcn_ds_consume;
16292 Value *Src0 = EmitScalarExpr(E->getArg(0));
16293 Function *F = CGM.getIntrinsic(Intrin, { Src0->getType() });
16294 return Builder.CreateCall(F, { Src0, Builder.getFalse() });
16295 }
16296 case AMDGPU::BI__builtin_amdgcn_ds_faddf:
16297 case AMDGPU::BI__builtin_amdgcn_ds_fminf:
16298 case AMDGPU::BI__builtin_amdgcn_ds_fmaxf: {
16299 Intrinsic::ID Intrin;
16300 switch (BuiltinID) {
16301 case AMDGPU::BI__builtin_amdgcn_ds_faddf:
16302 Intrin = Intrinsic::amdgcn_ds_fadd;
16303 break;
16304 case AMDGPU::BI__builtin_amdgcn_ds_fminf:
16305 Intrin = Intrinsic::amdgcn_ds_fmin;
16306 break;
16307 case AMDGPU::BI__builtin_amdgcn_ds_fmaxf:
16308 Intrin = Intrinsic::amdgcn_ds_fmax;
16309 break;
16310 }
16311 llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
16312 llvm::Value *Src1 = EmitScalarExpr(E->getArg(1));
16313 llvm::Value *Src2 = EmitScalarExpr(E->getArg(2));
16314 llvm::Value *Src3 = EmitScalarExpr(E->getArg(3));
16315 llvm::Value *Src4 = EmitScalarExpr(E->getArg(4));
16316 llvm::Function *F = CGM.getIntrinsic(Intrin, { Src1->getType() });
16317 llvm::FunctionType *FTy = F->getFunctionType();
16318 llvm::Type *PTy = FTy->getParamType(0);
16319 Src0 = Builder.CreatePointerBitCastOrAddrSpaceCast(Src0, PTy);
16320 return Builder.CreateCall(F, { Src0, Src1, Src2, Src3, Src4 });
16321 }
16322 case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64:
16323 case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f32:
16324 case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16:
16325 case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
16326 case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
16327 case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64:
16328 case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
16329 case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64: {
16330 Intrinsic::ID IID;
16331 llvm::Type *ArgTy = llvm::Type::getDoubleTy(getLLVMContext());
16332 switch (BuiltinID) {
16333 case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f32:
16334 ArgTy = llvm::Type::getFloatTy(getLLVMContext());
16335 IID = Intrinsic::amdgcn_global_atomic_fadd;
16336 break;
16337 case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16:
16338 ArgTy = llvm::FixedVectorType::get(
16339 llvm::Type::getHalfTy(getLLVMContext()), 2);
16340 IID = Intrinsic::amdgcn_global_atomic_fadd;
16341 break;
16342 case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64:
16343 IID = Intrinsic::amdgcn_global_atomic_fadd;
16344 break;
16345 case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
16346 IID = Intrinsic::amdgcn_global_atomic_fmin;
16347 break;
16348 case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
16349 IID = Intrinsic::amdgcn_global_atomic_fmax;
16350 break;
16351 case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64:
16352 IID = Intrinsic::amdgcn_flat_atomic_fadd;
16353 break;
16354 case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
16355 IID = Intrinsic::amdgcn_flat_atomic_fmin;
16356 break;
16357 case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64:
16358 IID = Intrinsic::amdgcn_flat_atomic_fmax;
16359 break;
16360 }
16361 llvm::Value *Addr = EmitScalarExpr(E->getArg(0));
16362 llvm::Value *Val = EmitScalarExpr(E->getArg(1));
16363 llvm::Function *F =
16364 CGM.getIntrinsic(IID, {ArgTy, Addr->getType(), Val->getType()});
16365 return Builder.CreateCall(F, {Addr, Val});
16366 }
16367 case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_f64:
16368 case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_f32: {
16369 Intrinsic::ID IID;
16370 llvm::Type *ArgTy;
16371 switch (BuiltinID) {
16372 case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_f32:
16373 ArgTy = llvm::Type::getFloatTy(getLLVMContext());
16374 IID = Intrinsic::amdgcn_ds_fadd;
16375 break;
16376 case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_f64:
16377 ArgTy = llvm::Type::getDoubleTy(getLLVMContext());
16378 IID = Intrinsic::amdgcn_ds_fadd;
16379 break;
16380 }
16381 llvm::Value *Addr = EmitScalarExpr(E->getArg(0));
16382 llvm::Value *Val = EmitScalarExpr(E->getArg(1));
16383 llvm::Constant *ZeroI32 = llvm::ConstantInt::getIntegerValue(
16384 llvm::Type::getInt32Ty(getLLVMContext()), APInt(32, 0, true));
16385 llvm::Constant *ZeroI1 = llvm::ConstantInt::getIntegerValue(
16386 llvm::Type::getInt1Ty(getLLVMContext()), APInt(1, 0));
16387 llvm::Function *F = CGM.getIntrinsic(IID, {ArgTy});
16388 return Builder.CreateCall(F, {Addr, Val, ZeroI32, ZeroI32, ZeroI1});
16389 }
16390 case AMDGPU::BI__builtin_amdgcn_read_exec: {
16391 CallInst *CI = cast<CallInst>(
16392 EmitSpecialRegisterBuiltin(*this, E, Int64Ty, Int64Ty, NormalRead, "exec"));
16393 CI->setConvergent();
16394 return CI;
16395 }
16396 case AMDGPU::BI__builtin_amdgcn_read_exec_lo:
16397 case AMDGPU::BI__builtin_amdgcn_read_exec_hi: {
16398 StringRef RegName = BuiltinID == AMDGPU::BI__builtin_amdgcn_read_exec_lo ?
16399 "exec_lo" : "exec_hi";
16400 CallInst *CI = cast<CallInst>(
16401 EmitSpecialRegisterBuiltin(*this, E, Int32Ty, Int32Ty, NormalRead, RegName));
16402 CI->setConvergent();
16403 return CI;
16404 }
16405 case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray:
16406 case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_h:
16407 case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_l:
16408 case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_lh: {
16409 llvm::Value *NodePtr = EmitScalarExpr(E->getArg(0));
16410 llvm::Value *RayExtent = EmitScalarExpr(E->getArg(1));
16411 llvm::Value *RayOrigin = EmitScalarExpr(E->getArg(2));
16412 llvm::Value *RayDir = EmitScalarExpr(E->getArg(3));
16413 llvm::Value *RayInverseDir = EmitScalarExpr(E->getArg(4));
16414 llvm::Value *TextureDescr = EmitScalarExpr(E->getArg(5));
16415
16416 Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_image_bvh_intersect_ray,
16417 {NodePtr->getType(), RayDir->getType()});
16418 return Builder.CreateCall(F, {NodePtr, RayExtent, RayOrigin, RayDir,
16419 RayInverseDir, TextureDescr});
16420 }
16421
16422 // amdgcn workitem
16423 case AMDGPU::BI__builtin_amdgcn_workitem_id_x:
16424 return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_x, 0, 1024);
16425 case AMDGPU::BI__builtin_amdgcn_workitem_id_y:
16426 return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_y, 0, 1024);
16427 case AMDGPU::BI__builtin_amdgcn_workitem_id_z:
16428 return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_z, 0, 1024);
16429
16430 // amdgcn workgroup size
16431 case AMDGPU::BI__builtin_amdgcn_workgroup_size_x:
16432 return EmitAMDGPUWorkGroupSize(*this, 0);
16433 case AMDGPU::BI__builtin_amdgcn_workgroup_size_y:
16434 return EmitAMDGPUWorkGroupSize(*this, 1);
16435 case AMDGPU::BI__builtin_amdgcn_workgroup_size_z:
16436 return EmitAMDGPUWorkGroupSize(*this, 2);
16437
16438 // amdgcn grid size
16439 case AMDGPU::BI__builtin_amdgcn_grid_size_x:
16440 return EmitAMDGPUGridSize(*this, 0);
16441 case AMDGPU::BI__builtin_amdgcn_grid_size_y:
16442 return EmitAMDGPUGridSize(*this, 1);
16443 case AMDGPU::BI__builtin_amdgcn_grid_size_z:
16444 return EmitAMDGPUGridSize(*this, 2);
16445
16446 // r600 intrinsics
16447 case AMDGPU::BI__builtin_r600_recipsqrt_ieee:
16448 case AMDGPU::BI__builtin_r600_recipsqrt_ieeef:
16449 return emitUnaryBuiltin(*this, E, Intrinsic::r600_recipsqrt_ieee);
16450 case AMDGPU::BI__builtin_r600_read_tidig_x:
16451 return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_x, 0, 1024);
16452 case AMDGPU::BI__builtin_r600_read_tidig_y:
16453 return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_y, 0, 1024);
16454 case AMDGPU::BI__builtin_r600_read_tidig_z:
16455 return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_z, 0, 1024);
16456 case AMDGPU::BI__builtin_amdgcn_alignbit: {
16457 llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
16458 llvm::Value *Src1 = EmitScalarExpr(E->getArg(1));
16459 llvm::Value *Src2 = EmitScalarExpr(E->getArg(2));
16460 Function *F = CGM.getIntrinsic(Intrinsic::fshr, Src0->getType());
16461 return Builder.CreateCall(F, { Src0, Src1, Src2 });
16462 }
16463
16464 case AMDGPU::BI__builtin_amdgcn_fence: {
16465 if (ProcessOrderScopeAMDGCN(EmitScalarExpr(E->getArg(0)),
16466 EmitScalarExpr(E->getArg(1)), AO, SSID))
16467 return Builder.CreateFence(AO, SSID);
16468 LLVM_FALLTHROUGH[[gnu::fallthrough]];
16469 }
16470 case AMDGPU::BI__builtin_amdgcn_atomic_inc32:
16471 case AMDGPU::BI__builtin_amdgcn_atomic_inc64:
16472 case AMDGPU::BI__builtin_amdgcn_atomic_dec32:
16473 case AMDGPU::BI__builtin_amdgcn_atomic_dec64: {
16474 unsigned BuiltinAtomicOp;
16475 llvm::Type *ResultType = ConvertType(E->getType());
16476
16477 switch (BuiltinID) {
16478 case AMDGPU::BI__builtin_amdgcn_atomic_inc32:
16479 case AMDGPU::BI__builtin_amdgcn_atomic_inc64:
16480 BuiltinAtomicOp = Intrinsic::amdgcn_atomic_inc;
16481 break;
16482 case AMDGPU::BI__builtin_amdgcn_atomic_dec32:
16483 case AMDGPU::BI__builtin_amdgcn_atomic_dec64:
16484 BuiltinAtomicOp = Intrinsic::amdgcn_atomic_dec;
16485 break;
16486 }
16487
16488 Value *Ptr = EmitScalarExpr(E->getArg(0));
16489 Value *Val = EmitScalarExpr(E->getArg(1));
16490
16491 llvm::Function *F =
16492 CGM.getIntrinsic(BuiltinAtomicOp, {ResultType, Ptr->getType()});
16493
16494 if (ProcessOrderScopeAMDGCN(EmitScalarExpr(E->getArg(2)),
16495 EmitScalarExpr(E->getArg(3)), AO, SSID)) {
16496
16497 // llvm.amdgcn.atomic.inc and llvm.amdgcn.atomic.dec expects ordering and
16498 // scope as unsigned values
16499 Value *MemOrder = Builder.getInt32(static_cast<int>(AO));
16500 Value *MemScope = Builder.getInt32(static_cast<int>(SSID));
16501
16502 QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType();
16503 bool Volatile =
16504 PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified();
16505 Value *IsVolatile = Builder.getInt1(static_cast<bool>(Volatile));
16506
16507 return Builder.CreateCall(F, {Ptr, Val, MemOrder, MemScope, IsVolatile});
16508 }
16509 LLVM_FALLTHROUGH[[gnu::fallthrough]];
16510 }
16511 default:
16512 return nullptr;
16513 }
16514}
16515
16516/// Handle a SystemZ function in which the final argument is a pointer
16517/// to an int that receives the post-instruction CC value. At the LLVM level
16518/// this is represented as a function that returns a {result, cc} pair.
16519static Value *EmitSystemZIntrinsicWithCC(CodeGenFunction &CGF,
16520 unsigned IntrinsicID,
16521 const CallExpr *E) {
16522 unsigned NumArgs = E->getNumArgs() - 1;
16523 SmallVector<Value *, 8> Args(NumArgs);
16524 for (unsigned I = 0; I < NumArgs; ++I)
16525 Args[I] = CGF.EmitScalarExpr(E->getArg(I));
16526 Address CCPtr = CGF.EmitPointerWithAlignment(E->getArg(NumArgs));
16527 Function *F = CGF.CGM.getIntrinsic(IntrinsicID);
16528 Value *Call = CGF.Builder.CreateCall(F, Args);
16529 Value *CC = CGF.Builder.CreateExtractValue(Call, 1);
16530 CGF.Builder.CreateStore(CC, CCPtr);
16531 return CGF.Builder.CreateExtractValue(Call, 0);
16532}
16533
16534Value *CodeGenFunction::EmitSystemZBuiltinExpr(unsigned BuiltinID,
16535 const CallExpr *E) {
16536 switch (BuiltinID) {
16537 case SystemZ::BI__builtin_tbegin: {
16538 Value *TDB = EmitScalarExpr(E->getArg(0));
16539 Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c);
16540 Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin);
16541 return Builder.CreateCall(F, {TDB, Control});
16542 }
16543 case SystemZ::BI__builtin_tbegin_nofloat: {
16544 Value *TDB = EmitScalarExpr(E->getArg(0));
16545 Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c);
16546 Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin_nofloat);
16547 return Builder.CreateCall(F, {TDB, Control});
16548 }
16549 case SystemZ::BI__builtin_tbeginc: {
16550 Value *TDB = llvm::ConstantPointerNull::get(Int8PtrTy);
16551 Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff08);
16552 Function *F = CGM.getIntrinsic(Intrinsic::s390_tbeginc);
16553 return Builder.CreateCall(F, {TDB, Control});
16554 }
16555 case SystemZ::BI__builtin_tabort: {
16556 Value *Data = EmitScalarExpr(E->getArg(0));
16557 Function *F = CGM.getIntrinsic(Intrinsic::s390_tabort);
16558 return Builder.CreateCall(F, Builder.CreateSExt(Data, Int64Ty, "tabort"));
16559 }
16560 case SystemZ::BI__builtin_non_tx_store: {
16561 Value *Address = EmitScalarExpr(E->getArg(0));
16562 Value *Data = EmitScalarExpr(E->getArg(1));
16563 Function *F = CGM.getIntrinsic(Intrinsic::s390_ntstg);
16564 return Builder.CreateCall(F, {Data, Address});
16565 }
16566
16567 // Vector builtins. Note that most vector builtins are mapped automatically
16568 // to target-specific LLVM intrinsics. The ones handled specially here can
16569 // be represented via standard LLVM IR, which is preferable to enable common
16570 // LLVM optimizations.
