Bug Summary

File:tools/clang/lib/CodeGen/CGBuiltin.cpp
Warning:line 3357, column 7
Undefined or garbage value returned to caller

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -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-eagerly-assume -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 -mrelocation-model pic -pic-level 2 -mthread-model posix -relaxed-aliasing -fmath-errno -masm-verbose -mconstructor-aliases -munwind-tables -fuse-init-array -target-cpu x86-64 -dwarf-column-info -debugger-tuning=gdb -momit-leaf-frame-pointer -ffunction-sections -fdata-sections -resource-dir /usr/lib/llvm-7/lib/clang/7.0.0 -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-7~svn338205/build-llvm/tools/clang/lib/CodeGen -I /build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen -I /build/llvm-toolchain-snapshot-7~svn338205/tools/clang/include -I /build/llvm-toolchain-snapshot-7~svn338205/build-llvm/tools/clang/include -I /build/llvm-toolchain-snapshot-7~svn338205/build-llvm/include -I /build/llvm-toolchain-snapshot-7~svn338205/include -U NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/8/../../../../include/c++/8 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/8/../../../../include/x86_64-linux-gnu/c++/8 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/8/../../../../include/x86_64-linux-gnu/c++/8 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/8/../../../../include/c++/8/backward -internal-isystem /usr/include/clang/7.0.0/include/ -internal-isystem /usr/local/include -internal-isystem /usr/lib/llvm-7/lib/clang/7.0.0/include -internal-externc-isystem /usr/lib/gcc/x86_64-linux-gnu/8/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-comment -std=c++11 -fdeprecated-macro -fdebug-compilation-dir /build/llvm-toolchain-snapshot-7~svn338205/build-llvm/tools/clang/lib/CodeGen -ferror-limit 19 -fmessage-length 0 -fvisibility-inlines-hidden -fobjc-runtime=gcc -fno-common -fdiagnostics-show-option -vectorize-loops -vectorize-slp -analyzer-output=html -analyzer-config stable-report-filename=true -o /tmp/scan-build-2018-07-29-043837-17923-1 -x c++ /build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp -faddrsig
1//===---- CGBuiltin.cpp - Emit LLVM Code for builtins ---------------------===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This contains code to emit Builtin calls as LLVM code.
11//
12//===----------------------------------------------------------------------===//
13
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 "TargetInfo.h"
22#include "clang/AST/ASTContext.h"
23#include "clang/AST/Decl.h"
24#include "clang/Analysis/Analyses/OSLog.h"
25#include "clang/Basic/TargetBuiltins.h"
26#include "clang/Basic/TargetInfo.h"
27#include "clang/CodeGen/CGFunctionInfo.h"
28#include "llvm/ADT/StringExtras.h"
29#include "llvm/IR/CallSite.h"
30#include "llvm/IR/DataLayout.h"
31#include "llvm/IR/InlineAsm.h"
32#include "llvm/IR/Intrinsics.h"
33#include "llvm/IR/MDBuilder.h"
34#include "llvm/Support/ConvertUTF.h"
35#include "llvm/Support/ScopedPrinter.h"
36#include "llvm/Support/TargetParser.h"
37#include <sstream>
38
39using namespace clang;
40using namespace CodeGen;
41using namespace llvm;
42
43static
44int64_t clamp(int64_t Value, int64_t Low, int64_t High) {
45 return std::min(High, std::max(Low, Value));
46}
47
48/// getBuiltinLibFunction - Given a builtin id for a function like
49/// "__builtin_fabsf", return a Function* for "fabsf".
50llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD,
51 unsigned BuiltinID) {
52 assert(Context.BuiltinInfo.isLibFunction(BuiltinID))(static_cast <bool> (Context.BuiltinInfo.isLibFunction(
BuiltinID)) ? void (0) : __assert_fail ("Context.BuiltinInfo.isLibFunction(BuiltinID)"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 52, __extension__ __PRETTY_FUNCTION__))
;
53
54 // Get the name, skip over the __builtin_ prefix (if necessary).
55 StringRef Name;
56 GlobalDecl D(FD);
57
58 // If the builtin has been declared explicitly with an assembler label,
59 // use the mangled name. This differs from the plain label on platforms
60 // that prefix labels.
61 if (FD->hasAttr<AsmLabelAttr>())
62 Name = getMangledName(D);
63 else
64 Name = Context.BuiltinInfo.getName(BuiltinID) + 10;
65
66 llvm::FunctionType *Ty =
67 cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType()));
68
69 return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false);
70}
71
72/// Emit the conversions required to turn the given value into an
73/// integer of the given size.
74static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V,
75 QualType T, llvm::IntegerType *IntType) {
76 V = CGF.EmitToMemory(V, T);
77
78 if (V->getType()->isPointerTy())
79 return CGF.Builder.CreatePtrToInt(V, IntType);
80
81 assert(V->getType() == IntType)(static_cast <bool> (V->getType() == IntType) ? void
(0) : __assert_fail ("V->getType() == IntType", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 81, __extension__ __PRETTY_FUNCTION__))
;
82 return V;
83}
84
85static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V,
86 QualType T, llvm::Type *ResultType) {
87 V = CGF.EmitFromMemory(V, T);
88
89 if (ResultType->isPointerTy())
90 return CGF.Builder.CreateIntToPtr(V, ResultType);
91
92 assert(V->getType() == ResultType)(static_cast <bool> (V->getType() == ResultType) ? void
(0) : __assert_fail ("V->getType() == ResultType", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 92, __extension__ __PRETTY_FUNCTION__))
;
93 return V;
94}
95
96/// Utility to insert an atomic instruction based on Instrinsic::ID
97/// and the expression node.
98static Value *MakeBinaryAtomicValue(CodeGenFunction &CGF,
99 llvm::AtomicRMWInst::BinOp Kind,
100 const CallExpr *E) {
101 QualType T = E->getType();
102 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 102, __extension__ __PRETTY_FUNCTION__))
;
103 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 104, __extension__ __PRETTY_FUNCTION__))
104 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 104, __extension__ __PRETTY_FUNCTION__))
;
105 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 105, __extension__ __PRETTY_FUNCTION__))
;
106
107 llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
108 unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
109
110 llvm::IntegerType *IntType =
111 llvm::IntegerType::get(CGF.getLLVMContext(),
112 CGF.getContext().getTypeSize(T));
113 llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
114
115 llvm::Value *Args[2];
116 Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
117 Args[1] = CGF.EmitScalarExpr(E->getArg(1));
118 llvm::Type *ValueType = Args[1]->getType();
119 Args[1] = EmitToInt(CGF, Args[1], T, IntType);
120
121 llvm::Value *Result = CGF.Builder.CreateAtomicRMW(
122 Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent);
123 return EmitFromInt(CGF, Result, T, ValueType);
124}
125
126static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) {
127 Value *Val = CGF.EmitScalarExpr(E->getArg(0));
128 Value *Address = CGF.EmitScalarExpr(E->getArg(1));
129
130 // Convert the type of the pointer to a pointer to the stored type.
131 Val = CGF.EmitToMemory(Val, E->getArg(0)->getType());
132 Value *BC = CGF.Builder.CreateBitCast(
133 Address, llvm::PointerType::getUnqual(Val->getType()), "cast");
134 LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType());
135 LV.setNontemporal(true);
136 CGF.EmitStoreOfScalar(Val, LV, false);
137 return nullptr;
138}
139
140static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) {
141 Value *Address = CGF.EmitScalarExpr(E->getArg(0));
142
143 LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType());
144 LV.setNontemporal(true);
145 return CGF.EmitLoadOfScalar(LV, E->getExprLoc());
146}
147
148static RValue EmitBinaryAtomic(CodeGenFunction &CGF,
149 llvm::AtomicRMWInst::BinOp Kind,
150 const CallExpr *E) {
151 return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E));
152}
153
154/// Utility to insert an atomic instruction based Instrinsic::ID and
155/// the expression node, where the return value is the result of the
156/// operation.
157static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF,
158 llvm::AtomicRMWInst::BinOp Kind,
159 const CallExpr *E,
160 Instruction::BinaryOps Op,
161 bool Invert = false) {
162 QualType T = E->getType();
163 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 163, __extension__ __PRETTY_FUNCTION__))
;
164 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 165, __extension__ __PRETTY_FUNCTION__))
165 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 165, __extension__ __PRETTY_FUNCTION__))
;
166 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 166, __extension__ __PRETTY_FUNCTION__))
;
167
168 llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
169 unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
170
171 llvm::IntegerType *IntType =
172 llvm::IntegerType::get(CGF.getLLVMContext(),
173 CGF.getContext().getTypeSize(T));
174 llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
175
176 llvm::Value *Args[2];
177 Args[1] = CGF.EmitScalarExpr(E->getArg(1));
178 llvm::Type *ValueType = Args[1]->getType();
179 Args[1] = EmitToInt(CGF, Args[1], T, IntType);
180 Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
181
182 llvm::Value *Result = CGF.Builder.CreateAtomicRMW(
183 Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent);
184 Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]);
185 if (Invert)
186 Result = CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result,
187 llvm::ConstantInt::get(IntType, -1));
188 Result = EmitFromInt(CGF, Result, T, ValueType);
189 return RValue::get(Result);
190}
191
192/// Utility to insert an atomic cmpxchg instruction.
193///
194/// @param CGF The current codegen function.
195/// @param E Builtin call expression to convert to cmpxchg.
196/// arg0 - address to operate on
197/// arg1 - value to compare with
198/// arg2 - new value
199/// @param ReturnBool Specifies whether to return success flag of
200/// cmpxchg result or the old value.
201///
202/// @returns result of cmpxchg, according to ReturnBool
203static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E,
204 bool ReturnBool) {
205 QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType();
206 llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
207 unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
208
209 llvm::IntegerType *IntType = llvm::IntegerType::get(
210 CGF.getLLVMContext(), CGF.getContext().getTypeSize(T));
211 llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
212
213 Value *Args[3];
214 Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
215 Args[1] = CGF.EmitScalarExpr(E->getArg(1));
216 llvm::Type *ValueType = Args[1]->getType();
217 Args[1] = EmitToInt(CGF, Args[1], T, IntType);
218 Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType);
219
220 Value *Pair = CGF.Builder.CreateAtomicCmpXchg(
221 Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent,
222 llvm::AtomicOrdering::SequentiallyConsistent);
223 if (ReturnBool)
224 // Extract boolean success flag and zext it to int.
225 return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1),
226 CGF.ConvertType(E->getType()));
227 else
228 // Extract old value and emit it using the same type as compare value.
229 return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T,
230 ValueType);
231}
232
233// Emit a simple mangled intrinsic that has 1 argument and a return type
234// matching the argument type.
235static Value *emitUnaryBuiltin(CodeGenFunction &CGF,
236 const CallExpr *E,
237 unsigned IntrinsicID) {
238 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
239
240 Value *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
241 return CGF.Builder.CreateCall(F, Src0);
242}
243
244// Emit an intrinsic that has 2 operands of the same type as its result.
245static Value *emitBinaryBuiltin(CodeGenFunction &CGF,
246 const CallExpr *E,
247 unsigned IntrinsicID) {
248 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
249 llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
250
251 Value *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
252 return CGF.Builder.CreateCall(F, { Src0, Src1 });
253}
254
255// Emit an intrinsic that has 3 operands of the same type as its result.
256static Value *emitTernaryBuiltin(CodeGenFunction &CGF,
257 const CallExpr *E,
258 unsigned IntrinsicID) {
259 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
260 llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
261 llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2));
262
263 Value *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
264 return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 });
265}
266
267// Emit an intrinsic that has 1 float or double operand, and 1 integer.
268static Value *emitFPIntBuiltin(CodeGenFunction &CGF,
269 const CallExpr *E,
270 unsigned IntrinsicID) {
271 llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
272 llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
273
274 Value *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
275 return CGF.Builder.CreateCall(F, {Src0, Src1});
276}
277
278/// EmitFAbs - Emit a call to @llvm.fabs().
279static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) {
280 Value *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType());
281 llvm::CallInst *Call = CGF.Builder.CreateCall(F, V);
282 Call->setDoesNotAccessMemory();
283 return Call;
284}
285
286/// Emit the computation of the sign bit for a floating point value. Returns
287/// the i1 sign bit value.
288static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) {
289 LLVMContext &C = CGF.CGM.getLLVMContext();
290
291 llvm::Type *Ty = V->getType();
292 int Width = Ty->getPrimitiveSizeInBits();
293 llvm::Type *IntTy = llvm::IntegerType::get(C, Width);
294 V = CGF.Builder.CreateBitCast(V, IntTy);
295 if (Ty->isPPC_FP128Ty()) {
296 // We want the sign bit of the higher-order double. The bitcast we just
297 // did works as if the double-double was stored to memory and then
298 // read as an i128. The "store" will put the higher-order double in the
299 // lower address in both little- and big-Endian modes, but the "load"
300 // will treat those bits as a different part of the i128: the low bits in
301 // little-Endian, the high bits in big-Endian. Therefore, on big-Endian
302 // we need to shift the high bits down to the low before truncating.
303 Width >>= 1;
304 if (CGF.getTarget().isBigEndian()) {
305 Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width);
306 V = CGF.Builder.CreateLShr(V, ShiftCst);
307 }
308 // We are truncating value in order to extract the higher-order
309 // double, which we will be using to extract the sign from.
310 IntTy = llvm::IntegerType::get(C, Width);
311 V = CGF.Builder.CreateTrunc(V, IntTy);
312 }
313 Value *Zero = llvm::Constant::getNullValue(IntTy);
314 return CGF.Builder.CreateICmpSLT(V, Zero);
315}
316
317static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD,
318 const CallExpr *E, llvm::Constant *calleeValue) {
319 CGCallee callee = CGCallee::forDirect(calleeValue, FD);
320 return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot());
321}
322
323/// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.*
324/// depending on IntrinsicID.
325///
326/// \arg CGF The current codegen function.
327/// \arg IntrinsicID The ID for the Intrinsic we wish to generate.
328/// \arg X The first argument to the llvm.*.with.overflow.*.
329/// \arg Y The second argument to the llvm.*.with.overflow.*.
330/// \arg Carry The carry returned by the llvm.*.with.overflow.*.
331/// \returns The result (i.e. sum/product) returned by the intrinsic.
332static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF,
333 const llvm::Intrinsic::ID IntrinsicID,
334 llvm::Value *X, llvm::Value *Y,
335 llvm::Value *&Carry) {
336 // Make sure we have integers of the same width.
337 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 339, __extension__ __PRETTY_FUNCTION__))
338 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 339, __extension__ __PRETTY_FUNCTION__))
339 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 339, __extension__ __PRETTY_FUNCTION__))
;
340
341 llvm::Value *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType());
342 llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y});
343 Carry = CGF.Builder.CreateExtractValue(Tmp, 1);
344 return CGF.Builder.CreateExtractValue(Tmp, 0);
345}
346
347static Value *emitRangedBuiltin(CodeGenFunction &CGF,
348 unsigned IntrinsicID,
349 int low, int high) {
350 llvm::MDBuilder MDHelper(CGF.getLLVMContext());
351 llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high));
352 Value *F = CGF.CGM.getIntrinsic(IntrinsicID, {});
353 llvm::Instruction *Call = CGF.Builder.CreateCall(F);
354 Call->setMetadata(llvm::LLVMContext::MD_range, RNode);
355 return Call;
356}
357
358namespace {
359 struct WidthAndSignedness {
360 unsigned Width;
361 bool Signed;
362 };
363}
364
365static WidthAndSignedness
366getIntegerWidthAndSignedness(const clang::ASTContext &context,
367 const clang::QualType Type) {
368 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 368, __extension__ __PRETTY_FUNCTION__))
;
369 unsigned Width = Type->isBooleanType() ? 1 : context.getTypeInfo(Type).Width;
370 bool Signed = Type->isSignedIntegerType();
371 return {Width, Signed};
372}
373
374// Given one or more integer types, this function produces an integer type that
375// encompasses them: any value in one of the given types could be expressed in
376// the encompassing type.
377static struct WidthAndSignedness
378EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) {
379 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 379, __extension__ __PRETTY_FUNCTION__))
;
380
381 // If any of the given types is signed, we must return a signed type.
382 bool Signed = false;
383 for (const auto &Type : Types) {
384 Signed |= Type.Signed;
385 }
386
387 // The encompassing type must have a width greater than or equal to the width
388 // of the specified types. Additionally, if the encompassing type is signed,
389 // its width must be strictly greater than the width of any unsigned types
390 // given.
391 unsigned Width = 0;
392 for (const auto &Type : Types) {
393 unsigned MinWidth = Type.Width + (Signed && !Type.Signed);
394 if (Width < MinWidth) {
395 Width = MinWidth;
396 }
397 }
398
399 return {Width, Signed};
400}
401
402Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) {
403 llvm::Type *DestType = Int8PtrTy;
404 if (ArgValue->getType() != DestType)
405 ArgValue =
406 Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data());
407
408 Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend;
409 return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue);
410}
411
412/// Checks if using the result of __builtin_object_size(p, @p From) in place of
413/// __builtin_object_size(p, @p To) is correct
414static bool areBOSTypesCompatible(int From, int To) {
415 // Note: Our __builtin_object_size implementation currently treats Type=0 and
416 // Type=2 identically. Encoding this implementation detail here may make
417 // improving __builtin_object_size difficult in the future, so it's omitted.
418 return From == To || (From == 0 && To == 1) || (From == 3 && To == 2);
419}
420
421static llvm::Value *
422getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) {
423 return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true);
424}
425
426llvm::Value *
427CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type,
428 llvm::IntegerType *ResType,
429 llvm::Value *EmittedE) {
430 uint64_t ObjectSize;
431 if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type))
432 return emitBuiltinObjectSize(E, Type, ResType, EmittedE);
433 return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true);
434}
435
436/// Returns a Value corresponding to the size of the given expression.
437/// This Value may be either of the following:
438/// - A llvm::Argument (if E is a param with the pass_object_size attribute on
439/// it)
440/// - A call to the @llvm.objectsize intrinsic
441///
442/// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null
443/// and we wouldn't otherwise try to reference a pass_object_size parameter,
444/// we'll call @llvm.objectsize on EmittedE, rather than emitting E.
445llvm::Value *
446CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type,
447 llvm::IntegerType *ResType,
448 llvm::Value *EmittedE) {
449 // We need to reference an argument if the pointer is a parameter with the
450 // pass_object_size attribute.
451 if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) {
452 auto *Param = dyn_cast<ParmVarDecl>(D->getDecl());
453 auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>();
454 if (Param != nullptr && PS != nullptr &&
455 areBOSTypesCompatible(PS->getType(), Type)) {
456 auto Iter = SizeArguments.find(Param);
457 assert(Iter != SizeArguments.end())(static_cast <bool> (Iter != SizeArguments.end()) ? void
(0) : __assert_fail ("Iter != SizeArguments.end()", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 457, __extension__ __PRETTY_FUNCTION__))
;
458
459 const ImplicitParamDecl *D = Iter->second;
460 auto DIter = LocalDeclMap.find(D);
461 assert(DIter != LocalDeclMap.end())(static_cast <bool> (DIter != LocalDeclMap.end()) ? void
(0) : __assert_fail ("DIter != LocalDeclMap.end()", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 461, __extension__ __PRETTY_FUNCTION__))
;
462
463 return EmitLoadOfScalar(DIter->second, /*volatile=*/false,
464 getContext().getSizeType(), E->getLocStart());
465 }
466 }
467
468 // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't
469 // evaluate E for side-effects. In either case, we shouldn't lower to
470 // @llvm.objectsize.
471 if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext())))
472 return getDefaultBuiltinObjectSizeResult(Type, ResType);
473
474 Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E);
475 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 476, __extension__ __PRETTY_FUNCTION__))
476 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 476, __extension__ __PRETTY_FUNCTION__))
;
477
478 Value *F = CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()});
479
480 // LLVM only supports 0 and 2, make sure that we pass along that as a boolean.
481 Value *Min = Builder.getInt1((Type & 2) != 0);
482 // For GCC compatibility, __builtin_object_size treat NULL as unknown size.
483 Value *NullIsUnknown = Builder.getTrue();
484 return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown});
485}
486
487namespace {
488/// A struct to generically desribe a bit test intrinsic.
489struct BitTest {
490 enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set };
491 enum InterlockingKind : uint8_t {
492 Unlocked,
493 Sequential,
494 Acquire,
495 Release,
496 NoFence
497 };
498
499 ActionKind Action;
500 InterlockingKind Interlocking;
501 bool Is64Bit;
502
503 static BitTest decodeBitTestBuiltin(unsigned BuiltinID);
504};
505} // namespace
506
507BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) {
508 switch (BuiltinID) {
509 // Main portable variants.
510 case Builtin::BI_bittest:
511 return {TestOnly, Unlocked, false};
512 case Builtin::BI_bittestandcomplement:
513 return {Complement, Unlocked, false};
514 case Builtin::BI_bittestandreset:
515 return {Reset, Unlocked, false};
516 case Builtin::BI_bittestandset:
517 return {Set, Unlocked, false};
518 case Builtin::BI_interlockedbittestandreset:
519 return {Reset, Sequential, false};
520 case Builtin::BI_interlockedbittestandset:
521 return {Set, Sequential, false};
522
523 // X86-specific 64-bit variants.
524 case Builtin::BI_bittest64:
525 return {TestOnly, Unlocked, true};
526 case Builtin::BI_bittestandcomplement64:
527 return {Complement, Unlocked, true};
528 case Builtin::BI_bittestandreset64:
529 return {Reset, Unlocked, true};
530 case Builtin::BI_bittestandset64:
531 return {Set, Unlocked, true};
532 case Builtin::BI_interlockedbittestandreset64:
533 return {Reset, Sequential, true};
534 case Builtin::BI_interlockedbittestandset64:
535 return {Set, Sequential, true};
536
537 // ARM/AArch64-specific ordering variants.
538 case Builtin::BI_interlockedbittestandset_acq:
539 return {Set, Acquire, false};
540 case Builtin::BI_interlockedbittestandset_rel:
541 return {Set, Release, false};
542 case Builtin::BI_interlockedbittestandset_nf:
543 return {Set, NoFence, false};
544 case Builtin::BI_interlockedbittestandreset_acq:
545 return {Reset, Acquire, false};
546 case Builtin::BI_interlockedbittestandreset_rel:
547 return {Reset, Release, false};
548 case Builtin::BI_interlockedbittestandreset_nf:
549 return {Reset, NoFence, false};
550 }
551 llvm_unreachable("expected only bittest intrinsics")::llvm::llvm_unreachable_internal("expected only bittest intrinsics"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 551)
;
552}
553
554static char bitActionToX86BTCode(BitTest::ActionKind A) {
555 switch (A) {
556 case BitTest::TestOnly: return '\0';
557 case BitTest::Complement: return 'c';
558 case BitTest::Reset: return 'r';
559 case BitTest::Set: return 's';
560 }
561 llvm_unreachable("invalid action")::llvm::llvm_unreachable_internal("invalid action", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 561)
;
562}
563
564static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF,
565 BitTest BT,
566 const CallExpr *E, Value *BitBase,
567 Value *BitPos) {
568 char Action = bitActionToX86BTCode(BT.Action);
569 char SizeSuffix = BT.Is64Bit ? 'q' : 'l';
570
571 // Build the assembly.
572 SmallString<64> Asm;
573 raw_svector_ostream AsmOS(Asm);
574 if (BT.Interlocking != BitTest::Unlocked)
575 AsmOS << "lock ";
576 AsmOS << "bt";
577 if (Action)
578 AsmOS << Action;
579 AsmOS << SizeSuffix << " $2, ($1)\n\tsetc ${0:b}";
580
581 // Build the constraints. FIXME: We should support immediates when possible.
582 std::string Constraints = "=r,r,r,~{cc},~{flags},~{fpsr}";
583 llvm::IntegerType *IntType = llvm::IntegerType::get(
584 CGF.getLLVMContext(),
585 CGF.getContext().getTypeSize(E->getArg(1)->getType()));
586 llvm::Type *IntPtrType = IntType->getPointerTo();
587 llvm::FunctionType *FTy =
588 llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false);
589
590 llvm::InlineAsm *IA =
591 llvm::InlineAsm::get(FTy, Asm, Constraints, /*SideEffects=*/true);
592 return CGF.Builder.CreateCall(IA, {BitBase, BitPos});
593}
594
595static llvm::AtomicOrdering
596getBitTestAtomicOrdering(BitTest::InterlockingKind I) {
597 switch (I) {
598 case BitTest::Unlocked: return llvm::AtomicOrdering::NotAtomic;
599 case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent;
600 case BitTest::Acquire: return llvm::AtomicOrdering::Acquire;
601 case BitTest::Release: return llvm::AtomicOrdering::Release;
602 case BitTest::NoFence: return llvm::AtomicOrdering::Monotonic;
603 }
604 llvm_unreachable("invalid interlocking")::llvm::llvm_unreachable_internal("invalid interlocking", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 604)
;
605}
606
607/// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of
608/// bits and a bit position and read and optionally modify the bit at that
609/// position. The position index can be arbitrarily large, i.e. it can be larger
610/// than 31 or 63, so we need an indexed load in the general case.
611static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF,
612 unsigned BuiltinID,
613 const CallExpr *E) {
614 Value *BitBase = CGF.EmitScalarExpr(E->getArg(0));
615 Value *BitPos = CGF.EmitScalarExpr(E->getArg(1));
616
617 BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID);
618
619 // X86 has special BT, BTC, BTR, and BTS instructions that handle the array
620 // indexing operation internally. Use them if possible.
621 llvm::Triple::ArchType Arch = CGF.getTarget().getTriple().getArch();
622 if (Arch == llvm::Triple::x86 || Arch == llvm::Triple::x86_64)
623 return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos);
624
625 // Otherwise, use generic code to load one byte and test the bit. Use all but
626 // the bottom three bits as the array index, and the bottom three bits to form
627 // a mask.
628 // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0;
629 Value *ByteIndex = CGF.Builder.CreateAShr(
630 BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx");
631 Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy);
632 Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8,
633 ByteIndex, "bittest.byteaddr"),
634 CharUnits::One());
635 Value *PosLow =
636 CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty),
637 llvm::ConstantInt::get(CGF.Int8Ty, 0x7));
638
639 // The updating instructions will need a mask.
640 Value *Mask = nullptr;
641 if (BT.Action != BitTest::TestOnly) {
642 Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow,
643 "bittest.mask");
644 }
645
646 // Check the action and ordering of the interlocked intrinsics.
647 llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking);
648
649 Value *OldByte = nullptr;
650 if (Ordering != llvm::AtomicOrdering::NotAtomic) {
651 // Emit a combined atomicrmw load/store operation for the interlocked
652 // intrinsics.
653 llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or;
654 if (BT.Action == BitTest::Reset) {
655 Mask = CGF.Builder.CreateNot(Mask);
656 RMWOp = llvm::AtomicRMWInst::And;
657 }
658 OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask,
659 Ordering);
660 } else {
661 // Emit a plain load for the non-interlocked intrinsics.
662 OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte");
663 Value *NewByte = nullptr;
664 switch (BT.Action) {
665 case BitTest::TestOnly:
666 // Don't store anything.
667 break;
668 case BitTest::Complement:
669 NewByte = CGF.Builder.CreateXor(OldByte, Mask);
670 break;
671 case BitTest::Reset:
672 NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask));
673 break;
674 case BitTest::Set:
675 NewByte = CGF.Builder.CreateOr(OldByte, Mask);
676 break;
677 }
678 if (NewByte)
679 CGF.Builder.CreateStore(NewByte, ByteAddr);
680 }
681
682 // However we loaded the old byte, either by plain load or atomicrmw, shift
683 // the bit into the low position and mask it to 0 or 1.
684 Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr");
685 return CGF.Builder.CreateAnd(
686 ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res");
687}
688
689namespace {
690enum class MSVCSetJmpKind {
691 _setjmpex,
692 _setjmp3,
693 _setjmp
694};
695}
696
697/// MSVC handles setjmp a bit differently on different platforms. On every
698/// architecture except 32-bit x86, the frame address is passed. On x86, extra
699/// parameters can be passed as variadic arguments, but we always pass none.
700static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind,
701 const CallExpr *E) {
702 llvm::Value *Arg1 = nullptr;
703 llvm::Type *Arg1Ty = nullptr;
704 StringRef Name;
705 bool IsVarArg = false;
706 if (SJKind == MSVCSetJmpKind::_setjmp3) {
707 Name = "_setjmp3";
708 Arg1Ty = CGF.Int32Ty;
709 Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0);
710 IsVarArg = true;
711 } else {
712 Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex";
713 Arg1Ty = CGF.Int8PtrTy;
714 Arg1 = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(Intrinsic::frameaddress),
715 llvm::ConstantInt::get(CGF.Int32Ty, 0));
716 }
717
718 // Mark the call site and declaration with ReturnsTwice.
719 llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty};
720 llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get(
721 CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex,
722 llvm::Attribute::ReturnsTwice);
723 llvm::Constant *SetJmpFn = CGF.CGM.CreateRuntimeFunction(
724 llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name,
725 ReturnsTwiceAttr, /*Local=*/true);
726
727 llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast(
728 CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy);
729 llvm::Value *Args[] = {Buf, Arg1};
730 llvm::CallSite CS = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args);
731 CS.setAttributes(ReturnsTwiceAttr);
732 return RValue::get(CS.getInstruction());
733}
734
735// Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code,
736// we handle them here.
737enum class CodeGenFunction::MSVCIntrin {
738 _BitScanForward,
739 _BitScanReverse,
740 _InterlockedAnd,
741 _InterlockedDecrement,
742 _InterlockedExchange,
743 _InterlockedExchangeAdd,
744 _InterlockedExchangeSub,
745 _InterlockedIncrement,
746 _InterlockedOr,
747 _InterlockedXor,
748 __fastfail,
749};
750
751Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID,
752 const CallExpr *E) {
753 switch (BuiltinID) {
754 case MSVCIntrin::_BitScanForward:
755 case MSVCIntrin::_BitScanReverse: {
756 Value *ArgValue = EmitScalarExpr(E->getArg(1));
757
758 llvm::Type *ArgType = ArgValue->getType();
759 llvm::Type *IndexType =
760 EmitScalarExpr(E->getArg(0))->getType()->getPointerElementType();
761 llvm::Type *ResultType = ConvertType(E->getType());
762
763 Value *ArgZero = llvm::Constant::getNullValue(ArgType);
764 Value *ResZero = llvm::Constant::getNullValue(ResultType);
765 Value *ResOne = llvm::ConstantInt::get(ResultType, 1);
766
767 BasicBlock *Begin = Builder.GetInsertBlock();
768 BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn);
769 Builder.SetInsertPoint(End);
770 PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result");
771
772 Builder.SetInsertPoint(Begin);
773 Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero);
774 BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn);
775 Builder.CreateCondBr(IsZero, End, NotZero);
776 Result->addIncoming(ResZero, Begin);
777
778 Builder.SetInsertPoint(NotZero);
779 Address IndexAddress = EmitPointerWithAlignment(E->getArg(0));
780
781 if (BuiltinID == MSVCIntrin::_BitScanForward) {
782 Value *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
783 Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()});
784 ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false);
785 Builder.CreateStore(ZeroCount, IndexAddress, false);
786 } else {
787 unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth();
788 Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1);
789
790 Value *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
791 Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()});
792 ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false);
793 Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount);
794 Builder.CreateStore(Index, IndexAddress, false);
795 }
796 Builder.CreateBr(End);
797 Result->addIncoming(ResOne, NotZero);
798
799 Builder.SetInsertPoint(End);
800 return Result;
801 }
802 case MSVCIntrin::_InterlockedAnd:
803 return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E);
804 case MSVCIntrin::_InterlockedExchange:
805 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E);
806 case MSVCIntrin::_InterlockedExchangeAdd:
807 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E);
808 case MSVCIntrin::_InterlockedExchangeSub:
809 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E);
810 case MSVCIntrin::_InterlockedOr:
811 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E);
812 case MSVCIntrin::_InterlockedXor:
813 return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E);
814
815 case MSVCIntrin::_InterlockedDecrement: {
816 llvm::Type *IntTy = ConvertType(E->getType());
817 AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
818 AtomicRMWInst::Sub,
819 EmitScalarExpr(E->getArg(0)),
820 ConstantInt::get(IntTy, 1),
821 llvm::AtomicOrdering::SequentiallyConsistent);
822 return Builder.CreateSub(RMWI, ConstantInt::get(IntTy, 1));
823 }
824 case MSVCIntrin::_InterlockedIncrement: {
825 llvm::Type *IntTy = ConvertType(E->getType());
826 AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
827 AtomicRMWInst::Add,
828 EmitScalarExpr(E->getArg(0)),
829 ConstantInt::get(IntTy, 1),
830 llvm::AtomicOrdering::SequentiallyConsistent);
831 return Builder.CreateAdd(RMWI, ConstantInt::get(IntTy, 1));
832 }
833
834 case MSVCIntrin::__fastfail: {
835 // Request immediate process termination from the kernel. The instruction
836 // sequences to do this are documented on MSDN:
837 // https://msdn.microsoft.com/en-us/library/dn774154.aspx
838 llvm::Triple::ArchType ISA = getTarget().getTriple().getArch();
839 StringRef Asm, Constraints;
840 switch (ISA) {
841 default:
842 ErrorUnsupported(E, "__fastfail call for this architecture");
843 break;
844 case llvm::Triple::x86:
845 case llvm::Triple::x86_64:
846 Asm = "int $$0x29";
847 Constraints = "{cx}";
848 break;
849 case llvm::Triple::thumb:
850 Asm = "udf #251";
851 Constraints = "{r0}";
852 break;
853 }
854 llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false);
855 llvm::InlineAsm *IA =
856 llvm::InlineAsm::get(FTy, Asm, Constraints, /*SideEffects=*/true);
857 llvm::AttributeList NoReturnAttr = llvm::AttributeList::get(
858 getLLVMContext(), llvm::AttributeList::FunctionIndex,
859 llvm::Attribute::NoReturn);
860 CallSite CS = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0)));
861 CS.setAttributes(NoReturnAttr);
862 return CS.getInstruction();
863 }
864 }
865 llvm_unreachable("Incorrect MSVC intrinsic!")::llvm::llvm_unreachable_internal("Incorrect MSVC intrinsic!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 865)
;
866}
867
868namespace {
869// ARC cleanup for __builtin_os_log_format
870struct CallObjCArcUse final : EHScopeStack::Cleanup {
871 CallObjCArcUse(llvm::Value *object) : object(object) {}
872 llvm::Value *object;
873
874 void Emit(CodeGenFunction &CGF, Flags flags) override {
875 CGF.EmitARCIntrinsicUse(object);
876 }
877};
878}
879
880Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E,
881 BuiltinCheckKind Kind) {
882 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 883, __extension__ __PRETTY_FUNCTION__))
883 && "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 883, __extension__ __PRETTY_FUNCTION__))
;
884
885 Value *ArgValue = EmitScalarExpr(E);
886 if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef())
887 return ArgValue;
888
889 SanitizerScope SanScope(this);
890 Value *Cond = Builder.CreateICmpNE(
891 ArgValue, llvm::Constant::getNullValue(ArgValue->getType()));
892 EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin),
893 SanitizerHandler::InvalidBuiltin,
894 {EmitCheckSourceLocation(E->getExprLoc()),
895 llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)},
896 None);
897 return ArgValue;
898}
899
900/// Get the argument type for arguments to os_log_helper.
901static CanQualType getOSLogArgType(ASTContext &C, int Size) {
902 QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false);
903 return C.getCanonicalType(UnsignedTy);
904}
905
906llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction(
907 const analyze_os_log::OSLogBufferLayout &Layout,
908 CharUnits BufferAlignment) {
909 ASTContext &Ctx = getContext();
910
911 llvm::SmallString<64> Name;
912 {
913 raw_svector_ostream OS(Name);
914 OS << "__os_log_helper";
915 OS << "_" << BufferAlignment.getQuantity();
916 OS << "_" << int(Layout.getSummaryByte());
917 OS << "_" << int(Layout.getNumArgsByte());
918 for (const auto &Item : Layout.Items)
919 OS << "_" << int(Item.getSizeByte()) << "_"
920 << int(Item.getDescriptorByte());
921 }
922
923 if (llvm::Function *F = CGM.getModule().getFunction(Name))
924 return F;
925
926 llvm::SmallVector<ImplicitParamDecl, 4> Params;
927 Params.emplace_back(Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"),
928 Ctx.VoidPtrTy, ImplicitParamDecl::Other);
929
930 for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) {
931 char Size = Layout.Items[I].getSizeByte();
932 if (!Size)
933 continue;
934
935 Params.emplace_back(
936 Ctx, nullptr, SourceLocation(),
937 &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)),
938 getOSLogArgType(Ctx, Size), ImplicitParamDecl::Other);
939 }
940
941 FunctionArgList Args;
942 for (auto &P : Params)
943 Args.push_back(&P);
944
945 // The helper function has linkonce_odr linkage to enable the linker to merge
946 // identical functions. To ensure the merging always happens, 'noinline' is
947 // attached to the function when compiling with -Oz.
948 const CGFunctionInfo &FI =
949 CGM.getTypes().arrangeBuiltinFunctionDeclaration(Ctx.VoidTy, Args);
950 llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI);
951 llvm::Function *Fn = llvm::Function::Create(
952 FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule());
953 Fn->setVisibility(llvm::GlobalValue::HiddenVisibility);
954 CGM.SetLLVMFunctionAttributes(nullptr, FI, Fn);
955 CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn);
956
957 // Attach 'noinline' at -Oz.
958 if (CGM.getCodeGenOpts().OptimizeSize == 2)
959 Fn->addFnAttr(llvm::Attribute::NoInline);
960
961 auto NL = ApplyDebugLocation::CreateEmpty(*this);
962 IdentifierInfo *II = &Ctx.Idents.get(Name);
963 FunctionDecl *FD = FunctionDecl::Create(
964 Ctx, Ctx.getTranslationUnitDecl(), SourceLocation(), SourceLocation(), II,
965 Ctx.VoidTy, nullptr, SC_PrivateExtern, false, false);
966
967 StartFunction(FD, Ctx.VoidTy, Fn, FI, Args);
968
969 // Create a scope with an artificial location for the body of this function.
970 auto AL = ApplyDebugLocation::CreateArtificial(*this);
971
972 CharUnits Offset;
973 Address BufAddr(Builder.CreateLoad(GetAddrOfLocalVar(&Params[0]), "buf"),
974 BufferAlignment);
975 Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()),
976 Builder.CreateConstByteGEP(BufAddr, Offset++, "summary"));
977 Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()),
978 Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs"));
979
980 unsigned I = 1;
981 for (const auto &Item : Layout.Items) {
982 Builder.CreateStore(
983 Builder.getInt8(Item.getDescriptorByte()),
984 Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor"));
985 Builder.CreateStore(
986 Builder.getInt8(Item.getSizeByte()),
987 Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize"));
988
989 CharUnits Size = Item.size();
990 if (!Size.getQuantity())
991 continue;
992
993 Address Arg = GetAddrOfLocalVar(&Params[I]);
994 Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData");
995 Addr = Builder.CreateBitCast(Addr, Arg.getPointer()->getType(),
996 "argDataCast");
997 Builder.CreateStore(Builder.CreateLoad(Arg), Addr);
998 Offset += Size;
999 ++I;
1000 }
1001
1002 FinishFunction();
1003
1004 return Fn;
1005}
1006
1007RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) {
1008 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1009, __extension__ __PRETTY_FUNCTION__))
1009 "__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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1009, __extension__ __PRETTY_FUNCTION__))
;
1010 ASTContext &Ctx = getContext();
1011 analyze_os_log::OSLogBufferLayout Layout;
1012 analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout);
1013 Address BufAddr = EmitPointerWithAlignment(E.getArg(0));
1014 llvm::SmallVector<llvm::Value *, 4> RetainableOperands;
1015
1016 // Ignore argument 1, the format string. It is not currently used.
1017 CallArgList Args;
1018 Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy);
1019
1020 for (const auto &Item : Layout.Items) {
1021 int Size = Item.getSizeByte();
1022 if (!Size)
1023 continue;
1024
1025 llvm::Value *ArgVal;
1026
1027 if (const Expr *TheExpr = Item.getExpr()) {
1028 ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false);
1029
1030 // Check if this is a retainable type.
1031 if (TheExpr->getType()->isObjCRetainableType()) {
1032 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1033, __extension__ __PRETTY_FUNCTION__))
1033 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1033, __extension__ __PRETTY_FUNCTION__))
;
1034 // Check if the object is constant, if not, save it in
1035 // RetainableOperands.
1036 if (!isa<Constant>(ArgVal))
1037 RetainableOperands.push_back(ArgVal);
1038 }
1039 } else {
1040 ArgVal = Builder.getInt32(Item.getConstValue().getQuantity());
1041 }
1042
1043 unsigned ArgValSize =
1044 CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType());
1045 llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(),
1046 ArgValSize);
1047 ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy);
1048 CanQualType ArgTy = getOSLogArgType(Ctx, Size);
1049 // If ArgVal has type x86_fp80, zero-extend ArgVal.
1050 ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy));
1051 Args.add(RValue::get(ArgVal), ArgTy);
1052 }
1053
1054 const CGFunctionInfo &FI =
1055 CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args);
1056 llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction(
1057 Layout, BufAddr.getAlignment());
1058 EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args);
1059
1060 // Push a clang.arc.use cleanup for each object in RetainableOperands. The
1061 // cleanup will cause the use to appear after the final log call, keeping
1062 // the object valid while it’s held in the log buffer. Note that if there’s
1063 // a release cleanup on the object, it will already be active; since
1064 // cleanups are emitted in reverse order, the use will occur before the
1065 // object is released.
1066 if (!RetainableOperands.empty() && getLangOpts().ObjCAutoRefCount &&
1067 CGM.getCodeGenOpts().OptimizationLevel != 0)
1068 for (llvm::Value *Object : RetainableOperands)
1069 pushFullExprCleanup<CallObjCArcUse>(getARCCleanupKind(), Object);
1070
1071 return RValue::get(BufAddr.getPointer());
1072}
1073
1074/// Determine if a binop is a checked mixed-sign multiply we can specialize.
1075static bool isSpecialMixedSignMultiply(unsigned BuiltinID,
1076 WidthAndSignedness Op1Info,
1077 WidthAndSignedness Op2Info,
1078 WidthAndSignedness ResultInfo) {
1079 return BuiltinID == Builtin::BI__builtin_mul_overflow &&
1080 Op1Info.Width == Op2Info.Width && Op1Info.Width >= ResultInfo.Width &&
1081 Op1Info.Signed != Op2Info.Signed;
1082}
1083
1084/// Emit a checked mixed-sign multiply. This is a cheaper specialization of
1085/// the generic checked-binop irgen.
1086static RValue
1087EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1,
1088 WidthAndSignedness Op1Info, const clang::Expr *Op2,
1089 WidthAndSignedness Op2Info,
1090 const clang::Expr *ResultArg, QualType ResultQTy,
1091 WidthAndSignedness ResultInfo) {
1092 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1094, __extension__ __PRETTY_FUNCTION__))
1093 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1094, __extension__ __PRETTY_FUNCTION__))
1094 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1094, __extension__ __PRETTY_FUNCTION__))
;
1095
1096 // Emit the signed and unsigned operands.
1097 const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2;
1098 const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1;
1099 llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp);
1100 llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp);
1101
1102 llvm::Type *OpTy = Signed->getType();
1103 llvm::Value *Zero = llvm::Constant::getNullValue(OpTy);
1104 Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg);
1105 llvm::Type *ResTy = ResultPtr.getElementType();
1106
1107 // Take the absolute value of the signed operand.
1108 llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero);
1109 llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed);
1110 llvm::Value *AbsSigned =
1111 CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed);
1112
1113 // Perform a checked unsigned multiplication.
1114 llvm::Value *UnsignedOverflow;
1115 llvm::Value *UnsignedResult =
1116 EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned,
1117 Unsigned, UnsignedOverflow);
1118
1119 llvm::Value *Overflow, *Result;
1120 if (ResultInfo.Signed) {
1121 // Signed overflow occurs if the result is greater than INT_MAX or lesser
1122 // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative).
1123 auto IntMax = llvm::APInt::getSignedMaxValue(ResultInfo.Width)
1124 .zextOrSelf(Op1Info.Width);
1125 llvm::Value *MaxResult =
1126 CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax),
1127 CGF.Builder.CreateZExt(IsNegative, OpTy));
1128 llvm::Value *SignedOverflow =
1129 CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult);
1130 Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow);
1131
1132 // Prepare the signed result (possibly by negating it).
1133 llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult);
1134 llvm::Value *SignedResult =
1135 CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult);
1136 Result = CGF.Builder.CreateTrunc(SignedResult, ResTy);
1137 } else {
1138 // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX.
1139 llvm::Value *Underflow = CGF.Builder.CreateAnd(
1140 IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult));
1141 Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow);
1142 if (ResultInfo.Width < Op1Info.Width) {
1143 auto IntMax =
1144 llvm::APInt::getMaxValue(ResultInfo.Width).zext(Op1Info.Width);
1145 llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT(
1146 UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax));
1147 Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow);
1148 }
1149
1150 // Negate the product if it would be negative in infinite precision.
1151 Result = CGF.Builder.CreateSelect(
1152 IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult);
1153
1154 Result = CGF.Builder.CreateTrunc(Result, ResTy);
1155 }
1156 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1156, __extension__ __PRETTY_FUNCTION__))
;
1157
1158 bool isVolatile =
1159 ResultArg->getType()->getPointeeType().isVolatileQualified();
1160 CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr,
1161 isVolatile);
1162 return RValue::get(Overflow);
1163}
1164
1165static llvm::Value *dumpRecord(CodeGenFunction &CGF, QualType RType,
1166 Value *&RecordPtr, CharUnits Align, Value *Func,
1167 int Lvl) {
1168 const auto *RT = RType->getAs<RecordType>();
1169 ASTContext &Context = CGF.getContext();
1170 RecordDecl *RD = RT->getDecl()->getDefinition();
1171 ASTContext &Ctx = RD->getASTContext();
1172 const ASTRecordLayout &RL = Ctx.getASTRecordLayout(RD);
1173 std::string Pad = std::string(Lvl * 4, ' ');
1174
1175 Value *GString =
1176 CGF.Builder.CreateGlobalStringPtr(RType.getAsString() + " {\n");
1177 Value *Res = CGF.Builder.CreateCall(Func, {GString});
1178
1179 static llvm::DenseMap<QualType, const char *> Types;
1180 if (Types.empty()) {
1181 Types[Context.CharTy] = "%c";
1182 Types[Context.BoolTy] = "%d";
1183 Types[Context.SignedCharTy] = "%hhd";
1184 Types[Context.UnsignedCharTy] = "%hhu";
1185 Types[Context.IntTy] = "%d";
1186 Types[Context.UnsignedIntTy] = "%u";
1187 Types[Context.LongTy] = "%ld";
1188 Types[Context.UnsignedLongTy] = "%lu";
1189 Types[Context.LongLongTy] = "%lld";
1190 Types[Context.UnsignedLongLongTy] = "%llu";
1191 Types[Context.ShortTy] = "%hd";
1192 Types[Context.UnsignedShortTy] = "%hu";
1193 Types[Context.VoidPtrTy] = "%p";
1194 Types[Context.FloatTy] = "%f";
1195 Types[Context.DoubleTy] = "%f";
1196 Types[Context.LongDoubleTy] = "%Lf";
1197 Types[Context.getPointerType(Context.CharTy)] = "%s";
1198 Types[Context.getPointerType(Context.getConstType(Context.CharTy))] = "%s";
1199 }
1200
1201 for (const auto *FD : RD->fields()) {
1202 uint64_t Off = RL.getFieldOffset(FD->getFieldIndex());
1203 Off = Ctx.toCharUnitsFromBits(Off).getQuantity();
1204
1205 Value *FieldPtr = RecordPtr;
1206 if (RD->isUnion())
1207 FieldPtr = CGF.Builder.CreatePointerCast(
1208 FieldPtr, CGF.ConvertType(Context.getPointerType(FD->getType())));
1209 else
1210 FieldPtr = CGF.Builder.CreateStructGEP(CGF.ConvertType(RType), FieldPtr,
1211 FD->getFieldIndex());
1212
1213 GString = CGF.Builder.CreateGlobalStringPtr(
1214 llvm::Twine(Pad)
1215 .concat(FD->getType().getAsString())
1216 .concat(llvm::Twine(' '))
1217 .concat(FD->getNameAsString())
1218 .concat(" : ")
1219 .str());
1220 Value *TmpRes = CGF.Builder.CreateCall(Func, {GString});
1221 Res = CGF.Builder.CreateAdd(Res, TmpRes);
1222
1223 QualType CanonicalType =
1224 FD->getType().getUnqualifiedType().getCanonicalType();
1225
1226 // We check whether we are in a recursive type
1227 if (CanonicalType->isRecordType()) {
1228 Value *TmpRes =
1229 dumpRecord(CGF, CanonicalType, FieldPtr, Align, Func, Lvl + 1);
1230 Res = CGF.Builder.CreateAdd(TmpRes, Res);
1231 continue;
1232 }
1233
1234 // We try to determine the best format to print the current field
1235 llvm::Twine Format = Types.find(CanonicalType) == Types.end()
1236 ? Types[Context.VoidPtrTy]
1237 : Types[CanonicalType];
1238
1239 Address FieldAddress = Address(FieldPtr, Align);
1240 FieldPtr = CGF.Builder.CreateLoad(FieldAddress);
1241
1242 // FIXME Need to handle bitfield here
1243 GString = CGF.Builder.CreateGlobalStringPtr(
1244 Format.concat(llvm::Twine('\n')).str());
1245 TmpRes = CGF.Builder.CreateCall(Func, {GString, FieldPtr});
1246 Res = CGF.Builder.CreateAdd(Res, TmpRes);
1247 }
1248
1249 GString = CGF.Builder.CreateGlobalStringPtr(Pad + "}\n");
1250 Value *TmpRes = CGF.Builder.CreateCall(Func, {GString});
1251 Res = CGF.Builder.CreateAdd(Res, TmpRes);
1252 return Res;
1253}
1254
1255RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
1256 unsigned BuiltinID, const CallExpr *E,
1257 ReturnValueSlot ReturnValue) {
1258 // See if we can constant fold this builtin. If so, don't emit it at all.
1259 Expr::EvalResult Result;
1260 if (E->EvaluateAsRValue(Result, CGM.getContext()) &&
1261 !Result.hasSideEffects()) {
1262 if (Result.Val.isInt())
1263 return RValue::get(llvm::ConstantInt::get(getLLVMContext(),
1264 Result.Val.getInt()));
1265 if (Result.Val.isFloat())
1266 return RValue::get(llvm::ConstantFP::get(getLLVMContext(),
1267 Result.Val.getFloat()));
1268 }
1269
1270 // There are LLVM math intrinsics/instructions corresponding to math library
1271 // functions except the LLVM op will never set errno while the math library
1272 // might. Also, math builtins have the same semantics as their math library
1273 // twins. Thus, we can transform math library and builtin calls to their
1274 // LLVM counterparts if the call is marked 'const' (known to never set errno).
1275 if (FD->hasAttr<ConstAttr>()) {
1276 switch (BuiltinID) {
1277 case Builtin::BIceil:
1278 case Builtin::BIceilf:
1279 case Builtin::BIceill:
1280 case Builtin::BI__builtin_ceil:
1281 case Builtin::BI__builtin_ceilf:
1282 case Builtin::BI__builtin_ceill:
1283 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::ceil));
1284
1285 case Builtin::BIcopysign:
1286 case Builtin::BIcopysignf:
1287 case Builtin::BIcopysignl:
1288 case Builtin::BI__builtin_copysign:
1289 case Builtin::BI__builtin_copysignf:
1290 case Builtin::BI__builtin_copysignl:
1291 case Builtin::BI__builtin_copysignf128:
1292 return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign));
1293
1294 case Builtin::BIcos:
1295 case Builtin::BIcosf:
1296 case Builtin::BIcosl:
1297 case Builtin::BI__builtin_cos:
1298 case Builtin::BI__builtin_cosf:
1299 case Builtin::BI__builtin_cosl:
1300 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::cos));
1301
1302 case Builtin::BIexp:
1303 case Builtin::BIexpf:
1304 case Builtin::BIexpl:
1305 case Builtin::BI__builtin_exp:
1306 case Builtin::BI__builtin_expf:
1307 case Builtin::BI__builtin_expl:
1308 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::exp));
1309
1310 case Builtin::BIexp2:
1311 case Builtin::BIexp2f:
1312 case Builtin::BIexp2l:
1313 case Builtin::BI__builtin_exp2:
1314 case Builtin::BI__builtin_exp2f:
1315 case Builtin::BI__builtin_exp2l:
1316 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::exp2));
1317
1318 case Builtin::BIfabs:
1319 case Builtin::BIfabsf:
1320 case Builtin::BIfabsl:
1321 case Builtin::BI__builtin_fabs:
1322 case Builtin::BI__builtin_fabsf:
1323 case Builtin::BI__builtin_fabsl:
1324 case Builtin::BI__builtin_fabsf128:
1325 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs));
1326
1327 case Builtin::BIfloor:
1328 case Builtin::BIfloorf:
1329 case Builtin::BIfloorl:
1330 case Builtin::BI__builtin_floor:
1331 case Builtin::BI__builtin_floorf:
1332 case Builtin::BI__builtin_floorl:
1333 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::floor));
1334
1335 case Builtin::BIfma:
1336 case Builtin::BIfmaf:
1337 case Builtin::BIfmal:
1338 case Builtin::BI__builtin_fma:
1339 case Builtin::BI__builtin_fmaf:
1340 case Builtin::BI__builtin_fmal:
1341 return RValue::get(emitTernaryBuiltin(*this, E, Intrinsic::fma));
1342
1343 case Builtin::BIfmax:
1344 case Builtin::BIfmaxf:
1345 case Builtin::BIfmaxl:
1346 case Builtin::BI__builtin_fmax:
1347 case Builtin::BI__builtin_fmaxf:
1348 case Builtin::BI__builtin_fmaxl:
1349 return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::maxnum));
1350
1351 case Builtin::BIfmin:
1352 case Builtin::BIfminf:
1353 case Builtin::BIfminl:
1354 case Builtin::BI__builtin_fmin:
1355 case Builtin::BI__builtin_fminf:
1356 case Builtin::BI__builtin_fminl:
1357 return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::minnum));
1358
1359 // fmod() is a special-case. It maps to the frem instruction rather than an
1360 // LLVM intrinsic.
1361 case Builtin::BIfmod:
1362 case Builtin::BIfmodf:
1363 case Builtin::BIfmodl:
1364 case Builtin::BI__builtin_fmod:
1365 case Builtin::BI__builtin_fmodf:
1366 case Builtin::BI__builtin_fmodl: {
1367 Value *Arg1 = EmitScalarExpr(E->getArg(0));
1368 Value *Arg2 = EmitScalarExpr(E->getArg(1));
1369 return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod"));
1370 }
1371
1372 case Builtin::BIlog:
1373 case Builtin::BIlogf:
1374 case Builtin::BIlogl:
1375 case Builtin::BI__builtin_log:
1376 case Builtin::BI__builtin_logf:
1377 case Builtin::BI__builtin_logl:
1378 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log));
1379
1380 case Builtin::BIlog10:
1381 case Builtin::BIlog10f:
1382 case Builtin::BIlog10l:
1383 case Builtin::BI__builtin_log10:
1384 case Builtin::BI__builtin_log10f:
1385 case Builtin::BI__builtin_log10l:
1386 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log10));
1387
1388 case Builtin::BIlog2:
1389 case Builtin::BIlog2f:
1390 case Builtin::BIlog2l:
1391 case Builtin::BI__builtin_log2:
1392 case Builtin::BI__builtin_log2f:
1393 case Builtin::BI__builtin_log2l:
1394 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log2));
1395
1396 case Builtin::BInearbyint:
1397 case Builtin::BInearbyintf:
1398 case Builtin::BInearbyintl:
1399 case Builtin::BI__builtin_nearbyint:
1400 case Builtin::BI__builtin_nearbyintf:
1401 case Builtin::BI__builtin_nearbyintl:
1402 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::nearbyint));
1403
1404 case Builtin::BIpow:
1405 case Builtin::BIpowf:
1406 case Builtin::BIpowl:
1407 case Builtin::BI__builtin_pow:
1408 case Builtin::BI__builtin_powf:
1409 case Builtin::BI__builtin_powl:
1410 return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::pow));
1411
1412 case Builtin::BIrint:
1413 case Builtin::BIrintf:
1414 case Builtin::BIrintl:
1415 case Builtin::BI__builtin_rint:
1416 case Builtin::BI__builtin_rintf:
1417 case Builtin::BI__builtin_rintl:
1418 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::rint));
1419
1420 case Builtin::BIround:
1421 case Builtin::BIroundf:
1422 case Builtin::BIroundl:
1423 case Builtin::BI__builtin_round:
1424 case Builtin::BI__builtin_roundf:
1425 case Builtin::BI__builtin_roundl:
1426 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::round));
1427
1428 case Builtin::BIsin:
1429 case Builtin::BIsinf:
1430 case Builtin::BIsinl:
1431 case Builtin::BI__builtin_sin:
1432 case Builtin::BI__builtin_sinf:
1433 case Builtin::BI__builtin_sinl:
1434 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::sin));
1435
1436 case Builtin::BIsqrt:
1437 case Builtin::BIsqrtf:
1438 case Builtin::BIsqrtl:
1439 case Builtin::BI__builtin_sqrt:
1440 case Builtin::BI__builtin_sqrtf:
1441 case Builtin::BI__builtin_sqrtl:
1442 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::sqrt));
1443
1444 case Builtin::BItrunc:
1445 case Builtin::BItruncf:
1446 case Builtin::BItruncl:
1447 case Builtin::BI__builtin_trunc:
1448 case Builtin::BI__builtin_truncf:
1449 case Builtin::BI__builtin_truncl:
1450 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::trunc));
1451
1452 default:
1453 break;
1454 }
1455 }
1456
1457 switch (BuiltinID) {
1458 default: break;
1459 case Builtin::BI__builtin___CFStringMakeConstantString:
1460 case Builtin::BI__builtin___NSStringMakeConstantString:
1461 return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType()));
1462 case Builtin::BI__builtin_stdarg_start:
1463 case Builtin::BI__builtin_va_start:
1464 case Builtin::BI__va_start:
1465 case Builtin::BI__builtin_va_end:
1466 return RValue::get(
1467 EmitVAStartEnd(BuiltinID == Builtin::BI__va_start
1468 ? EmitScalarExpr(E->getArg(0))
1469 : EmitVAListRef(E->getArg(0)).getPointer(),
1470 BuiltinID != Builtin::BI__builtin_va_end));
1471 case Builtin::BI__builtin_va_copy: {
1472 Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer();
1473 Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer();
1474
1475 llvm::Type *Type = Int8PtrTy;
1476
1477 DstPtr = Builder.CreateBitCast(DstPtr, Type);
1478 SrcPtr = Builder.CreateBitCast(SrcPtr, Type);
1479 return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy),
1480 {DstPtr, SrcPtr}));
1481 }
1482 case Builtin::BI__builtin_abs:
1483 case Builtin::BI__builtin_labs:
1484 case Builtin::BI__builtin_llabs: {
1485 // X < 0 ? -X : X
1486 // The negation has 'nsw' because abs of INT_MIN is undefined.
1487 Value *ArgValue = EmitScalarExpr(E->getArg(0));
1488 Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg");
1489 Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType());
1490 Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond");
1491 Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs");
1492 return RValue::get(Result);
1493 }
1494 case Builtin::BI__builtin_conj:
1495 case Builtin::BI__builtin_conjf:
1496 case Builtin::BI__builtin_conjl: {
1497 ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
1498 Value *Real = ComplexVal.first;
1499 Value *Imag = ComplexVal.second;
1500 Value *Zero =
1501 Imag->getType()->isFPOrFPVectorTy()
1502 ? llvm::ConstantFP::getZeroValueForNegation(Imag->getType())
1503 : llvm::Constant::getNullValue(Imag->getType());
1504
1505 Imag = Builder.CreateFSub(Zero, Imag, "sub");
1506 return RValue::getComplex(std::make_pair(Real, Imag));
1507 }
1508 case Builtin::BI__builtin_creal:
1509 case Builtin::BI__builtin_crealf:
1510 case Builtin::BI__builtin_creall:
1511 case Builtin::BIcreal:
1512 case Builtin::BIcrealf:
1513 case Builtin::BIcreall: {
1514 ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
1515 return RValue::get(ComplexVal.first);
1516 }
1517
1518 case Builtin::BI__builtin_dump_struct: {
1519 Value *Func = EmitScalarExpr(E->getArg(1)->IgnoreImpCasts());
1520 CharUnits Arg0Align = EmitPointerWithAlignment(E->getArg(0)).getAlignment();
1521
1522 const Expr *Arg0 = E->getArg(0)->IgnoreImpCasts();
1523 QualType Arg0Type = Arg0->getType()->getPointeeType();
1524
1525 Value *RecordPtr = EmitScalarExpr(Arg0);
1526 Value *Res = dumpRecord(*this, Arg0Type, RecordPtr, Arg0Align, Func, 0);
1527 return RValue::get(Res);
1528 }
1529
1530 case Builtin::BI__builtin_cimag:
1531 case Builtin::BI__builtin_cimagf:
1532 case Builtin::BI__builtin_cimagl:
1533 case Builtin::BIcimag:
1534 case Builtin::BIcimagf:
1535 case Builtin::BIcimagl: {
1536 ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
1537 return RValue::get(ComplexVal.second);
1538 }
1539
1540 case Builtin::BI__builtin_ctzs:
1541 case Builtin::BI__builtin_ctz:
1542 case Builtin::BI__builtin_ctzl:
1543 case Builtin::BI__builtin_ctzll: {
1544 Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero);
1545
1546 llvm::Type *ArgType = ArgValue->getType();
1547 Value *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
1548
1549 llvm::Type *ResultType = ConvertType(E->getType());
1550 Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef());
1551 Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef});
1552 if (Result->getType() != ResultType)
1553 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1554 "cast");
1555 return RValue::get(Result);
1556 }
1557 case Builtin::BI__builtin_clzs:
1558 case Builtin::BI__builtin_clz:
1559 case Builtin::BI__builtin_clzl:
1560 case Builtin::BI__builtin_clzll: {
1561 Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero);
1562
1563 llvm::Type *ArgType = ArgValue->getType();
1564 Value *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
1565
1566 llvm::Type *ResultType = ConvertType(E->getType());
1567 Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef());
1568 Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef});
1569 if (Result->getType() != ResultType)
1570 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1571 "cast");
1572 return RValue::get(Result);
1573 }
1574 case Builtin::BI__builtin_ffs:
1575 case Builtin::BI__builtin_ffsl:
1576 case Builtin::BI__builtin_ffsll: {
1577 // ffs(x) -> x ? cttz(x) + 1 : 0
1578 Value *ArgValue = EmitScalarExpr(E->getArg(0));
1579
1580 llvm::Type *ArgType = ArgValue->getType();
1581 Value *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
1582
1583 llvm::Type *ResultType = ConvertType(E->getType());
1584 Value *Tmp =
1585 Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}),
1586 llvm::ConstantInt::get(ArgType, 1));
1587 Value *Zero = llvm::Constant::getNullValue(ArgType);
1588 Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero");
1589 Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs");
1590 if (Result->getType() != ResultType)
1591 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1592 "cast");
1593 return RValue::get(Result);
1594 }
1595 case Builtin::BI__builtin_parity:
1596 case Builtin::BI__builtin_parityl:
1597 case Builtin::BI__builtin_parityll: {
1598 // parity(x) -> ctpop(x) & 1
1599 Value *ArgValue = EmitScalarExpr(E->getArg(0));
1600
1601 llvm::Type *ArgType = ArgValue->getType();
1602 Value *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType);
1603
1604 llvm::Type *ResultType = ConvertType(E->getType());
1605 Value *Tmp = Builder.CreateCall(F, ArgValue);
1606 Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1));
1607 if (Result->getType() != ResultType)
1608 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1609 "cast");
1610 return RValue::get(Result);
1611 }
1612 case Builtin::BI__popcnt16:
1613 case Builtin::BI__popcnt:
1614 case Builtin::BI__popcnt64:
1615 case Builtin::BI__builtin_popcount:
1616 case Builtin::BI__builtin_popcountl:
1617 case Builtin::BI__builtin_popcountll: {
1618 Value *ArgValue = EmitScalarExpr(E->getArg(0));
1619
1620 llvm::Type *ArgType = ArgValue->getType();
1621 Value *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType);
1622
1623 llvm::Type *ResultType = ConvertType(E->getType());
1624 Value *Result = Builder.CreateCall(F, ArgValue);
1625 if (Result->getType() != ResultType)
1626 Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1627 "cast");
1628 return RValue::get(Result);
1629 }
1630 case Builtin::BI_rotr8:
1631 case Builtin::BI_rotr16:
1632 case Builtin::BI_rotr:
1633 case Builtin::BI_lrotr:
1634 case Builtin::BI_rotr64: {
1635 Value *Val = EmitScalarExpr(E->getArg(0));
1636 Value *Shift = EmitScalarExpr(E->getArg(1));
1637
1638 llvm::Type *ArgType = Val->getType();
1639 Shift = Builder.CreateIntCast(Shift, ArgType, false);
1640 unsigned ArgWidth = ArgType->getIntegerBitWidth();
1641 Value *Mask = llvm::ConstantInt::get(ArgType, ArgWidth - 1);
1642
1643 Value *RightShiftAmt = Builder.CreateAnd(Shift, Mask);
1644 Value *RightShifted = Builder.CreateLShr(Val, RightShiftAmt);
1645 Value *LeftShiftAmt = Builder.CreateAnd(Builder.CreateNeg(Shift), Mask);
1646 Value *LeftShifted = Builder.CreateShl(Val, LeftShiftAmt);
1647 Value *Result = Builder.CreateOr(LeftShifted, RightShifted);
1648 return RValue::get(Result);
1649 }
1650 case Builtin::BI_rotl8:
1651 case Builtin::BI_rotl16:
1652 case Builtin::BI_rotl:
1653 case Builtin::BI_lrotl:
1654 case Builtin::BI_rotl64: {
1655 Value *Val = EmitScalarExpr(E->getArg(0));
1656 Value *Shift = EmitScalarExpr(E->getArg(1));
1657
1658 llvm::Type *ArgType = Val->getType();
1659 Shift = Builder.CreateIntCast(Shift, ArgType, false);
1660 unsigned ArgWidth = ArgType->getIntegerBitWidth();
1661 Value *Mask = llvm::ConstantInt::get(ArgType, ArgWidth - 1);
1662
1663 Value *LeftShiftAmt = Builder.CreateAnd(Shift, Mask);
1664 Value *LeftShifted = Builder.CreateShl(Val, LeftShiftAmt);
1665 Value *RightShiftAmt = Builder.CreateAnd(Builder.CreateNeg(Shift), Mask);
1666 Value *RightShifted = Builder.CreateLShr(Val, RightShiftAmt);
1667 Value *Result = Builder.CreateOr(LeftShifted, RightShifted);
1668 return RValue::get(Result);
1669 }
1670 case Builtin::BI__builtin_unpredictable: {
1671 // Always return the argument of __builtin_unpredictable. LLVM does not
1672 // handle this builtin. Metadata for this builtin should be added directly
1673 // to instructions such as branches or switches that use it.
1674 return RValue::get(EmitScalarExpr(E->getArg(0)));
1675 }
1676 case Builtin::BI__builtin_expect: {
1677 Value *ArgValue = EmitScalarExpr(E->getArg(0));
1678 llvm::Type *ArgType = ArgValue->getType();
1679
1680 Value *ExpectedValue = EmitScalarExpr(E->getArg(1));
1681 // Don't generate llvm.expect on -O0 as the backend won't use it for
1682 // anything.
1683 // Note, we still IRGen ExpectedValue because it could have side-effects.
1684 if (CGM.getCodeGenOpts().OptimizationLevel == 0)
1685 return RValue::get(ArgValue);
1686
1687 Value *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType);
1688 Value *Result =
1689 Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval");
1690 return RValue::get(Result);
1691 }
1692 case Builtin::BI__builtin_assume_aligned: {
1693 Value *PtrValue = EmitScalarExpr(E->getArg(0));
1694 Value *OffsetValue =
1695 (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr;
1696
1697 Value *AlignmentValue = EmitScalarExpr(E->getArg(1));
1698 ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue);
1699 unsigned Alignment = (unsigned) AlignmentCI->getZExtValue();
1700
1701 EmitAlignmentAssumption(PtrValue, Alignment, OffsetValue);
1702 return RValue::get(PtrValue);
1703 }
1704 case Builtin::BI__assume:
1705 case Builtin::BI__builtin_assume: {
1706 if (E->getArg(0)->HasSideEffects(getContext()))
1707 return RValue::get(nullptr);
1708
1709 Value *ArgValue = EmitScalarExpr(E->getArg(0));
1710 Value *FnAssume = CGM.getIntrinsic(Intrinsic::assume);
1711 return RValue::get(Builder.CreateCall(FnAssume, ArgValue));
1712 }
1713 case Builtin::BI__builtin_bswap16:
1714 case Builtin::BI__builtin_bswap32:
1715 case Builtin::BI__builtin_bswap64: {
1716 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap));
1717 }
1718 case Builtin::BI__builtin_bitreverse8:
1719 case Builtin::BI__builtin_bitreverse16:
1720 case Builtin::BI__builtin_bitreverse32:
1721 case Builtin::BI__builtin_bitreverse64: {
1722 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse));
1723 }
1724 case Builtin::BI__builtin_object_size: {
1725 unsigned Type =
1726 E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue();
1727 auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType()));
1728
1729 // We pass this builtin onto the optimizer so that it can figure out the
1730 // object size in more complex cases.
1731 return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType,
1732 /*EmittedE=*/nullptr));
1733 }
1734 case Builtin::BI__builtin_prefetch: {
1735 Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0));
1736 // FIXME: Technically these constants should of type 'int', yes?
1737 RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) :
1738 llvm::ConstantInt::get(Int32Ty, 0);
1739 Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) :
1740 llvm::ConstantInt::get(Int32Ty, 3);
1741 Value *Data = llvm::ConstantInt::get(Int32Ty, 1);
1742 Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
1743 return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data}));
1744 }
1745 case Builtin::BI__builtin_readcyclecounter: {
1746 Value *F = CGM.getIntrinsic(Intrinsic::readcyclecounter);
1747 return RValue::get(Builder.CreateCall(F));
1748 }
1749 case Builtin::BI__builtin___clear_cache: {
1750 Value *Begin = EmitScalarExpr(E->getArg(0));
1751 Value *End = EmitScalarExpr(E->getArg(1));
1752 Value *F = CGM.getIntrinsic(Intrinsic::clear_cache);
1753 return RValue::get(Builder.CreateCall(F, {Begin, End}));
1754 }
1755 case Builtin::BI__builtin_trap:
1756 return RValue::get(EmitTrapCall(Intrinsic::trap));
1757 case Builtin::BI__debugbreak:
1758 return RValue::get(EmitTrapCall(Intrinsic::debugtrap));
1759 case Builtin::BI__builtin_unreachable: {
1760 EmitUnreachable(E->getExprLoc());
1761
1762 // We do need to preserve an insertion point.
1763 EmitBlock(createBasicBlock("unreachable.cont"));
1764
1765 return RValue::get(nullptr);
1766 }
1767
1768 case Builtin::BI__builtin_powi:
1769 case Builtin::BI__builtin_powif:
1770 case Builtin::BI__builtin_powil: {
1771 Value *Base = EmitScalarExpr(E->getArg(0));
1772 Value *Exponent = EmitScalarExpr(E->getArg(1));
1773 llvm::Type *ArgType = Base->getType();
1774 Value *F = CGM.getIntrinsic(Intrinsic::powi, ArgType);
1775 return RValue::get(Builder.CreateCall(F, {Base, Exponent}));
1776 }
1777
1778 case Builtin::BI__builtin_isgreater:
1779 case Builtin::BI__builtin_isgreaterequal:
1780 case Builtin::BI__builtin_isless:
1781 case Builtin::BI__builtin_islessequal:
1782 case Builtin::BI__builtin_islessgreater:
1783 case Builtin::BI__builtin_isunordered: {
1784 // Ordered comparisons: we know the arguments to these are matching scalar
1785 // floating point values.
1786 Value *LHS = EmitScalarExpr(E->getArg(0));
1787 Value *RHS = EmitScalarExpr(E->getArg(1));
1788
1789 switch (BuiltinID) {
1790 default: llvm_unreachable("Unknown ordered comparison")::llvm::llvm_unreachable_internal("Unknown ordered comparison"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 1790)
;
1791 case Builtin::BI__builtin_isgreater:
1792 LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp");
1793 break;
1794 case Builtin::BI__builtin_isgreaterequal:
1795 LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp");
1796 break;
1797 case Builtin::BI__builtin_isless:
1798 LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp");
1799 break;
1800 case Builtin::BI__builtin_islessequal:
1801 LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp");
1802 break;
1803 case Builtin::BI__builtin_islessgreater:
1804 LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp");
1805 break;
1806 case Builtin::BI__builtin_isunordered:
1807 LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp");
1808 break;
1809 }
1810 // ZExt bool to int type.
1811 return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType())));
1812 }
1813 case Builtin::BI__builtin_isnan: {
1814 Value *V = EmitScalarExpr(E->getArg(0));
1815 V = Builder.CreateFCmpUNO(V, V, "cmp");
1816 return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType())));
1817 }
1818
1819 case Builtin::BIfinite:
1820 case Builtin::BI__finite:
1821 case Builtin::BIfinitef:
1822 case Builtin::BI__finitef:
1823 case Builtin::BIfinitel:
1824 case Builtin::BI__finitel:
1825 case Builtin::BI__builtin_isinf:
1826 case Builtin::BI__builtin_isfinite: {
1827 // isinf(x) --> fabs(x) == infinity
1828 // isfinite(x) --> fabs(x) != infinity
1829 // x != NaN via the ordered compare in either case.
1830 Value *V = EmitScalarExpr(E->getArg(0));
1831 Value *Fabs = EmitFAbs(*this, V);
1832 Constant *Infinity = ConstantFP::getInfinity(V->getType());
1833 CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf)
1834 ? CmpInst::FCMP_OEQ
1835 : CmpInst::FCMP_ONE;
1836 Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf");
1837 return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType())));
1838 }
1839
1840 case Builtin::BI__builtin_isinf_sign: {
1841 // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0
1842 Value *Arg = EmitScalarExpr(E->getArg(0));
1843 Value *AbsArg = EmitFAbs(*this, Arg);
1844 Value *IsInf = Builder.CreateFCmpOEQ(
1845 AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf");
1846 Value *IsNeg = EmitSignBit(*this, Arg);
1847
1848 llvm::Type *IntTy = ConvertType(E->getType());
1849 Value *Zero = Constant::getNullValue(IntTy);
1850 Value *One = ConstantInt::get(IntTy, 1);
1851 Value *NegativeOne = ConstantInt::get(IntTy, -1);
1852 Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One);
1853 Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero);
1854 return RValue::get(Result);
1855 }
1856
1857 case Builtin::BI__builtin_isnormal: {
1858 // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min
1859 Value *V = EmitScalarExpr(E->getArg(0));
1860 Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq");
1861
1862 Value *Abs = EmitFAbs(*this, V);
1863 Value *IsLessThanInf =
1864 Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf");
1865 APFloat Smallest = APFloat::getSmallestNormalized(
1866 getContext().getFloatTypeSemantics(E->getArg(0)->getType()));
1867 Value *IsNormal =
1868 Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest),
1869 "isnormal");
1870 V = Builder.CreateAnd(Eq, IsLessThanInf, "and");
1871 V = Builder.CreateAnd(V, IsNormal, "and");
1872 return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType())));
1873 }
1874
1875 case Builtin::BI__builtin_fpclassify: {
1876 Value *V = EmitScalarExpr(E->getArg(5));
1877 llvm::Type *Ty = ConvertType(E->getArg(5)->getType());
1878
1879 // Create Result
1880 BasicBlock *Begin = Builder.GetInsertBlock();
1881 BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn);
1882 Builder.SetInsertPoint(End);
1883 PHINode *Result =
1884 Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4,
1885 "fpclassify_result");
1886
1887 // if (V==0) return FP_ZERO
1888 Builder.SetInsertPoint(Begin);
1889 Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty),
1890 "iszero");
1891 Value *ZeroLiteral = EmitScalarExpr(E->getArg(4));
1892 BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn);
1893 Builder.CreateCondBr(IsZero, End, NotZero);
1894 Result->addIncoming(ZeroLiteral, Begin);
1895
1896 // if (V != V) return FP_NAN
1897 Builder.SetInsertPoint(NotZero);
1898 Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp");
1899 Value *NanLiteral = EmitScalarExpr(E->getArg(0));
1900 BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn);
1901 Builder.CreateCondBr(IsNan, End, NotNan);
1902 Result->addIncoming(NanLiteral, NotZero);
1903
1904 // if (fabs(V) == infinity) return FP_INFINITY
1905 Builder.SetInsertPoint(NotNan);
1906 Value *VAbs = EmitFAbs(*this, V);
1907 Value *IsInf =
1908 Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()),
1909 "isinf");
1910 Value *InfLiteral = EmitScalarExpr(E->getArg(1));
1911 BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn);
1912 Builder.CreateCondBr(IsInf, End, NotInf);
1913 Result->addIncoming(InfLiteral, NotNan);
1914
1915 // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL
1916 Builder.SetInsertPoint(NotInf);
1917 APFloat Smallest = APFloat::getSmallestNormalized(
1918 getContext().getFloatTypeSemantics(E->getArg(5)->getType()));
1919 Value *IsNormal =
1920 Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest),
1921 "isnormal");
1922 Value *NormalResult =
1923 Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)),
1924 EmitScalarExpr(E->getArg(3)));
1925 Builder.CreateBr(End);
1926 Result->addIncoming(NormalResult, NotInf);
1927
1928 // return Result
1929 Builder.SetInsertPoint(End);
1930 return RValue::get(Result);
1931 }
1932
1933 case Builtin::BIalloca:
1934 case Builtin::BI_alloca:
1935 case Builtin::BI__builtin_alloca: {
1936 Value *Size = EmitScalarExpr(E->getArg(0));
1937 const TargetInfo &TI = getContext().getTargetInfo();
1938 // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__.
1939 unsigned SuitableAlignmentInBytes =
1940 CGM.getContext()
1941 .toCharUnitsFromBits(TI.getSuitableAlign())
1942 .getQuantity();
1943 AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size);
1944 AI->setAlignment(SuitableAlignmentInBytes);
1945 return RValue::get(AI);
1946 }
1947
1948 case Builtin::BI__builtin_alloca_with_align: {
1949 Value *Size = EmitScalarExpr(E->getArg(0));
1950 Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1));
1951 auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue);
1952 unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue();
1953 unsigned AlignmentInBytes =
1954 CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getQuantity();
1955 AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size);
1956 AI->setAlignment(AlignmentInBytes);
1957 return RValue::get(AI);
1958 }
1959
1960 case Builtin::BIbzero:
1961 case Builtin::BI__builtin_bzero: {
1962 Address Dest = EmitPointerWithAlignment(E->getArg(0));
1963 Value *SizeVal = EmitScalarExpr(E->getArg(1));
1964 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
1965 E->getArg(0)->getExprLoc(), FD, 0);
1966 Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false);
1967 return RValue::get(nullptr);
1968 }
1969 case Builtin::BImemcpy:
1970 case Builtin::BI__builtin_memcpy: {
1971 Address Dest = EmitPointerWithAlignment(E->getArg(0));
1972 Address Src = EmitPointerWithAlignment(E->getArg(1));
1973 Value *SizeVal = EmitScalarExpr(E->getArg(2));
1974 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
1975 E->getArg(0)->getExprLoc(), FD, 0);
1976 EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(),
1977 E->getArg(1)->getExprLoc(), FD, 1);
1978 Builder.CreateMemCpy(Dest, Src, SizeVal, false);
1979 return RValue::get(Dest.getPointer());
1980 }
1981
1982 case Builtin::BI__builtin_char_memchr:
1983 BuiltinID = Builtin::BI__builtin_memchr;
1984 break;
1985
1986 case Builtin::BI__builtin___memcpy_chk: {
1987 // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2.
1988 llvm::APSInt Size, DstSize;
1989 if (!E->getArg(2)->EvaluateAsInt(Size, CGM.getContext()) ||
1990 !E->getArg(3)->EvaluateAsInt(DstSize, CGM.getContext()))
1991 break;
1992 if (Size.ugt(DstSize))
1993 break;
1994 Address Dest = EmitPointerWithAlignment(E->getArg(0));
1995 Address Src = EmitPointerWithAlignment(E->getArg(1));
1996 Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
1997 Builder.CreateMemCpy(Dest, Src, SizeVal, false);
1998 return RValue::get(Dest.getPointer());
1999 }
2000
2001 case Builtin::BI__builtin_objc_memmove_collectable: {
2002 Address DestAddr = EmitPointerWithAlignment(E->getArg(0));
2003 Address SrcAddr = EmitPointerWithAlignment(E->getArg(1));
2004 Value *SizeVal = EmitScalarExpr(E->getArg(2));
2005 CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this,
2006 DestAddr, SrcAddr, SizeVal);
2007 return RValue::get(DestAddr.getPointer());
2008 }
2009
2010 case Builtin::BI__builtin___memmove_chk: {
2011 // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2.
2012 llvm::APSInt Size, DstSize;
2013 if (!E->getArg(2)->EvaluateAsInt(Size, CGM.getContext()) ||
2014 !E->getArg(3)->EvaluateAsInt(DstSize, CGM.getContext()))
2015 break;
2016 if (Size.ugt(DstSize))
2017 break;
2018 Address Dest = EmitPointerWithAlignment(E->getArg(0));
2019 Address Src = EmitPointerWithAlignment(E->getArg(1));
2020 Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
2021 Builder.CreateMemMove(Dest, Src, SizeVal, false);
2022 return RValue::get(Dest.getPointer());
2023 }
2024
2025 case Builtin::BImemmove:
2026 case Builtin::BI__builtin_memmove: {
2027 Address Dest = EmitPointerWithAlignment(E->getArg(0));
2028 Address Src = EmitPointerWithAlignment(E->getArg(1));
2029 Value *SizeVal = EmitScalarExpr(E->getArg(2));
2030 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
2031 E->getArg(0)->getExprLoc(), FD, 0);
2032 EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(),
2033 E->getArg(1)->getExprLoc(), FD, 1);
2034 Builder.CreateMemMove(Dest, Src, SizeVal, false);
2035 return RValue::get(Dest.getPointer());
2036 }
2037 case Builtin::BImemset:
2038 case Builtin::BI__builtin_memset: {
2039 Address Dest = EmitPointerWithAlignment(E->getArg(0));
2040 Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)),
2041 Builder.getInt8Ty());
2042 Value *SizeVal = EmitScalarExpr(E->getArg(2));
2043 EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
2044 E->getArg(0)->getExprLoc(), FD, 0);
2045 Builder.CreateMemSet(Dest, ByteVal, SizeVal, false);
2046 return RValue::get(Dest.getPointer());
2047 }
2048 case Builtin::BI__builtin___memset_chk: {
2049 // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2.
2050 llvm::APSInt Size, DstSize;
2051 if (!E->getArg(2)->EvaluateAsInt(Size, CGM.getContext()) ||
2052 !E->getArg(3)->EvaluateAsInt(DstSize, CGM.getContext()))
2053 break;
2054 if (Size.ugt(DstSize))
2055 break;
2056 Address Dest = EmitPointerWithAlignment(E->getArg(0));
2057 Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)),
2058 Builder.getInt8Ty());
2059 Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
2060 Builder.CreateMemSet(Dest, ByteVal, SizeVal, false);
2061 return RValue::get(Dest.getPointer());
2062 }
2063 case Builtin::BI__builtin_wmemcmp: {
2064 // The MSVC runtime library does not provide a definition of wmemcmp, so we
2065 // need an inline implementation.
2066 if (!getTarget().getTriple().isOSMSVCRT())
2067 break;
2068
2069 llvm::Type *WCharTy = ConvertType(getContext().WCharTy);
2070
2071 Value *Dst = EmitScalarExpr(E->getArg(0));
2072 Value *Src = EmitScalarExpr(E->getArg(1));
2073 Value *Size = EmitScalarExpr(E->getArg(2));
2074
2075 BasicBlock *Entry = Builder.GetInsertBlock();
2076 BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt");
2077 BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt");
2078 BasicBlock *Next = createBasicBlock("wmemcmp.next");
2079 BasicBlock *Exit = createBasicBlock("wmemcmp.exit");
2080 Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0));
2081 Builder.CreateCondBr(SizeEq0, Exit, CmpGT);
2082
2083 EmitBlock(CmpGT);
2084 PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2);
2085 DstPhi->addIncoming(Dst, Entry);
2086 PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2);
2087 SrcPhi->addIncoming(Src, Entry);
2088 PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2);
2089 SizePhi->addIncoming(Size, Entry);
2090 CharUnits WCharAlign =
2091 getContext().getTypeAlignInChars(getContext().WCharTy);
2092 Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign);
2093 Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign);
2094 Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh);
2095 Builder.CreateCondBr(DstGtSrc, Exit, CmpLT);
2096
2097 EmitBlock(CmpLT);
2098 Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh);
2099 Builder.CreateCondBr(DstLtSrc, Exit, Next);
2100
2101 EmitBlock(Next);
2102 Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1);
2103 Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1);
2104 Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1));
2105 Value *NextSizeEq0 =
2106 Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0));
2107 Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT);
2108 DstPhi->addIncoming(NextDst, Next);
2109 SrcPhi->addIncoming(NextSrc, Next);
2110 SizePhi->addIncoming(NextSize, Next);
2111
2112 EmitBlock(Exit);
2113 PHINode *Ret = Builder.CreatePHI(IntTy, 4);
2114 Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry);
2115 Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT);
2116 Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT);
2117 Ret->addIncoming(ConstantInt::get(IntTy, 0), Next);
2118 return RValue::get(Ret);
2119 }
2120 case Builtin::BI__builtin_dwarf_cfa: {
2121 // The offset in bytes from the first argument to the CFA.
2122 //
2123 // Why on earth is this in the frontend? Is there any reason at
2124 // all that the backend can't reasonably determine this while
2125 // lowering llvm.eh.dwarf.cfa()?
2126 //
2127 // TODO: If there's a satisfactory reason, add a target hook for
2128 // this instead of hard-coding 0, which is correct for most targets.
2129 int32_t Offset = 0;
2130
2131 Value *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa);
2132 return RValue::get(Builder.CreateCall(F,
2133 llvm::ConstantInt::get(Int32Ty, Offset)));
2134 }
2135 case Builtin::BI__builtin_return_address: {
2136 Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0),
2137 getContext().UnsignedIntTy);
2138 Value *F = CGM.getIntrinsic(Intrinsic::returnaddress);
2139 return RValue::get(Builder.CreateCall(F, Depth));
2140 }
2141 case Builtin::BI_ReturnAddress: {
2142 Value *F = CGM.getIntrinsic(Intrinsic::returnaddress);
2143 return RValue::get(Builder.CreateCall(F, Builder.getInt32(0)));
2144 }
2145 case Builtin::BI__builtin_frame_address: {
2146 Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0),
2147 getContext().UnsignedIntTy);
2148 Value *F = CGM.getIntrinsic(Intrinsic::frameaddress);
2149 return RValue::get(Builder.CreateCall(F, Depth));
2150 }
2151 case Builtin::BI__builtin_extract_return_addr: {
2152 Value *Address = EmitScalarExpr(E->getArg(0));
2153 Value *Result = getTargetHooks().decodeReturnAddress(*this, Address);
2154 return RValue::get(Result);
2155 }
2156 case Builtin::BI__builtin_frob_return_addr: {
2157 Value *Address = EmitScalarExpr(E->getArg(0));
2158 Value *Result = getTargetHooks().encodeReturnAddress(*this, Address);
2159 return RValue::get(Result);
2160 }
2161 case Builtin::BI__builtin_dwarf_sp_column: {
2162 llvm::IntegerType *Ty
2163 = cast<llvm::IntegerType>(ConvertType(E->getType()));
2164 int Column = getTargetHooks().getDwarfEHStackPointer(CGM);
2165 if (Column == -1) {
2166 CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column");
2167 return RValue::get(llvm::UndefValue::get(Ty));
2168 }
2169 return RValue::get(llvm::ConstantInt::get(Ty, Column, true));
2170 }
2171 case Builtin::BI__builtin_init_dwarf_reg_size_table: {
2172 Value *Address = EmitScalarExpr(E->getArg(0));
2173 if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address))
2174 CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table");
2175 return RValue::get(llvm::UndefValue::get(ConvertType(E->getType())));
2176 }
2177 case Builtin::BI__builtin_eh_return: {
2178 Value *Int = EmitScalarExpr(E->getArg(0));
2179 Value *Ptr = EmitScalarExpr(E->getArg(1));
2180
2181 llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType());
2182 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 2183, __extension__ __PRETTY_FUNCTION__))
2183 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 2183, __extension__ __PRETTY_FUNCTION__))
;
2184 Value *F = CGM.getIntrinsic(IntTy->getBitWidth() == 32
2185 ? Intrinsic::eh_return_i32
2186 : Intrinsic::eh_return_i64);
2187 Builder.CreateCall(F, {Int, Ptr});
2188 Builder.CreateUnreachable();
2189
2190 // We do need to preserve an insertion point.
2191 EmitBlock(createBasicBlock("builtin_eh_return.cont"));
2192
2193 return RValue::get(nullptr);
2194 }
2195 case Builtin::BI__builtin_unwind_init: {
2196 Value *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init);
2197 return RValue::get(Builder.CreateCall(F));
2198 }
2199 case Builtin::BI__builtin_extend_pointer: {
2200 // Extends a pointer to the size of an _Unwind_Word, which is
2201 // uint64_t on all platforms. Generally this gets poked into a
2202 // register and eventually used as an address, so if the
2203 // addressing registers are wider than pointers and the platform
2204 // doesn't implicitly ignore high-order bits when doing
2205 // addressing, we need to make sure we zext / sext based on
2206 // the platform's expectations.
2207 //
2208 // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html
2209
2210 // Cast the pointer to intptr_t.
2211 Value *Ptr = EmitScalarExpr(E->getArg(0));
2212 Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast");
2213
2214 // If that's 64 bits, we're done.
2215 if (IntPtrTy->getBitWidth() == 64)
2216 return RValue::get(Result);
2217
2218 // Otherwise, ask the codegen data what to do.
2219 if (getTargetHooks().extendPointerWithSExt())
2220 return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext"));
2221 else
2222 return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext"));
2223 }
2224 case Builtin::BI__builtin_setjmp: {
2225 // Buffer is a void**.
2226 Address Buf = EmitPointerWithAlignment(E->getArg(0));
2227
2228 // Store the frame pointer to the setjmp buffer.
2229 Value *FrameAddr =
2230 Builder.CreateCall(CGM.getIntrinsic(Intrinsic::frameaddress),
2231 ConstantInt::get(Int32Ty, 0));
2232 Builder.CreateStore(FrameAddr, Buf);
2233
2234 // Store the stack pointer to the setjmp buffer.
2235 Value *StackAddr =
2236 Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave));
2237 Address StackSaveSlot =
2238 Builder.CreateConstInBoundsGEP(Buf, 2, getPointerSize());
2239 Builder.CreateStore(StackAddr, StackSaveSlot);
2240
2241 // Call LLVM's EH setjmp, which is lightweight.
2242 Value *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp);
2243 Buf = Builder.CreateBitCast(Buf, Int8PtrTy);
2244 return RValue::get(Builder.CreateCall(F, Buf.getPointer()));
2245 }
2246 case Builtin::BI__builtin_longjmp: {
2247 Value *Buf = EmitScalarExpr(E->getArg(0));
2248 Buf = Builder.CreateBitCast(Buf, Int8PtrTy);
2249
2250 // Call LLVM's EH longjmp, which is lightweight.
2251 Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf);
2252
2253 // longjmp doesn't return; mark this as unreachable.
2254 Builder.CreateUnreachable();
2255
2256 // We do need to preserve an insertion point.
2257 EmitBlock(createBasicBlock("longjmp.cont"));
2258
2259 return RValue::get(nullptr);
2260 }
2261 case Builtin::BI__sync_fetch_and_add:
2262 case Builtin::BI__sync_fetch_and_sub:
2263 case Builtin::BI__sync_fetch_and_or:
2264 case Builtin::BI__sync_fetch_and_and:
2265 case Builtin::BI__sync_fetch_and_xor:
2266 case Builtin::BI__sync_fetch_and_nand:
2267 case Builtin::BI__sync_add_and_fetch:
2268 case Builtin::BI__sync_sub_and_fetch:
2269 case Builtin::BI__sync_and_and_fetch:
2270 case Builtin::BI__sync_or_and_fetch:
2271 case Builtin::BI__sync_xor_and_fetch:
2272 case Builtin::BI__sync_nand_and_fetch:
2273 case Builtin::BI__sync_val_compare_and_swap:
2274 case Builtin::BI__sync_bool_compare_and_swap:
2275 case Builtin::BI__sync_lock_test_and_set:
2276 case Builtin::BI__sync_lock_release:
2277 case Builtin::BI__sync_swap:
2278 llvm_unreachable("Shouldn't make it through sema")::llvm::llvm_unreachable_internal("Shouldn't make it through sema"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 2278)
;
2279 case Builtin::BI__sync_fetch_and_add_1:
2280 case Builtin::BI__sync_fetch_and_add_2:
2281 case Builtin::BI__sync_fetch_and_add_4:
2282 case Builtin::BI__sync_fetch_and_add_8:
2283 case Builtin::BI__sync_fetch_and_add_16:
2284 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E);
2285 case Builtin::BI__sync_fetch_and_sub_1:
2286 case Builtin::BI__sync_fetch_and_sub_2:
2287 case Builtin::BI__sync_fetch_and_sub_4:
2288 case Builtin::BI__sync_fetch_and_sub_8:
2289 case Builtin::BI__sync_fetch_and_sub_16:
2290 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E);
2291 case Builtin::BI__sync_fetch_and_or_1:
2292 case Builtin::BI__sync_fetch_and_or_2:
2293 case Builtin::BI__sync_fetch_and_or_4:
2294 case Builtin::BI__sync_fetch_and_or_8:
2295 case Builtin::BI__sync_fetch_and_or_16:
2296 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E);
2297 case Builtin::BI__sync_fetch_and_and_1:
2298 case Builtin::BI__sync_fetch_and_and_2:
2299 case Builtin::BI__sync_fetch_and_and_4:
2300 case Builtin::BI__sync_fetch_and_and_8:
2301 case Builtin::BI__sync_fetch_and_and_16:
2302 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E);
2303 case Builtin::BI__sync_fetch_and_xor_1:
2304 case Builtin::BI__sync_fetch_and_xor_2:
2305 case Builtin::BI__sync_fetch_and_xor_4:
2306 case Builtin::BI__sync_fetch_and_xor_8:
2307 case Builtin::BI__sync_fetch_and_xor_16:
2308 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E);
2309 case Builtin::BI__sync_fetch_and_nand_1:
2310 case Builtin::BI__sync_fetch_and_nand_2:
2311 case Builtin::BI__sync_fetch_and_nand_4:
2312 case Builtin::BI__sync_fetch_and_nand_8:
2313 case Builtin::BI__sync_fetch_and_nand_16:
2314 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E);
2315
2316 // Clang extensions: not overloaded yet.
2317 case Builtin::BI__sync_fetch_and_min:
2318 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E);
2319 case Builtin::BI__sync_fetch_and_max:
2320 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E);
2321 case Builtin::BI__sync_fetch_and_umin:
2322 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E);
2323 case Builtin::BI__sync_fetch_and_umax:
2324 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E);
2325
2326 case Builtin::BI__sync_add_and_fetch_1:
2327 case Builtin::BI__sync_add_and_fetch_2:
2328 case Builtin::BI__sync_add_and_fetch_4:
2329 case Builtin::BI__sync_add_and_fetch_8:
2330 case Builtin::BI__sync_add_and_fetch_16:
2331 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E,
2332 llvm::Instruction::Add);
2333 case Builtin::BI__sync_sub_and_fetch_1:
2334 case Builtin::BI__sync_sub_and_fetch_2:
2335 case Builtin::BI__sync_sub_and_fetch_4:
2336 case Builtin::BI__sync_sub_and_fetch_8:
2337 case Builtin::BI__sync_sub_and_fetch_16:
2338 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E,
2339 llvm::Instruction::Sub);
2340 case Builtin::BI__sync_and_and_fetch_1:
2341 case Builtin::BI__sync_and_and_fetch_2:
2342 case Builtin::BI__sync_and_and_fetch_4:
2343 case Builtin::BI__sync_and_and_fetch_8:
2344 case Builtin::BI__sync_and_and_fetch_16:
2345 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E,
2346 llvm::Instruction::And);
2347 case Builtin::BI__sync_or_and_fetch_1:
2348 case Builtin::BI__sync_or_and_fetch_2:
2349 case Builtin::BI__sync_or_and_fetch_4:
2350 case Builtin::BI__sync_or_and_fetch_8:
2351 case Builtin::BI__sync_or_and_fetch_16:
2352 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E,
2353 llvm::Instruction::Or);
2354 case Builtin::BI__sync_xor_and_fetch_1:
2355 case Builtin::BI__sync_xor_and_fetch_2:
2356 case Builtin::BI__sync_xor_and_fetch_4:
2357 case Builtin::BI__sync_xor_and_fetch_8:
2358 case Builtin::BI__sync_xor_and_fetch_16:
2359 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E,
2360 llvm::Instruction::Xor);
2361 case Builtin::BI__sync_nand_and_fetch_1:
2362 case Builtin::BI__sync_nand_and_fetch_2:
2363 case Builtin::BI__sync_nand_and_fetch_4:
2364 case Builtin::BI__sync_nand_and_fetch_8:
2365 case Builtin::BI__sync_nand_and_fetch_16:
2366 return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E,
2367 llvm::Instruction::And, true);
2368
2369 case Builtin::BI__sync_val_compare_and_swap_1:
2370 case Builtin::BI__sync_val_compare_and_swap_2:
2371 case Builtin::BI__sync_val_compare_and_swap_4:
2372 case Builtin::BI__sync_val_compare_and_swap_8:
2373 case Builtin::BI__sync_val_compare_and_swap_16:
2374 return RValue::get(MakeAtomicCmpXchgValue(*this, E, false));
2375
2376 case Builtin::BI__sync_bool_compare_and_swap_1:
2377 case Builtin::BI__sync_bool_compare_and_swap_2:
2378 case Builtin::BI__sync_bool_compare_and_swap_4:
2379 case Builtin::BI__sync_bool_compare_and_swap_8:
2380 case Builtin::BI__sync_bool_compare_and_swap_16:
2381 return RValue::get(MakeAtomicCmpXchgValue(*this, E, true));
2382
2383 case Builtin::BI__sync_swap_1:
2384 case Builtin::BI__sync_swap_2:
2385 case Builtin::BI__sync_swap_4:
2386 case Builtin::BI__sync_swap_8:
2387 case Builtin::BI__sync_swap_16:
2388 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E);
2389
2390 case Builtin::BI__sync_lock_test_and_set_1:
2391 case Builtin::BI__sync_lock_test_and_set_2:
2392 case Builtin::BI__sync_lock_test_and_set_4:
2393 case Builtin::BI__sync_lock_test_and_set_8:
2394 case Builtin::BI__sync_lock_test_and_set_16:
2395 return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E);
2396
2397 case Builtin::BI__sync_lock_release_1:
2398 case Builtin::BI__sync_lock_release_2:
2399 case Builtin::BI__sync_lock_release_4:
2400 case Builtin::BI__sync_lock_release_8:
2401 case Builtin::BI__sync_lock_release_16: {
2402 Value *Ptr = EmitScalarExpr(E->getArg(0));
2403 QualType ElTy = E->getArg(0)->getType()->getPointeeType();
2404 CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy);
2405 llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(),
2406 StoreSize.getQuantity() * 8);
2407 Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo());
2408 llvm::StoreInst *Store =
2409 Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr,
2410 StoreSize);
2411 Store->setAtomic(llvm::AtomicOrdering::Release);
2412 return RValue::get(nullptr);
2413 }
2414
2415 case Builtin::BI__sync_synchronize: {
2416 // We assume this is supposed to correspond to a C++0x-style
2417 // sequentially-consistent fence (i.e. this is only usable for
2418 // synchronization, not device I/O or anything like that). This intrinsic
2419 // is really badly designed in the sense that in theory, there isn't
2420 // any way to safely use it... but in practice, it mostly works
2421 // to use it with non-atomic loads and stores to get acquire/release
2422 // semantics.
2423 Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent);
2424 return RValue::get(nullptr);
2425 }
2426
2427 case Builtin::BI__builtin_nontemporal_load:
2428 return RValue::get(EmitNontemporalLoad(*this, E));
2429 case Builtin::BI__builtin_nontemporal_store:
2430 return RValue::get(EmitNontemporalStore(*this, E));
2431 case Builtin::BI__c11_atomic_is_lock_free:
2432 case Builtin::BI__atomic_is_lock_free: {
2433 // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the
2434 // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since
2435 // _Atomic(T) is always properly-aligned.
2436 const char *LibCallName = "__atomic_is_lock_free";
2437 CallArgList Args;
2438 Args.add(RValue::get(EmitScalarExpr(E->getArg(0))),
2439 getContext().getSizeType());
2440 if (BuiltinID == Builtin::BI__atomic_is_lock_free)
2441 Args.add(RValue::get(EmitScalarExpr(E->getArg(1))),
2442 getContext().VoidPtrTy);
2443 else
2444 Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)),
2445 getContext().VoidPtrTy);
2446 const CGFunctionInfo &FuncInfo =
2447 CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args);
2448 llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo);
2449 llvm::Constant *Func = CGM.CreateRuntimeFunction(FTy, LibCallName);
2450 return EmitCall(FuncInfo, CGCallee::forDirect(Func),
2451 ReturnValueSlot(), Args);
2452 }
2453
2454 case Builtin::BI__atomic_test_and_set: {
2455 // Look at the argument type to determine whether this is a volatile
2456 // operation. The parameter type is always volatile.
2457 QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType();
2458 bool Volatile =
2459 PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified();
2460
2461 Value *Ptr = EmitScalarExpr(E->getArg(0));
2462 unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace();
2463 Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace));
2464 Value *NewVal = Builder.getInt8(1);
2465 Value *Order = EmitScalarExpr(E->getArg(1));
2466 if (isa<llvm::ConstantInt>(Order)) {
2467 int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
2468 AtomicRMWInst *Result = nullptr;
2469 switch (ord) {
2470 case 0: // memory_order_relaxed
2471 default: // invalid order
2472 Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2473 llvm::AtomicOrdering::Monotonic);
2474 break;
2475 case 1: // memory_order_consume
2476 case 2: // memory_order_acquire
2477 Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2478 llvm::AtomicOrdering::Acquire);
2479 break;
2480 case 3: // memory_order_release
2481 Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2482 llvm::AtomicOrdering::Release);
2483 break;
2484 case 4: // memory_order_acq_rel
2485
2486 Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2487 llvm::AtomicOrdering::AcquireRelease);
2488 break;
2489 case 5: // memory_order_seq_cst
2490 Result = Builder.CreateAtomicRMW(
2491 llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2492 llvm::AtomicOrdering::SequentiallyConsistent);
2493 break;
2494 }
2495 Result->setVolatile(Volatile);
2496 return RValue::get(Builder.CreateIsNotNull(Result, "tobool"));
2497 }
2498
2499 llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
2500
2501 llvm::BasicBlock *BBs[5] = {
2502 createBasicBlock("monotonic", CurFn),
2503 createBasicBlock("acquire", CurFn),
2504 createBasicBlock("release", CurFn),
2505 createBasicBlock("acqrel", CurFn),
2506 createBasicBlock("seqcst", CurFn)
2507 };
2508 llvm::AtomicOrdering Orders[5] = {
2509 llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire,
2510 llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease,
2511 llvm::AtomicOrdering::SequentiallyConsistent};
2512
2513 Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
2514 llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]);
2515
2516 Builder.SetInsertPoint(ContBB);
2517 PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set");
2518
2519 for (unsigned i = 0; i < 5; ++i) {
2520 Builder.SetInsertPoint(BBs[i]);
2521 AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg,
2522 Ptr, NewVal, Orders[i]);
2523 RMW->setVolatile(Volatile);
2524 Result->addIncoming(RMW, BBs[i]);
2525 Builder.CreateBr(ContBB);
2526 }
2527
2528 SI->addCase(Builder.getInt32(0), BBs[0]);
2529 SI->addCase(Builder.getInt32(1), BBs[1]);
2530 SI->addCase(Builder.getInt32(2), BBs[1]);
2531 SI->addCase(Builder.getInt32(3), BBs[2]);
2532 SI->addCase(Builder.getInt32(4), BBs[3]);
2533 SI->addCase(Builder.getInt32(5), BBs[4]);
2534
2535 Builder.SetInsertPoint(ContBB);
2536 return RValue::get(Builder.CreateIsNotNull(Result, "tobool"));
2537 }
2538
2539 case Builtin::BI__atomic_clear: {
2540 QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType();
2541 bool Volatile =
2542 PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified();
2543
2544 Address Ptr = EmitPointerWithAlignment(E->getArg(0));
2545 unsigned AddrSpace = Ptr.getPointer()->getType()->getPointerAddressSpace();
2546 Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace));
2547 Value *NewVal = Builder.getInt8(0);
2548 Value *Order = EmitScalarExpr(E->getArg(1));
2549 if (isa<llvm::ConstantInt>(Order)) {
2550 int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
2551 StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile);
2552 switch (ord) {
2553 case 0: // memory_order_relaxed
2554 default: // invalid order
2555 Store->setOrdering(llvm::AtomicOrdering::Monotonic);
2556 break;
2557 case 3: // memory_order_release
2558 Store->setOrdering(llvm::AtomicOrdering::Release);
2559 break;
2560 case 5: // memory_order_seq_cst
2561 Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent);
2562 break;
2563 }
2564 return RValue::get(nullptr);
2565 }
2566
2567 llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
2568
2569 llvm::BasicBlock *BBs[3] = {
2570 createBasicBlock("monotonic", CurFn),
2571 createBasicBlock("release", CurFn),
2572 createBasicBlock("seqcst", CurFn)
2573 };
2574 llvm::AtomicOrdering Orders[3] = {
2575 llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release,
2576 llvm::AtomicOrdering::SequentiallyConsistent};
2577
2578 Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
2579 llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]);
2580
2581 for (unsigned i = 0; i < 3; ++i) {
2582 Builder.SetInsertPoint(BBs[i]);
2583 StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile);
2584 Store->setOrdering(Orders[i]);
2585 Builder.CreateBr(ContBB);
2586 }
2587
2588 SI->addCase(Builder.getInt32(0), BBs[0]);
2589 SI->addCase(Builder.getInt32(3), BBs[1]);
2590 SI->addCase(Builder.getInt32(5), BBs[2]);
2591
2592 Builder.SetInsertPoint(ContBB);
2593 return RValue::get(nullptr);
2594 }
2595
2596 case Builtin::BI__atomic_thread_fence:
2597 case Builtin::BI__atomic_signal_fence:
2598 case Builtin::BI__c11_atomic_thread_fence:
2599 case Builtin::BI__c11_atomic_signal_fence: {
2600 llvm::SyncScope::ID SSID;
2601 if (BuiltinID == Builtin::BI__atomic_signal_fence ||
2602 BuiltinID == Builtin::BI__c11_atomic_signal_fence)
2603 SSID = llvm::SyncScope::SingleThread;
2604 else
2605 SSID = llvm::SyncScope::System;
2606 Value *Order = EmitScalarExpr(E->getArg(0));
2607 if (isa<llvm::ConstantInt>(Order)) {
2608 int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
2609 switch (ord) {
2610 case 0: // memory_order_relaxed
2611 default: // invalid order
2612 break;
2613 case 1: // memory_order_consume
2614 case 2: // memory_order_acquire
2615 Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID);
2616 break;
2617 case 3: // memory_order_release
2618 Builder.CreateFence(llvm::AtomicOrdering::Release, SSID);
2619 break;
2620 case 4: // memory_order_acq_rel
2621 Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID);
2622 break;
2623 case 5: // memory_order_seq_cst
2624 Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID);
2625 break;
2626 }
2627 return RValue::get(nullptr);
2628 }
2629
2630 llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB;
2631 AcquireBB = createBasicBlock("acquire", CurFn);
2632 ReleaseBB = createBasicBlock("release", CurFn);
2633 AcqRelBB = createBasicBlock("acqrel", CurFn);
2634 SeqCstBB = createBasicBlock("seqcst", CurFn);
2635 llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
2636
2637 Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
2638 llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB);
2639
2640 Builder.SetInsertPoint(AcquireBB);
2641 Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID);
2642 Builder.CreateBr(ContBB);
2643 SI->addCase(Builder.getInt32(1), AcquireBB);
2644 SI->addCase(Builder.getInt32(2), AcquireBB);
2645
2646 Builder.SetInsertPoint(ReleaseBB);
2647 Builder.CreateFence(llvm::AtomicOrdering::Release, SSID);
2648 Builder.CreateBr(ContBB);
2649 SI->addCase(Builder.getInt32(3), ReleaseBB);
2650
2651 Builder.SetInsertPoint(AcqRelBB);
2652 Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID);
2653 Builder.CreateBr(ContBB);
2654 SI->addCase(Builder.getInt32(4), AcqRelBB);
2655
2656 Builder.SetInsertPoint(SeqCstBB);
2657 Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID);
2658 Builder.CreateBr(ContBB);
2659 SI->addCase(Builder.getInt32(5), SeqCstBB);
2660
2661 Builder.SetInsertPoint(ContBB);
2662 return RValue::get(nullptr);
2663 }
2664
2665 case Builtin::BI__builtin_signbit:
2666 case Builtin::BI__builtin_signbitf:
2667 case Builtin::BI__builtin_signbitl: {
2668 return RValue::get(
2669 Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))),
2670 ConvertType(E->getType())));
2671 }
2672 case Builtin::BI__annotation: {
2673 // Re-encode each wide string to UTF8 and make an MDString.
2674 SmallVector<Metadata *, 1> Strings;
2675 for (const Expr *Arg : E->arguments()) {
2676 const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts());
2677 assert(Str->getCharByteWidth() == 2)(static_cast <bool> (Str->getCharByteWidth() == 2) ?
void (0) : __assert_fail ("Str->getCharByteWidth() == 2",
"/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 2677, __extension__ __PRETTY_FUNCTION__))
;
2678 StringRef WideBytes = Str->getBytes();
2679 std::string StrUtf8;
2680 if (!convertUTF16ToUTF8String(
2681 makeArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) {
2682 CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument");
2683 continue;
2684 }
2685 Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8));
2686 }
2687
2688 // Build and MDTuple of MDStrings and emit the intrinsic call.
2689 llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {});
2690 MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings);
2691 Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple));
2692 return RValue::getIgnored();
2693 }
2694 case Builtin::BI__builtin_annotation: {
2695 llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0));
2696 llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::annotation,
2697 AnnVal->getType());
2698
2699 // Get the annotation string, go through casts. Sema requires this to be a
2700 // non-wide string literal, potentially casted, so the cast<> is safe.
2701 const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts();
2702 StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString();
2703 return RValue::get(EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc()));
2704 }
2705 case Builtin::BI__builtin_addcb:
2706 case Builtin::BI__builtin_addcs:
2707 case Builtin::BI__builtin_addc:
2708 case Builtin::BI__builtin_addcl:
2709 case Builtin::BI__builtin_addcll:
2710 case Builtin::BI__builtin_subcb:
2711 case Builtin::BI__builtin_subcs:
2712 case Builtin::BI__builtin_subc:
2713 case Builtin::BI__builtin_subcl:
2714 case Builtin::BI__builtin_subcll: {
2715
2716 // We translate all of these builtins from expressions of the form:
2717 // int x = ..., y = ..., carryin = ..., carryout, result;
2718 // result = __builtin_addc(x, y, carryin, &carryout);
2719 //
2720 // to LLVM IR of the form:
2721 //
2722 // %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y)
2723 // %tmpsum1 = extractvalue {i32, i1} %tmp1, 0
2724 // %carry1 = extractvalue {i32, i1} %tmp1, 1
2725 // %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1,
2726 // i32 %carryin)
2727 // %result = extractvalue {i32, i1} %tmp2, 0
2728 // %carry2 = extractvalue {i32, i1} %tmp2, 1
2729 // %tmp3 = or i1 %carry1, %carry2
2730 // %tmp4 = zext i1 %tmp3 to i32
2731 // store i32 %tmp4, i32* %carryout
2732
2733 // Scalarize our inputs.
2734 llvm::Value *X = EmitScalarExpr(E->getArg(0));
2735 llvm::Value *Y = EmitScalarExpr(E->getArg(1));
2736 llvm::Value *Carryin = EmitScalarExpr(E->getArg(2));
2737 Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3));
2738
2739 // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow.
2740 llvm::Intrinsic::ID IntrinsicId;
2741 switch (BuiltinID) {
2742 default: llvm_unreachable("Unknown multiprecision builtin id.")::llvm::llvm_unreachable_internal("Unknown multiprecision builtin id."
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 2742)
;
2743 case Builtin::BI__builtin_addcb:
2744 case Builtin::BI__builtin_addcs:
2745 case Builtin::BI__builtin_addc:
2746 case Builtin::BI__builtin_addcl:
2747 case Builtin::BI__builtin_addcll:
2748 IntrinsicId = llvm::Intrinsic::uadd_with_overflow;
2749 break;
2750 case Builtin::BI__builtin_subcb:
2751 case Builtin::BI__builtin_subcs:
2752 case Builtin::BI__builtin_subc:
2753 case Builtin::BI__builtin_subcl:
2754 case Builtin::BI__builtin_subcll:
2755 IntrinsicId = llvm::Intrinsic::usub_with_overflow;
2756 break;
2757 }
2758
2759 // Construct our resulting LLVM IR expression.
2760 llvm::Value *Carry1;
2761 llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId,
2762 X, Y, Carry1);
2763 llvm::Value *Carry2;
2764 llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId,
2765 Sum1, Carryin, Carry2);
2766 llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2),
2767 X->getType());
2768 Builder.CreateStore(CarryOut, CarryOutPtr);
2769 return RValue::get(Sum2);
2770 }
2771
2772 case Builtin::BI__builtin_add_overflow:
2773 case Builtin::BI__builtin_sub_overflow:
2774 case Builtin::BI__builtin_mul_overflow: {
2775 const clang::Expr *LeftArg = E->getArg(0);
2776 const clang::Expr *RightArg = E->getArg(1);
2777 const clang::Expr *ResultArg = E->getArg(2);
2778
2779 clang::QualType ResultQTy =
2780 ResultArg->getType()->castAs<PointerType>()->getPointeeType();
2781
2782 WidthAndSignedness LeftInfo =
2783 getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType());
2784 WidthAndSignedness RightInfo =
2785 getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType());
2786 WidthAndSignedness ResultInfo =
2787 getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy);
2788
2789 // Handle mixed-sign multiplication as a special case, because adding
2790 // runtime or backend support for our generic irgen would be too expensive.
2791 if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo))
2792 return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg,
2793 RightInfo, ResultArg, ResultQTy,
2794 ResultInfo);
2795
2796 WidthAndSignedness EncompassingInfo =
2797 EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo});
2798
2799 llvm::Type *EncompassingLLVMTy =
2800 llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width);
2801
2802 llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy);
2803
2804 llvm::Intrinsic::ID IntrinsicId;
2805 switch (BuiltinID) {
2806 default:
2807 llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id."
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 2807)
;
2808 case Builtin::BI__builtin_add_overflow:
2809 IntrinsicId = EncompassingInfo.Signed
2810 ? llvm::Intrinsic::sadd_with_overflow
2811 : llvm::Intrinsic::uadd_with_overflow;
2812 break;
2813 case Builtin::BI__builtin_sub_overflow:
2814 IntrinsicId = EncompassingInfo.Signed
2815 ? llvm::Intrinsic::ssub_with_overflow
2816 : llvm::Intrinsic::usub_with_overflow;
2817 break;
2818 case Builtin::BI__builtin_mul_overflow:
2819 IntrinsicId = EncompassingInfo.Signed
2820 ? llvm::Intrinsic::smul_with_overflow
2821 : llvm::Intrinsic::umul_with_overflow;
2822 break;
2823 }
2824
2825 llvm::Value *Left = EmitScalarExpr(LeftArg);
2826 llvm::Value *Right = EmitScalarExpr(RightArg);
2827 Address ResultPtr = EmitPointerWithAlignment(ResultArg);
2828
2829 // Extend each operand to the encompassing type.
2830 Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed);
2831 Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed);
2832
2833 // Perform the operation on the extended values.
2834 llvm::Value *Overflow, *Result;
2835 Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow);
2836
2837 if (EncompassingInfo.Width > ResultInfo.Width) {
2838 // The encompassing type is wider than the result type, so we need to
2839 // truncate it.
2840 llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy);
2841
2842 // To see if the truncation caused an overflow, we will extend
2843 // the result and then compare it to the original result.
2844 llvm::Value *ResultTruncExt = Builder.CreateIntCast(
2845 ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed);
2846 llvm::Value *TruncationOverflow =
2847 Builder.CreateICmpNE(Result, ResultTruncExt);
2848
2849 Overflow = Builder.CreateOr(Overflow, TruncationOverflow);
2850 Result = ResultTrunc;
2851 }
2852
2853 // Finally, store the result using the pointer.
2854 bool isVolatile =
2855 ResultArg->getType()->getPointeeType().isVolatileQualified();
2856 Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile);
2857
2858 return RValue::get(Overflow);
2859 }
2860
2861 case Builtin::BI__builtin_uadd_overflow:
2862 case Builtin::BI__builtin_uaddl_overflow:
2863 case Builtin::BI__builtin_uaddll_overflow:
2864 case Builtin::BI__builtin_usub_overflow:
2865 case Builtin::BI__builtin_usubl_overflow:
2866 case Builtin::BI__builtin_usubll_overflow:
2867 case Builtin::BI__builtin_umul_overflow:
2868 case Builtin::BI__builtin_umull_overflow:
2869 case Builtin::BI__builtin_umulll_overflow:
2870 case Builtin::BI__builtin_sadd_overflow:
2871 case Builtin::BI__builtin_saddl_overflow:
2872 case Builtin::BI__builtin_saddll_overflow:
2873 case Builtin::BI__builtin_ssub_overflow:
2874 case Builtin::BI__builtin_ssubl_overflow:
2875 case Builtin::BI__builtin_ssubll_overflow:
2876 case Builtin::BI__builtin_smul_overflow:
2877 case Builtin::BI__builtin_smull_overflow:
2878 case Builtin::BI__builtin_smulll_overflow: {
2879
2880 // We translate all of these builtins directly to the relevant llvm IR node.
2881
2882 // Scalarize our inputs.
2883 llvm::Value *X = EmitScalarExpr(E->getArg(0));
2884 llvm::Value *Y = EmitScalarExpr(E->getArg(1));
2885 Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2));
2886
2887 // Decide which of the overflow intrinsics we are lowering to:
2888 llvm::Intrinsic::ID IntrinsicId;
2889 switch (BuiltinID) {
2890 default: llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id."
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 2890)
;
2891 case Builtin::BI__builtin_uadd_overflow:
2892 case Builtin::BI__builtin_uaddl_overflow:
2893 case Builtin::BI__builtin_uaddll_overflow:
2894 IntrinsicId = llvm::Intrinsic::uadd_with_overflow;
2895 break;
2896 case Builtin::BI__builtin_usub_overflow:
2897 case Builtin::BI__builtin_usubl_overflow:
2898 case Builtin::BI__builtin_usubll_overflow:
2899 IntrinsicId = llvm::Intrinsic::usub_with_overflow;
2900 break;
2901 case Builtin::BI__builtin_umul_overflow:
2902 case Builtin::BI__builtin_umull_overflow:
2903 case Builtin::BI__builtin_umulll_overflow:
2904 IntrinsicId = llvm::Intrinsic::umul_with_overflow;
2905 break;
2906 case Builtin::BI__builtin_sadd_overflow:
2907 case Builtin::BI__builtin_saddl_overflow:
2908 case Builtin::BI__builtin_saddll_overflow:
2909 IntrinsicId = llvm::Intrinsic::sadd_with_overflow;
2910 break;
2911 case Builtin::BI__builtin_ssub_overflow:
2912 case Builtin::BI__builtin_ssubl_overflow:
2913 case Builtin::BI__builtin_ssubll_overflow:
2914 IntrinsicId = llvm::Intrinsic::ssub_with_overflow;
2915 break;
2916 case Builtin::BI__builtin_smul_overflow:
2917 case Builtin::BI__builtin_smull_overflow:
2918 case Builtin::BI__builtin_smulll_overflow:
2919 IntrinsicId = llvm::Intrinsic::smul_with_overflow;
2920 break;
2921 }
2922
2923
2924 llvm::Value *Carry;
2925 llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry);
2926 Builder.CreateStore(Sum, SumOutPtr);
2927
2928 return RValue::get(Carry);
2929 }
2930 case Builtin::BI__builtin_addressof:
2931 return RValue::get(EmitLValue(E->getArg(0)).getPointer());
2932 case Builtin::BI__builtin_operator_new:
2933 return EmitBuiltinNewDeleteCall(
2934 E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false);
2935 case Builtin::BI__builtin_operator_delete:
2936 return EmitBuiltinNewDeleteCall(
2937 E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true);
2938
2939 case Builtin::BI__noop:
2940 // __noop always evaluates to an integer literal zero.
2941 return RValue::get(ConstantInt::get(IntTy, 0));
2942 case Builtin::BI__builtin_call_with_static_chain: {
2943 const CallExpr *Call = cast<CallExpr>(E->getArg(0));
2944 const Expr *Chain = E->getArg(1);
2945 return EmitCall(Call->getCallee()->getType(),
2946 EmitCallee(Call->getCallee()), Call, ReturnValue,
2947 EmitScalarExpr(Chain));
2948 }
2949 case Builtin::BI_InterlockedExchange8:
2950 case Builtin::BI_InterlockedExchange16:
2951 case Builtin::BI_InterlockedExchange:
2952 case Builtin::BI_InterlockedExchangePointer:
2953 return RValue::get(
2954 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E));
2955 case Builtin::BI_InterlockedCompareExchangePointer: {
2956 llvm::Type *RTy;
2957 llvm::IntegerType *IntType =
2958 IntegerType::get(getLLVMContext(),
2959 getContext().getTypeSize(E->getType()));
2960 llvm::Type *IntPtrType = IntType->getPointerTo();
2961
2962 llvm::Value *Destination =
2963 Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType);
2964
2965 llvm::Value *Exchange = EmitScalarExpr(E->getArg(1));
2966 RTy = Exchange->getType();
2967 Exchange = Builder.CreatePtrToInt(Exchange, IntType);
2968
2969 llvm::Value *Comparand =
2970 Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType);
2971
2972 auto Result =
2973 Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange,
2974 AtomicOrdering::SequentiallyConsistent,
2975 AtomicOrdering::SequentiallyConsistent);
2976 Result->setVolatile(true);
2977
2978 return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result,
2979 0),
2980 RTy));
2981 }
2982 case Builtin::BI_InterlockedCompareExchange8:
2983 case Builtin::BI_InterlockedCompareExchange16:
2984 case Builtin::BI_InterlockedCompareExchange:
2985 case Builtin::BI_InterlockedCompareExchange64: {
2986 AtomicCmpXchgInst *CXI = Builder.CreateAtomicCmpXchg(
2987 EmitScalarExpr(E->getArg(0)),
2988 EmitScalarExpr(E->getArg(2)),
2989 EmitScalarExpr(E->getArg(1)),
2990 AtomicOrdering::SequentiallyConsistent,
2991 AtomicOrdering::SequentiallyConsistent);
2992 CXI->setVolatile(true);
2993 return RValue::get(Builder.CreateExtractValue(CXI, 0));
2994 }
2995 case Builtin::BI_InterlockedIncrement16:
2996 case Builtin::BI_InterlockedIncrement:
2997 return RValue::get(
2998 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E));
2999 case Builtin::BI_InterlockedDecrement16:
3000 case Builtin::BI_InterlockedDecrement:
3001 return RValue::get(
3002 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E));
3003 case Builtin::BI_InterlockedAnd8:
3004 case Builtin::BI_InterlockedAnd16:
3005 case Builtin::BI_InterlockedAnd:
3006 return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E));
3007 case Builtin::BI_InterlockedExchangeAdd8:
3008 case Builtin::BI_InterlockedExchangeAdd16:
3009 case Builtin::BI_InterlockedExchangeAdd:
3010 return RValue::get(
3011 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E));
3012 case Builtin::BI_InterlockedExchangeSub8:
3013 case Builtin::BI_InterlockedExchangeSub16:
3014 case Builtin::BI_InterlockedExchangeSub:
3015 return RValue::get(
3016 EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E));
3017 case Builtin::BI_InterlockedOr8:
3018 case Builtin::BI_InterlockedOr16:
3019 case Builtin::BI_InterlockedOr:
3020 return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E));
3021 case Builtin::BI_InterlockedXor8:
3022 case Builtin::BI_InterlockedXor16:
3023 case Builtin::BI_InterlockedXor:
3024 return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E));
3025
3026 case Builtin::BI_bittest64:
3027 case Builtin::BI_bittest:
3028 case Builtin::BI_bittestandcomplement64:
3029 case Builtin::BI_bittestandcomplement:
3030 case Builtin::BI_bittestandreset64:
3031 case Builtin::BI_bittestandreset:
3032 case Builtin::BI_bittestandset64:
3033 case Builtin::BI_bittestandset:
3034 case Builtin::BI_interlockedbittestandreset:
3035 case Builtin::BI_interlockedbittestandreset64:
3036 case Builtin::BI_interlockedbittestandset64:
3037 case Builtin::BI_interlockedbittestandset:
3038 case Builtin::BI_interlockedbittestandset_acq:
3039 case Builtin::BI_interlockedbittestandset_rel:
3040 case Builtin::BI_interlockedbittestandset_nf:
3041 case Builtin::BI_interlockedbittestandreset_acq:
3042 case Builtin::BI_interlockedbittestandreset_rel:
3043 case Builtin::BI_interlockedbittestandreset_nf:
3044 return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E));
3045
3046 case Builtin::BI__exception_code:
3047 case Builtin::BI_exception_code:
3048 return RValue::get(EmitSEHExceptionCode());
3049 case Builtin::BI__exception_info:
3050 case Builtin::BI_exception_info:
3051 return RValue::get(EmitSEHExceptionInfo());
3052 case Builtin::BI__abnormal_termination:
3053 case Builtin::BI_abnormal_termination:
3054 return RValue::get(EmitSEHAbnormalTermination());
3055 case Builtin::BI_setjmpex:
3056 if (getTarget().getTriple().isOSMSVCRT())
3057 return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E);
3058 break;
3059 case Builtin::BI_setjmp:
3060 if (getTarget().getTriple().isOSMSVCRT()) {
3061 if (getTarget().getTriple().getArch() == llvm::Triple::x86)
3062 return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E);
3063 else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64)
3064 return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E);
3065 return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E);
3066 }
3067 break;
3068
3069 case Builtin::BI__GetExceptionInfo: {
3070 if (llvm::GlobalVariable *GV =
3071 CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType()))
3072 return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy));
3073 break;
3074 }
3075
3076 case Builtin::BI__fastfail:
3077 return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E));
3078
3079 case Builtin::BI__builtin_coro_size: {
3080 auto & Context = getContext();
3081 auto SizeTy = Context.getSizeType();
3082 auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy));
3083 Value *F = CGM.getIntrinsic(Intrinsic::coro_size, T);
3084 return RValue::get(Builder.CreateCall(F));
3085 }
3086
3087 case Builtin::BI__builtin_coro_id:
3088 return EmitCoroutineIntrinsic(E, Intrinsic::coro_id);
3089 case Builtin::BI__builtin_coro_promise:
3090 return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise);
3091 case Builtin::BI__builtin_coro_resume:
3092 return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume);
3093 case Builtin::BI__builtin_coro_frame:
3094 return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame);
3095 case Builtin::BI__builtin_coro_noop:
3096 return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop);
3097 case Builtin::BI__builtin_coro_free:
3098 return EmitCoroutineIntrinsic(E, Intrinsic::coro_free);
3099 case Builtin::BI__builtin_coro_destroy:
3100 return EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy);
3101 case Builtin::BI__builtin_coro_done:
3102 return EmitCoroutineIntrinsic(E, Intrinsic::coro_done);
3103 case Builtin::BI__builtin_coro_alloc:
3104 return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc);
3105 case Builtin::BI__builtin_coro_begin:
3106 return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin);
3107 case Builtin::BI__builtin_coro_end:
3108 return EmitCoroutineIntrinsic(E, Intrinsic::coro_end);
3109 case Builtin::BI__builtin_coro_suspend:
3110 return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend);
3111 case Builtin::BI__builtin_coro_param:
3112 return EmitCoroutineIntrinsic(E, Intrinsic::coro_param);
3113
3114 // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions
3115 case Builtin::BIread_pipe:
3116 case Builtin::BIwrite_pipe: {
3117 Value *Arg0 = EmitScalarExpr(E->getArg(0)),
3118 *Arg1 = EmitScalarExpr(E->getArg(1));
3119 CGOpenCLRuntime OpenCLRT(CGM);
3120 Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
3121 Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
3122
3123 // Type of the generic packet parameter.
3124 unsigned GenericAS =
3125 getContext().getTargetAddressSpace(LangAS::opencl_generic);
3126 llvm::Type *I8PTy = llvm::PointerType::get(
3127 llvm::Type::getInt8Ty(getLLVMContext()), GenericAS);
3128
3129 // Testing which overloaded version we should generate the call for.
3130 if (2U == E->getNumArgs()) {
3131 const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2"
3132 : "__write_pipe_2";
3133 // Creating a generic function type to be able to call with any builtin or
3134 // user defined type.
3135 llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty};
3136 llvm::FunctionType *FTy = llvm::FunctionType::get(
3137 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3138 Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy);
3139 return RValue::get(
3140 Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3141 {Arg0, BCast, PacketSize, PacketAlign}));
3142 } else {
3143 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3144, __extension__ __PRETTY_FUNCTION__))
3144 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3144, __extension__ __PRETTY_FUNCTION__))
;
3145 const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4"
3146 : "__write_pipe_4";
3147
3148 llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy,
3149 Int32Ty, Int32Ty};
3150 Value *Arg2 = EmitScalarExpr(E->getArg(2)),
3151 *Arg3 = EmitScalarExpr(E->getArg(3));
3152 llvm::FunctionType *FTy = llvm::FunctionType::get(
3153 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3154 Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy);
3155 // We know the third argument is an integer type, but we may need to cast
3156 // it to i32.
3157 if (Arg2->getType() != Int32Ty)
3158 Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty);
3159 return RValue::get(Builder.CreateCall(
3160 CGM.CreateRuntimeFunction(FTy, Name),
3161 {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign}));
3162 }
3163 }
3164 // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write
3165 // functions
3166 case Builtin::BIreserve_read_pipe:
3167 case Builtin::BIreserve_write_pipe:
3168 case Builtin::BIwork_group_reserve_read_pipe:
3169 case Builtin::BIwork_group_reserve_write_pipe:
3170 case Builtin::BIsub_group_reserve_read_pipe:
3171 case Builtin::BIsub_group_reserve_write_pipe: {
3172 // Composing the mangled name for the function.
3173 const char *Name;
3174 if (BuiltinID == Builtin::BIreserve_read_pipe)
3175 Name = "__reserve_read_pipe";
3176 else if (BuiltinID == Builtin::BIreserve_write_pipe)
3177 Name = "__reserve_write_pipe";
3178 else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe)
3179 Name = "__work_group_reserve_read_pipe";
3180 else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe)
3181 Name = "__work_group_reserve_write_pipe";
3182 else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe)
3183 Name = "__sub_group_reserve_read_pipe";
3184 else
3185 Name = "__sub_group_reserve_write_pipe";
3186
3187 Value *Arg0 = EmitScalarExpr(E->getArg(0)),
3188 *Arg1 = EmitScalarExpr(E->getArg(1));
3189 llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy);
3190 CGOpenCLRuntime OpenCLRT(CGM);
3191 Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
3192 Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
3193
3194 // Building the generic function prototype.
3195 llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty};
3196 llvm::FunctionType *FTy = llvm::FunctionType::get(
3197 ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3198 // We know the second argument is an integer type, but we may need to cast
3199 // it to i32.
3200 if (Arg1->getType() != Int32Ty)
3201 Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty);
3202 return RValue::get(
3203 Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3204 {Arg0, Arg1, PacketSize, PacketAlign}));
3205 }
3206 // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write
3207 // functions
3208 case Builtin::BIcommit_read_pipe:
3209 case Builtin::BIcommit_write_pipe:
3210 case Builtin::BIwork_group_commit_read_pipe:
3211 case Builtin::BIwork_group_commit_write_pipe:
3212 case Builtin::BIsub_group_commit_read_pipe:
3213 case Builtin::BIsub_group_commit_write_pipe: {
3214 const char *Name;
3215 if (BuiltinID == Builtin::BIcommit_read_pipe)
3216 Name = "__commit_read_pipe";
3217 else if (BuiltinID == Builtin::BIcommit_write_pipe)
3218 Name = "__commit_write_pipe";
3219 else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe)
3220 Name = "__work_group_commit_read_pipe";
3221 else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe)
3222 Name = "__work_group_commit_write_pipe";
3223 else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe)
3224 Name = "__sub_group_commit_read_pipe";
3225 else
3226 Name = "__sub_group_commit_write_pipe";
3227
3228 Value *Arg0 = EmitScalarExpr(E->getArg(0)),
3229 *Arg1 = EmitScalarExpr(E->getArg(1));
3230 CGOpenCLRuntime OpenCLRT(CGM);
3231 Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
3232 Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
3233
3234 // Building the generic function prototype.
3235 llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty};
3236 llvm::FunctionType *FTy =
3237 llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()),
3238 llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3239
3240 return RValue::get(
3241 Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3242 {Arg0, Arg1, PacketSize, PacketAlign}));
3243 }
3244 // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions
3245 case Builtin::BIget_pipe_num_packets:
3246 case Builtin::BIget_pipe_max_packets: {
3247 const char *BaseName;
3248 const PipeType *PipeTy = E->getArg(0)->getType()->getAs<PipeType>();
3249 if (BuiltinID == Builtin::BIget_pipe_num_packets)
3250 BaseName = "__get_pipe_num_packets";
3251 else
3252 BaseName = "__get_pipe_max_packets";
3253 auto Name = std::string(BaseName) +
3254 std::string(PipeTy->isReadOnly() ? "_ro" : "_wo");
3255
3256 // Building the generic function prototype.
3257 Value *Arg0 = EmitScalarExpr(E->getArg(0));
3258 CGOpenCLRuntime OpenCLRT(CGM);
3259 Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
3260 Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
3261 llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty};
3262 llvm::FunctionType *FTy = llvm::FunctionType::get(
3263 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3264
3265 return RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3266 {Arg0, PacketSize, PacketAlign}));
3267 }
3268
3269 // OpenCL v2.0 s6.13.9 - Address space qualifier functions.
3270 case Builtin::BIto_global:
3271 case Builtin::BIto_local:
3272 case Builtin::BIto_private: {
3273 auto Arg0 = EmitScalarExpr(E->getArg(0));
3274 auto NewArgT = llvm::PointerType::get(Int8Ty,
3275 CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic));
3276 auto NewRetT = llvm::PointerType::get(Int8Ty,
3277 CGM.getContext().getTargetAddressSpace(
3278 E->getType()->getPointeeType().getAddressSpace()));
3279 auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false);
3280 llvm::Value *NewArg;
3281 if (Arg0->getType()->getPointerAddressSpace() !=
3282 NewArgT->getPointerAddressSpace())
3283 NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT);
3284 else
3285 NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT);
3286 auto NewName = std::string("__") + E->getDirectCallee()->getName().str();
3287 auto NewCall =
3288 Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg});
3289 return RValue::get(Builder.CreateBitOrPointerCast(NewCall,
3290 ConvertType(E->getType())));
3291 }
3292
3293 // OpenCL v2.0, s6.13.17 - Enqueue kernel function.
3294 // It contains four different overload formats specified in Table 6.13.17.1.
3295 case Builtin::BIenqueue_kernel: {
3296 StringRef Name; // Generated function call name
3297 unsigned NumArgs = E->getNumArgs();
3298
3299 llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy);
3300 llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
3301 getContext().getTargetAddressSpace(LangAS::opencl_generic));
3302
3303 llvm::Value *Queue = EmitScalarExpr(E->getArg(0));
3304 llvm::Value *Flags = EmitScalarExpr(E->getArg(1));
3305 LValue NDRangeL = EmitAggExprToLValue(E->getArg(2));
3306 llvm::Value *Range = NDRangeL.getAddress().getPointer();
3307 llvm::Type *RangeTy = NDRangeL.getAddress().getType();
3308
3309 if (NumArgs == 4) {
3310 // The most basic form of the call with parameters:
3311 // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void)
3312 Name = "__enqueue_kernel_basic";
3313 llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy,
3314 GenericVoidPtrTy};
3315 llvm::FunctionType *FTy = llvm::FunctionType::get(
3316 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3317
3318 auto Info =
3319 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3));
3320 llvm::Value *Kernel =
3321 Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3322 llvm::Value *Block =
3323 Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3324
3325 AttrBuilder B;
3326 B.addAttribute(Attribute::ByVal);
3327 llvm::AttributeList ByValAttrSet =
3328 llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B);
3329
3330 auto RTCall =
3331 Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet),
3332 {Queue, Flags, Range, Kernel, Block});
3333 RTCall->setAttributes(ByValAttrSet);
3334 return RValue::get(RTCall);
3335 }
3336 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3336, __extension__ __PRETTY_FUNCTION__))
;
3337
3338 // Create a temporary array to hold the sizes of local pointer arguments
3339 // for the block. \p First is the position of the first size argument.