16571
16572 case SystemZ::BI__builtin_s390_vpopctb:
16573 case SystemZ::BI__builtin_s390_vpopcth:
16574 case SystemZ::BI__builtin_s390_vpopctf:
16575 case SystemZ::BI__builtin_s390_vpopctg: {
16576 llvm::Type *ResultType = ConvertType(E->getType());
16577 Value *X = EmitScalarExpr(E->getArg(0));
16578 Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
16579 return Builder.CreateCall(F, X);
16580 }
16581
16582 case SystemZ::BI__builtin_s390_vclzb:
16583 case SystemZ::BI__builtin_s390_vclzh:
16584 case SystemZ::BI__builtin_s390_vclzf:
16585 case SystemZ::BI__builtin_s390_vclzg: {
16586 llvm::Type *ResultType = ConvertType(E->getType());
16587 Value *X = EmitScalarExpr(E->getArg(0));
16588 Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
16589 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType);
16590 return Builder.CreateCall(F, {X, Undef});
16591 }
16592
16593 case SystemZ::BI__builtin_s390_vctzb:
16594 case SystemZ::BI__builtin_s390_vctzh:
16595 case SystemZ::BI__builtin_s390_vctzf:
16596 case SystemZ::BI__builtin_s390_vctzg: {
16597 llvm::Type *ResultType = ConvertType(E->getType());
16598 Value *X = EmitScalarExpr(E->getArg(0));
16599 Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
16600 Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType);
16601 return Builder.CreateCall(F, {X, Undef});
16602 }
16603
16604 case SystemZ::BI__builtin_s390_vfsqsb:
16605 case SystemZ::BI__builtin_s390_vfsqdb: {
16606 llvm::Type *ResultType = ConvertType(E->getType());
16607 Value *X = EmitScalarExpr(E->getArg(0));
16608 if (Builder.getIsFPConstrained()) {
16609 Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_sqrt, ResultType);
16610 return Builder.CreateConstrainedFPCall(F, { X });
16611 } else {
16612 Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType);
16613 return Builder.CreateCall(F, X);
16614 }
16615 }
16616 case SystemZ::BI__builtin_s390_vfmasb:
16617 case SystemZ::BI__builtin_s390_vfmadb: {
16618 llvm::Type *ResultType = ConvertType(E->getType());
16619 Value *X = EmitScalarExpr(E->getArg(0));
16620 Value *Y = EmitScalarExpr(E->getArg(1));
16621 Value *Z = EmitScalarExpr(E->getArg(2));
16622 if (Builder.getIsFPConstrained()) {
16623 Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType);
16624 return Builder.CreateConstrainedFPCall(F, {X, Y, Z});
16625 } else {
16626 Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
16627 return Builder.CreateCall(F, {X, Y, Z});
16628 }
16629 }
16630 case SystemZ::BI__builtin_s390_vfmssb:
16631 case SystemZ::BI__builtin_s390_vfmsdb: {
16632 llvm::Type *ResultType = ConvertType(E->getType());
16633 Value *X = EmitScalarExpr(E->getArg(0));
16634 Value *Y = EmitScalarExpr(E->getArg(1));
16635 Value *Z = EmitScalarExpr(E->getArg(2));
16636 if (Builder.getIsFPConstrained()) {
16637 Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType);
16638 return Builder.CreateConstrainedFPCall(F, {X, Y, Builder.CreateFNeg(Z, "neg")});
16639 } else {
16640 Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
16641 return Builder.CreateCall(F, {X, Y, Builder.CreateFNeg(Z, "neg")});
16642 }
16643 }
16644 case SystemZ::BI__builtin_s390_vfnmasb:
16645 case SystemZ::BI__builtin_s390_vfnmadb: {
16646 llvm::Type *ResultType = ConvertType(E->getType());
16647 Value *X = EmitScalarExpr(E->getArg(0));
16648 Value *Y = EmitScalarExpr(E->getArg(1));
16649 Value *Z = EmitScalarExpr(E->getArg(2));
16650 if (Builder.getIsFPConstrained()) {
16651 Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType);
16652 return Builder.CreateFNeg(Builder.CreateConstrainedFPCall(F, {X, Y, Z}), "neg");
16653 } else {
16654 Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
16655 return Builder.CreateFNeg(Builder.CreateCall(F, {X, Y, Z}), "neg");
16656 }
16657 }
16658 case SystemZ::BI__builtin_s390_vfnmssb:
16659 case SystemZ::BI__builtin_s390_vfnmsdb: {
16660 llvm::Type *ResultType = ConvertType(E->getType());
16661 Value *X = EmitScalarExpr(E->getArg(0));
16662 Value *Y = EmitScalarExpr(E->getArg(1));
16663 Value *Z = EmitScalarExpr(E->getArg(2));
16664 if (Builder.getIsFPConstrained()) {
16665 Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType);
16666 Value *NegZ = Builder.CreateFNeg(Z, "sub");
16667 return Builder.CreateFNeg(Builder.CreateConstrainedFPCall(F, {X, Y, NegZ}));
16668 } else {
16669 Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
16670 Value *NegZ = Builder.CreateFNeg(Z, "neg");
16671 return Builder.CreateFNeg(Builder.CreateCall(F, {X, Y, NegZ}));
16672 }
16673 }
16674 case SystemZ::BI__builtin_s390_vflpsb:
16675 case SystemZ::BI__builtin_s390_vflpdb: {
16676 llvm::Type *ResultType = ConvertType(E->getType());
16677 Value *X = EmitScalarExpr(E->getArg(0));
16678 Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
16679 return Builder.CreateCall(F, X);
16680 }
16681 case SystemZ::BI__builtin_s390_vflnsb:
16682 case SystemZ::BI__builtin_s390_vflndb: {
16683 llvm::Type *ResultType = ConvertType(E->getType());
16684 Value *X = EmitScalarExpr(E->getArg(0));
16685 Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
16686 return Builder.CreateFNeg(Builder.CreateCall(F, X), "neg");
16687 }
16688 case SystemZ::BI__builtin_s390_vfisb:
16689 case SystemZ::BI__builtin_s390_vfidb: {
16690 llvm::Type *ResultType = ConvertType(E->getType());
16691 Value *X = EmitScalarExpr(E->getArg(0));
16692 // Constant-fold the M4 and M5 mask arguments.
16693 llvm::APSInt M4 = *E->getArg(1)->getIntegerConstantExpr(getContext());
16694 llvm::APSInt M5 = *E->getArg(2)->getIntegerConstantExpr(getContext());
16695 // Check whether this instance can be represented via a LLVM standard
16696 // intrinsic. We only support some combinations of M4 and M5.
16697 Intrinsic::ID ID = Intrinsic::not_intrinsic;
16698 Intrinsic::ID CI;
16699 switch (M4.getZExtValue()) {
16700 default: break;
16701 case 0: // IEEE-inexact exception allowed
16702 switch (M5.getZExtValue()) {
16703 default: break;
16704 case 0: ID = Intrinsic::rint;
16705 CI = Intrinsic::experimental_constrained_rint; break;
16706 }
16707 break;
16708 case 4: // IEEE-inexact exception suppressed
16709 switch (M5.getZExtValue()) {
16710 default: break;
16711 case 0: ID = Intrinsic::nearbyint;
16712 CI = Intrinsic::experimental_constrained_nearbyint; break;
16713 case 1: ID = Intrinsic::round;
16714 CI = Intrinsic::experimental_constrained_round; break;
16715 case 5: ID = Intrinsic::trunc;
16716 CI = Intrinsic::experimental_constrained_trunc; break;
16717 case 6: ID = Intrinsic::ceil;
16718 CI = Intrinsic::experimental_constrained_ceil; break;
16719 case 7: ID = Intrinsic::floor;
16720 CI = Intrinsic::experimental_constrained_floor; break;
16721 }
16722 break;
16723 }
16724 if (ID != Intrinsic::not_intrinsic) {
16725 if (Builder.getIsFPConstrained()) {
16726 Function *F = CGM.getIntrinsic(CI, ResultType);
16727 return Builder.CreateConstrainedFPCall(F, X);
16728 } else {
16729 Function *F = CGM.getIntrinsic(ID, ResultType);
16730 return Builder.CreateCall(F, X);
16731 }
16732 }
16733 switch (BuiltinID) { // FIXME: constrained version?
16734 case SystemZ::BI__builtin_s390_vfisb: ID = Intrinsic::s390_vfisb; break;
16735 case SystemZ::BI__builtin_s390_vfidb: ID = Intrinsic::s390_vfidb; break;
16736 default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 16736)
;
16737 }
16738 Function *F = CGM.getIntrinsic(ID);
16739 Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
16740 Value *M5Value = llvm::ConstantInt::get(getLLVMContext(), M5);
16741 return Builder.CreateCall(F, {X, M4Value, M5Value});
16742 }
16743 case SystemZ::BI__builtin_s390_vfmaxsb:
16744 case SystemZ::BI__builtin_s390_vfmaxdb: {
16745 llvm::Type *ResultType = ConvertType(E->getType());
16746 Value *X = EmitScalarExpr(E->getArg(0));
16747 Value *Y = EmitScalarExpr(E->getArg(1));
16748 // Constant-fold the M4 mask argument.
16749 llvm::APSInt M4 = *E->getArg(2)->getIntegerConstantExpr(getContext());
16750 // Check whether this instance can be represented via a LLVM standard
16751 // intrinsic. We only support some values of M4.
16752 Intrinsic::ID ID = Intrinsic::not_intrinsic;
16753 Intrinsic::ID CI;
16754 switch (M4.getZExtValue()) {
16755 default: break;
16756 case 4: ID = Intrinsic::maxnum;
16757 CI = Intrinsic::experimental_constrained_maxnum; break;
16758 }
16759 if (ID != Intrinsic::not_intrinsic) {
16760 if (Builder.getIsFPConstrained()) {
16761 Function *F = CGM.getIntrinsic(CI, ResultType);
16762 return Builder.CreateConstrainedFPCall(F, {X, Y});
16763 } else {
16764 Function *F = CGM.getIntrinsic(ID, ResultType);
16765 return Builder.CreateCall(F, {X, Y});
16766 }
16767 }
16768 switch (BuiltinID) {
16769 case SystemZ::BI__builtin_s390_vfmaxsb: ID = Intrinsic::s390_vfmaxsb; break;
16770 case SystemZ::BI__builtin_s390_vfmaxdb: ID = Intrinsic::s390_vfmaxdb; break;
16771 default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 16771)
;
16772 }
16773 Function *F = CGM.getIntrinsic(ID);
16774 Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
16775 return Builder.CreateCall(F, {X, Y, M4Value});
16776 }
16777 case SystemZ::BI__builtin_s390_vfminsb:
16778 case SystemZ::BI__builtin_s390_vfmindb: {
16779 llvm::Type *ResultType = ConvertType(E->getType());
16780 Value *X = EmitScalarExpr(E->getArg(0));
16781 Value *Y = EmitScalarExpr(E->getArg(1));
16782 // Constant-fold the M4 mask argument.
16783 llvm::APSInt M4 = *E->getArg(2)->getIntegerConstantExpr(getContext());
16784 // Check whether this instance can be represented via a LLVM standard
16785 // intrinsic. We only support some values of M4.
16786 Intrinsic::ID ID = Intrinsic::not_intrinsic;
16787 Intrinsic::ID CI;
16788 switch (M4.getZExtValue()) {
16789 default: break;
16790 case 4: ID = Intrinsic::minnum;
16791 CI = Intrinsic::experimental_constrained_minnum; break;
16792 }
16793 if (ID != Intrinsic::not_intrinsic) {
16794 if (Builder.getIsFPConstrained()) {
16795 Function *F = CGM.getIntrinsic(CI, ResultType);
16796 return Builder.CreateConstrainedFPCall(F, {X, Y});
16797 } else {
16798 Function *F = CGM.getIntrinsic(ID, ResultType);
16799 return Builder.CreateCall(F, {X, Y});
16800 }
16801 }
16802 switch (BuiltinID) {
16803 case SystemZ::BI__builtin_s390_vfminsb: ID = Intrinsic::s390_vfminsb; break;
16804 case SystemZ::BI__builtin_s390_vfmindb: ID = Intrinsic::s390_vfmindb; break;
16805 default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 16805)
;
16806 }
16807 Function *F = CGM.getIntrinsic(ID);
16808 Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
16809 return Builder.CreateCall(F, {X, Y, M4Value});
16810 }
16811
16812 case SystemZ::BI__builtin_s390_vlbrh:
16813 case SystemZ::BI__builtin_s390_vlbrf:
16814 case SystemZ::BI__builtin_s390_vlbrg: {
16815 llvm::Type *ResultType = ConvertType(E->getType());
16816 Value *X = EmitScalarExpr(E->getArg(0));
16817 Function *F = CGM.getIntrinsic(Intrinsic::bswap, ResultType);
16818 return Builder.CreateCall(F, X);
16819 }
16820
16821 // Vector intrinsics that output the post-instruction CC value.
16822
16823#define INTRINSIC_WITH_CC(NAME) \
16824 case SystemZ::BI__builtin_##NAME: \
16825 return EmitSystemZIntrinsicWithCC(*this, Intrinsic::NAME, E)
16826
16827 INTRINSIC_WITH_CC(s390_vpkshs);
16828 INTRINSIC_WITH_CC(s390_vpksfs);
16829 INTRINSIC_WITH_CC(s390_vpksgs);
16830
16831 INTRINSIC_WITH_CC(s390_vpklshs);
16832 INTRINSIC_WITH_CC(s390_vpklsfs);
16833 INTRINSIC_WITH_CC(s390_vpklsgs);
16834
16835 INTRINSIC_WITH_CC(s390_vceqbs);
16836 INTRINSIC_WITH_CC(s390_vceqhs);
16837 INTRINSIC_WITH_CC(s390_vceqfs);
16838 INTRINSIC_WITH_CC(s390_vceqgs);
16839
16840 INTRINSIC_WITH_CC(s390_vchbs);
16841 INTRINSIC_WITH_CC(s390_vchhs);
16842 INTRINSIC_WITH_CC(s390_vchfs);
16843 INTRINSIC_WITH_CC(s390_vchgs);
16844
16845 INTRINSIC_WITH_CC(s390_vchlbs);
16846 INTRINSIC_WITH_CC(s390_vchlhs);
16847 INTRINSIC_WITH_CC(s390_vchlfs);
16848 INTRINSIC_WITH_CC(s390_vchlgs);
16849
16850 INTRINSIC_WITH_CC(s390_vfaebs);
16851 INTRINSIC_WITH_CC(s390_vfaehs);
16852 INTRINSIC_WITH_CC(s390_vfaefs);
16853
16854 INTRINSIC_WITH_CC(s390_vfaezbs);
16855 INTRINSIC_WITH_CC(s390_vfaezhs);
16856 INTRINSIC_WITH_CC(s390_vfaezfs);
16857
16858 INTRINSIC_WITH_CC(s390_vfeebs);
16859 INTRINSIC_WITH_CC(s390_vfeehs);
16860 INTRINSIC_WITH_CC(s390_vfeefs);
16861
16862 INTRINSIC_WITH_CC(s390_vfeezbs);
16863 INTRINSIC_WITH_CC(s390_vfeezhs);
16864 INTRINSIC_WITH_CC(s390_vfeezfs);
16865
16866 INTRINSIC_WITH_CC(s390_vfenebs);
16867 INTRINSIC_WITH_CC(s390_vfenehs);
16868 INTRINSIC_WITH_CC(s390_vfenefs);
16869
16870 INTRINSIC_WITH_CC(s390_vfenezbs);
16871 INTRINSIC_WITH_CC(s390_vfenezhs);
16872 INTRINSIC_WITH_CC(s390_vfenezfs);
16873
16874 INTRINSIC_WITH_CC(s390_vistrbs);
16875 INTRINSIC_WITH_CC(s390_vistrhs);
16876 INTRINSIC_WITH_CC(s390_vistrfs);
16877
16878 INTRINSIC_WITH_CC(s390_vstrcbs);
16879 INTRINSIC_WITH_CC(s390_vstrchs);
16880 INTRINSIC_WITH_CC(s390_vstrcfs);
16881
16882 INTRINSIC_WITH_CC(s390_vstrczbs);
16883 INTRINSIC_WITH_CC(s390_vstrczhs);
16884 INTRINSIC_WITH_CC(s390_vstrczfs);
16885
16886 INTRINSIC_WITH_CC(s390_vfcesbs);
16887 INTRINSIC_WITH_CC(s390_vfcedbs);
16888 INTRINSIC_WITH_CC(s390_vfchsbs);
16889 INTRINSIC_WITH_CC(s390_vfchdbs);
16890 INTRINSIC_WITH_CC(s390_vfchesbs);
16891 INTRINSIC_WITH_CC(s390_vfchedbs);
16892
16893 INTRINSIC_WITH_CC(s390_vftcisb);
16894 INTRINSIC_WITH_CC(s390_vftcidb);
16895
16896 INTRINSIC_WITH_CC(s390_vstrsb);
16897 INTRINSIC_WITH_CC(s390_vstrsh);
16898 INTRINSIC_WITH_CC(s390_vstrsf);
16899
16900 INTRINSIC_WITH_CC(s390_vstrszb);
16901 INTRINSIC_WITH_CC(s390_vstrszh);
16902 INTRINSIC_WITH_CC(s390_vstrszf);
16903
16904#undef INTRINSIC_WITH_CC
16905
16906 default:
16907 return nullptr;
16908 }
16909}
16910
16911namespace {
16912// Helper classes for mapping MMA builtins to particular LLVM intrinsic variant.
16913struct NVPTXMmaLdstInfo {
16914 unsigned NumResults; // Number of elements to load/store
16915 // Intrinsic IDs for row/col variants. 0 if particular layout is unsupported.