3340 auto CreateArrayForSizeVar = [=](unsigned First) {
3341 auto *AT = llvm::ArrayType::get(SizeTy, NumArgs - First);
3342 auto *Arr = Builder.CreateAlloca(AT);
3343 llvm::Value *Ptr;
1
'Ptr' declared without an initial value
3344 // Each of the following arguments specifies the size of the corresponding
3345 // argument passed to the enqueued block.
3346 auto *Zero = llvm::ConstantInt::get(IntTy, 0);
3347 for (unsigned I = First; I < NumArgs; ++I) {
2
Assuming 'I' is >= 'NumArgs'
3
Loop condition is false. Execution continues on line 3357
3348 auto *Index = llvm::ConstantInt::get(IntTy, I - First);
3349 auto *GEP = Builder.CreateGEP(Arr, {Zero, Index});
3350 if (I == First)
3351 Ptr = GEP;
3352 auto *V =
3353 Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy);
3354 Builder.CreateAlignedStore(
3355 V, GEP, CGM.getDataLayout().getPrefTypeAlignment(SizeTy));
3356 }
3357 return Ptr;
4
Undefined or garbage value returned to caller
3358 };
3359
3360 // Could have events and/or varargs.
3361 if (E->getArg(3)->getType()->isBlockPointerType()) {
3362 // No events passed, but has variadic arguments.
3363 Name = "__enqueue_kernel_varargs";
3364 auto Info =
3365 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3));
3366 llvm::Value *Kernel =
3367 Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3368 auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3369 auto *PtrToSizeArray = CreateArrayForSizeVar(4);
3370
3371 // Create a vector of the arguments, as well as a constant value to
3372 // express to the runtime the number of variadic arguments.
3373 std::vector<llvm::Value *> Args = {
3374 Queue, Flags, Range,
3375 Kernel, Block, ConstantInt::get(IntTy, NumArgs - 4),
3376 PtrToSizeArray};
3377 std::vector<llvm::Type *> ArgTys = {
3378 QueueTy, IntTy, RangeTy,
3379 GenericVoidPtrTy, GenericVoidPtrTy, IntTy,
3380 PtrToSizeArray->getType()};
3381
3382 llvm::FunctionType *FTy = llvm::FunctionType::get(
3383 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3384 return RValue::get(
3385 Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3386 llvm::ArrayRef<llvm::Value *>(Args)));
3387 }
3388 // Any calls now have event arguments passed.
3389 if (NumArgs >= 7) {
3390 llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy);
3391 llvm::Type *EventPtrTy = EventTy->getPointerTo(
3392 CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic));
3393
3394 llvm::Value *NumEvents =
3395 Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty);
3396 llvm::Value *EventList =
3397 E->getArg(4)->getType()->isArrayType()
3398 ? EmitArrayToPointerDecay(E->getArg(4)).getPointer()
3399 : EmitScalarExpr(E->getArg(4));
3400 llvm::Value *ClkEvent = EmitScalarExpr(E->getArg(5));
3401 // Convert to generic address space.
3402 EventList = Builder.CreatePointerCast(EventList, EventPtrTy);
3403 ClkEvent = Builder.CreatePointerCast(ClkEvent, EventPtrTy);
3404 auto Info =
3405 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6));
3406 llvm::Value *Kernel =
3407 Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3408 llvm::Value *Block =
3409 Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3410
3411 std::vector<llvm::Type *> ArgTys = {
3412 QueueTy, Int32Ty, RangeTy, Int32Ty,
3413 EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy};
3414
3415 std::vector<llvm::Value *> Args = {Queue, Flags, Range, NumEvents,
3416 EventList, ClkEvent, Kernel, Block};
3417
3418 if (NumArgs == 7) {
3419 // Has events but no variadics.
3420 Name = "__enqueue_kernel_basic_events";
3421 llvm::FunctionType *FTy = llvm::FunctionType::get(
3422 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3423 return RValue::get(
3424 Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3425 llvm::ArrayRef<llvm::Value *>(Args)));
3426 }
3427 // Has event info and variadics
3428 // Pass the number of variadics to the runtime function too.
3429 Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7));
3430 ArgTys.push_back(Int32Ty);
3431 Name = "__enqueue_kernel_events_varargs";
3432
3433 auto *PtrToSizeArray = CreateArrayForSizeVar(7);
3434 Args.push_back(PtrToSizeArray);
3435 ArgTys.push_back(PtrToSizeArray->getType());
3436
3437 llvm::FunctionType *FTy = llvm::FunctionType::get(
3438 Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3439 return RValue::get(
3440 Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3441 llvm::ArrayRef<llvm::Value *>(Args)));
3442 }
3443 LLVM_FALLTHROUGH[[clang::fallthrough]];
3444 }
3445 // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block
3446 // parameter.
3447 case Builtin::BIget_kernel_work_group_size: {
3448 llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
3449 getContext().getTargetAddressSpace(LangAS::opencl_generic));
3450 auto Info =
3451 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0));
3452 Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3453 Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3454 return RValue::get(Builder.CreateCall(
3455 CGM.CreateRuntimeFunction(
3456 llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy},
3457 false),
3458 "__get_kernel_work_group_size_impl"),
3459 {Kernel, Arg}));
3460 }
3461 case Builtin::BIget_kernel_preferred_work_group_size_multiple: {
3462 llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
3463 getContext().getTargetAddressSpace(LangAS::opencl_generic));
3464 auto Info =
3465 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0));
3466 Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3467 Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3468 return RValue::get(Builder.CreateCall(
3469 CGM.CreateRuntimeFunction(
3470 llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy},
3471 false),
3472 "__get_kernel_preferred_work_group_size_multiple_impl"),
3473 {Kernel, Arg}));
3474 }
3475 case Builtin::BIget_kernel_max_sub_group_size_for_ndrange:
3476 case Builtin::BIget_kernel_sub_group_count_for_ndrange: {
3477 llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
3478 getContext().getTargetAddressSpace(LangAS::opencl_generic));
3479 LValue NDRangeL = EmitAggExprToLValue(E->getArg(0));
3480 llvm::Value *NDRange = NDRangeL.getAddress().getPointer();
3481 auto Info =
3482 CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1));
3483 Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3484 Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3485 const char *Name =
3486 BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange
3487 ? "__get_kernel_max_sub_group_size_for_ndrange_impl"
3488 : "__get_kernel_sub_group_count_for_ndrange_impl";
3489 return RValue::get(Builder.CreateCall(
3490 CGM.CreateRuntimeFunction(
3491 llvm::FunctionType::get(
3492 IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy},
3493 false),
3494 Name),
3495 {NDRange, Kernel, Block}));
3496 }
3497
3498 case Builtin::BI__builtin_store_half:
3499 case Builtin::BI__builtin_store_halff: {
3500 Value *Val = EmitScalarExpr(E->getArg(0));
3501 Address Address = EmitPointerWithAlignment(E->getArg(1));
3502 Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy());
3503 return RValue::get(Builder.CreateStore(HalfVal, Address));
3504 }
3505 case Builtin::BI__builtin_load_half: {
3506 Address Address = EmitPointerWithAlignment(E->getArg(0));
3507 Value *HalfVal = Builder.CreateLoad(Address);
3508 return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy()));
3509 }
3510 case Builtin::BI__builtin_load_halff: {
3511 Address Address = EmitPointerWithAlignment(E->getArg(0));
3512 Value *HalfVal = Builder.CreateLoad(Address);
3513 return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy()));
3514 }
3515 case Builtin::BIprintf:
3516 if (getTarget().getTriple().isNVPTX())
3517 return EmitNVPTXDevicePrintfCallExpr(E, ReturnValue);
3518 break;
3519 case Builtin::BI__builtin_canonicalize:
3520 case Builtin::BI__builtin_canonicalizef:
3521 case Builtin::BI__builtin_canonicalizel:
3522 return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize));
3523
3524 case Builtin::BI__builtin_thread_pointer: {
3525 if (!getContext().getTargetInfo().isTLSSupported())
3526 CGM.ErrorUnsupported(E, "__builtin_thread_pointer");
3527 // Fall through - it's already mapped to the intrinsic by GCCBuiltin.
3528 break;
3529 }
3530 case Builtin::BI__builtin_os_log_format:
3531 return emitBuiltinOSLogFormat(*E);
3532
3533 case Builtin::BI__builtin_os_log_format_buffer_size: {
3534 analyze_os_log::OSLogBufferLayout Layout;
3535 analyze_os_log::computeOSLogBufferLayout(CGM.getContext(), E, Layout);
3536 return RValue::get(ConstantInt::get(ConvertType(E->getType()),
3537 Layout.size().getQuantity()));
3538 }
3539
3540 case Builtin::BI__xray_customevent: {
3541 if (!ShouldXRayInstrumentFunction())
3542 return RValue::getIgnored();
3543
3544 if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has(
3545 XRayInstrKind::Custom))
3546 return RValue::getIgnored();
3547
3548 if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>())
3549 if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents())
3550 return RValue::getIgnored();
3551
3552 Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent);
3553 auto FTy = F->getFunctionType();
3554 auto Arg0 = E->getArg(0);
3555 auto Arg0Val = EmitScalarExpr(Arg0);
3556 auto Arg0Ty = Arg0->getType();
3557 auto PTy0 = FTy->getParamType(0);
3558 if (PTy0 != Arg0Val->getType()) {
3559 if (Arg0Ty->isArrayType())
3560 Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer();
3561 else
3562 Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0);
3563 }
3564 auto Arg1 = EmitScalarExpr(E->getArg(1));
3565 auto PTy1 = FTy->getParamType(1);
3566 if (PTy1 != Arg1->getType())
3567 Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1);
3568 return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1}));
3569 }
3570
3571 case Builtin::BI__xray_typedevent: {
3572 // TODO: There should be a way to always emit events even if the current
3573 // function is not instrumented. Losing events in a stream can cripple
3574 // a trace.
3575 if (!ShouldXRayInstrumentFunction())
3576 return RValue::getIgnored();
3577
3578 if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has(
3579 XRayInstrKind::Typed))
3580 return RValue::getIgnored();
3581
3582 if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>())
3583 if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents())
3584 return RValue::getIgnored();
3585
3586 Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent);
3587 auto FTy = F->getFunctionType();
3588 auto Arg0 = EmitScalarExpr(E->getArg(0));
3589 auto PTy0 = FTy->getParamType(0);
3590 if (PTy0 != Arg0->getType())
3591 Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0);
3592 auto Arg1 = E->getArg(1);
3593 auto Arg1Val = EmitScalarExpr(Arg1);
3594 auto Arg1Ty = Arg1->getType();
3595 auto PTy1 = FTy->getParamType(1);
3596 if (PTy1 != Arg1Val->getType()) {
3597 if (Arg1Ty->isArrayType())
3598 Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer();
3599 else
3600 Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1);
3601 }
3602 auto Arg2 = EmitScalarExpr(E->getArg(2));
3603 auto PTy2 = FTy->getParamType(2);
3604 if (PTy2 != Arg2->getType())
3605 Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2);
3606 return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2}));
3607 }
3608
3609 case Builtin::BI__builtin_ms_va_start:
3610 case Builtin::BI__builtin_ms_va_end:
3611 return RValue::get(
3612 EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(),
3613 BuiltinID == Builtin::BI__builtin_ms_va_start));
3614
3615 case Builtin::BI__builtin_ms_va_copy: {
3616 // Lower this manually. We can't reliably determine whether or not any
3617 // given va_copy() is for a Win64 va_list from the calling convention
3618 // alone, because it's legal to do this from a System V ABI function.
3619 // With opaque pointer types, we won't have enough information in LLVM
3620 // IR to determine this from the argument types, either. Best to do it
3621 // now, while we have enough information.
3622 Address DestAddr = EmitMSVAListRef(E->getArg(0));
3623 Address SrcAddr = EmitMSVAListRef(E->getArg(1));
3624
3625 llvm::Type *BPP = Int8PtrPtrTy;
3626
3627 DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"),
3628 DestAddr.getAlignment());
3629 SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"),
3630 SrcAddr.getAlignment());
3631
3632 Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val");
3633 return RValue::get(Builder.CreateStore(ArgPtr, DestAddr));
3634 }
3635 }
3636
3637 // If this is an alias for a lib function (e.g. __builtin_sin), emit
3638 // the call using the normal call path, but using the unmangled
3639 // version of the function name.
3640 if (getContext().BuiltinInfo.isLibFunction(BuiltinID))
3641 return emitLibraryCall(*this, FD, E,
3642 CGM.getBuiltinLibFunction(FD, BuiltinID));
3643
3644 // If this is a predefined lib function (e.g. malloc), emit the call
3645 // using exactly the normal call path.
3646 if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID))
3647 return emitLibraryCall(*this, FD, E,
3648 cast<llvm::Constant>(EmitScalarExpr(E->getCallee())));
3649
3650 // Check that a call to a target specific builtin has the correct target
3651 // features.
3652 // This is down here to avoid non-target specific builtins, however, if
3653 // generic builtins start to require generic target features then we
3654 // can move this up to the beginning of the function.
3655 checkTargetFeatures(E, FD);
3656
3657 if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID))
3658 LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth);
3659
3660 // See if we have a target specific intrinsic.
3661 const char *Name = getContext().BuiltinInfo.getName(BuiltinID);
3662 Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic;
3663 StringRef Prefix =
3664 llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch());
3665 if (!Prefix.empty()) {
3666 IntrinsicID = Intrinsic::getIntrinsicForGCCBuiltin(Prefix.data(), Name);
3667 // NOTE we don't need to perform a compatibility flag check here since the
3668 // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the
3669 // MS builtins via ALL_MS_LANGUAGES and are filtered earlier.
3670 if (IntrinsicID == Intrinsic::not_intrinsic)
3671 IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name);
3672 }
3673
3674 if (IntrinsicID != Intrinsic::not_intrinsic) {
3675 SmallVector<Value*, 16> Args;
3676
3677 // Find out if any arguments are required to be integer constant
3678 // expressions.
3679 unsigned ICEArguments = 0;
3680 ASTContext::GetBuiltinTypeError Error;
3681 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
3682 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3682, __extension__ __PRETTY_FUNCTION__))
;
3683
3684 Function *F = CGM.getIntrinsic(IntrinsicID);
3685 llvm::FunctionType *FTy = F->getFunctionType();
3686
3687 for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) {
3688 Value *ArgValue;
3689 // If this is a normal argument, just emit it as a scalar.
3690 if ((ICEArguments & (1 << i)) == 0) {
3691 ArgValue = EmitScalarExpr(E->getArg(i));
3692 } else {
3693 // If this is required to be a constant, constant fold it so that we
3694 // know that the generated intrinsic gets a ConstantInt.
3695 llvm::APSInt Result;
3696 bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result,getContext());
3697 assert(IsConst && "Constant arg isn't actually constant?")(static_cast <bool> (IsConst && "Constant arg isn't actually constant?"
) ? void (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\""
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3697, __extension__ __PRETTY_FUNCTION__))
;
3698 (void)IsConst;
3699 ArgValue = llvm::ConstantInt::get(getLLVMContext(), Result);
3700 }
3701
3702 // If the intrinsic arg type is different from the builtin arg type
3703 // we need to do a bit cast.
3704 llvm::Type *PTy = FTy->getParamType(i);
3705 if (PTy != ArgValue->getType()) {
3706 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3707, __extension__ __PRETTY_FUNCTION__))
3707 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3707, __extension__ __PRETTY_FUNCTION__))
;
3708 ArgValue = Builder.CreateBitCast(ArgValue, PTy);
3709 }
3710
3711 Args.push_back(ArgValue);
3712 }
3713
3714 Value *V = Builder.CreateCall(F, Args);
3715 QualType BuiltinRetType = E->getType();
3716
3717 llvm::Type *RetTy = VoidTy;
3718 if (!BuiltinRetType->isVoidType())
3719 RetTy = ConvertType(BuiltinRetType);
3720
3721 if (RetTy != V->getType()) {
3722 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3723, __extension__ __PRETTY_FUNCTION__))
3723 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3723, __extension__ __PRETTY_FUNCTION__))
;
3724 V = Builder.CreateBitCast(V, RetTy);
3725 }
3726
3727 return RValue::get(V);
3728 }
3729
3730 // See if we have a target specific builtin that needs to be lowered.
3731 if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E))
3732 return RValue::get(V);
3733
3734 ErrorUnsupported(E, "builtin function");
3735
3736 // Unknown builtin, for now just dump it out and return undef.
3737 return GetUndefRValue(E->getType());
3738}
3739
3740static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF,
3741 unsigned BuiltinID, const CallExpr *E,
3742 llvm::Triple::ArchType Arch) {
3743 switch (Arch) {
3744 case llvm::Triple::arm:
3745 case llvm::Triple::armeb:
3746 case llvm::Triple::thumb:
3747 case llvm::Triple::thumbeb:
3748 return CGF->EmitARMBuiltinExpr(BuiltinID, E, Arch);
3749 case llvm::Triple::aarch64:
3750 case llvm::Triple::aarch64_be:
3751 return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch);
3752 case llvm::Triple::x86:
3753 case llvm::Triple::x86_64:
3754 return CGF->EmitX86BuiltinExpr(BuiltinID, E);
3755 case llvm::Triple::ppc:
3756 case llvm::Triple::ppc64:
3757 case llvm::Triple::ppc64le:
3758 return CGF->EmitPPCBuiltinExpr(BuiltinID, E);
3759 case llvm::Triple::r600:
3760 case llvm::Triple::amdgcn:
3761 return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E);
3762 case llvm::Triple::systemz:
3763 return CGF->EmitSystemZBuiltinExpr(BuiltinID, E);
3764 case llvm::Triple::nvptx:
3765 case llvm::Triple::nvptx64:
3766 return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E);
3767 case llvm::Triple::wasm32:
3768 case llvm::Triple::wasm64:
3769 return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E);
3770 case llvm::Triple::hexagon:
3771 return CGF->EmitHexagonBuiltinExpr(BuiltinID, E);
3772 default:
3773 return nullptr;
3774 }
3775}
3776
3777Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID,
3778 const CallExpr *E) {
3779 if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) {
3780 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3780, __extension__ __PRETTY_FUNCTION__))
;
3781 return EmitTargetArchBuiltinExpr(
3782 this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E,
3783 getContext().getAuxTargetInfo()->getTriple().getArch());
3784 }
3785
3786 return EmitTargetArchBuiltinExpr(this, BuiltinID, E,
3787 getTarget().getTriple().getArch());
3788}
3789
3790static llvm::VectorType *GetNeonType(CodeGenFunction *CGF,
3791 NeonTypeFlags TypeFlags,
3792 bool HasLegalHalfType=true,
3793 bool V1Ty=false) {
3794 int IsQuad = TypeFlags.isQuad();
3795 switch (TypeFlags.getEltType()) {
3796 case NeonTypeFlags::Int8:
3797 case NeonTypeFlags::Poly8:
3798 return llvm::VectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad));
3799 case NeonTypeFlags::Int16:
3800 case NeonTypeFlags::Poly16:
3801 return llvm::VectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad));
3802 case NeonTypeFlags::Float16:
3803 if (HasLegalHalfType)
3804 return llvm::VectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad));
3805 else
3806 return llvm::VectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad));
3807 case NeonTypeFlags::Int32:
3808 return llvm::VectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad));
3809 case NeonTypeFlags::Int64:
3810 case NeonTypeFlags::Poly64:
3811 return llvm::VectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad));
3812 case NeonTypeFlags::Poly128:
3813 // FIXME: i128 and f128 doesn't get fully support in Clang and llvm.
3814 // There is a lot of i128 and f128 API missing.
3815 // so we use v16i8 to represent poly128 and get pattern matched.
3816 return llvm::VectorType::get(CGF->Int8Ty, 16);
3817 case NeonTypeFlags::Float32:
3818 return llvm::VectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad));
3819 case NeonTypeFlags::Float64:
3820 return llvm::VectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad));
3821 }
3822 llvm_unreachable("Unknown vector element type!")::llvm::llvm_unreachable_internal("Unknown vector element type!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3822)
;
3823}
3824
3825static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF,
3826 NeonTypeFlags IntTypeFlags) {
3827 int IsQuad = IntTypeFlags.isQuad();
3828 switch (IntTypeFlags.getEltType()) {
3829 case NeonTypeFlags::Int16:
3830 return llvm::VectorType::get(CGF->HalfTy, (4 << IsQuad));
3831 case NeonTypeFlags::Int32:
3832 return llvm::VectorType::get(CGF->FloatTy, (2 << IsQuad));
3833 case NeonTypeFlags::Int64:
3834 return llvm::VectorType::get(CGF->DoubleTy, (1 << IsQuad));
3835 default:
3836 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 3836)
;
3837 }
3838}
3839
3840Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) {
3841 unsigned nElts = V->getType()->getVectorNumElements();
3842 Value* SV = llvm::ConstantVector::getSplat(nElts, C);
3843 return Builder.CreateShuffleVector(V, V, SV, "lane");
3844}
3845
3846Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops,
3847 const char *name,
3848 unsigned shift, bool rightshift) {
3849 unsigned j = 0;
3850 for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
3851 ai != ae; ++ai, ++j)
3852 if (shift > 0 && shift == j)
3853 Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift);
3854 else
3855 Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name);
3856
3857 return Builder.CreateCall(F, Ops, name);
3858}
3859
3860Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty,
3861 bool neg) {
3862 int SV = cast<ConstantInt>(V)->getSExtValue();
3863 return ConstantInt::get(Ty, neg ? -SV : SV);
3864}
3865
3866// Right-shift a vector by a constant.
3867Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift,
3868 llvm::Type *Ty, bool usgn,
3869 const char *name) {
3870 llvm::VectorType *VTy = cast<llvm::VectorType>(Ty);
3871
3872 int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue();
3873 int EltSize = VTy->getScalarSizeInBits();
3874
3875 Vec = Builder.CreateBitCast(Vec, Ty);
3876
3877 // lshr/ashr are undefined when the shift amount is equal to the vector
3878 // element size.
3879 if (ShiftAmt == EltSize) {
3880 if (usgn) {
3881 // Right-shifting an unsigned value by its size yields 0.
3882 return llvm::ConstantAggregateZero::get(VTy);
3883 } else {
3884 // Right-shifting a signed value by its size is equivalent
3885 // to a shift of size-1.
3886 --ShiftAmt;
3887 Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt);
3888 }
3889 }
3890
3891 Shift = EmitNeonShiftVector(Shift, Ty, false);
3892 if (usgn)
3893 return Builder.CreateLShr(Vec, Shift, name);
3894 else
3895 return Builder.CreateAShr(Vec, Shift, name);
3896}
3897
3898enum {
3899 AddRetType = (1 << 0),
3900 Add1ArgType = (1 << 1),
3901 Add2ArgTypes = (1 << 2),
3902
3903 VectorizeRetType = (1 << 3),
3904 VectorizeArgTypes = (1 << 4),
3905
3906 InventFloatType = (1 << 5),
3907 UnsignedAlts = (1 << 6),
3908
3909 Use64BitVectors = (1 << 7),
3910 Use128BitVectors = (1 << 8),
3911
3912 Vectorize1ArgType = Add1ArgType | VectorizeArgTypes,
3913 VectorRet = AddRetType | VectorizeRetType,
3914 VectorRetGetArgs01 =
3915 AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes,
3916 FpCmpzModifiers =
3917 AddRetType | VectorizeRetType | Add1ArgType | InventFloatType
3918};
3919
3920namespace {
3921struct NeonIntrinsicInfo {
3922 const char *NameHint;
3923 unsigned BuiltinID;
3924 unsigned LLVMIntrinsic;
3925 unsigned AltLLVMIntrinsic;
3926 unsigned TypeModifier;
3927
3928 bool operator<(unsigned RHSBuiltinID) const {
3929 return BuiltinID < RHSBuiltinID;
3930 }
3931 bool operator<(const NeonIntrinsicInfo &TE) const {
3932 return BuiltinID < TE.BuiltinID;
3933 }
3934};
3935} // end anonymous namespace
3936
3937#define NEONMAP0(NameBase) \
3938 { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 }
3939
3940#define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \
3941 { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \
3942 Intrinsic::LLVMIntrinsic, 0, TypeModifier }
3943
3944#define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \
3945 { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \
3946 Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \
3947 TypeModifier }
3948
3949static const NeonIntrinsicInfo ARMSIMDIntrinsicMap [] = {
3950 NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts),
3951 NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts),
3952 NEONMAP1(vabs_v, arm_neon_vabs, 0),
3953 NEONMAP1(vabsq_v, arm_neon_vabs, 0),
3954 NEONMAP0(vaddhn_v),
3955 NEONMAP1(vaesdq_v, arm_neon_aesd, 0),
3956 NEONMAP1(vaeseq_v, arm_neon_aese, 0),
3957 NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0),
3958 NEONMAP1(vaesmcq_v, arm_neon_aesmc, 0),
3959 NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType),
3960 NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType),
3961 NEONMAP1(vcage_v, arm_neon_vacge, 0),
3962 NEONMAP1(vcageq_v, arm_neon_vacge, 0),
3963 NEONMAP1(vcagt_v, arm_neon_vacgt, 0),
3964 NEONMAP1(vcagtq_v, arm_neon_vacgt, 0),
3965 NEONMAP1(vcale_v, arm_neon_vacge, 0),
3966 NEONMAP1(vcaleq_v, arm_neon_vacge, 0),
3967 NEONMAP1(vcalt_v, arm_neon_vacgt, 0),
3968 NEONMAP1(vcaltq_v, arm_neon_vacgt, 0),
3969 NEONMAP0(vceqz_v),
3970 NEONMAP0(vceqzq_v),
3971 NEONMAP0(vcgez_v),
3972 NEONMAP0(vcgezq_v),
3973 NEONMAP0(vcgtz_v),
3974 NEONMAP0(vcgtzq_v),
3975 NEONMAP0(vclez_v),
3976 NEONMAP0(vclezq_v),
3977 NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType),
3978 NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType),
3979 NEONMAP0(vcltz_v),
3980 NEONMAP0(vcltzq_v),
3981 NEONMAP1(vclz_v, ctlz, Add1ArgType),
3982 NEONMAP1(vclzq_v, ctlz, Add1ArgType),
3983 NEONMAP1(vcnt_v, ctpop, Add1ArgType),
3984 NEONMAP1(vcntq_v, ctpop, Add1ArgType),
3985 NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0),
3986 NEONMAP0(vcvt_f16_v),
3987 NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0),
3988 NEONMAP0(vcvt_f32_v),
3989 NEONMAP2(vcvt_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
3990 NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
3991 NEONMAP1(vcvt_n_s16_v, arm_neon_vcvtfp2fxs, 0),
3992 NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0),
3993 NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0),
3994 NEONMAP1(vcvt_n_u16_v, arm_neon_vcvtfp2fxu, 0),
3995 NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0),
3996 NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0),
3997 NEONMAP0(vcvt_s16_v),
3998 NEONMAP0(vcvt_s32_v),
3999 NEONMAP0(vcvt_s64_v),
4000 NEONMAP0(vcvt_u16_v),
4001 NEONMAP0(vcvt_u32_v),
4002 NEONMAP0(vcvt_u64_v),
4003 NEONMAP1(vcvta_s16_v, arm_neon_vcvtas, 0),
4004 NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0),
4005 NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0),
4006 NEONMAP1(vcvta_u16_v, arm_neon_vcvtau, 0),
4007 NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0),
4008 NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0),
4009 NEONMAP1(vcvtaq_s16_v, arm_neon_vcvtas, 0),
4010 NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0),
4011 NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0),
4012 NEONMAP1(vcvtaq_u16_v, arm_neon_vcvtau, 0),
4013 NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0),
4014 NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0),
4015 NEONMAP1(vcvtm_s16_v, arm_neon_vcvtms, 0),
4016 NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0),
4017 NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0),
4018 NEONMAP1(vcvtm_u16_v, arm_neon_vcvtmu, 0),
4019 NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0),
4020 NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0),
4021 NEONMAP1(vcvtmq_s16_v, arm_neon_vcvtms, 0),
4022 NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0),
4023 NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0),
4024 NEONMAP1(vcvtmq_u16_v, arm_neon_vcvtmu, 0),
4025 NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0),
4026 NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0),
4027 NEONMAP1(vcvtn_s16_v, arm_neon_vcvtns, 0),
4028 NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0),
4029 NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0),
4030 NEONMAP1(vcvtn_u16_v, arm_neon_vcvtnu, 0),
4031 NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0),
4032 NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0),
4033 NEONMAP1(vcvtnq_s16_v, arm_neon_vcvtns, 0),
4034 NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0),
4035 NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0),
4036 NEONMAP1(vcvtnq_u16_v, arm_neon_vcvtnu, 0),
4037 NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0),
4038 NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0),
4039 NEONMAP1(vcvtp_s16_v, arm_neon_vcvtps, 0),
4040 NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0),
4041 NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0),
4042 NEONMAP1(vcvtp_u16_v, arm_neon_vcvtpu, 0),
4043 NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0),
4044 NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0),
4045 NEONMAP1(vcvtpq_s16_v, arm_neon_vcvtps, 0),
4046 NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0),
4047 NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0),
4048 NEONMAP1(vcvtpq_u16_v, arm_neon_vcvtpu, 0),
4049 NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0),
4050 NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0),
4051 NEONMAP0(vcvtq_f16_v),
4052 NEONMAP0(vcvtq_f32_v),
4053 NEONMAP2(vcvtq_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
4054 NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
4055 NEONMAP1(vcvtq_n_s16_v, arm_neon_vcvtfp2fxs, 0),
4056 NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0),
4057 NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0),
4058 NEONMAP1(vcvtq_n_u16_v, arm_neon_vcvtfp2fxu, 0),
4059 NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0),
4060 NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0),
4061 NEONMAP0(vcvtq_s16_v),
4062 NEONMAP0(vcvtq_s32_v),
4063 NEONMAP0(vcvtq_s64_v),
4064 NEONMAP0(vcvtq_u16_v),
4065 NEONMAP0(vcvtq_u32_v),
4066 NEONMAP0(vcvtq_u64_v),
4067 NEONMAP2(vdot_v, arm_neon_udot, arm_neon_sdot, 0),
4068 NEONMAP2(vdotq_v, arm_neon_udot, arm_neon_sdot, 0),
4069 NEONMAP0(vext_v),
4070 NEONMAP0(vextq_v),
4071 NEONMAP0(vfma_v),
4072 NEONMAP0(vfmaq_v),
4073 NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts),
4074 NEONMAP2(vhaddq_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts),
4075 NEONMAP2(vhsub_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts),
4076 NEONMAP2(vhsubq_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts),
4077 NEONMAP0(vld1_dup_v),
4078 NEONMAP1(vld1_v, arm_neon_vld1, 0),
4079 NEONMAP1(vld1_x2_v, arm_neon_vld1x2, 0),
4080 NEONMAP1(vld1_x3_v, arm_neon_vld1x3, 0),
4081 NEONMAP1(vld1_x4_v, arm_neon_vld1x4, 0),
4082 NEONMAP0(vld1q_dup_v),
4083 NEONMAP1(vld1q_v, arm_neon_vld1, 0),
4084 NEONMAP1(vld1q_x2_v, arm_neon_vld1x2, 0),
4085 NEONMAP1(vld1q_x3_v, arm_neon_vld1x3, 0),
4086 NEONMAP1(vld1q_x4_v, arm_neon_vld1x4, 0),
4087 NEONMAP1(vld2_dup_v, arm_neon_vld2dup, 0),
4088 NEONMAP1(vld2_lane_v, arm_neon_vld2lane, 0),
4089 NEONMAP1(vld2_v, arm_neon_vld2, 0),
4090 NEONMAP1(vld2q_dup_v, arm_neon_vld2dup, 0),
4091 NEONMAP1(vld2q_lane_v, arm_neon_vld2lane, 0),
4092 NEONMAP1(vld2q_v, arm_neon_vld2, 0),
4093 NEONMAP1(vld3_dup_v, arm_neon_vld3dup, 0),
4094 NEONMAP1(vld3_lane_v, arm_neon_vld3lane, 0),
4095 NEONMAP1(vld3_v, arm_neon_vld3, 0),
4096 NEONMAP1(vld3q_dup_v, arm_neon_vld3dup, 0),
4097 NEONMAP1(vld3q_lane_v, arm_neon_vld3lane, 0),
4098 NEONMAP1(vld3q_v, arm_neon_vld3, 0),
4099 NEONMAP1(vld4_dup_v, arm_neon_vld4dup, 0),
4100 NEONMAP1(vld4_lane_v, arm_neon_vld4lane, 0),
4101 NEONMAP1(vld4_v, arm_neon_vld4, 0),
4102 NEONMAP1(vld4q_dup_v, arm_neon_vld4dup, 0),
4103 NEONMAP1(vld4q_lane_v, arm_neon_vld4lane, 0),
4104 NEONMAP1(vld4q_v, arm_neon_vld4, 0),
4105 NEONMAP2(vmax_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts),
4106 NEONMAP1(vmaxnm_v, arm_neon_vmaxnm, Add1ArgType),
4107 NEONMAP1(vmaxnmq_v, arm_neon_vmaxnm, Add1ArgType),
4108 NEONMAP2(vmaxq_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts),
4109 NEONMAP2(vmin_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts),
4110 NEONMAP1(vminnm_v, arm_neon_vminnm, Add1ArgType),
4111 NEONMAP1(vminnmq_v, arm_neon_vminnm, Add1ArgType),
4112 NEONMAP2(vminq_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts),
4113 NEONMAP0(vmovl_v),
4114 NEONMAP0(vmovn_v),
4115 NEONMAP1(vmul_v, arm_neon_vmulp, Add1ArgType),
4116 NEONMAP0(vmull_v),
4117 NEONMAP1(vmulq_v, arm_neon_vmulp, Add1ArgType),
4118 NEONMAP2(vpadal_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts),
4119 NEONMAP2(vpadalq_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts),
4120 NEONMAP1(vpadd_v, arm_neon_vpadd, Add1ArgType),
4121 NEONMAP2(vpaddl_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts),
4122 NEONMAP2(vpaddlq_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts),
4123 NEONMAP1(vpaddq_v, arm_neon_vpadd, Add1ArgType),
4124 NEONMAP2(vpmax_v, arm_neon_vpmaxu, arm_neon_vpmaxs, Add1ArgType | UnsignedAlts),
4125 NEONMAP2(vpmin_v, arm_neon_vpminu, arm_neon_vpmins, Add1ArgType | UnsignedAlts),
4126 NEONMAP1(vqabs_v, arm_neon_vqabs, Add1ArgType),
4127 NEONMAP1(vqabsq_v, arm_neon_vqabs, Add1ArgType),
4128 NEONMAP2(vqadd_v, arm_neon_vqaddu, arm_neon_vqadds, Add1ArgType | UnsignedAlts),
4129 NEONMAP2(vqaddq_v, arm_neon_vqaddu, arm_neon_vqadds, Add1ArgType | UnsignedAlts),
4130 NEONMAP2(vqdmlal_v, arm_neon_vqdmull, arm_neon_vqadds, 0),
4131 NEONMAP2(vqdmlsl_v, arm_neon_vqdmull, arm_neon_vqsubs, 0),
4132 NEONMAP1(vqdmulh_v, arm_neon_vqdmulh, Add1ArgType),
4133 NEONMAP1(vqdmulhq_v, arm_neon_vqdmulh, Add1ArgType),
4134 NEONMAP1(vqdmull_v, arm_neon_vqdmull, Add1ArgType),
4135 NEONMAP2(vqmovn_v, arm_neon_vqmovnu, arm_neon_vqmovns, Add1ArgType | UnsignedAlts),
4136 NEONMAP1(vqmovun_v, arm_neon_vqmovnsu, Add1ArgType),
4137 NEONMAP1(vqneg_v, arm_neon_vqneg, Add1ArgType),
4138 NEONMAP1(vqnegq_v, arm_neon_vqneg, Add1ArgType),
4139 NEONMAP1(vqrdmulh_v, arm_neon_vqrdmulh, Add1ArgType),
4140 NEONMAP1(vqrdmulhq_v, arm_neon_vqrdmulh, Add1ArgType),
4141 NEONMAP2(vqrshl_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts),
4142 NEONMAP2(vqrshlq_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts),
4143 NEONMAP2(vqshl_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts),
4144 NEONMAP2(vqshl_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts),
4145 NEONMAP2(vqshlq_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts),
4146 NEONMAP2(vqshlq_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts),
4147 NEONMAP1(vqshlu_n_v, arm_neon_vqshiftsu, 0),
4148 NEONMAP1(vqshluq_n_v, arm_neon_vqshiftsu, 0),
4149 NEONMAP2(vqsub_v, arm_neon_vqsubu, arm_neon_vqsubs, Add1ArgType | UnsignedAlts),
4150 NEONMAP2(vqsubq_v, arm_neon_vqsubu, arm_neon_vqsubs, Add1ArgType | UnsignedAlts),
4151 NEONMAP1(vraddhn_v, arm_neon_vraddhn, Add1ArgType),
4152 NEONMAP2(vrecpe_v, arm_neon_vrecpe, arm_neon_vrecpe, 0),
4153 NEONMAP2(vrecpeq_v, arm_neon_vrecpe, arm_neon_vrecpe, 0),
4154 NEONMAP1(vrecps_v, arm_neon_vrecps, Add1ArgType),
4155 NEONMAP1(vrecpsq_v, arm_neon_vrecps, Add1ArgType),
4156 NEONMAP2(vrhadd_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts),
4157 NEONMAP2(vrhaddq_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts),
4158 NEONMAP1(vrnd_v, arm_neon_vrintz, Add1ArgType),
4159 NEONMAP1(vrnda_v, arm_neon_vrinta, Add1ArgType),
4160 NEONMAP1(vrndaq_v, arm_neon_vrinta, Add1ArgType),
4161 NEONMAP0(vrndi_v),
4162 NEONMAP0(vrndiq_v),
4163 NEONMAP1(vrndm_v, arm_neon_vrintm, Add1ArgType),
4164 NEONMAP1(vrndmq_v, arm_neon_vrintm, Add1ArgType),
4165 NEONMAP1(vrndn_v, arm_neon_vrintn, Add1ArgType),
4166 NEONMAP1(vrndnq_v, arm_neon_vrintn, Add1ArgType),
4167 NEONMAP1(vrndp_v, arm_neon_vrintp, Add1ArgType),
4168 NEONMAP1(vrndpq_v, arm_neon_vrintp, Add1ArgType),
4169 NEONMAP1(vrndq_v, arm_neon_vrintz, Add1ArgType),
4170 NEONMAP1(vrndx_v, arm_neon_vrintx, Add1ArgType),
4171 NEONMAP1(vrndxq_v, arm_neon_vrintx, Add1ArgType),
4172 NEONMAP2(vrshl_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts),
4173 NEONMAP2(vrshlq_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts),
4174 NEONMAP2(vrshr_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts),
4175 NEONMAP2(vrshrq_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts),
4176 NEONMAP2(vrsqrte_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0),
4177 NEONMAP2(vrsqrteq_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0),
4178 NEONMAP1(vrsqrts_v, arm_neon_vrsqrts, Add1ArgType),
4179 NEONMAP1(vrsqrtsq_v, arm_neon_vrsqrts, Add1ArgType),
4180 NEONMAP1(vrsubhn_v, arm_neon_vrsubhn, Add1ArgType),
4181 NEONMAP1(vsha1su0q_v, arm_neon_sha1su0, 0),
4182 NEONMAP1(vsha1su1q_v, arm_neon_sha1su1, 0),
4183 NEONMAP1(vsha256h2q_v, arm_neon_sha256h2, 0),
4184 NEONMAP1(vsha256hq_v, arm_neon_sha256h, 0),
4185 NEONMAP1(vsha256su0q_v, arm_neon_sha256su0, 0),
4186 NEONMAP1(vsha256su1q_v, arm_neon_sha256su1, 0),
4187 NEONMAP0(vshl_n_v),
4188 NEONMAP2(vshl_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts),
4189 NEONMAP0(vshll_n_v),
4190 NEONMAP0(vshlq_n_v),
4191 NEONMAP2(vshlq_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts),
4192 NEONMAP0(vshr_n_v),
4193 NEONMAP0(vshrn_n_v),
4194 NEONMAP0(vshrq_n_v),
4195 NEONMAP1(vst1_v, arm_neon_vst1, 0),
4196 NEONMAP1(vst1_x2_v, arm_neon_vst1x2, 0),
4197 NEONMAP1(vst1_x3_v, arm_neon_vst1x3, 0),
4198 NEONMAP1(vst1_x4_v, arm_neon_vst1x4, 0),
4199 NEONMAP1(vst1q_v, arm_neon_vst1, 0),
4200 NEONMAP1(vst1q_x2_v, arm_neon_vst1x2, 0),
4201 NEONMAP1(vst1q_x3_v, arm_neon_vst1x3, 0),
4202 NEONMAP1(vst1q_x4_v, arm_neon_vst1x4, 0),
4203 NEONMAP1(vst2_lane_v, arm_neon_vst2lane, 0),
4204 NEONMAP1(vst2_v, arm_neon_vst2, 0),
4205 NEONMAP1(vst2q_lane_v, arm_neon_vst2lane, 0),
4206 NEONMAP1(vst2q_v, arm_neon_vst2, 0),
4207 NEONMAP1(vst3_lane_v, arm_neon_vst3lane, 0),
4208 NEONMAP1(vst3_v, arm_neon_vst3, 0),
4209 NEONMAP1(vst3q_lane_v, arm_neon_vst3lane, 0),
4210 NEONMAP1(vst3q_v, arm_neon_vst3, 0),
4211 NEONMAP1(vst4_lane_v, arm_neon_vst4lane, 0),
4212 NEONMAP1(vst4_v, arm_neon_vst4, 0),
4213 NEONMAP1(vst4q_lane_v, arm_neon_vst4lane, 0),
4214 NEONMAP1(vst4q_v, arm_neon_vst4, 0),
4215 NEONMAP0(vsubhn_v),
4216 NEONMAP0(vtrn_v),
4217 NEONMAP0(vtrnq_v),
4218 NEONMAP0(vtst_v),
4219 NEONMAP0(vtstq_v),
4220 NEONMAP0(vuzp_v),
4221 NEONMAP0(vuzpq_v),
4222 NEONMAP0(vzip_v),
4223 NEONMAP0(vzipq_v)
4224};
4225
4226static const NeonIntrinsicInfo AArch64SIMDIntrinsicMap[] = {
4227 NEONMAP1(vabs_v, aarch64_neon_abs, 0),
4228 NEONMAP1(vabsq_v, aarch64_neon_abs, 0),
4229 NEONMAP0(vaddhn_v),
4230 NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0),
4231 NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0),
4232 NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0),
4233 NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0),
4234 NEONMAP1(vcage_v, aarch64_neon_facge, 0),
4235 NEONMAP1(vcageq_v, aarch64_neon_facge, 0),
4236 NEONMAP1(vcagt_v, aarch64_neon_facgt, 0),
4237 NEONMAP1(vcagtq_v, aarch64_neon_facgt, 0),
4238 NEONMAP1(vcale_v, aarch64_neon_facge, 0),
4239 NEONMAP1(vcaleq_v, aarch64_neon_facge, 0),
4240 NEONMAP1(vcalt_v, aarch64_neon_facgt, 0),
4241 NEONMAP1(vcaltq_v, aarch64_neon_facgt, 0),
4242 NEONMAP0(vceqz_v),
4243 NEONMAP0(vceqzq_v),
4244 NEONMAP0(vcgez_v),
4245 NEONMAP0(vcgezq_v),
4246 NEONMAP0(vcgtz_v),
4247 NEONMAP0(vcgtzq_v),
4248 NEONMAP0(vclez_v),
4249 NEONMAP0(vclezq_v),
4250 NEONMAP1(vcls_v, aarch64_neon_cls, Add1ArgType),
4251 NEONMAP1(vclsq_v, aarch64_neon_cls, Add1ArgType),
4252 NEONMAP0(vcltz_v),
4253 NEONMAP0(vcltzq_v),
4254 NEONMAP1(vclz_v, ctlz, Add1ArgType),
4255 NEONMAP1(vclzq_v, ctlz, Add1ArgType),
4256 NEONMAP1(vcnt_v, ctpop, Add1ArgType),
4257 NEONMAP1(vcntq_v, ctpop, Add1ArgType),
4258 NEONMAP1(vcvt_f16_f32, aarch64_neon_vcvtfp2hf, 0),
4259 NEONMAP0(vcvt_f16_v),
4260 NEONMAP1(vcvt_f32_f16, aarch64_neon_vcvthf2fp, 0),
4261 NEONMAP0(vcvt_f32_v),
4262 NEONMAP2(vcvt_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4263 NEONMAP2(vcvt_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4264 NEONMAP2(vcvt_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4265 NEONMAP1(vcvt_n_s16_v, aarch64_neon_vcvtfp2fxs, 0),
4266 NEONMAP1(vcvt_n_s32_v, aarch64_neon_vcvtfp2fxs, 0),
4267 NEONMAP1(vcvt_n_s64_v, aarch64_neon_vcvtfp2fxs, 0),
4268 NEONMAP1(vcvt_n_u16_v, aarch64_neon_vcvtfp2fxu, 0),
4269 NEONMAP1(vcvt_n_u32_v, aarch64_neon_vcvtfp2fxu, 0),
4270 NEONMAP1(vcvt_n_u64_v, aarch64_neon_vcvtfp2fxu, 0),
4271 NEONMAP0(vcvtq_f16_v),
4272 NEONMAP0(vcvtq_f32_v),
4273 NEONMAP2(vcvtq_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4274 NEONMAP2(vcvtq_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4275 NEONMAP2(vcvtq_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4276 NEONMAP1(vcvtq_n_s16_v, aarch64_neon_vcvtfp2fxs, 0),
4277 NEONMAP1(vcvtq_n_s32_v, aarch64_neon_vcvtfp2fxs, 0),
4278 NEONMAP1(vcvtq_n_s64_v, aarch64_neon_vcvtfp2fxs, 0),
4279 NEONMAP1(vcvtq_n_u16_v, aarch64_neon_vcvtfp2fxu, 0),
4280 NEONMAP1(vcvtq_n_u32_v, aarch64_neon_vcvtfp2fxu, 0),
4281 NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0),
4282 NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType),
4283 NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0),
4284 NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0),
4285 NEONMAP0(vext_v),
4286 NEONMAP0(vextq_v),
4287 NEONMAP0(vfma_v),
4288 NEONMAP0(vfmaq_v),
4289 NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts),
4290 NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts),
4291 NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts),
4292 NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts),
4293 NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0),
4294 NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0),
4295 NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0),
4296 NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0),
4297 NEONMAP1(vld1q_x3_v, aarch64_neon_ld1x3, 0),
4298 NEONMAP1(vld1q_x4_v, aarch64_neon_ld1x4, 0),
4299 NEONMAP0(vmovl_v),
4300 NEONMAP0(vmovn_v),
4301 NEONMAP1(vmul_v, aarch64_neon_pmul, Add1ArgType),
4302 NEONMAP1(vmulq_v, aarch64_neon_pmul, Add1ArgType),
4303 NEONMAP1(vpadd_v, aarch64_neon_addp, Add1ArgType),
4304 NEONMAP2(vpaddl_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts),
4305 NEONMAP2(vpaddlq_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts),
4306 NEONMAP1(vpaddq_v, aarch64_neon_addp, Add1ArgType),
4307 NEONMAP1(vqabs_v, aarch64_neon_sqabs, Add1ArgType),
4308 NEONMAP1(vqabsq_v, aarch64_neon_sqabs, Add1ArgType),
4309 NEONMAP2(vqadd_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts),
4310 NEONMAP2(vqaddq_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts),
4311 NEONMAP2(vqdmlal_v, aarch64_neon_sqdmull, aarch64_neon_sqadd, 0),
4312 NEONMAP2(vqdmlsl_v, aarch64_neon_sqdmull, aarch64_neon_sqsub, 0),
4313 NEONMAP1(vqdmulh_v, aarch64_neon_sqdmulh, Add1ArgType),
4314 NEONMAP1(vqdmulhq_v, aarch64_neon_sqdmulh, Add1ArgType),
4315 NEONMAP1(vqdmull_v, aarch64_neon_sqdmull, Add1ArgType),
4316 NEONMAP2(vqmovn_v, aarch64_neon_uqxtn, aarch64_neon_sqxtn, Add1ArgType | UnsignedAlts),
4317 NEONMAP1(vqmovun_v, aarch64_neon_sqxtun, Add1ArgType),
4318 NEONMAP1(vqneg_v, aarch64_neon_sqneg, Add1ArgType),
4319 NEONMAP1(vqnegq_v, aarch64_neon_sqneg, Add1ArgType),
4320 NEONMAP1(vqrdmulh_v, aarch64_neon_sqrdmulh, Add1ArgType),
4321 NEONMAP1(vqrdmulhq_v, aarch64_neon_sqrdmulh, Add1ArgType),
4322 NEONMAP2(vqrshl_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts),
4323 NEONMAP2(vqrshlq_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts),
4324 NEONMAP2(vqshl_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl, UnsignedAlts),
4325 NEONMAP2(vqshl_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts),
4326 NEONMAP2(vqshlq_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl,UnsignedAlts),
4327 NEONMAP2(vqshlq_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts),
4328 NEONMAP1(vqshlu_n_v, aarch64_neon_sqshlu, 0),
4329 NEONMAP1(vqshluq_n_v, aarch64_neon_sqshlu, 0),
4330 NEONMAP2(vqsub_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts),
4331 NEONMAP2(vqsubq_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts),
4332 NEONMAP1(vraddhn_v, aarch64_neon_raddhn, Add1ArgType),
4333 NEONMAP2(vrecpe_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0),
4334 NEONMAP2(vrecpeq_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0),
4335 NEONMAP1(vrecps_v, aarch64_neon_frecps, Add1ArgType),
4336 NEONMAP1(vrecpsq_v, aarch64_neon_frecps, Add1ArgType),
4337 NEONMAP2(vrhadd_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts),
4338 NEONMAP2(vrhaddq_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts),
4339 NEONMAP0(vrndi_v),
4340 NEONMAP0(vrndiq_v),
4341 NEONMAP2(vrshl_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts),
4342 NEONMAP2(vrshlq_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts),
4343 NEONMAP2(vrshr_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts),
4344 NEONMAP2(vrshrq_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts),
4345 NEONMAP2(vrsqrte_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0),
4346 NEONMAP2(vrsqrteq_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0),
4347 NEONMAP1(vrsqrts_v, aarch64_neon_frsqrts, Add1ArgType),
4348 NEONMAP1(vrsqrtsq_v, aarch64_neon_frsqrts, Add1ArgType),
4349 NEONMAP1(vrsubhn_v, aarch64_neon_rsubhn, Add1ArgType),
4350 NEONMAP1(vsha1su0q_v, aarch64_crypto_sha1su0, 0),
4351 NEONMAP1(vsha1su1q_v, aarch64_crypto_sha1su1, 0),
4352 NEONMAP1(vsha256h2q_v, aarch64_crypto_sha256h2, 0),
4353 NEONMAP1(vsha256hq_v, aarch64_crypto_sha256h, 0),
4354 NEONMAP1(vsha256su0q_v, aarch64_crypto_sha256su0, 0),
4355 NEONMAP1(vsha256su1q_v, aarch64_crypto_sha256su1, 0),
4356 NEONMAP0(vshl_n_v),
4357 NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts),
4358 NEONMAP0(vshll_n_v),
4359 NEONMAP0(vshlq_n_v),
4360 NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts),
4361 NEONMAP0(vshr_n_v),
4362 NEONMAP0(vshrn_n_v),
4363 NEONMAP0(vshrq_n_v),
4364 NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0),
4365 NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0),
4366 NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0),
4367 NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0),
4368 NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0),
4369 NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0),
4370 NEONMAP0(vsubhn_v),
4371 NEONMAP0(vtst_v),
4372 NEONMAP0(vtstq_v),
4373};
4374
4375static const NeonIntrinsicInfo AArch64SISDIntrinsicMap[] = {
4376 NEONMAP1(vabdd_f64, aarch64_sisd_fabd, Add1ArgType),
4377 NEONMAP1(vabds_f32, aarch64_sisd_fabd, Add1ArgType),
4378 NEONMAP1(vabsd_s64, aarch64_neon_abs, Add1ArgType),
4379 NEONMAP1(vaddlv_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType),
4380 NEONMAP1(vaddlv_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType),
4381 NEONMAP1(vaddlvq_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType),
4382 NEONMAP1(vaddlvq_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType),
4383 NEONMAP1(vaddv_f32, aarch64_neon_faddv, AddRetType | Add1ArgType),
4384 NEONMAP1(vaddv_s32, aarch64_neon_saddv, AddRetType | Add1ArgType),
4385 NEONMAP1(vaddv_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4386 NEONMAP1(vaddvq_f32, aarch64_neon_faddv, AddRetType | Add1ArgType),
4387 NEONMAP1(vaddvq_f64, aarch64_neon_faddv, AddRetType | Add1ArgType),
4388 NEONMAP1(vaddvq_s32, aarch64_neon_saddv, AddRetType | Add1ArgType),
4389 NEONMAP1(vaddvq_s64, aarch64_neon_saddv, AddRetType | Add1ArgType),
4390 NEONMAP1(vaddvq_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4391 NEONMAP1(vaddvq_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4392 NEONMAP1(vcaged_f64, aarch64_neon_facge, AddRetType | Add1ArgType),
4393 NEONMAP1(vcages_f32, aarch64_neon_facge, AddRetType | Add1ArgType),
4394 NEONMAP1(vcagtd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType),
4395 NEONMAP1(vcagts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType),
4396 NEONMAP1(vcaled_f64, aarch64_neon_facge, AddRetType | Add1ArgType),
4397 NEONMAP1(vcales_f32, aarch64_neon_facge, AddRetType | Add1ArgType),
4398 NEONMAP1(vcaltd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType),
4399 NEONMAP1(vcalts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType),
4400 NEONMAP1(vcvtad_s64_f64, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
4401 NEONMAP1(vcvtad_u64_f64, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
4402 NEONMAP1(vcvtas_s32_f32, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
4403 NEONMAP1(vcvtas_u32_f32, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
4404 NEONMAP1(vcvtd_n_f64_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
4405 NEONMAP1(vcvtd_n_f64_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
4406 NEONMAP1(vcvtd_n_s64_f64, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
4407 NEONMAP1(vcvtd_n_u64_f64, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
4408 NEONMAP1(vcvtmd_s64_f64, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
4409 NEONMAP1(vcvtmd_u64_f64, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
4410 NEONMAP1(vcvtms_s32_f32, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
4411 NEONMAP1(vcvtms_u32_f32, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
4412 NEONMAP1(vcvtnd_s64_f64, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
4413 NEONMAP1(vcvtnd_u64_f64, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
4414 NEONMAP1(vcvtns_s32_f32, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
4415 NEONMAP1(vcvtns_u32_f32, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
4416 NEONMAP1(vcvtpd_s64_f64, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
4417 NEONMAP1(vcvtpd_u64_f64, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
4418 NEONMAP1(vcvtps_s32_f32, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
4419 NEONMAP1(vcvtps_u32_f32, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
4420 NEONMAP1(vcvts_n_f32_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
4421 NEONMAP1(vcvts_n_f32_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
4422 NEONMAP1(vcvts_n_s32_f32, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
4423 NEONMAP1(vcvts_n_u32_f32, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
4424 NEONMAP1(vcvtxd_f32_f64, aarch64_sisd_fcvtxn, 0),
4425 NEONMAP1(vmaxnmv_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4426 NEONMAP1(vmaxnmvq_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4427 NEONMAP1(vmaxnmvq_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4428 NEONMAP1(vmaxv_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4429 NEONMAP1(vmaxv_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType),
4430 NEONMAP1(vmaxv_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType),
4431 NEONMAP1(vmaxvq_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4432 NEONMAP1(vmaxvq_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4433 NEONMAP1(vmaxvq_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType),
4434 NEONMAP1(vmaxvq_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType),
4435 NEONMAP1(vminnmv_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4436 NEONMAP1(vminnmvq_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4437 NEONMAP1(vminnmvq_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4438 NEONMAP1(vminv_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
4439 NEONMAP1(vminv_s32, aarch64_neon_sminv, AddRetType | Add1ArgType),
4440 NEONMAP1(vminv_u32, aarch64_neon_uminv, AddRetType | Add1ArgType),
4441 NEONMAP1(vminvq_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
4442 NEONMAP1(vminvq_f64, aarch64_neon_fminv, AddRetType | Add1ArgType),
4443 NEONMAP1(vminvq_s32, aarch64_neon_sminv, AddRetType | Add1ArgType),
4444 NEONMAP1(vminvq_u32, aarch64_neon_uminv, AddRetType | Add1ArgType),
4445 NEONMAP1(vmull_p64, aarch64_neon_pmull64, 0),
4446 NEONMAP1(vmulxd_f64, aarch64_neon_fmulx, Add1ArgType),
4447 NEONMAP1(vmulxs_f32, aarch64_neon_fmulx, Add1ArgType),
4448 NEONMAP1(vpaddd_s64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4449 NEONMAP1(vpaddd_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4450 NEONMAP1(vpmaxnmqd_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4451 NEONMAP1(vpmaxnms_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4452 NEONMAP1(vpmaxqd_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4453 NEONMAP1(vpmaxs_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4454 NEONMAP1(vpminnmqd_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4455 NEONMAP1(vpminnms_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4456 NEONMAP1(vpminqd_f64, aarch64_neon_fminv, AddRetType | Add1ArgType),
4457 NEONMAP1(vpmins_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
4458 NEONMAP1(vqabsb_s8, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors),
4459 NEONMAP1(vqabsd_s64, aarch64_neon_sqabs, Add1ArgType),
4460 NEONMAP1(vqabsh_s16, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors),
4461 NEONMAP1(vqabss_s32, aarch64_neon_sqabs, Add1ArgType),
4462 NEONMAP1(vqaddb_s8, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors),
4463 NEONMAP1(vqaddb_u8, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors),
4464 NEONMAP1(vqaddd_s64, aarch64_neon_sqadd, Add1ArgType),
4465 NEONMAP1(vqaddd_u64, aarch64_neon_uqadd, Add1ArgType),
4466 NEONMAP1(vqaddh_s16, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors),
4467 NEONMAP1(vqaddh_u16, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors),
4468 NEONMAP1(vqadds_s32, aarch64_neon_sqadd, Add1ArgType),
4469 NEONMAP1(vqadds_u32, aarch64_neon_uqadd, Add1ArgType),
4470 NEONMAP1(vqdmulhh_s16, aarch64_neon_sqdmulh, Vectorize1ArgType | Use64BitVectors),
4471 NEONMAP1(vqdmulhs_s32, aarch64_neon_sqdmulh, Add1ArgType),
4472 NEONMAP1(vqdmullh_s16, aarch64_neon_sqdmull, VectorRet | Use128BitVectors),
4473 NEONMAP1(vqdmulls_s32, aarch64_neon_sqdmulls_scalar, 0),
4474 NEONMAP1(vqmovnd_s64, aarch64_neon_scalar_sqxtn, AddRetType | Add1ArgType),
4475 NEONMAP1(vqmovnd_u64, aarch64_neon_scalar_uqxtn, AddRetType | Add1ArgType),
4476 NEONMAP1(vqmovnh_s16, aarch64_neon_sqxtn, VectorRet | Use64BitVectors),
4477 NEONMAP1(vqmovnh_u16, aarch64_neon_uqxtn, VectorRet | Use64BitVectors),
4478 NEONMAP1(vqmovns_s32, aarch64_neon_sqxtn, VectorRet | Use64BitVectors),
4479 NEONMAP1(vqmovns_u32, aarch64_neon_uqxtn, VectorRet | Use64BitVectors),
4480 NEONMAP1(vqmovund_s64, aarch64_neon_scalar_sqxtun, AddRetType | Add1ArgType),
4481 NEONMAP1(vqmovunh_s16, aarch64_neon_sqxtun, VectorRet | Use64BitVectors),
4482 NEONMAP1(vqmovuns_s32, aarch64_neon_sqxtun, VectorRet | Use64BitVectors),
4483 NEONMAP1(vqnegb_s8, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors),
4484 NEONMAP1(vqnegd_s64, aarch64_neon_sqneg, Add1ArgType),
4485 NEONMAP1(vqnegh_s16, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors),
4486 NEONMAP1(vqnegs_s32, aarch64_neon_sqneg, Add1ArgType),
4487 NEONMAP1(vqrdmulhh_s16, aarch64_neon_sqrdmulh, Vectorize1ArgType | Use64BitVectors),
4488 NEONMAP1(vqrdmulhs_s32, aarch64_neon_sqrdmulh, Add1ArgType),
4489 NEONMAP1(vqrshlb_s8, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors),
4490 NEONMAP1(vqrshlb_u8, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors),
4491 NEONMAP1(vqrshld_s64, aarch64_neon_sqrshl, Add1ArgType),
4492 NEONMAP1(vqrshld_u64, aarch64_neon_uqrshl, Add1ArgType),
4493 NEONMAP1(vqrshlh_s16, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors),
4494 NEONMAP1(vqrshlh_u16, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors),
4495 NEONMAP1(vqrshls_s32, aarch64_neon_sqrshl, Add1ArgType),
4496 NEONMAP1(vqrshls_u32, aarch64_neon_uqrshl, Add1ArgType),
4497 NEONMAP1(vqrshrnd_n_s64, aarch64_neon_sqrshrn, AddRetType),
4498 NEONMAP1(vqrshrnd_n_u64, aarch64_neon_uqrshrn, AddRetType),
4499 NEONMAP1(vqrshrnh_n_s16, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors),
4500 NEONMAP1(vqrshrnh_n_u16, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors),
4501 NEONMAP1(vqrshrns_n_s32, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors),
4502 NEONMAP1(vqrshrns_n_u32, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors),
4503 NEONMAP1(vqrshrund_n_s64, aarch64_neon_sqrshrun, AddRetType),
4504 NEONMAP1(vqrshrunh_n_s16, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors),
4505 NEONMAP1(vqrshruns_n_s32, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors),
4506 NEONMAP1(vqshlb_n_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
4507 NEONMAP1(vqshlb_n_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
4508 NEONMAP1(vqshlb_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
4509 NEONMAP1(vqshlb_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
4510 NEONMAP1(vqshld_s64, aarch64_neon_sqshl, Add1ArgType),
4511 NEONMAP1(vqshld_u64, aarch64_neon_uqshl, Add1ArgType),
4512 NEONMAP1(vqshlh_n_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
4513 NEONMAP1(vqshlh_n_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
4514 NEONMAP1(vqshlh_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
4515 NEONMAP1(vqshlh_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
4516 NEONMAP1(vqshls_n_s32, aarch64_neon_sqshl, Add1ArgType),
4517 NEONMAP1(vqshls_n_u32, aarch64_neon_uqshl, Add1ArgType),
4518 NEONMAP1(vqshls_s32, aarch64_neon_sqshl, Add1ArgType),
4519 NEONMAP1(vqshls_u32, aarch64_neon_uqshl, Add1ArgType),
4520 NEONMAP1(vqshlub_n_s8, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors),
4521 NEONMAP1(vqshluh_n_s16, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors),
4522 NEONMAP1(vqshlus_n_s32, aarch64_neon_sqshlu, Add1ArgType),
4523 NEONMAP1(vqshrnd_n_s64, aarch64_neon_sqshrn, AddRetType),
4524 NEONMAP1(vqshrnd_n_u64, aarch64_neon_uqshrn, AddRetType),
4525 NEONMAP1(vqshrnh_n_s16, aarch64_neon_sqshrn, VectorRet | Use64BitVectors),
4526 NEONMAP1(vqshrnh_n_u16, aarch64_neon_uqshrn, VectorRet | Use64BitVectors),
4527 NEONMAP1(vqshrns_n_s32, aarch64_neon_sqshrn, VectorRet | Use64BitVectors),
4528 NEONMAP1(vqshrns_n_u32, aarch64_neon_uqshrn, VectorRet | Use64BitVectors),
4529 NEONMAP1(vqshrund_n_s64, aarch64_neon_sqshrun, AddRetType),
4530 NEONMAP1(vqshrunh_n_s16, aarch64_neon_sqshrun, VectorRet | Use64BitVectors),
4531 NEONMAP1(vqshruns_n_s32, aarch64_neon_sqshrun, VectorRet | Use64BitVectors),
4532 NEONMAP1(vqsubb_s8, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors),
4533 NEONMAP1(vqsubb_u8, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors),
4534 NEONMAP1(vqsubd_s64, aarch64_neon_sqsub, Add1ArgType),
4535 NEONMAP1(vqsubd_u64, aarch64_neon_uqsub, Add1ArgType),
4536 NEONMAP1(vqsubh_s16, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors),
4537 NEONMAP1(vqsubh_u16, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors),
4538 NEONMAP1(vqsubs_s32, aarch64_neon_sqsub, Add1ArgType),
4539 NEONMAP1(vqsubs_u32, aarch64_neon_uqsub, Add1ArgType),
4540 NEONMAP1(vrecped_f64, aarch64_neon_frecpe, Add1ArgType),
4541 NEONMAP1(vrecpes_f32, aarch64_neon_frecpe, Add1ArgType),
4542 NEONMAP1(vrecpxd_f64, aarch64_neon_frecpx, Add1ArgType),
4543 NEONMAP1(vrecpxs_f32, aarch64_neon_frecpx, Add1ArgType),
4544 NEONMAP1(vrshld_s64, aarch64_neon_srshl, Add1ArgType),
4545 NEONMAP1(vrshld_u64, aarch64_neon_urshl, Add1ArgType),
4546 NEONMAP1(vrsqrted_f64, aarch64_neon_frsqrte, Add1ArgType),
4547 NEONMAP1(vrsqrtes_f32, aarch64_neon_frsqrte, Add1ArgType),
4548 NEONMAP1(vrsqrtsd_f64, aarch64_neon_frsqrts, Add1ArgType),
4549 NEONMAP1(vrsqrtss_f32, aarch64_neon_frsqrts, Add1ArgType),
4550 NEONMAP1(vsha1cq_u32, aarch64_crypto_sha1c, 0),
4551 NEONMAP1(vsha1h_u32, aarch64_crypto_sha1h, 0),
4552 NEONMAP1(vsha1mq_u32, aarch64_crypto_sha1m, 0),
4553 NEONMAP1(vsha1pq_u32, aarch64_crypto_sha1p, 0),
4554 NEONMAP1(vshld_s64, aarch64_neon_sshl, Add1ArgType),
4555 NEONMAP1(vshld_u64, aarch64_neon_ushl, Add1ArgType),
4556 NEONMAP1(vslid_n_s64, aarch64_neon_vsli, Vectorize1ArgType),
4557 NEONMAP1(vslid_n_u64, aarch64_neon_vsli, Vectorize1ArgType),
4558 NEONMAP1(vsqaddb_u8, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors),
4559 NEONMAP1(vsqaddd_u64, aarch64_neon_usqadd, Add1ArgType),
4560 NEONMAP1(vsqaddh_u16, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors),
4561 NEONMAP1(vsqadds_u32, aarch64_neon_usqadd, Add1ArgType),
4562 NEONMAP1(vsrid_n_s64, aarch64_neon_vsri, Vectorize1ArgType),
4563 NEONMAP1(vsrid_n_u64, aarch64_neon_vsri, Vectorize1ArgType),
4564 NEONMAP1(vuqaddb_s8, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors),
4565 NEONMAP1(vuqaddd_s64, aarch64_neon_suqadd, Add1ArgType),
4566 NEONMAP1(vuqaddh_s16, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors),
4567 NEONMAP1(vuqadds_s32, aarch64_neon_suqadd, Add1ArgType),
4568 // FP16 scalar intrinisics go here.
4569 NEONMAP1(vabdh_f16, aarch64_sisd_fabd, Add1ArgType),
4570 NEONMAP1(vcvtah_s32_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
4571 NEONMAP1(vcvtah_s64_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
4572 NEONMAP1(vcvtah_u32_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
4573 NEONMAP1(vcvtah_u64_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
4574 NEONMAP1(vcvth_n_f16_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
4575 NEONMAP1(vcvth_n_f16_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
4576 NEONMAP1(vcvth_n_f16_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
4577 NEONMAP1(vcvth_n_f16_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
4578 NEONMAP1(vcvth_n_s32_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
4579 NEONMAP1(vcvth_n_s64_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
4580 NEONMAP1(vcvth_n_u32_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
4581 NEONMAP1(vcvth_n_u64_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
4582 NEONMAP1(vcvtmh_s32_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
4583 NEONMAP1(vcvtmh_s64_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
4584 NEONMAP1(vcvtmh_u32_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
4585 NEONMAP1(vcvtmh_u64_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
4586 NEONMAP1(vcvtnh_s32_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
4587 NEONMAP1(vcvtnh_s64_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
4588 NEONMAP1(vcvtnh_u32_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
4589 NEONMAP1(vcvtnh_u64_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
4590 NEONMAP1(vcvtph_s32_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
4591 NEONMAP1(vcvtph_s64_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
4592 NEONMAP1(vcvtph_u32_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
4593 NEONMAP1(vcvtph_u64_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
4594 NEONMAP1(vmulxh_f16, aarch64_neon_fmulx, Add1ArgType),
4595 NEONMAP1(vrecpeh_f16, aarch64_neon_frecpe, Add1ArgType),
4596 NEONMAP1(vrecpxh_f16, aarch64_neon_frecpx, Add1ArgType),
4597 NEONMAP1(vrsqrteh_f16, aarch64_neon_frsqrte, Add1ArgType),
4598 NEONMAP1(vrsqrtsh_f16, aarch64_neon_frsqrts, Add1ArgType),
4599};
4600
4601#undef NEONMAP0
4602#undef NEONMAP1
4603#undef NEONMAP2
4604
4605static bool NEONSIMDIntrinsicsProvenSorted = false;
4606
4607static bool AArch64SIMDIntrinsicsProvenSorted = false;
4608static bool AArch64SISDIntrinsicsProvenSorted = false;
4609
4610
4611static const NeonIntrinsicInfo *
4612findNeonIntrinsicInMap(ArrayRef<NeonIntrinsicInfo> IntrinsicMap,
4613 unsigned BuiltinID, bool &MapProvenSorted) {
4614
4615#ifndef NDEBUG
4616 if (!MapProvenSorted) {
4617 assert(std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap)))(static_cast <bool> (std::is_sorted(std::begin(IntrinsicMap
), std::end(IntrinsicMap))) ? void (0) : __assert_fail ("std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap))"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 4617, __extension__ __PRETTY_FUNCTION__))
;
4618 MapProvenSorted = true;
4619 }
4620#endif
4621
4622 const NeonIntrinsicInfo *Builtin =
4623 std::lower_bound(IntrinsicMap.begin(), IntrinsicMap.end(), BuiltinID);
4624
4625 if (Builtin != IntrinsicMap.end() && Builtin->BuiltinID == BuiltinID)
4626 return Builtin;
4627
4628 return nullptr;
4629}
4630
4631Function *CodeGenFunction::LookupNeonLLVMIntrinsic(unsigned IntrinsicID,
4632 unsigned Modifier,
4633 llvm::Type *ArgType,
4634 const CallExpr *E) {
4635 int VectorSize = 0;
4636 if (Modifier & Use64BitVectors)
4637 VectorSize = 64;
4638 else if (Modifier & Use128BitVectors)
4639 VectorSize = 128;
4640
4641 // Return type.
4642 SmallVector<llvm::Type *, 3> Tys;
4643 if (Modifier & AddRetType) {
4644 llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext()));
4645 if (Modifier & VectorizeRetType)
4646 Ty = llvm::VectorType::get(
4647 Ty, VectorSize ? VectorSize / Ty->getPrimitiveSizeInBits() : 1);
4648
4649 Tys.push_back(Ty);
4650 }
4651
4652 // Arguments.
4653 if (Modifier & VectorizeArgTypes) {
4654 int Elts = VectorSize ? VectorSize / ArgType->getPrimitiveSizeInBits() : 1;
4655 ArgType = llvm::VectorType::get(ArgType, Elts);
4656 }
4657
4658 if (Modifier & (Add1ArgType | Add2ArgTypes))
4659 Tys.push_back(ArgType);
4660
4661 if (Modifier & Add2ArgTypes)
4662 Tys.push_back(ArgType);
4663
4664 if (Modifier & InventFloatType)
4665 Tys.push_back(FloatTy);
4666
4667 return CGM.getIntrinsic(IntrinsicID, Tys);
4668}
4669
4670static Value *EmitCommonNeonSISDBuiltinExpr(CodeGenFunction &CGF,
4671 const NeonIntrinsicInfo &SISDInfo,
4672 SmallVectorImpl<Value *> &Ops,
4673 const CallExpr *E) {
4674 unsigned BuiltinID = SISDInfo.BuiltinID;
4675 unsigned int Int = SISDInfo.LLVMIntrinsic;
4676 unsigned Modifier = SISDInfo.TypeModifier;
4677 const char *s = SISDInfo.NameHint;
4678
4679 switch (BuiltinID) {
4680 case NEON::BI__builtin_neon_vcled_s64:
4681 case NEON::BI__builtin_neon_vcled_u64:
4682 case NEON::BI__builtin_neon_vcles_f32:
4683 case NEON::BI__builtin_neon_vcled_f64:
4684 case NEON::BI__builtin_neon_vcltd_s64:
4685 case NEON::BI__builtin_neon_vcltd_u64:
4686 case NEON::BI__builtin_neon_vclts_f32:
4687 case NEON::BI__builtin_neon_vcltd_f64:
4688 case NEON::BI__builtin_neon_vcales_f32:
4689 case NEON::BI__builtin_neon_vcaled_f64:
4690 case NEON::BI__builtin_neon_vcalts_f32:
4691 case NEON::BI__builtin_neon_vcaltd_f64:
4692 // Only one direction of comparisons actually exist, cmle is actually a cmge
4693 // with swapped operands. The table gives us the right intrinsic but we
4694 // still need to do the swap.
4695 std::swap(Ops[0], Ops[1]);
4696 break;
4697 }
4698
4699 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 4699, __extension__ __PRETTY_FUNCTION__))
;
4700
4701 // Determine the type(s) of this overloaded AArch64 intrinsic.
4702 const Expr *Arg = E->getArg(0);
4703 llvm::Type *ArgTy = CGF.ConvertType(Arg->getType());
4704 Function *F = CGF.LookupNeonLLVMIntrinsic(Int, Modifier, ArgTy, E);
4705
4706 int j = 0;
4707 ConstantInt *C0 = ConstantInt::get(CGF.SizeTy, 0);
4708 for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
4709 ai != ae; ++ai, ++j) {
4710 llvm::Type *ArgTy = ai->getType();
4711 if (Ops[j]->getType()->getPrimitiveSizeInBits() ==
4712 ArgTy->getPrimitiveSizeInBits())
4713 continue;
4714
4715 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 4715, __extension__ __PRETTY_FUNCTION__))
;
4716 // The constant argument to an _n_ intrinsic always has Int32Ty, so truncate
4717 // it before inserting.
4718 Ops[j] =
4719 CGF.Builder.CreateTruncOrBitCast(Ops[j], ArgTy->getVectorElementType());
4720 Ops[j] =
4721 CGF.Builder.CreateInsertElement(UndefValue::get(ArgTy), Ops[j], C0);
4722 }
4723
4724 Value *Result = CGF.EmitNeonCall(F, Ops, s);
4725 llvm::Type *ResultType = CGF.ConvertType(E->getType());
4726 if (ResultType->getPrimitiveSizeInBits() <
4727 Result->getType()->getPrimitiveSizeInBits())
4728 return CGF.Builder.CreateExtractElement(Result, C0);
4729
4730 return CGF.Builder.CreateBitCast(Result, ResultType, s);
4731}
4732
4733Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
4734 unsigned BuiltinID, unsigned LLVMIntrinsic, unsigned AltLLVMIntrinsic,
4735 const char *NameHint, unsigned Modifier, const CallExpr *E,
4736 SmallVectorImpl<llvm::Value *> &Ops, Address PtrOp0, Address PtrOp1,
4737 llvm::Triple::ArchType Arch) {
4738 // Get the last argument, which specifies the vector type.
4739 llvm::APSInt NeonTypeConst;
4740 const Expr *Arg = E->getArg(E->getNumArgs() - 1);
4741 if (!Arg->isIntegerConstantExpr(NeonTypeConst, getContext()))
4742 return nullptr;
4743
4744 // Determine the type of this overloaded NEON intrinsic.
4745 NeonTypeFlags Type(NeonTypeConst.getZExtValue());
4746 bool Usgn = Type.isUnsigned();
4747 bool Quad = Type.isQuad();
4748 const bool HasLegalHalfType = getTarget().hasLegalHalfType();
4749
4750 llvm::VectorType *VTy = GetNeonType(this, Type, HasLegalHalfType);
4751 llvm::Type *Ty = VTy;
4752 if (!Ty)
4753 return nullptr;
4754
4755 auto getAlignmentValue32 = [&](Address addr) -> Value* {
4756 return Builder.getInt32(addr.getAlignment().getQuantity());
4757 };
4758
4759 unsigned Int = LLVMIntrinsic;
4760 if ((Modifier & UnsignedAlts) && !Usgn)
4761 Int = AltLLVMIntrinsic;
4762
4763 switch (BuiltinID) {
4764 default: break;
4765 case NEON::BI__builtin_neon_vabs_v:
4766 case NEON::BI__builtin_neon_vabsq_v:
4767 if (VTy->getElementType()->isFloatingPointTy())
4768 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, Ty), Ops, "vabs");
4769 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), Ops, "vabs");
4770 case NEON::BI__builtin_neon_vaddhn_v: {
4771 llvm::VectorType *SrcTy =
4772 llvm::VectorType::getExtendedElementVectorType(VTy);
4773
4774 // %sum = add <4 x i32> %lhs, %rhs
4775 Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
4776 Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy);
4777 Ops[0] = Builder.CreateAdd(Ops[0], Ops[1], "vaddhn");
4778
4779 // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16>
4780 Constant *ShiftAmt =
4781 ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2);
4782 Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vaddhn");
4783
4784 // %res = trunc <4 x i32> %high to <4 x i16>
4785 return Builder.CreateTrunc(Ops[0], VTy, "vaddhn");
4786 }
4787 case NEON::BI__builtin_neon_vcale_v:
4788 case NEON::BI__builtin_neon_vcaleq_v:
4789 case NEON::BI__builtin_neon_vcalt_v:
4790 case NEON::BI__builtin_neon_vcaltq_v:
4791 std::swap(Ops[0], Ops[1]);
4792 LLVM_FALLTHROUGH[[clang::fallthrough]];
4793 case NEON::BI__builtin_neon_vcage_v:
4794 case NEON::BI__builtin_neon_vcageq_v:
4795 case NEON::BI__builtin_neon_vcagt_v:
4796 case NEON::BI__builtin_neon_vcagtq_v: {
4797 llvm::Type *Ty;
4798 switch (VTy->getScalarSizeInBits()) {
4799 default: llvm_unreachable("unexpected type")::llvm::llvm_unreachable_internal("unexpected type", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 4799)
;
4800 case 32:
4801 Ty = FloatTy;
4802 break;
4803 case 64:
4804 Ty = DoubleTy;
4805 break;
4806 case 16:
4807 Ty = HalfTy;
4808 break;
4809 }
4810 llvm::Type *VecFlt = llvm::VectorType::get(Ty, VTy->getNumElements());
4811 llvm::Type *Tys[] = { VTy, VecFlt };
4812 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
4813 return EmitNeonCall(F, Ops, NameHint);
4814 }
4815 case NEON::BI__builtin_neon_vceqz_v:
4816 case NEON::BI__builtin_neon_vceqzq_v:
4817 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OEQ,
4818 ICmpInst::ICMP_EQ, "vceqz");
4819 case NEON::BI__builtin_neon_vcgez_v:
4820 case NEON::BI__builtin_neon_vcgezq_v:
4821 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGE,
4822 ICmpInst::ICMP_SGE, "vcgez");
4823 case NEON::BI__builtin_neon_vclez_v:
4824 case NEON::BI__builtin_neon_vclezq_v:
4825 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLE,
4826 ICmpInst::ICMP_SLE, "vclez");
4827 case NEON::BI__builtin_neon_vcgtz_v:
4828 case NEON::BI__builtin_neon_vcgtzq_v:
4829 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGT,
4830 ICmpInst::ICMP_SGT, "vcgtz");
4831 case NEON::BI__builtin_neon_vcltz_v:
4832 case NEON::BI__builtin_neon_vcltzq_v:
4833 return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLT,
4834 ICmpInst::ICMP_SLT, "vcltz");
4835 case NEON::BI__builtin_neon_vclz_v:
4836 case NEON::BI__builtin_neon_vclzq_v:
4837 // We generate target-independent intrinsic, which needs a second argument
4838 // for whether or not clz of zero is undefined; on ARM it isn't.
4839 Ops.push_back(Builder.getInt1(getTarget().isCLZForZeroUndef()));
4840 break;
4841 case NEON::BI__builtin_neon_vcvt_f32_v:
4842 case NEON::BI__builtin_neon_vcvtq_f32_v:
4843 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
4844 Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float32, false, Quad),
4845 HasLegalHalfType);
4846 return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
4847 : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
4848 case NEON::BI__builtin_neon_vcvt_f16_v:
4849 case NEON::BI__builtin_neon_vcvtq_f16_v:
4850 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
4851 Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float16, false, Quad),
4852 HasLegalHalfType);
4853 return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
4854 : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
4855 case NEON::BI__builtin_neon_vcvt_n_f16_v:
4856 case NEON::BI__builtin_neon_vcvt_n_f32_v:
4857 case NEON::BI__builtin_neon_vcvt_n_f64_v:
4858 case NEON::BI__builtin_neon_vcvtq_n_f16_v:
4859 case NEON::BI__builtin_neon_vcvtq_n_f32_v:
4860 case NEON::BI__builtin_neon_vcvtq_n_f64_v: {
4861 llvm::Type *Tys[2] = { GetFloatNeonType(this, Type), Ty };
4862 Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic;
4863 Function *F = CGM.getIntrinsic(Int, Tys);
4864 return EmitNeonCall(F, Ops, "vcvt_n");
4865 }
4866 case NEON::BI__builtin_neon_vcvt_n_s16_v:
4867 case NEON::BI__builtin_neon_vcvt_n_s32_v:
4868 case NEON::BI__builtin_neon_vcvt_n_u16_v:
4869 case NEON::BI__builtin_neon_vcvt_n_u32_v:
4870 case NEON::BI__builtin_neon_vcvt_n_s64_v:
4871 case NEON::BI__builtin_neon_vcvt_n_u64_v:
4872 case NEON::BI__builtin_neon_vcvtq_n_s16_v:
4873 case NEON::BI__builtin_neon_vcvtq_n_s32_v:
4874 case NEON::BI__builtin_neon_vcvtq_n_u16_v:
4875 case NEON::BI__builtin_neon_vcvtq_n_u32_v:
4876 case NEON::BI__builtin_neon_vcvtq_n_s64_v:
4877 case NEON::BI__builtin_neon_vcvtq_n_u64_v: {
4878 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
4879 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
4880 return EmitNeonCall(F, Ops, "vcvt_n");
4881 }
4882 case NEON::BI__builtin_neon_vcvt_s32_v:
4883 case NEON::BI__builtin_neon_vcvt_u32_v:
4884 case NEON::BI__builtin_neon_vcvt_s64_v:
4885 case NEON::BI__builtin_neon_vcvt_u64_v:
4886 case NEON::BI__builtin_neon_vcvt_s16_v:
4887 case NEON::BI__builtin_neon_vcvt_u16_v:
4888 case NEON::BI__builtin_neon_vcvtq_s32_v:
4889 case NEON::BI__builtin_neon_vcvtq_u32_v:
4890 case NEON::BI__builtin_neon_vcvtq_s64_v:
4891 case NEON::BI__builtin_neon_vcvtq_u64_v:
4892 case NEON::BI__builtin_neon_vcvtq_s16_v:
4893 case NEON::BI__builtin_neon_vcvtq_u16_v: {
4894 Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type));
4895 return Usgn ? Builder.CreateFPToUI(Ops[0], Ty, "vcvt")
4896 : Builder.CreateFPToSI(Ops[0], Ty, "vcvt");
4897 }
4898 case NEON::BI__builtin_neon_vcvta_s16_v:
4899 case NEON::BI__builtin_neon_vcvta_s32_v:
4900 case NEON::BI__builtin_neon_vcvta_s64_v:
4901 case NEON::BI__builtin_neon_vcvta_u16_v:
4902 case NEON::BI__builtin_neon_vcvta_u32_v:
4903 case NEON::BI__builtin_neon_vcvta_u64_v:
4904 case NEON::BI__builtin_neon_vcvtaq_s16_v:
4905 case NEON::BI__builtin_neon_vcvtaq_s32_v:
4906 case NEON::BI__builtin_neon_vcvtaq_s64_v:
4907 case NEON::BI__builtin_neon_vcvtaq_u16_v:
4908 case NEON::BI__builtin_neon_vcvtaq_u32_v:
4909 case NEON::BI__builtin_neon_vcvtaq_u64_v:
4910 case NEON::BI__builtin_neon_vcvtn_s16_v:
4911 case NEON::BI__builtin_neon_vcvtn_s32_v:
4912 case NEON::BI__builtin_neon_vcvtn_s64_v:
4913 case NEON::BI__builtin_neon_vcvtn_u16_v:
4914 case NEON::BI__builtin_neon_vcvtn_u32_v:
4915 case NEON::BI__builtin_neon_vcvtn_u64_v:
4916 case NEON::BI__builtin_neon_vcvtnq_s16_v:
4917 case NEON::BI__builtin_neon_vcvtnq_s32_v:
4918 case NEON::BI__builtin_neon_vcvtnq_s64_v:
4919 case NEON::BI__builtin_neon_vcvtnq_u16_v:
4920 case NEON::BI__builtin_neon_vcvtnq_u32_v:
4921 case NEON::BI__builtin_neon_vcvtnq_u64_v:
4922 case NEON::BI__builtin_neon_vcvtp_s16_v:
4923 case NEON::BI__builtin_neon_vcvtp_s32_v:
4924 case NEON::BI__builtin_neon_vcvtp_s64_v:
4925 case NEON::BI__builtin_neon_vcvtp_u16_v:
4926 case NEON::BI__builtin_neon_vcvtp_u32_v:
4927 case NEON::BI__builtin_neon_vcvtp_u64_v:
4928 case NEON::BI__builtin_neon_vcvtpq_s16_v:
4929 case NEON::BI__builtin_neon_vcvtpq_s32_v:
4930 case NEON::BI__builtin_neon_vcvtpq_s64_v:
4931 case NEON::BI__builtin_neon_vcvtpq_u16_v:
4932 case NEON::BI__builtin_neon_vcvtpq_u32_v:
4933 case NEON::BI__builtin_neon_vcvtpq_u64_v:
4934 case NEON::BI__builtin_neon_vcvtm_s16_v:
4935 case NEON::BI__builtin_neon_vcvtm_s32_v:
4936 case NEON::BI__builtin_neon_vcvtm_s64_v:
4937 case NEON::BI__builtin_neon_vcvtm_u16_v:
4938 case NEON::BI__builtin_neon_vcvtm_u32_v:
4939 case NEON::BI__builtin_neon_vcvtm_u64_v:
4940 case NEON::BI__builtin_neon_vcvtmq_s16_v:
4941 case NEON::BI__builtin_neon_vcvtmq_s32_v:
4942 case NEON::BI__builtin_neon_vcvtmq_s64_v:
4943 case NEON::BI__builtin_neon_vcvtmq_u16_v:
4944 case NEON::BI__builtin_neon_vcvtmq_u32_v:
4945 case NEON::BI__builtin_neon_vcvtmq_u64_v: {
4946 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
4947 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint);
4948 }
4949 case NEON::BI__builtin_neon_vext_v:
4950 case NEON::BI__builtin_neon_vextq_v: {
4951 int CV = cast<ConstantInt>(Ops[2])->getSExtValue();
4952 SmallVector<uint32_t, 16> Indices;
4953 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
4954 Indices.push_back(i+CV);
4955
4956 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
4957 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
4958 return Builder.CreateShuffleVector(Ops[0], Ops[1], Indices, "vext");
4959 }
4960 case NEON::BI__builtin_neon_vfma_v:
4961 case NEON::BI__builtin_neon_vfmaq_v: {
4962 Value *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
4963 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
4964 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
4965 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
4966
4967 // NEON intrinsic puts accumulator first, unlike the LLVM fma.
4968 return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
4969 }
4970 case NEON::BI__builtin_neon_vld1_v:
4971 case NEON::BI__builtin_neon_vld1q_v: {
4972 llvm::Type *Tys[] = {Ty, Int8PtrTy};
4973 Ops.push_back(getAlignmentValue32(PtrOp0));
4974 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "vld1");
4975 }
4976 case NEON::BI__builtin_neon_vld1_x2_v:
4977 case NEON::BI__builtin_neon_vld1q_x2_v:
4978 case NEON::BI__builtin_neon_vld1_x3_v:
4979 case NEON::BI__builtin_neon_vld1q_x3_v:
4980 case NEON::BI__builtin_neon_vld1_x4_v:
4981 case NEON::BI__builtin_neon_vld1q_x4_v: {
4982 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getVectorElementType());
4983 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
4984 llvm::Type *Tys[2] = { VTy, PTy };
4985 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
4986 Ops[1] = Builder.CreateCall(F, Ops[1], "vld1xN");
4987 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
4988 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
4989 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
4990 }
4991 case NEON::BI__builtin_neon_vld2_v:
4992 case NEON::BI__builtin_neon_vld2q_v:
4993 case NEON::BI__builtin_neon_vld3_v:
4994 case NEON::BI__builtin_neon_vld3q_v:
4995 case NEON::BI__builtin_neon_vld4_v:
4996 case NEON::BI__builtin_neon_vld4q_v:
4997 case NEON::BI__builtin_neon_vld2_dup_v:
4998 case NEON::BI__builtin_neon_vld2q_dup_v:
4999 case NEON::BI__builtin_neon_vld3_dup_v:
5000 case NEON::BI__builtin_neon_vld3q_dup_v:
5001 case NEON::BI__builtin_neon_vld4_dup_v:
5002 case NEON::BI__builtin_neon_vld4q_dup_v: {
5003 llvm::Type *Tys[] = {Ty, Int8PtrTy};
5004 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
5005 Value *Align = getAlignmentValue32(PtrOp1);
5006 Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, NameHint);
5007 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
5008 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5009 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
5010 }
5011 case NEON::BI__builtin_neon_vld1_dup_v:
5012 case NEON::BI__builtin_neon_vld1q_dup_v: {
5013 Value *V = UndefValue::get(Ty);
5014 Ty = llvm::PointerType::getUnqual(VTy->getElementType());
5015 PtrOp0 = Builder.CreateBitCast(PtrOp0, Ty);
5016 LoadInst *Ld = Builder.CreateLoad(PtrOp0);
5017 llvm::Constant *CI = ConstantInt::get(SizeTy, 0);
5018 Ops[0] = Builder.CreateInsertElement(V, Ld, CI);
5019 return EmitNeonSplat(Ops[0], CI);
5020 }
5021 case NEON::BI__builtin_neon_vld2_lane_v:
5022 case NEON::BI__builtin_neon_vld2q_lane_v:
5023 case NEON::BI__builtin_neon_vld3_lane_v:
5024 case NEON::BI__builtin_neon_vld3q_lane_v:
5025 case NEON::BI__builtin_neon_vld4_lane_v:
5026 case NEON::BI__builtin_neon_vld4q_lane_v: {
5027 llvm::Type *Tys[] = {Ty, Int8PtrTy};
5028 Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
5029 for (unsigned I = 2; I < Ops.size() - 1; ++I)
5030 Ops[I] = Builder.CreateBitCast(Ops[I], Ty);
5031 Ops.push_back(getAlignmentValue32(PtrOp1));
5032 Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), NameHint);
5033 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
5034 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5035 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
5036 }
5037 case NEON::BI__builtin_neon_vmovl_v: {
5038 llvm::Type *DTy =llvm::VectorType::getTruncatedElementVectorType(VTy);
5039 Ops[0] = Builder.CreateBitCast(Ops[0], DTy);
5040 if (Usgn)
5041 return Builder.CreateZExt(Ops[0], Ty, "vmovl");
5042 return Builder.CreateSExt(Ops[0], Ty, "vmovl");
5043 }
5044 case NEON::BI__builtin_neon_vmovn_v: {
5045 llvm::Type *QTy = llvm::VectorType::getExtendedElementVectorType(VTy);
5046 Ops[0] = Builder.CreateBitCast(Ops[0], QTy);
5047 return Builder.CreateTrunc(Ops[0], Ty, "vmovn");
5048 }
5049 case NEON::BI__builtin_neon_vmull_v:
5050 // FIXME: the integer vmull operations could be emitted in terms of pure
5051 // LLVM IR (2 exts followed by a mul). Unfortunately LLVM has a habit of
5052 // hoisting the exts outside loops. Until global ISel comes along that can
5053 // see through such movement this leads to bad CodeGen. So we need an
5054 // intrinsic for now.
5055 Int = Usgn ? Intrinsic::arm_neon_vmullu : Intrinsic::arm_neon_vmulls;
5056 Int = Type.isPoly() ? (unsigned)Intrinsic::arm_neon_vmullp : Int;
5057 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull");
5058 case NEON::BI__builtin_neon_vpadal_v:
5059 case NEON::BI__builtin_neon_vpadalq_v: {
5060 // The source operand type has twice as many elements of half the size.
5061 unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits();
5062 llvm::Type *EltTy =
5063 llvm::IntegerType::get(getLLVMContext(), EltBits / 2);
5064 llvm::Type *NarrowTy =
5065 llvm::VectorType::get(EltTy, VTy->getNumElements() * 2);
5066 llvm::Type *Tys[2] = { Ty, NarrowTy };
5067 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint);
5068 }
5069 case NEON::BI__builtin_neon_vpaddl_v:
5070 case NEON::BI__builtin_neon_vpaddlq_v: {
5071 // The source operand type has twice as many elements of half the size.
5072 unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits();
5073 llvm::Type *EltTy = llvm::IntegerType::get(getLLVMContext(), EltBits / 2);
5074 llvm::Type *NarrowTy =
5075 llvm::VectorType::get(EltTy, VTy->getNumElements() * 2);
5076 llvm::Type *Tys[2] = { Ty, NarrowTy };
5077 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vpaddl");
5078 }
5079 case NEON::BI__builtin_neon_vqdmlal_v:
5080 case NEON::BI__builtin_neon_vqdmlsl_v: {
5081 SmallVector<Value *, 2> MulOps(Ops.begin() + 1, Ops.end());
5082 Ops[1] =
5083 EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), MulOps, "vqdmlal");
5084 Ops.resize(2);
5085 return EmitNeonCall(CGM.getIntrinsic(AltLLVMIntrinsic, Ty), Ops, NameHint);
5086 }
5087 case NEON::BI__builtin_neon_vqshl_n_v:
5088 case NEON::BI__builtin_neon_vqshlq_n_v:
5089 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshl_n",
5090 1, false);
5091 case NEON::BI__builtin_neon_vqshlu_n_v:
5092 case NEON::BI__builtin_neon_vqshluq_n_v:
5093 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshlu_n",
5094 1, false);
5095 case NEON::BI__builtin_neon_vrecpe_v:
5096 case NEON::BI__builtin_neon_vrecpeq_v:
5097 case NEON::BI__builtin_neon_vrsqrte_v:
5098 case NEON::BI__builtin_neon_vrsqrteq_v:
5099 Int = Ty->isFPOrFPVectorTy() ? LLVMIntrinsic : AltLLVMIntrinsic;
5100 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint);
5101 case NEON::BI__builtin_neon_vrndi_v:
5102 case NEON::BI__builtin_neon_vrndiq_v:
5103 Int = Intrinsic::nearbyint;
5104 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint);
5105 case NEON::BI__builtin_neon_vrshr_n_v:
5106 case NEON::BI__builtin_neon_vrshrq_n_v:
5107 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshr_n",
5108 1, true);
5109 case NEON::BI__builtin_neon_vshl_n_v:
5110 case NEON::BI__builtin_neon_vshlq_n_v:
5111 Ops[1] = EmitNeonShiftVector(Ops[1], Ty, false);
5112 return Builder.CreateShl(Builder.CreateBitCast(Ops[0],Ty), Ops[1],
5113 "vshl_n");
5114 case NEON::BI__builtin_neon_vshll_n_v: {
5115 llvm::Type *SrcTy = llvm::VectorType::getTruncatedElementVectorType(VTy);
5116 Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
5117 if (Usgn)
5118 Ops[0] = Builder.CreateZExt(Ops[0], VTy);
5119 else
5120 Ops[0] = Builder.CreateSExt(Ops[0], VTy);
5121 Ops[1] = EmitNeonShiftVector(Ops[1], VTy, false);
5122 return Builder.CreateShl(Ops[0], Ops[1], "vshll_n");
5123 }
5124 case NEON::BI__builtin_neon_vshrn_n_v: {
5125 llvm::Type *SrcTy = llvm::VectorType::getExtendedElementVectorType(VTy);
5126 Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
5127 Ops[1] = EmitNeonShiftVector(Ops[1], SrcTy, false);
5128 if (Usgn)
5129 Ops[0] = Builder.CreateLShr(Ops[0], Ops[1]);
5130 else
5131 Ops[0] = Builder.CreateAShr(Ops[0], Ops[1]);
5132 return Builder.CreateTrunc(Ops[0], Ty, "vshrn_n");
5133 }
5134 case NEON::BI__builtin_neon_vshr_n_v:
5135 case NEON::BI__builtin_neon_vshrq_n_v:
5136 return EmitNeonRShiftImm(Ops[0], Ops[1], Ty, Usgn, "vshr_n");
5137 case NEON::BI__builtin_neon_vst1_v:
5138 case NEON::BI__builtin_neon_vst1q_v:
5139 case NEON::BI__builtin_neon_vst2_v:
5140 case NEON::BI__builtin_neon_vst2q_v:
5141 case NEON::BI__builtin_neon_vst3_v:
5142 case NEON::BI__builtin_neon_vst3q_v:
5143 case NEON::BI__builtin_neon_vst4_v:
5144 case NEON::BI__builtin_neon_vst4q_v:
5145 case NEON::BI__builtin_neon_vst2_lane_v:
5146 case NEON::BI__builtin_neon_vst2q_lane_v:
5147 case NEON::BI__builtin_neon_vst3_lane_v:
5148 case NEON::BI__builtin_neon_vst3q_lane_v:
5149 case NEON::BI__builtin_neon_vst4_lane_v:
5150 case NEON::BI__builtin_neon_vst4q_lane_v: {
5151 llvm::Type *Tys[] = {Int8PtrTy, Ty};
5152 Ops.push_back(getAlignmentValue32(PtrOp0));
5153 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "");
5154 }
5155 case NEON::BI__builtin_neon_vst1_x2_v:
5156 case NEON::BI__builtin_neon_vst1q_x2_v:
5157 case NEON::BI__builtin_neon_vst1_x3_v:
5158 case NEON::BI__builtin_neon_vst1q_x3_v:
5159 case NEON::BI__builtin_neon_vst1_x4_v:
5160 case NEON::BI__builtin_neon_vst1q_x4_v: {
5161 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getVectorElementType());
5162 // TODO: Currently in AArch32 mode the pointer operand comes first, whereas
5163 // in AArch64 it comes last. We may want to stick to one or another.
5164 if (Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_be) {
5165 llvm::Type *Tys[2] = { VTy, PTy };
5166 std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
5167 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "");
5168 }
5169 llvm::Type *Tys[2] = { PTy, VTy };
5170 return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "");
5171 }
5172 case NEON::BI__builtin_neon_vsubhn_v: {
5173 llvm::VectorType *SrcTy =
5174 llvm::VectorType::getExtendedElementVectorType(VTy);
5175
5176 // %sum = add <4 x i32> %lhs, %rhs
5177 Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
5178 Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy);
5179 Ops[0] = Builder.CreateSub(Ops[0], Ops[1], "vsubhn");
5180
5181 // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16>
5182 Constant *ShiftAmt =
5183 ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2);
5184 Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vsubhn");
5185
5186 // %res = trunc <4 x i32> %high to <4 x i16>
5187 return Builder.CreateTrunc(Ops[0], VTy, "vsubhn");
5188 }
5189 case NEON::BI__builtin_neon_vtrn_v:
5190 case NEON::BI__builtin_neon_vtrnq_v: {
5191 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
5192 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5193 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
5194 Value *SV = nullptr;
5195
5196 for (unsigned vi = 0; vi != 2; ++vi) {
5197 SmallVector<uint32_t, 16> Indices;
5198 for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
5199 Indices.push_back(i+vi);
5200 Indices.push_back(i+e+vi);
5201 }
5202 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
5203 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn");
5204 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
5205 }
5206 return SV;
5207 }
5208 case NEON::BI__builtin_neon_vtst_v:
5209 case NEON::BI__builtin_neon_vtstq_v: {
5210 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5211 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5212 Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]);
5213 Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0],
5214 ConstantAggregateZero::get(Ty));
5215 return Builder.CreateSExt(Ops[0], Ty, "vtst");
5216 }
5217 case NEON::BI__builtin_neon_vuzp_v:
5218 case NEON::BI__builtin_neon_vuzpq_v: {
5219 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
5220 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5221 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
5222 Value *SV = nullptr;
5223
5224 for (unsigned vi = 0; vi != 2; ++vi) {
5225 SmallVector<uint32_t, 16> Indices;
5226 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
5227 Indices.push_back(2*i+vi);
5228
5229 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
5230 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp");
5231 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
5232 }
5233 return SV;
5234 }
5235 case NEON::BI__builtin_neon_vzip_v:
5236 case NEON::BI__builtin_neon_vzipq_v: {
5237 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
5238 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5239 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
5240 Value *SV = nullptr;
5241
5242 for (unsigned vi = 0; vi != 2; ++vi) {
5243 SmallVector<uint32_t, 16> Indices;
5244 for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
5245 Indices.push_back((i + vi*e) >> 1);
5246 Indices.push_back(((i + vi*e) >> 1)+e);
5247 }
5248 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
5249 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip");
5250 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
5251 }
5252 return SV;
5253 }
5254 case NEON::BI__builtin_neon_vdot_v:
5255 case NEON::BI__builtin_neon_vdotq_v: {
5256 llvm::Type *InputTy =
5257 llvm::VectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8);
5258 llvm::Type *Tys[2] = { Ty, InputTy };
5259 Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic;
5260 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vdot");
5261 }
5262 }
5263
5264 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5264, __extension__ __PRETTY_FUNCTION__))
;
5265
5266 // Determine the type(s) of this overloaded AArch64 intrinsic.
5267 Function *F = LookupNeonLLVMIntrinsic(Int, Modifier, Ty, E);
5268
5269 Value *Result = EmitNeonCall(F, Ops, NameHint);
5270 llvm::Type *ResultType = ConvertType(E->getType());
5271 // AArch64 intrinsic one-element vector type cast to
5272 // scalar type expected by the builtin
5273 return Builder.CreateBitCast(Result, ResultType, NameHint);
5274}
5275
5276Value *CodeGenFunction::EmitAArch64CompareBuiltinExpr(
5277 Value *Op, llvm::Type *Ty, const CmpInst::Predicate Fp,
5278 const CmpInst::Predicate Ip, const Twine &Name) {
5279 llvm::Type *OTy = Op->getType();
5280
5281 // FIXME: this is utterly horrific. We should not be looking at previous
5282 // codegen context to find out what needs doing. Unfortunately TableGen
5283 // currently gives us exactly the same calls for vceqz_f32 and vceqz_s32
5284 // (etc).
5285 if (BitCastInst *BI = dyn_cast<BitCastInst>(Op))
5286 OTy = BI->getOperand(0)->getType();
5287
5288 Op = Builder.CreateBitCast(Op, OTy);
5289 if (OTy->getScalarType()->isFloatingPointTy()) {
5290 Op = Builder.CreateFCmp(Fp, Op, Constant::getNullValue(OTy));
5291 } else {
5292 Op = Builder.CreateICmp(Ip, Op, Constant::getNullValue(OTy));
5293 }
5294 return Builder.CreateSExt(Op, Ty, Name);
5295}
5296
5297static Value *packTBLDVectorList(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
5298 Value *ExtOp, Value *IndexOp,
5299 llvm::Type *ResTy, unsigned IntID,
5300 const char *Name) {
5301 SmallVector<Value *, 2> TblOps;
5302 if (ExtOp)
5303 TblOps.push_back(ExtOp);
5304
5305 // Build a vector containing sequential number like (0, 1, 2, ..., 15)
5306 SmallVector<uint32_t, 16> Indices;
5307 llvm::VectorType *TblTy = cast<llvm::VectorType>(Ops[0]->getType());
5308 for (unsigned i = 0, e = TblTy->getNumElements(); i != e; ++i) {
5309 Indices.push_back(2*i);
5310 Indices.push_back(2*i+1);
5311 }
5312
5313 int PairPos = 0, End = Ops.size() - 1;
5314 while (PairPos < End) {
5315 TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos],
5316 Ops[PairPos+1], Indices,
5317 Name));
5318 PairPos += 2;
5319 }
5320
5321 // If there's an odd number of 64-bit lookup table, fill the high 64-bit
5322 // of the 128-bit lookup table with zero.
5323 if (PairPos == End) {
5324 Value *ZeroTbl = ConstantAggregateZero::get(TblTy);
5325 TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos],
5326 ZeroTbl, Indices, Name));
5327 }
5328
5329 Function *TblF;
5330 TblOps.push_back(IndexOp);
5331 TblF = CGF.CGM.getIntrinsic(IntID, ResTy);
5332
5333 return CGF.EmitNeonCall(TblF, TblOps, Name);
5334}
5335
5336Value *CodeGenFunction::GetValueForARMHint(unsigned BuiltinID) {
5337 unsigned Value;
5338 switch (BuiltinID) {
5339 default:
5340 return nullptr;
5341 case ARM::BI__builtin_arm_nop:
5342 Value = 0;
5343 break;
5344 case ARM::BI__builtin_arm_yield:
5345 case ARM::BI__yield:
5346 Value = 1;
5347 break;
5348 case ARM::BI__builtin_arm_wfe:
5349 case ARM::BI__wfe:
5350 Value = 2;
5351 break;
5352 case ARM::BI__builtin_arm_wfi:
5353 case ARM::BI__wfi:
5354 Value = 3;
5355 break;
5356 case ARM::BI__builtin_arm_sev:
5357 case ARM::BI__sev:
5358 Value = 4;
5359 break;
5360 case ARM::BI__builtin_arm_sevl:
5361 case ARM::BI__sevl:
5362 Value = 5;
5363 break;
5364 }
5365
5366 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_hint),
5367 llvm::ConstantInt::get(Int32Ty, Value));
5368}
5369
5370// Generates the IR for the read/write special register builtin,
5371// ValueType is the type of the value that is to be written or read,
5372// RegisterType is the type of the register being written to or read from.
5373static Value *EmitSpecialRegisterBuiltin(CodeGenFunction &CGF,
5374 const CallExpr *E,
5375 llvm::Type *RegisterType,
5376 llvm::Type *ValueType,
5377 bool IsRead,
5378 StringRef SysReg = "") {
5379 // write and register intrinsics only support 32 and 64 bit operations.
5380 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5381, __extension__ __PRETTY_FUNCTION__))
5381 && "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5381, __extension__ __PRETTY_FUNCTION__))
;
5382
5383 CodeGen::CGBuilderTy &Builder = CGF.Builder;
5384 CodeGen::CodeGenModule &CGM = CGF.CGM;
5385 LLVMContext &Context = CGM.getLLVMContext();
5386
5387 if (SysReg.empty()) {
5388 const Expr *SysRegStrExpr = E->getArg(0)->IgnoreParenCasts();
5389 SysReg = cast<clang::StringLiteral>(SysRegStrExpr)->getString();
5390 }
5391
5392 llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysReg) };
5393 llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops);
5394 llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName);
5395
5396 llvm::Type *Types[] = { RegisterType };
5397
5398 bool MixedTypes = RegisterType->isIntegerTy(64) && ValueType->isIntegerTy(32);
5399 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5400, __extension__ __PRETTY_FUNCTION__))
5400 && "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5400, __extension__ __PRETTY_FUNCTION__))
;
5401
5402 if (IsRead) {
5403 llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types);
5404 llvm::Value *Call = Builder.CreateCall(F, Metadata);
5405
5406 if (MixedTypes)
5407 // Read into 64 bit register and then truncate result to 32 bit.
5408 return Builder.CreateTrunc(Call, ValueType);
5409
5410 if (ValueType->isPointerTy())
5411 // Have i32/i64 result (Call) but want to return a VoidPtrTy (i8*).
5412 return Builder.CreateIntToPtr(Call, ValueType);
5413
5414 return Call;
5415 }
5416
5417 llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types);
5418 llvm::Value *ArgValue = CGF.EmitScalarExpr(E->getArg(1));
5419 if (MixedTypes) {
5420 // Extend 32 bit write value to 64 bit to pass to write.
5421 ArgValue = Builder.CreateZExt(ArgValue, RegisterType);
5422 return Builder.CreateCall(F, { Metadata, ArgValue });
5423 }
5424
5425 if (ValueType->isPointerTy()) {
5426 // Have VoidPtrTy ArgValue but want to return an i32/i64.
5427 ArgValue = Builder.CreatePtrToInt(ArgValue, RegisterType);
5428 return Builder.CreateCall(F, { Metadata, ArgValue });
5429 }
5430
5431 return Builder.CreateCall(F, { Metadata, ArgValue });
5432}
5433
5434/// Return true if BuiltinID is an overloaded Neon intrinsic with an extra
5435/// argument that specifies the vector type.
5436static bool HasExtraNeonArgument(unsigned BuiltinID) {
5437 switch (BuiltinID) {
5438 default: break;
5439 case NEON::BI__builtin_neon_vget_lane_i8:
5440 case NEON::BI__builtin_neon_vget_lane_i16:
5441 case NEON::BI__builtin_neon_vget_lane_i32:
5442 case NEON::BI__builtin_neon_vget_lane_i64:
5443 case NEON::BI__builtin_neon_vget_lane_f32:
5444 case NEON::BI__builtin_neon_vgetq_lane_i8:
5445 case NEON::BI__builtin_neon_vgetq_lane_i16:
5446 case NEON::BI__builtin_neon_vgetq_lane_i32:
5447 case NEON::BI__builtin_neon_vgetq_lane_i64:
5448 case NEON::BI__builtin_neon_vgetq_lane_f32:
5449 case NEON::BI__builtin_neon_vset_lane_i8:
5450 case NEON::BI__builtin_neon_vset_lane_i16:
5451 case NEON::BI__builtin_neon_vset_lane_i32:
5452 case NEON::BI__builtin_neon_vset_lane_i64:
5453 case NEON::BI__builtin_neon_vset_lane_f32:
5454 case NEON::BI__builtin_neon_vsetq_lane_i8:
5455 case NEON::BI__builtin_neon_vsetq_lane_i16:
5456 case NEON::BI__builtin_neon_vsetq_lane_i32:
5457 case NEON::BI__builtin_neon_vsetq_lane_i64:
5458 case NEON::BI__builtin_neon_vsetq_lane_f32:
5459 case NEON::BI__builtin_neon_vsha1h_u32:
5460 case NEON::BI__builtin_neon_vsha1cq_u32:
5461 case NEON::BI__builtin_neon_vsha1pq_u32:
5462 case NEON::BI__builtin_neon_vsha1mq_u32:
5463 case clang::ARM::BI_MoveToCoprocessor:
5464 case clang::ARM::BI_MoveToCoprocessor2:
5465 return false;
5466 }
5467 return true;
5468}
5469
5470Value *CodeGenFunction::EmitISOVolatileLoad(const CallExpr *E) {
5471 Value *Ptr = EmitScalarExpr(E->getArg(0));
5472 QualType ElTy = E->getArg(0)->getType()->getPointeeType();
5473 CharUnits LoadSize = getContext().getTypeSizeInChars(ElTy);
5474 llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(),
5475 LoadSize.getQuantity() * 8);
5476 Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo());
5477 llvm::LoadInst *Load =
5478 Builder.CreateAlignedLoad(Ptr, LoadSize);
5479 Load->setVolatile(true);
5480 return Load;
5481}
5482
5483Value *CodeGenFunction::EmitISOVolatileStore(const CallExpr *E) {
5484 Value *Ptr = EmitScalarExpr(E->getArg(0));
5485 Value *Value = EmitScalarExpr(E->getArg(1));
5486 QualType ElTy = E->getArg(0)->getType()->getPointeeType();
5487 CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy);
5488 llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(),
5489 StoreSize.getQuantity() * 8);
5490 Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo());
5491 llvm::StoreInst *Store =
5492 Builder.CreateAlignedStore(Value, Ptr,
5493 StoreSize);
5494 Store->setVolatile(true);
5495 return Store;
5496}
5497
5498Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
5499 const CallExpr *E,
5500 llvm::Triple::ArchType Arch) {
5501 if (auto Hint = GetValueForARMHint(BuiltinID))
5502 return Hint;
5503
5504 if (BuiltinID == ARM::BI__emit) {
5505 bool IsThumb = getTarget().getTriple().getArch() == llvm::Triple::thumb;
5506 llvm::FunctionType *FTy =
5507 llvm::FunctionType::get(VoidTy, /*Variadic=*/false);
5508
5509 APSInt Value;
5510 if (!E->getArg(0)->EvaluateAsInt(Value, CGM.getContext()))
5511 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5511)
;
5512
5513 uint64_t ZExtValue = Value.zextOrTrunc(IsThumb ? 16 : 32).getZExtValue();
5514
5515 llvm::InlineAsm *Emit =
5516 IsThumb ? InlineAsm::get(FTy, ".inst.n 0x" + utohexstr(ZExtValue), "",
5517 /*SideEffects=*/true)
5518 : InlineAsm::get(FTy, ".inst 0x" + utohexstr(ZExtValue), "",
5519 /*SideEffects=*/true);
5520
5521 return Builder.CreateCall(Emit);
5522 }
5523
5524 if (BuiltinID == ARM::BI__builtin_arm_dbg) {
5525 Value *Option = EmitScalarExpr(E->getArg(0));
5526 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_dbg), Option);
5527 }
5528
5529 if (BuiltinID == ARM::BI__builtin_arm_prefetch) {
5530 Value *Address = EmitScalarExpr(E->getArg(0));
5531 Value *RW = EmitScalarExpr(E->getArg(1));
5532 Value *IsData = EmitScalarExpr(E->getArg(2));
5533
5534 // Locality is not supported on ARM target
5535 Value *Locality = llvm::ConstantInt::get(Int32Ty, 3);
5536
5537 Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
5538 return Builder.CreateCall(F, {Address, RW, Locality, IsData});
5539 }
5540
5541 if (BuiltinID == ARM::BI__builtin_arm_rbit) {
5542 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
5543 return Builder.CreateCall(
5544 CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
5545 }
5546
5547 if (BuiltinID == ARM::BI__clear_cache) {
5548 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5548, __extension__ __PRETTY_FUNCTION__))
;
5549 const FunctionDecl *FD = E->getDirectCallee();
5550 Value *Ops[2];
5551 for (unsigned i = 0; i < 2; i++)
5552 Ops[i] = EmitScalarExpr(E->getArg(i));
5553 llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType());
5554 llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty);
5555 StringRef Name = FD->getName();
5556 return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops);
5557 }
5558
5559 if (BuiltinID == ARM::BI__builtin_arm_mcrr ||
5560 BuiltinID == ARM::BI__builtin_arm_mcrr2) {
5561 Function *F;
5562
5563 switch (BuiltinID) {
5564 default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5564)
;
5565 case ARM::BI__builtin_arm_mcrr:
5566 F = CGM.getIntrinsic(Intrinsic::arm_mcrr);
5567 break;
5568 case ARM::BI__builtin_arm_mcrr2:
5569 F = CGM.getIntrinsic(Intrinsic::arm_mcrr2);
5570 break;
5571 }
5572
5573 // MCRR{2} instruction has 5 operands but
5574 // the intrinsic has 4 because Rt and Rt2
5575 // are represented as a single unsigned 64
5576 // bit integer in the intrinsic definition
5577 // but internally it's represented as 2 32
5578 // bit integers.
5579
5580 Value *Coproc = EmitScalarExpr(E->getArg(0));
5581 Value *Opc1 = EmitScalarExpr(E->getArg(1));
5582 Value *RtAndRt2 = EmitScalarExpr(E->getArg(2));
5583 Value *CRm = EmitScalarExpr(E->getArg(3));
5584
5585 Value *C1 = llvm::ConstantInt::get(Int64Ty, 32);
5586 Value *Rt = Builder.CreateTruncOrBitCast(RtAndRt2, Int32Ty);
5587 Value *Rt2 = Builder.CreateLShr(RtAndRt2, C1);
5588 Rt2 = Builder.CreateTruncOrBitCast(Rt2, Int32Ty);
5589
5590 return Builder.CreateCall(F, {Coproc, Opc1, Rt, Rt2, CRm});
5591 }
5592
5593 if (BuiltinID == ARM::BI__builtin_arm_mrrc ||
5594 BuiltinID == ARM::BI__builtin_arm_mrrc2) {
5595 Function *F;
5596
5597 switch (BuiltinID) {
5598 default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5598)
;
5599 case ARM::BI__builtin_arm_mrrc:
5600 F = CGM.getIntrinsic(Intrinsic::arm_mrrc);
5601 break;
5602 case ARM::BI__builtin_arm_mrrc2:
5603 F = CGM.getIntrinsic(Intrinsic::arm_mrrc2);
5604 break;
5605 }
5606
5607 Value *Coproc = EmitScalarExpr(E->getArg(0));
5608 Value *Opc1 = EmitScalarExpr(E->getArg(1));
5609 Value *CRm = EmitScalarExpr(E->getArg(2));
5610 Value *RtAndRt2 = Builder.CreateCall(F, {Coproc, Opc1, CRm});
5611
5612 // Returns an unsigned 64 bit integer, represented
5613 // as two 32 bit integers.
5614
5615 Value *Rt = Builder.CreateExtractValue(RtAndRt2, 1);
5616 Value *Rt1 = Builder.CreateExtractValue(RtAndRt2, 0);
5617 Rt = Builder.CreateZExt(Rt, Int64Ty);
5618 Rt1 = Builder.CreateZExt(Rt1, Int64Ty);
5619
5620 Value *ShiftCast = llvm::ConstantInt::get(Int64Ty, 32);
5621 RtAndRt2 = Builder.CreateShl(Rt, ShiftCast, "shl", true);
5622 RtAndRt2 = Builder.CreateOr(RtAndRt2, Rt1);
5623
5624 return Builder.CreateBitCast(RtAndRt2, ConvertType(E->getType()));
5625 }
5626
5627 if (BuiltinID == ARM::BI__builtin_arm_ldrexd ||
5628 ((BuiltinID == ARM::BI__builtin_arm_ldrex ||
5629 BuiltinID == ARM::BI__builtin_arm_ldaex) &&
5630 getContext().getTypeSize(E->getType()) == 64) ||
5631 BuiltinID == ARM::BI__ldrexd) {
5632 Function *F;
5633
5634 switch (BuiltinID) {
5635 default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5635)
;
5636 case ARM::BI__builtin_arm_ldaex:
5637 F = CGM.getIntrinsic(Intrinsic::arm_ldaexd);
5638 break;
5639 case ARM::BI__builtin_arm_ldrexd:
5640 case ARM::BI__builtin_arm_ldrex:
5641 case ARM::BI__ldrexd:
5642 F = CGM.getIntrinsic(Intrinsic::arm_ldrexd);
5643 break;
5644 }
5645
5646 Value *LdPtr = EmitScalarExpr(E->getArg(0));
5647 Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy),
5648 "ldrexd");
5649
5650 Value *Val0 = Builder.CreateExtractValue(Val, 1);
5651 Value *Val1 = Builder.CreateExtractValue(Val, 0);
5652 Val0 = Builder.CreateZExt(Val0, Int64Ty);
5653 Val1 = Builder.CreateZExt(Val1, Int64Ty);
5654
5655 Value *ShiftCst = llvm::ConstantInt::get(Int64Ty, 32);
5656 Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */);
5657 Val = Builder.CreateOr(Val, Val1);
5658 return Builder.CreateBitCast(Val, ConvertType(E->getType()));
5659 }
5660
5661 if (BuiltinID == ARM::BI__builtin_arm_ldrex ||
5662 BuiltinID == ARM::BI__builtin_arm_ldaex) {
5663 Value *LoadAddr = EmitScalarExpr(E->getArg(0));
5664
5665 QualType Ty = E->getType();
5666 llvm::Type *RealResTy = ConvertType(Ty);
5667 llvm::Type *PtrTy = llvm::IntegerType::get(
5668 getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo();
5669 LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy);
5670
5671 Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_ldaex
5672 ? Intrinsic::arm_ldaex
5673 : Intrinsic::arm_ldrex,
5674 PtrTy);
5675 Value *Val = Builder.CreateCall(F, LoadAddr, "ldrex");
5676
5677 if (RealResTy->isPointerTy())
5678 return Builder.CreateIntToPtr(Val, RealResTy);
5679 else {
5680 llvm::Type *IntResTy = llvm::IntegerType::get(
5681 getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy));
5682 Val = Builder.CreateTruncOrBitCast(Val, IntResTy);
5683 return Builder.CreateBitCast(Val, RealResTy);
5684 }
5685 }
5686
5687 if (BuiltinID == ARM::BI__builtin_arm_strexd ||
5688 ((BuiltinID == ARM::BI__builtin_arm_stlex ||
5689 BuiltinID == ARM::BI__builtin_arm_strex) &&
5690 getContext().getTypeSize(E->getArg(0)->getType()) == 64)) {
5691 Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex
5692 ? Intrinsic::arm_stlexd
5693 : Intrinsic::arm_strexd);
5694 llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty);
5695
5696 Address Tmp = CreateMemTemp(E->getArg(0)->getType());
5697 Value *Val = EmitScalarExpr(E->getArg(0));
5698 Builder.CreateStore(Val, Tmp);
5699
5700 Address LdPtr = Builder.CreateBitCast(Tmp,llvm::PointerType::getUnqual(STy));
5701 Val = Builder.CreateLoad(LdPtr);
5702
5703 Value *Arg0 = Builder.CreateExtractValue(Val, 0);
5704 Value *Arg1 = Builder.CreateExtractValue(Val, 1);
5705 Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), Int8PtrTy);
5706 return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "strexd");
5707 }
5708
5709 if (BuiltinID == ARM::BI__builtin_arm_strex ||
5710 BuiltinID == ARM::BI__builtin_arm_stlex) {
5711 Value *StoreVal = EmitScalarExpr(E->getArg(0));
5712 Value *StoreAddr = EmitScalarExpr(E->getArg(1));
5713
5714 QualType Ty = E->getArg(0)->getType();
5715 llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(),
5716 getContext().getTypeSize(Ty));
5717 StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo());
5718
5719 if (StoreVal->getType()->isPointerTy())
5720 StoreVal = Builder.CreatePtrToInt(StoreVal, Int32Ty);
5721 else {
5722 llvm::Type *IntTy = llvm::IntegerType::get(
5723 getLLVMContext(),
5724 CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType()));
5725 StoreVal = Builder.CreateBitCast(StoreVal, IntTy);
5726 StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int32Ty);
5727 }
5728
5729 Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex
5730 ? Intrinsic::arm_stlex
5731 : Intrinsic::arm_strex,
5732 StoreAddr->getType());
5733 return Builder.CreateCall(F, {StoreVal, StoreAddr}, "strex");
5734 }
5735
5736 switch (BuiltinID) {
5737 case ARM::BI__iso_volatile_load8:
5738 case ARM::BI__iso_volatile_load16:
5739 case ARM::BI__iso_volatile_load32:
5740 case ARM::BI__iso_volatile_load64:
5741 return EmitISOVolatileLoad(E);
5742 case ARM::BI__iso_volatile_store8:
5743 case ARM::BI__iso_volatile_store16:
5744 case ARM::BI__iso_volatile_store32:
5745 case ARM::BI__iso_volatile_store64:
5746 return EmitISOVolatileStore(E);
5747 }
5748
5749 if (BuiltinID == ARM::BI__builtin_arm_clrex) {
5750 Function *F = CGM.getIntrinsic(Intrinsic::arm_clrex);
5751 return Builder.CreateCall(F);
5752 }
5753
5754 // CRC32
5755 Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic;
5756 switch (BuiltinID) {
5757 case ARM::BI__builtin_arm_crc32b:
5758 CRCIntrinsicID = Intrinsic::arm_crc32b; break;
5759 case ARM::BI__builtin_arm_crc32cb:
5760 CRCIntrinsicID = Intrinsic::arm_crc32cb; break;
5761 case ARM::BI__builtin_arm_crc32h:
5762 CRCIntrinsicID = Intrinsic::arm_crc32h; break;
5763 case ARM::BI__builtin_arm_crc32ch:
5764 CRCIntrinsicID = Intrinsic::arm_crc32ch; break;
5765 case ARM::BI__builtin_arm_crc32w:
5766 case ARM::BI__builtin_arm_crc32d:
5767 CRCIntrinsicID = Intrinsic::arm_crc32w; break;
5768 case ARM::BI__builtin_arm_crc32cw:
5769 case ARM::BI__builtin_arm_crc32cd:
5770 CRCIntrinsicID = Intrinsic::arm_crc32cw; break;
5771 }
5772
5773 if (CRCIntrinsicID != Intrinsic::not_intrinsic) {
5774 Value *Arg0 = EmitScalarExpr(E->getArg(0));
5775 Value *Arg1 = EmitScalarExpr(E->getArg(1));
5776
5777 // crc32{c,}d intrinsics are implemnted as two calls to crc32{c,}w
5778 // intrinsics, hence we need different codegen for these cases.
5779 if (BuiltinID == ARM::BI__builtin_arm_crc32d ||
5780 BuiltinID == ARM::BI__builtin_arm_crc32cd) {
5781 Value *C1 = llvm::ConstantInt::get(Int64Ty, 32);
5782 Value *Arg1a = Builder.CreateTruncOrBitCast(Arg1, Int32Ty);
5783 Value *Arg1b = Builder.CreateLShr(Arg1, C1);
5784 Arg1b = Builder.CreateTruncOrBitCast(Arg1b, Int32Ty);
5785
5786 Function *F = CGM.getIntrinsic(CRCIntrinsicID);
5787 Value *Res = Builder.CreateCall(F, {Arg0, Arg1a});
5788 return Builder.CreateCall(F, {Res, Arg1b});
5789 } else {
5790 Arg1 = Builder.CreateZExtOrBitCast(Arg1, Int32Ty);
5791
5792 Function *F = CGM.getIntrinsic(CRCIntrinsicID);
5793 return Builder.CreateCall(F, {Arg0, Arg1});
5794 }
5795 }
5796
5797 if (BuiltinID == ARM::BI__builtin_arm_rsr ||
5798 BuiltinID == ARM::BI__builtin_arm_rsr64 ||
5799 BuiltinID == ARM::BI__builtin_arm_rsrp ||
5800 BuiltinID == ARM::BI__builtin_arm_wsr ||
5801 BuiltinID == ARM::BI__builtin_arm_wsr64 ||
5802 BuiltinID == ARM::BI__builtin_arm_wsrp) {
5803
5804 bool IsRead = BuiltinID == ARM::BI__builtin_arm_rsr ||
5805 BuiltinID == ARM::BI__builtin_arm_rsr64 ||
5806 BuiltinID == ARM::BI__builtin_arm_rsrp;
5807
5808 bool IsPointerBuiltin = BuiltinID == ARM::BI__builtin_arm_rsrp ||
5809 BuiltinID == ARM::BI__builtin_arm_wsrp;
5810
5811 bool Is64Bit = BuiltinID == ARM::BI__builtin_arm_rsr64 ||
5812 BuiltinID == ARM::BI__builtin_arm_wsr64;
5813
5814 llvm::Type *ValueType;
5815 llvm::Type *RegisterType;
5816 if (IsPointerBuiltin) {
5817 ValueType = VoidPtrTy;
5818 RegisterType = Int32Ty;
5819 } else if (Is64Bit) {
5820 ValueType = RegisterType = Int64Ty;
5821 } else {
5822 ValueType = RegisterType = Int32Ty;
5823 }
5824
5825 return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, IsRead);
5826 }
5827
5828 // Find out if any arguments are required to be integer constant
5829 // expressions.
5830 unsigned ICEArguments = 0;
5831 ASTContext::GetBuiltinTypeError Error;
5832 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
5833 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5833, __extension__ __PRETTY_FUNCTION__))
;
5834
5835 auto getAlignmentValue32 = [&](Address addr) -> Value* {
5836 return Builder.getInt32(addr.getAlignment().getQuantity());
5837 };
5838
5839 Address PtrOp0 = Address::invalid();
5840 Address PtrOp1 = Address::invalid();
5841 SmallVector<Value*, 4> Ops;
5842 bool HasExtraArg = HasExtraNeonArgument(BuiltinID);
5843 unsigned NumArgs = E->getNumArgs() - (HasExtraArg ? 1 : 0);
5844 for (unsigned i = 0, e = NumArgs; i != e; i++) {
5845 if (i == 0) {
5846 switch (BuiltinID) {
5847 case NEON::BI__builtin_neon_vld1_v:
5848 case NEON::BI__builtin_neon_vld1q_v:
5849 case NEON::BI__builtin_neon_vld1q_lane_v:
5850 case NEON::BI__builtin_neon_vld1_lane_v:
5851 case NEON::BI__builtin_neon_vld1_dup_v:
5852 case NEON::BI__builtin_neon_vld1q_dup_v:
5853 case NEON::BI__builtin_neon_vst1_v:
5854 case NEON::BI__builtin_neon_vst1q_v:
5855 case NEON::BI__builtin_neon_vst1q_lane_v:
5856 case NEON::BI__builtin_neon_vst1_lane_v:
5857 case NEON::BI__builtin_neon_vst2_v:
5858 case NEON::BI__builtin_neon_vst2q_v:
5859 case NEON::BI__builtin_neon_vst2_lane_v:
5860 case NEON::BI__builtin_neon_vst2q_lane_v:
5861 case NEON::BI__builtin_neon_vst3_v:
5862 case NEON::BI__builtin_neon_vst3q_v:
5863 case NEON::BI__builtin_neon_vst3_lane_v:
5864 case NEON::BI__builtin_neon_vst3q_lane_v:
5865 case NEON::BI__builtin_neon_vst4_v:
5866 case NEON::BI__builtin_neon_vst4q_v:
5867 case NEON::BI__builtin_neon_vst4_lane_v:
5868 case NEON::BI__builtin_neon_vst4q_lane_v:
5869 // Get the alignment for the argument in addition to the value;
5870 // we'll use it later.
5871 PtrOp0 = EmitPointerWithAlignment(E->getArg(0));
5872 Ops.push_back(PtrOp0.getPointer());
5873 continue;
5874 }
5875 }
5876 if (i == 1) {
5877 switch (BuiltinID) {
5878 case NEON::BI__builtin_neon_vld2_v:
5879 case NEON::BI__builtin_neon_vld2q_v:
5880 case NEON::BI__builtin_neon_vld3_v:
5881 case NEON::BI__builtin_neon_vld3q_v:
5882 case NEON::BI__builtin_neon_vld4_v:
5883 case NEON::BI__builtin_neon_vld4q_v:
5884 case NEON::BI__builtin_neon_vld2_lane_v:
5885 case NEON::BI__builtin_neon_vld2q_lane_v:
5886 case NEON::BI__builtin_neon_vld3_lane_v:
5887 case NEON::BI__builtin_neon_vld3q_lane_v:
5888 case NEON::BI__builtin_neon_vld4_lane_v:
5889 case NEON::BI__builtin_neon_vld4q_lane_v:
5890 case NEON::BI__builtin_neon_vld2_dup_v:
5891 case NEON::BI__builtin_neon_vld2q_dup_v:
5892 case NEON::BI__builtin_neon_vld3_dup_v:
5893 case NEON::BI__builtin_neon_vld3q_dup_v:
5894 case NEON::BI__builtin_neon_vld4_dup_v:
5895 case NEON::BI__builtin_neon_vld4q_dup_v:
5896 // Get the alignment for the argument in addition to the value;
5897 // we'll use it later.
5898 PtrOp1 = EmitPointerWithAlignment(E->getArg(1));
5899 Ops.push_back(PtrOp1.getPointer());
5900 continue;
5901 }
5902 }
5903
5904 if ((ICEArguments & (1 << i)) == 0) {
5905 Ops.push_back(EmitScalarExpr(E->getArg(i)));
5906 } else {
5907 // If this is required to be a constant, constant fold it so that we know
5908 // that the generated intrinsic gets a ConstantInt.
5909 llvm::APSInt Result;
5910 bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext());
5911 assert(IsConst && "Constant arg isn't actually constant?")(static_cast <bool> (IsConst && "Constant arg isn't actually constant?"
) ? void (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\""
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5911, __extension__ __PRETTY_FUNCTION__))
; (void)IsConst;
5912 Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result));
5913 }
5914 }
5915
5916 switch (BuiltinID) {
5917 default: break;
5918
5919 case NEON::BI__builtin_neon_vget_lane_i8:
5920 case NEON::BI__builtin_neon_vget_lane_i16:
5921 case NEON::BI__builtin_neon_vget_lane_i32:
5922 case NEON::BI__builtin_neon_vget_lane_i64:
5923 case NEON::BI__builtin_neon_vget_lane_f32:
5924 case NEON::BI__builtin_neon_vgetq_lane_i8:
5925 case NEON::BI__builtin_neon_vgetq_lane_i16:
5926 case NEON::BI__builtin_neon_vgetq_lane_i32:
5927 case NEON::BI__builtin_neon_vgetq_lane_i64:
5928 case NEON::BI__builtin_neon_vgetq_lane_f32:
5929 return Builder.CreateExtractElement(Ops[0], Ops[1], "vget_lane");
5930
5931 case NEON::BI__builtin_neon_vrndns_f32: {
5932 Value *Arg = EmitScalarExpr(E->getArg(0));
5933 llvm::Type *Tys[] = {Arg->getType()};
5934 Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vrintn, Tys);
5935 return Builder.CreateCall(F, {Arg}, "vrndn"); }
5936
5937 case NEON::BI__builtin_neon_vset_lane_i8:
5938 case NEON::BI__builtin_neon_vset_lane_i16:
5939 case NEON::BI__builtin_neon_vset_lane_i32:
5940 case NEON::BI__builtin_neon_vset_lane_i64:
5941 case NEON::BI__builtin_neon_vset_lane_f32:
5942 case NEON::BI__builtin_neon_vsetq_lane_i8:
5943 case NEON::BI__builtin_neon_vsetq_lane_i16:
5944 case NEON::BI__builtin_neon_vsetq_lane_i32:
5945 case NEON::BI__builtin_neon_vsetq_lane_i64:
5946 case NEON::BI__builtin_neon_vsetq_lane_f32:
5947 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
5948
5949 case NEON::BI__builtin_neon_vsha1h_u32:
5950 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1h), Ops,
5951 "vsha1h");
5952 case NEON::BI__builtin_neon_vsha1cq_u32:
5953 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1c), Ops,
5954 "vsha1h");
5955 case NEON::BI__builtin_neon_vsha1pq_u32:
5956 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1p), Ops,
5957 "vsha1h");
5958 case NEON::BI__builtin_neon_vsha1mq_u32:
5959 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1m), Ops,
5960 "vsha1h");
5961
5962 // The ARM _MoveToCoprocessor builtins put the input register value as
5963 // the first argument, but the LLVM intrinsic expects it as the third one.
5964 case ARM::BI_MoveToCoprocessor:
5965 case ARM::BI_MoveToCoprocessor2: {
5966 Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI_MoveToCoprocessor ?
5967 Intrinsic::arm_mcr : Intrinsic::arm_mcr2);
5968 return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0],
5969 Ops[3], Ops[4], Ops[5]});
5970 }
5971 case ARM::BI_BitScanForward:
5972 case ARM::BI_BitScanForward64:
5973 return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E);
5974 case ARM::BI_BitScanReverse:
5975 case ARM::BI_BitScanReverse64:
5976 return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E);
5977
5978 case ARM::BI_InterlockedAnd64:
5979 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E);
5980 case ARM::BI_InterlockedExchange64:
5981 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E);
5982 case ARM::BI_InterlockedExchangeAdd64:
5983 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E);
5984 case ARM::BI_InterlockedExchangeSub64:
5985 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E);
5986 case ARM::BI_InterlockedOr64:
5987 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E);
5988 case ARM::BI_InterlockedXor64:
5989 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E);
5990 case ARM::BI_InterlockedDecrement64:
5991 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E);
5992 case ARM::BI_InterlockedIncrement64:
5993 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E);
5994 }
5995
5996 // Get the last argument, which specifies the vector type.
5997 assert(HasExtraArg)(static_cast <bool> (HasExtraArg) ? void (0) : __assert_fail
("HasExtraArg", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 5997, __extension__ __PRETTY_FUNCTION__))
;
5998 llvm::APSInt Result;
5999 const Expr *Arg = E->getArg(E->getNumArgs()-1);
6000 if (!Arg->isIntegerConstantExpr(Result, getContext()))
6001 return nullptr;
6002
6003 if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f ||
6004 BuiltinID == ARM::BI__builtin_arm_vcvtr_d) {
6005 // Determine the overloaded type of this builtin.
6006 llvm::Type *Ty;
6007 if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f)
6008 Ty = FloatTy;
6009 else
6010 Ty = DoubleTy;
6011
6012 // Determine whether this is an unsigned conversion or not.
6013 bool usgn = Result.getZExtValue() == 1;
6014 unsigned Int = usgn ? Intrinsic::arm_vcvtru : Intrinsic::arm_vcvtr;
6015
6016 // Call the appropriate intrinsic.
6017 Function *F = CGM.getIntrinsic(Int, Ty);
6018 return Builder.CreateCall(F, Ops, "vcvtr");
6019 }
6020
6021 // Determine the type of this overloaded NEON intrinsic.
6022 NeonTypeFlags Type(Result.getZExtValue());
6023 bool usgn = Type.isUnsigned();
6024 bool rightShift = false;
6025
6026 llvm::VectorType *VTy = GetNeonType(this, Type,
6027 getTarget().hasLegalHalfType());
6028 llvm::Type *Ty = VTy;
6029 if (!Ty)
6030 return nullptr;
6031
6032 // Many NEON builtins have identical semantics and uses in ARM and
6033 // AArch64. Emit these in a single function.