16916 unsigned IID_col;
16917 unsigned IID_row;
16918};
16919
16920#define MMA_INTR(geom_op_type, layout) \
16921 Intrinsic::nvvm_wmma_##geom_op_type##_##layout##_stride
16922#define MMA_LDST(n, geom_op_type) \
16923 { n, MMA_INTR(geom_op_type, col), MMA_INTR(geom_op_type, row) }
16924
16925static NVPTXMmaLdstInfo getNVPTXMmaLdstInfo(unsigned BuiltinID) {
16926 switch (BuiltinID) {
16927 // FP MMA loads
16928 case NVPTX::BI__hmma_m16n16k16_ld_a:
16929 return MMA_LDST(8, m16n16k16_load_a_f16);
16930 case NVPTX::BI__hmma_m16n16k16_ld_b:
16931 return MMA_LDST(8, m16n16k16_load_b_f16);
16932 case NVPTX::BI__hmma_m16n16k16_ld_c_f16:
16933 return MMA_LDST(4, m16n16k16_load_c_f16);
16934 case NVPTX::BI__hmma_m16n16k16_ld_c_f32:
16935 return MMA_LDST(8, m16n16k16_load_c_f32);
16936 case NVPTX::BI__hmma_m32n8k16_ld_a:
16937 return MMA_LDST(8, m32n8k16_load_a_f16);
16938 case NVPTX::BI__hmma_m32n8k16_ld_b:
16939 return MMA_LDST(8, m32n8k16_load_b_f16);
16940 case NVPTX::BI__hmma_m32n8k16_ld_c_f16:
16941 return MMA_LDST(4, m32n8k16_load_c_f16);
16942 case NVPTX::BI__hmma_m32n8k16_ld_c_f32:
16943 return MMA_LDST(8, m32n8k16_load_c_f32);
16944 case NVPTX::BI__hmma_m8n32k16_ld_a:
16945 return MMA_LDST(8, m8n32k16_load_a_f16);
16946 case NVPTX::BI__hmma_m8n32k16_ld_b:
16947 return MMA_LDST(8, m8n32k16_load_b_f16);
16948 case NVPTX::BI__hmma_m8n32k16_ld_c_f16:
16949 return MMA_LDST(4, m8n32k16_load_c_f16);
16950 case NVPTX::BI__hmma_m8n32k16_ld_c_f32:
16951 return MMA_LDST(8, m8n32k16_load_c_f32);
16952
16953 // Integer MMA loads
16954 case NVPTX::BI__imma_m16n16k16_ld_a_s8:
16955 return MMA_LDST(2, m16n16k16_load_a_s8);
16956 case NVPTX::BI__imma_m16n16k16_ld_a_u8:
16957 return MMA_LDST(2, m16n16k16_load_a_u8);
16958 case NVPTX::BI__imma_m16n16k16_ld_b_s8:
16959 return MMA_LDST(2, m16n16k16_load_b_s8);
16960 case NVPTX::BI__imma_m16n16k16_ld_b_u8:
16961 return MMA_LDST(2, m16n16k16_load_b_u8);
16962 case NVPTX::BI__imma_m16n16k16_ld_c:
16963 return MMA_LDST(8, m16n16k16_load_c_s32);
16964 case NVPTX::BI__imma_m32n8k16_ld_a_s8:
16965 return MMA_LDST(4, m32n8k16_load_a_s8);
16966 case NVPTX::BI__imma_m32n8k16_ld_a_u8:
16967 return MMA_LDST(4, m32n8k16_load_a_u8);
16968 case NVPTX::BI__imma_m32n8k16_ld_b_s8:
16969 return MMA_LDST(1, m32n8k16_load_b_s8);
16970 case NVPTX::BI__imma_m32n8k16_ld_b_u8:
16971 return MMA_LDST(1, m32n8k16_load_b_u8);
16972 case NVPTX::BI__imma_m32n8k16_ld_c:
16973 return MMA_LDST(8, m32n8k16_load_c_s32);
16974 case NVPTX::BI__imma_m8n32k16_ld_a_s8:
16975 return MMA_LDST(1, m8n32k16_load_a_s8);
16976 case NVPTX::BI__imma_m8n32k16_ld_a_u8:
16977 return MMA_LDST(1, m8n32k16_load_a_u8);
16978 case NVPTX::BI__imma_m8n32k16_ld_b_s8:
16979 return MMA_LDST(4, m8n32k16_load_b_s8);
16980 case NVPTX::BI__imma_m8n32k16_ld_b_u8:
16981 return MMA_LDST(4, m8n32k16_load_b_u8);
16982 case NVPTX::BI__imma_m8n32k16_ld_c:
16983 return MMA_LDST(8, m8n32k16_load_c_s32);
16984
16985 // Sub-integer MMA loads.
16986 // Only row/col layout is supported by A/B fragments.
16987 case NVPTX::BI__imma_m8n8k32_ld_a_s4:
16988 return {1, 0, MMA_INTR(m8n8k32_load_a_s4, row)};
16989 case NVPTX::BI__imma_m8n8k32_ld_a_u4:
16990 return {1, 0, MMA_INTR(m8n8k32_load_a_u4, row)};
16991 case NVPTX::BI__imma_m8n8k32_ld_b_s4:
16992 return {1, MMA_INTR(m8n8k32_load_b_s4, col), 0};
16993 case NVPTX::BI__imma_m8n8k32_ld_b_u4:
16994 return {1, MMA_INTR(m8n8k32_load_b_u4, col), 0};
16995 case NVPTX::BI__imma_m8n8k32_ld_c:
16996 return MMA_LDST(2, m8n8k32_load_c_s32);
16997 case NVPTX::BI__bmma_m8n8k128_ld_a_b1:
16998 return {1, 0, MMA_INTR(m8n8k128_load_a_b1, row)};
16999 case NVPTX::BI__bmma_m8n8k128_ld_b_b1:
17000 return {1, MMA_INTR(m8n8k128_load_b_b1, col), 0};
17001 case NVPTX::BI__bmma_m8n8k128_ld_c:
17002 return MMA_LDST(2, m8n8k128_load_c_s32);
17003
17004 // Double MMA loads
17005 case NVPTX::BI__dmma_m8n8k4_ld_a:
17006 return MMA_LDST(1, m8n8k4_load_a_f64);
17007 case NVPTX::BI__dmma_m8n8k4_ld_b:
17008 return MMA_LDST(1, m8n8k4_load_b_f64);
17009 case NVPTX::BI__dmma_m8n8k4_ld_c:
17010 return MMA_LDST(2, m8n8k4_load_c_f64);
17011
17012 // Alternate float MMA loads
17013 case NVPTX::BI__mma_bf16_m16n16k16_ld_a:
17014 return MMA_LDST(4, m16n16k16_load_a_bf16);
17015 case NVPTX::BI__mma_bf16_m16n16k16_ld_b:
17016 return MMA_LDST(4, m16n16k16_load_b_bf16);
17017 case NVPTX::BI__mma_bf16_m8n32k16_ld_a:
17018 return MMA_LDST(2, m8n32k16_load_a_bf16);
17019 case NVPTX::BI__mma_bf16_m8n32k16_ld_b:
17020 return MMA_LDST(8, m8n32k16_load_b_bf16);
17021 case NVPTX::BI__mma_bf16_m32n8k16_ld_a:
17022 return MMA_LDST(8, m32n8k16_load_a_bf16);
17023 case NVPTX::BI__mma_bf16_m32n8k16_ld_b:
17024 return MMA_LDST(2, m32n8k16_load_b_bf16);
17025 case NVPTX::BI__mma_tf32_m16n16k8_ld_a:
17026 return MMA_LDST(4, m16n16k8_load_a_tf32);
17027 case NVPTX::BI__mma_tf32_m16n16k8_ld_b:
17028 return MMA_LDST(2, m16n16k8_load_b_tf32);
17029 case NVPTX::BI__mma_tf32_m16n16k8_ld_c:
17030 return MMA_LDST(8, m16n16k8_load_c_f32);
17031
17032 // NOTE: We need to follow inconsitent naming scheme used by NVCC. Unlike
17033 // PTX and LLVM IR where stores always use fragment D, NVCC builtins always
17034 // use fragment C for both loads and stores.
17035 // FP MMA stores.
17036 case NVPTX::BI__hmma_m16n16k16_st_c_f16:
17037 return MMA_LDST(4, m16n16k16_store_d_f16);
17038 case NVPTX::BI__hmma_m16n16k16_st_c_f32:
17039 return MMA_LDST(8, m16n16k16_store_d_f32);
17040 case NVPTX::BI__hmma_m32n8k16_st_c_f16:
17041 return MMA_LDST(4, m32n8k16_store_d_f16);
17042 case NVPTX::BI__hmma_m32n8k16_st_c_f32:
17043 return MMA_LDST(8, m32n8k16_store_d_f32);
17044 case NVPTX::BI__hmma_m8n32k16_st_c_f16:
17045 return MMA_LDST(4, m8n32k16_store_d_f16);
17046 case NVPTX::BI__hmma_m8n32k16_st_c_f32:
17047 return MMA_LDST(8, m8n32k16_store_d_f32);
17048
17049 // Integer and sub-integer MMA stores.
17050 // Another naming quirk. Unlike other MMA builtins that use PTX types in the
17051 // name, integer loads/stores use LLVM's i32.
17052 case NVPTX::BI__imma_m16n16k16_st_c_i32:
17053 return MMA_LDST(8, m16n16k16_store_d_s32);
17054 case NVPTX::BI__imma_m32n8k16_st_c_i32:
17055 return MMA_LDST(8, m32n8k16_store_d_s32);
17056 case NVPTX::BI__imma_m8n32k16_st_c_i32:
17057 return MMA_LDST(8, m8n32k16_store_d_s32);
17058 case NVPTX::BI__imma_m8n8k32_st_c_i32:
17059 return MMA_LDST(2, m8n8k32_store_d_s32);
17060 case NVPTX::BI__bmma_m8n8k128_st_c_i32:
17061 return MMA_LDST(2, m8n8k128_store_d_s32);
17062
17063 // Double MMA store
17064 case NVPTX::BI__dmma_m8n8k4_st_c_f64:
17065 return MMA_LDST(2, m8n8k4_store_d_f64);
17066
17067 // Alternate float MMA store
17068 case NVPTX::BI__mma_m16n16k8_st_c_f32:
17069 return MMA_LDST(8, m16n16k8_store_d_f32);
17070
17071 default:
17072 llvm_unreachable("Unknown MMA builtin")::llvm::llvm_unreachable_internal("Unknown MMA builtin", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 17072)
;
17073 }
17074}
17075#undef MMA_LDST
17076#undef MMA_INTR
17077
17078
17079struct NVPTXMmaInfo {
17080 unsigned NumEltsA;
17081 unsigned NumEltsB;
17082 unsigned NumEltsC;
17083 unsigned NumEltsD;
17084
17085 // Variants are ordered by layout-A/layout-B/satf, where 'row' has priority
17086 // over 'col' for layout. The index of non-satf variants is expected to match
17087 // the undocumented layout constants used by CUDA's mma.hpp.
17088 std::array<unsigned, 8> Variants;
17089
17090 unsigned getMMAIntrinsic(int Layout, bool Satf) {
17091 unsigned Index = Layout + 4 * Satf;
17092 if (Index >= Variants.size())
17093 return 0;
17094 return Variants[Index];
17095 }
17096};
17097
17098 // Returns an intrinsic that matches Layout and Satf for valid combinations of
17099 // Layout and Satf, 0 otherwise.
17100static NVPTXMmaInfo getNVPTXMmaInfo(unsigned BuiltinID) {
17101 // clang-format off
17102#define MMA_VARIANTS(geom, type) \
17103 Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type, \
17104 Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \
17105 Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type, \
17106 Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type
17107#define MMA_SATF_VARIANTS(geom, type) \
17108 MMA_VARIANTS(geom, type), \
17109 Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type##_satfinite, \
17110 Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \
17111 Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type##_satfinite, \
17112 Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type##_satfinite
17113// Sub-integer MMA only supports row.col layout.
17114#define MMA_VARIANTS_I4(geom, type) \
17115 0, \
17116 Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \
17117 0, \
17118 0, \
17119 0, \
17120 Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \
17121 0, \
17122 0
17123// b1 MMA does not support .satfinite.
17124#define MMA_VARIANTS_B1_XOR(geom, type) \
17125 0, \
17126 Intrinsic::nvvm_wmma_##geom##_mma_xor_popc_row_col_##type, \
17127 0, \
17128 0, \
17129 0, \
17130 0, \
17131 0, \
17132 0
17133#define MMA_VARIANTS_B1_AND(geom, type) \
17134 0, \
17135 Intrinsic::nvvm_wmma_##geom##_mma_and_popc_row_col_##type, \
17136 0, \
17137 0, \
17138 0, \
17139 0, \
17140 0, \
17141 0
17142 // clang-format on
17143 switch (BuiltinID) {
17144 // FP MMA
17145 // Note that 'type' argument of MMA_SATF_VARIANTS uses D_C notation, while
17146 // NumEltsN of return value are ordered as A,B,C,D.