6034 auto IntrinsicMap = makeArrayRef(ARMSIMDIntrinsicMap);
6035 const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap(
6036 IntrinsicMap, BuiltinID, NEONSIMDIntrinsicsProvenSorted);
6037 if (Builtin)
6038 return EmitCommonNeonBuiltinExpr(
6039 Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic,
6040 Builtin->NameHint, Builtin->TypeModifier, E, Ops, PtrOp0, PtrOp1, Arch);
6041
6042 unsigned Int;
6043 switch (BuiltinID) {
6044 default: return nullptr;
6045 case NEON::BI__builtin_neon_vld1q_lane_v:
6046 // Handle 64-bit integer elements as a special case. Use shuffles of
6047 // one-element vectors to avoid poor code for i64 in the backend.
6048 if (VTy->getElementType()->isIntegerTy(64)) {
6049 // Extract the other lane.
6050 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6051 uint32_t Lane = cast<ConstantInt>(Ops[2])->getZExtValue();
6052 Value *SV = llvm::ConstantVector::get(ConstantInt::get(Int32Ty, 1-Lane));
6053 Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV);
6054 // Load the value as a one-element vector.
6055 Ty = llvm::VectorType::get(VTy->getElementType(), 1);
6056 llvm::Type *Tys[] = {Ty, Int8PtrTy};
6057 Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vld1, Tys);
6058 Value *Align = getAlignmentValue32(PtrOp0);
6059 Value *Ld = Builder.CreateCall(F, {Ops[0], Align});
6060 // Combine them.
6061 uint32_t Indices[] = {1 - Lane, Lane};
6062 SV = llvm::ConstantDataVector::get(getLLVMContext(), Indices);
6063 return Builder.CreateShuffleVector(Ops[1], Ld, SV, "vld1q_lane");
6064 }
6065 LLVM_FALLTHROUGH[[clang::fallthrough]];
6066 case NEON::BI__builtin_neon_vld1_lane_v: {
6067 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6068 PtrOp0 = Builder.CreateElementBitCast(PtrOp0, VTy->getElementType());
6069 Value *Ld = Builder.CreateLoad(PtrOp0);
6070 return Builder.CreateInsertElement(Ops[1], Ld, Ops[2], "vld1_lane");
6071 }
6072 case NEON::BI__builtin_neon_vqrshrn_n_v:
6073 Int =
6074 usgn ? Intrinsic::arm_neon_vqrshiftnu : Intrinsic::arm_neon_vqrshiftns;
6075 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n",
6076 1, true);
6077 case NEON::BI__builtin_neon_vqrshrun_n_v:
6078 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqrshiftnsu, Ty),
6079 Ops, "vqrshrun_n", 1, true);
6080 case NEON::BI__builtin_neon_vqshrn_n_v:
6081 Int = usgn ? Intrinsic::arm_neon_vqshiftnu : Intrinsic::arm_neon_vqshiftns;
6082 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n",
6083 1, true);
6084 case NEON::BI__builtin_neon_vqshrun_n_v:
6085 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqshiftnsu, Ty),
6086 Ops, "vqshrun_n", 1, true);
6087 case NEON::BI__builtin_neon_vrecpe_v:
6088 case NEON::BI__builtin_neon_vrecpeq_v:
6089 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrecpe, Ty),
6090 Ops, "vrecpe");
6091 case NEON::BI__builtin_neon_vrshrn_n_v:
6092 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrshiftn, Ty),
6093 Ops, "vrshrn_n", 1, true);
6094 case NEON::BI__builtin_neon_vrsra_n_v:
6095 case NEON::BI__builtin_neon_vrsraq_n_v:
6096 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6097 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6098 Ops[2] = EmitNeonShiftVector(Ops[2], Ty, true);
6099 Int = usgn ? Intrinsic::arm_neon_vrshiftu : Intrinsic::arm_neon_vrshifts;
6100 Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Ty), {Ops[1], Ops[2]});
6101 return Builder.CreateAdd(Ops[0], Ops[1], "vrsra_n");
6102 case NEON::BI__builtin_neon_vsri_n_v:
6103 case NEON::BI__builtin_neon_vsriq_n_v:
6104 rightShift = true;
6105 LLVM_FALLTHROUGH[[clang::fallthrough]];
6106 case NEON::BI__builtin_neon_vsli_n_v:
6107 case NEON::BI__builtin_neon_vsliq_n_v:
6108 Ops[2] = EmitNeonShiftVector(Ops[2], Ty, rightShift);
6109 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vshiftins, Ty),
6110 Ops, "vsli_n");
6111 case NEON::BI__builtin_neon_vsra_n_v:
6112 case NEON::BI__builtin_neon_vsraq_n_v:
6113 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6114 Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n");
6115 return Builder.CreateAdd(Ops[0], Ops[1]);
6116 case NEON::BI__builtin_neon_vst1q_lane_v:
6117 // Handle 64-bit integer elements as a special case. Use a shuffle to get
6118 // a one-element vector and avoid poor code for i64 in the backend.
6119 if (VTy->getElementType()->isIntegerTy(64)) {
6120 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6121 Value *SV = llvm::ConstantVector::get(cast<llvm::Constant>(Ops[2]));
6122 Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV);
6123 Ops[2] = getAlignmentValue32(PtrOp0);
6124 llvm::Type *Tys[] = {Int8PtrTy, Ops[1]->getType()};
6125 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_neon_vst1,
6126 Tys), Ops);
6127 }
6128 LLVM_FALLTHROUGH[[clang::fallthrough]];
6129 case NEON::BI__builtin_neon_vst1_lane_v: {
6130 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6131 Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]);
6132 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
6133 auto St = Builder.CreateStore(Ops[1], Builder.CreateBitCast(PtrOp0, Ty));
6134 return St;
6135 }
6136 case NEON::BI__builtin_neon_vtbl1_v:
6137 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl1),
6138 Ops, "vtbl1");
6139 case NEON::BI__builtin_neon_vtbl2_v:
6140 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl2),
6141 Ops, "vtbl2");
6142 case NEON::BI__builtin_neon_vtbl3_v:
6143 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl3),
6144 Ops, "vtbl3");
6145 case NEON::BI__builtin_neon_vtbl4_v:
6146 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl4),
6147 Ops, "vtbl4");
6148 case NEON::BI__builtin_neon_vtbx1_v:
6149 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx1),
6150 Ops, "vtbx1");
6151 case NEON::BI__builtin_neon_vtbx2_v:
6152 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx2),
6153 Ops, "vtbx2");
6154 case NEON::BI__builtin_neon_vtbx3_v:
6155 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx3),
6156 Ops, "vtbx3");
6157 case NEON::BI__builtin_neon_vtbx4_v:
6158 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx4),
6159 Ops, "vtbx4");
6160 }
6161}
6162
6163static Value *EmitAArch64TblBuiltinExpr(CodeGenFunction &CGF, unsigned BuiltinID,
6164 const CallExpr *E,
6165 SmallVectorImpl<Value *> &Ops,
6166 llvm::Triple::ArchType Arch) {
6167 unsigned int Int = 0;
6168 const char *s = nullptr;
6169
6170 switch (BuiltinID) {
6171 default:
6172 return nullptr;
6173 case NEON::BI__builtin_neon_vtbl1_v:
6174 case NEON::BI__builtin_neon_vqtbl1_v:
6175 case NEON::BI__builtin_neon_vqtbl1q_v:
6176 case NEON::BI__builtin_neon_vtbl2_v:
6177 case NEON::BI__builtin_neon_vqtbl2_v:
6178 case NEON::BI__builtin_neon_vqtbl2q_v:
6179 case NEON::BI__builtin_neon_vtbl3_v:
6180 case NEON::BI__builtin_neon_vqtbl3_v:
6181 case NEON::BI__builtin_neon_vqtbl3q_v:
6182 case NEON::BI__builtin_neon_vtbl4_v:
6183 case NEON::BI__builtin_neon_vqtbl4_v:
6184 case NEON::BI__builtin_neon_vqtbl4q_v:
6185 break;
6186 case NEON::BI__builtin_neon_vtbx1_v:
6187 case NEON::BI__builtin_neon_vqtbx1_v:
6188 case NEON::BI__builtin_neon_vqtbx1q_v:
6189 case NEON::BI__builtin_neon_vtbx2_v:
6190 case NEON::BI__builtin_neon_vqtbx2_v:
6191 case NEON::BI__builtin_neon_vqtbx2q_v:
6192 case NEON::BI__builtin_neon_vtbx3_v:
6193 case NEON::BI__builtin_neon_vqtbx3_v:
6194 case NEON::BI__builtin_neon_vqtbx3q_v:
6195 case NEON::BI__builtin_neon_vtbx4_v:
6196 case NEON::BI__builtin_neon_vqtbx4_v:
6197 case NEON::BI__builtin_neon_vqtbx4q_v:
6198 break;
6199 }
6200
6201 assert(E->getNumArgs() >= 3)(static_cast <bool> (E->getNumArgs() >= 3) ? void
(0) : __assert_fail ("E->getNumArgs() >= 3", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6201, __extension__ __PRETTY_FUNCTION__))
;
6202
6203 // Get the last argument, which specifies the vector type.
6204 llvm::APSInt Result;
6205 const Expr *Arg = E->getArg(E->getNumArgs() - 1);
6206 if (!Arg->isIntegerConstantExpr(Result, CGF.getContext()))
6207 return nullptr;
6208
6209 // Determine the type of this overloaded NEON intrinsic.
6210 NeonTypeFlags Type(Result.getZExtValue());
6211 llvm::VectorType *Ty = GetNeonType(&CGF, Type);
6212 if (!Ty)
6213 return nullptr;
6214
6215 CodeGen::CGBuilderTy &Builder = CGF.Builder;
6216
6217 // AArch64 scalar builtins are not overloaded, they do not have an extra
6218 // argument that specifies the vector type, need to handle each case.
6219 switch (BuiltinID) {
6220 case NEON::BI__builtin_neon_vtbl1_v: {
6221 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 1), nullptr,
6222 Ops[1], Ty, Intrinsic::aarch64_neon_tbl1,
6223 "vtbl1");
6224 }
6225 case NEON::BI__builtin_neon_vtbl2_v: {
6226 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 2), nullptr,
6227 Ops[2], Ty, Intrinsic::aarch64_neon_tbl1,
6228 "vtbl1");
6229 }
6230 case NEON::BI__builtin_neon_vtbl3_v: {
6231 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 3), nullptr,
6232 Ops[3], Ty, Intrinsic::aarch64_neon_tbl2,
6233 "vtbl2");
6234 }
6235 case NEON::BI__builtin_neon_vtbl4_v: {
6236 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 4), nullptr,
6237 Ops[4], Ty, Intrinsic::aarch64_neon_tbl2,
6238 "vtbl2");
6239 }
6240 case NEON::BI__builtin_neon_vtbx1_v: {
6241 Value *TblRes =
6242 packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 1), nullptr, Ops[2],
6243 Ty, Intrinsic::aarch64_neon_tbl1, "vtbl1");
6244
6245 llvm::Constant *EightV = ConstantInt::get(Ty, 8);
6246 Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[2], EightV);
6247 CmpRes = Builder.CreateSExt(CmpRes, Ty);
6248
6249 Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]);
6250 Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes);
6251 return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx");
6252 }
6253 case NEON::BI__builtin_neon_vtbx2_v: {
6254 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 2), Ops[0],
6255 Ops[3], Ty, Intrinsic::aarch64_neon_tbx1,
6256 "vtbx1");
6257 }
6258 case NEON::BI__builtin_neon_vtbx3_v: {
6259 Value *TblRes =
6260 packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 3), nullptr, Ops[4],
6261 Ty, Intrinsic::aarch64_neon_tbl2, "vtbl2");
6262
6263 llvm::Constant *TwentyFourV = ConstantInt::get(Ty, 24);
6264 Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[4],
6265 TwentyFourV);
6266 CmpRes = Builder.CreateSExt(CmpRes, Ty);
6267
6268 Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]);
6269 Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes);
6270 return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx");
6271 }
6272 case NEON::BI__builtin_neon_vtbx4_v: {
6273 return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 4), Ops[0],
6274 Ops[5], Ty, Intrinsic::aarch64_neon_tbx2,
6275 "vtbx2");
6276 }
6277 case NEON::BI__builtin_neon_vqtbl1_v:
6278 case NEON::BI__builtin_neon_vqtbl1q_v:
6279 Int = Intrinsic::aarch64_neon_tbl1; s = "vtbl1"; break;
6280 case NEON::BI__builtin_neon_vqtbl2_v:
6281 case NEON::BI__builtin_neon_vqtbl2q_v: {
6282 Int = Intrinsic::aarch64_neon_tbl2; s = "vtbl2"; break;
6283 case NEON::BI__builtin_neon_vqtbl3_v:
6284 case NEON::BI__builtin_neon_vqtbl3q_v:
6285 Int = Intrinsic::aarch64_neon_tbl3; s = "vtbl3"; break;
6286 case NEON::BI__builtin_neon_vqtbl4_v:
6287 case NEON::BI__builtin_neon_vqtbl4q_v:
6288 Int = Intrinsic::aarch64_neon_tbl4; s = "vtbl4"; break;
6289 case NEON::BI__builtin_neon_vqtbx1_v:
6290 case NEON::BI__builtin_neon_vqtbx1q_v:
6291 Int = Intrinsic::aarch64_neon_tbx1; s = "vtbx1"; break;
6292 case NEON::BI__builtin_neon_vqtbx2_v:
6293 case NEON::BI__builtin_neon_vqtbx2q_v:
6294 Int = Intrinsic::aarch64_neon_tbx2; s = "vtbx2"; break;
6295 case NEON::BI__builtin_neon_vqtbx3_v:
6296 case NEON::BI__builtin_neon_vqtbx3q_v:
6297 Int = Intrinsic::aarch64_neon_tbx3; s = "vtbx3"; break;
6298 case NEON::BI__builtin_neon_vqtbx4_v:
6299 case NEON::BI__builtin_neon_vqtbx4q_v:
6300 Int = Intrinsic::aarch64_neon_tbx4; s = "vtbx4"; break;
6301 }
6302 }
6303
6304 if (!Int)
6305 return nullptr;
6306
6307 Function *F = CGF.CGM.getIntrinsic(Int, Ty);
6308 return CGF.EmitNeonCall(F, Ops, s);
6309}
6310
6311Value *CodeGenFunction::vectorWrapScalar16(Value *Op) {
6312 llvm::Type *VTy = llvm::VectorType::get(Int16Ty, 4);
6313 Op = Builder.CreateBitCast(Op, Int16Ty);
6314 Value *V = UndefValue::get(VTy);
6315 llvm::Constant *CI = ConstantInt::get(SizeTy, 0);
6316 Op = Builder.CreateInsertElement(V, Op, CI);
6317 return Op;
6318}
6319
6320Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
6321 const CallExpr *E,
6322 llvm::Triple::ArchType Arch) {
6323 unsigned HintID = static_cast<unsigned>(-1);
6324 switch (BuiltinID) {
6325 default: break;
6326 case AArch64::BI__builtin_arm_nop:
6327 HintID = 0;
6328 break;
6329 case AArch64::BI__builtin_arm_yield:
6330 case AArch64::BI__yield:
6331 HintID = 1;
6332 break;
6333 case AArch64::BI__builtin_arm_wfe:
6334 case AArch64::BI__wfe:
6335 HintID = 2;
6336 break;
6337 case AArch64::BI__builtin_arm_wfi:
6338 case AArch64::BI__wfi:
6339 HintID = 3;
6340 break;
6341 case AArch64::BI__builtin_arm_sev:
6342 case AArch64::BI__sev:
6343 HintID = 4;
6344 break;
6345 case AArch64::BI__builtin_arm_sevl:
6346 case AArch64::BI__sevl:
6347 HintID = 5;
6348 break;
6349 }
6350
6351 if (HintID != static_cast<unsigned>(-1)) {
6352 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_hint);
6353 return Builder.CreateCall(F, llvm::ConstantInt::get(Int32Ty, HintID));
6354 }
6355
6356 if (BuiltinID == AArch64::BI__builtin_arm_prefetch) {
6357 Value *Address = EmitScalarExpr(E->getArg(0));
6358 Value *RW = EmitScalarExpr(E->getArg(1));
6359 Value *CacheLevel = EmitScalarExpr(E->getArg(2));
6360 Value *RetentionPolicy = EmitScalarExpr(E->getArg(3));
6361 Value *IsData = EmitScalarExpr(E->getArg(4));
6362
6363 Value *Locality = nullptr;
6364 if (cast<llvm::ConstantInt>(RetentionPolicy)->isZero()) {
6365 // Temporal fetch, needs to convert cache level to locality.
6366 Locality = llvm::ConstantInt::get(Int32Ty,
6367 -cast<llvm::ConstantInt>(CacheLevel)->getValue() + 3);
6368 } else {
6369 // Streaming fetch.
6370 Locality = llvm::ConstantInt::get(Int32Ty, 0);
6371 }
6372
6373 // FIXME: We need AArch64 specific LLVM intrinsic if we want to specify
6374 // PLDL3STRM or PLDL2STRM.
6375 Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
6376 return Builder.CreateCall(F, {Address, RW, Locality, IsData});
6377 }
6378
6379 if (BuiltinID == AArch64::BI__builtin_arm_rbit) {
6380 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6381, __extension__ __PRETTY_FUNCTION__))
6381 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6381, __extension__ __PRETTY_FUNCTION__))
;
6382 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
6383 return Builder.CreateCall(
6384 CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
6385 }
6386 if (BuiltinID == AArch64::BI__builtin_arm_rbit64) {
6387 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6388, __extension__ __PRETTY_FUNCTION__))
6388 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6388, __extension__ __PRETTY_FUNCTION__))
;
6389 llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
6390 return Builder.CreateCall(
6391 CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
6392 }
6393
6394 if (BuiltinID == AArch64::BI__clear_cache) {
6395 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6395, __extension__ __PRETTY_FUNCTION__))
;
6396 const FunctionDecl *FD = E->getDirectCallee();
6397 Value *Ops[2];
6398 for (unsigned i = 0; i < 2; i++)
6399 Ops[i] = EmitScalarExpr(E->getArg(i));
6400 llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType());
6401 llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty);
6402 StringRef Name = FD->getName();
6403 return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops);
6404 }
6405
6406 if ((BuiltinID == AArch64::BI__builtin_arm_ldrex ||
6407 BuiltinID == AArch64::BI__builtin_arm_ldaex) &&
6408 getContext().getTypeSize(E->getType()) == 128) {
6409 Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex
6410 ? Intrinsic::aarch64_ldaxp
6411 : Intrinsic::aarch64_ldxp);
6412
6413 Value *LdPtr = EmitScalarExpr(E->getArg(0));
6414 Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy),
6415 "ldxp");
6416
6417 Value *Val0 = Builder.CreateExtractValue(Val, 1);
6418 Value *Val1 = Builder.CreateExtractValue(Val, 0);
6419 llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128);
6420 Val0 = Builder.CreateZExt(Val0, Int128Ty);
6421 Val1 = Builder.CreateZExt(Val1, Int128Ty);
6422
6423 Value *ShiftCst = llvm::ConstantInt::get(Int128Ty, 64);
6424 Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */);
6425 Val = Builder.CreateOr(Val, Val1);
6426 return Builder.CreateBitCast(Val, ConvertType(E->getType()));
6427 } else if (BuiltinID == AArch64::BI__builtin_arm_ldrex ||
6428 BuiltinID == AArch64::BI__builtin_arm_ldaex) {
6429 Value *LoadAddr = EmitScalarExpr(E->getArg(0));
6430
6431 QualType Ty = E->getType();
6432 llvm::Type *RealResTy = ConvertType(Ty);
6433 llvm::Type *PtrTy = llvm::IntegerType::get(
6434 getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo();
6435 LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy);
6436
6437 Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex
6438 ? Intrinsic::aarch64_ldaxr
6439 : Intrinsic::aarch64_ldxr,
6440 PtrTy);
6441 Value *Val = Builder.CreateCall(F, LoadAddr, "ldxr");
6442
6443 if (RealResTy->isPointerTy())
6444 return Builder.CreateIntToPtr(Val, RealResTy);
6445
6446 llvm::Type *IntResTy = llvm::IntegerType::get(
6447 getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy));
6448 Val = Builder.CreateTruncOrBitCast(Val, IntResTy);
6449 return Builder.CreateBitCast(Val, RealResTy);
6450 }
6451
6452 if ((BuiltinID == AArch64::BI__builtin_arm_strex ||
6453 BuiltinID == AArch64::BI__builtin_arm_stlex) &&
6454 getContext().getTypeSize(E->getArg(0)->getType()) == 128) {
6455 Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex
6456 ? Intrinsic::aarch64_stlxp
6457 : Intrinsic::aarch64_stxp);
6458 llvm::Type *STy = llvm::StructType::get(Int64Ty, Int64Ty);
6459
6460 Address Tmp = CreateMemTemp(E->getArg(0)->getType());
6461 EmitAnyExprToMem(E->getArg(0), Tmp, Qualifiers(), /*init*/ true);
6462
6463 Tmp = Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(STy));
6464 llvm::Value *Val = Builder.CreateLoad(Tmp);
6465
6466 Value *Arg0 = Builder.CreateExtractValue(Val, 0);
6467 Value *Arg1 = Builder.CreateExtractValue(Val, 1);
6468 Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)),
6469 Int8PtrTy);
6470 return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "stxp");
6471 }
6472
6473 if (BuiltinID == AArch64::BI__builtin_arm_strex ||
6474 BuiltinID == AArch64::BI__builtin_arm_stlex) {
6475 Value *StoreVal = EmitScalarExpr(E->getArg(0));
6476 Value *StoreAddr = EmitScalarExpr(E->getArg(1));
6477
6478 QualType Ty = E->getArg(0)->getType();
6479 llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(),
6480 getContext().getTypeSize(Ty));
6481 StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo());
6482
6483 if (StoreVal->getType()->isPointerTy())
6484 StoreVal = Builder.CreatePtrToInt(StoreVal, Int64Ty);
6485 else {
6486 llvm::Type *IntTy = llvm::IntegerType::get(
6487 getLLVMContext(),
6488 CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType()));
6489 StoreVal = Builder.CreateBitCast(StoreVal, IntTy);
6490 StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int64Ty);
6491 }
6492
6493 Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex
6494 ? Intrinsic::aarch64_stlxr
6495 : Intrinsic::aarch64_stxr,
6496 StoreAddr->getType());
6497 return Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr");
6498 }
6499
6500 if (BuiltinID == AArch64::BI__builtin_arm_clrex) {
6501 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_clrex);
6502 return Builder.CreateCall(F);
6503 }
6504
6505 // CRC32
6506 Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic;
6507 switch (BuiltinID) {
6508 case AArch64::BI__builtin_arm_crc32b:
6509 CRCIntrinsicID = Intrinsic::aarch64_crc32b; break;
6510 case AArch64::BI__builtin_arm_crc32cb:
6511 CRCIntrinsicID = Intrinsic::aarch64_crc32cb; break;
6512 case AArch64::BI__builtin_arm_crc32h:
6513 CRCIntrinsicID = Intrinsic::aarch64_crc32h; break;
6514 case AArch64::BI__builtin_arm_crc32ch:
6515 CRCIntrinsicID = Intrinsic::aarch64_crc32ch; break;
6516 case AArch64::BI__builtin_arm_crc32w:
6517 CRCIntrinsicID = Intrinsic::aarch64_crc32w; break;
6518 case AArch64::BI__builtin_arm_crc32cw:
6519 CRCIntrinsicID = Intrinsic::aarch64_crc32cw; break;
6520 case AArch64::BI__builtin_arm_crc32d:
6521 CRCIntrinsicID = Intrinsic::aarch64_crc32x; break;
6522 case AArch64::BI__builtin_arm_crc32cd:
6523 CRCIntrinsicID = Intrinsic::aarch64_crc32cx; break;
6524 }
6525
6526 if (CRCIntrinsicID != Intrinsic::not_intrinsic) {
6527 Value *Arg0 = EmitScalarExpr(E->getArg(0));
6528 Value *Arg1 = EmitScalarExpr(E->getArg(1));
6529 Function *F = CGM.getIntrinsic(CRCIntrinsicID);
6530
6531 llvm::Type *DataTy = F->getFunctionType()->getParamType(1);
6532 Arg1 = Builder.CreateZExtOrBitCast(Arg1, DataTy);
6533
6534 return Builder.CreateCall(F, {Arg0, Arg1});
6535 }
6536
6537 if (BuiltinID == AArch64::BI__builtin_arm_rsr ||
6538 BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
6539 BuiltinID == AArch64::BI__builtin_arm_rsrp ||
6540 BuiltinID == AArch64::BI__builtin_arm_wsr ||
6541 BuiltinID == AArch64::BI__builtin_arm_wsr64 ||
6542 BuiltinID == AArch64::BI__builtin_arm_wsrp) {
6543
6544 bool IsRead = BuiltinID == AArch64::BI__builtin_arm_rsr ||
6545 BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
6546 BuiltinID == AArch64::BI__builtin_arm_rsrp;
6547
6548 bool IsPointerBuiltin = BuiltinID == AArch64::BI__builtin_arm_rsrp ||
6549 BuiltinID == AArch64::BI__builtin_arm_wsrp;
6550
6551 bool Is64Bit = BuiltinID != AArch64::BI__builtin_arm_rsr &&
6552 BuiltinID != AArch64::BI__builtin_arm_wsr;
6553
6554 llvm::Type *ValueType;
6555 llvm::Type *RegisterType = Int64Ty;
6556 if (IsPointerBuiltin) {
6557 ValueType = VoidPtrTy;
6558 } else if (Is64Bit) {
6559 ValueType = Int64Ty;
6560 } else {
6561 ValueType = Int32Ty;
6562 }
6563
6564 return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, IsRead);
6565 }
6566
6567 // Find out if any arguments are required to be integer constant
6568 // expressions.
6569 unsigned ICEArguments = 0;
6570 ASTContext::GetBuiltinTypeError Error;
6571 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
6572 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6572, __extension__ __PRETTY_FUNCTION__))
;
6573
6574 llvm::SmallVector<Value*, 4> Ops;
6575 for (unsigned i = 0, e = E->getNumArgs() - 1; i != e; i++) {
6576 if ((ICEArguments & (1 << i)) == 0) {
6577 Ops.push_back(EmitScalarExpr(E->getArg(i)));
6578 } else {
6579 // If this is required to be a constant, constant fold it so that we know
6580 // that the generated intrinsic gets a ConstantInt.
6581 llvm::APSInt Result;
6582 bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext());
6583 assert(IsConst && "Constant arg isn't actually constant?")(static_cast <bool> (IsConst && "Constant arg isn't actually constant?"
) ? void (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\""
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6583, __extension__ __PRETTY_FUNCTION__))
;
6584 (void)IsConst;
6585 Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result));
6586 }
6587 }
6588
6589 auto SISDMap = makeArrayRef(AArch64SISDIntrinsicMap);
6590 const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap(
6591 SISDMap, BuiltinID, AArch64SISDIntrinsicsProvenSorted);
6592
6593 if (Builtin) {
6594 Ops.push_back(EmitScalarExpr(E->getArg(E->getNumArgs() - 1)));
6595 Value *Result = EmitCommonNeonSISDBuiltinExpr(*this, *Builtin, Ops, E);
6596 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6596, __extension__ __PRETTY_FUNCTION__))
;
6597 return Result;
6598 }
6599
6600 llvm::APSInt Result;
6601 const Expr *Arg = E->getArg(E->getNumArgs()-1);
6602 NeonTypeFlags Type(0);
6603 if (Arg->isIntegerConstantExpr(Result, getContext()))
6604 // Determine the type of this overloaded NEON intrinsic.
6605 Type = NeonTypeFlags(Result.getZExtValue());
6606
6607 bool usgn = Type.isUnsigned();
6608 bool quad = Type.isQuad();
6609
6610 // Handle non-overloaded intrinsics first.
6611 switch (BuiltinID) {
6612 default: break;
6613 case NEON::BI__builtin_neon_vabsh_f16:
6614 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6615 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, HalfTy), Ops, "vabs");
6616 case NEON::BI__builtin_neon_vldrq_p128: {
6617 llvm::Type *Int128Ty = llvm::Type::getIntNTy(getLLVMContext(), 128);
6618 llvm::Type *Int128PTy = llvm::PointerType::get(Int128Ty, 0);
6619 Value *Ptr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int128PTy);
6620 return Builder.CreateAlignedLoad(Int128Ty, Ptr,
6621 CharUnits::fromQuantity(16));
6622 }
6623 case NEON::BI__builtin_neon_vstrq_p128: {
6624 llvm::Type *Int128PTy = llvm::Type::getIntNPtrTy(getLLVMContext(), 128);
6625 Value *Ptr = Builder.CreateBitCast(Ops[0], Int128PTy);
6626 return Builder.CreateDefaultAlignedStore(EmitScalarExpr(E->getArg(1)), Ptr);
6627 }
6628 case NEON::BI__builtin_neon_vcvts_u32_f32:
6629 case NEON::BI__builtin_neon_vcvtd_u64_f64:
6630 usgn = true;
6631 LLVM_FALLTHROUGH[[clang::fallthrough]];
6632 case NEON::BI__builtin_neon_vcvts_s32_f32:
6633 case NEON::BI__builtin_neon_vcvtd_s64_f64: {
6634 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6635 bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64;
6636 llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty;
6637 llvm::Type *FTy = Is64 ? DoubleTy : FloatTy;
6638 Ops[0] = Builder.CreateBitCast(Ops[0], FTy);
6639 if (usgn)
6640 return Builder.CreateFPToUI(Ops[0], InTy);
6641 return Builder.CreateFPToSI(Ops[0], InTy);
6642 }
6643 case NEON::BI__builtin_neon_vcvts_f32_u32:
6644 case NEON::BI__builtin_neon_vcvtd_f64_u64:
6645 usgn = true;
6646 LLVM_FALLTHROUGH[[clang::fallthrough]];
6647 case NEON::BI__builtin_neon_vcvts_f32_s32:
6648 case NEON::BI__builtin_neon_vcvtd_f64_s64: {
6649 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6650 bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64;
6651 llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty;
6652 llvm::Type *FTy = Is64 ? DoubleTy : FloatTy;
6653 Ops[0] = Builder.CreateBitCast(Ops[0], InTy);
6654 if (usgn)
6655 return Builder.CreateUIToFP(Ops[0], FTy);
6656 return Builder.CreateSIToFP(Ops[0], FTy);
6657 }
6658 case NEON::BI__builtin_neon_vcvth_f16_u16:
6659 case NEON::BI__builtin_neon_vcvth_f16_u32:
6660 case NEON::BI__builtin_neon_vcvth_f16_u64:
6661 usgn = true;
6662 // FALL THROUGH
6663 case NEON::BI__builtin_neon_vcvth_f16_s16:
6664 case NEON::BI__builtin_neon_vcvth_f16_s32:
6665 case NEON::BI__builtin_neon_vcvth_f16_s64: {
6666 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6667 llvm::Type *FTy = HalfTy;
6668 llvm::Type *InTy;
6669 if (Ops[0]->getType()->getPrimitiveSizeInBits() == 64)
6670 InTy = Int64Ty;
6671 else if (Ops[0]->getType()->getPrimitiveSizeInBits() == 32)
6672 InTy = Int32Ty;
6673 else
6674 InTy = Int16Ty;
6675 Ops[0] = Builder.CreateBitCast(Ops[0], InTy);
6676 if (usgn)
6677 return Builder.CreateUIToFP(Ops[0], FTy);
6678 return Builder.CreateSIToFP(Ops[0], FTy);
6679 }
6680 case NEON::BI__builtin_neon_vcvth_u16_f16:
6681 usgn = true;
6682 // FALL THROUGH
6683 case NEON::BI__builtin_neon_vcvth_s16_f16: {
6684 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6685 Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
6686 if (usgn)
6687 return Builder.CreateFPToUI(Ops[0], Int16Ty);
6688 return Builder.CreateFPToSI(Ops[0], Int16Ty);
6689 }
6690 case NEON::BI__builtin_neon_vcvth_u32_f16:
6691 usgn = true;
6692 // FALL THROUGH
6693 case NEON::BI__builtin_neon_vcvth_s32_f16: {
6694 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6695 Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
6696 if (usgn)
6697 return Builder.CreateFPToUI(Ops[0], Int32Ty);
6698 return Builder.CreateFPToSI(Ops[0], Int32Ty);
6699 }
6700 case NEON::BI__builtin_neon_vcvth_u64_f16:
6701 usgn = true;
6702 // FALL THROUGH
6703 case NEON::BI__builtin_neon_vcvth_s64_f16: {
6704 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6705 Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
6706 if (usgn)
6707 return Builder.CreateFPToUI(Ops[0], Int64Ty);
6708 return Builder.CreateFPToSI(Ops[0], Int64Ty);
6709 }
6710 case NEON::BI__builtin_neon_vcvtah_u16_f16:
6711 case NEON::BI__builtin_neon_vcvtmh_u16_f16:
6712 case NEON::BI__builtin_neon_vcvtnh_u16_f16:
6713 case NEON::BI__builtin_neon_vcvtph_u16_f16:
6714 case NEON::BI__builtin_neon_vcvtah_s16_f16:
6715 case NEON::BI__builtin_neon_vcvtmh_s16_f16:
6716 case NEON::BI__builtin_neon_vcvtnh_s16_f16:
6717 case NEON::BI__builtin_neon_vcvtph_s16_f16: {
6718 unsigned Int;
6719 llvm::Type* InTy = Int32Ty;
6720 llvm::Type* FTy = HalfTy;
6721 llvm::Type *Tys[2] = {InTy, FTy};
6722 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6723 switch (BuiltinID) {
6724 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6724)
;
6725 case NEON::BI__builtin_neon_vcvtah_u16_f16:
6726 Int = Intrinsic::aarch64_neon_fcvtau; break;
6727 case NEON::BI__builtin_neon_vcvtmh_u16_f16:
6728 Int = Intrinsic::aarch64_neon_fcvtmu; break;
6729 case NEON::BI__builtin_neon_vcvtnh_u16_f16:
6730 Int = Intrinsic::aarch64_neon_fcvtnu; break;
6731 case NEON::BI__builtin_neon_vcvtph_u16_f16:
6732 Int = Intrinsic::aarch64_neon_fcvtpu; break;
6733 case NEON::BI__builtin_neon_vcvtah_s16_f16:
6734 Int = Intrinsic::aarch64_neon_fcvtas; break;
6735 case NEON::BI__builtin_neon_vcvtmh_s16_f16:
6736 Int = Intrinsic::aarch64_neon_fcvtms; break;
6737 case NEON::BI__builtin_neon_vcvtnh_s16_f16:
6738 Int = Intrinsic::aarch64_neon_fcvtns; break;
6739 case NEON::BI__builtin_neon_vcvtph_s16_f16:
6740 Int = Intrinsic::aarch64_neon_fcvtps; break;
6741 }
6742 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvt");
6743 return Builder.CreateTrunc(Ops[0], Int16Ty);
6744 }
6745 case NEON::BI__builtin_neon_vcaleh_f16:
6746 case NEON::BI__builtin_neon_vcalth_f16:
6747 case NEON::BI__builtin_neon_vcageh_f16:
6748 case NEON::BI__builtin_neon_vcagth_f16: {
6749 unsigned Int;
6750 llvm::Type* InTy = Int32Ty;
6751 llvm::Type* FTy = HalfTy;
6752 llvm::Type *Tys[2] = {InTy, FTy};
6753 Ops.push_back(EmitScalarExpr(E->getArg(1)));
6754 switch (BuiltinID) {
6755 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6755)
;
6756 case NEON::BI__builtin_neon_vcageh_f16:
6757 Int = Intrinsic::aarch64_neon_facge; break;
6758 case NEON::BI__builtin_neon_vcagth_f16:
6759 Int = Intrinsic::aarch64_neon_facgt; break;
6760 case NEON::BI__builtin_neon_vcaleh_f16:
6761 Int = Intrinsic::aarch64_neon_facge; std::swap(Ops[0], Ops[1]); break;
6762 case NEON::BI__builtin_neon_vcalth_f16:
6763 Int = Intrinsic::aarch64_neon_facgt; std::swap(Ops[0], Ops[1]); break;
6764 }
6765 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "facg");
6766 return Builder.CreateTrunc(Ops[0], Int16Ty);
6767 }
6768 case NEON::BI__builtin_neon_vcvth_n_s16_f16:
6769 case NEON::BI__builtin_neon_vcvth_n_u16_f16: {
6770 unsigned Int;
6771 llvm::Type* InTy = Int32Ty;
6772 llvm::Type* FTy = HalfTy;
6773 llvm::Type *Tys[2] = {InTy, FTy};
6774 Ops.push_back(EmitScalarExpr(E->getArg(1)));
6775 switch (BuiltinID) {
6776 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6776)
;
6777 case NEON::BI__builtin_neon_vcvth_n_s16_f16:
6778 Int = Intrinsic::aarch64_neon_vcvtfp2fxs; break;
6779 case NEON::BI__builtin_neon_vcvth_n_u16_f16:
6780 Int = Intrinsic::aarch64_neon_vcvtfp2fxu; break;
6781 }
6782 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n");
6783 return Builder.CreateTrunc(Ops[0], Int16Ty);
6784 }
6785 case NEON::BI__builtin_neon_vcvth_n_f16_s16:
6786 case NEON::BI__builtin_neon_vcvth_n_f16_u16: {
6787 unsigned Int;
6788 llvm::Type* FTy = HalfTy;
6789 llvm::Type* InTy = Int32Ty;
6790 llvm::Type *Tys[2] = {FTy, InTy};
6791 Ops.push_back(EmitScalarExpr(E->getArg(1)));
6792 switch (BuiltinID) {
6793 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6793)
;
6794 case NEON::BI__builtin_neon_vcvth_n_f16_s16:
6795 Int = Intrinsic::aarch64_neon_vcvtfxs2fp;
6796 Ops[0] = Builder.CreateSExt(Ops[0], InTy, "sext");
6797 break;
6798 case NEON::BI__builtin_neon_vcvth_n_f16_u16:
6799 Int = Intrinsic::aarch64_neon_vcvtfxu2fp;
6800 Ops[0] = Builder.CreateZExt(Ops[0], InTy);
6801 break;
6802 }
6803 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n");
6804 }
6805 case NEON::BI__builtin_neon_vpaddd_s64: {
6806 llvm::Type *Ty = llvm::VectorType::get(Int64Ty, 2);
6807 Value *Vec = EmitScalarExpr(E->getArg(0));
6808 // The vector is v2f64, so make sure it's bitcast to that.
6809 Vec = Builder.CreateBitCast(Vec, Ty, "v2i64");
6810 llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
6811 llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
6812 Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
6813 Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
6814 // Pairwise addition of a v2f64 into a scalar f64.
6815 return Builder.CreateAdd(Op0, Op1, "vpaddd");
6816 }
6817 case NEON::BI__builtin_neon_vpaddd_f64: {
6818 llvm::Type *Ty =
6819 llvm::VectorType::get(DoubleTy, 2);
6820 Value *Vec = EmitScalarExpr(E->getArg(0));
6821 // The vector is v2f64, so make sure it's bitcast to that.
6822 Vec = Builder.CreateBitCast(Vec, Ty, "v2f64");
6823 llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
6824 llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
6825 Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
6826 Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
6827 // Pairwise addition of a v2f64 into a scalar f64.
6828 return Builder.CreateFAdd(Op0, Op1, "vpaddd");
6829 }
6830 case NEON::BI__builtin_neon_vpadds_f32: {
6831 llvm::Type *Ty =
6832 llvm::VectorType::get(FloatTy, 2);
6833 Value *Vec = EmitScalarExpr(E->getArg(0));
6834 // The vector is v2f32, so make sure it's bitcast to that.
6835 Vec = Builder.CreateBitCast(Vec, Ty, "v2f32");
6836 llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
6837 llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
6838 Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
6839 Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
6840 // Pairwise addition of a v2f32 into a scalar f32.
6841 return Builder.CreateFAdd(Op0, Op1, "vpaddd");
6842 }
6843 case NEON::BI__builtin_neon_vceqzd_s64:
6844 case NEON::BI__builtin_neon_vceqzd_f64:
6845 case NEON::BI__builtin_neon_vceqzs_f32:
6846 case NEON::BI__builtin_neon_vceqzh_f16:
6847 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6848 return EmitAArch64CompareBuiltinExpr(
6849 Ops[0], ConvertType(E->getCallReturnType(getContext())),
6850 ICmpInst::FCMP_OEQ, ICmpInst::ICMP_EQ, "vceqz");
6851 case NEON::BI__builtin_neon_vcgezd_s64:
6852 case NEON::BI__builtin_neon_vcgezd_f64:
6853 case NEON::BI__builtin_neon_vcgezs_f32:
6854 case NEON::BI__builtin_neon_vcgezh_f16:
6855 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6856 return EmitAArch64CompareBuiltinExpr(
6857 Ops[0], ConvertType(E->getCallReturnType(getContext())),
6858 ICmpInst::FCMP_OGE, ICmpInst::ICMP_SGE, "vcgez");
6859 case NEON::BI__builtin_neon_vclezd_s64:
6860 case NEON::BI__builtin_neon_vclezd_f64:
6861 case NEON::BI__builtin_neon_vclezs_f32:
6862 case NEON::BI__builtin_neon_vclezh_f16:
6863 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6864 return EmitAArch64CompareBuiltinExpr(
6865 Ops[0], ConvertType(E->getCallReturnType(getContext())),
6866 ICmpInst::FCMP_OLE, ICmpInst::ICMP_SLE, "vclez");
6867 case NEON::BI__builtin_neon_vcgtzd_s64:
6868 case NEON::BI__builtin_neon_vcgtzd_f64:
6869 case NEON::BI__builtin_neon_vcgtzs_f32:
6870 case NEON::BI__builtin_neon_vcgtzh_f16:
6871 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6872 return EmitAArch64CompareBuiltinExpr(
6873 Ops[0], ConvertType(E->getCallReturnType(getContext())),
6874 ICmpInst::FCMP_OGT, ICmpInst::ICMP_SGT, "vcgtz");
6875 case NEON::BI__builtin_neon_vcltzd_s64:
6876 case NEON::BI__builtin_neon_vcltzd_f64:
6877 case NEON::BI__builtin_neon_vcltzs_f32:
6878 case NEON::BI__builtin_neon_vcltzh_f16:
6879 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6880 return EmitAArch64CompareBuiltinExpr(
6881 Ops[0], ConvertType(E->getCallReturnType(getContext())),
6882 ICmpInst::FCMP_OLT, ICmpInst::ICMP_SLT, "vcltz");
6883
6884 case NEON::BI__builtin_neon_vceqzd_u64: {
6885 Ops.push_back(EmitScalarExpr(E->getArg(0)));
6886 Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
6887 Ops[0] =
6888 Builder.CreateICmpEQ(Ops[0], llvm::Constant::getNullValue(Int64Ty));
6889 return Builder.CreateSExt(Ops[0], Int64Ty, "vceqzd");
6890 }
6891 case NEON::BI__builtin_neon_vceqd_f64:
6892 case NEON::BI__builtin_neon_vcled_f64:
6893 case NEON::BI__builtin_neon_vcltd_f64:
6894 case NEON::BI__builtin_neon_vcged_f64:
6895 case NEON::BI__builtin_neon_vcgtd_f64: {
6896 llvm::CmpInst::Predicate P;
6897 switch (BuiltinID) {
6898 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6898)
;
6899 case NEON::BI__builtin_neon_vceqd_f64: P = llvm::FCmpInst::FCMP_OEQ; break;
6900 case NEON::BI__builtin_neon_vcled_f64: P = llvm::FCmpInst::FCMP_OLE; break;
6901 case NEON::BI__builtin_neon_vcltd_f64: P = llvm::FCmpInst::FCMP_OLT; break;
6902 case NEON::BI__builtin_neon_vcged_f64: P = llvm::FCmpInst::FCMP_OGE; break;
6903 case NEON::BI__builtin_neon_vcgtd_f64: P = llvm::FCmpInst::FCMP_OGT; break;
6904 }
6905 Ops.push_back(EmitScalarExpr(E->getArg(1)));
6906 Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
6907 Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy);
6908 Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
6909 return Builder.CreateSExt(Ops[0], Int64Ty, "vcmpd");
6910 }
6911 case NEON::BI__builtin_neon_vceqs_f32:
6912 case NEON::BI__builtin_neon_vcles_f32:
6913 case NEON::BI__builtin_neon_vclts_f32:
6914 case NEON::BI__builtin_neon_vcges_f32:
6915 case NEON::BI__builtin_neon_vcgts_f32: {
6916 llvm::CmpInst::Predicate P;
6917 switch (BuiltinID) {
6918 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6918)
;
6919 case NEON::BI__builtin_neon_vceqs_f32: P = llvm::FCmpInst::FCMP_OEQ; break;
6920 case NEON::BI__builtin_neon_vcles_f32: P = llvm::FCmpInst::FCMP_OLE; break;
6921 case NEON::BI__builtin_neon_vclts_f32: P = llvm::FCmpInst::FCMP_OLT; break;
6922 case NEON::BI__builtin_neon_vcges_f32: P = llvm::FCmpInst::FCMP_OGE; break;
6923 case NEON::BI__builtin_neon_vcgts_f32: P = llvm::FCmpInst::FCMP_OGT; break;
6924 }
6925 Ops.push_back(EmitScalarExpr(E->getArg(1)));
6926 Ops[0] = Builder.CreateBitCast(Ops[0], FloatTy);
6927 Ops[1] = Builder.CreateBitCast(Ops[1], FloatTy);
6928 Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
6929 return Builder.CreateSExt(Ops[0], Int32Ty, "vcmpd");
6930 }
6931 case NEON::BI__builtin_neon_vceqh_f16:
6932 case NEON::BI__builtin_neon_vcleh_f16:
6933 case NEON::BI__builtin_neon_vclth_f16:
6934 case NEON::BI__builtin_neon_vcgeh_f16:
6935 case NEON::BI__builtin_neon_vcgth_f16: {
6936 llvm::CmpInst::Predicate P;
6937 switch (BuiltinID) {
6938 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6938)
;
6939 case NEON::BI__builtin_neon_vceqh_f16: P = llvm::FCmpInst::FCMP_OEQ; break;
6940 case NEON::BI__builtin_neon_vcleh_f16: P = llvm::FCmpInst::FCMP_OLE; break;
6941 case NEON::BI__builtin_neon_vclth_f16: P = llvm::FCmpInst::FCMP_OLT; break;
6942 case NEON::BI__builtin_neon_vcgeh_f16: P = llvm::FCmpInst::FCMP_OGE; break;
6943 case NEON::BI__builtin_neon_vcgth_f16: P = llvm::FCmpInst::FCMP_OGT; break;
6944 }
6945 Ops.push_back(EmitScalarExpr(E->getArg(1)));
6946 Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
6947 Ops[1] = Builder.CreateBitCast(Ops[1], HalfTy);
6948 Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
6949 return Builder.CreateSExt(Ops[0], Int16Ty, "vcmpd");
6950 }
6951 case NEON::BI__builtin_neon_vceqd_s64:
6952 case NEON::BI__builtin_neon_vceqd_u64:
6953 case NEON::BI__builtin_neon_vcgtd_s64:
6954 case NEON::BI__builtin_neon_vcgtd_u64:
6955 case NEON::BI__builtin_neon_vcltd_s64:
6956 case NEON::BI__builtin_neon_vcltd_u64:
6957 case NEON::BI__builtin_neon_vcged_u64:
6958 case NEON::BI__builtin_neon_vcged_s64:
6959 case NEON::BI__builtin_neon_vcled_u64:
6960 case NEON::BI__builtin_neon_vcled_s64: {
6961 llvm::CmpInst::Predicate P;
6962 switch (BuiltinID) {
6963 default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 6963)
;
6964 case NEON::BI__builtin_neon_vceqd_s64:
6965 case NEON::BI__builtin_neon_vceqd_u64:P = llvm::ICmpInst::ICMP_EQ;break;
6966 case NEON::BI__builtin_neon_vcgtd_s64:P = llvm::ICmpInst::ICMP_SGT;break;
6967 case NEON::BI__builtin_neon_vcgtd_u64:P = llvm::ICmpInst::ICMP_UGT;break;
6968 case NEON::BI__builtin_neon_vcltd_s64:P = llvm::ICmpInst::ICMP_SLT;break;
6969 case NEON::BI__builtin_neon_vcltd_u64:P = llvm::ICmpInst::ICMP_ULT;break;
6970 case NEON::BI__builtin_neon_vcged_u64:P = llvm::ICmpInst::ICMP_UGE;break;
6971 case NEON::BI__builtin_neon_vcged_s64:P = llvm::ICmpInst::ICMP_SGE;break;
6972 case NEON::BI__builtin_neon_vcled_u64:P = llvm::ICmpInst::ICMP_ULE;break;
6973 case NEON::BI__builtin_neon_vcled_s64:P = llvm::ICmpInst::ICMP_SLE;break;
6974 }
6975 Ops.push_back(EmitScalarExpr(E->getArg(1)));
6976 Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
6977 Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
6978 Ops[0] = Builder.CreateICmp(P, Ops[0], Ops[1]);
6979 return Builder.CreateSExt(Ops[0], Int64Ty, "vceqd");
6980 }
6981 case NEON::BI__builtin_neon_vtstd_s64:
6982 case NEON::BI__builtin_neon_vtstd_u64: {
6983 Ops.push_back(EmitScalarExpr(E->getArg(1)));
6984 Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
6985 Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
6986 Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]);
6987 Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0],
6988 llvm::Constant::getNullValue(Int64Ty));
6989 return Builder.CreateSExt(Ops[0], Int64Ty, "vtstd");
6990 }
6991 case NEON::BI__builtin_neon_vset_lane_i8:
6992 case NEON::BI__builtin_neon_vset_lane_i16:
6993 case NEON::BI__builtin_neon_vset_lane_i32:
6994 case NEON::BI__builtin_neon_vset_lane_i64:
6995 case NEON::BI__builtin_neon_vset_lane_f32:
6996 case NEON::BI__builtin_neon_vsetq_lane_i8:
6997 case NEON::BI__builtin_neon_vsetq_lane_i16:
6998 case NEON::BI__builtin_neon_vsetq_lane_i32:
6999 case NEON::BI__builtin_neon_vsetq_lane_i64:
7000 case NEON::BI__builtin_neon_vsetq_lane_f32:
7001 Ops.push_back(EmitScalarExpr(E->getArg(2)));
7002 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
7003 case NEON::BI__builtin_neon_vset_lane_f64:
7004 // The vector type needs a cast for the v1f64 variant.
7005 Ops[1] = Builder.CreateBitCast(Ops[1],
7006 llvm::VectorType::get(DoubleTy, 1));
7007 Ops.push_back(EmitScalarExpr(E->getArg(2)));
7008 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
7009 case NEON::BI__builtin_neon_vsetq_lane_f64:
7010 // The vector type needs a cast for the v2f64 variant.
7011 Ops[1] = Builder.CreateBitCast(Ops[1],
7012 llvm::VectorType::get(DoubleTy, 2));
7013 Ops.push_back(EmitScalarExpr(E->getArg(2)));
7014 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
7015
7016 case NEON::BI__builtin_neon_vget_lane_i8:
7017 case NEON::BI__builtin_neon_vdupb_lane_i8:
7018 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int8Ty, 8));
7019 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7020 "vget_lane");
7021 case NEON::BI__builtin_neon_vgetq_lane_i8:
7022 case NEON::BI__builtin_neon_vdupb_laneq_i8:
7023 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int8Ty, 16));
7024 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7025 "vgetq_lane");
7026 case NEON::BI__builtin_neon_vget_lane_i16:
7027 case NEON::BI__builtin_neon_vduph_lane_i16:
7028 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int16Ty, 4));
7029 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7030 "vget_lane");
7031 case NEON::BI__builtin_neon_vgetq_lane_i16:
7032 case NEON::BI__builtin_neon_vduph_laneq_i16:
7033 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int16Ty, 8));
7034 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7035 "vgetq_lane");
7036 case NEON::BI__builtin_neon_vget_lane_i32:
7037 case NEON::BI__builtin_neon_vdups_lane_i32:
7038 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 2));
7039 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7040 "vget_lane");
7041 case NEON::BI__builtin_neon_vdups_lane_f32:
7042 Ops[0] = Builder.CreateBitCast(Ops[0],
7043 llvm::VectorType::get(FloatTy, 2));
7044 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7045 "vdups_lane");
7046 case NEON::BI__builtin_neon_vgetq_lane_i32:
7047 case NEON::BI__builtin_neon_vdups_laneq_i32:
7048 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4));
7049 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7050 "vgetq_lane");
7051 case NEON::BI__builtin_neon_vget_lane_i64:
7052 case NEON::BI__builtin_neon_vdupd_lane_i64:
7053 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 1));
7054 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7055 "vget_lane");
7056 case NEON::BI__builtin_neon_vdupd_lane_f64:
7057 Ops[0] = Builder.CreateBitCast(Ops[0],
7058 llvm::VectorType::get(DoubleTy, 1));
7059 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7060 "vdupd_lane");
7061 case NEON::BI__builtin_neon_vgetq_lane_i64:
7062 case NEON::BI__builtin_neon_vdupd_laneq_i64:
7063 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2));
7064 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7065 "vgetq_lane");
7066 case NEON::BI__builtin_neon_vget_lane_f32:
7067 Ops[0] = Builder.CreateBitCast(Ops[0],
7068 llvm::VectorType::get(FloatTy, 2));
7069 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7070 "vget_lane");
7071 case NEON::BI__builtin_neon_vget_lane_f64:
7072 Ops[0] = Builder.CreateBitCast(Ops[0],
7073 llvm::VectorType::get(DoubleTy, 1));
7074 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7075 "vget_lane");
7076 case NEON::BI__builtin_neon_vgetq_lane_f32:
7077 case NEON::BI__builtin_neon_vdups_laneq_f32:
7078 Ops[0] = Builder.CreateBitCast(Ops[0],
7079 llvm::VectorType::get(FloatTy, 4));
7080 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7081 "vgetq_lane");
7082 case NEON::BI__builtin_neon_vgetq_lane_f64:
7083 case NEON::BI__builtin_neon_vdupd_laneq_f64:
7084 Ops[0] = Builder.CreateBitCast(Ops[0],
7085 llvm::VectorType::get(DoubleTy, 2));
7086 return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7087 "vgetq_lane");
7088 case NEON::BI__builtin_neon_vaddh_f16:
7089 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7090 return Builder.CreateFAdd(Ops[0], Ops[1], "vaddh");
7091 case NEON::BI__builtin_neon_vsubh_f16:
7092 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7093 return Builder.CreateFSub(Ops[0], Ops[1], "vsubh");
7094 case NEON::BI__builtin_neon_vmulh_f16:
7095 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7096 return Builder.CreateFMul(Ops[0], Ops[1], "vmulh");
7097 case NEON::BI__builtin_neon_vdivh_f16:
7098 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7099 return Builder.CreateFDiv(Ops[0], Ops[1], "vdivh");
7100 case NEON::BI__builtin_neon_vfmah_f16: {
7101 Value *F = CGM.getIntrinsic(Intrinsic::fma, HalfTy);
7102 // NEON intrinsic puts accumulator first, unlike the LLVM fma.
7103 return Builder.CreateCall(F,
7104 {EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), Ops[0]});
7105 }
7106 case NEON::BI__builtin_neon_vfmsh_f16: {
7107 Value *F = CGM.getIntrinsic(Intrinsic::fma, HalfTy);
7108 Value *Zero = llvm::ConstantFP::getZeroValueForNegation(HalfTy);
7109 Value* Sub = Builder.CreateFSub(Zero, EmitScalarExpr(E->getArg(1)), "vsubh");
7110 // NEON intrinsic puts accumulator first, unlike the LLVM fma.
7111 return Builder.CreateCall(F, {Sub, EmitScalarExpr(E->getArg(2)), Ops[0]});
7112 }
7113 case NEON::BI__builtin_neon_vaddd_s64:
7114 case NEON::BI__builtin_neon_vaddd_u64:
7115 return Builder.CreateAdd(Ops[0], EmitScalarExpr(E->getArg(1)), "vaddd");
7116 case NEON::BI__builtin_neon_vsubd_s64:
7117 case NEON::BI__builtin_neon_vsubd_u64:
7118 return Builder.CreateSub(Ops[0], EmitScalarExpr(E->getArg(1)), "vsubd");
7119 case NEON::BI__builtin_neon_vqdmlalh_s16:
7120 case NEON::BI__builtin_neon_vqdmlslh_s16: {
7121 SmallVector<Value *, 2> ProductOps;
7122 ProductOps.push_back(vectorWrapScalar16(Ops[1]));
7123 ProductOps.push_back(vectorWrapScalar16(EmitScalarExpr(E->getArg(2))));
7124 llvm::Type *VTy = llvm::VectorType::get(Int32Ty, 4);
7125 Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy),
7126 ProductOps, "vqdmlXl");
7127 Constant *CI = ConstantInt::get(SizeTy, 0);
7128 Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0");
7129
7130 unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlalh_s16
7131 ? Intrinsic::aarch64_neon_sqadd
7132 : Intrinsic::aarch64_neon_sqsub;
7133 return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int32Ty), Ops, "vqdmlXl");
7134 }
7135 case NEON::BI__builtin_neon_vqshlud_n_s64: {
7136 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7137 Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty);
7138 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqshlu, Int64Ty),
7139 Ops, "vqshlu_n");
7140 }
7141 case NEON::BI__builtin_neon_vqshld_n_u64:
7142 case NEON::BI__builtin_neon_vqshld_n_s64: {
7143 unsigned Int = BuiltinID == NEON::BI__builtin_neon_vqshld_n_u64
7144 ? Intrinsic::aarch64_neon_uqshl
7145 : Intrinsic::aarch64_neon_sqshl;
7146 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7147 Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty);
7148 return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vqshl_n");
7149 }
7150 case NEON::BI__builtin_neon_vrshrd_n_u64:
7151 case NEON::BI__builtin_neon_vrshrd_n_s64: {
7152 unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrshrd_n_u64
7153 ? Intrinsic::aarch64_neon_urshl
7154 : Intrinsic::aarch64_neon_srshl;
7155 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7156 int SV = cast<ConstantInt>(Ops[1])->getSExtValue();
7157 Ops[1] = ConstantInt::get(Int64Ty, -SV);
7158 return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vrshr_n");
7159 }
7160 case NEON::BI__builtin_neon_vrsrad_n_u64:
7161 case NEON::BI__builtin_neon_vrsrad_n_s64: {
7162 unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrsrad_n_u64
7163 ? Intrinsic::aarch64_neon_urshl
7164 : Intrinsic::aarch64_neon_srshl;
7165 Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
7166 Ops.push_back(Builder.CreateNeg(EmitScalarExpr(E->getArg(2))));
7167 Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Int64Ty),
7168 {Ops[1], Builder.CreateSExt(Ops[2], Int64Ty)});
7169 return Builder.CreateAdd(Ops[0], Builder.CreateBitCast(Ops[1], Int64Ty));
7170 }
7171 case NEON::BI__builtin_neon_vshld_n_s64:
7172 case NEON::BI__builtin_neon_vshld_n_u64: {
7173 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
7174 return Builder.CreateShl(
7175 Ops[0], ConstantInt::get(Int64Ty, Amt->getZExtValue()), "shld_n");
7176 }
7177 case NEON::BI__builtin_neon_vshrd_n_s64: {
7178 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
7179 return Builder.CreateAShr(
7180 Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
7181 Amt->getZExtValue())),
7182 "shrd_n");
7183 }
7184 case NEON::BI__builtin_neon_vshrd_n_u64: {
7185 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
7186 uint64_t ShiftAmt = Amt->getZExtValue();
7187 // Right-shifting an unsigned value by its size yields 0.
7188 if (ShiftAmt == 64)
7189 return ConstantInt::get(Int64Ty, 0);
7190 return Builder.CreateLShr(Ops[0], ConstantInt::get(Int64Ty, ShiftAmt),
7191 "shrd_n");
7192 }
7193 case NEON::BI__builtin_neon_vsrad_n_s64: {
7194 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2)));
7195 Ops[1] = Builder.CreateAShr(
7196 Ops[1], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
7197 Amt->getZExtValue())),
7198 "shrd_n");
7199 return Builder.CreateAdd(Ops[0], Ops[1]);
7200 }
7201 case NEON::BI__builtin_neon_vsrad_n_u64: {
7202 llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2)));
7203 uint64_t ShiftAmt = Amt->getZExtValue();
7204 // Right-shifting an unsigned value by its size yields 0.
7205 // As Op + 0 = Op, return Ops[0] directly.
7206 if (ShiftAmt == 64)
7207 return Ops[0];
7208 Ops[1] = Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, ShiftAmt),
7209 "shrd_n");
7210 return Builder.CreateAdd(Ops[0], Ops[1]);
7211 }
7212 case NEON::BI__builtin_neon_vqdmlalh_lane_s16:
7213 case NEON::BI__builtin_neon_vqdmlalh_laneq_s16:
7214 case NEON::BI__builtin_neon_vqdmlslh_lane_s16:
7215 case NEON::BI__builtin_neon_vqdmlslh_laneq_s16: {
7216 Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)),
7217 "lane");
7218 SmallVector<Value *, 2> ProductOps;
7219 ProductOps.push_back(vectorWrapScalar16(Ops[1]));
7220 ProductOps.push_back(vectorWrapScalar16(Ops[2]));
7221 llvm::Type *VTy = llvm::VectorType::get(Int32Ty, 4);
7222 Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy),
7223 ProductOps, "vqdmlXl");
7224 Constant *CI = ConstantInt::get(SizeTy, 0);
7225 Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0");
7226 Ops.pop_back();
7227
7228 unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlalh_lane_s16 ||
7229 BuiltinID == NEON::BI__builtin_neon_vqdmlalh_laneq_s16)
7230 ? Intrinsic::aarch64_neon_sqadd
7231 : Intrinsic::aarch64_neon_sqsub;
7232 return EmitNeonCall(CGM.getIntrinsic(AccInt, Int32Ty), Ops, "vqdmlXl");
7233 }
7234 case NEON::BI__builtin_neon_vqdmlals_s32:
7235 case NEON::BI__builtin_neon_vqdmlsls_s32: {
7236 SmallVector<Value *, 2> ProductOps;
7237 ProductOps.push_back(Ops[1]);
7238 ProductOps.push_back(EmitScalarExpr(E->getArg(2)));
7239 Ops[1] =
7240 EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar),
7241 ProductOps, "vqdmlXl");
7242
7243 unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlals_s32
7244 ? Intrinsic::aarch64_neon_sqadd
7245 : Intrinsic::aarch64_neon_sqsub;
7246 return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int64Ty), Ops, "vqdmlXl");
7247 }
7248 case NEON::BI__builtin_neon_vqdmlals_lane_s32:
7249 case NEON::BI__builtin_neon_vqdmlals_laneq_s32:
7250 case NEON::BI__builtin_neon_vqdmlsls_lane_s32:
7251 case NEON::BI__builtin_neon_vqdmlsls_laneq_s32: {
7252 Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)),
7253 "lane");
7254 SmallVector<Value *, 2> ProductOps;
7255 ProductOps.push_back(Ops[1]);
7256 ProductOps.push_back(Ops[2]);
7257 Ops[1] =
7258 EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar),
7259 ProductOps, "vqdmlXl");
7260 Ops.pop_back();
7261
7262 unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlals_lane_s32 ||
7263 BuiltinID == NEON::BI__builtin_neon_vqdmlals_laneq_s32)
7264 ? Intrinsic::aarch64_neon_sqadd
7265 : Intrinsic::aarch64_neon_sqsub;
7266 return EmitNeonCall(CGM.getIntrinsic(AccInt, Int64Ty), Ops, "vqdmlXl");
7267 }
7268 }
7269
7270 llvm::VectorType *VTy = GetNeonType(this, Type);
7271 llvm::Type *Ty = VTy;
7272 if (!Ty)
7273 return nullptr;
7274
7275 // Not all intrinsics handled by the common case work for AArch64 yet, so only
7276 // defer to common code if it's been added to our special map.
7277 Builtin = findNeonIntrinsicInMap(AArch64SIMDIntrinsicMap, BuiltinID,
7278 AArch64SIMDIntrinsicsProvenSorted);
7279
7280 if (Builtin)
7281 return EmitCommonNeonBuiltinExpr(
7282 Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic,
7283 Builtin->NameHint, Builtin->TypeModifier, E, Ops,
7284 /*never use addresses*/ Address::invalid(), Address::invalid(), Arch);
7285
7286 if (Value *V = EmitAArch64TblBuiltinExpr(*this, BuiltinID, E, Ops, Arch))
7287 return V;
7288
7289 unsigned Int;
7290 switch (BuiltinID) {
7291 default: return nullptr;
7292 case NEON::BI__builtin_neon_vbsl_v:
7293 case NEON::BI__builtin_neon_vbslq_v: {
7294 llvm::Type *BitTy = llvm::VectorType::getInteger(VTy);
7295 Ops[0] = Builder.CreateBitCast(Ops[0], BitTy, "vbsl");
7296 Ops[1] = Builder.CreateBitCast(Ops[1], BitTy, "vbsl");
7297 Ops[2] = Builder.CreateBitCast(Ops[2], BitTy, "vbsl");
7298
7299 Ops[1] = Builder.CreateAnd(Ops[0], Ops[1], "vbsl");
7300 Ops[2] = Builder.CreateAnd(Builder.CreateNot(Ops[0]), Ops[2], "vbsl");
7301 Ops[0] = Builder.CreateOr(Ops[1], Ops[2], "vbsl");
7302 return Builder.CreateBitCast(Ops[0], Ty);
7303 }
7304 case NEON::BI__builtin_neon_vfma_lane_v:
7305 case NEON::BI__builtin_neon_vfmaq_lane_v: { // Only used for FP types
7306 // The ARM builtins (and instructions) have the addend as the first
7307 // operand, but the 'fma' intrinsics have it last. Swap it around here.
7308 Value *Addend = Ops[0];
7309 Value *Multiplicand = Ops[1];
7310 Value *LaneSource = Ops[2];
7311 Ops[0] = Multiplicand;
7312 Ops[1] = LaneSource;
7313 Ops[2] = Addend;
7314
7315 // Now adjust things to handle the lane access.
7316 llvm::Type *SourceTy = BuiltinID == NEON::BI__builtin_neon_vfmaq_lane_v ?
7317 llvm::VectorType::get(VTy->getElementType(), VTy->getNumElements() / 2) :
7318 VTy;
7319 llvm::Constant *cst = cast<Constant>(Ops[3]);
7320 Value *SV = llvm::ConstantVector::getSplat(VTy->getNumElements(), cst);
7321 Ops[1] = Builder.CreateBitCast(Ops[1], SourceTy);
7322 Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV, "lane");
7323
7324 Ops.pop_back();
7325 Int = Intrinsic::fma;
7326 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "fmla");
7327 }
7328 case NEON::BI__builtin_neon_vfma_laneq_v: {
7329 llvm::VectorType *VTy = cast<llvm::VectorType>(Ty);
7330 // v1f64 fma should be mapped to Neon scalar f64 fma
7331 if (VTy && VTy->getElementType() == DoubleTy) {
7332 Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
7333 Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy);
7334 llvm::Type *VTy = GetNeonType(this,
7335 NeonTypeFlags(NeonTypeFlags::Float64, false, true));
7336 Ops[2] = Builder.CreateBitCast(Ops[2], VTy);
7337 Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract");
7338 Value *F = CGM.getIntrinsic(Intrinsic::fma, DoubleTy);
7339 Value *Result = Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
7340 return Builder.CreateBitCast(Result, Ty);
7341 }
7342 Value *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
7343 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
7344 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7345
7346 llvm::Type *STy = llvm::VectorType::get(VTy->getElementType(),
7347 VTy->getNumElements() * 2);
7348 Ops[2] = Builder.CreateBitCast(Ops[2], STy);
7349 Value* SV = llvm::ConstantVector::getSplat(VTy->getNumElements(),
7350 cast<ConstantInt>(Ops[3]));
7351 Ops[2] = Builder.CreateShuffleVector(Ops[2], Ops[2], SV, "lane");
7352
7353 return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]});
7354 }
7355 case NEON::BI__builtin_neon_vfmaq_laneq_v: {
7356 Value *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
7357 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
7358 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7359
7360 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
7361 Ops[2] = EmitNeonSplat(Ops[2], cast<ConstantInt>(Ops[3]));
7362 return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]});
7363 }
7364 case NEON::BI__builtin_neon_vfmah_lane_f16:
7365 case NEON::BI__builtin_neon_vfmas_lane_f32:
7366 case NEON::BI__builtin_neon_vfmah_laneq_f16:
7367 case NEON::BI__builtin_neon_vfmas_laneq_f32:
7368 case NEON::BI__builtin_neon_vfmad_lane_f64:
7369 case NEON::BI__builtin_neon_vfmad_laneq_f64: {
7370 Ops.push_back(EmitScalarExpr(E->getArg(3)));
7371 llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext()));
7372 Value *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
7373 Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract");
7374 return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
7375 }
7376 case NEON::BI__builtin_neon_vmull_v:
7377 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7378 Int = usgn ? Intrinsic::aarch64_neon_umull : Intrinsic::aarch64_neon_smull;
7379 if (Type.isPoly()) Int = Intrinsic::aarch64_neon_pmull;
7380 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull");
7381 case NEON::BI__builtin_neon_vmax_v:
7382 case NEON::BI__builtin_neon_vmaxq_v:
7383 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7384 Int = usgn ? Intrinsic::aarch64_neon_umax : Intrinsic::aarch64_neon_smax;
7385 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmax;
7386 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmax");
7387 case NEON::BI__builtin_neon_vmaxh_f16: {
7388 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7389 Int = Intrinsic::aarch64_neon_fmax;
7390 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmax");
7391 }
7392 case NEON::BI__builtin_neon_vmin_v:
7393 case NEON::BI__builtin_neon_vminq_v:
7394 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7395 Int = usgn ? Intrinsic::aarch64_neon_umin : Intrinsic::aarch64_neon_smin;
7396 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmin;
7397 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmin");
7398 case NEON::BI__builtin_neon_vminh_f16: {
7399 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7400 Int = Intrinsic::aarch64_neon_fmin;
7401 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmin");
7402 }
7403 case NEON::BI__builtin_neon_vabd_v:
7404 case NEON::BI__builtin_neon_vabdq_v:
7405 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7406 Int = usgn ? Intrinsic::aarch64_neon_uabd : Intrinsic::aarch64_neon_sabd;
7407 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fabd;
7408 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vabd");
7409 case NEON::BI__builtin_neon_vpadal_v:
7410 case NEON::BI__builtin_neon_vpadalq_v: {
7411 unsigned ArgElts = VTy->getNumElements();
7412 llvm::IntegerType *EltTy = cast<IntegerType>(VTy->getElementType());
7413 unsigned BitWidth = EltTy->getBitWidth();
7414 llvm::Type *ArgTy = llvm::VectorType::get(
7415 llvm::IntegerType::get(getLLVMContext(), BitWidth/2), 2*ArgElts);
7416 llvm::Type* Tys[2] = { VTy, ArgTy };
7417 Int = usgn ? Intrinsic::aarch64_neon_uaddlp : Intrinsic::aarch64_neon_saddlp;
7418 SmallVector<llvm::Value*, 1> TmpOps;
7419 TmpOps.push_back(Ops[1]);
7420 Function *F = CGM.getIntrinsic(Int, Tys);
7421 llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vpadal");
7422 llvm::Value *addend = Builder.CreateBitCast(Ops[0], tmp->getType());
7423 return Builder.CreateAdd(tmp, addend);
7424 }
7425 case NEON::BI__builtin_neon_vpmin_v:
7426 case NEON::BI__builtin_neon_vpminq_v:
7427 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7428 Int = usgn ? Intrinsic::aarch64_neon_uminp : Intrinsic::aarch64_neon_sminp;
7429 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fminp;
7430 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmin");
7431 case NEON::BI__builtin_neon_vpmax_v:
7432 case NEON::BI__builtin_neon_vpmaxq_v:
7433 // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7434 Int = usgn ? Intrinsic::aarch64_neon_umaxp : Intrinsic::aarch64_neon_smaxp;
7435 if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmaxp;
7436 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmax");
7437 case NEON::BI__builtin_neon_vminnm_v:
7438 case NEON::BI__builtin_neon_vminnmq_v:
7439 Int = Intrinsic::aarch64_neon_fminnm;
7440 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vminnm");
7441 case NEON::BI__builtin_neon_vminnmh_f16:
7442 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7443 Int = Intrinsic::aarch64_neon_fminnm;
7444 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vminnm");
7445 case NEON::BI__builtin_neon_vmaxnm_v:
7446 case NEON::BI__builtin_neon_vmaxnmq_v:
7447 Int = Intrinsic::aarch64_neon_fmaxnm;
7448 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmaxnm");
7449 case NEON::BI__builtin_neon_vmaxnmh_f16:
7450 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7451 Int = Intrinsic::aarch64_neon_fmaxnm;
7452 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmaxnm");
7453 case NEON::BI__builtin_neon_vrecpss_f32: {
7454 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7455 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, FloatTy),
7456 Ops, "vrecps");
7457 }
7458 case NEON::BI__builtin_neon_vrecpsd_f64:
7459 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7460 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, DoubleTy),
7461 Ops, "vrecps");
7462 case NEON::BI__builtin_neon_vrecpsh_f16:
7463 Ops.push_back(EmitScalarExpr(E->getArg(1)));
7464 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, HalfTy),
7465 Ops, "vrecps");
7466 case NEON::BI__builtin_neon_vqshrun_n_v:
7467 Int = Intrinsic::aarch64_neon_sqshrun;
7468 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrun_n");
7469 case NEON::BI__builtin_neon_vqrshrun_n_v:
7470 Int = Intrinsic::aarch64_neon_sqrshrun;
7471 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrun_n");
7472 case NEON::BI__builtin_neon_vqshrn_n_v:
7473 Int = usgn ? Intrinsic::aarch64_neon_uqshrn : Intrinsic::aarch64_neon_sqshrn;
7474 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n");
7475 case NEON::BI__builtin_neon_vrshrn_n_v:
7476 Int = Intrinsic::aarch64_neon_rshrn;
7477 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshrn_n");
7478 case NEON::BI__builtin_neon_vqrshrn_n_v:
7479 Int = usgn ? Intrinsic::aarch64_neon_uqrshrn : Intrinsic::aarch64_neon_sqrshrn;
7480 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n");
7481 case NEON::BI__builtin_neon_vrndah_f16: {
7482 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7483 Int = Intrinsic::round;
7484 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrnda");
7485 }
7486 case NEON::BI__builtin_neon_vrnda_v:
7487 case NEON::BI__builtin_neon_vrndaq_v: {
7488 Int = Intrinsic::round;
7489 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnda");
7490 }
7491 case NEON::BI__builtin_neon_vrndih_f16: {
7492 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7493 Int = Intrinsic::nearbyint;
7494 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndi");
7495 }
7496 case NEON::BI__builtin_neon_vrndmh_f16: {
7497 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7498 Int = Intrinsic::floor;
7499 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndm");
7500 }
7501 case NEON::BI__builtin_neon_vrndm_v:
7502 case NEON::BI__builtin_neon_vrndmq_v: {
7503 Int = Intrinsic::floor;
7504 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndm");
7505 }
7506 case NEON::BI__builtin_neon_vrndnh_f16: {
7507 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7508 Int = Intrinsic::aarch64_neon_frintn;
7509 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndn");
7510 }
7511 case NEON::BI__builtin_neon_vrndn_v:
7512 case NEON::BI__builtin_neon_vrndnq_v: {
7513 Int = Intrinsic::aarch64_neon_frintn;
7514 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndn");
7515 }
7516 case NEON::BI__builtin_neon_vrndns_f32: {
7517 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7518 Int = Intrinsic::aarch64_neon_frintn;
7519 return EmitNeonCall(CGM.getIntrinsic(Int, FloatTy), Ops, "vrndn");
7520 }
7521 case NEON::BI__builtin_neon_vrndph_f16: {
7522 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7523 Int = Intrinsic::ceil;
7524 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndp");
7525 }
7526 case NEON::BI__builtin_neon_vrndp_v:
7527 case NEON::BI__builtin_neon_vrndpq_v: {
7528 Int = Intrinsic::ceil;
7529 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndp");
7530 }
7531 case NEON::BI__builtin_neon_vrndxh_f16: {
7532 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7533 Int = Intrinsic::rint;
7534 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndx");
7535 }
7536 case NEON::BI__builtin_neon_vrndx_v:
7537 case NEON::BI__builtin_neon_vrndxq_v: {
7538 Int = Intrinsic::rint;
7539 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndx");
7540 }
7541 case NEON::BI__builtin_neon_vrndh_f16: {
7542 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7543 Int = Intrinsic::trunc;
7544 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndz");
7545 }
7546 case NEON::BI__builtin_neon_vrnd_v:
7547 case NEON::BI__builtin_neon_vrndq_v: {
7548 Int = Intrinsic::trunc;
7549 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndz");
7550 }
7551 case NEON::BI__builtin_neon_vcvt_f64_v:
7552 case NEON::BI__builtin_neon_vcvtq_f64_v:
7553 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
7554 Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, quad));
7555 return usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
7556 : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
7557 case NEON::BI__builtin_neon_vcvt_f64_f32: {
7558 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 7559, __extension__ __PRETTY_FUNCTION__))
7559 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 7559, __extension__ __PRETTY_FUNCTION__))
;
7560 NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float32, false, false);
7561 Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag));
7562
7563 return Builder.CreateFPExt(Ops[0], Ty, "vcvt");
7564 }
7565 case NEON::BI__builtin_neon_vcvt_f32_f64: {
7566 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 7567, __extension__ __PRETTY_FUNCTION__))
7567 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 7567, __extension__ __PRETTY_FUNCTION__))
;
7568 NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float64, false, true);
7569 Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag));
7570
7571 return Builder.CreateFPTrunc(Ops[0], Ty, "vcvt");
7572 }
7573 case NEON::BI__builtin_neon_vcvt_s32_v:
7574 case NEON::BI__builtin_neon_vcvt_u32_v:
7575 case NEON::BI__builtin_neon_vcvt_s64_v:
7576 case NEON::BI__builtin_neon_vcvt_u64_v:
7577 case NEON::BI__builtin_neon_vcvt_s16_v:
7578 case NEON::BI__builtin_neon_vcvt_u16_v:
7579 case NEON::BI__builtin_neon_vcvtq_s32_v:
7580 case NEON::BI__builtin_neon_vcvtq_u32_v:
7581 case NEON::BI__builtin_neon_vcvtq_s64_v:
7582 case NEON::BI__builtin_neon_vcvtq_u64_v:
7583 case NEON::BI__builtin_neon_vcvtq_s16_v:
7584 case NEON::BI__builtin_neon_vcvtq_u16_v: {
7585 Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type));
7586 if (usgn)
7587 return Builder.CreateFPToUI(Ops[0], Ty);
7588 return Builder.CreateFPToSI(Ops[0], Ty);
7589 }
7590 case NEON::BI__builtin_neon_vcvta_s16_v:
7591 case NEON::BI__builtin_neon_vcvta_u16_v:
7592 case NEON::BI__builtin_neon_vcvta_s32_v:
7593 case NEON::BI__builtin_neon_vcvtaq_s16_v:
7594 case NEON::BI__builtin_neon_vcvtaq_s32_v:
7595 case NEON::BI__builtin_neon_vcvta_u32_v:
7596 case NEON::BI__builtin_neon_vcvtaq_u16_v:
7597 case NEON::BI__builtin_neon_vcvtaq_u32_v:
7598 case NEON::BI__builtin_neon_vcvta_s64_v:
7599 case NEON::BI__builtin_neon_vcvtaq_s64_v:
7600 case NEON::BI__builtin_neon_vcvta_u64_v:
7601 case NEON::BI__builtin_neon_vcvtaq_u64_v: {
7602 Int = usgn ? Intrinsic::aarch64_neon_fcvtau : Intrinsic::aarch64_neon_fcvtas;
7603 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
7604 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvta");
7605 }
7606 case NEON::BI__builtin_neon_vcvtm_s16_v:
7607 case NEON::BI__builtin_neon_vcvtm_s32_v:
7608 case NEON::BI__builtin_neon_vcvtmq_s16_v:
7609 case NEON::BI__builtin_neon_vcvtmq_s32_v:
7610 case NEON::BI__builtin_neon_vcvtm_u16_v:
7611 case NEON::BI__builtin_neon_vcvtm_u32_v:
7612 case NEON::BI__builtin_neon_vcvtmq_u16_v:
7613 case NEON::BI__builtin_neon_vcvtmq_u32_v:
7614 case NEON::BI__builtin_neon_vcvtm_s64_v:
7615 case NEON::BI__builtin_neon_vcvtmq_s64_v:
7616 case NEON::BI__builtin_neon_vcvtm_u64_v:
7617 case NEON::BI__builtin_neon_vcvtmq_u64_v: {
7618 Int = usgn ? Intrinsic::aarch64_neon_fcvtmu : Intrinsic::aarch64_neon_fcvtms;
7619 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
7620 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtm");
7621 }
7622 case NEON::BI__builtin_neon_vcvtn_s16_v:
7623 case NEON::BI__builtin_neon_vcvtn_s32_v:
7624 case NEON::BI__builtin_neon_vcvtnq_s16_v:
7625 case NEON::BI__builtin_neon_vcvtnq_s32_v:
7626 case NEON::BI__builtin_neon_vcvtn_u16_v:
7627 case NEON::BI__builtin_neon_vcvtn_u32_v:
7628 case NEON::BI__builtin_neon_vcvtnq_u16_v:
7629 case NEON::BI__builtin_neon_vcvtnq_u32_v:
7630 case NEON::BI__builtin_neon_vcvtn_s64_v:
7631 case NEON::BI__builtin_neon_vcvtnq_s64_v:
7632 case NEON::BI__builtin_neon_vcvtn_u64_v:
7633 case NEON::BI__builtin_neon_vcvtnq_u64_v: {
7634 Int = usgn ? Intrinsic::aarch64_neon_fcvtnu : Intrinsic::aarch64_neon_fcvtns;
7635 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
7636 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtn");
7637 }
7638 case NEON::BI__builtin_neon_vcvtp_s16_v:
7639 case NEON::BI__builtin_neon_vcvtp_s32_v:
7640 case NEON::BI__builtin_neon_vcvtpq_s16_v:
7641 case NEON::BI__builtin_neon_vcvtpq_s32_v:
7642 case NEON::BI__builtin_neon_vcvtp_u16_v:
7643 case NEON::BI__builtin_neon_vcvtp_u32_v:
7644 case NEON::BI__builtin_neon_vcvtpq_u16_v:
7645 case NEON::BI__builtin_neon_vcvtpq_u32_v:
7646 case NEON::BI__builtin_neon_vcvtp_s64_v:
7647 case NEON::BI__builtin_neon_vcvtpq_s64_v:
7648 case NEON::BI__builtin_neon_vcvtp_u64_v:
7649 case NEON::BI__builtin_neon_vcvtpq_u64_v: {
7650 Int = usgn ? Intrinsic::aarch64_neon_fcvtpu : Intrinsic::aarch64_neon_fcvtps;
7651 llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
7652 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtp");
7653 }
7654 case NEON::BI__builtin_neon_vmulx_v:
7655 case NEON::BI__builtin_neon_vmulxq_v: {
7656 Int = Intrinsic::aarch64_neon_fmulx;
7657 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmulx");
7658 }
7659 case NEON::BI__builtin_neon_vmulxh_lane_f16:
7660 case NEON::BI__builtin_neon_vmulxh_laneq_f16: {
7661 // vmulx_lane should be mapped to Neon scalar mulx after
7662 // extracting the scalar element
7663 Ops.push_back(EmitScalarExpr(E->getArg(2)));
7664 Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract");
7665 Ops.pop_back();
7666 Int = Intrinsic::aarch64_neon_fmulx;
7667 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmulx");
7668 }
7669 case NEON::BI__builtin_neon_vmul_lane_v:
7670 case NEON::BI__builtin_neon_vmul_laneq_v: {
7671 // v1f64 vmul_lane should be mapped to Neon scalar mul lane
7672 bool Quad = false;
7673 if (BuiltinID == NEON::BI__builtin_neon_vmul_laneq_v)
7674 Quad = true;
7675 Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
7676 llvm::Type *VTy = GetNeonType(this,
7677 NeonTypeFlags(NeonTypeFlags::Float64, false, Quad));
7678 Ops[1] = Builder.CreateBitCast(Ops[1], VTy);
7679 Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract");
7680 Value *Result = Builder.CreateFMul(Ops[0], Ops[1]);
7681 return Builder.CreateBitCast(Result, Ty);
7682 }
7683 case NEON::BI__builtin_neon_vnegd_s64:
7684 return Builder.CreateNeg(EmitScalarExpr(E->getArg(0)), "vnegd");
7685 case NEON::BI__builtin_neon_vnegh_f16:
7686 return Builder.CreateFNeg(EmitScalarExpr(E->getArg(0)), "vnegh");
7687 case NEON::BI__builtin_neon_vpmaxnm_v:
7688 case NEON::BI__builtin_neon_vpmaxnmq_v: {
7689 Int = Intrinsic::aarch64_neon_fmaxnmp;
7690 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmaxnm");
7691 }
7692 case NEON::BI__builtin_neon_vpminnm_v:
7693 case NEON::BI__builtin_neon_vpminnmq_v: {
7694 Int = Intrinsic::aarch64_neon_fminnmp;
7695 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpminnm");
7696 }
7697 case NEON::BI__builtin_neon_vsqrth_f16: {
7698 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7699 Int = Intrinsic::sqrt;
7700 return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vsqrt");
7701 }
7702 case NEON::BI__builtin_neon_vsqrt_v:
7703 case NEON::BI__builtin_neon_vsqrtq_v: {
7704 Int = Intrinsic::sqrt;
7705 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
7706 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqrt");
7707 }
7708 case NEON::BI__builtin_neon_vrbit_v:
7709 case NEON::BI__builtin_neon_vrbitq_v: {
7710 Int = Intrinsic::aarch64_neon_rbit;
7711 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrbit");
7712 }
7713 case NEON::BI__builtin_neon_vaddv_u8:
7714 // FIXME: These are handled by the AArch64 scalar code.
7715 usgn = true;
7716 LLVM_FALLTHROUGH[[clang::fallthrough]];
7717 case NEON::BI__builtin_neon_vaddv_s8: {
7718 Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
7719 Ty = Int32Ty;
7720 VTy = llvm::VectorType::get(Int8Ty, 8);
7721 llvm::Type *Tys[2] = { Ty, VTy };
7722 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7723 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
7724 return Builder.CreateTrunc(Ops[0], Int8Ty);
7725 }
7726 case NEON::BI__builtin_neon_vaddv_u16:
7727 usgn = true;
7728 LLVM_FALLTHROUGH[[clang::fallthrough]];
7729 case NEON::BI__builtin_neon_vaddv_s16: {
7730 Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
7731 Ty = Int32Ty;
7732 VTy = llvm::VectorType::get(Int16Ty, 4);
7733 llvm::Type *Tys[2] = { Ty, VTy };
7734 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7735 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
7736 return Builder.CreateTrunc(Ops[0], Int16Ty);
7737 }
7738 case NEON::BI__builtin_neon_vaddvq_u8:
7739 usgn = true;
7740 LLVM_FALLTHROUGH[[clang::fallthrough]];
7741 case NEON::BI__builtin_neon_vaddvq_s8: {
7742 Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
7743 Ty = Int32Ty;
7744 VTy = llvm::VectorType::get(Int8Ty, 16);
7745 llvm::Type *Tys[2] = { Ty, VTy };
7746 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7747 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
7748 return Builder.CreateTrunc(Ops[0], Int8Ty);
7749 }
7750 case NEON::BI__builtin_neon_vaddvq_u16:
7751 usgn = true;
7752 LLVM_FALLTHROUGH[[clang::fallthrough]];
7753 case NEON::BI__builtin_neon_vaddvq_s16: {
7754 Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
7755 Ty = Int32Ty;
7756 VTy = llvm::VectorType::get(Int16Ty, 8);
7757 llvm::Type *Tys[2] = { Ty, VTy };
7758 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7759 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
7760 return Builder.CreateTrunc(Ops[0], Int16Ty);
7761 }
7762 case NEON::BI__builtin_neon_vmaxv_u8: {
7763 Int = Intrinsic::aarch64_neon_umaxv;
7764 Ty = Int32Ty;
7765 VTy = llvm::VectorType::get(Int8Ty, 8);
7766 llvm::Type *Tys[2] = { Ty, VTy };
7767 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7768 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7769 return Builder.CreateTrunc(Ops[0], Int8Ty);
7770 }
7771 case NEON::BI__builtin_neon_vmaxv_u16: {
7772 Int = Intrinsic::aarch64_neon_umaxv;
7773 Ty = Int32Ty;
7774 VTy = llvm::VectorType::get(Int16Ty, 4);
7775 llvm::Type *Tys[2] = { Ty, VTy };
7776 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7777 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7778 return Builder.CreateTrunc(Ops[0], Int16Ty);
7779 }
7780 case NEON::BI__builtin_neon_vmaxvq_u8: {
7781 Int = Intrinsic::aarch64_neon_umaxv;
7782 Ty = Int32Ty;
7783 VTy = llvm::VectorType::get(Int8Ty, 16);
7784 llvm::Type *Tys[2] = { Ty, VTy };
7785 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7786 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7787 return Builder.CreateTrunc(Ops[0], Int8Ty);
7788 }
7789 case NEON::BI__builtin_neon_vmaxvq_u16: {
7790 Int = Intrinsic::aarch64_neon_umaxv;
7791 Ty = Int32Ty;
7792 VTy = llvm::VectorType::get(Int16Ty, 8);
7793 llvm::Type *Tys[2] = { Ty, VTy };
7794 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7795 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7796 return Builder.CreateTrunc(Ops[0], Int16Ty);
7797 }
7798 case NEON::BI__builtin_neon_vmaxv_s8: {
7799 Int = Intrinsic::aarch64_neon_smaxv;
7800 Ty = Int32Ty;
7801 VTy = llvm::VectorType::get(Int8Ty, 8);
7802 llvm::Type *Tys[2] = { Ty, VTy };
7803 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7804 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7805 return Builder.CreateTrunc(Ops[0], Int8Ty);
7806 }
7807 case NEON::BI__builtin_neon_vmaxv_s16: {
7808 Int = Intrinsic::aarch64_neon_smaxv;
7809 Ty = Int32Ty;
7810 VTy = llvm::VectorType::get(Int16Ty, 4);
7811 llvm::Type *Tys[2] = { Ty, VTy };
7812 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7813 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7814 return Builder.CreateTrunc(Ops[0], Int16Ty);
7815 }
7816 case NEON::BI__builtin_neon_vmaxvq_s8: {
7817 Int = Intrinsic::aarch64_neon_smaxv;
7818 Ty = Int32Ty;
7819 VTy = llvm::VectorType::get(Int8Ty, 16);
7820 llvm::Type *Tys[2] = { Ty, VTy };
7821 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7822 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7823 return Builder.CreateTrunc(Ops[0], Int8Ty);
7824 }
7825 case NEON::BI__builtin_neon_vmaxvq_s16: {
7826 Int = Intrinsic::aarch64_neon_smaxv;
7827 Ty = Int32Ty;
7828 VTy = llvm::VectorType::get(Int16Ty, 8);
7829 llvm::Type *Tys[2] = { Ty, VTy };
7830 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7831 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7832 return Builder.CreateTrunc(Ops[0], Int16Ty);
7833 }
7834 case NEON::BI__builtin_neon_vmaxv_f16: {
7835 Int = Intrinsic::aarch64_neon_fmaxv;
7836 Ty = HalfTy;
7837 VTy = llvm::VectorType::get(HalfTy, 4);
7838 llvm::Type *Tys[2] = { Ty, VTy };
7839 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7840 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7841 return Builder.CreateTrunc(Ops[0], HalfTy);
7842 }
7843 case NEON::BI__builtin_neon_vmaxvq_f16: {
7844 Int = Intrinsic::aarch64_neon_fmaxv;
7845 Ty = HalfTy;
7846 VTy = llvm::VectorType::get(HalfTy, 8);
7847 llvm::Type *Tys[2] = { Ty, VTy };
7848 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7849 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
7850 return Builder.CreateTrunc(Ops[0], HalfTy);
7851 }
7852 case NEON::BI__builtin_neon_vminv_u8: {
7853 Int = Intrinsic::aarch64_neon_uminv;
7854 Ty = Int32Ty;
7855 VTy = llvm::VectorType::get(Int8Ty, 8);
7856 llvm::Type *Tys[2] = { Ty, VTy };
7857 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7858 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7859 return Builder.CreateTrunc(Ops[0], Int8Ty);
7860 }
7861 case NEON::BI__builtin_neon_vminv_u16: {
7862 Int = Intrinsic::aarch64_neon_uminv;
7863 Ty = Int32Ty;
7864 VTy = llvm::VectorType::get(Int16Ty, 4);
7865 llvm::Type *Tys[2] = { Ty, VTy };
7866 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7867 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7868 return Builder.CreateTrunc(Ops[0], Int16Ty);
7869 }
7870 case NEON::BI__builtin_neon_vminvq_u8: {
7871 Int = Intrinsic::aarch64_neon_uminv;
7872 Ty = Int32Ty;
7873 VTy = llvm::VectorType::get(Int8Ty, 16);
7874 llvm::Type *Tys[2] = { Ty, VTy };
7875 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7876 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7877 return Builder.CreateTrunc(Ops[0], Int8Ty);
7878 }
7879 case NEON::BI__builtin_neon_vminvq_u16: {
7880 Int = Intrinsic::aarch64_neon_uminv;
7881 Ty = Int32Ty;
7882 VTy = llvm::VectorType::get(Int16Ty, 8);
7883 llvm::Type *Tys[2] = { Ty, VTy };
7884 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7885 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7886 return Builder.CreateTrunc(Ops[0], Int16Ty);
7887 }
7888 case NEON::BI__builtin_neon_vminv_s8: {
7889 Int = Intrinsic::aarch64_neon_sminv;
7890 Ty = Int32Ty;
7891 VTy = llvm::VectorType::get(Int8Ty, 8);
7892 llvm::Type *Tys[2] = { Ty, VTy };
7893 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7894 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7895 return Builder.CreateTrunc(Ops[0], Int8Ty);
7896 }
7897 case NEON::BI__builtin_neon_vminv_s16: {
7898 Int = Intrinsic::aarch64_neon_sminv;
7899 Ty = Int32Ty;
7900 VTy = llvm::VectorType::get(Int16Ty, 4);
7901 llvm::Type *Tys[2] = { Ty, VTy };
7902 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7903 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7904 return Builder.CreateTrunc(Ops[0], Int16Ty);
7905 }
7906 case NEON::BI__builtin_neon_vminvq_s8: {
7907 Int = Intrinsic::aarch64_neon_sminv;
7908 Ty = Int32Ty;
7909 VTy = llvm::VectorType::get(Int8Ty, 16);
7910 llvm::Type *Tys[2] = { Ty, VTy };
7911 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7912 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7913 return Builder.CreateTrunc(Ops[0], Int8Ty);
7914 }
7915 case NEON::BI__builtin_neon_vminvq_s16: {
7916 Int = Intrinsic::aarch64_neon_sminv;
7917 Ty = Int32Ty;
7918 VTy = llvm::VectorType::get(Int16Ty, 8);
7919 llvm::Type *Tys[2] = { Ty, VTy };
7920 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7921 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7922 return Builder.CreateTrunc(Ops[0], Int16Ty);
7923 }
7924 case NEON::BI__builtin_neon_vminv_f16: {
7925 Int = Intrinsic::aarch64_neon_fminv;
7926 Ty = HalfTy;
7927 VTy = llvm::VectorType::get(HalfTy, 4);
7928 llvm::Type *Tys[2] = { Ty, VTy };
7929 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7930 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7931 return Builder.CreateTrunc(Ops[0], HalfTy);
7932 }
7933 case NEON::BI__builtin_neon_vminvq_f16: {
7934 Int = Intrinsic::aarch64_neon_fminv;
7935 Ty = HalfTy;
7936 VTy = llvm::VectorType::get(HalfTy, 8);
7937 llvm::Type *Tys[2] = { Ty, VTy };
7938 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7939 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
7940 return Builder.CreateTrunc(Ops[0], HalfTy);
7941 }
7942 case NEON::BI__builtin_neon_vmaxnmv_f16: {
7943 Int = Intrinsic::aarch64_neon_fmaxnmv;
7944 Ty = HalfTy;
7945 VTy = llvm::VectorType::get(HalfTy, 4);
7946 llvm::Type *Tys[2] = { Ty, VTy };
7947 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7948 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv");
7949 return Builder.CreateTrunc(Ops[0], HalfTy);
7950 }
7951 case NEON::BI__builtin_neon_vmaxnmvq_f16: {
7952 Int = Intrinsic::aarch64_neon_fmaxnmv;
7953 Ty = HalfTy;
7954 VTy = llvm::VectorType::get(HalfTy, 8);
7955 llvm::Type *Tys[2] = { Ty, VTy };
7956 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7957 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv");
7958 return Builder.CreateTrunc(Ops[0], HalfTy);
7959 }
7960 case NEON::BI__builtin_neon_vminnmv_f16: {
7961 Int = Intrinsic::aarch64_neon_fminnmv;
7962 Ty = HalfTy;
7963 VTy = llvm::VectorType::get(HalfTy, 4);
7964 llvm::Type *Tys[2] = { Ty, VTy };
7965 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7966 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv");
7967 return Builder.CreateTrunc(Ops[0], HalfTy);
7968 }
7969 case NEON::BI__builtin_neon_vminnmvq_f16: {
7970 Int = Intrinsic::aarch64_neon_fminnmv;
7971 Ty = HalfTy;
7972 VTy = llvm::VectorType::get(HalfTy, 8);
7973 llvm::Type *Tys[2] = { Ty, VTy };
7974 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7975 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv");
7976 return Builder.CreateTrunc(Ops[0], HalfTy);
7977 }
7978 case NEON::BI__builtin_neon_vmul_n_f64: {
7979 Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
7980 Value *RHS = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), DoubleTy);
7981 return Builder.CreateFMul(Ops[0], RHS);
7982 }
7983 case NEON::BI__builtin_neon_vaddlv_u8: {
7984 Int = Intrinsic::aarch64_neon_uaddlv;
7985 Ty = Int32Ty;
7986 VTy = llvm::VectorType::get(Int8Ty, 8);
7987 llvm::Type *Tys[2] = { Ty, VTy };
7988 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7989 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
7990 return Builder.CreateTrunc(Ops[0], Int16Ty);
7991 }
7992 case NEON::BI__builtin_neon_vaddlv_u16: {
7993 Int = Intrinsic::aarch64_neon_uaddlv;
7994 Ty = Int32Ty;
7995 VTy = llvm::VectorType::get(Int16Ty, 4);
7996 llvm::Type *Tys[2] = { Ty, VTy };
7997 Ops.push_back(EmitScalarExpr(E->getArg(0)));
7998 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
7999 }
8000 case NEON::BI__builtin_neon_vaddlvq_u8: {
8001 Int = Intrinsic::aarch64_neon_uaddlv;
8002 Ty = Int32Ty;
8003 VTy = llvm::VectorType::get(Int8Ty, 16);
8004 llvm::Type *Tys[2] = { Ty, VTy };
8005 Ops.push_back(EmitScalarExpr(E->getArg(0)));
8006 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8007 return Builder.CreateTrunc(Ops[0], Int16Ty);
8008 }
8009 case NEON::BI__builtin_neon_vaddlvq_u16: {
8010 Int = Intrinsic::aarch64_neon_uaddlv;
8011 Ty = Int32Ty;
8012 VTy = llvm::VectorType::get(Int16Ty, 8);
8013 llvm::Type *Tys[2] = { Ty, VTy };
8014 Ops.push_back(EmitScalarExpr(E->getArg(0)));
8015 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8016 }
8017 case NEON::BI__builtin_neon_vaddlv_s8: {
8018 Int = Intrinsic::aarch64_neon_saddlv;
8019 Ty = Int32Ty;
8020 VTy = llvm::VectorType::get(Int8Ty, 8);
8021 llvm::Type *Tys[2] = { Ty, VTy };
8022 Ops.push_back(EmitScalarExpr(E->getArg(0)));
8023 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8024 return Builder.CreateTrunc(Ops[0], Int16Ty);
8025 }
8026 case NEON::BI__builtin_neon_vaddlv_s16: {
8027 Int = Intrinsic::aarch64_neon_saddlv;
8028 Ty = Int32Ty;
8029 VTy = llvm::VectorType::get(Int16Ty, 4);
8030 llvm::Type *Tys[2] = { Ty, VTy };
8031 Ops.push_back(EmitScalarExpr(E->getArg(0)));
8032 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8033 }
8034 case NEON::BI__builtin_neon_vaddlvq_s8: {
8035 Int = Intrinsic::aarch64_neon_saddlv;
8036 Ty = Int32Ty;
8037 VTy = llvm::VectorType::get(Int8Ty, 16);
8038 llvm::Type *Tys[2] = { Ty, VTy };
8039 Ops.push_back(EmitScalarExpr(E->getArg(0)));
8040 Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8041 return Builder.CreateTrunc(Ops[0], Int16Ty);
8042 }
8043 case NEON::BI__builtin_neon_vaddlvq_s16: {
8044 Int = Intrinsic::aarch64_neon_saddlv;
8045 Ty = Int32Ty;
8046 VTy = llvm::VectorType::get(Int16Ty, 8);
8047 llvm::Type *Tys[2] = { Ty, VTy };
8048 Ops.push_back(EmitScalarExpr(E->getArg(0)));
8049 return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8050 }
8051 case NEON::BI__builtin_neon_vsri_n_v:
8052 case NEON::BI__builtin_neon_vsriq_n_v: {
8053 Int = Intrinsic::aarch64_neon_vsri;
8054 llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty);
8055 return EmitNeonCall(Intrin, Ops, "vsri_n");
8056 }
8057 case NEON::BI__builtin_neon_vsli_n_v:
8058 case NEON::BI__builtin_neon_vsliq_n_v: {
8059 Int = Intrinsic::aarch64_neon_vsli;
8060 llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty);
8061 return EmitNeonCall(Intrin, Ops, "vsli_n");
8062 }
8063 case NEON::BI__builtin_neon_vsra_n_v:
8064 case NEON::BI__builtin_neon_vsraq_n_v:
8065 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8066 Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n");
8067 return Builder.CreateAdd(Ops[0], Ops[1]);
8068 case NEON::BI__builtin_neon_vrsra_n_v:
8069 case NEON::BI__builtin_neon_vrsraq_n_v: {
8070 Int = usgn ? Intrinsic::aarch64_neon_urshl : Intrinsic::aarch64_neon_srshl;
8071 SmallVector<llvm::Value*,2> TmpOps;
8072 TmpOps.push_back(Ops[1]);
8073 TmpOps.push_back(Ops[2]);
8074 Function* F = CGM.getIntrinsic(Int, Ty);
8075 llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vrshr_n", 1, true);
8076 Ops[0] = Builder.CreateBitCast(Ops[0], VTy);
8077 return Builder.CreateAdd(Ops[0], tmp);
8078 }
8079 case NEON::BI__builtin_neon_vld1_v:
8080 case NEON::BI__builtin_neon_vld1q_v: {
8081 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy));
8082 auto Alignment = CharUnits::fromQuantity(
8083 BuiltinID == NEON::BI__builtin_neon_vld1_v ? 8 : 16);
8084 return Builder.CreateAlignedLoad(VTy, Ops[0], Alignment);
8085 }
8086 case NEON::BI__builtin_neon_vst1_v:
8087 case NEON::BI__builtin_neon_vst1q_v:
8088 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy));
8089 Ops[1] = Builder.CreateBitCast(Ops[1], VTy);
8090 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8091 case NEON::BI__builtin_neon_vld1_lane_v:
8092 case NEON::BI__builtin_neon_vld1q_lane_v: {
8093 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8094 Ty = llvm::PointerType::getUnqual(VTy->getElementType());
8095 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8096 auto Alignment = CharUnits::fromQuantity(
8097 BuiltinID == NEON::BI__builtin_neon_vld1_lane_v ? 8 : 16);
8098 Ops[0] =
8099 Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment);
8100 return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vld1_lane");
8101 }
8102 case NEON::BI__builtin_neon_vld1_dup_v:
8103 case NEON::BI__builtin_neon_vld1q_dup_v: {
8104 Value *V = UndefValue::get(Ty);
8105 Ty = llvm::PointerType::getUnqual(VTy->getElementType());
8106 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8107 auto Alignment = CharUnits::fromQuantity(
8108 BuiltinID == NEON::BI__builtin_neon_vld1_dup_v ? 8 : 16);
8109 Ops[0] =
8110 Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment);
8111 llvm::Constant *CI = ConstantInt::get(Int32Ty, 0);
8112 Ops[0] = Builder.CreateInsertElement(V, Ops[0], CI);
8113 return EmitNeonSplat(Ops[0], CI);
8114 }
8115 case NEON::BI__builtin_neon_vst1_lane_v:
8116 case NEON::BI__builtin_neon_vst1q_lane_v:
8117 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8118 Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]);
8119 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
8120 return Builder.CreateDefaultAlignedStore(Ops[1],
8121 Builder.CreateBitCast(Ops[0], Ty));
8122 case NEON::BI__builtin_neon_vld2_v:
8123 case NEON::BI__builtin_neon_vld2q_v: {
8124 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
8125 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8126 llvm::Type *Tys[2] = { VTy, PTy };
8127 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2, Tys);
8128 Ops[1] = Builder.CreateCall(F, Ops[1], "vld2");
8129 Ops[0] = Builder.CreateBitCast(Ops[0],
8130 llvm::PointerType::getUnqual(Ops[1]->getType()));
8131 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8132 }
8133 case NEON::BI__builtin_neon_vld3_v:
8134 case NEON::BI__builtin_neon_vld3q_v: {
8135 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
8136 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8137 llvm::Type *Tys[2] = { VTy, PTy };
8138 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3, Tys);
8139 Ops[1] = Builder.CreateCall(F, Ops[1], "vld3");
8140 Ops[0] = Builder.CreateBitCast(Ops[0],
8141 llvm::PointerType::getUnqual(Ops[1]->getType()));
8142 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8143 }
8144 case NEON::BI__builtin_neon_vld4_v:
8145 case NEON::BI__builtin_neon_vld4q_v: {
8146 llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
8147 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8148 llvm::Type *Tys[2] = { VTy, PTy };
8149 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4, Tys);
8150 Ops[1] = Builder.CreateCall(F, Ops[1], "vld4");
8151 Ops[0] = Builder.CreateBitCast(Ops[0],
8152 llvm::PointerType::getUnqual(Ops[1]->getType()));
8153 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8154 }
8155 case NEON::BI__builtin_neon_vld2_dup_v:
8156 case NEON::BI__builtin_neon_vld2q_dup_v: {
8157 llvm::Type *PTy =
8158 llvm::PointerType::getUnqual(VTy->getElementType());
8159 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8160 llvm::Type *Tys[2] = { VTy, PTy };
8161 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2r, Tys);
8162 Ops[1] = Builder.CreateCall(F, Ops[1], "vld2");
8163 Ops[0] = Builder.CreateBitCast(Ops[0],
8164 llvm::PointerType::getUnqual(Ops[1]->getType()));
8165 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8166 }
8167 case NEON::BI__builtin_neon_vld3_dup_v:
8168 case NEON::BI__builtin_neon_vld3q_dup_v: {
8169 llvm::Type *PTy =
8170 llvm::PointerType::getUnqual(VTy->getElementType());
8171 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8172 llvm::Type *Tys[2] = { VTy, PTy };
8173 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3r, Tys);
8174 Ops[1] = Builder.CreateCall(F, Ops[1], "vld3");
8175 Ops[0] = Builder.CreateBitCast(Ops[0],
8176 llvm::PointerType::getUnqual(Ops[1]->getType()));
8177 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8178 }
8179 case NEON::BI__builtin_neon_vld4_dup_v:
8180 case NEON::BI__builtin_neon_vld4q_dup_v: {
8181 llvm::Type *PTy =
8182 llvm::PointerType::getUnqual(VTy->getElementType());
8183 Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8184 llvm::Type *Tys[2] = { VTy, PTy };
8185 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4r, Tys);
8186 Ops[1] = Builder.CreateCall(F, Ops[1], "vld4");
8187 Ops[0] = Builder.CreateBitCast(Ops[0],
8188 llvm::PointerType::getUnqual(Ops[1]->getType()));
8189 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8190 }
8191 case NEON::BI__builtin_neon_vld2_lane_v:
8192 case NEON::BI__builtin_neon_vld2q_lane_v: {
8193 llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
8194 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2lane, Tys);
8195 Ops.push_back(Ops[1]);
8196 Ops.erase(Ops.begin()+1);
8197 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8198 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8199 Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty);
8200 Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld2_lane");
8201 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
8202 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8203 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8204 }
8205 case NEON::BI__builtin_neon_vld3_lane_v:
8206 case NEON::BI__builtin_neon_vld3q_lane_v: {
8207 llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
8208 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3lane, Tys);
8209 Ops.push_back(Ops[1]);
8210 Ops.erase(Ops.begin()+1);
8211 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8212 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8213 Ops[3] = Builder.CreateBitCast(Ops[3], Ty);
8214 Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty);
8215 Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld3_lane");
8216 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
8217 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8218 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8219 }
8220 case NEON::BI__builtin_neon_vld4_lane_v:
8221 case NEON::BI__builtin_neon_vld4q_lane_v: {
8222 llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
8223 Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4lane, Tys);
8224 Ops.push_back(Ops[1]);
8225 Ops.erase(Ops.begin()+1);
8226 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8227 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8228 Ops[3] = Builder.CreateBitCast(Ops[3], Ty);
8229 Ops[4] = Builder.CreateBitCast(Ops[4], Ty);
8230 Ops[5] = Builder.CreateZExt(Ops[5], Int64Ty);
8231 Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld4_lane");
8232 Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
8233 Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8234 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8235 }
8236 case NEON::BI__builtin_neon_vst2_v:
8237 case NEON::BI__builtin_neon_vst2q_v: {
8238 Ops.push_back(Ops[0]);
8239 Ops.erase(Ops.begin());
8240 llvm::Type *Tys[2] = { VTy, Ops[2]->getType() };
8241 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2, Tys),
8242 Ops, "");
8243 }
8244 case NEON::BI__builtin_neon_vst2_lane_v:
8245 case NEON::BI__builtin_neon_vst2q_lane_v: {
8246 Ops.push_back(Ops[0]);
8247 Ops.erase(Ops.begin());
8248 Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty);
8249 llvm::Type *Tys[2] = { VTy, Ops[3]->getType() };
8250 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2lane, Tys),
8251 Ops, "");
8252 }
8253 case NEON::BI__builtin_neon_vst3_v:
8254 case NEON::BI__builtin_neon_vst3q_v: {
8255 Ops.push_back(Ops[0]);
8256 Ops.erase(Ops.begin());
8257 llvm::Type *Tys[2] = { VTy, Ops[3]->getType() };
8258 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3, Tys),
8259 Ops, "");
8260 }
8261 case NEON::BI__builtin_neon_vst3_lane_v:
8262 case NEON::BI__builtin_neon_vst3q_lane_v: {
8263 Ops.push_back(Ops[0]);
8264 Ops.erase(Ops.begin());
8265 Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty);
8266 llvm::Type *Tys[2] = { VTy, Ops[4]->getType() };
8267 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3lane, Tys),
8268 Ops, "");
8269 }
8270 case NEON::BI__builtin_neon_vst4_v:
8271 case NEON::BI__builtin_neon_vst4q_v: {
8272 Ops.push_back(Ops[0]);
8273 Ops.erase(Ops.begin());
8274 llvm::Type *Tys[2] = { VTy, Ops[4]->getType() };
8275 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4, Tys),
8276 Ops, "");
8277 }
8278 case NEON::BI__builtin_neon_vst4_lane_v:
8279 case NEON::BI__builtin_neon_vst4q_lane_v: {
8280 Ops.push_back(Ops[0]);
8281 Ops.erase(Ops.begin());
8282 Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty);
8283 llvm::Type *Tys[2] = { VTy, Ops[5]->getType() };
8284 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4lane, Tys),
8285 Ops, "");
8286 }
8287 case NEON::BI__builtin_neon_vtrn_v:
8288 case NEON::BI__builtin_neon_vtrnq_v: {
8289 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
8290 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8291 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8292 Value *SV = nullptr;
8293
8294 for (unsigned vi = 0; vi != 2; ++vi) {
8295 SmallVector<uint32_t, 16> Indices;
8296 for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
8297 Indices.push_back(i+vi);
8298 Indices.push_back(i+e+vi);
8299 }
8300 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
8301 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn");
8302 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
8303 }
8304 return SV;
8305 }
8306 case NEON::BI__builtin_neon_vuzp_v:
8307 case NEON::BI__builtin_neon_vuzpq_v: {
8308 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
8309 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8310 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8311 Value *SV = nullptr;
8312
8313 for (unsigned vi = 0; vi != 2; ++vi) {
8314 SmallVector<uint32_t, 16> Indices;
8315 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
8316 Indices.push_back(2*i+vi);
8317
8318 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
8319 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp");
8320 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
8321 }
8322 return SV;
8323 }
8324 case NEON::BI__builtin_neon_vzip_v:
8325 case NEON::BI__builtin_neon_vzipq_v: {
8326 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
8327 Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8328 Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8329 Value *SV = nullptr;
8330
8331 for (unsigned vi = 0; vi != 2; ++vi) {
8332 SmallVector<uint32_t, 16> Indices;
8333 for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
8334 Indices.push_back((i + vi*e) >> 1);
8335 Indices.push_back(((i + vi*e) >> 1)+e);
8336 }
8337 Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
8338 SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip");
8339 SV = Builder.CreateDefaultAlignedStore(SV, Addr);
8340 }
8341 return SV;
8342 }
8343 case NEON::BI__builtin_neon_vqtbl1q_v: {
8344 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl1, Ty),
8345 Ops, "vtbl1");
8346 }
8347 case NEON::BI__builtin_neon_vqtbl2q_v: {
8348 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl2, Ty),
8349 Ops, "vtbl2");
8350 }
8351 case NEON::BI__builtin_neon_vqtbl3q_v: {
8352 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl3, Ty),
8353 Ops, "vtbl3");
8354 }
8355 case NEON::BI__builtin_neon_vqtbl4q_v: {
8356 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl4, Ty),
8357 Ops, "vtbl4");
8358 }
8359 case NEON::BI__builtin_neon_vqtbx1q_v: {
8360 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx1, Ty),
8361 Ops, "vtbx1");
8362 }
8363 case NEON::BI__builtin_neon_vqtbx2q_v: {
8364 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx2, Ty),
8365 Ops, "vtbx2");
8366 }
8367 case NEON::BI__builtin_neon_vqtbx3q_v: {
8368 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx3, Ty),
8369 Ops, "vtbx3");
8370 }
8371 case NEON::BI__builtin_neon_vqtbx4q_v: {
8372 return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx4, Ty),
8373 Ops, "vtbx4");
8374 }
8375 case NEON::BI__builtin_neon_vsqadd_v:
8376 case NEON::BI__builtin_neon_vsqaddq_v: {
8377 Int = Intrinsic::aarch64_neon_usqadd;
8378 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqadd");
8379 }
8380 case NEON::BI__builtin_neon_vuqadd_v:
8381 case NEON::BI__builtin_neon_vuqaddq_v: {
8382 Int = Intrinsic::aarch64_neon_suqadd;
8383 return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vuqadd");
8384 }
8385 case AArch64::BI__iso_volatile_load8:
8386 case AArch64::BI__iso_volatile_load16:
8387 case AArch64::BI__iso_volatile_load32:
8388 case AArch64::BI__iso_volatile_load64:
8389 return EmitISOVolatileLoad(E);
8390 case AArch64::BI__iso_volatile_store8:
8391 case AArch64::BI__iso_volatile_store16:
8392 case AArch64::BI__iso_volatile_store32:
8393 case AArch64::BI__iso_volatile_store64:
8394 return EmitISOVolatileStore(E);
8395 case AArch64::BI_BitScanForward:
8396 case AArch64::BI_BitScanForward64:
8397 return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E);
8398 case AArch64::BI_BitScanReverse:
8399 case AArch64::BI_BitScanReverse64:
8400 return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E);
8401 case AArch64::BI_InterlockedAnd64:
8402 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E);
8403 case AArch64::BI_InterlockedExchange64:
8404 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E);
8405 case AArch64::BI_InterlockedExchangeAdd64:
8406 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E);
8407 case AArch64::BI_InterlockedExchangeSub64:
8408 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E);
8409 case AArch64::BI_InterlockedOr64:
8410 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E);
8411 case AArch64::BI_InterlockedXor64:
8412 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E);
8413 case AArch64::BI_InterlockedDecrement64:
8414 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E);
8415 case AArch64::BI_InterlockedIncrement64:
8416 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E);
8417 }
8418}
8419
8420llvm::Value *CodeGenFunction::
8421BuildVector(ArrayRef<llvm::Value*> Ops) {
8422 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8423, __extension__ __PRETTY_FUNCTION__))
8423 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8423, __extension__ __PRETTY_FUNCTION__))
;
8424 bool AllConstants = true;
8425 for (unsigned i = 0, e = Ops.size(); i != e && AllConstants; ++i)
8426 AllConstants &= isa<Constant>(Ops[i]);
8427
8428 // If this is a constant vector, create a ConstantVector.
8429 if (AllConstants) {
8430 SmallVector<llvm::Constant*, 16> CstOps;
8431 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
8432 CstOps.push_back(cast<Constant>(Ops[i]));
8433 return llvm::ConstantVector::get(CstOps);
8434 }
8435
8436 // Otherwise, insertelement the values to build the vector.
8437 Value *Result =
8438 llvm::UndefValue::get(llvm::VectorType::get(Ops[0]->getType(), Ops.size()));
8439
8440 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
8441 Result = Builder.CreateInsertElement(Result, Ops[i], Builder.getInt32(i));
8442
8443 return Result;
8444}
8445
8446// Convert the mask from an integer type to a vector of i1.
8447static Value *getMaskVecValue(CodeGenFunction &CGF, Value *Mask,
8448 unsigned NumElts) {
8449
8450 llvm::VectorType *MaskTy = llvm::VectorType::get(CGF.Builder.getInt1Ty(),
8451 cast<IntegerType>(Mask->getType())->getBitWidth());
8452 Value *MaskVec = CGF.Builder.CreateBitCast(Mask, MaskTy);
8453
8454 // If we have less than 8 elements, then the starting mask was an i8 and
8455 // we need to extract down to the right number of elements.
8456 if (NumElts < 8) {
8457 uint32_t Indices[4];
8458 for (unsigned i = 0; i != NumElts; ++i)
8459 Indices[i] = i;
8460 MaskVec = CGF.Builder.CreateShuffleVector(MaskVec, MaskVec,
8461 makeArrayRef(Indices, NumElts),
8462 "extract");
8463 }
8464 return MaskVec;
8465}
8466
8467static Value *EmitX86MaskedStore(CodeGenFunction &CGF,
8468 ArrayRef<Value *> Ops,
8469 unsigned Align) {
8470 // Cast the pointer to right type.
8471 Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
8472 llvm::PointerType::getUnqual(Ops[1]->getType()));
8473
8474 Value *MaskVec = getMaskVecValue(CGF, Ops[2],
8475 Ops[1]->getType()->getVectorNumElements());
8476
8477 return CGF.Builder.CreateMaskedStore(Ops[1], Ptr, Align, MaskVec);
8478}
8479
8480static Value *EmitX86MaskedLoad(CodeGenFunction &CGF,
8481 ArrayRef<Value *> Ops, unsigned Align) {
8482 // Cast the pointer to right type.
8483 Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
8484 llvm::PointerType::getUnqual(Ops[1]->getType()));
8485
8486 Value *MaskVec = getMaskVecValue(CGF, Ops[2],
8487 Ops[1]->getType()->getVectorNumElements());
8488
8489 return CGF.Builder.CreateMaskedLoad(Ptr, Align, MaskVec, Ops[1]);
8490}
8491
8492static Value *EmitX86ExpandLoad(CodeGenFunction &CGF,
8493 ArrayRef<Value *> Ops) {
8494 llvm::Type *ResultTy = Ops[1]->getType();
8495 llvm::Type *PtrTy = ResultTy->getVectorElementType();
8496
8497 // Cast the pointer to element type.
8498 Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
8499 llvm::PointerType::getUnqual(PtrTy));
8500
8501 Value *MaskVec = getMaskVecValue(CGF, Ops[2],
8502 ResultTy->getVectorNumElements());
8503
8504 llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_expandload,
8505 ResultTy);
8506 return CGF.Builder.CreateCall(F, { Ptr, MaskVec, Ops[1] });
8507}
8508
8509static Value *EmitX86CompressStore(CodeGenFunction &CGF,
8510 ArrayRef<Value *> Ops) {
8511 llvm::Type *ResultTy = Ops[1]->getType();
8512 llvm::Type *PtrTy = ResultTy->getVectorElementType();
8513
8514 // Cast the pointer to element type.
8515 Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
8516 llvm::PointerType::getUnqual(PtrTy));
8517
8518 Value *MaskVec = getMaskVecValue(CGF, Ops[2],
8519 ResultTy->getVectorNumElements());
8520
8521 llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_compressstore,
8522 ResultTy);
8523 return CGF.Builder.CreateCall(F, { Ops[1], Ptr, MaskVec });
8524}
8525
8526static Value *EmitX86MaskLogic(CodeGenFunction &CGF, Instruction::BinaryOps Opc,
8527 unsigned NumElts, ArrayRef<Value *> Ops,
8528 bool InvertLHS = false) {
8529 Value *LHS = getMaskVecValue(CGF, Ops[0], NumElts);
8530 Value *RHS = getMaskVecValue(CGF, Ops[1], NumElts);
8531
8532 if (InvertLHS)
8533 LHS = CGF.Builder.CreateNot(LHS);
8534
8535 return CGF.Builder.CreateBitCast(CGF.Builder.CreateBinOp(Opc, LHS, RHS),
8536 CGF.Builder.getIntNTy(std::max(NumElts, 8U)));
8537}
8538
8539static Value *EmitX86Select(CodeGenFunction &CGF,
8540 Value *Mask, Value *Op0, Value *Op1) {
8541
8542 // If the mask is all ones just return first argument.
8543 if (const auto *C = dyn_cast<Constant>(Mask))
8544 if (C->isAllOnesValue())
8545 return Op0;
8546
8547 Mask = getMaskVecValue(CGF, Mask, Op0->getType()->getVectorNumElements());
8548
8549 return CGF.Builder.CreateSelect(Mask, Op0, Op1);
8550}
8551
8552static Value *EmitX86ScalarSelect(CodeGenFunction &CGF,
8553 Value *Mask, Value *Op0, Value *Op1) {
8554 // If the mask is all ones just return first argument.
8555 if (const auto *C = dyn_cast<Constant>(Mask))
8556 if (C->isAllOnesValue())
8557 return Op0;
8558
8559 llvm::VectorType *MaskTy =
8560 llvm::VectorType::get(CGF.Builder.getInt1Ty(),
8561 Mask->getType()->getIntegerBitWidth());
8562 Mask = CGF.Builder.CreateBitCast(Mask, MaskTy);
8563 Mask = CGF.Builder.CreateExtractElement(Mask, (uint64_t)0);
8564 return CGF.Builder.CreateSelect(Mask, Op0, Op1);
8565}
8566
8567static Value *EmitX86MaskedCompareResult(CodeGenFunction &CGF, Value *Cmp,
8568 unsigned NumElts, Value *MaskIn) {
8569 if (MaskIn) {
8570 const auto *C = dyn_cast<Constant>(MaskIn);
8571 if (!C || !C->isAllOnesValue())
8572 Cmp = CGF.Builder.CreateAnd(Cmp, getMaskVecValue(CGF, MaskIn, NumElts));
8573 }
8574
8575 if (NumElts < 8) {
8576 uint32_t Indices[8];
8577 for (unsigned i = 0; i != NumElts; ++i)
8578 Indices[i] = i;
8579 for (unsigned i = NumElts; i != 8; ++i)
8580 Indices[i] = i % NumElts + NumElts;
8581 Cmp = CGF.Builder.CreateShuffleVector(
8582 Cmp, llvm::Constant::getNullValue(Cmp->getType()), Indices);
8583 }
8584
8585 return CGF.Builder.CreateBitCast(Cmp,
8586 IntegerType::get(CGF.getLLVMContext(),
8587 std::max(NumElts, 8U)));
8588}
8589
8590static Value *EmitX86MaskedCompare(CodeGenFunction &CGF, unsigned CC,
8591 bool Signed, ArrayRef<Value *> Ops) {
8592 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8593, __extension__ __PRETTY_FUNCTION__))
8593 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8593, __extension__ __PRETTY_FUNCTION__))
;
8594 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
8595 Value *Cmp;
8596
8597 if (CC == 3) {
8598 Cmp = Constant::getNullValue(
8599 llvm::VectorType::get(CGF.Builder.getInt1Ty(), NumElts));
8600 } else if (CC == 7) {
8601 Cmp = Constant::getAllOnesValue(
8602 llvm::VectorType::get(CGF.Builder.getInt1Ty(), NumElts));
8603 } else {
8604 ICmpInst::Predicate Pred;
8605 switch (CC) {
8606 default: llvm_unreachable("Unknown condition code")::llvm::llvm_unreachable_internal("Unknown condition code", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8606)
;
8607 case 0: Pred = ICmpInst::ICMP_EQ; break;
8608 case 1: Pred = Signed ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; break;
8609 case 2: Pred = Signed ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; break;
8610 case 4: Pred = ICmpInst::ICMP_NE; break;
8611 case 5: Pred = Signed ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; break;
8612 case 6: Pred = Signed ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; break;
8613 }
8614 Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]);
8615 }
8616
8617 Value *MaskIn = nullptr;
8618 if (Ops.size() == 4)
8619 MaskIn = Ops[3];
8620
8621 return EmitX86MaskedCompareResult(CGF, Cmp, NumElts, MaskIn);
8622}
8623
8624static Value *EmitX86ConvertToMask(CodeGenFunction &CGF, Value *In) {
8625 Value *Zero = Constant::getNullValue(In->getType());
8626 return EmitX86MaskedCompare(CGF, 1, true, { In, Zero });
8627}
8628
8629static Value *EmitX86Abs(CodeGenFunction &CGF, ArrayRef<Value *> Ops) {
8630
8631 llvm::Type *Ty = Ops[0]->getType();
8632 Value *Zero = llvm::Constant::getNullValue(Ty);
8633 Value *Sub = CGF.Builder.CreateSub(Zero, Ops[0]);
8634 Value *Cmp = CGF.Builder.CreateICmp(ICmpInst::ICMP_SGT, Ops[0], Zero);
8635 Value *Res = CGF.Builder.CreateSelect(Cmp, Ops[0], Sub);
8636 return Res;
8637}
8638
8639static Value *EmitX86MinMax(CodeGenFunction &CGF, ICmpInst::Predicate Pred,
8640 ArrayRef<Value *> Ops) {
8641 Value *Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]);
8642 Value *Res = CGF.Builder.CreateSelect(Cmp, Ops[0], Ops[1]);
8643
8644 assert(Ops.size() == 2)(static_cast <bool> (Ops.size() == 2) ? void (0) : __assert_fail
("Ops.size() == 2", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8644, __extension__ __PRETTY_FUNCTION__))
;
8645 return Res;
8646}
8647
8648// Lowers X86 FMA intrinsics to IR.
8649static Value *EmitX86FMAExpr(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
8650 unsigned BuiltinID, bool IsAddSub) {
8651
8652 bool Subtract = false;
8653 Intrinsic::ID IID = Intrinsic::not_intrinsic;
8654 switch (BuiltinID) {
8655 default: break;
8656 case clang::X86::BI__builtin_ia32_vfmsubps512_mask3:
8657 Subtract = true;
8658 LLVM_FALLTHROUGH[[clang::fallthrough]];
8659 case clang::X86::BI__builtin_ia32_vfmaddps512_mask:
8660 case clang::X86::BI__builtin_ia32_vfmaddps512_maskz:
8661 case clang::X86::BI__builtin_ia32_vfmaddps512_mask3:
8662 IID = llvm::Intrinsic::x86_avx512_vfmadd_ps_512; break;
8663 case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3:
8664 Subtract = true;
8665 LLVM_FALLTHROUGH[[clang::fallthrough]];
8666 case clang::X86::BI__builtin_ia32_vfmaddpd512_mask:
8667 case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz:
8668 case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3:
8669 IID = llvm::Intrinsic::x86_avx512_vfmadd_pd_512; break;
8670 case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3:
8671 Subtract = true;
8672 LLVM_FALLTHROUGH[[clang::fallthrough]];
8673 case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask:
8674 case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz:
8675 case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3:
8676 IID = llvm::Intrinsic::x86_avx512_vfmaddsub_ps_512;
8677 break;
8678 case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
8679 Subtract = true;
8680 LLVM_FALLTHROUGH[[clang::fallthrough]];
8681 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask:
8682 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
8683 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
8684 IID = llvm::Intrinsic::x86_avx512_vfmaddsub_pd_512;
8685 break;
8686 }
8687
8688 Value *A = Ops[0];
8689 Value *B = Ops[1];
8690 Value *C = Ops[2];
8691
8692 if (Subtract)
8693 C = CGF.Builder.CreateFNeg(C);
8694
8695 Value *Res;
8696
8697 // Only handle in case of _MM_FROUND_CUR_DIRECTION/4 (no rounding).
8698 if (IID != Intrinsic::not_intrinsic &&
8699 cast<llvm::ConstantInt>(Ops.back())->getZExtValue() != (uint64_t)4) {
8700 Function *Intr = CGF.CGM.getIntrinsic(IID);
8701 Res = CGF.Builder.CreateCall(Intr, {A, B, C, Ops.back() });
8702 } else {
8703 llvm::Type *Ty = A->getType();
8704 Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ty);
8705 Res = CGF.Builder.CreateCall(FMA, {A, B, C} );
8706
8707 if (IsAddSub) {
8708 // Negate even elts in C using a mask.
8709 unsigned NumElts = Ty->getVectorNumElements();
8710 SmallVector<uint32_t, 16> Indices(NumElts);
8711 for (unsigned i = 0; i != NumElts; ++i)
8712 Indices[i] = i + (i % 2) * NumElts;
8713
8714 Value *NegC = CGF.Builder.CreateFNeg(C);
8715 Value *FMSub = CGF.Builder.CreateCall(FMA, {A, B, NegC} );
8716 Res = CGF.Builder.CreateShuffleVector(FMSub, Res, Indices);
8717 }
8718 }
8719
8720 // Handle any required masking.
8721 Value *MaskFalseVal = nullptr;
8722 switch (BuiltinID) {
8723 case clang::X86::BI__builtin_ia32_vfmaddps512_mask:
8724 case clang::X86::BI__builtin_ia32_vfmaddpd512_mask:
8725 case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask:
8726 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask:
8727 MaskFalseVal = Ops[0];
8728 break;
8729 case clang::X86::BI__builtin_ia32_vfmaddps512_maskz:
8730 case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz:
8731 case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz:
8732 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
8733 MaskFalseVal = Constant::getNullValue(Ops[0]->getType());
8734 break;
8735 case clang::X86::BI__builtin_ia32_vfmsubps512_mask3:
8736 case clang::X86::BI__builtin_ia32_vfmaddps512_mask3:
8737 case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3:
8738 case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3:
8739 case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3:
8740 case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3:
8741 case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
8742 case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
8743 MaskFalseVal = Ops[2];
8744 break;
8745 }
8746
8747 if (MaskFalseVal)
8748 return EmitX86Select(CGF, Ops[3], Res, MaskFalseVal);
8749
8750 return Res;
8751}
8752
8753static Value *
8754EmitScalarFMAExpr(CodeGenFunction &CGF, MutableArrayRef<Value *> Ops,
8755 Value *Upper, bool ZeroMask = false, unsigned PTIdx = 0,
8756 bool NegAcc = false) {
8757 unsigned Rnd = 4;
8758 if (Ops.size() > 4)
8759 Rnd = cast<llvm::ConstantInt>(Ops[4])->getZExtValue();
8760
8761 if (NegAcc)
8762 Ops[2] = CGF.Builder.CreateFNeg(Ops[2]);
8763
8764 Ops[0] = CGF.Builder.CreateExtractElement(Ops[0], (uint64_t)0);
8765 Ops[1] = CGF.Builder.CreateExtractElement(Ops[1], (uint64_t)0);
8766 Ops[2] = CGF.Builder.CreateExtractElement(Ops[2], (uint64_t)0);
8767 Value *Res;
8768 if (Rnd != 4) {
8769 Intrinsic::ID IID = Ops[0]->getType()->getPrimitiveSizeInBits() == 32 ?
8770 Intrinsic::x86_avx512_vfmadd_f32 :
8771 Intrinsic::x86_avx512_vfmadd_f64;
8772 Res = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID),
8773 {Ops[0], Ops[1], Ops[2], Ops[4]});
8774 } else {
8775 Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ops[0]->getType());
8776 Res = CGF.Builder.CreateCall(FMA, Ops.slice(0, 3));
8777 }
8778 // If we have more than 3 arguments, we need to do masking.
8779 if (Ops.size() > 3) {
8780 Value *PassThru = ZeroMask ? Constant::getNullValue(Res->getType())
8781 : Ops[PTIdx];
8782
8783 // If we negated the accumulator and the its the PassThru value we need to
8784 // bypass the negate. Conveniently Upper should be the same thing in this
8785 // case.
8786 if (NegAcc && PTIdx == 2)
8787 PassThru = CGF.Builder.CreateExtractElement(Upper, (uint64_t)0);
8788
8789 Res = EmitX86ScalarSelect(CGF, Ops[3], Res, PassThru);
8790 }
8791 return CGF.Builder.CreateInsertElement(Upper, Res, (uint64_t)0);
8792}
8793
8794static Value *EmitX86Muldq(CodeGenFunction &CGF, bool IsSigned,
8795 ArrayRef<Value *> Ops) {
8796 llvm::Type *Ty = Ops[0]->getType();
8797 // Arguments have a vXi32 type so cast to vXi64.
8798 Ty = llvm::VectorType::get(CGF.Int64Ty,
8799 Ty->getPrimitiveSizeInBits() / 64);
8800 Value *LHS = CGF.Builder.CreateBitCast(Ops[0], Ty);
8801 Value *RHS = CGF.Builder.CreateBitCast(Ops[1], Ty);
8802
8803 if (IsSigned) {
8804 // Shift left then arithmetic shift right.
8805 Constant *ShiftAmt = ConstantInt::get(Ty, 32);
8806 LHS = CGF.Builder.CreateShl(LHS, ShiftAmt);
8807 LHS = CGF.Builder.CreateAShr(LHS, ShiftAmt);
8808 RHS = CGF.Builder.CreateShl(RHS, ShiftAmt);
8809 RHS = CGF.Builder.CreateAShr(RHS, ShiftAmt);
8810 } else {
8811 // Clear the upper bits.
8812 Constant *Mask = ConstantInt::get(Ty, 0xffffffff);
8813 LHS = CGF.Builder.CreateAnd(LHS, Mask);
8814 RHS = CGF.Builder.CreateAnd(RHS, Mask);
8815 }
8816
8817 return CGF.Builder.CreateMul(LHS, RHS);
8818}
8819
8820// Emit a masked pternlog intrinsic. This only exists because the header has to
8821// use a macro and we aren't able to pass the input argument to a pternlog
8822// builtin and a select builtin without evaluating it twice.
8823static Value *EmitX86Ternlog(CodeGenFunction &CGF, bool ZeroMask,
8824 ArrayRef<Value *> Ops) {
8825 llvm::Type *Ty = Ops[0]->getType();
8826
8827 unsigned VecWidth = Ty->getPrimitiveSizeInBits();
8828 unsigned EltWidth = Ty->getScalarSizeInBits();
8829 Intrinsic::ID IID;
8830 if (VecWidth == 128 && EltWidth == 32)
8831 IID = Intrinsic::x86_avx512_pternlog_d_128;
8832 else if (VecWidth == 256 && EltWidth == 32)
8833 IID = Intrinsic::x86_avx512_pternlog_d_256;
8834 else if (VecWidth == 512 && EltWidth == 32)
8835 IID = Intrinsic::x86_avx512_pternlog_d_512;
8836 else if (VecWidth == 128 && EltWidth == 64)
8837 IID = Intrinsic::x86_avx512_pternlog_q_128;
8838 else if (VecWidth == 256 && EltWidth == 64)
8839 IID = Intrinsic::x86_avx512_pternlog_q_256;
8840 else if (VecWidth == 512 && EltWidth == 64)
8841 IID = Intrinsic::x86_avx512_pternlog_q_512;
8842 else
8843 llvm_unreachable("Unexpected intrinsic")::llvm::llvm_unreachable_internal("Unexpected intrinsic", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8843)
;
8844
8845 Value *Ternlog = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID),
8846 Ops.drop_back());
8847 Value *PassThru = ZeroMask ? ConstantAggregateZero::get(Ty) : Ops[0];
8848 return EmitX86Select(CGF, Ops[4], Ternlog, PassThru);
8849}
8850
8851static Value *EmitX86SExtMask(CodeGenFunction &CGF, Value *Op,
8852 llvm::Type *DstTy) {
8853 unsigned NumberOfElements = DstTy->getVectorNumElements();
8854 Value *Mask = getMaskVecValue(CGF, Op, NumberOfElements);
8855 return CGF.Builder.CreateSExt(Mask, DstTy, "vpmovm2");
8856}
8857
8858Value *CodeGenFunction::EmitX86CpuIs(const CallExpr *E) {
8859 const Expr *CPUExpr = E->getArg(0)->IgnoreParenCasts();
8860 StringRef CPUStr = cast<clang::StringLiteral>(CPUExpr)->getString();
8861 return EmitX86CpuIs(CPUStr);
8862}
8863
8864Value *CodeGenFunction::EmitX86CpuIs(StringRef CPUStr) {
8865
8866 llvm::Type *Int32Ty = Builder.getInt32Ty();
8867
8868 // Matching the struct layout from the compiler-rt/libgcc structure that is
8869 // filled in:
8870 // unsigned int __cpu_vendor;
8871 // unsigned int __cpu_type;
8872 // unsigned int __cpu_subtype;
8873 // unsigned int __cpu_features[1];
8874 llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty,
8875 llvm::ArrayType::get(Int32Ty, 1));
8876
8877 // Grab the global __cpu_model.