17147 case NVPTX::BI__hmma_m16n16k16_mma_f16f16:
17148 return {8, 8, 4, 4, {{MMA_SATF_VARIANTS(m16n16k16, f16_f16)}}};
17149 case NVPTX::BI__hmma_m16n16k16_mma_f32f16:
17150 return {8, 8, 4, 8, {{MMA_SATF_VARIANTS(m16n16k16, f32_f16)}}};
17151 case NVPTX::BI__hmma_m16n16k16_mma_f16f32:
17152 return {8, 8, 8, 4, {{MMA_SATF_VARIANTS(m16n16k16, f16_f32)}}};
17153 case NVPTX::BI__hmma_m16n16k16_mma_f32f32:
17154 return {8, 8, 8, 8, {{MMA_SATF_VARIANTS(m16n16k16, f32_f32)}}};
17155 case NVPTX::BI__hmma_m32n8k16_mma_f16f16:
17156 return {8, 8, 4, 4, {{MMA_SATF_VARIANTS(m32n8k16, f16_f16)}}};
17157 case NVPTX::BI__hmma_m32n8k16_mma_f32f16:
17158 return {8, 8, 4, 8, {{MMA_SATF_VARIANTS(m32n8k16, f32_f16)}}};
17159 case NVPTX::BI__hmma_m32n8k16_mma_f16f32:
17160 return {8, 8, 8, 4, {{MMA_SATF_VARIANTS(m32n8k16, f16_f32)}}};
17161 case NVPTX::BI__hmma_m32n8k16_mma_f32f32:
17162 return {8, 8, 8, 8, {{MMA_SATF_VARIANTS(m32n8k16, f32_f32)}}};
17163 case NVPTX::BI__hmma_m8n32k16_mma_f16f16:
17164 return {8, 8, 4, 4, {{MMA_SATF_VARIANTS(m8n32k16, f16_f16)}}};
17165 case NVPTX::BI__hmma_m8n32k16_mma_f32f16:
17166 return {8, 8, 4, 8, {{MMA_SATF_VARIANTS(m8n32k16, f32_f16)}}};
17167 case NVPTX::BI__hmma_m8n32k16_mma_f16f32:
17168 return {8, 8, 8, 4, {{MMA_SATF_VARIANTS(m8n32k16, f16_f32)}}};
17169 case NVPTX::BI__hmma_m8n32k16_mma_f32f32:
17170 return {8, 8, 8, 8, {{MMA_SATF_VARIANTS(m8n32k16, f32_f32)}}};
17171
17172 // Integer MMA
17173 case NVPTX::BI__imma_m16n16k16_mma_s8:
17174 return {2, 2, 8, 8, {{MMA_SATF_VARIANTS(m16n16k16, s8)}}};
17175 case NVPTX::BI__imma_m16n16k16_mma_u8:
17176 return {2, 2, 8, 8, {{MMA_SATF_VARIANTS(m16n16k16, u8)}}};
17177 case NVPTX::BI__imma_m32n8k16_mma_s8:
17178 return {4, 1, 8, 8, {{MMA_SATF_VARIANTS(m32n8k16, s8)}}};
17179 case NVPTX::BI__imma_m32n8k16_mma_u8:
17180 return {4, 1, 8, 8, {{MMA_SATF_VARIANTS(m32n8k16, u8)}}};
17181 case NVPTX::BI__imma_m8n32k16_mma_s8:
17182 return {1, 4, 8, 8, {{MMA_SATF_VARIANTS(m8n32k16, s8)}}};
17183 case NVPTX::BI__imma_m8n32k16_mma_u8:
17184 return {1, 4, 8, 8, {{MMA_SATF_VARIANTS(m8n32k16, u8)}}};
17185
17186 // Sub-integer MMA
17187 case NVPTX::BI__imma_m8n8k32_mma_s4:
17188 return {1, 1, 2, 2, {{MMA_VARIANTS_I4(m8n8k32, s4)}}};
17189 case NVPTX::BI__imma_m8n8k32_mma_u4:
17190 return {1, 1, 2, 2, {{MMA_VARIANTS_I4(m8n8k32, u4)}}};
17191 case NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1:
17192 return {1, 1, 2, 2, {{MMA_VARIANTS_B1_XOR(m8n8k128, b1)}}};
17193 case NVPTX::BI__bmma_m8n8k128_mma_and_popc_b1:
17194 return {1, 1, 2, 2, {{MMA_VARIANTS_B1_AND(m8n8k128, b1)}}};
17195
17196 // Double MMA
17197 case NVPTX::BI__dmma_m8n8k4_mma_f64:
17198 return {1, 1, 2, 2, {{MMA_VARIANTS(m8n8k4, f64)}}};
17199
17200 // Alternate FP MMA
17201 case NVPTX::BI__mma_bf16_m16n16k16_mma_f32:
17202 return {4, 4, 8, 8, {{MMA_VARIANTS(m16n16k16, bf16)}}};
17203 case NVPTX::BI__mma_bf16_m8n32k16_mma_f32:
17204 return {2, 8, 8, 8, {{MMA_VARIANTS(m8n32k16, bf16)}}};
17205 case NVPTX::BI__mma_bf16_m32n8k16_mma_f32:
17206 return {8, 2, 8, 8, {{MMA_VARIANTS(m32n8k16, bf16)}}};
17207 case NVPTX::BI__mma_tf32_m16n16k8_mma_f32:
17208 return {4, 4, 8, 8, {{MMA_VARIANTS(m16n16k8, tf32)}}};
17209 default:
17210 llvm_unreachable("Unexpected builtin ID.")::llvm::llvm_unreachable_internal("Unexpected builtin ID.", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 17210)
;
17211 }
17212#undef MMA_VARIANTS
17213#undef MMA_SATF_VARIANTS
17214#undef MMA_VARIANTS_I4
17215#undef MMA_VARIANTS_B1_AND
17216#undef MMA_VARIANTS_B1_XOR
17217}
17218
17219} // namespace
17220
17221Value *
17222CodeGenFunction::EmitNVPTXBuiltinExpr(unsigned BuiltinID, const CallExpr *E) {
17223 auto MakeLdg = [&](unsigned IntrinsicID) {
17224 Value *Ptr = EmitScalarExpr(E->getArg(0));
17225 clang::CharUnits Align =
17226 CGM.getNaturalPointeeTypeAlignment(E->getArg(0)->getType());
17227 return Builder.CreateCall(
17228 CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(),
17229 Ptr->getType()}),
17230 {Ptr, ConstantInt::get(Builder.getInt32Ty(), Align.getQuantity())});
17231 };
17232 auto MakeScopedAtomic = [&](unsigned IntrinsicID) {
17233 Value *Ptr = EmitScalarExpr(E->getArg(0));
17234 return Builder.CreateCall(
17235 CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(),
17236 Ptr->getType()}),
17237 {Ptr, EmitScalarExpr(E->getArg(1))});
17238 };
17239 switch (BuiltinID) {
17240 case NVPTX::BI__nvvm_atom_add_gen_i:
17241 case NVPTX::BI__nvvm_atom_add_gen_l:
17242 case NVPTX::BI__nvvm_atom_add_gen_ll:
17243 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Add, E);
17244
17245 case NVPTX::BI__nvvm_atom_sub_gen_i:
17246 case NVPTX::BI__nvvm_atom_sub_gen_l:
17247 case NVPTX::BI__nvvm_atom_sub_gen_ll:
17248 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Sub, E);
17249
17250 case NVPTX::BI__nvvm_atom_and_gen_i:
17251 case NVPTX::BI__nvvm_atom_and_gen_l:
17252 case NVPTX::BI__nvvm_atom_and_gen_ll:
17253 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::And, E);
17254
17255 case NVPTX::BI__nvvm_atom_or_gen_i:
17256 case NVPTX::BI__nvvm_atom_or_gen_l:
17257 case NVPTX::BI__nvvm_atom_or_gen_ll:
17258 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Or, E);
17259
17260 case NVPTX::BI__nvvm_atom_xor_gen_i:
17261 case NVPTX::BI__nvvm_atom_xor_gen_l:
17262 case NVPTX::BI__nvvm_atom_xor_gen_ll:
17263 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xor, E);
17264
17265 case NVPTX::BI__nvvm_atom_xchg_gen_i:
17266 case NVPTX::BI__nvvm_atom_xchg_gen_l:
17267 case NVPTX::BI__nvvm_atom_xchg_gen_ll:
17268 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xchg, E);
17269
17270 case NVPTX::BI__nvvm_atom_max_gen_i:
17271 case NVPTX::BI__nvvm_atom_max_gen_l:
17272 case NVPTX::BI__nvvm_atom_max_gen_ll:
17273 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Max, E);
17274
17275 case NVPTX::BI__nvvm_atom_max_gen_ui:
17276 case NVPTX::BI__nvvm_atom_max_gen_ul:
17277 case NVPTX::BI__nvvm_atom_max_gen_ull:
17278 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMax, E);
17279
17280 case NVPTX::BI__nvvm_atom_min_gen_i:
17281 case NVPTX::BI__nvvm_atom_min_gen_l:
17282 case NVPTX::BI__nvvm_atom_min_gen_ll:
17283 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Min, E);
17284
17285 case NVPTX::BI__nvvm_atom_min_gen_ui:
17286 case NVPTX::BI__nvvm_atom_min_gen_ul:
17287 case NVPTX::BI__nvvm_atom_min_gen_ull:
17288 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMin, E);
17289
17290 case NVPTX::BI__nvvm_atom_cas_gen_i:
17291 case NVPTX::BI__nvvm_atom_cas_gen_l:
17292 case NVPTX::BI__nvvm_atom_cas_gen_ll:
17293 // __nvvm_atom_cas_gen_* should return the old value rather than the
17294 // success flag.
17295 return MakeAtomicCmpXchgValue(*this, E, /*ReturnBool=*/false);
17296
17297 case NVPTX::BI__nvvm_atom_add_gen_f:
17298 case NVPTX::BI__nvvm_atom_add_gen_d: {
17299 Value *Ptr = EmitScalarExpr(E->getArg(0));
17300 Value *Val = EmitScalarExpr(E->getArg(1));
17301 return Builder.CreateAtomicRMW(llvm::AtomicRMWInst::FAdd, Ptr, Val,
17302 AtomicOrdering::SequentiallyConsistent);
17303 }
17304
17305 case NVPTX::BI__nvvm_atom_inc_gen_ui: {
17306 Value *Ptr = EmitScalarExpr(E->getArg(0));
17307 Value *Val = EmitScalarExpr(E->getArg(1));
17308 Function *FnALI32 =
17309 CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_inc_32, Ptr->getType());
17310 return Builder.CreateCall(FnALI32, {Ptr, Val});
17311 }
17312
17313 case NVPTX::BI__nvvm_atom_dec_gen_ui: {
17314 Value *Ptr = EmitScalarExpr(E->getArg(0));
17315 Value *Val = EmitScalarExpr(E->getArg(1));
17316 Function *FnALD32 =
17317 CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_dec_32, Ptr->getType());
17318 return Builder.CreateCall(FnALD32, {Ptr, Val});
17319 }
17320
17321 case NVPTX::BI__nvvm_ldg_c:
17322 case NVPTX::BI__nvvm_ldg_c2:
17323 case NVPTX::BI__nvvm_ldg_c4:
17324 case NVPTX::BI__nvvm_ldg_s:
17325 case NVPTX::BI__nvvm_ldg_s2:
17326 case NVPTX::BI__nvvm_ldg_s4:
17327 case NVPTX::BI__nvvm_ldg_i:
17328 case NVPTX::BI__nvvm_ldg_i2:
17329 case NVPTX::BI__nvvm_ldg_i4:
17330 case NVPTX::BI__nvvm_ldg_l:
17331 case NVPTX::BI__nvvm_ldg_ll:
17332 case NVPTX::BI__nvvm_ldg_ll2:
17333 case NVPTX::BI__nvvm_ldg_uc:
17334 case NVPTX::BI__nvvm_ldg_uc2:
17335 case NVPTX::BI__nvvm_ldg_uc4:
17336 case NVPTX::BI__nvvm_ldg_us:
17337 case NVPTX::BI__nvvm_ldg_us2:
17338 case NVPTX::BI__nvvm_ldg_us4:
17339 case NVPTX::BI__nvvm_ldg_ui:
17340 case NVPTX::BI__nvvm_ldg_ui2:
17341 case NVPTX::BI__nvvm_ldg_ui4:
17342 case NVPTX::BI__nvvm_ldg_ul:
17343 case NVPTX::BI__nvvm_ldg_ull:
17344 case NVPTX::BI__nvvm_ldg_ull2:
17345 // PTX Interoperability section 2.2: "For a vector with an even number of
17346 // elements, its alignment is set to number of elements times the alignment
17347 // of its member: n*alignof(t)."
17348 return MakeLdg(Intrinsic::nvvm_ldg_global_i);
17349 case NVPTX::BI__nvvm_ldg_f:
17350 case NVPTX::BI__nvvm_ldg_f2:
17351 case NVPTX::BI__nvvm_ldg_f4:
17352 case NVPTX::BI__nvvm_ldg_d:
17353 case NVPTX::BI__nvvm_ldg_d2:
17354 return MakeLdg(Intrinsic::nvvm_ldg_global_f);
17355
17356 case NVPTX::BI__nvvm_atom_cta_add_gen_i:
17357 case NVPTX::BI__nvvm_atom_cta_add_gen_l:
17358 case NVPTX::BI__nvvm_atom_cta_add_gen_ll:
17359 return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_cta);
17360 case NVPTX::BI__nvvm_atom_sys_add_gen_i:
17361 case NVPTX::BI__nvvm_atom_sys_add_gen_l:
17362 case NVPTX::BI__nvvm_atom_sys_add_gen_ll:
17363 return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_sys);
17364 case NVPTX::BI__nvvm_atom_cta_add_gen_f:
17365 case NVPTX::BI__nvvm_atom_cta_add_gen_d:
17366 return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_cta);
17367 case NVPTX::BI__nvvm_atom_sys_add_gen_f:
17368 case NVPTX::BI__nvvm_atom_sys_add_gen_d:
17369 return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_sys);
17370 case NVPTX::BI__nvvm_atom_cta_xchg_gen_i:
17371 case NVPTX::BI__nvvm_atom_cta_xchg_gen_l:
17372 case NVPTX::BI__nvvm_atom_cta_xchg_gen_ll:
17373 return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_cta);
17374 case NVPTX::BI__nvvm_atom_sys_xchg_gen_i:
17375 case NVPTX::BI__nvvm_atom_sys_xchg_gen_l:
17376 case NVPTX::BI__nvvm_atom_sys_xchg_gen_ll:
17377 return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_sys);
17378 case NVPTX::BI__nvvm_atom_cta_max_gen_i:
17379 case NVPTX::BI__nvvm_atom_cta_max_gen_ui:
17380 case NVPTX::BI__nvvm_atom_cta_max_gen_l:
17381 case NVPTX::BI__nvvm_atom_cta_max_gen_ul:
17382 case NVPTX::BI__nvvm_atom_cta_max_gen_ll:
17383 case NVPTX::BI__nvvm_atom_cta_max_gen_ull:
17384 return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_cta);
17385 case NVPTX::BI__nvvm_atom_sys_max_gen_i:
17386 case NVPTX::BI__nvvm_atom_sys_max_gen_ui:
17387 case NVPTX::BI__nvvm_atom_sys_max_gen_l:
17388 case NVPTX::BI__nvvm_atom_sys_max_gen_ul:
17389 case NVPTX::BI__nvvm_atom_sys_max_gen_ll:
17390 case NVPTX::BI__nvvm_atom_sys_max_gen_ull:
17391 return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_sys);
17392 case NVPTX::BI__nvvm_atom_cta_min_gen_i:
17393 case NVPTX::BI__nvvm_atom_cta_min_gen_ui:
17394 case NVPTX::BI__nvvm_atom_cta_min_gen_l:
17395 case NVPTX::BI__nvvm_atom_cta_min_gen_ul:
17396 case NVPTX::BI__nvvm_atom_cta_min_gen_ll:
17397 case NVPTX::BI__nvvm_atom_cta_min_gen_ull:
17398 return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_cta);
17399 case NVPTX::BI__nvvm_atom_sys_min_gen_i:
17400 case NVPTX::BI__nvvm_atom_sys_min_gen_ui:
17401 case NVPTX::BI__nvvm_atom_sys_min_gen_l:
17402 case NVPTX::BI__nvvm_atom_sys_min_gen_ul:
17403 case NVPTX::BI__nvvm_atom_sys_min_gen_ll:
17404 case NVPTX::BI__nvvm_atom_sys_min_gen_ull:
17405 return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_sys);
17406 case NVPTX::BI__nvvm_atom_cta_inc_gen_ui:
17407 return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_cta);
17408 case NVPTX::BI__nvvm_atom_cta_dec_gen_ui:
17409 return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_cta);
17410 case NVPTX::BI__nvvm_atom_sys_inc_gen_ui:
17411 return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_sys);
17412 case NVPTX::BI__nvvm_atom_sys_dec_gen_ui:
17413 return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_sys);
17414 case NVPTX::BI__nvvm_atom_cta_and_gen_i:
17415 case NVPTX::BI__nvvm_atom_cta_and_gen_l:
17416 case NVPTX::BI__nvvm_atom_cta_and_gen_ll:
17417 return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_cta);
17418 case NVPTX::BI__nvvm_atom_sys_and_gen_i:
17419 case NVPTX::BI__nvvm_atom_sys_and_gen_l:
17420 case NVPTX::BI__nvvm_atom_sys_and_gen_ll:
17421 return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_sys);
17422 case NVPTX::BI__nvvm_atom_cta_or_gen_i:
17423 case NVPTX::BI__nvvm_atom_cta_or_gen_l:
17424 case NVPTX::BI__nvvm_atom_cta_or_gen_ll:
17425 return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_cta);
17426 case NVPTX::BI__nvvm_atom_sys_or_gen_i:
17427 case NVPTX::BI__nvvm_atom_sys_or_gen_l:
17428 case NVPTX::BI__nvvm_atom_sys_or_gen_ll:
17429 return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_sys);
17430 case NVPTX::BI__nvvm_atom_cta_xor_gen_i:
17431 case NVPTX::BI__nvvm_atom_cta_xor_gen_l:
17432 case NVPTX::BI__nvvm_atom_cta_xor_gen_ll:
17433 return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_cta);
17434 case NVPTX::BI__nvvm_atom_sys_xor_gen_i:
17435 case NVPTX::BI__nvvm_atom_sys_xor_gen_l:
17436 case NVPTX::BI__nvvm_atom_sys_xor_gen_ll:
17437 return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_sys);
17438 case NVPTX::BI__nvvm_atom_cta_cas_gen_i:
17439 case NVPTX::BI__nvvm_atom_cta_cas_gen_l:
17440 case NVPTX::BI__nvvm_atom_cta_cas_gen_ll: {
17441 Value *Ptr = EmitScalarExpr(E->getArg(0));
17442 return Builder.CreateCall(
17443 CGM.getIntrinsic(
17444 Intrinsic::nvvm_atomic_cas_gen_i_cta,
17445 {Ptr->getType()->getPointerElementType(), Ptr->getType()}),
17446 {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))});
17447 }
17448 case NVPTX::BI__nvvm_atom_sys_cas_gen_i:
17449 case NVPTX::BI__nvvm_atom_sys_cas_gen_l:
17450 case NVPTX::BI__nvvm_atom_sys_cas_gen_ll: {
17451 Value *Ptr = EmitScalarExpr(E->getArg(0));
17452 return Builder.CreateCall(
17453 CGM.getIntrinsic(
17454 Intrinsic::nvvm_atomic_cas_gen_i_sys,
17455 {Ptr->getType()->getPointerElementType(), Ptr->getType()}),
17456 {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))});
17457 }
17458 case NVPTX::BI__nvvm_match_all_sync_i32p:
17459 case NVPTX::BI__nvvm_match_all_sync_i64p: {
17460 Value *Mask = EmitScalarExpr(E->getArg(0));
17461 Value *Val = EmitScalarExpr(E->getArg(1));
17462 Address PredOutPtr = EmitPointerWithAlignment(E->getArg(2));
17463 Value *ResultPair = Builder.CreateCall(
17464 CGM.getIntrinsic(BuiltinID == NVPTX::BI__nvvm_match_all_sync_i32p
17465 ? Intrinsic::nvvm_match_all_sync_i32p
17466 : Intrinsic::nvvm_match_all_sync_i64p),
17467 {Mask, Val});
17468 Value *Pred = Builder.CreateZExt(Builder.CreateExtractValue(ResultPair, 1),
17469 PredOutPtr.getElementType());
17470 Builder.CreateStore(Pred, PredOutPtr);
17471 return Builder.CreateExtractValue(ResultPair, 0);
17472 }
17473
17474 // FP MMA loads
17475 case NVPTX::BI__hmma_m16n16k16_ld_a:
17476 case NVPTX::BI__hmma_m16n16k16_ld_b:
17477 case NVPTX::BI__hmma_m16n16k16_ld_c_f16:
17478 case NVPTX::BI__hmma_m16n16k16_ld_c_f32:
17479 case NVPTX::BI__hmma_m32n8k16_ld_a:
17480 case NVPTX::BI__hmma_m32n8k16_ld_b:
17481 case NVPTX::BI__hmma_m32n8k16_ld_c_f16:
17482 case NVPTX::BI__hmma_m32n8k16_ld_c_f32:
17483 case NVPTX::BI__hmma_m8n32k16_ld_a:
17484 case NVPTX::BI__hmma_m8n32k16_ld_b:
17485 case NVPTX::BI__hmma_m8n32k16_ld_c_f16:
17486 case NVPTX::BI__hmma_m8n32k16_ld_c_f32:
17487 // Integer MMA loads.
17488 case NVPTX::BI__imma_m16n16k16_ld_a_s8:
17489 case NVPTX::BI__imma_m16n16k16_ld_a_u8:
17490 case NVPTX::BI__imma_m16n16k16_ld_b_s8:
17491 case NVPTX::BI__imma_m16n16k16_ld_b_u8:
17492 case NVPTX::BI__imma_m16n16k16_ld_c:
17493 case NVPTX::BI__imma_m32n8k16_ld_a_s8:
17494 case NVPTX::BI__imma_m32n8k16_ld_a_u8:
17495 case NVPTX::BI__imma_m32n8k16_ld_b_s8:
17496 case NVPTX::BI__imma_m32n8k16_ld_b_u8:
17497 case NVPTX::BI__imma_m32n8k16_ld_c:
17498 case NVPTX::BI__imma_m8n32k16_ld_a_s8:
17499 case NVPTX::BI__imma_m8n32k16_ld_a_u8:
17500 case NVPTX::BI__imma_m8n32k16_ld_b_s8:
17501 case NVPTX::BI__imma_m8n32k16_ld_b_u8:
17502 case NVPTX::BI__imma_m8n32k16_ld_c:
17503 // Sub-integer MMA loads.
17504 case NVPTX::BI__imma_m8n8k32_ld_a_s4:
17505 case NVPTX::BI__imma_m8n8k32_ld_a_u4:
17506 case NVPTX::BI__imma_m8n8k32_ld_b_s4:
17507 case NVPTX::BI__imma_m8n8k32_ld_b_u4:
17508 case NVPTX::BI__imma_m8n8k32_ld_c:
17509 case NVPTX::BI__bmma_m8n8k128_ld_a_b1:
17510 case NVPTX::BI__bmma_m8n8k128_ld_b_b1:
17511 case NVPTX::BI__bmma_m8n8k128_ld_c:
17512 // Double MMA loads.
17513 case NVPTX::BI__dmma_m8n8k4_ld_a:
17514 case NVPTX::BI__dmma_m8n8k4_ld_b:
17515 case NVPTX::BI__dmma_m8n8k4_ld_c:
17516 // Alternate float MMA loads.
17517 case NVPTX::BI__mma_bf16_m16n16k16_ld_a:
17518 case NVPTX::BI__mma_bf16_m16n16k16_ld_b:
17519 case NVPTX::BI__mma_bf16_m8n32k16_ld_a:
17520 case NVPTX::BI__mma_bf16_m8n32k16_ld_b:
17521 case NVPTX::BI__mma_bf16_m32n8k16_ld_a:
17522 case NVPTX::BI__mma_bf16_m32n8k16_ld_b:
17523 case NVPTX::BI__mma_tf32_m16n16k8_ld_a:
17524 case NVPTX::BI__mma_tf32_m16n16k8_ld_b:
17525 case NVPTX::BI__mma_tf32_m16n16k8_ld_c: {
17526 Address Dst = EmitPointerWithAlignment(E->getArg(0));
17527 Value *Src = EmitScalarExpr(E->getArg(1));
17528 Value *Ldm = EmitScalarExpr(E->getArg(2));
17529 Optional<llvm::APSInt> isColMajorArg =
17530 E->getArg(3)->getIntegerConstantExpr(getContext());
17531 if (!isColMajorArg)
17532 return nullptr;
17533 bool isColMajor = isColMajorArg->getSExtValue();
17534 NVPTXMmaLdstInfo II = getNVPTXMmaLdstInfo(BuiltinID);
17535 unsigned IID = isColMajor ? II.IID_col : II.IID_row;
17536 if (IID == 0)
17537 return nullptr;
17538
17539 Value *Result =
17540 Builder.CreateCall(CGM.getIntrinsic(IID, Src->getType()), {Src, Ldm});
17541
17542 // Save returned values.
17543 assert(II.NumResults)(static_cast <bool> (II.NumResults) ? void (0) : __assert_fail
("II.NumResults", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 17543, __extension__ __PRETTY_FUNCTION__))
;
17544 if (II.NumResults == 1) {
17545 Builder.CreateAlignedStore(Result, Dst.getPointer(),
17546 CharUnits::fromQuantity(4));
17547 } else {
17548 for (unsigned i = 0; i < II.NumResults; ++i) {
17549 Builder.CreateAlignedStore(
17550 Builder.CreateBitCast(Builder.CreateExtractValue(Result, i),
17551 Dst.getElementType()),
17552 Builder.CreateGEP(Dst.getElementType(), Dst.getPointer(),
17553 llvm::ConstantInt::get(IntTy, i)),
17554 CharUnits::fromQuantity(4));
17555 }
17556 }
17557 return Result;
17558 }
17559
17560 case NVPTX::BI__hmma_m16n16k16_st_c_f16:
17561 case NVPTX::BI__hmma_m16n16k16_st_c_f32:
17562 case NVPTX::BI__hmma_m32n8k16_st_c_f16:
17563 case NVPTX::BI__hmma_m32n8k16_st_c_f32:
17564 case NVPTX::BI__hmma_m8n32k16_st_c_f16:
17565 case NVPTX::BI__hmma_m8n32k16_st_c_f32:
17566 case NVPTX::BI__imma_m16n16k16_st_c_i32:
17567 case NVPTX::BI__imma_m32n8k16_st_c_i32:
17568 case NVPTX::BI__imma_m8n32k16_st_c_i32:
17569 case NVPTX::BI__imma_m8n8k32_st_c_i32:
17570 case NVPTX::BI__bmma_m8n8k128_st_c_i32:
17571 case NVPTX::BI__dmma_m8n8k4_st_c_f64:
17572 case NVPTX::BI__mma_m16n16k8_st_c_f32: {
17573 Value *Dst = EmitScalarExpr(E->getArg(0));
17574 Address Src = EmitPointerWithAlignment(E->getArg(1));
17575 Value *Ldm = EmitScalarExpr(E->getArg(2));
17576 Optional<llvm::APSInt> isColMajorArg =
17577 E->getArg(3)->getIntegerConstantExpr(getContext());
17578 if (!isColMajorArg)
17579 return nullptr;
17580 bool isColMajor = isColMajorArg->getSExtValue();
17581 NVPTXMmaLdstInfo II = getNVPTXMmaLdstInfo(BuiltinID);
17582 unsigned IID = isColMajor ? II.IID_col : II.IID_row;
17583 if (IID == 0)
17584 return nullptr;
17585 Function *Intrinsic =
17586 CGM.getIntrinsic(IID, Dst->getType());
17587 llvm::Type *ParamType = Intrinsic->getFunctionType()->getParamType(1);
17588 SmallVector<Value *, 10> Values = {Dst};
17589 for (unsigned i = 0; i < II.NumResults; ++i) {
17590 Value *V = Builder.CreateAlignedLoad(
17591 Src.getElementType(),
17592 Builder.CreateGEP(Src.getElementType(), Src.getPointer(),
17593 llvm::ConstantInt::get(IntTy, i)),
17594 CharUnits::fromQuantity(4));
17595 Values.push_back(Builder.CreateBitCast(V, ParamType));
17596 }
17597 Values.push_back(Ldm);
17598 Value *Result = Builder.CreateCall(Intrinsic, Values);
17599 return Result;
17600 }
17601
17602 // BI__hmma_m16n16k16_mma_<Dtype><CType>(d, a, b, c, layout, satf) -->
17603 // Intrinsic::nvvm_wmma_m16n16k16_mma_sync<layout A,B><DType><CType><Satf>
17604 case NVPTX::BI__hmma_m16n16k16_mma_f16f16:
17605 case NVPTX::BI__hmma_m16n16k16_mma_f32f16:
17606 case NVPTX::BI__hmma_m16n16k16_mma_f32f32:
17607 case NVPTX::BI__hmma_m16n16k16_mma_f16f32:
17608 case NVPTX::BI__hmma_m32n8k16_mma_f16f16:
17609 case NVPTX::BI__hmma_m32n8k16_mma_f32f16:
17610 case NVPTX::BI__hmma_m32n8k16_mma_f32f32:
17611 case NVPTX::BI__hmma_m32n8k16_mma_f16f32:
17612 case NVPTX::BI__hmma_m8n32k16_mma_f16f16:
17613 case NVPTX::BI__hmma_m8n32k16_mma_f32f16:
17614 case NVPTX::BI__hmma_m8n32k16_mma_f32f32:
17615 case NVPTX::BI__hmma_m8n32k16_mma_f16f32:
17616 case NVPTX::BI__imma_m16n16k16_mma_s8:
17617 case NVPTX::BI__imma_m16n16k16_mma_u8:
17618 case NVPTX::BI__imma_m32n8k16_mma_s8:
17619 case NVPTX::BI__imma_m32n8k16_mma_u8:
17620 case NVPTX::BI__imma_m8n32k16_mma_s8:
17621 case NVPTX::BI__imma_m8n32k16_mma_u8:
17622 case NVPTX::BI__imma_m8n8k32_mma_s4:
17623 case NVPTX::BI__imma_m8n8k32_mma_u4:
17624 case NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1:
17625 case NVPTX::BI__bmma_m8n8k128_mma_and_popc_b1:
17626 case NVPTX::BI__dmma_m8n8k4_mma_f64:
17627 case NVPTX::BI__mma_bf16_m16n16k16_mma_f32:
17628 case NVPTX::BI__mma_bf16_m8n32k16_mma_f32:
17629 case NVPTX::BI__mma_bf16_m32n8k16_mma_f32:
17630 case NVPTX::BI__mma_tf32_m16n16k8_mma_f32: {
17631 Address Dst = EmitPointerWithAlignment(E->getArg(0));
17632 Address SrcA = EmitPointerWithAlignment(E->getArg(1));
17633 Address SrcB = EmitPointerWithAlignment(E->getArg(2));
17634 Address SrcC = EmitPointerWithAlignment(E->getArg(3));
17635 Optional<llvm::APSInt> LayoutArg =
17636 E->getArg(4)->getIntegerConstantExpr(getContext());
17637 if (!LayoutArg)
17638 return nullptr;
17639 int Layout = LayoutArg->getSExtValue();
17640 if (Layout < 0 || Layout > 3)
17641 return nullptr;
17642 llvm::APSInt SatfArg;
17643 if (BuiltinID == NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1 ||
17644 BuiltinID == NVPTX::BI__bmma_m8n8k128_mma_and_popc_b1)
17645 SatfArg = 0; // .b1 does not have satf argument.
17646 else if (Optional<llvm::APSInt> OptSatfArg =
17647 E->getArg(5)->getIntegerConstantExpr(getContext()))
17648 SatfArg = *OptSatfArg;
17649 else
17650 return nullptr;
17651 bool Satf = SatfArg.getSExtValue();
17652 NVPTXMmaInfo MI = getNVPTXMmaInfo(BuiltinID);
17653 unsigned IID = MI.getMMAIntrinsic(Layout, Satf);
17654 if (IID == 0) // Unsupported combination of Layout/Satf.
17655 return nullptr;
17656
17657 SmallVector<Value *, 24> Values;
17658 Function *Intrinsic = CGM.getIntrinsic(IID);
17659 llvm::Type *AType = Intrinsic->getFunctionType()->getParamType(0);
17660 // Load A
17661 for (unsigned i = 0; i < MI.NumEltsA; ++i) {
17662 Value *V = Builder.CreateAlignedLoad(
17663 SrcA.getElementType(),
17664 Builder.CreateGEP(SrcA.getElementType(), SrcA.getPointer(),
17665 llvm::ConstantInt::get(IntTy, i)),
17666 CharUnits::fromQuantity(4));
17667 Values.push_back(Builder.CreateBitCast(V, AType));
17668 }
17669 // Load B
17670 llvm::Type *BType = Intrinsic->getFunctionType()->getParamType(MI.NumEltsA);
17671 for (unsigned i = 0; i < MI.NumEltsB; ++i) {
17672 Value *V = Builder.CreateAlignedLoad(
17673 SrcB.getElementType(),
17674 Builder.CreateGEP(SrcB.getElementType(), SrcB.getPointer(),
17675 llvm::ConstantInt::get(IntTy, i)),
17676 CharUnits::fromQuantity(4));
17677 Values.push_back(Builder.CreateBitCast(V, BType));
17678 }
17679 // Load C
17680 llvm::Type *CType =
17681 Intrinsic->getFunctionType()->getParamType(MI.NumEltsA + MI.NumEltsB);
17682 for (unsigned i = 0; i < MI.NumEltsC; ++i) {
17683 Value *V = Builder.CreateAlignedLoad(
17684 SrcC.getElementType(),
17685 Builder.CreateGEP(SrcC.getElementType(), SrcC.getPointer(),
17686 llvm::ConstantInt::get(IntTy, i)),
17687 CharUnits::fromQuantity(4));
17688 Values.push_back(Builder.CreateBitCast(V, CType));
17689 }
17690 Value *Result = Builder.CreateCall(Intrinsic, Values);
17691 llvm::Type *DType = Dst.getElementType();
17692 for (unsigned i = 0; i < MI.NumEltsD; ++i)
17693 Builder.CreateAlignedStore(
17694 Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), DType),
17695 Builder.CreateGEP(Dst.getElementType(), Dst.getPointer(),
17696 llvm::ConstantInt::get(IntTy, i)),
17697 CharUnits::fromQuantity(4));
17698 return Result;
17699 }
17700 default:
17701 return nullptr;
17702 }
17703}
17704
17705namespace {
17706struct BuiltinAlignArgs {
17707 llvm::Value *Src = nullptr;
17708 llvm::Type *SrcType = nullptr;
17709 llvm::Value *Alignment = nullptr;
17710 llvm::Value *Mask = nullptr;
17711 llvm::IntegerType *IntType = nullptr;
17712
17713 BuiltinAlignArgs(const CallExpr *E, CodeGenFunction &CGF) {
17714 QualType AstType = E->getArg(0)->getType();
17715 if (AstType->isArrayType())
17716 Src = CGF.EmitArrayToPointerDecay(E->getArg(0)).getPointer();
17717 else
17718 Src = CGF.EmitScalarExpr(E->getArg(0));
17719 SrcType = Src->getType();
17720 if (SrcType->isPointerTy()) {
17721 IntType = IntegerType::get(
17722 CGF.getLLVMContext(),
17723 CGF.CGM.getDataLayout().getIndexTypeSizeInBits(SrcType));
17724 } else {
17725 assert(SrcType->isIntegerTy())(static_cast <bool> (SrcType->isIntegerTy()) ? void (
0) : __assert_fail ("SrcType->isIntegerTy()", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 17725, __extension__ __PRETTY_FUNCTION__))
;
17726 IntType = cast<llvm::IntegerType>(SrcType);
17727 }
17728 Alignment = CGF.EmitScalarExpr(E->getArg(1));
17729 Alignment = CGF.Builder.CreateZExtOrTrunc(Alignment, IntType, "alignment");
17730 auto *One = llvm::ConstantInt::get(IntType, 1);
17731 Mask = CGF.Builder.CreateSub(Alignment, One, "mask");
17732 }
17733};
17734} // namespace
17735
17736/// Generate (x & (y-1)) == 0.