8878 llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model");
8879
8880 // Calculate the index needed to access the correct field based on the
8881 // range. Also adjust the expected value.
8882 unsigned Index;
8883 unsigned Value;
8884 std::tie(Index, Value) = StringSwitch<std::pair<unsigned, unsigned>>(CPUStr)
8885#define X86_VENDOR(ENUM, STRING) \
8886 .Case(STRING, {0u, static_cast<unsigned>(llvm::X86::ENUM)})
8887#define X86_CPU_TYPE_COMPAT_WITH_ALIAS(ARCHNAME, ENUM, STR, ALIAS) \
8888 .Cases(STR, ALIAS, {1u, static_cast<unsigned>(llvm::X86::ENUM)})
8889#define X86_CPU_TYPE_COMPAT(ARCHNAME, ENUM, STR) \
8890 .Case(STR, {1u, static_cast<unsigned>(llvm::X86::ENUM)})
8891#define X86_CPU_SUBTYPE_COMPAT(ARCHNAME, ENUM, STR) \
8892 .Case(STR, {2u, static_cast<unsigned>(llvm::X86::ENUM)})
8893#include "llvm/Support/X86TargetParser.def"
8894 .Default({0, 0});
8895 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8895, __extension__ __PRETTY_FUNCTION__))
;
8896
8897 // Grab the appropriate field from __cpu_model.
8898 llvm::Value *Idxs[] = {ConstantInt::get(Int32Ty, 0),
8899 ConstantInt::get(Int32Ty, Index)};
8900 llvm::Value *CpuValue = Builder.CreateGEP(STy, CpuModel, Idxs);
8901 CpuValue = Builder.CreateAlignedLoad(CpuValue, CharUnits::fromQuantity(4));
8902
8903 // Check the value of the field against the requested value.
8904 return Builder.CreateICmpEQ(CpuValue,
8905 llvm::ConstantInt::get(Int32Ty, Value));
8906}
8907
8908Value *CodeGenFunction::EmitX86CpuSupports(const CallExpr *E) {
8909 const Expr *FeatureExpr = E->getArg(0)->IgnoreParenCasts();
8910 StringRef FeatureStr = cast<StringLiteral>(FeatureExpr)->getString();
8911 return EmitX86CpuSupports(FeatureStr);
8912}
8913
8914uint32_t
8915CodeGenFunction::GetX86CpuSupportsMask(ArrayRef<StringRef> FeatureStrs) {
8916 // Processor features and mapping to processor feature value.
8917 uint32_t FeaturesMask = 0;
8918 for (const StringRef &FeatureStr : FeatureStrs) {
8919 unsigned Feature =
8920 StringSwitch<unsigned>(FeatureStr)
8921#define X86_FEATURE_COMPAT(VAL, ENUM, STR) .Case(STR, VAL)
8922#include "llvm/Support/X86TargetParser.def"
8923 ;
8924 FeaturesMask |= (1U << Feature);
8925 }
8926 return FeaturesMask;
8927}
8928
8929Value *CodeGenFunction::EmitX86CpuSupports(ArrayRef<StringRef> FeatureStrs) {
8930 return EmitX86CpuSupports(GetX86CpuSupportsMask(FeatureStrs));
8931}
8932
8933llvm::Value *CodeGenFunction::EmitX86CpuSupports(uint32_t FeaturesMask) {
8934 // Matching the struct layout from the compiler-rt/libgcc structure that is
8935 // filled in:
8936 // unsigned int __cpu_vendor;
8937 // unsigned int __cpu_type;
8938 // unsigned int __cpu_subtype;
8939 // unsigned int __cpu_features[1];
8940 llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty,
8941 llvm::ArrayType::get(Int32Ty, 1));
8942
8943 // Grab the global __cpu_model.
8944 llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model");
8945
8946 // Grab the first (0th) element from the field __cpu_features off of the
8947 // global in the struct STy.
8948 Value *Idxs[] = {ConstantInt::get(Int32Ty, 0), ConstantInt::get(Int32Ty, 3),
8949 ConstantInt::get(Int32Ty, 0)};
8950 Value *CpuFeatures = Builder.CreateGEP(STy, CpuModel, Idxs);
8951 Value *Features =
8952 Builder.CreateAlignedLoad(CpuFeatures, CharUnits::fromQuantity(4));
8953
8954 // Check the value of the bit corresponding to the feature requested.
8955 Value *Bitset = Builder.CreateAnd(
8956 Features, llvm::ConstantInt::get(Int32Ty, FeaturesMask));
8957 return Builder.CreateICmpNE(Bitset, llvm::ConstantInt::get(Int32Ty, 0));
8958}
8959
8960Value *CodeGenFunction::EmitX86CpuInit() {
8961 llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy,
8962 /*Variadic*/ false);
8963 llvm::Constant *Func = CGM.CreateRuntimeFunction(FTy, "__cpu_indicator_init");
8964 return Builder.CreateCall(Func);
8965}
8966
8967Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
8968 const CallExpr *E) {
8969 if (BuiltinID == X86::BI__builtin_cpu_is)
8970 return EmitX86CpuIs(E);
8971 if (BuiltinID == X86::BI__builtin_cpu_supports)
8972 return EmitX86CpuSupports(E);
8973 if (BuiltinID == X86::BI__builtin_cpu_init)
8974 return EmitX86CpuInit();
8975
8976 SmallVector<Value*, 4> Ops;
8977
8978 // Find out if any arguments are required to be integer constant expressions.
8979 unsigned ICEArguments = 0;
8980 ASTContext::GetBuiltinTypeError Error;
8981 getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
8982 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8982, __extension__ __PRETTY_FUNCTION__))
;
8983
8984 for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) {
8985 // If this is a normal argument, just emit it as a scalar.
8986 if ((ICEArguments & (1 << i)) == 0) {
8987 Ops.push_back(EmitScalarExpr(E->getArg(i)));
8988 continue;
8989 }
8990
8991 // If this is required to be a constant, constant fold it so that we know
8992 // that the generated intrinsic gets a ConstantInt.
8993 llvm::APSInt Result;
8994 bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext());
8995 assert(IsConst && "Constant arg isn't actually constant?")(static_cast <bool> (IsConst && "Constant arg isn't actually constant?"
) ? void (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\""
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 8995, __extension__ __PRETTY_FUNCTION__))
; (void)IsConst;
8996 Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result));
8997 }
8998
8999 // These exist so that the builtin that takes an immediate can be bounds
9000 // checked by clang to avoid passing bad immediates to the backend. Since
9001 // AVX has a larger immediate than SSE we would need separate builtins to
9002 // do the different bounds checking. Rather than create a clang specific
9003 // SSE only builtin, this implements eight separate builtins to match gcc
9004 // implementation.
9005 auto getCmpIntrinsicCall = [this, &Ops](Intrinsic::ID ID, unsigned Imm) {
9006 Ops.push_back(llvm::ConstantInt::get(Int8Ty, Imm));
9007 llvm::Function *F = CGM.getIntrinsic(ID);
9008 return Builder.CreateCall(F, Ops);
9009 };
9010
9011 // For the vector forms of FP comparisons, translate the builtins directly to
9012 // IR.
9013 // TODO: The builtins could be removed if the SSE header files used vector
9014 // extension comparisons directly (vector ordered/unordered may need
9015 // additional support via __builtin_isnan()).
9016 auto getVectorFCmpIR = [this, &Ops](CmpInst::Predicate Pred) {
9017 Value *Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]);
9018 llvm::VectorType *FPVecTy = cast<llvm::VectorType>(Ops[0]->getType());
9019 llvm::VectorType *IntVecTy = llvm::VectorType::getInteger(FPVecTy);
9020 Value *Sext = Builder.CreateSExt(Cmp, IntVecTy);
9021 return Builder.CreateBitCast(Sext, FPVecTy);
9022 };
9023
9024 switch (BuiltinID) {
9025 default: return nullptr;
9026 case X86::BI_mm_prefetch: {
9027 Value *Address = Ops[0];
9028 ConstantInt *C = cast<ConstantInt>(Ops[1]);
9029 Value *RW = ConstantInt::get(Int32Ty, (C->getZExtValue() >> 2) & 0x1);
9030 Value *Locality = ConstantInt::get(Int32Ty, C->getZExtValue() & 0x3);
9031 Value *Data = ConstantInt::get(Int32Ty, 1);
9032 Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
9033 return Builder.CreateCall(F, {Address, RW, Locality, Data});
9034 }
9035 case X86::BI_mm_clflush: {
9036 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_clflush),
9037 Ops[0]);
9038 }
9039 case X86::BI_mm_lfence: {
9040 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_lfence));
9041 }
9042 case X86::BI_mm_mfence: {
9043 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_mfence));
9044 }
9045 case X86::BI_mm_sfence: {
9046 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_sfence));
9047 }
9048 case X86::BI_mm_pause: {
9049 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_pause));
9050 }
9051 case X86::BI__rdtsc: {
9052 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtsc));
9053 }
9054 case X86::BI__builtin_ia32_undef128:
9055 case X86::BI__builtin_ia32_undef256:
9056 case X86::BI__builtin_ia32_undef512:
9057 // The x86 definition of "undef" is not the same as the LLVM definition
9058 // (PR32176). We leave optimizing away an unnecessary zero constant to the
9059 // IR optimizer and backend.
9060 // TODO: If we had a "freeze" IR instruction to generate a fixed undef
9061 // value, we should use that here instead of a zero.
9062 return llvm::Constant::getNullValue(ConvertType(E->getType()));
9063 case X86::BI__builtin_ia32_vec_init_v8qi:
9064 case X86::BI__builtin_ia32_vec_init_v4hi:
9065 case X86::BI__builtin_ia32_vec_init_v2si:
9066 return Builder.CreateBitCast(BuildVector(Ops),
9067 llvm::Type::getX86_MMXTy(getLLVMContext()));
9068 case X86::BI__builtin_ia32_vec_ext_v2si:
9069 case X86::BI__builtin_ia32_vec_ext_v16qi:
9070 case X86::BI__builtin_ia32_vec_ext_v8hi:
9071 case X86::BI__builtin_ia32_vec_ext_v4si:
9072 case X86::BI__builtin_ia32_vec_ext_v4sf:
9073 case X86::BI__builtin_ia32_vec_ext_v2di:
9074 case X86::BI__builtin_ia32_vec_ext_v32qi:
9075 case X86::BI__builtin_ia32_vec_ext_v16hi:
9076 case X86::BI__builtin_ia32_vec_ext_v8si:
9077 case X86::BI__builtin_ia32_vec_ext_v4di: {
9078 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9079 uint64_t Index = cast<ConstantInt>(Ops[1])->getZExtValue();
9080 Index &= NumElts - 1;
9081 // These builtins exist so we can ensure the index is an ICE and in range.
9082 // Otherwise we could just do this in the header file.
9083 return Builder.CreateExtractElement(Ops[0], Index);
9084 }
9085 case X86::BI__builtin_ia32_vec_set_v16qi:
9086 case X86::BI__builtin_ia32_vec_set_v8hi:
9087 case X86::BI__builtin_ia32_vec_set_v4si:
9088 case X86::BI__builtin_ia32_vec_set_v2di:
9089 case X86::BI__builtin_ia32_vec_set_v32qi:
9090 case X86::BI__builtin_ia32_vec_set_v16hi:
9091 case X86::BI__builtin_ia32_vec_set_v8si:
9092 case X86::BI__builtin_ia32_vec_set_v4di: {
9093 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9094 unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue();
9095 Index &= NumElts - 1;
9096 // These builtins exist so we can ensure the index is an ICE and in range.
9097 // Otherwise we could just do this in the header file.
9098 return Builder.CreateInsertElement(Ops[0], Ops[1], Index);
9099 }
9100 case X86::BI_mm_setcsr:
9101 case X86::BI__builtin_ia32_ldmxcsr: {
9102 Address Tmp = CreateMemTemp(E->getArg(0)->getType());
9103 Builder.CreateStore(Ops[0], Tmp);
9104 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_ldmxcsr),
9105 Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy));
9106 }
9107 case X86::BI_mm_getcsr:
9108 case X86::BI__builtin_ia32_stmxcsr: {
9109 Address Tmp = CreateMemTemp(E->getType());
9110 Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_stmxcsr),
9111 Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy));
9112 return Builder.CreateLoad(Tmp, "stmxcsr");
9113 }
9114 case X86::BI__builtin_ia32_xsave:
9115 case X86::BI__builtin_ia32_xsave64:
9116 case X86::BI__builtin_ia32_xrstor:
9117 case X86::BI__builtin_ia32_xrstor64:
9118 case X86::BI__builtin_ia32_xsaveopt:
9119 case X86::BI__builtin_ia32_xsaveopt64:
9120 case X86::BI__builtin_ia32_xrstors:
9121 case X86::BI__builtin_ia32_xrstors64:
9122 case X86::BI__builtin_ia32_xsavec:
9123 case X86::BI__builtin_ia32_xsavec64:
9124 case X86::BI__builtin_ia32_xsaves:
9125 case X86::BI__builtin_ia32_xsaves64: {
9126 Intrinsic::ID ID;
9127#define INTRINSIC_X86_XSAVE_ID(NAME) \
9128 case X86::BI__builtin_ia32_##NAME: \
9129 ID = Intrinsic::x86_##NAME; \
9130 break
9131 switch (BuiltinID) {
9132 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 9132)
;
9133 INTRINSIC_X86_XSAVE_ID(xsave);
9134 INTRINSIC_X86_XSAVE_ID(xsave64);
9135 INTRINSIC_X86_XSAVE_ID(xrstor);
9136 INTRINSIC_X86_XSAVE_ID(xrstor64);
9137 INTRINSIC_X86_XSAVE_ID(xsaveopt);
9138 INTRINSIC_X86_XSAVE_ID(xsaveopt64);
9139 INTRINSIC_X86_XSAVE_ID(xrstors);
9140 INTRINSIC_X86_XSAVE_ID(xrstors64);
9141 INTRINSIC_X86_XSAVE_ID(xsavec);
9142 INTRINSIC_X86_XSAVE_ID(xsavec64);
9143 INTRINSIC_X86_XSAVE_ID(xsaves);
9144 INTRINSIC_X86_XSAVE_ID(xsaves64);
9145 }
9146#undef INTRINSIC_X86_XSAVE_ID
9147 Value *Mhi = Builder.CreateTrunc(
9148 Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, 32)), Int32Ty);
9149 Value *Mlo = Builder.CreateTrunc(Ops[1], Int32Ty);
9150 Ops[1] = Mhi;
9151 Ops.push_back(Mlo);
9152 return Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
9153 }
9154 case X86::BI__builtin_ia32_storedqudi128_mask:
9155 case X86::BI__builtin_ia32_storedqusi128_mask:
9156 case X86::BI__builtin_ia32_storedquhi128_mask:
9157 case X86::BI__builtin_ia32_storedquqi128_mask:
9158 case X86::BI__builtin_ia32_storeupd128_mask:
9159 case X86::BI__builtin_ia32_storeups128_mask:
9160 case X86::BI__builtin_ia32_storedqudi256_mask:
9161 case X86::BI__builtin_ia32_storedqusi256_mask:
9162 case X86::BI__builtin_ia32_storedquhi256_mask:
9163 case X86::BI__builtin_ia32_storedquqi256_mask:
9164 case X86::BI__builtin_ia32_storeupd256_mask:
9165 case X86::BI__builtin_ia32_storeups256_mask:
9166 case X86::BI__builtin_ia32_storedqudi512_mask:
9167 case X86::BI__builtin_ia32_storedqusi512_mask:
9168 case X86::BI__builtin_ia32_storedquhi512_mask:
9169 case X86::BI__builtin_ia32_storedquqi512_mask:
9170 case X86::BI__builtin_ia32_storeupd512_mask:
9171 case X86::BI__builtin_ia32_storeups512_mask:
9172 return EmitX86MaskedStore(*this, Ops, 1);
9173
9174 case X86::BI__builtin_ia32_storess128_mask:
9175 case X86::BI__builtin_ia32_storesd128_mask: {
9176 return EmitX86MaskedStore(*this, Ops, 1);
9177 }
9178 case X86::BI__builtin_ia32_vpopcntb_128:
9179 case X86::BI__builtin_ia32_vpopcntd_128:
9180 case X86::BI__builtin_ia32_vpopcntq_128:
9181 case X86::BI__builtin_ia32_vpopcntw_128:
9182 case X86::BI__builtin_ia32_vpopcntb_256:
9183 case X86::BI__builtin_ia32_vpopcntd_256:
9184 case X86::BI__builtin_ia32_vpopcntq_256:
9185 case X86::BI__builtin_ia32_vpopcntw_256:
9186 case X86::BI__builtin_ia32_vpopcntb_512:
9187 case X86::BI__builtin_ia32_vpopcntd_512:
9188 case X86::BI__builtin_ia32_vpopcntq_512:
9189 case X86::BI__builtin_ia32_vpopcntw_512: {
9190 llvm::Type *ResultType = ConvertType(E->getType());
9191 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
9192 return Builder.CreateCall(F, Ops);
9193 }
9194 case X86::BI__builtin_ia32_cvtmask2b128:
9195 case X86::BI__builtin_ia32_cvtmask2b256:
9196 case X86::BI__builtin_ia32_cvtmask2b512:
9197 case X86::BI__builtin_ia32_cvtmask2w128:
9198 case X86::BI__builtin_ia32_cvtmask2w256:
9199 case X86::BI__builtin_ia32_cvtmask2w512:
9200 case X86::BI__builtin_ia32_cvtmask2d128:
9201 case X86::BI__builtin_ia32_cvtmask2d256:
9202 case X86::BI__builtin_ia32_cvtmask2d512:
9203 case X86::BI__builtin_ia32_cvtmask2q128:
9204 case X86::BI__builtin_ia32_cvtmask2q256:
9205 case X86::BI__builtin_ia32_cvtmask2q512:
9206 return EmitX86SExtMask(*this, Ops[0], ConvertType(E->getType()));
9207
9208 case X86::BI__builtin_ia32_cvtb2mask128:
9209 case X86::BI__builtin_ia32_cvtb2mask256:
9210 case X86::BI__builtin_ia32_cvtb2mask512:
9211 case X86::BI__builtin_ia32_cvtw2mask128:
9212 case X86::BI__builtin_ia32_cvtw2mask256:
9213 case X86::BI__builtin_ia32_cvtw2mask512:
9214 case X86::BI__builtin_ia32_cvtd2mask128:
9215 case X86::BI__builtin_ia32_cvtd2mask256:
9216 case X86::BI__builtin_ia32_cvtd2mask512:
9217 case X86::BI__builtin_ia32_cvtq2mask128:
9218 case X86::BI__builtin_ia32_cvtq2mask256:
9219 case X86::BI__builtin_ia32_cvtq2mask512:
9220 return EmitX86ConvertToMask(*this, Ops[0]);
9221
9222 case X86::BI__builtin_ia32_vfmaddss3:
9223 case X86::BI__builtin_ia32_vfmaddsd3:
9224 case X86::BI__builtin_ia32_vfmaddss3_mask:
9225 case X86::BI__builtin_ia32_vfmaddsd3_mask:
9226 return EmitScalarFMAExpr(*this, Ops, Ops[0]);
9227 case X86::BI__builtin_ia32_vfmaddss:
9228 case X86::BI__builtin_ia32_vfmaddsd:
9229 return EmitScalarFMAExpr(*this, Ops,
9230 Constant::getNullValue(Ops[0]->getType()));
9231 case X86::BI__builtin_ia32_vfmaddss3_maskz:
9232 case X86::BI__builtin_ia32_vfmaddsd3_maskz:
9233 return EmitScalarFMAExpr(*this, Ops, Ops[0], /*ZeroMask*/true);
9234 case X86::BI__builtin_ia32_vfmaddss3_mask3:
9235 case X86::BI__builtin_ia32_vfmaddsd3_mask3:
9236 return EmitScalarFMAExpr(*this, Ops, Ops[2], /*ZeroMask*/false, 2);
9237 case X86::BI__builtin_ia32_vfmsubss3_mask3:
9238 case X86::BI__builtin_ia32_vfmsubsd3_mask3:
9239 return EmitScalarFMAExpr(*this, Ops, Ops[2], /*ZeroMask*/false, 2,
9240 /*NegAcc*/true);
9241 case X86::BI__builtin_ia32_vfmaddps:
9242 case X86::BI__builtin_ia32_vfmaddpd:
9243 case X86::BI__builtin_ia32_vfmaddps256:
9244 case X86::BI__builtin_ia32_vfmaddpd256:
9245 case X86::BI__builtin_ia32_vfmaddps512_mask:
9246 case X86::BI__builtin_ia32_vfmaddps512_maskz:
9247 case X86::BI__builtin_ia32_vfmaddps512_mask3:
9248 case X86::BI__builtin_ia32_vfmsubps512_mask3:
9249 case X86::BI__builtin_ia32_vfmaddpd512_mask:
9250 case X86::BI__builtin_ia32_vfmaddpd512_maskz:
9251 case X86::BI__builtin_ia32_vfmaddpd512_mask3:
9252 case X86::BI__builtin_ia32_vfmsubpd512_mask3:
9253 return EmitX86FMAExpr(*this, Ops, BuiltinID, /*IsAddSub*/false);
9254 case X86::BI__builtin_ia32_vfmaddsubps:
9255 case X86::BI__builtin_ia32_vfmaddsubpd:
9256 case X86::BI__builtin_ia32_vfmaddsubps256:
9257 case X86::BI__builtin_ia32_vfmaddsubpd256:
9258 case X86::BI__builtin_ia32_vfmaddsubps512_mask:
9259 case X86::BI__builtin_ia32_vfmaddsubps512_maskz:
9260 case X86::BI__builtin_ia32_vfmaddsubps512_mask3:
9261 case X86::BI__builtin_ia32_vfmsubaddps512_mask3:
9262 case X86::BI__builtin_ia32_vfmaddsubpd512_mask:
9263 case X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
9264 case X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
9265 case X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
9266 return EmitX86FMAExpr(*this, Ops, BuiltinID, /*IsAddSub*/true);
9267
9268 case X86::BI__builtin_ia32_movdqa32store128_mask:
9269 case X86::BI__builtin_ia32_movdqa64store128_mask:
9270 case X86::BI__builtin_ia32_storeaps128_mask:
9271 case X86::BI__builtin_ia32_storeapd128_mask:
9272 case X86::BI__builtin_ia32_movdqa32store256_mask:
9273 case X86::BI__builtin_ia32_movdqa64store256_mask:
9274 case X86::BI__builtin_ia32_storeaps256_mask:
9275 case X86::BI__builtin_ia32_storeapd256_mask:
9276 case X86::BI__builtin_ia32_movdqa32store512_mask:
9277 case X86::BI__builtin_ia32_movdqa64store512_mask:
9278 case X86::BI__builtin_ia32_storeaps512_mask:
9279 case X86::BI__builtin_ia32_storeapd512_mask: {
9280 unsigned Align =
9281 getContext().getTypeAlignInChars(E->getArg(1)->getType()).getQuantity();
9282 return EmitX86MaskedStore(*this, Ops, Align);
9283 }
9284 case X86::BI__builtin_ia32_loadups128_mask:
9285 case X86::BI__builtin_ia32_loadups256_mask:
9286 case X86::BI__builtin_ia32_loadups512_mask:
9287 case X86::BI__builtin_ia32_loadupd128_mask:
9288 case X86::BI__builtin_ia32_loadupd256_mask:
9289 case X86::BI__builtin_ia32_loadupd512_mask:
9290 case X86::BI__builtin_ia32_loaddquqi128_mask:
9291 case X86::BI__builtin_ia32_loaddquqi256_mask:
9292 case X86::BI__builtin_ia32_loaddquqi512_mask:
9293 case X86::BI__builtin_ia32_loaddquhi128_mask:
9294 case X86::BI__builtin_ia32_loaddquhi256_mask:
9295 case X86::BI__builtin_ia32_loaddquhi512_mask:
9296 case X86::BI__builtin_ia32_loaddqusi128_mask:
9297 case X86::BI__builtin_ia32_loaddqusi256_mask:
9298 case X86::BI__builtin_ia32_loaddqusi512_mask:
9299 case X86::BI__builtin_ia32_loaddqudi128_mask:
9300 case X86::BI__builtin_ia32_loaddqudi256_mask:
9301 case X86::BI__builtin_ia32_loaddqudi512_mask:
9302 return EmitX86MaskedLoad(*this, Ops, 1);
9303
9304 case X86::BI__builtin_ia32_loadss128_mask:
9305 case X86::BI__builtin_ia32_loadsd128_mask:
9306 return EmitX86MaskedLoad(*this, Ops, 1);
9307
9308 case X86::BI__builtin_ia32_loadaps128_mask:
9309 case X86::BI__builtin_ia32_loadaps256_mask:
9310 case X86::BI__builtin_ia32_loadaps512_mask:
9311 case X86::BI__builtin_ia32_loadapd128_mask:
9312 case X86::BI__builtin_ia32_loadapd256_mask:
9313 case X86::BI__builtin_ia32_loadapd512_mask:
9314 case X86::BI__builtin_ia32_movdqa32load128_mask:
9315 case X86::BI__builtin_ia32_movdqa32load256_mask:
9316 case X86::BI__builtin_ia32_movdqa32load512_mask:
9317 case X86::BI__builtin_ia32_movdqa64load128_mask:
9318 case X86::BI__builtin_ia32_movdqa64load256_mask:
9319 case X86::BI__builtin_ia32_movdqa64load512_mask: {
9320 unsigned Align =
9321 getContext().getTypeAlignInChars(E->getArg(1)->getType()).getQuantity();
9322 return EmitX86MaskedLoad(*this, Ops, Align);
9323 }
9324
9325 case X86::BI__builtin_ia32_expandloaddf128_mask:
9326 case X86::BI__builtin_ia32_expandloaddf256_mask:
9327 case X86::BI__builtin_ia32_expandloaddf512_mask:
9328 case X86::BI__builtin_ia32_expandloadsf128_mask:
9329 case X86::BI__builtin_ia32_expandloadsf256_mask:
9330 case X86::BI__builtin_ia32_expandloadsf512_mask:
9331 case X86::BI__builtin_ia32_expandloaddi128_mask:
9332 case X86::BI__builtin_ia32_expandloaddi256_mask:
9333 case X86::BI__builtin_ia32_expandloaddi512_mask:
9334 case X86::BI__builtin_ia32_expandloadsi128_mask:
9335 case X86::BI__builtin_ia32_expandloadsi256_mask:
9336 case X86::BI__builtin_ia32_expandloadsi512_mask:
9337 case X86::BI__builtin_ia32_expandloadhi128_mask:
9338 case X86::BI__builtin_ia32_expandloadhi256_mask:
9339 case X86::BI__builtin_ia32_expandloadhi512_mask:
9340 case X86::BI__builtin_ia32_expandloadqi128_mask:
9341 case X86::BI__builtin_ia32_expandloadqi256_mask:
9342 case X86::BI__builtin_ia32_expandloadqi512_mask:
9343 return EmitX86ExpandLoad(*this, Ops);
9344
9345 case X86::BI__builtin_ia32_compressstoredf128_mask:
9346 case X86::BI__builtin_ia32_compressstoredf256_mask:
9347 case X86::BI__builtin_ia32_compressstoredf512_mask:
9348 case X86::BI__builtin_ia32_compressstoresf128_mask:
9349 case X86::BI__builtin_ia32_compressstoresf256_mask:
9350 case X86::BI__builtin_ia32_compressstoresf512_mask:
9351 case X86::BI__builtin_ia32_compressstoredi128_mask:
9352 case X86::BI__builtin_ia32_compressstoredi256_mask:
9353 case X86::BI__builtin_ia32_compressstoredi512_mask:
9354 case X86::BI__builtin_ia32_compressstoresi128_mask:
9355 case X86::BI__builtin_ia32_compressstoresi256_mask:
9356 case X86::BI__builtin_ia32_compressstoresi512_mask:
9357 case X86::BI__builtin_ia32_compressstorehi128_mask:
9358 case X86::BI__builtin_ia32_compressstorehi256_mask:
9359 case X86::BI__builtin_ia32_compressstorehi512_mask:
9360 case X86::BI__builtin_ia32_compressstoreqi128_mask:
9361 case X86::BI__builtin_ia32_compressstoreqi256_mask:
9362 case X86::BI__builtin_ia32_compressstoreqi512_mask:
9363 return EmitX86CompressStore(*this, Ops);
9364
9365 case X86::BI__builtin_ia32_storehps:
9366 case X86::BI__builtin_ia32_storelps: {
9367 llvm::Type *PtrTy = llvm::PointerType::getUnqual(Int64Ty);
9368 llvm::Type *VecTy = llvm::VectorType::get(Int64Ty, 2);
9369
9370 // cast val v2i64
9371 Ops[1] = Builder.CreateBitCast(Ops[1], VecTy, "cast");
9372
9373 // extract (0, 1)
9374 unsigned Index = BuiltinID == X86::BI__builtin_ia32_storelps ? 0 : 1;
9375 Ops[1] = Builder.CreateExtractElement(Ops[1], Index, "extract");
9376
9377 // cast pointer to i64 & store
9378 Ops[0] = Builder.CreateBitCast(Ops[0], PtrTy);
9379 return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
9380 }
9381 case X86::BI__builtin_ia32_vextractf128_pd256:
9382 case X86::BI__builtin_ia32_vextractf128_ps256:
9383 case X86::BI__builtin_ia32_vextractf128_si256:
9384 case X86::BI__builtin_ia32_extract128i256:
9385 case X86::BI__builtin_ia32_extractf64x4_mask:
9386 case X86::BI__builtin_ia32_extractf32x4_mask:
9387 case X86::BI__builtin_ia32_extracti64x4_mask:
9388 case X86::BI__builtin_ia32_extracti32x4_mask:
9389 case X86::BI__builtin_ia32_extractf32x8_mask:
9390 case X86::BI__builtin_ia32_extracti32x8_mask:
9391 case X86::BI__builtin_ia32_extractf32x4_256_mask:
9392 case X86::BI__builtin_ia32_extracti32x4_256_mask:
9393 case X86::BI__builtin_ia32_extractf64x2_256_mask:
9394 case X86::BI__builtin_ia32_extracti64x2_256_mask:
9395 case X86::BI__builtin_ia32_extractf64x2_512_mask:
9396 case X86::BI__builtin_ia32_extracti64x2_512_mask: {
9397 llvm::Type *DstTy = ConvertType(E->getType());
9398 unsigned NumElts = DstTy->getVectorNumElements();
9399 unsigned SrcNumElts = Ops[0]->getType()->getVectorNumElements();
9400 unsigned SubVectors = SrcNumElts / NumElts;
9401 unsigned Index = cast<ConstantInt>(Ops[1])->getZExtValue();
9402 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 9402, __extension__ __PRETTY_FUNCTION__))
;
9403 Index &= SubVectors - 1; // Remove any extra bits.
9404 Index *= NumElts;
9405
9406 uint32_t Indices[16];
9407 for (unsigned i = 0; i != NumElts; ++i)
9408 Indices[i] = i + Index;
9409
9410 Value *Res = Builder.CreateShuffleVector(Ops[0],
9411 UndefValue::get(Ops[0]->getType()),
9412 makeArrayRef(Indices, NumElts),
9413 "extract");
9414
9415 if (Ops.size() == 4)
9416 Res = EmitX86Select(*this, Ops[3], Res, Ops[2]);
9417
9418 return Res;
9419 }
9420 case X86::BI__builtin_ia32_vinsertf128_pd256:
9421 case X86::BI__builtin_ia32_vinsertf128_ps256:
9422 case X86::BI__builtin_ia32_vinsertf128_si256:
9423 case X86::BI__builtin_ia32_insert128i256:
9424 case X86::BI__builtin_ia32_insertf64x4:
9425 case X86::BI__builtin_ia32_insertf32x4:
9426 case X86::BI__builtin_ia32_inserti64x4:
9427 case X86::BI__builtin_ia32_inserti32x4:
9428 case X86::BI__builtin_ia32_insertf32x8:
9429 case X86::BI__builtin_ia32_inserti32x8:
9430 case X86::BI__builtin_ia32_insertf32x4_256:
9431 case X86::BI__builtin_ia32_inserti32x4_256:
9432 case X86::BI__builtin_ia32_insertf64x2_256:
9433 case X86::BI__builtin_ia32_inserti64x2_256:
9434 case X86::BI__builtin_ia32_insertf64x2_512:
9435 case X86::BI__builtin_ia32_inserti64x2_512: {
9436 unsigned DstNumElts = Ops[0]->getType()->getVectorNumElements();
9437 unsigned SrcNumElts = Ops[1]->getType()->getVectorNumElements();
9438 unsigned SubVectors = DstNumElts / SrcNumElts;
9439 unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue();
9440 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 9440, __extension__ __PRETTY_FUNCTION__))
;
9441 Index &= SubVectors - 1; // Remove any extra bits.
9442 Index *= SrcNumElts;
9443
9444 uint32_t Indices[16];
9445 for (unsigned i = 0; i != DstNumElts; ++i)
9446 Indices[i] = (i >= SrcNumElts) ? SrcNumElts + (i % SrcNumElts) : i;
9447
9448 Value *Op1 = Builder.CreateShuffleVector(Ops[1],
9449 UndefValue::get(Ops[1]->getType()),
9450 makeArrayRef(Indices, DstNumElts),
9451 "widen");
9452
9453 for (unsigned i = 0; i != DstNumElts; ++i) {
9454 if (i >= Index && i < (Index + SrcNumElts))
9455 Indices[i] = (i - Index) + DstNumElts;
9456 else
9457 Indices[i] = i;
9458 }
9459
9460 return Builder.CreateShuffleVector(Ops[0], Op1,
9461 makeArrayRef(Indices, DstNumElts),
9462 "insert");
9463 }
9464 case X86::BI__builtin_ia32_pmovqd512_mask:
9465 case X86::BI__builtin_ia32_pmovwb512_mask: {
9466 Value *Res = Builder.CreateTrunc(Ops[0], Ops[1]->getType());
9467 return EmitX86Select(*this, Ops[2], Res, Ops[1]);
9468 }
9469 case X86::BI__builtin_ia32_pmovdb512_mask:
9470 case X86::BI__builtin_ia32_pmovdw512_mask:
9471 case X86::BI__builtin_ia32_pmovqw512_mask: {
9472 if (const auto *C = dyn_cast<Constant>(Ops[2]))
9473 if (C->isAllOnesValue())
9474 return Builder.CreateTrunc(Ops[0], Ops[1]->getType());
9475
9476 Intrinsic::ID IID;
9477 switch (BuiltinID) {
9478 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 9478)
;
9479 case X86::BI__builtin_ia32_pmovdb512_mask:
9480 IID = Intrinsic::x86_avx512_mask_pmov_db_512;
9481 break;
9482 case X86::BI__builtin_ia32_pmovdw512_mask:
9483 IID = Intrinsic::x86_avx512_mask_pmov_dw_512;
9484 break;
9485 case X86::BI__builtin_ia32_pmovqw512_mask:
9486 IID = Intrinsic::x86_avx512_mask_pmov_qw_512;
9487 break;
9488 }
9489
9490 Function *Intr = CGM.getIntrinsic(IID);
9491 return Builder.CreateCall(Intr, Ops);
9492 }
9493 case X86::BI__builtin_ia32_pblendw128:
9494 case X86::BI__builtin_ia32_blendpd:
9495 case X86::BI__builtin_ia32_blendps:
9496 case X86::BI__builtin_ia32_blendpd256:
9497 case X86::BI__builtin_ia32_blendps256:
9498 case X86::BI__builtin_ia32_pblendw256:
9499 case X86::BI__builtin_ia32_pblendd128:
9500 case X86::BI__builtin_ia32_pblendd256: {
9501 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9502 unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
9503
9504 uint32_t Indices[16];
9505 // If there are more than 8 elements, the immediate is used twice so make
9506 // sure we handle that.
9507 for (unsigned i = 0; i != NumElts; ++i)
9508 Indices[i] = ((Imm >> (i % 8)) & 0x1) ? NumElts + i : i;
9509
9510 return Builder.CreateShuffleVector(Ops[0], Ops[1],
9511 makeArrayRef(Indices, NumElts),
9512 "blend");
9513 }
9514 case X86::BI__builtin_ia32_pshuflw:
9515 case X86::BI__builtin_ia32_pshuflw256:
9516 case X86::BI__builtin_ia32_pshuflw512: {
9517 uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
9518 llvm::Type *Ty = Ops[0]->getType();
9519 unsigned NumElts = Ty->getVectorNumElements();
9520
9521 // Splat the 8-bits of immediate 4 times to help the loop wrap around.
9522 Imm = (Imm & 0xff) * 0x01010101;
9523
9524 uint32_t Indices[32];
9525 for (unsigned l = 0; l != NumElts; l += 8) {
9526 for (unsigned i = 0; i != 4; ++i) {
9527 Indices[l + i] = l + (Imm & 3);
9528 Imm >>= 2;
9529 }
9530 for (unsigned i = 4; i != 8; ++i)
9531 Indices[l + i] = l + i;
9532 }
9533
9534 return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
9535 makeArrayRef(Indices, NumElts),
9536 "pshuflw");
9537 }
9538 case X86::BI__builtin_ia32_pshufhw:
9539 case X86::BI__builtin_ia32_pshufhw256:
9540 case X86::BI__builtin_ia32_pshufhw512: {
9541 uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
9542 llvm::Type *Ty = Ops[0]->getType();
9543 unsigned NumElts = Ty->getVectorNumElements();
9544
9545 // Splat the 8-bits of immediate 4 times to help the loop wrap around.
9546 Imm = (Imm & 0xff) * 0x01010101;
9547
9548 uint32_t Indices[32];
9549 for (unsigned l = 0; l != NumElts; l += 8) {
9550 for (unsigned i = 0; i != 4; ++i)
9551 Indices[l + i] = l + i;
9552 for (unsigned i = 4; i != 8; ++i) {
9553 Indices[l + i] = l + 4 + (Imm & 3);
9554 Imm >>= 2;
9555 }
9556 }
9557
9558 return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
9559 makeArrayRef(Indices, NumElts),
9560 "pshufhw");
9561 }
9562 case X86::BI__builtin_ia32_pshufd:
9563 case X86::BI__builtin_ia32_pshufd256:
9564 case X86::BI__builtin_ia32_pshufd512:
9565 case X86::BI__builtin_ia32_vpermilpd:
9566 case X86::BI__builtin_ia32_vpermilps:
9567 case X86::BI__builtin_ia32_vpermilpd256:
9568 case X86::BI__builtin_ia32_vpermilps256:
9569 case X86::BI__builtin_ia32_vpermilpd512:
9570 case X86::BI__builtin_ia32_vpermilps512: {
9571 uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
9572 llvm::Type *Ty = Ops[0]->getType();
9573 unsigned NumElts = Ty->getVectorNumElements();
9574 unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128;
9575 unsigned NumLaneElts = NumElts / NumLanes;
9576
9577 // Splat the 8-bits of immediate 4 times to help the loop wrap around.
9578 Imm = (Imm & 0xff) * 0x01010101;
9579
9580 uint32_t Indices[16];
9581 for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
9582 for (unsigned i = 0; i != NumLaneElts; ++i) {
9583 Indices[i + l] = (Imm % NumLaneElts) + l;
9584 Imm /= NumLaneElts;
9585 }
9586 }
9587
9588 return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
9589 makeArrayRef(Indices, NumElts),
9590 "permil");
9591 }
9592 case X86::BI__builtin_ia32_shufpd:
9593 case X86::BI__builtin_ia32_shufpd256:
9594 case X86::BI__builtin_ia32_shufpd512:
9595 case X86::BI__builtin_ia32_shufps:
9596 case X86::BI__builtin_ia32_shufps256:
9597 case X86::BI__builtin_ia32_shufps512: {
9598 uint32_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
9599 llvm::Type *Ty = Ops[0]->getType();
9600 unsigned NumElts = Ty->getVectorNumElements();
9601 unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128;
9602 unsigned NumLaneElts = NumElts / NumLanes;
9603
9604 // Splat the 8-bits of immediate 4 times to help the loop wrap around.
9605 Imm = (Imm & 0xff) * 0x01010101;
9606
9607 uint32_t Indices[16];
9608 for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
9609 for (unsigned i = 0; i != NumLaneElts; ++i) {
9610 unsigned Index = Imm % NumLaneElts;
9611 Imm /= NumLaneElts;
9612 if (i >= (NumLaneElts / 2))
9613 Index += NumElts;
9614 Indices[l + i] = l + Index;
9615 }
9616 }
9617
9618 return Builder.CreateShuffleVector(Ops[0], Ops[1],
9619 makeArrayRef(Indices, NumElts),
9620 "shufp");
9621 }
9622 case X86::BI__builtin_ia32_permdi256:
9623 case X86::BI__builtin_ia32_permdf256:
9624 case X86::BI__builtin_ia32_permdi512:
9625 case X86::BI__builtin_ia32_permdf512: {
9626 unsigned Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
9627 llvm::Type *Ty = Ops[0]->getType();
9628 unsigned NumElts = Ty->getVectorNumElements();
9629
9630 // These intrinsics operate on 256-bit lanes of four 64-bit elements.
9631 uint32_t Indices[8];
9632 for (unsigned l = 0; l != NumElts; l += 4)
9633 for (unsigned i = 0; i != 4; ++i)
9634 Indices[l + i] = l + ((Imm >> (2 * i)) & 0x3);
9635
9636 return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
9637 makeArrayRef(Indices, NumElts),
9638 "perm");
9639 }
9640 case X86::BI__builtin_ia32_palignr128:
9641 case X86::BI__builtin_ia32_palignr256:
9642 case X86::BI__builtin_ia32_palignr512: {
9643 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff;
9644
9645 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9646 assert(NumElts % 16 == 0)(static_cast <bool> (NumElts % 16 == 0) ? void (0) : __assert_fail
("NumElts % 16 == 0", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 9646, __extension__ __PRETTY_FUNCTION__))
;
9647
9648 // If palignr is shifting the pair of vectors more than the size of two
9649 // lanes, emit zero.
9650 if (ShiftVal >= 32)
9651 return llvm::Constant::getNullValue(ConvertType(E->getType()));
9652
9653 // If palignr is shifting the pair of input vectors more than one lane,
9654 // but less than two lanes, convert to shifting in zeroes.
9655 if (ShiftVal > 16) {
9656 ShiftVal -= 16;
9657 Ops[1] = Ops[0];
9658 Ops[0] = llvm::Constant::getNullValue(Ops[0]->getType());
9659 }
9660
9661 uint32_t Indices[64];
9662 // 256-bit palignr operates on 128-bit lanes so we need to handle that
9663 for (unsigned l = 0; l != NumElts; l += 16) {
9664 for (unsigned i = 0; i != 16; ++i) {
9665 unsigned Idx = ShiftVal + i;
9666 if (Idx >= 16)
9667 Idx += NumElts - 16; // End of lane, switch operand.
9668 Indices[l + i] = Idx + l;
9669 }
9670 }
9671
9672 return Builder.CreateShuffleVector(Ops[1], Ops[0],
9673 makeArrayRef(Indices, NumElts),
9674 "palignr");
9675 }
9676 case X86::BI__builtin_ia32_alignd128:
9677 case X86::BI__builtin_ia32_alignd256:
9678 case X86::BI__builtin_ia32_alignd512:
9679 case X86::BI__builtin_ia32_alignq128:
9680 case X86::BI__builtin_ia32_alignq256:
9681 case X86::BI__builtin_ia32_alignq512: {
9682 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9683 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff;
9684
9685 // Mask the shift amount to width of two vectors.
9686 ShiftVal &= (2 * NumElts) - 1;
9687
9688 uint32_t Indices[16];
9689 for (unsigned i = 0; i != NumElts; ++i)
9690 Indices[i] = i + ShiftVal;
9691
9692 return Builder.CreateShuffleVector(Ops[1], Ops[0],
9693 makeArrayRef(Indices, NumElts),
9694 "valign");
9695 }
9696 case X86::BI__builtin_ia32_shuf_f32x4_256:
9697 case X86::BI__builtin_ia32_shuf_f64x2_256:
9698 case X86::BI__builtin_ia32_shuf_i32x4_256:
9699 case X86::BI__builtin_ia32_shuf_i64x2_256:
9700 case X86::BI__builtin_ia32_shuf_f32x4:
9701 case X86::BI__builtin_ia32_shuf_f64x2:
9702 case X86::BI__builtin_ia32_shuf_i32x4:
9703 case X86::BI__builtin_ia32_shuf_i64x2: {
9704 unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
9705 llvm::Type *Ty = Ops[0]->getType();
9706 unsigned NumElts = Ty->getVectorNumElements();
9707 unsigned NumLanes = Ty->getPrimitiveSizeInBits() == 512 ? 4 : 2;
9708 unsigned NumLaneElts = NumElts / NumLanes;
9709
9710 uint32_t Indices[16];
9711 for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
9712 unsigned Index = (Imm % NumLanes) * NumLaneElts;
9713 Imm /= NumLanes; // Discard the bits we just used.
9714 if (l >= (NumElts / 2))
9715 Index += NumElts; // Switch to other source.
9716 for (unsigned i = 0; i != NumLaneElts; ++i) {
9717 Indices[l + i] = Index + i;
9718 }
9719 }
9720
9721 return Builder.CreateShuffleVector(Ops[0], Ops[1],
9722 makeArrayRef(Indices, NumElts),
9723 "shuf");
9724 }
9725
9726 case X86::BI__builtin_ia32_vperm2f128_pd256:
9727 case X86::BI__builtin_ia32_vperm2f128_ps256:
9728 case X86::BI__builtin_ia32_vperm2f128_si256:
9729 case X86::BI__builtin_ia32_permti256: {
9730 unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
9731 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9732
9733 // This takes a very simple approach since there are two lanes and a
9734 // shuffle can have 2 inputs. So we reserve the first input for the first
9735 // lane and the second input for the second lane. This may result in
9736 // duplicate sources, but this can be dealt with in the backend.
9737
9738 Value *OutOps[2];
9739 uint32_t Indices[8];
9740 for (unsigned l = 0; l != 2; ++l) {
9741 // Determine the source for this lane.
9742 if (Imm & (1 << ((l * 4) + 3)))
9743 OutOps[l] = llvm::ConstantAggregateZero::get(Ops[0]->getType());
9744 else if (Imm & (1 << ((l * 4) + 1)))
9745 OutOps[l] = Ops[1];
9746 else
9747 OutOps[l] = Ops[0];
9748
9749 for (unsigned i = 0; i != NumElts/2; ++i) {
9750 // Start with ith element of the source for this lane.
9751 unsigned Idx = (l * NumElts) + i;
9752 // If bit 0 of the immediate half is set, switch to the high half of
9753 // the source.
9754 if (Imm & (1 << (l * 4)))
9755 Idx += NumElts/2;
9756 Indices[(l * (NumElts/2)) + i] = Idx;
9757 }
9758 }
9759
9760 return Builder.CreateShuffleVector(OutOps[0], OutOps[1],
9761 makeArrayRef(Indices, NumElts),
9762 "vperm");
9763 }
9764
9765 case X86::BI__builtin_ia32_pslldqi128_byteshift:
9766 case X86::BI__builtin_ia32_pslldqi256_byteshift:
9767 case X86::BI__builtin_ia32_pslldqi512_byteshift: {
9768 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
9769 llvm::Type *ResultType = Ops[0]->getType();
9770 // Builtin type is vXi64 so multiply by 8 to get bytes.
9771 unsigned NumElts = ResultType->getVectorNumElements() * 8;
9772
9773 // If pslldq is shifting the vector more than 15 bytes, emit zero.
9774 if (ShiftVal >= 16)
9775 return llvm::Constant::getNullValue(ResultType);
9776
9777 uint32_t Indices[64];
9778 // 256/512-bit pslldq operates on 128-bit lanes so we need to handle that
9779 for (unsigned l = 0; l != NumElts; l += 16) {
9780 for (unsigned i = 0; i != 16; ++i) {
9781 unsigned Idx = NumElts + i - ShiftVal;
9782 if (Idx < NumElts) Idx -= NumElts - 16; // end of lane, switch operand.
9783 Indices[l + i] = Idx + l;
9784 }
9785 }
9786
9787 llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, NumElts);
9788 Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast");
9789 Value *Zero = llvm::Constant::getNullValue(VecTy);
9790 Value *SV = Builder.CreateShuffleVector(Zero, Cast,
9791 makeArrayRef(Indices, NumElts),
9792 "pslldq");
9793 return Builder.CreateBitCast(SV, Ops[0]->getType(), "cast");
9794 }
9795 case X86::BI__builtin_ia32_psrldqi128_byteshift:
9796 case X86::BI__builtin_ia32_psrldqi256_byteshift:
9797 case X86::BI__builtin_ia32_psrldqi512_byteshift: {
9798 unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
9799 llvm::Type *ResultType = Ops[0]->getType();
9800 // Builtin type is vXi64 so multiply by 8 to get bytes.
9801 unsigned NumElts = ResultType->getVectorNumElements() * 8;
9802
9803 // If psrldq is shifting the vector more than 15 bytes, emit zero.
9804 if (ShiftVal >= 16)
9805 return llvm::Constant::getNullValue(ResultType);
9806
9807 uint32_t Indices[64];
9808 // 256/512-bit psrldq operates on 128-bit lanes so we need to handle that
9809 for (unsigned l = 0; l != NumElts; l += 16) {
9810 for (unsigned i = 0; i != 16; ++i) {
9811 unsigned Idx = i + ShiftVal;
9812 if (Idx >= 16) Idx += NumElts - 16; // end of lane, switch operand.
9813 Indices[l + i] = Idx + l;
9814 }
9815 }
9816
9817 llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, NumElts);
9818 Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast");
9819 Value *Zero = llvm::Constant::getNullValue(VecTy);
9820 Value *SV = Builder.CreateShuffleVector(Cast, Zero,
9821 makeArrayRef(Indices, NumElts),
9822 "psrldq");
9823 return Builder.CreateBitCast(SV, ResultType, "cast");
9824 }
9825 case X86::BI__builtin_ia32_movnti:
9826 case X86::BI__builtin_ia32_movnti64:
9827 case X86::BI__builtin_ia32_movntsd:
9828 case X86::BI__builtin_ia32_movntss: {
9829 llvm::MDNode *Node = llvm::MDNode::get(
9830 getLLVMContext(), llvm::ConstantAsMetadata::get(Builder.getInt32(1)));
9831
9832 Value *Ptr = Ops[0];
9833 Value *Src = Ops[1];
9834
9835 // Extract the 0'th element of the source vector.
9836 if (BuiltinID == X86::BI__builtin_ia32_movntsd ||
9837 BuiltinID == X86::BI__builtin_ia32_movntss)
9838 Src = Builder.CreateExtractElement(Src, (uint64_t)0, "extract");
9839
9840 // Convert the type of the pointer to a pointer to the stored type.
9841 Value *BC = Builder.CreateBitCast(
9842 Ptr, llvm::PointerType::getUnqual(Src->getType()), "cast");
9843
9844 // Unaligned nontemporal store of the scalar value.
9845 StoreInst *SI = Builder.CreateDefaultAlignedStore(Src, BC);
9846 SI->setMetadata(CGM.getModule().getMDKindID("nontemporal"), Node);
9847 SI->setAlignment(1);
9848 return SI;
9849 }
9850
9851 case X86::BI__builtin_ia32_selectb_128:
9852 case X86::BI__builtin_ia32_selectb_256:
9853 case X86::BI__builtin_ia32_selectb_512:
9854 case X86::BI__builtin_ia32_selectw_128:
9855 case X86::BI__builtin_ia32_selectw_256:
9856 case X86::BI__builtin_ia32_selectw_512:
9857 case X86::BI__builtin_ia32_selectd_128:
9858 case X86::BI__builtin_ia32_selectd_256:
9859 case X86::BI__builtin_ia32_selectd_512:
9860 case X86::BI__builtin_ia32_selectq_128:
9861 case X86::BI__builtin_ia32_selectq_256:
9862 case X86::BI__builtin_ia32_selectq_512:
9863 case X86::BI__builtin_ia32_selectps_128:
9864 case X86::BI__builtin_ia32_selectps_256:
9865 case X86::BI__builtin_ia32_selectps_512:
9866 case X86::BI__builtin_ia32_selectpd_128:
9867 case X86::BI__builtin_ia32_selectpd_256:
9868 case X86::BI__builtin_ia32_selectpd_512:
9869 return EmitX86Select(*this, Ops[0], Ops[1], Ops[2]);
9870 case X86::BI__builtin_ia32_selectss_128:
9871 case X86::BI__builtin_ia32_selectsd_128: {
9872 Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0);
9873 Value *B = Builder.CreateExtractElement(Ops[2], (uint64_t)0);
9874 A = EmitX86ScalarSelect(*this, Ops[0], A, B);
9875 return Builder.CreateInsertElement(Ops[1], A, (uint64_t)0);
9876 }
9877 case X86::BI__builtin_ia32_cmpb128_mask:
9878 case X86::BI__builtin_ia32_cmpb256_mask:
9879 case X86::BI__builtin_ia32_cmpb512_mask:
9880 case X86::BI__builtin_ia32_cmpw128_mask:
9881 case X86::BI__builtin_ia32_cmpw256_mask:
9882 case X86::BI__builtin_ia32_cmpw512_mask:
9883 case X86::BI__builtin_ia32_cmpd128_mask:
9884 case X86::BI__builtin_ia32_cmpd256_mask:
9885 case X86::BI__builtin_ia32_cmpd512_mask:
9886 case X86::BI__builtin_ia32_cmpq128_mask:
9887 case X86::BI__builtin_ia32_cmpq256_mask:
9888 case X86::BI__builtin_ia32_cmpq512_mask: {
9889 unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
9890 return EmitX86MaskedCompare(*this, CC, true, Ops);
9891 }
9892 case X86::BI__builtin_ia32_ucmpb128_mask:
9893 case X86::BI__builtin_ia32_ucmpb256_mask:
9894 case X86::BI__builtin_ia32_ucmpb512_mask:
9895 case X86::BI__builtin_ia32_ucmpw128_mask:
9896 case X86::BI__builtin_ia32_ucmpw256_mask:
9897 case X86::BI__builtin_ia32_ucmpw512_mask:
9898 case X86::BI__builtin_ia32_ucmpd128_mask:
9899 case X86::BI__builtin_ia32_ucmpd256_mask:
9900 case X86::BI__builtin_ia32_ucmpd512_mask:
9901 case X86::BI__builtin_ia32_ucmpq128_mask:
9902 case X86::BI__builtin_ia32_ucmpq256_mask:
9903 case X86::BI__builtin_ia32_ucmpq512_mask: {
9904 unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
9905 return EmitX86MaskedCompare(*this, CC, false, Ops);
9906 }
9907
9908 case X86::BI__builtin_ia32_kortestchi:
9909 case X86::BI__builtin_ia32_kortestzhi: {
9910 Value *Or = EmitX86MaskLogic(*this, Instruction::Or, 16, Ops);
9911 Value *C;
9912 if (BuiltinID == X86::BI__builtin_ia32_kortestchi)
9913 C = llvm::Constant::getAllOnesValue(Builder.getInt16Ty());
9914 else
9915 C = llvm::Constant::getNullValue(Builder.getInt16Ty());
9916 Value *Cmp = Builder.CreateICmpEQ(Or, C);
9917 return Builder.CreateZExt(Cmp, ConvertType(E->getType()));
9918 }
9919
9920 case X86::BI__builtin_ia32_kandhi:
9921 return EmitX86MaskLogic(*this, Instruction::And, 16, Ops);
9922 case X86::BI__builtin_ia32_kandnhi:
9923 return EmitX86MaskLogic(*this, Instruction::And, 16, Ops, true);
9924 case X86::BI__builtin_ia32_korhi:
9925 return EmitX86MaskLogic(*this, Instruction::Or, 16, Ops);
9926 case X86::BI__builtin_ia32_kxnorhi:
9927 return EmitX86MaskLogic(*this, Instruction::Xor, 16, Ops, true);
9928 case X86::BI__builtin_ia32_kxorhi:
9929 return EmitX86MaskLogic(*this, Instruction::Xor, 16, Ops);
9930 case X86::BI__builtin_ia32_knothi: {
9931 Ops[0] = getMaskVecValue(*this, Ops[0], 16);
9932 return Builder.CreateBitCast(Builder.CreateNot(Ops[0]),
9933 Builder.getInt16Ty());
9934 }
9935
9936 case X86::BI__builtin_ia32_kunpckdi:
9937 case X86::BI__builtin_ia32_kunpcksi:
9938 case X86::BI__builtin_ia32_kunpckhi: {
9939 unsigned NumElts = Ops[0]->getType()->getScalarSizeInBits();
9940 Value *LHS = getMaskVecValue(*this, Ops[0], NumElts);
9941 Value *RHS = getMaskVecValue(*this, Ops[1], NumElts);
9942 uint32_t Indices[64];
9943 for (unsigned i = 0; i != NumElts; ++i)
9944 Indices[i] = i;
9945
9946 // First extract half of each vector. This gives better codegen than
9947 // doing it in a single shuffle.
9948 LHS = Builder.CreateShuffleVector(LHS, LHS,
9949 makeArrayRef(Indices, NumElts / 2));
9950 RHS = Builder.CreateShuffleVector(RHS, RHS,
9951 makeArrayRef(Indices, NumElts / 2));
9952 // Concat the vectors.
9953 // NOTE: Operands are swapped to match the intrinsic definition.
9954 Value *Res = Builder.CreateShuffleVector(RHS, LHS,
9955 makeArrayRef(Indices, NumElts));
9956 return Builder.CreateBitCast(Res, Ops[0]->getType());
9957 }
9958
9959 case X86::BI__builtin_ia32_vplzcntd_128:
9960 case X86::BI__builtin_ia32_vplzcntd_256:
9961 case X86::BI__builtin_ia32_vplzcntd_512:
9962 case X86::BI__builtin_ia32_vplzcntq_128:
9963 case X86::BI__builtin_ia32_vplzcntq_256:
9964 case X86::BI__builtin_ia32_vplzcntq_512: {
9965 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType());
9966 return Builder.CreateCall(F, {Ops[0],Builder.getInt1(false)});
9967 }
9968 case X86::BI__builtin_ia32_sqrtss:
9969 case X86::BI__builtin_ia32_sqrtsd: {
9970 Value *A = Builder.CreateExtractElement(Ops[0], (uint64_t)0);
9971 Function *F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType());
9972 A = Builder.CreateCall(F, {A});
9973 return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0);
9974 }
9975 case X86::BI__builtin_ia32_sqrtsd_round_mask:
9976 case X86::BI__builtin_ia32_sqrtss_round_mask: {
9977 unsigned CC = cast<llvm::ConstantInt>(Ops[4])->getZExtValue();
9978 // Support only if the rounding mode is 4 (AKA CUR_DIRECTION),
9979 // otherwise keep the intrinsic.
9980 if (CC != 4) {
9981 Intrinsic::ID IID = BuiltinID == X86::BI__builtin_ia32_sqrtsd_round_mask ?
9982 Intrinsic::x86_avx512_mask_sqrt_sd :
9983 Intrinsic::x86_avx512_mask_sqrt_ss;
9984 return Builder.CreateCall(CGM.getIntrinsic(IID), Ops);
9985 }
9986 Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0);
9987 Function *F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType());
9988 A = Builder.CreateCall(F, A);
9989 Value *Src = Builder.CreateExtractElement(Ops[2], (uint64_t)0);
9990 A = EmitX86ScalarSelect(*this, Ops[3], A, Src);
9991 return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0);
9992 }
9993 case X86::BI__builtin_ia32_sqrtpd256:
9994 case X86::BI__builtin_ia32_sqrtpd:
9995 case X86::BI__builtin_ia32_sqrtps256:
9996 case X86::BI__builtin_ia32_sqrtps:
9997 case X86::BI__builtin_ia32_sqrtps512:
9998 case X86::BI__builtin_ia32_sqrtpd512: {
9999 if (Ops.size() == 2) {
10000 unsigned CC = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
10001 // Support only if the rounding mode is 4 (AKA CUR_DIRECTION),
10002 // otherwise keep the intrinsic.
10003 if (CC != 4) {
10004 Intrinsic::ID IID = BuiltinID == X86::BI__builtin_ia32_sqrtps512 ?
10005 Intrinsic::x86_avx512_sqrt_ps_512 :
10006 Intrinsic::x86_avx512_sqrt_pd_512;
10007 return Builder.CreateCall(CGM.getIntrinsic(IID), Ops);
10008 }
10009 }
10010 Function *F = CGM.getIntrinsic(Intrinsic::sqrt, Ops[0]->getType());
10011 return Builder.CreateCall(F, Ops[0]);
10012 }
10013 case X86::BI__builtin_ia32_pabsb128:
10014 case X86::BI__builtin_ia32_pabsw128:
10015 case X86::BI__builtin_ia32_pabsd128:
10016 case X86::BI__builtin_ia32_pabsb256:
10017 case X86::BI__builtin_ia32_pabsw256:
10018 case X86::BI__builtin_ia32_pabsd256:
10019 case X86::BI__builtin_ia32_pabsq128:
10020 case X86::BI__builtin_ia32_pabsq256:
10021 case X86::BI__builtin_ia32_pabsb512:
10022 case X86::BI__builtin_ia32_pabsw512:
10023 case X86::BI__builtin_ia32_pabsd512:
10024 case X86::BI__builtin_ia32_pabsq512:
10025 return EmitX86Abs(*this, Ops);
10026
10027 case X86::BI__builtin_ia32_pmaxsb128:
10028 case X86::BI__builtin_ia32_pmaxsw128:
10029 case X86::BI__builtin_ia32_pmaxsd128:
10030 case X86::BI__builtin_ia32_pmaxsq128:
10031 case X86::BI__builtin_ia32_pmaxsb256:
10032 case X86::BI__builtin_ia32_pmaxsw256:
10033 case X86::BI__builtin_ia32_pmaxsd256:
10034 case X86::BI__builtin_ia32_pmaxsq256:
10035 case X86::BI__builtin_ia32_pmaxsb512:
10036 case X86::BI__builtin_ia32_pmaxsw512:
10037 case X86::BI__builtin_ia32_pmaxsd512:
10038 case X86::BI__builtin_ia32_pmaxsq512:
10039 return EmitX86MinMax(*this, ICmpInst::ICMP_SGT, Ops);
10040 case X86::BI__builtin_ia32_pmaxub128:
10041 case X86::BI__builtin_ia32_pmaxuw128:
10042 case X86::BI__builtin_ia32_pmaxud128:
10043 case X86::BI__builtin_ia32_pmaxuq128:
10044 case X86::BI__builtin_ia32_pmaxub256:
10045 case X86::BI__builtin_ia32_pmaxuw256:
10046 case X86::BI__builtin_ia32_pmaxud256:
10047 case X86::BI__builtin_ia32_pmaxuq256:
10048 case X86::BI__builtin_ia32_pmaxub512:
10049 case X86::BI__builtin_ia32_pmaxuw512:
10050 case X86::BI__builtin_ia32_pmaxud512:
10051 case X86::BI__builtin_ia32_pmaxuq512:
10052 return EmitX86MinMax(*this, ICmpInst::ICMP_UGT, Ops);
10053 case X86::BI__builtin_ia32_pminsb128:
10054 case X86::BI__builtin_ia32_pminsw128:
10055 case X86::BI__builtin_ia32_pminsd128:
10056 case X86::BI__builtin_ia32_pminsq128:
10057 case X86::BI__builtin_ia32_pminsb256:
10058 case X86::BI__builtin_ia32_pminsw256:
10059 case X86::BI__builtin_ia32_pminsd256:
10060 case X86::BI__builtin_ia32_pminsq256:
10061 case X86::BI__builtin_ia32_pminsb512:
10062 case X86::BI__builtin_ia32_pminsw512:
10063 case X86::BI__builtin_ia32_pminsd512:
10064 case X86::BI__builtin_ia32_pminsq512:
10065 return EmitX86MinMax(*this, ICmpInst::ICMP_SLT, Ops);
10066 case X86::BI__builtin_ia32_pminub128:
10067 case X86::BI__builtin_ia32_pminuw128:
10068 case X86::BI__builtin_ia32_pminud128:
10069 case X86::BI__builtin_ia32_pminuq128:
10070 case X86::BI__builtin_ia32_pminub256:
10071 case X86::BI__builtin_ia32_pminuw256:
10072 case X86::BI__builtin_ia32_pminud256:
10073 case X86::BI__builtin_ia32_pminuq256:
10074 case X86::BI__builtin_ia32_pminub512:
10075 case X86::BI__builtin_ia32_pminuw512:
10076 case X86::BI__builtin_ia32_pminud512:
10077 case X86::BI__builtin_ia32_pminuq512:
10078 return EmitX86MinMax(*this, ICmpInst::ICMP_ULT, Ops);
10079
10080 case X86::BI__builtin_ia32_pmuludq128:
10081 case X86::BI__builtin_ia32_pmuludq256:
10082 case X86::BI__builtin_ia32_pmuludq512:
10083 return EmitX86Muldq(*this, /*IsSigned*/false, Ops);
10084
10085 case X86::BI__builtin_ia32_pmuldq128:
10086 case X86::BI__builtin_ia32_pmuldq256:
10087 case X86::BI__builtin_ia32_pmuldq512:
10088 return EmitX86Muldq(*this, /*IsSigned*/true, Ops);
10089
10090 case X86::BI__builtin_ia32_pternlogd512_mask:
10091 case X86::BI__builtin_ia32_pternlogq512_mask:
10092 case X86::BI__builtin_ia32_pternlogd128_mask:
10093 case X86::BI__builtin_ia32_pternlogd256_mask:
10094 case X86::BI__builtin_ia32_pternlogq128_mask:
10095 case X86::BI__builtin_ia32_pternlogq256_mask:
10096 return EmitX86Ternlog(*this, /*ZeroMask*/false, Ops);
10097
10098 case X86::BI__builtin_ia32_pternlogd512_maskz:
10099 case X86::BI__builtin_ia32_pternlogq512_maskz:
10100 case X86::BI__builtin_ia32_pternlogd128_maskz:
10101 case X86::BI__builtin_ia32_pternlogd256_maskz:
10102 case X86::BI__builtin_ia32_pternlogq128_maskz:
10103 case X86::BI__builtin_ia32_pternlogq256_maskz:
10104 return EmitX86Ternlog(*this, /*ZeroMask*/true, Ops);
10105
10106 // 3DNow!