17737RValue CodeGenFunction::EmitBuiltinIsAligned(const CallExpr *E) {
17738 BuiltinAlignArgs Args(E, *this);
17739 llvm::Value *SrcAddress = Args.Src;
17740 if (Args.SrcType->isPointerTy())
17741 SrcAddress =
17742 Builder.CreateBitOrPointerCast(Args.Src, Args.IntType, "src_addr");
17743 return RValue::get(Builder.CreateICmpEQ(
17744 Builder.CreateAnd(SrcAddress, Args.Mask, "set_bits"),
17745 llvm::Constant::getNullValue(Args.IntType), "is_aligned"));
17746}
17747
17748/// Generate (x & ~(y-1)) to align down or ((x+(y-1)) & ~(y-1)) to align up.
17749/// Note: For pointer types we can avoid ptrtoint/inttoptr pairs by using the
17750/// llvm.ptrmask instrinsic (with a GEP before in the align_up case).
17751/// TODO: actually use ptrmask once most optimization passes know about it.
17752RValue CodeGenFunction::EmitBuiltinAlignTo(const CallExpr *E, bool AlignUp) {
17753 BuiltinAlignArgs Args(E, *this);
17754 llvm::Value *SrcAddr = Args.Src;
17755 if (Args.Src->getType()->isPointerTy())
17756 SrcAddr = Builder.CreatePtrToInt(Args.Src, Args.IntType, "intptr");
17757 llvm::Value *SrcForMask = SrcAddr;
17758 if (AlignUp) {
17759 // When aligning up we have to first add the mask to ensure we go over the
17760 // next alignment value and then align down to the next valid multiple.
17761 // By adding the mask, we ensure that align_up on an already aligned
17762 // value will not change the value.
17763 SrcForMask = Builder.CreateAdd(SrcForMask, Args.Mask, "over_boundary");
17764 }
17765 // Invert the mask to only clear the lower bits.
17766 llvm::Value *InvertedMask = Builder.CreateNot(Args.Mask, "inverted_mask");
17767 llvm::Value *Result =
17768 Builder.CreateAnd(SrcForMask, InvertedMask, "aligned_result");
17769 if (Args.Src->getType()->isPointerTy()) {
17770 /// TODO: Use ptrmask instead of ptrtoint+gep once it is optimized well.
17771 // Result = Builder.CreateIntrinsic(
17772 // Intrinsic::ptrmask, {Args.SrcType, SrcForMask->getType(), Args.IntType},
17773 // {SrcForMask, NegatedMask}, nullptr, "aligned_result");
17774 Result->setName("aligned_intptr");
17775 llvm::Value *Difference = Builder.CreateSub(Result, SrcAddr, "diff");
17776 // The result must point to the same underlying allocation. This means we
17777 // can use an inbounds GEP to enable better optimization.
17778 Value *Base = EmitCastToVoidPtr(Args.Src);
17779 if (getLangOpts().isSignedOverflowDefined())
17780 Result = Builder.CreateGEP(Int8Ty, Base, Difference, "aligned_result");
17781 else
17782 Result = EmitCheckedInBoundsGEP(Base, Difference,
17783 /*SignedIndices=*/true,
17784 /*isSubtraction=*/!AlignUp,
17785 E->getExprLoc(), "aligned_result");
17786 Result = Builder.CreatePointerCast(Result, Args.SrcType);
17787 // Emit an alignment assumption to ensure that the new alignment is
17788 // propagated to loads/stores, etc.
17789 emitAlignmentAssumption(Result, E, E->getExprLoc(), Args.Alignment);
17790 }
17791 assert(Result->getType() == Args.SrcType)(static_cast <bool> (Result->getType() == Args.SrcType
) ? void (0) : __assert_fail ("Result->getType() == Args.SrcType"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 17791, __extension__ __PRETTY_FUNCTION__))
;
17792 return RValue::get(Result);
17793}
17794
17795Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
17796 const CallExpr *E) {
17797 switch (BuiltinID) {
17798 case WebAssembly::BI__builtin_wasm_memory_size: {
17799 llvm::Type *ResultType = ConvertType(E->getType());
17800 Value *I = EmitScalarExpr(E->getArg(0));
17801 Function *Callee =
17802 CGM.getIntrinsic(Intrinsic::wasm_memory_size, ResultType);
17803 return Builder.CreateCall(Callee, I);
17804 }
17805 case WebAssembly::BI__builtin_wasm_memory_grow: {
17806 llvm::Type *ResultType = ConvertType(E->getType());
17807 Value *Args[] = {EmitScalarExpr(E->getArg(0)),
17808 EmitScalarExpr(E->getArg(1))};
17809 Function *Callee =
17810 CGM.getIntrinsic(Intrinsic::wasm_memory_grow, ResultType);
17811 return Builder.CreateCall(Callee, Args);
17812 }
17813 case WebAssembly::BI__builtin_wasm_tls_size: {
17814 llvm::Type *ResultType = ConvertType(E->getType());
17815 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_size, ResultType);
17816 return Builder.CreateCall(Callee);
17817 }
17818 case WebAssembly::BI__builtin_wasm_tls_align: {
17819 llvm::Type *ResultType = ConvertType(E->getType());
17820 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_align, ResultType);
17821 return Builder.CreateCall(Callee);
17822 }
17823 case WebAssembly::BI__builtin_wasm_tls_base: {
17824 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_base);
17825 return Builder.CreateCall(Callee);
17826 }
17827 case WebAssembly::BI__builtin_wasm_throw: {
17828 Value *Tag = EmitScalarExpr(E->getArg(0));
17829 Value *Obj = EmitScalarExpr(E->getArg(1));
17830 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_throw);
17831 return Builder.CreateCall(Callee, {Tag, Obj});
17832 }
17833 case WebAssembly::BI__builtin_wasm_rethrow: {
17834 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_rethrow);
17835 return Builder.CreateCall(Callee);
17836 }
17837 case WebAssembly::BI__builtin_wasm_memory_atomic_wait32: {
17838 Value *Addr = EmitScalarExpr(E->getArg(0));
17839 Value *Expected = EmitScalarExpr(E->getArg(1));
17840 Value *Timeout = EmitScalarExpr(E->getArg(2));
17841 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_atomic_wait32);
17842 return Builder.CreateCall(Callee, {Addr, Expected, Timeout});
17843 }
17844 case WebAssembly::BI__builtin_wasm_memory_atomic_wait64: {
17845 Value *Addr = EmitScalarExpr(E->getArg(0));
17846 Value *Expected = EmitScalarExpr(E->getArg(1));
17847 Value *Timeout = EmitScalarExpr(E->getArg(2));
17848 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_atomic_wait64);
17849 return Builder.CreateCall(Callee, {Addr, Expected, Timeout});
17850 }
17851 case WebAssembly::BI__builtin_wasm_memory_atomic_notify: {
17852 Value *Addr = EmitScalarExpr(E->getArg(0));
17853 Value *Count = EmitScalarExpr(E->getArg(1));
17854 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_atomic_notify);
17855 return Builder.CreateCall(Callee, {Addr, Count});
17856 }
17857 case WebAssembly::BI__builtin_wasm_trunc_s_i32_f32:
17858 case WebAssembly::BI__builtin_wasm_trunc_s_i32_f64:
17859 case WebAssembly::BI__builtin_wasm_trunc_s_i64_f32:
17860 case WebAssembly::BI__builtin_wasm_trunc_s_i64_f64: {
17861 Value *Src = EmitScalarExpr(E->getArg(0));
17862 llvm::Type *ResT = ConvertType(E->getType());
17863 Function *Callee =
17864 CGM.getIntrinsic(Intrinsic::wasm_trunc_signed, {ResT, Src->getType()});
17865 return Builder.CreateCall(Callee, {Src});
17866 }
17867 case WebAssembly::BI__builtin_wasm_trunc_u_i32_f32:
17868 case WebAssembly::BI__builtin_wasm_trunc_u_i32_f64:
17869 case WebAssembly::BI__builtin_wasm_trunc_u_i64_f32:
17870 case WebAssembly::BI__builtin_wasm_trunc_u_i64_f64: {
17871 Value *Src = EmitScalarExpr(E->getArg(0));
17872 llvm::Type *ResT = ConvertType(E->getType());
17873 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_unsigned,
17874 {ResT, Src->getType()});
17875 return Builder.CreateCall(Callee, {Src});
17876 }
17877 case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f32:
17878 case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f64:
17879 case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f32:
17880 case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f64:
17881 case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32x4_f32x4: {
17882 Value *Src = EmitScalarExpr(E->getArg(0));
17883 llvm::Type *ResT = ConvertType(E->getType());
17884 Function *Callee =
17885 CGM.getIntrinsic(Intrinsic::fptosi_sat, {ResT, Src->getType()});
17886 return Builder.CreateCall(Callee, {Src});
17887 }
17888 case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f32:
17889 case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f64:
17890 case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f32:
17891 case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f64:
17892 case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32x4_f32x4: {
17893 Value *Src = EmitScalarExpr(E->getArg(0));
17894 llvm::Type *ResT = ConvertType(E->getType());
17895 Function *Callee =
17896 CGM.getIntrinsic(Intrinsic::fptoui_sat, {ResT, Src->getType()});
17897 return Builder.CreateCall(Callee, {Src});
17898 }
17899 case WebAssembly::BI__builtin_wasm_min_f32:
17900 case WebAssembly::BI__builtin_wasm_min_f64:
17901 case WebAssembly::BI__builtin_wasm_min_f32x4:
17902 case WebAssembly::BI__builtin_wasm_min_f64x2: {
17903 Value *LHS = EmitScalarExpr(E->getArg(0));
17904 Value *RHS = EmitScalarExpr(E->getArg(1));
17905 Function *Callee =
17906 CGM.getIntrinsic(Intrinsic::minimum, ConvertType(E->getType()));
17907 return Builder.CreateCall(Callee, {LHS, RHS});
17908 }
17909 case WebAssembly::BI__builtin_wasm_max_f32:
17910 case WebAssembly::BI__builtin_wasm_max_f64:
17911 case WebAssembly::BI__builtin_wasm_max_f32x4:
17912 case WebAssembly::BI__builtin_wasm_max_f64x2: {
17913 Value *LHS = EmitScalarExpr(E->getArg(0));
17914 Value *RHS = EmitScalarExpr(E->getArg(1));
17915 Function *Callee =
17916 CGM.getIntrinsic(Intrinsic::maximum, ConvertType(E->getType()));
17917 return Builder.CreateCall(Callee, {LHS, RHS});
17918 }
17919 case WebAssembly::BI__builtin_wasm_pmin_f32x4:
17920 case WebAssembly::BI__builtin_wasm_pmin_f64x2: {
17921 Value *LHS = EmitScalarExpr(E->getArg(0));
17922 Value *RHS = EmitScalarExpr(E->getArg(1));
17923 Function *Callee =
17924 CGM.getIntrinsic(Intrinsic::wasm_pmin, ConvertType(E->getType()));
17925 return Builder.CreateCall(Callee, {LHS, RHS});
17926 }
17927 case WebAssembly::BI__builtin_wasm_pmax_f32x4:
17928 case WebAssembly::BI__builtin_wasm_pmax_f64x2: {
17929 Value *LHS = EmitScalarExpr(E->getArg(0));
17930 Value *RHS = EmitScalarExpr(E->getArg(1));
17931 Function *Callee =
17932 CGM.getIntrinsic(Intrinsic::wasm_pmax, ConvertType(E->getType()));
17933 return Builder.CreateCall(Callee, {LHS, RHS});
17934 }
17935 case WebAssembly::BI__builtin_wasm_ceil_f32x4:
17936 case WebAssembly::BI__builtin_wasm_floor_f32x4:
17937 case WebAssembly::BI__builtin_wasm_trunc_f32x4:
17938 case WebAssembly::BI__builtin_wasm_nearest_f32x4:
17939 case WebAssembly::BI__builtin_wasm_ceil_f64x2:
17940 case WebAssembly::BI__builtin_wasm_floor_f64x2:
17941 case WebAssembly::BI__builtin_wasm_trunc_f64x2:
17942 case WebAssembly::BI__builtin_wasm_nearest_f64x2: {
17943 unsigned IntNo;
17944 switch (BuiltinID) {
17945 case WebAssembly::BI__builtin_wasm_ceil_f32x4:
17946 case WebAssembly::BI__builtin_wasm_ceil_f64x2:
17947 IntNo = Intrinsic::ceil;
17948 break;
17949 case WebAssembly::BI__builtin_wasm_floor_f32x4:
17950 case WebAssembly::BI__builtin_wasm_floor_f64x2:
17951 IntNo = Intrinsic::floor;
17952 break;
17953 case WebAssembly::BI__builtin_wasm_trunc_f32x4:
17954 case WebAssembly::BI__builtin_wasm_trunc_f64x2:
17955 IntNo = Intrinsic::trunc;
17956 break;
17957 case WebAssembly::BI__builtin_wasm_nearest_f32x4:
17958 case WebAssembly::BI__builtin_wasm_nearest_f64x2:
17959 IntNo = Intrinsic::nearbyint;
17960 break;
17961 default:
17962 llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 17962)
;
17963 }
17964 Value *Value = EmitScalarExpr(E->getArg(0));
17965 Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType()));
17966 return Builder.CreateCall(Callee, Value);
17967 }
17968 case WebAssembly::BI__builtin_wasm_swizzle_i8x16: {
17969 Value *Src = EmitScalarExpr(E->getArg(0));
17970 Value *Indices = EmitScalarExpr(E->getArg(1));
17971 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_swizzle);
17972 return Builder.CreateCall(Callee, {Src, Indices});
17973 }
17974 case WebAssembly::BI__builtin_wasm_add_sat_s_i8x16:
17975 case WebAssembly::BI__builtin_wasm_add_sat_u_i8x16:
17976 case WebAssembly::BI__builtin_wasm_add_sat_s_i16x8:
17977 case WebAssembly::BI__builtin_wasm_add_sat_u_i16x8:
17978 case WebAssembly::BI__builtin_wasm_sub_sat_s_i8x16:
17979 case WebAssembly::BI__builtin_wasm_sub_sat_u_i8x16:
17980 case WebAssembly::BI__builtin_wasm_sub_sat_s_i16x8:
17981 case WebAssembly::BI__builtin_wasm_sub_sat_u_i16x8: {
17982 unsigned IntNo;
17983 switch (BuiltinID) {
17984 case WebAssembly::BI__builtin_wasm_add_sat_s_i8x16:
17985 case WebAssembly::BI__builtin_wasm_add_sat_s_i16x8:
17986 IntNo = Intrinsic::sadd_sat;
17987 break;
17988 case WebAssembly::BI__builtin_wasm_add_sat_u_i8x16:
17989 case WebAssembly::BI__builtin_wasm_add_sat_u_i16x8:
17990 IntNo = Intrinsic::uadd_sat;
17991 break;
17992 case WebAssembly::BI__builtin_wasm_sub_sat_s_i8x16:
17993 case WebAssembly::BI__builtin_wasm_sub_sat_s_i16x8:
17994 IntNo = Intrinsic::wasm_sub_sat_signed;
17995 break;
17996 case WebAssembly::BI__builtin_wasm_sub_sat_u_i8x16:
17997 case WebAssembly::BI__builtin_wasm_sub_sat_u_i16x8:
17998 IntNo = Intrinsic::wasm_sub_sat_unsigned;
17999 break;
18000 default:
18001 llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18001)
;
18002 }
18003 Value *LHS = EmitScalarExpr(E->getArg(0));
18004 Value *RHS = EmitScalarExpr(E->getArg(1));
18005 Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType()));
18006 return Builder.CreateCall(Callee, {LHS, RHS});
18007 }
18008 case WebAssembly::BI__builtin_wasm_abs_i8x16:
18009 case WebAssembly::BI__builtin_wasm_abs_i16x8:
18010 case WebAssembly::BI__builtin_wasm_abs_i32x4:
18011 case WebAssembly::BI__builtin_wasm_abs_i64x2: {
18012 Value *Vec = EmitScalarExpr(E->getArg(0));
18013 Value *Neg = Builder.CreateNeg(Vec, "neg");
18014 Constant *Zero = llvm::Constant::getNullValue(Vec->getType());
18015 Value *ICmp = Builder.CreateICmpSLT(Vec, Zero, "abscond");
18016 return Builder.CreateSelect(ICmp, Neg, Vec, "abs");
18017 }
18018 case WebAssembly::BI__builtin_wasm_min_s_i8x16:
18019 case WebAssembly::BI__builtin_wasm_min_u_i8x16:
18020 case WebAssembly::BI__builtin_wasm_max_s_i8x16:
18021 case WebAssembly::BI__builtin_wasm_max_u_i8x16:
18022 case WebAssembly::BI__builtin_wasm_min_s_i16x8:
18023 case WebAssembly::BI__builtin_wasm_min_u_i16x8:
18024 case WebAssembly::BI__builtin_wasm_max_s_i16x8:
18025 case WebAssembly::BI__builtin_wasm_max_u_i16x8:
18026 case WebAssembly::BI__builtin_wasm_min_s_i32x4:
18027 case WebAssembly::BI__builtin_wasm_min_u_i32x4:
18028 case WebAssembly::BI__builtin_wasm_max_s_i32x4:
18029 case WebAssembly::BI__builtin_wasm_max_u_i32x4: {
18030 Value *LHS = EmitScalarExpr(E->getArg(0));
18031 Value *RHS = EmitScalarExpr(E->getArg(1));
18032 Value *ICmp;
18033 switch (BuiltinID) {
18034 case WebAssembly::BI__builtin_wasm_min_s_i8x16:
18035 case WebAssembly::BI__builtin_wasm_min_s_i16x8:
18036 case WebAssembly::BI__builtin_wasm_min_s_i32x4:
18037 ICmp = Builder.CreateICmpSLT(LHS, RHS);
18038 break;
18039 case WebAssembly::BI__builtin_wasm_min_u_i8x16:
18040 case WebAssembly::BI__builtin_wasm_min_u_i16x8:
18041 case WebAssembly::BI__builtin_wasm_min_u_i32x4:
18042 ICmp = Builder.CreateICmpULT(LHS, RHS);
18043 break;
18044 case WebAssembly::BI__builtin_wasm_max_s_i8x16:
18045 case WebAssembly::BI__builtin_wasm_max_s_i16x8:
18046 case WebAssembly::BI__builtin_wasm_max_s_i32x4:
18047 ICmp = Builder.CreateICmpSGT(LHS, RHS);
18048 break;
18049 case WebAssembly::BI__builtin_wasm_max_u_i8x16:
18050 case WebAssembly::BI__builtin_wasm_max_u_i16x8:
18051 case WebAssembly::BI__builtin_wasm_max_u_i32x4:
18052 ICmp = Builder.CreateICmpUGT(LHS, RHS);
18053 break;
18054 default:
18055 llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18055)
;
18056 }
18057 return Builder.CreateSelect(ICmp, LHS, RHS);
18058 }
18059 case WebAssembly::BI__builtin_wasm_avgr_u_i8x16:
18060 case WebAssembly::BI__builtin_wasm_avgr_u_i16x8: {
18061 Value *LHS = EmitScalarExpr(E->getArg(0));
18062 Value *RHS = EmitScalarExpr(E->getArg(1));
18063 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_avgr_unsigned,
18064 ConvertType(E->getType()));
18065 return Builder.CreateCall(Callee, {LHS, RHS});
18066 }
18067 case WebAssembly::BI__builtin_wasm_q15mulr_sat_s_i16x8: {
18068 Value *LHS = EmitScalarExpr(E->getArg(0));
18069 Value *RHS = EmitScalarExpr(E->getArg(1));
18070 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_q15mulr_sat_signed);
18071 return Builder.CreateCall(Callee, {LHS, RHS});
18072 }
18073 case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_s_i16x8:
18074 case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_u_i16x8:
18075 case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_s_i32x4:
18076 case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_u_i32x4: {
18077 Value *Vec = EmitScalarExpr(E->getArg(0));
18078 unsigned IntNo;
18079 switch (BuiltinID) {
18080 case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_s_i16x8:
18081 case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_s_i32x4:
18082 IntNo = Intrinsic::wasm_extadd_pairwise_signed;
18083 break;
18084 case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_u_i16x8:
18085 case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_u_i32x4:
18086 IntNo = Intrinsic::wasm_extadd_pairwise_unsigned;
18087 break;
18088 default:
18089 llvm_unreachable("unexptected builtin ID")::llvm::llvm_unreachable_internal("unexptected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18089)
;
18090 }
18091
18092 Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType()));
18093 return Builder.CreateCall(Callee, Vec);
18094 }
18095 case WebAssembly::BI__builtin_wasm_bitselect: {
18096 Value *V1 = EmitScalarExpr(E->getArg(0));
18097 Value *V2 = EmitScalarExpr(E->getArg(1));
18098 Value *C = EmitScalarExpr(E->getArg(2));
18099 Function *Callee =
18100 CGM.getIntrinsic(Intrinsic::wasm_bitselect, ConvertType(E->getType()));
18101 return Builder.CreateCall(Callee, {V1, V2, C});
18102 }
18103 case WebAssembly::BI__builtin_wasm_dot_s_i32x4_i16x8: {
18104 Value *LHS = EmitScalarExpr(E->getArg(0));
18105 Value *RHS = EmitScalarExpr(E->getArg(1));
18106 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_dot);
18107 return Builder.CreateCall(Callee, {LHS, RHS});
18108 }
18109 case WebAssembly::BI__builtin_wasm_popcnt_i8x16: {
18110 Value *Vec = EmitScalarExpr(E->getArg(0));
18111 Function *Callee =
18112 CGM.getIntrinsic(Intrinsic::ctpop, ConvertType(E->getType()));
18113 return Builder.CreateCall(Callee, {Vec});
18114 }
18115 case WebAssembly::BI__builtin_wasm_any_true_v128:
18116 case WebAssembly::BI__builtin_wasm_all_true_i8x16:
18117 case WebAssembly::BI__builtin_wasm_all_true_i16x8:
18118 case WebAssembly::BI__builtin_wasm_all_true_i32x4:
18119 case WebAssembly::BI__builtin_wasm_all_true_i64x2: {
18120 unsigned IntNo;
18121 switch (BuiltinID) {
18122 case WebAssembly::BI__builtin_wasm_any_true_v128:
18123 IntNo = Intrinsic::wasm_anytrue;
18124 break;
18125 case WebAssembly::BI__builtin_wasm_all_true_i8x16:
18126 case WebAssembly::BI__builtin_wasm_all_true_i16x8:
18127 case WebAssembly::BI__builtin_wasm_all_true_i32x4:
18128 case WebAssembly::BI__builtin_wasm_all_true_i64x2:
18129 IntNo = Intrinsic::wasm_alltrue;
18130 break;
18131 default:
18132 llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18132)
;
18133 }
18134 Value *Vec = EmitScalarExpr(E->getArg(0));
18135 Function *Callee = CGM.getIntrinsic(IntNo, Vec->getType());
18136 return Builder.CreateCall(Callee, {Vec});
18137 }
18138 case WebAssembly::BI__builtin_wasm_bitmask_i8x16:
18139 case WebAssembly::BI__builtin_wasm_bitmask_i16x8:
18140 case WebAssembly::BI__builtin_wasm_bitmask_i32x4:
18141 case WebAssembly::BI__builtin_wasm_bitmask_i64x2: {
18142 Value *Vec = EmitScalarExpr(E->getArg(0));
18143 Function *Callee =
18144 CGM.getIntrinsic(Intrinsic::wasm_bitmask, Vec->getType());
18145 return Builder.CreateCall(Callee, {Vec});
18146 }
18147 case WebAssembly::BI__builtin_wasm_abs_f32x4:
18148 case WebAssembly::BI__builtin_wasm_abs_f64x2: {
18149 Value *Vec = EmitScalarExpr(E->getArg(0));
18150 Function *Callee = CGM.getIntrinsic(Intrinsic::fabs, Vec->getType());
18151 return Builder.CreateCall(Callee, {Vec});
18152 }
18153 case WebAssembly::BI__builtin_wasm_sqrt_f32x4:
18154 case WebAssembly::BI__builtin_wasm_sqrt_f64x2: {
18155 Value *Vec = EmitScalarExpr(E->getArg(0));
18156 Function *Callee = CGM.getIntrinsic(Intrinsic::sqrt, Vec->getType());
18157 return Builder.CreateCall(Callee, {Vec});
18158 }
18159 case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8:
18160 case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8:
18161 case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4:
18162 case WebAssembly::BI__builtin_wasm_narrow_u_i16x8_i32x4: {
18163 Value *Low = EmitScalarExpr(E->getArg(0));
18164 Value *High = EmitScalarExpr(E->getArg(1));
18165 unsigned IntNo;
18166 switch (BuiltinID) {
18167 case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8:
18168 case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4:
18169 IntNo = Intrinsic::wasm_narrow_signed;
18170 break;
18171 case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8:
18172 case WebAssembly::BI__builtin_wasm_narrow_u_i16x8_i32x4:
18173 IntNo = Intrinsic::wasm_narrow_unsigned;
18174 break;
18175 default:
18176 llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18176)
;
18177 }
18178 Function *Callee =
18179 CGM.getIntrinsic(IntNo, {ConvertType(E->getType()), Low->getType()});
18180 return Builder.CreateCall(Callee, {Low, High});
18181 }
18182 case WebAssembly::BI__builtin_wasm_trunc_sat_zero_s_f64x2_i32x4:
18183 case WebAssembly::BI__builtin_wasm_trunc_sat_zero_u_f64x2_i32x4: {
18184 Value *Vec = EmitScalarExpr(E->getArg(0));
18185 unsigned IntNo;
18186 switch (BuiltinID) {
18187 case WebAssembly::BI__builtin_wasm_trunc_sat_zero_s_f64x2_i32x4:
18188 IntNo = Intrinsic::fptosi_sat;
18189 break;
18190 case WebAssembly::BI__builtin_wasm_trunc_sat_zero_u_f64x2_i32x4:
18191 IntNo = Intrinsic::fptoui_sat;
18192 break;
18193 default:
18194 llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18194)
;
18195 }
18196 llvm::Type *SrcT = Vec->getType();
18197 llvm::Type *TruncT =
18198 SrcT->getWithNewType(llvm::IntegerType::get(getLLVMContext(), 32));
18199 Function *Callee = CGM.getIntrinsic(IntNo, {TruncT, SrcT});
18200 Value *Trunc = Builder.CreateCall(Callee, Vec);
18201 Value *Splat = Builder.CreateVectorSplat(2, Builder.getInt32(0));
18202 Value *ConcatMask =
18203 llvm::ConstantVector::get({Builder.getInt32(0), Builder.getInt32(1),
18204 Builder.getInt32(2), Builder.getInt32(3)});
18205 return Builder.CreateShuffleVector(Trunc, Splat, ConcatMask);
18206 }
18207 case WebAssembly::BI__builtin_wasm_shuffle_i8x16: {
18208 Value *Ops[18];
18209 size_t OpIdx = 0;
18210 Ops[OpIdx++] = EmitScalarExpr(E->getArg(0));
18211 Ops[OpIdx++] = EmitScalarExpr(E->getArg(1));
18212 while (OpIdx < 18) {
18213 Optional<llvm::APSInt> LaneConst =
18214 E->getArg(OpIdx)->getIntegerConstantExpr(getContext());
18215 assert(LaneConst && "Constant arg isn't actually constant?")(static_cast <bool> (LaneConst && "Constant arg isn't actually constant?"
) ? void (0) : __assert_fail ("LaneConst && \"Constant arg isn't actually constant?\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18215, __extension__ __PRETTY_FUNCTION__))
;
18216 Ops[OpIdx++] = llvm::ConstantInt::get(getLLVMContext(), *LaneConst);
18217 }
18218 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_shuffle);
18219 return Builder.CreateCall(Callee, Ops);
18220 }
18221 default:
18222 return nullptr;
18223 }
18224}
18225
18226static std::pair<Intrinsic::ID, unsigned>
18227getIntrinsicForHexagonNonGCCBuiltin(unsigned BuiltinID) {
18228 struct Info {
18229 unsigned BuiltinID;
18230 Intrinsic::ID IntrinsicID;
18231 unsigned VecLen;
18232 };
18233 Info Infos[] = {
18234#define CUSTOM_BUILTIN_MAPPING(x,s) \
18235 { Hexagon::BI__builtin_HEXAGON_##x, Intrinsic::hexagon_##x, s },
18236 CUSTOM_BUILTIN_MAPPING(L2_loadrub_pci, 0)
18237 CUSTOM_BUILTIN_MAPPING(L2_loadrb_pci, 0)
18238 CUSTOM_BUILTIN_MAPPING(L2_loadruh_pci, 0)
18239 CUSTOM_BUILTIN_MAPPING(L2_loadrh_pci, 0)
18240 CUSTOM_BUILTIN_MAPPING(L2_loadri_pci, 0)
18241 CUSTOM_BUILTIN_MAPPING(L2_loadrd_pci, 0)
18242 CUSTOM_BUILTIN_MAPPING(L2_loadrub_pcr, 0)
18243 CUSTOM_BUILTIN_MAPPING(L2_loadrb_pcr, 0)
18244 CUSTOM_BUILTIN_MAPPING(L2_loadruh_pcr, 0)
18245 CUSTOM_BUILTIN_MAPPING(L2_loadrh_pcr, 0)
18246 CUSTOM_BUILTIN_MAPPING(L2_loadri_pcr, 0)
18247 CUSTOM_BUILTIN_MAPPING(L2_loadrd_pcr, 0)
18248 CUSTOM_BUILTIN_MAPPING(S2_storerb_pci, 0)
18249 CUSTOM_BUILTIN_MAPPING(S2_storerh_pci, 0)
18250 CUSTOM_BUILTIN_MAPPING(S2_storerf_pci, 0)
18251 CUSTOM_BUILTIN_MAPPING(S2_storeri_pci, 0)
18252 CUSTOM_BUILTIN_MAPPING(S2_storerd_pci, 0)
18253 CUSTOM_BUILTIN_MAPPING(S2_storerb_pcr, 0)
18254 CUSTOM_BUILTIN_MAPPING(S2_storerh_pcr, 0)
18255 CUSTOM_BUILTIN_MAPPING(S2_storerf_pcr, 0)
18256 CUSTOM_BUILTIN_MAPPING(S2_storeri_pcr, 0)
18257 CUSTOM_BUILTIN_MAPPING(S2_storerd_pcr, 0)
18258 CUSTOM_BUILTIN_MAPPING(V6_vmaskedstoreq, 64)
18259 CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorenq, 64)
18260 CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentq, 64)
18261 CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentnq, 64)
18262 CUSTOM_BUILTIN_MAPPING(V6_vmaskedstoreq_128B, 128)
18263 CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorenq_128B, 128)
18264 CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentq_128B, 128)
18265 CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentnq_128B, 128)
18266#include "clang/Basic/BuiltinsHexagonMapCustomDep.def"
18267#undef CUSTOM_BUILTIN_MAPPING
18268 };
18269
18270 auto CmpInfo = [] (Info A, Info B) { return A.BuiltinID < B.BuiltinID; };
18271 static const bool SortOnce = (llvm::sort(Infos, CmpInfo), true);
18272 (void)SortOnce;
18273
18274 const Info *F = std::lower_bound(std::begin(Infos), std::end(Infos),
18275 Info{BuiltinID, 0, 0}, CmpInfo);
18276 if (F == std::end(Infos) || F->BuiltinID != BuiltinID)
18277 return {Intrinsic::not_intrinsic, 0};
18278
18279 return {F->IntrinsicID, F->VecLen};
18280}
18281
18282Value *CodeGenFunction::EmitHexagonBuiltinExpr(unsigned BuiltinID,
18283 const CallExpr *E) {
18284 Intrinsic::ID ID;
18285 unsigned VecLen;
18286 std::tie(ID, VecLen) = getIntrinsicForHexagonNonGCCBuiltin(BuiltinID);
18287
18288 auto MakeCircOp = [this, E](unsigned IntID, bool IsLoad) {
18289 // The base pointer is passed by address, so it needs to be loaded.
18290 Address A = EmitPointerWithAlignment(E->getArg(0));
18291 Address BP = Address(
18292 Builder.CreateBitCast(A.getPointer(), Int8PtrPtrTy), A.getAlignment());
18293 llvm::Value *Base = Builder.CreateLoad(BP);
18294 // The treatment of both loads and stores is the same: the arguments for
18295 // the builtin are the same as the arguments for the intrinsic.
18296 // Load:
18297 // builtin(Base, Inc, Mod, Start) -> intr(Base, Inc, Mod, Start)
18298 // builtin(Base, Mod, Start) -> intr(Base, Mod, Start)
18299 // Store:
18300 // builtin(Base, Inc, Mod, Val, Start) -> intr(Base, Inc, Mod, Val, Start)
18301 // builtin(Base, Mod, Val, Start) -> intr(Base, Mod, Val, Start)
18302 SmallVector<llvm::Value*,5> Ops = { Base };
18303 for (unsigned i = 1, e = E->getNumArgs(); i != e; ++i)
18304 Ops.push_back(EmitScalarExpr(E->getArg(i)));
18305
18306 llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops);
18307 // The load intrinsics generate two results (Value, NewBase), stores
18308 // generate one (NewBase). The new base address needs to be stored.
18309 llvm::Value *NewBase = IsLoad ? Builder.CreateExtractValue(Result, 1)
18310 : Result;
18311 llvm::Value *LV = Builder.CreateBitCast(
18312 EmitScalarExpr(E->getArg(0)), NewBase->getType()->getPointerTo());
18313 Address Dest = EmitPointerWithAlignment(E->getArg(0));
18314 llvm::Value *RetVal =
18315 Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment());
18316 if (IsLoad)
18317 RetVal = Builder.CreateExtractValue(Result, 0);
18318 return RetVal;
18319 };
18320
18321 // Handle the conversion of bit-reverse load intrinsics to bit code.
18322 // The intrinsic call after this function only reads from memory and the
18323 // write to memory is dealt by the store instruction.
18324 auto MakeBrevLd = [this, E](unsigned IntID, llvm::Type *DestTy) {
18325 // The intrinsic generates one result, which is the new value for the base
18326 // pointer. It needs to be returned. The result of the load instruction is
18327 // passed to intrinsic by address, so the value needs to be stored.
18328 llvm::Value *BaseAddress =
18329 Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int8PtrTy);
18330
18331 // Expressions like &(*pt++) will be incremented per evaluation.
18332 // EmitPointerWithAlignment and EmitScalarExpr evaluates the expression
18333 // per call.
18334 Address DestAddr = EmitPointerWithAlignment(E->getArg(1));
18335 DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), Int8PtrTy),
18336 DestAddr.getAlignment());
18337 llvm::Value *DestAddress = DestAddr.getPointer();
18338
18339 // Operands are Base, Dest, Modifier.
18340 // The intrinsic format in LLVM IR is defined as
18341 // { ValueType, i8* } (i8*, i32).
18342 llvm::Value *Result = Builder.CreateCall(
18343 CGM.getIntrinsic(IntID), {BaseAddress, EmitScalarExpr(E->getArg(2))});
18344
18345 // The value needs to be stored as the variable is passed by reference.
18346 llvm::Value *DestVal = Builder.CreateExtractValue(Result, 0);
18347
18348 // The store needs to be truncated to fit the destination type.
18349 // While i32 and i64 are natively supported on Hexagon, i8 and i16 needs
18350 // to be handled with stores of respective destination type.
18351 DestVal = Builder.CreateTrunc(DestVal, DestTy);
18352
18353 llvm::Value *DestForStore =
18354 Builder.CreateBitCast(DestAddress, DestVal->getType()->getPointerTo());
18355 Builder.CreateAlignedStore(DestVal, DestForStore, DestAddr.getAlignment());
18356 // The updated value of the base pointer is returned.
18357 return Builder.CreateExtractValue(Result, 1);
18358 };
18359
18360 auto V2Q = [this, VecLen] (llvm::Value *Vec) {
18361 Intrinsic::ID ID = VecLen == 128 ? Intrinsic::hexagon_V6_vandvrt_128B
18362 : Intrinsic::hexagon_V6_vandvrt;
18363 return Builder.CreateCall(CGM.getIntrinsic(ID),
18364 {Vec, Builder.getInt32(-1)});
18365 };
18366 auto Q2V = [this, VecLen] (llvm::Value *Pred) {
18367 Intrinsic::ID ID = VecLen == 128 ? Intrinsic::hexagon_V6_vandqrt_128B
18368 : Intrinsic::hexagon_V6_vandqrt;
18369 return Builder.CreateCall(CGM.getIntrinsic(ID),
18370 {Pred, Builder.getInt32(-1)});
18371 };
18372
18373 switch (BuiltinID) {
18374 // These intrinsics return a tuple {Vector, VectorPred} in LLVM IR,
18375 // and the corresponding C/C++ builtins use loads/stores to update
18376 // the predicate.
18377 case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry:
18378 case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry_128B:
18379 case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry:
18380 case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry_128B: {
18381 // Get the type from the 0-th argument.
18382 llvm::Type *VecType = ConvertType(E->getArg(0)->getType());
18383 Address PredAddr = Builder.CreateBitCast(
18384 EmitPointerWithAlignment(E->getArg(2)), VecType->getPointerTo(0));
18385 llvm::Value *PredIn = V2Q(Builder.CreateLoad(PredAddr));
18386 llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID),
18387 {EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), PredIn});
18388
18389 llvm::Value *PredOut = Builder.CreateExtractValue(Result, 1);
18390 Builder.CreateAlignedStore(Q2V(PredOut), PredAddr.getPointer(),
18391 PredAddr.getAlignment());
18392 return Builder.CreateExtractValue(Result, 0);
18393 }
18394
18395 case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pci:
18396 case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pci:
18397 case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pci:
18398 case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pci:
18399 case Hexagon::BI__builtin_HEXAGON_L2_loadri_pci:
18400 case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pci:
18401 case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pcr:
18402 case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pcr:
18403 case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pcr:
18404 case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pcr:
18405 case Hexagon::BI__builtin_HEXAGON_L2_loadri_pcr:
18406 case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pcr:
18407 return MakeCircOp(ID, /*IsLoad=*/true);
18408 case Hexagon::BI__builtin_HEXAGON_S2_storerb_pci:
18409 case Hexagon::BI__builtin_HEXAGON_S2_storerh_pci:
18410 case Hexagon::BI__builtin_HEXAGON_S2_storerf_pci:
18411 case Hexagon::BI__builtin_HEXAGON_S2_storeri_pci:
18412 case Hexagon::BI__builtin_HEXAGON_S2_storerd_pci:
18413 case Hexagon::BI__builtin_HEXAGON_S2_storerb_pcr:
18414 case Hexagon::BI__builtin_HEXAGON_S2_storerh_pcr:
18415 case Hexagon::BI__builtin_HEXAGON_S2_storerf_pcr:
18416 case Hexagon::BI__builtin_HEXAGON_S2_storeri_pcr:
18417 case Hexagon::BI__builtin_HEXAGON_S2_storerd_pcr:
18418 return MakeCircOp(ID, /*IsLoad=*/false);
18419 case Hexagon::BI__builtin_brev_ldub:
18420 return MakeBrevLd(Intrinsic::hexagon_L2_loadrub_pbr, Int8Ty);
18421 case Hexagon::BI__builtin_brev_ldb:
18422 return MakeBrevLd(Intrinsic::hexagon_L2_loadrb_pbr, Int8Ty);
18423 case Hexagon::BI__builtin_brev_lduh:
18424 return MakeBrevLd(Intrinsic::hexagon_L2_loadruh_pbr, Int16Ty);
18425 case Hexagon::BI__builtin_brev_ldh:
18426 return MakeBrevLd(Intrinsic::hexagon_L2_loadrh_pbr, Int16Ty);
18427 case Hexagon::BI__builtin_brev_ldw:
18428 return MakeBrevLd(Intrinsic::hexagon_L2_loadri_pbr, Int32Ty);
18429 case Hexagon::BI__builtin_brev_ldd:
18430 return MakeBrevLd(Intrinsic::hexagon_L2_loadrd_pbr, Int64Ty);
18431
18432 default: {
18433 if (ID == Intrinsic::not_intrinsic)
18434 return nullptr;
18435
18436 auto IsVectorPredTy = [](llvm::Type *T) {
18437 return T->isVectorTy() &&
18438 cast<llvm::VectorType>(T)->getElementType()->isIntegerTy(1);
18439 };
18440
18441 llvm::Function *IntrFn = CGM.getIntrinsic(ID);
18442 llvm::FunctionType *IntrTy = IntrFn->getFunctionType();
18443 SmallVector<llvm::Value*,4> Ops;
18444 for (unsigned i = 0, e = IntrTy->getNumParams(); i != e; ++i) {
18445 llvm::Type *T = IntrTy->getParamType(i);
18446 const Expr *A = E->getArg(i);
18447 if (IsVectorPredTy(T)) {
18448 // There will be an implicit cast to a boolean vector. Strip it.
18449 if (auto *Cast = dyn_cast<ImplicitCastExpr>(A)) {
18450 if (Cast->getCastKind() == CK_BitCast)
18451 A = Cast->getSubExpr();
18452 }
18453 Ops.push_back(V2Q(EmitScalarExpr(A)));
18454 } else {
18455 Ops.push_back(EmitScalarExpr(A));
18456 }
18457 }
18458
18459 llvm::Value *Call = Builder.CreateCall(IntrFn, Ops);
18460 if (IsVectorPredTy(IntrTy->getReturnType()))
18461 Call = Q2V(Call);
18462
18463 return Call;
18464 } // default
18465 } // switch
18466
18467 return nullptr;
18468}
18469
18470Value *CodeGenFunction::EmitRISCVBuiltinExpr(unsigned BuiltinID,
18471 const CallExpr *E,
18472 ReturnValueSlot ReturnValue) {
18473 SmallVector<Value *, 4> Ops;
18474 llvm::Type *ResultType = ConvertType(E->getType());
18475
18476 for (unsigned i = 0, e = E->getNumArgs(); i != e; i++)
18477 Ops.push_back(EmitScalarExpr(E->getArg(i)));
18478
18479 Intrinsic::ID ID = Intrinsic::not_intrinsic;
18480 unsigned NF = 1;
18481
18482 // Required for overloaded intrinsics.
18483 llvm::SmallVector<llvm::Type *, 2> IntrinsicTypes;
18484 switch (BuiltinID) {
18485 default: llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18485)
;
18486 case RISCV::BI__builtin_riscv_orc_b_32:
18487 case RISCV::BI__builtin_riscv_orc_b_64:
18488 case RISCV::BI__builtin_riscv_clmul:
18489 case RISCV::BI__builtin_riscv_clmulh:
18490 case RISCV::BI__builtin_riscv_clmulr:
18491 case RISCV::BI__builtin_riscv_bcompress_32:
18492 case RISCV::BI__builtin_riscv_bcompress_64:
18493 case RISCV::BI__builtin_riscv_bdecompress_32:
18494 case RISCV::BI__builtin_riscv_bdecompress_64:
18495 case RISCV::BI__builtin_riscv_grev_32:
18496 case RISCV::BI__builtin_riscv_grev_64:
18497 case RISCV::BI__builtin_riscv_gorc_32:
18498 case RISCV::BI__builtin_riscv_gorc_64:
18499 case RISCV::BI__builtin_riscv_shfl_32:
18500 case RISCV::BI__builtin_riscv_shfl_64:
18501 case RISCV::BI__builtin_riscv_unshfl_32:
18502 case RISCV::BI__builtin_riscv_unshfl_64:
18503 case RISCV::BI__builtin_riscv_xperm_n:
18504 case RISCV::BI__builtin_riscv_xperm_b:
18505 case RISCV::BI__builtin_riscv_xperm_h:
18506 case RISCV::BI__builtin_riscv_xperm_w:
18507 case RISCV::BI__builtin_riscv_crc32_b:
18508 case RISCV::BI__builtin_riscv_crc32_h:
18509 case RISCV::BI__builtin_riscv_crc32_w:
18510 case RISCV::BI__builtin_riscv_crc32_d:
18511 case RISCV::BI__builtin_riscv_crc32c_b:
18512 case RISCV::BI__builtin_riscv_crc32c_h:
18513 case RISCV::BI__builtin_riscv_crc32c_w:
18514 case RISCV::BI__builtin_riscv_crc32c_d: {
18515 switch (BuiltinID) {
18516 default: llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18516)
;
18517 // Zbb
18518 case RISCV::BI__builtin_riscv_orc_b_32:
18519 case RISCV::BI__builtin_riscv_orc_b_64:
18520 ID = Intrinsic::riscv_orc_b;
18521 break;
18522
18523 // Zbc
18524 case RISCV::BI__builtin_riscv_clmul:
18525 ID = Intrinsic::riscv_clmul;
18526 break;
18527 case RISCV::BI__builtin_riscv_clmulh:
18528 ID = Intrinsic::riscv_clmulh;
18529 break;
18530 case RISCV::BI__builtin_riscv_clmulr:
18531 ID = Intrinsic::riscv_clmulr;
18532 break;
18533
18534 // Zbe
18535 case RISCV::BI__builtin_riscv_bcompress_32:
18536 case RISCV::BI__builtin_riscv_bcompress_64:
18537 ID = Intrinsic::riscv_bcompress;
18538 break;
18539 case RISCV::BI__builtin_riscv_bdecompress_32:
18540 case RISCV::BI__builtin_riscv_bdecompress_64:
18541 ID = Intrinsic::riscv_bdecompress;
18542 break;
18543
18544 // Zbp
18545 case RISCV::BI__builtin_riscv_grev_32:
18546 case RISCV::BI__builtin_riscv_grev_64:
18547 ID = Intrinsic::riscv_grev;
18548 break;
18549 case RISCV::BI__builtin_riscv_gorc_32:
18550 case RISCV::BI__builtin_riscv_gorc_64:
18551 ID = Intrinsic::riscv_gorc;
18552 break;
18553 case RISCV::BI__builtin_riscv_shfl_32:
18554 case RISCV::BI__builtin_riscv_shfl_64:
18555 ID = Intrinsic::riscv_shfl;
18556 break;
18557 case RISCV::BI__builtin_riscv_unshfl_32:
18558 case RISCV::BI__builtin_riscv_unshfl_64:
18559 ID = Intrinsic::riscv_unshfl;
18560 break;
18561 case RISCV::BI__builtin_riscv_xperm_n:
18562 ID = Intrinsic::riscv_xperm_n;
18563 break;
18564 case RISCV::BI__builtin_riscv_xperm_b:
18565 ID = Intrinsic::riscv_xperm_b;
18566 break;
18567 case RISCV::BI__builtin_riscv_xperm_h:
18568 ID = Intrinsic::riscv_xperm_h;
18569 break;
18570 case RISCV::BI__builtin_riscv_xperm_w:
18571 ID = Intrinsic::riscv_xperm_w;
18572 break;
18573
18574 // Zbr
18575 case RISCV::BI__builtin_riscv_crc32_b:
18576 ID = Intrinsic::riscv_crc32_b;
18577 break;
18578 case RISCV::BI__builtin_riscv_crc32_h:
18579 ID = Intrinsic::riscv_crc32_h;
18580 break;
18581 case RISCV::BI__builtin_riscv_crc32_w:
18582 ID = Intrinsic::riscv_crc32_w;
18583 break;
18584 case RISCV::BI__builtin_riscv_crc32_d:
18585 ID = Intrinsic::riscv_crc32_d;
18586 break;
18587 case RISCV::BI__builtin_riscv_crc32c_b:
18588 ID = Intrinsic::riscv_crc32c_b;
18589 break;
18590 case RISCV::BI__builtin_riscv_crc32c_h:
18591 ID = Intrinsic::riscv_crc32c_h;
18592 break;
18593 case RISCV::BI__builtin_riscv_crc32c_w:
18594 ID = Intrinsic::riscv_crc32c_w;
18595 break;
18596 case RISCV::BI__builtin_riscv_crc32c_d:
18597 ID = Intrinsic::riscv_crc32c_d;
18598 break;
18599 }
18600
18601 IntrinsicTypes = {ResultType};
18602 break;
18603 }
18604 // Vector builtins are handled from here.
18605#include "clang/Basic/riscv_vector_builtin_cg.inc"
18606 }
18607
18608 assert(ID != Intrinsic::not_intrinsic)(static_cast <bool> (ID != Intrinsic::not_intrinsic) ? void
(0) : __assert_fail ("ID != Intrinsic::not_intrinsic", "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/clang/lib/CodeGen/CGBuiltin.cpp"
, 18608, __extension__ __PRETTY_FUNCTION__))
;
18609
18610 llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes);
18611 return Builder.CreateCall(F, Ops, "");
18612}