10107 case X86::BI__builtin_ia32_pswapdsf:
10108 case X86::BI__builtin_ia32_pswapdsi: {
10109 llvm::Type *MMXTy = llvm::Type::getX86_MMXTy(getLLVMContext());
10110 Ops[0] = Builder.CreateBitCast(Ops[0], MMXTy, "cast");
10111 llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_3dnowa_pswapd);
10112 return Builder.CreateCall(F, Ops, "pswapd");
10113 }
10114 case X86::BI__builtin_ia32_rdrand16_step:
10115 case X86::BI__builtin_ia32_rdrand32_step:
10116 case X86::BI__builtin_ia32_rdrand64_step:
10117 case X86::BI__builtin_ia32_rdseed16_step:
10118 case X86::BI__builtin_ia32_rdseed32_step:
10119 case X86::BI__builtin_ia32_rdseed64_step: {
10120 Intrinsic::ID ID;
10121 switch (BuiltinID) {
10122 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10122)
;
10123 case X86::BI__builtin_ia32_rdrand16_step:
10124 ID = Intrinsic::x86_rdrand_16;
10125 break;
10126 case X86::BI__builtin_ia32_rdrand32_step:
10127 ID = Intrinsic::x86_rdrand_32;
10128 break;
10129 case X86::BI__builtin_ia32_rdrand64_step:
10130 ID = Intrinsic::x86_rdrand_64;
10131 break;
10132 case X86::BI__builtin_ia32_rdseed16_step:
10133 ID = Intrinsic::x86_rdseed_16;
10134 break;
10135 case X86::BI__builtin_ia32_rdseed32_step:
10136 ID = Intrinsic::x86_rdseed_32;
10137 break;
10138 case X86::BI__builtin_ia32_rdseed64_step:
10139 ID = Intrinsic::x86_rdseed_64;
10140 break;
10141 }
10142
10143 Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID));
10144 Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 0),
10145 Ops[0]);
10146 return Builder.CreateExtractValue(Call, 1);
10147 }
10148
10149 case X86::BI__builtin_ia32_fpclassps128_mask:
10150 case X86::BI__builtin_ia32_fpclassps256_mask:
10151 case X86::BI__builtin_ia32_fpclassps512_mask:
10152 case X86::BI__builtin_ia32_fpclasspd128_mask:
10153 case X86::BI__builtin_ia32_fpclasspd256_mask:
10154 case X86::BI__builtin_ia32_fpclasspd512_mask: {
10155 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
10156 Value *MaskIn = Ops[2];
10157 Ops.erase(&Ops[2]);
10158
10159 Intrinsic::ID ID;
10160 switch (BuiltinID) {
10161 default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10161)
;
10162 case X86::BI__builtin_ia32_fpclassps128_mask:
10163 ID = Intrinsic::x86_avx512_fpclass_ps_128;
10164 break;
10165 case X86::BI__builtin_ia32_fpclassps256_mask:
10166 ID = Intrinsic::x86_avx512_fpclass_ps_256;
10167 break;
10168 case X86::BI__builtin_ia32_fpclassps512_mask:
10169 ID = Intrinsic::x86_avx512_fpclass_ps_512;
10170 break;
10171 case X86::BI__builtin_ia32_fpclasspd128_mask:
10172 ID = Intrinsic::x86_avx512_fpclass_pd_128;
10173 break;
10174 case X86::BI__builtin_ia32_fpclasspd256_mask:
10175 ID = Intrinsic::x86_avx512_fpclass_pd_256;
10176 break;
10177 case X86::BI__builtin_ia32_fpclasspd512_mask:
10178 ID = Intrinsic::x86_avx512_fpclass_pd_512;
10179 break;
10180 }
10181
10182 Value *Fpclass = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
10183 return EmitX86MaskedCompareResult(*this, Fpclass, NumElts, MaskIn);
10184 }
10185
10186 // packed comparison intrinsics
10187 case X86::BI__builtin_ia32_cmpeqps:
10188 case X86::BI__builtin_ia32_cmpeqpd:
10189 return getVectorFCmpIR(CmpInst::FCMP_OEQ);
10190 case X86::BI__builtin_ia32_cmpltps:
10191 case X86::BI__builtin_ia32_cmpltpd:
10192 return getVectorFCmpIR(CmpInst::FCMP_OLT);
10193 case X86::BI__builtin_ia32_cmpleps:
10194 case X86::BI__builtin_ia32_cmplepd:
10195 return getVectorFCmpIR(CmpInst::FCMP_OLE);
10196 case X86::BI__builtin_ia32_cmpunordps:
10197 case X86::BI__builtin_ia32_cmpunordpd:
10198 return getVectorFCmpIR(CmpInst::FCMP_UNO);
10199 case X86::BI__builtin_ia32_cmpneqps:
10200 case X86::BI__builtin_ia32_cmpneqpd:
10201 return getVectorFCmpIR(CmpInst::FCMP_UNE);
10202 case X86::BI__builtin_ia32_cmpnltps:
10203 case X86::BI__builtin_ia32_cmpnltpd:
10204 return getVectorFCmpIR(CmpInst::FCMP_UGE);
10205 case X86::BI__builtin_ia32_cmpnleps:
10206 case X86::BI__builtin_ia32_cmpnlepd:
10207 return getVectorFCmpIR(CmpInst::FCMP_UGT);
10208 case X86::BI__builtin_ia32_cmpordps:
10209 case X86::BI__builtin_ia32_cmpordpd:
10210 return getVectorFCmpIR(CmpInst::FCMP_ORD);
10211 case X86::BI__builtin_ia32_cmpps:
10212 case X86::BI__builtin_ia32_cmpps256:
10213 case X86::BI__builtin_ia32_cmppd:
10214 case X86::BI__builtin_ia32_cmppd256:
10215 case X86::BI__builtin_ia32_cmpps128_mask:
10216 case X86::BI__builtin_ia32_cmpps256_mask:
10217 case X86::BI__builtin_ia32_cmpps512_mask:
10218 case X86::BI__builtin_ia32_cmppd128_mask:
10219 case X86::BI__builtin_ia32_cmppd256_mask:
10220 case X86::BI__builtin_ia32_cmppd512_mask: {
10221 // Lowering vector comparisons to fcmp instructions, while
10222 // ignoring signalling behaviour requested
10223 // ignoring rounding mode requested
10224 // This is is only possible as long as FENV_ACCESS is not implemented.
10225 // See also: https://reviews.llvm.org/D45616
10226
10227 // The third argument is the comparison condition, and integer in the
10228 // range [0, 31]
10229 unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x1f;
10230
10231 // Lowering to IR fcmp instruction.
10232 // Ignoring requested signaling behaviour,
10233 // e.g. both _CMP_GT_OS & _CMP_GT_OQ are translated to FCMP_OGT.
10234 FCmpInst::Predicate Pred;
10235 switch (CC) {
10236 case 0x00: Pred = FCmpInst::FCMP_OEQ; break;
10237 case 0x01: Pred = FCmpInst::FCMP_OLT; break;
10238 case 0x02: Pred = FCmpInst::FCMP_OLE; break;
10239 case 0x03: Pred = FCmpInst::FCMP_UNO; break;
10240 case 0x04: Pred = FCmpInst::FCMP_UNE; break;
10241 case 0x05: Pred = FCmpInst::FCMP_UGE; break;
10242 case 0x06: Pred = FCmpInst::FCMP_UGT; break;
10243 case 0x07: Pred = FCmpInst::FCMP_ORD; break;
10244 case 0x08: Pred = FCmpInst::FCMP_UEQ; break;
10245 case 0x09: Pred = FCmpInst::FCMP_ULT; break;
10246 case 0x0a: Pred = FCmpInst::FCMP_ULE; break;
10247 case 0x0b: Pred = FCmpInst::FCMP_FALSE; break;
10248 case 0x0c: Pred = FCmpInst::FCMP_ONE; break;
10249 case 0x0d: Pred = FCmpInst::FCMP_OGE; break;
10250 case 0x0e: Pred = FCmpInst::FCMP_OGT; break;
10251 case 0x0f: Pred = FCmpInst::FCMP_TRUE; break;
10252 case 0x10: Pred = FCmpInst::FCMP_OEQ; break;
10253 case 0x11: Pred = FCmpInst::FCMP_OLT; break;
10254 case 0x12: Pred = FCmpInst::FCMP_OLE; break;
10255 case 0x13: Pred = FCmpInst::FCMP_UNO; break;
10256 case 0x14: Pred = FCmpInst::FCMP_UNE; break;
10257 case 0x15: Pred = FCmpInst::FCMP_UGE; break;
10258 case 0x16: Pred = FCmpInst::FCMP_UGT; break;
10259 case 0x17: Pred = FCmpInst::FCMP_ORD; break;
10260 case 0x18: Pred = FCmpInst::FCMP_UEQ; break;
10261 case 0x19: Pred = FCmpInst::FCMP_ULT; break;
10262 case 0x1a: Pred = FCmpInst::FCMP_ULE; break;
10263 case 0x1b: Pred = FCmpInst::FCMP_FALSE; break;
10264 case 0x1c: Pred = FCmpInst::FCMP_ONE; break;
10265 case 0x1d: Pred = FCmpInst::FCMP_OGE; break;
10266 case 0x1e: Pred = FCmpInst::FCMP_OGT; break;
10267 case 0x1f: Pred = FCmpInst::FCMP_TRUE; break;
10268 default: llvm_unreachable("Unhandled CC")::llvm::llvm_unreachable_internal("Unhandled CC", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10268)
;
10269 }
10270
10271 // Builtins without the _mask suffix return a vector of integers
10272 // of the same width as the input vectors
10273 switch (BuiltinID) {
10274 case X86::BI__builtin_ia32_cmpps512_mask:
10275 case X86::BI__builtin_ia32_cmppd512_mask:
10276 case X86::BI__builtin_ia32_cmpps128_mask:
10277 case X86::BI__builtin_ia32_cmpps256_mask:
10278 case X86::BI__builtin_ia32_cmppd128_mask:
10279 case X86::BI__builtin_ia32_cmppd256_mask: {
10280 unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
10281 Value *Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]);
10282 return EmitX86MaskedCompareResult(*this, Cmp, NumElts, Ops[3]);
10283 }
10284 default:
10285 return getVectorFCmpIR(Pred);
10286 }
10287 }
10288
10289 // SSE scalar comparison intrinsics
10290 case X86::BI__builtin_ia32_cmpeqss:
10291 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 0);
10292 case X86::BI__builtin_ia32_cmpltss:
10293 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 1);
10294 case X86::BI__builtin_ia32_cmpless:
10295 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 2);
10296 case X86::BI__builtin_ia32_cmpunordss:
10297 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 3);
10298 case X86::BI__builtin_ia32_cmpneqss:
10299 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 4);
10300 case X86::BI__builtin_ia32_cmpnltss:
10301 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 5);
10302 case X86::BI__builtin_ia32_cmpnless:
10303 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 6);
10304 case X86::BI__builtin_ia32_cmpordss:
10305 return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 7);
10306 case X86::BI__builtin_ia32_cmpeqsd:
10307 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 0);
10308 case X86::BI__builtin_ia32_cmpltsd:
10309 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 1);
10310 case X86::BI__builtin_ia32_cmplesd:
10311 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 2);
10312 case X86::BI__builtin_ia32_cmpunordsd:
10313 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 3);
10314 case X86::BI__builtin_ia32_cmpneqsd:
10315 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 4);
10316 case X86::BI__builtin_ia32_cmpnltsd:
10317 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 5);
10318 case X86::BI__builtin_ia32_cmpnlesd:
10319 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 6);
10320 case X86::BI__builtin_ia32_cmpordsd:
10321 return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 7);
10322
10323 case X86::BI__emul:
10324 case X86::BI__emulu: {
10325 llvm::Type *Int64Ty = llvm::IntegerType::get(getLLVMContext(), 64);
10326 bool isSigned = (BuiltinID == X86::BI__emul);
10327 Value *LHS = Builder.CreateIntCast(Ops[0], Int64Ty, isSigned);
10328 Value *RHS = Builder.CreateIntCast(Ops[1], Int64Ty, isSigned);
10329 return Builder.CreateMul(LHS, RHS, "", !isSigned, isSigned);
10330 }
10331 case X86::BI__mulh:
10332 case X86::BI__umulh:
10333 case X86::BI_mul128:
10334 case X86::BI_umul128: {
10335 llvm::Type *ResType = ConvertType(E->getType());
10336 llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128);
10337
10338 bool IsSigned = (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI_mul128);
10339 Value *LHS = Builder.CreateIntCast(Ops[0], Int128Ty, IsSigned);
10340 Value *RHS = Builder.CreateIntCast(Ops[1], Int128Ty, IsSigned);
10341
10342 Value *MulResult, *HigherBits;
10343 if (IsSigned) {
10344 MulResult = Builder.CreateNSWMul(LHS, RHS);
10345 HigherBits = Builder.CreateAShr(MulResult, 64);
10346 } else {
10347 MulResult = Builder.CreateNUWMul(LHS, RHS);
10348 HigherBits = Builder.CreateLShr(MulResult, 64);
10349 }
10350 HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned);
10351
10352 if (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI__umulh)
10353 return HigherBits;
10354
10355 Address HighBitsAddress = EmitPointerWithAlignment(E->getArg(2));
10356 Builder.CreateStore(HigherBits, HighBitsAddress);
10357 return Builder.CreateIntCast(MulResult, ResType, IsSigned);
10358 }
10359
10360 case X86::BI__faststorefence: {
10361 return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent,
10362 llvm::SyncScope::System);
10363 }
10364 case X86::BI_ReadWriteBarrier:
10365 case X86::BI_ReadBarrier:
10366 case X86::BI_WriteBarrier: {
10367 return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent,
10368 llvm::SyncScope::SingleThread);
10369 }
10370 case X86::BI_BitScanForward:
10371 case X86::BI_BitScanForward64:
10372 return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E);
10373 case X86::BI_BitScanReverse:
10374 case X86::BI_BitScanReverse64:
10375 return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E);
10376
10377 case X86::BI_InterlockedAnd64:
10378 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E);
10379 case X86::BI_InterlockedExchange64:
10380 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E);
10381 case X86::BI_InterlockedExchangeAdd64:
10382 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E);
10383 case X86::BI_InterlockedExchangeSub64:
10384 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E);
10385 case X86::BI_InterlockedOr64:
10386 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E);
10387 case X86::BI_InterlockedXor64:
10388 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E);
10389 case X86::BI_InterlockedDecrement64:
10390 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E);
10391 case X86::BI_InterlockedIncrement64:
10392 return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E);
10393 case X86::BI_InterlockedCompareExchange128: {
10394 // InterlockedCompareExchange128 doesn't directly refer to 128bit ints,
10395 // instead it takes pointers to 64bit ints for Destination and
10396 // ComparandResult, and exchange is taken as two 64bit ints (high & low).
10397 // The previous value is written to ComparandResult, and success is
10398 // returned.
10399
10400 llvm::Type *Int128Ty = Builder.getInt128Ty();
10401 llvm::Type *Int128PtrTy = Int128Ty->getPointerTo();
10402
10403 Value *Destination =
10404 Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int128PtrTy);
10405 Value *ExchangeHigh128 =
10406 Builder.CreateZExt(EmitScalarExpr(E->getArg(1)), Int128Ty);
10407 Value *ExchangeLow128 =
10408 Builder.CreateZExt(EmitScalarExpr(E->getArg(2)), Int128Ty);
10409 Address ComparandResult(
10410 Builder.CreateBitCast(EmitScalarExpr(E->getArg(3)), Int128PtrTy),
10411 getContext().toCharUnitsFromBits(128));
10412
10413 Value *Exchange = Builder.CreateOr(
10414 Builder.CreateShl(ExchangeHigh128, 64, "", false, false),
10415 ExchangeLow128);
10416
10417 Value *Comparand = Builder.CreateLoad(ComparandResult);
10418
10419 AtomicCmpXchgInst *CXI =
10420 Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange,
10421 AtomicOrdering::SequentiallyConsistent,
10422 AtomicOrdering::SequentiallyConsistent);
10423 CXI->setVolatile(true);
10424
10425 // Write the result back to the inout pointer.
10426 Builder.CreateStore(Builder.CreateExtractValue(CXI, 0), ComparandResult);
10427
10428 // Get the success boolean and zero extend it to i8.
10429 Value *Success = Builder.CreateExtractValue(CXI, 1);
10430 return Builder.CreateZExt(Success, ConvertType(E->getType()));
10431 }
10432
10433 case X86::BI_AddressOfReturnAddress: {
10434 Value *F = CGM.getIntrinsic(Intrinsic::addressofreturnaddress);
10435 return Builder.CreateCall(F);
10436 }
10437 case X86::BI__stosb: {
10438 // We treat __stosb as a volatile memset - it may not generate "rep stosb"
10439 // instruction, but it will create a memset that won't be optimized away.
10440 return Builder.CreateMemSet(Ops[0], Ops[1], Ops[2], 1, true);
10441 }
10442 case X86::BI__ud2:
10443 // llvm.trap makes a ud2a instruction on x86.
10444 return EmitTrapCall(Intrinsic::trap);
10445 case X86::BI__int2c: {
10446 // This syscall signals a driver assertion failure in x86 NT kernels.
10447 llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false);
10448 llvm::InlineAsm *IA =
10449 llvm::InlineAsm::get(FTy, "int $$0x2c", "", /*SideEffects=*/true);
10450 llvm::AttributeList NoReturnAttr = llvm::AttributeList::get(
10451 getLLVMContext(), llvm::AttributeList::FunctionIndex,
10452 llvm::Attribute::NoReturn);
10453 CallSite CS = Builder.CreateCall(IA);
10454 CS.setAttributes(NoReturnAttr);
10455 return CS.getInstruction();
10456 }
10457 case X86::BI__readfsbyte:
10458 case X86::BI__readfsword:
10459 case X86::BI__readfsdword:
10460 case X86::BI__readfsqword: {
10461 llvm::Type *IntTy = ConvertType(E->getType());
10462 Value *Ptr = Builder.CreateIntToPtr(EmitScalarExpr(E->getArg(0)),
10463 llvm::PointerType::get(IntTy, 257));
10464 LoadInst *Load = Builder.CreateAlignedLoad(
10465 IntTy, Ptr, getContext().getTypeAlignInChars(E->getType()));
10466 Load->setVolatile(true);
10467 return Load;
10468 }
10469 case X86::BI__readgsbyte:
10470 case X86::BI__readgsword:
10471 case X86::BI__readgsdword:
10472 case X86::BI__readgsqword: {
10473 llvm::Type *IntTy = ConvertType(E->getType());
10474 Value *Ptr = Builder.CreateIntToPtr(EmitScalarExpr(E->getArg(0)),
10475 llvm::PointerType::get(IntTy, 256));
10476 LoadInst *Load = Builder.CreateAlignedLoad(
10477 IntTy, Ptr, getContext().getTypeAlignInChars(E->getType()));
10478 Load->setVolatile(true);
10479 return Load;
10480 }
10481 }
10482}
10483
10484
10485Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID,
10486 const CallExpr *E) {
10487 SmallVector<Value*, 4> Ops;
10488
10489 for (unsigned i = 0, e = E->getNumArgs(); i != e; i++)
10490 Ops.push_back(EmitScalarExpr(E->getArg(i)));
10491
10492 Intrinsic::ID ID = Intrinsic::not_intrinsic;
10493
10494 switch (BuiltinID) {
10495 default: return nullptr;
10496
10497 // __builtin_ppc_get_timebase is GCC 4.8+'s PowerPC-specific name for what we
10498 // call __builtin_readcyclecounter.
10499 case PPC::BI__builtin_ppc_get_timebase:
10500 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::readcyclecounter));
10501
10502 // vec_ld, vec_xl_be, vec_lvsl, vec_lvsr
10503 case PPC::BI__builtin_altivec_lvx:
10504 case PPC::BI__builtin_altivec_lvxl:
10505 case PPC::BI__builtin_altivec_lvebx:
10506 case PPC::BI__builtin_altivec_lvehx:
10507 case PPC::BI__builtin_altivec_lvewx:
10508 case PPC::BI__builtin_altivec_lvsl:
10509 case PPC::BI__builtin_altivec_lvsr:
10510 case PPC::BI__builtin_vsx_lxvd2x:
10511 case PPC::BI__builtin_vsx_lxvw4x:
10512 case PPC::BI__builtin_vsx_lxvd2x_be:
10513 case PPC::BI__builtin_vsx_lxvw4x_be:
10514 case PPC::BI__builtin_vsx_lxvl:
10515 case PPC::BI__builtin_vsx_lxvll:
10516 {
10517 if(BuiltinID == PPC::BI__builtin_vsx_lxvl ||
10518 BuiltinID == PPC::BI__builtin_vsx_lxvll){
10519 Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy);
10520 }else {
10521 Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy);
10522 Ops[0] = Builder.CreateGEP(Ops[1], Ops[0]);
10523 Ops.pop_back();
10524 }
10525
10526 switch (BuiltinID) {
10527 default: llvm_unreachable("Unsupported ld/lvsl/lvsr intrinsic!")::llvm::llvm_unreachable_internal("Unsupported ld/lvsl/lvsr intrinsic!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10527)
;
10528 case PPC::BI__builtin_altivec_lvx:
10529 ID = Intrinsic::ppc_altivec_lvx;
10530 break;
10531 case PPC::BI__builtin_altivec_lvxl:
10532 ID = Intrinsic::ppc_altivec_lvxl;
10533 break;
10534 case PPC::BI__builtin_altivec_lvebx:
10535 ID = Intrinsic::ppc_altivec_lvebx;
10536 break;
10537 case PPC::BI__builtin_altivec_lvehx:
10538 ID = Intrinsic::ppc_altivec_lvehx;
10539 break;
10540 case PPC::BI__builtin_altivec_lvewx:
10541 ID = Intrinsic::ppc_altivec_lvewx;
10542 break;
10543 case PPC::BI__builtin_altivec_lvsl:
10544 ID = Intrinsic::ppc_altivec_lvsl;
10545 break;
10546 case PPC::BI__builtin_altivec_lvsr:
10547 ID = Intrinsic::ppc_altivec_lvsr;
10548 break;
10549 case PPC::BI__builtin_vsx_lxvd2x:
10550 ID = Intrinsic::ppc_vsx_lxvd2x;
10551 break;
10552 case PPC::BI__builtin_vsx_lxvw4x:
10553 ID = Intrinsic::ppc_vsx_lxvw4x;
10554 break;
10555 case PPC::BI__builtin_vsx_lxvd2x_be:
10556 ID = Intrinsic::ppc_vsx_lxvd2x_be;
10557 break;
10558 case PPC::BI__builtin_vsx_lxvw4x_be:
10559 ID = Intrinsic::ppc_vsx_lxvw4x_be;
10560 break;
10561 case PPC::BI__builtin_vsx_lxvl:
10562 ID = Intrinsic::ppc_vsx_lxvl;
10563 break;
10564 case PPC::BI__builtin_vsx_lxvll:
10565 ID = Intrinsic::ppc_vsx_lxvll;
10566 break;
10567 }
10568 llvm::Function *F = CGM.getIntrinsic(ID);
10569 return Builder.CreateCall(F, Ops, "");
10570 }
10571
10572 // vec_st, vec_xst_be
10573 case PPC::BI__builtin_altivec_stvx:
10574 case PPC::BI__builtin_altivec_stvxl:
10575 case PPC::BI__builtin_altivec_stvebx:
10576 case PPC::BI__builtin_altivec_stvehx:
10577 case PPC::BI__builtin_altivec_stvewx:
10578 case PPC::BI__builtin_vsx_stxvd2x:
10579 case PPC::BI__builtin_vsx_stxvw4x:
10580 case PPC::BI__builtin_vsx_stxvd2x_be:
10581 case PPC::BI__builtin_vsx_stxvw4x_be:
10582 case PPC::BI__builtin_vsx_stxvl:
10583 case PPC::BI__builtin_vsx_stxvll:
10584 {
10585 if(BuiltinID == PPC::BI__builtin_vsx_stxvl ||
10586 BuiltinID == PPC::BI__builtin_vsx_stxvll ){
10587 Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy);
10588 }else {
10589 Ops[2] = Builder.CreateBitCast(Ops[2], Int8PtrTy);
10590 Ops[1] = Builder.CreateGEP(Ops[2], Ops[1]);
10591 Ops.pop_back();
10592 }
10593
10594 switch (BuiltinID) {
10595 default: llvm_unreachable("Unsupported st intrinsic!")::llvm::llvm_unreachable_internal("Unsupported st intrinsic!"
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10595)
;
10596 case PPC::BI__builtin_altivec_stvx:
10597 ID = Intrinsic::ppc_altivec_stvx;
10598 break;
10599 case PPC::BI__builtin_altivec_stvxl:
10600 ID = Intrinsic::ppc_altivec_stvxl;
10601 break;
10602 case PPC::BI__builtin_altivec_stvebx:
10603 ID = Intrinsic::ppc_altivec_stvebx;
10604 break;
10605 case PPC::BI__builtin_altivec_stvehx:
10606 ID = Intrinsic::ppc_altivec_stvehx;
10607 break;
10608 case PPC::BI__builtin_altivec_stvewx:
10609 ID = Intrinsic::ppc_altivec_stvewx;
10610 break;
10611 case PPC::BI__builtin_vsx_stxvd2x:
10612 ID = Intrinsic::ppc_vsx_stxvd2x;
10613 break;
10614 case PPC::BI__builtin_vsx_stxvw4x:
10615 ID = Intrinsic::ppc_vsx_stxvw4x;
10616 break;
10617 case PPC::BI__builtin_vsx_stxvd2x_be:
10618 ID = Intrinsic::ppc_vsx_stxvd2x_be;
10619 break;
10620 case PPC::BI__builtin_vsx_stxvw4x_be:
10621 ID = Intrinsic::ppc_vsx_stxvw4x_be;
10622 break;
10623 case PPC::BI__builtin_vsx_stxvl:
10624 ID = Intrinsic::ppc_vsx_stxvl;
10625 break;
10626 case PPC::BI__builtin_vsx_stxvll:
10627 ID = Intrinsic::ppc_vsx_stxvll;
10628 break;
10629 }
10630 llvm::Function *F = CGM.getIntrinsic(ID);
10631 return Builder.CreateCall(F, Ops, "");
10632 }
10633 // Square root
10634 case PPC::BI__builtin_vsx_xvsqrtsp:
10635 case PPC::BI__builtin_vsx_xvsqrtdp: {
10636 llvm::Type *ResultType = ConvertType(E->getType());
10637 Value *X = EmitScalarExpr(E->getArg(0));
10638 ID = Intrinsic::sqrt;
10639 llvm::Function *F = CGM.getIntrinsic(ID, ResultType);
10640 return Builder.CreateCall(F, X);
10641 }
10642 // Count leading zeros
10643 case PPC::BI__builtin_altivec_vclzb:
10644 case PPC::BI__builtin_altivec_vclzh:
10645 case PPC::BI__builtin_altivec_vclzw:
10646 case PPC::BI__builtin_altivec_vclzd: {
10647 llvm::Type *ResultType = ConvertType(E->getType());
10648 Value *X = EmitScalarExpr(E->getArg(0));
10649 Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
10650 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType);
10651 return Builder.CreateCall(F, {X, Undef});
10652 }
10653 case PPC::BI__builtin_altivec_vctzb:
10654 case PPC::BI__builtin_altivec_vctzh:
10655 case PPC::BI__builtin_altivec_vctzw:
10656 case PPC::BI__builtin_altivec_vctzd: {
10657 llvm::Type *ResultType = ConvertType(E->getType());
10658 Value *X = EmitScalarExpr(E->getArg(0));
10659 Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
10660 Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType);
10661 return Builder.CreateCall(F, {X, Undef});
10662 }
10663 case PPC::BI__builtin_altivec_vpopcntb:
10664 case PPC::BI__builtin_altivec_vpopcnth:
10665 case PPC::BI__builtin_altivec_vpopcntw:
10666 case PPC::BI__builtin_altivec_vpopcntd: {
10667 llvm::Type *ResultType = ConvertType(E->getType());
10668 Value *X = EmitScalarExpr(E->getArg(0));
10669 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
10670 return Builder.CreateCall(F, X);
10671 }
10672 // Copy sign
10673 case PPC::BI__builtin_vsx_xvcpsgnsp:
10674 case PPC::BI__builtin_vsx_xvcpsgndp: {
10675 llvm::Type *ResultType = ConvertType(E->getType());
10676 Value *X = EmitScalarExpr(E->getArg(0));
10677 Value *Y = EmitScalarExpr(E->getArg(1));
10678 ID = Intrinsic::copysign;
10679 llvm::Function *F = CGM.getIntrinsic(ID, ResultType);
10680 return Builder.CreateCall(F, {X, Y});
10681 }
10682 // Rounding/truncation
10683 case PPC::BI__builtin_vsx_xvrspip:
10684 case PPC::BI__builtin_vsx_xvrdpip:
10685 case PPC::BI__builtin_vsx_xvrdpim:
10686 case PPC::BI__builtin_vsx_xvrspim:
10687 case PPC::BI__builtin_vsx_xvrdpi:
10688 case PPC::BI__builtin_vsx_xvrspi:
10689 case PPC::BI__builtin_vsx_xvrdpic:
10690 case PPC::BI__builtin_vsx_xvrspic:
10691 case PPC::BI__builtin_vsx_xvrdpiz:
10692 case PPC::BI__builtin_vsx_xvrspiz: {
10693 llvm::Type *ResultType = ConvertType(E->getType());
10694 Value *X = EmitScalarExpr(E->getArg(0));
10695 if (BuiltinID == PPC::BI__builtin_vsx_xvrdpim ||
10696 BuiltinID == PPC::BI__builtin_vsx_xvrspim)
10697 ID = Intrinsic::floor;
10698 else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpi ||
10699 BuiltinID == PPC::BI__builtin_vsx_xvrspi)
10700 ID = Intrinsic::round;
10701 else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpic ||
10702 BuiltinID == PPC::BI__builtin_vsx_xvrspic)
10703 ID = Intrinsic::nearbyint;
10704 else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpip ||
10705 BuiltinID == PPC::BI__builtin_vsx_xvrspip)
10706 ID = Intrinsic::ceil;
10707 else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpiz ||
10708 BuiltinID == PPC::BI__builtin_vsx_xvrspiz)
10709 ID = Intrinsic::trunc;
10710 llvm::Function *F = CGM.getIntrinsic(ID, ResultType);
10711 return Builder.CreateCall(F, X);
10712 }
10713
10714 // Absolute value
10715 case PPC::BI__builtin_vsx_xvabsdp:
10716 case PPC::BI__builtin_vsx_xvabssp: {
10717 llvm::Type *ResultType = ConvertType(E->getType());
10718 Value *X = EmitScalarExpr(E->getArg(0));
10719 llvm::Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
10720 return Builder.CreateCall(F, X);
10721 }
10722
10723 // FMA variations
10724 case PPC::BI__builtin_vsx_xvmaddadp:
10725 case PPC::BI__builtin_vsx_xvmaddasp:
10726 case PPC::BI__builtin_vsx_xvnmaddadp:
10727 case PPC::BI__builtin_vsx_xvnmaddasp:
10728 case PPC::BI__builtin_vsx_xvmsubadp:
10729 case PPC::BI__builtin_vsx_xvmsubasp:
10730 case PPC::BI__builtin_vsx_xvnmsubadp:
10731 case PPC::BI__builtin_vsx_xvnmsubasp: {
10732 llvm::Type *ResultType = ConvertType(E->getType());
10733 Value *X = EmitScalarExpr(E->getArg(0));
10734 Value *Y = EmitScalarExpr(E->getArg(1));
10735 Value *Z = EmitScalarExpr(E->getArg(2));
10736 Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
10737 llvm::Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
10738 switch (BuiltinID) {
10739 case PPC::BI__builtin_vsx_xvmaddadp:
10740 case PPC::BI__builtin_vsx_xvmaddasp:
10741 return Builder.CreateCall(F, {X, Y, Z});
10742 case PPC::BI__builtin_vsx_xvnmaddadp:
10743 case PPC::BI__builtin_vsx_xvnmaddasp:
10744 return Builder.CreateFSub(Zero,
10745 Builder.CreateCall(F, {X, Y, Z}), "sub");
10746 case PPC::BI__builtin_vsx_xvmsubadp:
10747 case PPC::BI__builtin_vsx_xvmsubasp:
10748 return Builder.CreateCall(F,
10749 {X, Y, Builder.CreateFSub(Zero, Z, "sub")});
10750 case PPC::BI__builtin_vsx_xvnmsubadp:
10751 case PPC::BI__builtin_vsx_xvnmsubasp:
10752 Value *FsubRes =
10753 Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")});
10754 return Builder.CreateFSub(Zero, FsubRes, "sub");
10755 }
10756 llvm_unreachable("Unknown FMA operation")::llvm::llvm_unreachable_internal("Unknown FMA operation", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10756)
;
10757 return nullptr; // Suppress no-return warning
10758 }
10759
10760 case PPC::BI__builtin_vsx_insertword: {
10761 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxinsertw);
10762
10763 // Third argument is a compile time constant int. It must be clamped to
10764 // to the range [0, 12].
10765 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
10766 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10767, __extension__ __PRETTY_FUNCTION__))
10767 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10767, __extension__ __PRETTY_FUNCTION__))
;
10768 const int64_t MaxIndex = 12;
10769 int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex);
10770
10771 // The builtin semantics don't exactly match the xxinsertw instructions
10772 // semantics (which ppc_vsx_xxinsertw follows). The builtin extracts the
10773 // word from the first argument, and inserts it in the second argument. The
10774 // instruction extracts the word from its second input register and inserts
10775 // it into its first input register, so swap the first and second arguments.
10776 std::swap(Ops[0], Ops[1]);
10777
10778 // Need to cast the second argument from a vector of unsigned int to a
10779 // vector of long long.
10780 Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int64Ty, 2));
10781
10782 if (getTarget().isLittleEndian()) {
10783 // Create a shuffle mask of (1, 0)
10784 Constant *ShuffleElts[2] = { ConstantInt::get(Int32Ty, 1),
10785 ConstantInt::get(Int32Ty, 0)
10786 };
10787 Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts);
10788
10789 // Reverse the double words in the vector we will extract from.
10790 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2));
10791 Ops[0] = Builder.CreateShuffleVector(Ops[0], Ops[0], ShuffleMask);
10792
10793 // Reverse the index.
10794 Index = MaxIndex - Index;
10795 }
10796
10797 // Intrinsic expects the first arg to be a vector of int.
10798 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4));
10799 Ops[2] = ConstantInt::getSigned(Int32Ty, Index);
10800 return Builder.CreateCall(F, Ops);
10801 }
10802
10803 case PPC::BI__builtin_vsx_extractuword: {
10804 llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxextractuw);
10805
10806 // Intrinsic expects the first argument to be a vector of doublewords.
10807 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2));
10808
10809 // The second argument is a compile time constant int that needs to
10810 // be clamped to the range [0, 12].
10811 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[1]);
10812 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10813, __extension__ __PRETTY_FUNCTION__))
10813 "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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10813, __extension__ __PRETTY_FUNCTION__))
;
10814 const int64_t MaxIndex = 12;
10815 int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex);
10816
10817 if (getTarget().isLittleEndian()) {
10818 // Reverse the index.
10819 Index = MaxIndex - Index;
10820 Ops[1] = ConstantInt::getSigned(Int32Ty, Index);
10821
10822 // Emit the call, then reverse the double words of the results vector.
10823 Value *Call = Builder.CreateCall(F, Ops);
10824
10825 // Create a shuffle mask of (1, 0)
10826 Constant *ShuffleElts[2] = { ConstantInt::get(Int32Ty, 1),
10827 ConstantInt::get(Int32Ty, 0)
10828 };
10829 Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts);
10830
10831 Value *ShuffleCall = Builder.CreateShuffleVector(Call, Call, ShuffleMask);
10832 return ShuffleCall;
10833 } else {
10834 Ops[1] = ConstantInt::getSigned(Int32Ty, Index);
10835 return Builder.CreateCall(F, Ops);
10836 }
10837 }
10838
10839 case PPC::BI__builtin_vsx_xxpermdi: {
10840 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
10841 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10841, __extension__ __PRETTY_FUNCTION__))
;
10842
10843 unsigned Index = ArgCI->getZExtValue();
10844 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2));
10845 Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int64Ty, 2));
10846
10847 // Account for endianness by treating this as just a shuffle. So we use the
10848 // same indices for both LE and BE in order to produce expected results in
10849 // both cases.
10850 unsigned ElemIdx0 = (Index & 2) >> 1;
10851 unsigned ElemIdx1 = 2 + (Index & 1);
10852
10853 Constant *ShuffleElts[2] = {ConstantInt::get(Int32Ty, ElemIdx0),
10854 ConstantInt::get(Int32Ty, ElemIdx1)};
10855 Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts);
10856
10857 Value *ShuffleCall =
10858 Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleMask);
10859 QualType BIRetType = E->getType();
10860 auto RetTy = ConvertType(BIRetType);
10861 return Builder.CreateBitCast(ShuffleCall, RetTy);
10862 }
10863
10864 case PPC::BI__builtin_vsx_xxsldwi: {
10865 ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
10866 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-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 10866, __extension__ __PRETTY_FUNCTION__))
;
10867 unsigned Index = ArgCI->getZExtValue() & 0x3;
10868 Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4));
10869 Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int32Ty, 4));
10870
10871 // Create a shuffle mask
10872 unsigned ElemIdx0;
10873 unsigned ElemIdx1;
10874 unsigned ElemIdx2;
10875 unsigned ElemIdx3;
10876 if (getTarget().isLittleEndian()) {
10877 // Little endian element N comes from element 8+N-Index of the
10878 // concatenated wide vector (of course, using modulo arithmetic on
10879 // the total number of elements).
10880 ElemIdx0 = (8 - Index) % 8;
10881 ElemIdx1 = (9 - Index) % 8;
10882 ElemIdx2 = (10 - Index) % 8;
10883 ElemIdx3 = (11 - Index) % 8;
10884 } else {
10885 // Big endian ElemIdx<N> = Index + N
10886 ElemIdx0 = Index;
10887 ElemIdx1 = Index + 1;
10888 ElemIdx2 = Index + 2;
10889 ElemIdx3 = Index + 3;
10890 }
10891
10892 Constant *ShuffleElts[4] = {ConstantInt::get(Int32Ty, ElemIdx0),
10893 ConstantInt::get(Int32Ty, ElemIdx1),
10894 ConstantInt::get(Int32Ty, ElemIdx2),
10895 ConstantInt::get(Int32Ty, ElemIdx3)};
10896
10897 Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts);
10898 Value *ShuffleCall =
10899 Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleMask);
10900 QualType BIRetType = E->getType();
10901 auto RetTy = ConvertType(BIRetType);
10902 return Builder.CreateBitCast(ShuffleCall, RetTy);
10903 }
10904 }
10905}
10906
10907Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
10908 const CallExpr *E) {
10909 switch (BuiltinID) {
10910 case AMDGPU::BI__builtin_amdgcn_div_scale:
10911 case AMDGPU::BI__builtin_amdgcn_div_scalef: {
10912 // Translate from the intrinsics's struct return to the builtin's out
10913 // argument.
10914
10915 Address FlagOutPtr = EmitPointerWithAlignment(E->getArg(3));
10916
10917 llvm::Value *X = EmitScalarExpr(E->getArg(0));
10918 llvm::Value *Y = EmitScalarExpr(E->getArg(1));
10919 llvm::Value *Z = EmitScalarExpr(E->getArg(2));
10920
10921 llvm::Value *Callee = CGM.getIntrinsic(Intrinsic::amdgcn_div_scale,
10922 X->getType());
10923
10924 llvm::Value *Tmp = Builder.CreateCall(Callee, {X, Y, Z});
10925
10926 llvm::Value *Result = Builder.CreateExtractValue(Tmp, 0);
10927 llvm::Value *Flag = Builder.CreateExtractValue(Tmp, 1);
10928
10929 llvm::Type *RealFlagType
10930 = FlagOutPtr.getPointer()->getType()->getPointerElementType();
10931
10932 llvm::Value *FlagExt = Builder.CreateZExt(Flag, RealFlagType);
10933 Builder.CreateStore(FlagExt, FlagOutPtr);
10934 return Result;
10935 }
10936 case AMDGPU::BI__builtin_amdgcn_div_fmas:
10937 case AMDGPU::BI__builtin_amdgcn_div_fmasf: {
10938 llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
10939 llvm::Value *Src1 = EmitScalarExpr(E->getArg(1));
10940 llvm::Value *Src2 = EmitScalarExpr(E->getArg(2));
10941 llvm::Value *Src3 = EmitScalarExpr(E->getArg(3));
10942
10943 llvm::Value *F = CGM.getIntrinsic(Intrinsic::amdgcn_div_fmas,
10944 Src0->getType());
10945 llvm::Value *Src3ToBool = Builder.CreateIsNotNull(Src3);
10946 return Builder.CreateCall(F, {Src0, Src1, Src2, Src3ToBool});
10947 }
10948
10949 case AMDGPU::BI__builtin_amdgcn_ds_swizzle:
10950 return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_ds_swizzle);
10951 case AMDGPU::BI__builtin_amdgcn_mov_dpp: {
10952 llvm::SmallVector<llvm::Value *, 5> Args;
10953 for (unsigned I = 0; I != 5; ++I)
10954 Args.push_back(EmitScalarExpr(E->getArg(I)));
10955 Value *F = CGM.getIntrinsic(Intrinsic::amdgcn_mov_dpp,
10956 Args[0]->getType());
10957 return Builder.CreateCall(F, Args);
10958 }
10959 case AMDGPU::BI__builtin_amdgcn_div_fixup:
10960 case AMDGPU::BI__builtin_amdgcn_div_fixupf:
10961 case AMDGPU::BI__builtin_amdgcn_div_fixuph:
10962 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_div_fixup);
10963 case AMDGPU::BI__builtin_amdgcn_trig_preop:
10964 case AMDGPU::BI__builtin_amdgcn_trig_preopf:
10965 return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_trig_preop);
10966 case AMDGPU::BI__builtin_amdgcn_rcp:
10967 case AMDGPU::BI__builtin_amdgcn_rcpf:
10968 case AMDGPU::BI__builtin_amdgcn_rcph:
10969 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rcp);
10970 case AMDGPU::BI__builtin_amdgcn_rsq:
10971 case AMDGPU::BI__builtin_amdgcn_rsqf:
10972 case AMDGPU::BI__builtin_amdgcn_rsqh:
10973 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq);
10974 case AMDGPU::BI__builtin_amdgcn_rsq_clamp:
10975 case AMDGPU::BI__builtin_amdgcn_rsq_clampf:
10976 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq_clamp);
10977 case AMDGPU::BI__builtin_amdgcn_sinf:
10978 case AMDGPU::BI__builtin_amdgcn_sinh:
10979 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_sin);
10980 case AMDGPU::BI__builtin_amdgcn_cosf:
10981 case AMDGPU::BI__builtin_amdgcn_cosh:
10982 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_cos);
10983 case AMDGPU::BI__builtin_amdgcn_log_clampf:
10984 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_log_clamp);
10985 case AMDGPU::BI__builtin_amdgcn_ldexp:
10986 case AMDGPU::BI__builtin_amdgcn_ldexpf:
10987 case AMDGPU::BI__builtin_amdgcn_ldexph:
10988 return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_ldexp);
10989 case AMDGPU::BI__builtin_amdgcn_frexp_mant:
10990 case AMDGPU::BI__builtin_amdgcn_frexp_mantf:
10991 case AMDGPU::BI__builtin_amdgcn_frexp_manth:
10992 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_frexp_mant);
10993 case AMDGPU::BI__builtin_amdgcn_frexp_exp:
10994 case AMDGPU::BI__builtin_amdgcn_frexp_expf: {
10995 Value *Src0 = EmitScalarExpr(E->getArg(0));
10996 Value *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp,
10997 { Builder.getInt32Ty(), Src0->getType() });
10998 return Builder.CreateCall(F, Src0);
10999 }
11000 case AMDGPU::BI__builtin_amdgcn_frexp_exph: {
11001 Value *Src0 = EmitScalarExpr(E->getArg(0));
11002 Value *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp,
11003 { Builder.getInt16Ty(), Src0->getType() });
11004 return Builder.CreateCall(F, Src0);
11005 }
11006 case AMDGPU::BI__builtin_amdgcn_fract:
11007 case AMDGPU::BI__builtin_amdgcn_fractf:
11008 case AMDGPU::BI__builtin_amdgcn_fracth:
11009 return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_fract);
11010 case AMDGPU::BI__builtin_amdgcn_lerp:
11011 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_lerp);
11012 case AMDGPU::BI__builtin_amdgcn_uicmp:
11013 case AMDGPU::BI__builtin_amdgcn_uicmpl:
11014 case AMDGPU::BI__builtin_amdgcn_sicmp:
11015 case AMDGPU::BI__builtin_amdgcn_sicmpl:
11016 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_icmp);
11017 case AMDGPU::BI__builtin_amdgcn_fcmp:
11018 case AMDGPU::BI__builtin_amdgcn_fcmpf:
11019 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fcmp);
11020 case AMDGPU::BI__builtin_amdgcn_class:
11021 case AMDGPU::BI__builtin_amdgcn_classf:
11022 case AMDGPU::BI__builtin_amdgcn_classh:
11023 return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_class);
11024 case AMDGPU::BI__builtin_amdgcn_fmed3f:
11025 case AMDGPU::BI__builtin_amdgcn_fmed3h:
11026 return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fmed3);
11027 case AMDGPU::BI__builtin_amdgcn_read_exec: {
11028 CallInst *CI = cast<CallInst>(
11029 EmitSpecialRegisterBuiltin(*this, E, Int64Ty, Int64Ty, true, "exec"));
11030 CI->setConvergent();
11031 return CI;
11032 }
11033 case AMDGPU::BI__builtin_amdgcn_read_exec_lo:
11034 case AMDGPU::BI__builtin_amdgcn_read_exec_hi: {
11035 StringRef RegName = BuiltinID == AMDGPU::BI__builtin_amdgcn_read_exec_lo ?
11036 "exec_lo" : "exec_hi";
11037 CallInst *CI = cast<CallInst>(
11038 EmitSpecialRegisterBuiltin(*this, E, Int32Ty, Int32Ty, true, RegName));
11039 CI->setConvergent();
11040 return CI;
11041 }
11042 case AMDGPU::BI__builtin_amdgcn_ds_faddf:
11043 case AMDGPU::BI__builtin_amdgcn_ds_fminf:
11044 case AMDGPU::BI__builtin_amdgcn_ds_fmaxf: {
11045 llvm::SmallVector<llvm::Value *, 5> Args;
11046 for (unsigned I = 0; I != 5; ++I)
11047 Args.push_back(EmitScalarExpr(E->getArg(I)));
11048 const llvm::Type *PtrTy = Args[0]->getType();
11049 // check pointer parameter
11050 if (!PtrTy->isPointerTy() ||
11051 E->getArg(0)
11052 ->getType()
11053 ->getPointeeType()
11054 .getQualifiers()
11055 .getAddressSpace() != LangAS::opencl_local ||
11056 !PtrTy->getPointerElementType()->isFloatTy()) {
11057 CGM.Error(E->getArg(0)->getLocStart(),
11058 "parameter should have type \"local float*\"");
11059 return nullptr;
11060 }
11061 // check float parameter
11062 if (!Args[1]->getType()->isFloatTy()) {
11063 CGM.Error(E->getArg(1)->getLocStart(),
11064 "parameter should have type \"float\"");
11065 return nullptr;
11066 }
11067
11068 Intrinsic::ID ID;
11069 switch (BuiltinID) {
11070 case AMDGPU::BI__builtin_amdgcn_ds_faddf:
11071 ID = Intrinsic::amdgcn_ds_fadd;
11072 break;
11073 case AMDGPU::BI__builtin_amdgcn_ds_fminf:
11074 ID = Intrinsic::amdgcn_ds_fmin;
11075 break;
11076 case AMDGPU::BI__builtin_amdgcn_ds_fmaxf:
11077 ID = Intrinsic::amdgcn_ds_fmax;
11078 break;
11079 default:
11080 llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11080)
;
11081 }
11082 Value *F = CGM.getIntrinsic(ID);
11083 return Builder.CreateCall(F, Args);
11084 }
11085
11086 // amdgcn workitem
11087 case AMDGPU::BI__builtin_amdgcn_workitem_id_x:
11088 return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_x, 0, 1024);
11089 case AMDGPU::BI__builtin_amdgcn_workitem_id_y:
11090 return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_y, 0, 1024);
11091 case AMDGPU::BI__builtin_amdgcn_workitem_id_z:
11092 return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_z, 0, 1024);
11093
11094 // r600 intrinsics
11095 case AMDGPU::BI__builtin_r600_recipsqrt_ieee:
11096 case AMDGPU::BI__builtin_r600_recipsqrt_ieeef:
11097 return emitUnaryBuiltin(*this, E, Intrinsic::r600_recipsqrt_ieee);
11098 case AMDGPU::BI__builtin_r600_read_tidig_x:
11099 return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_x, 0, 1024);
11100 case AMDGPU::BI__builtin_r600_read_tidig_y:
11101 return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_y, 0, 1024);
11102 case AMDGPU::BI__builtin_r600_read_tidig_z:
11103 return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_z, 0, 1024);
11104 default:
11105 return nullptr;
11106 }
11107}
11108
11109/// Handle a SystemZ function in which the final argument is a pointer
11110/// to an int that receives the post-instruction CC value. At the LLVM level
11111/// this is represented as a function that returns a {result, cc} pair.
11112static Value *EmitSystemZIntrinsicWithCC(CodeGenFunction &CGF,
11113 unsigned IntrinsicID,
11114 const CallExpr *E) {
11115 unsigned NumArgs = E->getNumArgs() - 1;
11116 SmallVector<Value *, 8> Args(NumArgs);
11117 for (unsigned I = 0; I < NumArgs; ++I)
11118 Args[I] = CGF.EmitScalarExpr(E->getArg(I));
11119 Address CCPtr = CGF.EmitPointerWithAlignment(E->getArg(NumArgs));
11120 Value *F = CGF.CGM.getIntrinsic(IntrinsicID);
11121 Value *Call = CGF.Builder.CreateCall(F, Args);
11122 Value *CC = CGF.Builder.CreateExtractValue(Call, 1);
11123 CGF.Builder.CreateStore(CC, CCPtr);
11124 return CGF.Builder.CreateExtractValue(Call, 0);
11125}
11126
11127Value *CodeGenFunction::EmitSystemZBuiltinExpr(unsigned BuiltinID,
11128 const CallExpr *E) {
11129 switch (BuiltinID) {
11130 case SystemZ::BI__builtin_tbegin: {
11131 Value *TDB = EmitScalarExpr(E->getArg(0));
11132 Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c);
11133 Value *F = CGM.getIntrinsic(Intrinsic::s390_tbegin);
11134 return Builder.CreateCall(F, {TDB, Control});
11135 }
11136 case SystemZ::BI__builtin_tbegin_nofloat: {
11137 Value *TDB = EmitScalarExpr(E->getArg(0));
11138 Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c);
11139 Value *F = CGM.getIntrinsic(Intrinsic::s390_tbegin_nofloat);
11140 return Builder.CreateCall(F, {TDB, Control});
11141 }
11142 case SystemZ::BI__builtin_tbeginc: {
11143 Value *TDB = llvm::ConstantPointerNull::get(Int8PtrTy);
11144 Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff08);
11145 Value *F = CGM.getIntrinsic(Intrinsic::s390_tbeginc);
11146 return Builder.CreateCall(F, {TDB, Control});
11147 }
11148 case SystemZ::BI__builtin_tabort: {
11149 Value *Data = EmitScalarExpr(E->getArg(0));
11150 Value *F = CGM.getIntrinsic(Intrinsic::s390_tabort);
11151 return Builder.CreateCall(F, Builder.CreateSExt(Data, Int64Ty, "tabort"));
11152 }
11153 case SystemZ::BI__builtin_non_tx_store: {
11154 Value *Address = EmitScalarExpr(E->getArg(0));
11155 Value *Data = EmitScalarExpr(E->getArg(1));
11156 Value *F = CGM.getIntrinsic(Intrinsic::s390_ntstg);
11157 return Builder.CreateCall(F, {Data, Address});
11158 }
11159
11160 // Vector builtins. Note that most vector builtins are mapped automatically
11161 // to target-specific LLVM intrinsics. The ones handled specially here can
11162 // be represented via standard LLVM IR, which is preferable to enable common
11163 // LLVM optimizations.
11164
11165 case SystemZ::BI__builtin_s390_vpopctb:
11166 case SystemZ::BI__builtin_s390_vpopcth:
11167 case SystemZ::BI__builtin_s390_vpopctf:
11168 case SystemZ::BI__builtin_s390_vpopctg: {
11169 llvm::Type *ResultType = ConvertType(E->getType());
11170 Value *X = EmitScalarExpr(E->getArg(0));
11171 Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
11172 return Builder.CreateCall(F, X);
11173 }
11174
11175 case SystemZ::BI__builtin_s390_vclzb:
11176 case SystemZ::BI__builtin_s390_vclzh:
11177 case SystemZ::BI__builtin_s390_vclzf:
11178 case SystemZ::BI__builtin_s390_vclzg: {
11179 llvm::Type *ResultType = ConvertType(E->getType());
11180 Value *X = EmitScalarExpr(E->getArg(0));
11181 Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
11182 Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType);
11183 return Builder.CreateCall(F, {X, Undef});
11184 }
11185
11186 case SystemZ::BI__builtin_s390_vctzb:
11187 case SystemZ::BI__builtin_s390_vctzh:
11188 case SystemZ::BI__builtin_s390_vctzf:
11189 case SystemZ::BI__builtin_s390_vctzg: {
11190 llvm::Type *ResultType = ConvertType(E->getType());
11191 Value *X = EmitScalarExpr(E->getArg(0));
11192 Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
11193 Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType);
11194 return Builder.CreateCall(F, {X, Undef});
11195 }
11196
11197 case SystemZ::BI__builtin_s390_vfsqsb:
11198 case SystemZ::BI__builtin_s390_vfsqdb: {
11199 llvm::Type *ResultType = ConvertType(E->getType());
11200 Value *X = EmitScalarExpr(E->getArg(0));
11201 Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType);
11202 return Builder.CreateCall(F, X);
11203 }
11204 case SystemZ::BI__builtin_s390_vfmasb:
11205 case SystemZ::BI__builtin_s390_vfmadb: {
11206 llvm::Type *ResultType = ConvertType(E->getType());
11207 Value *X = EmitScalarExpr(E->getArg(0));
11208 Value *Y = EmitScalarExpr(E->getArg(1));
11209 Value *Z = EmitScalarExpr(E->getArg(2));
11210 Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
11211 return Builder.CreateCall(F, {X, Y, Z});
11212 }
11213 case SystemZ::BI__builtin_s390_vfmssb:
11214 case SystemZ::BI__builtin_s390_vfmsdb: {
11215 llvm::Type *ResultType = ConvertType(E->getType());
11216 Value *X = EmitScalarExpr(E->getArg(0));
11217 Value *Y = EmitScalarExpr(E->getArg(1));
11218 Value *Z = EmitScalarExpr(E->getArg(2));
11219 Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
11220 Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
11221 return Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")});
11222 }
11223 case SystemZ::BI__builtin_s390_vfnmasb:
11224 case SystemZ::BI__builtin_s390_vfnmadb: {
11225 llvm::Type *ResultType = ConvertType(E->getType());
11226 Value *X = EmitScalarExpr(E->getArg(0));
11227 Value *Y = EmitScalarExpr(E->getArg(1));
11228 Value *Z = EmitScalarExpr(E->getArg(2));
11229 Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
11230 Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
11231 return Builder.CreateFSub(Zero, Builder.CreateCall(F, {X, Y, Z}), "sub");
11232 }
11233 case SystemZ::BI__builtin_s390_vfnmssb:
11234 case SystemZ::BI__builtin_s390_vfnmsdb: {
11235 llvm::Type *ResultType = ConvertType(E->getType());
11236 Value *X = EmitScalarExpr(E->getArg(0));
11237 Value *Y = EmitScalarExpr(E->getArg(1));
11238 Value *Z = EmitScalarExpr(E->getArg(2));
11239 Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
11240 Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
11241 Value *NegZ = Builder.CreateFSub(Zero, Z, "sub");
11242 return Builder.CreateFSub(Zero, Builder.CreateCall(F, {X, Y, NegZ}));
11243 }
11244 case SystemZ::BI__builtin_s390_vflpsb:
11245 case SystemZ::BI__builtin_s390_vflpdb: {
11246 llvm::Type *ResultType = ConvertType(E->getType());
11247 Value *X = EmitScalarExpr(E->getArg(0));
11248 Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
11249 return Builder.CreateCall(F, X);
11250 }
11251 case SystemZ::BI__builtin_s390_vflnsb:
11252 case SystemZ::BI__builtin_s390_vflndb: {
11253 llvm::Type *ResultType = ConvertType(E->getType());
11254 Value *X = EmitScalarExpr(E->getArg(0));
11255 Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
11256 Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
11257 return Builder.CreateFSub(Zero, Builder.CreateCall(F, X), "sub");
11258 }
11259 case SystemZ::BI__builtin_s390_vfisb:
11260 case SystemZ::BI__builtin_s390_vfidb: {
11261 llvm::Type *ResultType = ConvertType(E->getType());
11262 Value *X = EmitScalarExpr(E->getArg(0));
11263 // Constant-fold the M4 and M5 mask arguments.
11264 llvm::APSInt M4, M5;
11265 bool IsConstM4 = E->getArg(1)->isIntegerConstantExpr(M4, getContext());
11266 bool IsConstM5 = E->getArg(2)->isIntegerConstantExpr(M5, getContext());
11267 assert(IsConstM4 && IsConstM5 && "Constant arg isn't actually constant?")(static_cast <bool> (IsConstM4 && IsConstM5 &&
"Constant arg isn't actually constant?") ? void (0) : __assert_fail
("IsConstM4 && IsConstM5 && \"Constant arg isn't actually constant?\""
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11267, __extension__ __PRETTY_FUNCTION__))
;
11268 (void)IsConstM4; (void)IsConstM5;
11269 // Check whether this instance can be represented via a LLVM standard
11270 // intrinsic. We only support some combinations of M4 and M5.
11271 Intrinsic::ID ID = Intrinsic::not_intrinsic;
11272 switch (M4.getZExtValue()) {
11273 default: break;
11274 case 0: // IEEE-inexact exception allowed
11275 switch (M5.getZExtValue()) {
11276 default: break;
11277 case 0: ID = Intrinsic::rint; break;
11278 }
11279 break;
11280 case 4: // IEEE-inexact exception suppressed
11281 switch (M5.getZExtValue()) {
11282 default: break;
11283 case 0: ID = Intrinsic::nearbyint; break;
11284 case 1: ID = Intrinsic::round; break;
11285 case 5: ID = Intrinsic::trunc; break;
11286 case 6: ID = Intrinsic::ceil; break;
11287 case 7: ID = Intrinsic::floor; break;
11288 }
11289 break;
11290 }
11291 if (ID != Intrinsic::not_intrinsic) {
11292 Function *F = CGM.getIntrinsic(ID, ResultType);
11293 return Builder.CreateCall(F, X);
11294 }
11295 switch (BuiltinID) {
11296 case SystemZ::BI__builtin_s390_vfisb: ID = Intrinsic::s390_vfisb; break;
11297 case SystemZ::BI__builtin_s390_vfidb: ID = Intrinsic::s390_vfidb; break;
11298 default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11298)
;
11299 }
11300 Function *F = CGM.getIntrinsic(ID);
11301 Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
11302 Value *M5Value = llvm::ConstantInt::get(getLLVMContext(), M5);
11303 return Builder.CreateCall(F, {X, M4Value, M5Value});
11304 }
11305 case SystemZ::BI__builtin_s390_vfmaxsb:
11306 case SystemZ::BI__builtin_s390_vfmaxdb: {
11307 llvm::Type *ResultType = ConvertType(E->getType());
11308 Value *X = EmitScalarExpr(E->getArg(0));
11309 Value *Y = EmitScalarExpr(E->getArg(1));
11310 // Constant-fold the M4 mask argument.
11311 llvm::APSInt M4;
11312 bool IsConstM4 = E->getArg(2)->isIntegerConstantExpr(M4, getContext());
11313 assert(IsConstM4 && "Constant arg isn't actually constant?")(static_cast <bool> (IsConstM4 && "Constant arg isn't actually constant?"
) ? void (0) : __assert_fail ("IsConstM4 && \"Constant arg isn't actually constant?\""
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11313, __extension__ __PRETTY_FUNCTION__))
;
11314 (void)IsConstM4;
11315 // Check whether this instance can be represented via a LLVM standard
11316 // intrinsic. We only support some values of M4.
11317 Intrinsic::ID ID = Intrinsic::not_intrinsic;
11318 switch (M4.getZExtValue()) {
11319 default: break;
11320 case 4: ID = Intrinsic::maxnum; break;
11321 }
11322 if (ID != Intrinsic::not_intrinsic) {
11323 Function *F = CGM.getIntrinsic(ID, ResultType);
11324 return Builder.CreateCall(F, {X, Y});
11325 }
11326 switch (BuiltinID) {
11327 case SystemZ::BI__builtin_s390_vfmaxsb: ID = Intrinsic::s390_vfmaxsb; break;
11328 case SystemZ::BI__builtin_s390_vfmaxdb: ID = Intrinsic::s390_vfmaxdb; break;
11329 default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11329)
;
11330 }
11331 Function *F = CGM.getIntrinsic(ID);
11332 Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
11333 return Builder.CreateCall(F, {X, Y, M4Value});
11334 }
11335 case SystemZ::BI__builtin_s390_vfminsb:
11336 case SystemZ::BI__builtin_s390_vfmindb: {
11337 llvm::Type *ResultType = ConvertType(E->getType());
11338 Value *X = EmitScalarExpr(E->getArg(0));
11339 Value *Y = EmitScalarExpr(E->getArg(1));
11340 // Constant-fold the M4 mask argument.
11341 llvm::APSInt M4;
11342 bool IsConstM4 = E->getArg(2)->isIntegerConstantExpr(M4, getContext());
11343 assert(IsConstM4 && "Constant arg isn't actually constant?")(static_cast <bool> (IsConstM4 && "Constant arg isn't actually constant?"
) ? void (0) : __assert_fail ("IsConstM4 && \"Constant arg isn't actually constant?\""
, "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11343, __extension__ __PRETTY_FUNCTION__))
;
11344 (void)IsConstM4;
11345 // Check whether this instance can be represented via a LLVM standard
11346 // intrinsic. We only support some values of M4.
11347 Intrinsic::ID ID = Intrinsic::not_intrinsic;
11348 switch (M4.getZExtValue()) {
11349 default: break;
11350 case 4: ID = Intrinsic::minnum; break;
11351 }
11352 if (ID != Intrinsic::not_intrinsic) {
11353 Function *F = CGM.getIntrinsic(ID, ResultType);
11354 return Builder.CreateCall(F, {X, Y});
11355 }
11356 switch (BuiltinID) {
11357 case SystemZ::BI__builtin_s390_vfminsb: ID = Intrinsic::s390_vfminsb; break;
11358 case SystemZ::BI__builtin_s390_vfmindb: ID = Intrinsic::s390_vfmindb; break;
11359 default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11359)
;
11360 }
11361 Function *F = CGM.getIntrinsic(ID);
11362 Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
11363 return Builder.CreateCall(F, {X, Y, M4Value});
11364 }
11365
11366 // Vector intrisincs that output the post-instruction CC value.
11367
11368#define INTRINSIC_WITH_CC(NAME) \
11369 case SystemZ::BI__builtin_##NAME: \
11370 return EmitSystemZIntrinsicWithCC(*this, Intrinsic::NAME, E)
11371
11372 INTRINSIC_WITH_CC(s390_vpkshs);
11373 INTRINSIC_WITH_CC(s390_vpksfs);
11374 INTRINSIC_WITH_CC(s390_vpksgs);
11375
11376 INTRINSIC_WITH_CC(s390_vpklshs);
11377 INTRINSIC_WITH_CC(s390_vpklsfs);
11378 INTRINSIC_WITH_CC(s390_vpklsgs);
11379
11380 INTRINSIC_WITH_CC(s390_vceqbs);
11381 INTRINSIC_WITH_CC(s390_vceqhs);
11382 INTRINSIC_WITH_CC(s390_vceqfs);
11383 INTRINSIC_WITH_CC(s390_vceqgs);
11384
11385 INTRINSIC_WITH_CC(s390_vchbs);
11386 INTRINSIC_WITH_CC(s390_vchhs);
11387 INTRINSIC_WITH_CC(s390_vchfs);
11388 INTRINSIC_WITH_CC(s390_vchgs);
11389
11390 INTRINSIC_WITH_CC(s390_vchlbs);
11391 INTRINSIC_WITH_CC(s390_vchlhs);
11392 INTRINSIC_WITH_CC(s390_vchlfs);
11393 INTRINSIC_WITH_CC(s390_vchlgs);
11394
11395 INTRINSIC_WITH_CC(s390_vfaebs);
11396 INTRINSIC_WITH_CC(s390_vfaehs);
11397 INTRINSIC_WITH_CC(s390_vfaefs);
11398
11399 INTRINSIC_WITH_CC(s390_vfaezbs);
11400 INTRINSIC_WITH_CC(s390_vfaezhs);
11401 INTRINSIC_WITH_CC(s390_vfaezfs);
11402
11403 INTRINSIC_WITH_CC(s390_vfeebs);
11404 INTRINSIC_WITH_CC(s390_vfeehs);
11405 INTRINSIC_WITH_CC(s390_vfeefs);
11406
11407 INTRINSIC_WITH_CC(s390_vfeezbs);
11408 INTRINSIC_WITH_CC(s390_vfeezhs);
11409 INTRINSIC_WITH_CC(s390_vfeezfs);
11410
11411 INTRINSIC_WITH_CC(s390_vfenebs);
11412 INTRINSIC_WITH_CC(s390_vfenehs);
11413 INTRINSIC_WITH_CC(s390_vfenefs);
11414
11415 INTRINSIC_WITH_CC(s390_vfenezbs);
11416 INTRINSIC_WITH_CC(s390_vfenezhs);
11417 INTRINSIC_WITH_CC(s390_vfenezfs);
11418
11419 INTRINSIC_WITH_CC(s390_vistrbs);
11420 INTRINSIC_WITH_CC(s390_vistrhs);
11421 INTRINSIC_WITH_CC(s390_vistrfs);
11422
11423 INTRINSIC_WITH_CC(s390_vstrcbs);
11424 INTRINSIC_WITH_CC(s390_vstrchs);
11425 INTRINSIC_WITH_CC(s390_vstrcfs);
11426
11427 INTRINSIC_WITH_CC(s390_vstrczbs);
11428 INTRINSIC_WITH_CC(s390_vstrczhs);
11429 INTRINSIC_WITH_CC(s390_vstrczfs);
11430
11431 INTRINSIC_WITH_CC(s390_vfcesbs);
11432 INTRINSIC_WITH_CC(s390_vfcedbs);
11433 INTRINSIC_WITH_CC(s390_vfchsbs);
11434 INTRINSIC_WITH_CC(s390_vfchdbs);
11435 INTRINSIC_WITH_CC(s390_vfchesbs);
11436 INTRINSIC_WITH_CC(s390_vfchedbs);
11437
11438 INTRINSIC_WITH_CC(s390_vftcisb);
11439 INTRINSIC_WITH_CC(s390_vftcidb);
11440
11441#undef INTRINSIC_WITH_CC
11442
11443 default:
11444 return nullptr;
11445 }
11446}
11447
11448Value *CodeGenFunction::EmitNVPTXBuiltinExpr(unsigned BuiltinID,
11449 const CallExpr *E) {
11450 auto MakeLdg = [&](unsigned IntrinsicID) {
11451 Value *Ptr = EmitScalarExpr(E->getArg(0));
11452 clang::CharUnits Align =
11453 getNaturalPointeeTypeAlignment(E->getArg(0)->getType());
11454 return Builder.CreateCall(
11455 CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(),
11456 Ptr->getType()}),
11457 {Ptr, ConstantInt::get(Builder.getInt32Ty(), Align.getQuantity())});
11458 };
11459 auto MakeScopedAtomic = [&](unsigned IntrinsicID) {
11460 Value *Ptr = EmitScalarExpr(E->getArg(0));
11461 return Builder.CreateCall(
11462 CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(),
11463 Ptr->getType()}),
11464 {Ptr, EmitScalarExpr(E->getArg(1))});
11465 };
11466 switch (BuiltinID) {
11467 case NVPTX::BI__nvvm_atom_add_gen_i:
11468 case NVPTX::BI__nvvm_atom_add_gen_l:
11469 case NVPTX::BI__nvvm_atom_add_gen_ll:
11470 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Add, E);
11471
11472 case NVPTX::BI__nvvm_atom_sub_gen_i:
11473 case NVPTX::BI__nvvm_atom_sub_gen_l:
11474 case NVPTX::BI__nvvm_atom_sub_gen_ll:
11475 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Sub, E);
11476
11477 case NVPTX::BI__nvvm_atom_and_gen_i:
11478 case NVPTX::BI__nvvm_atom_and_gen_l:
11479 case NVPTX::BI__nvvm_atom_and_gen_ll:
11480 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::And, E);
11481
11482 case NVPTX::BI__nvvm_atom_or_gen_i:
11483 case NVPTX::BI__nvvm_atom_or_gen_l:
11484 case NVPTX::BI__nvvm_atom_or_gen_ll:
11485 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Or, E);
11486
11487 case NVPTX::BI__nvvm_atom_xor_gen_i:
11488 case NVPTX::BI__nvvm_atom_xor_gen_l:
11489 case NVPTX::BI__nvvm_atom_xor_gen_ll:
11490 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xor, E);
11491
11492 case NVPTX::BI__nvvm_atom_xchg_gen_i:
11493 case NVPTX::BI__nvvm_atom_xchg_gen_l:
11494 case NVPTX::BI__nvvm_atom_xchg_gen_ll:
11495 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xchg, E);
11496
11497 case NVPTX::BI__nvvm_atom_max_gen_i:
11498 case NVPTX::BI__nvvm_atom_max_gen_l:
11499 case NVPTX::BI__nvvm_atom_max_gen_ll:
11500 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Max, E);
11501
11502 case NVPTX::BI__nvvm_atom_max_gen_ui:
11503 case NVPTX::BI__nvvm_atom_max_gen_ul:
11504 case NVPTX::BI__nvvm_atom_max_gen_ull:
11505 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMax, E);
11506
11507 case NVPTX::BI__nvvm_atom_min_gen_i:
11508 case NVPTX::BI__nvvm_atom_min_gen_l:
11509 case NVPTX::BI__nvvm_atom_min_gen_ll:
11510 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Min, E);
11511
11512 case NVPTX::BI__nvvm_atom_min_gen_ui:
11513 case NVPTX::BI__nvvm_atom_min_gen_ul:
11514 case NVPTX::BI__nvvm_atom_min_gen_ull:
11515 return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMin, E);
11516
11517 case NVPTX::BI__nvvm_atom_cas_gen_i:
11518 case NVPTX::BI__nvvm_atom_cas_gen_l:
11519 case NVPTX::BI__nvvm_atom_cas_gen_ll:
11520 // __nvvm_atom_cas_gen_* should return the old value rather than the
11521 // success flag.
11522 return MakeAtomicCmpXchgValue(*this, E, /*ReturnBool=*/false);
11523
11524 case NVPTX::BI__nvvm_atom_add_gen_f: {
11525 Value *Ptr = EmitScalarExpr(E->getArg(0));
11526 Value *Val = EmitScalarExpr(E->getArg(1));
11527 // atomicrmw only deals with integer arguments so we need to use
11528 // LLVM's nvvm_atomic_load_add_f32 intrinsic for that.
11529 Value *FnALAF32 =
11530 CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_add_f32, Ptr->getType());
11531 return Builder.CreateCall(FnALAF32, {Ptr, Val});
11532 }
11533
11534 case NVPTX::BI__nvvm_atom_add_gen_d: {
11535 Value *Ptr = EmitScalarExpr(E->getArg(0));
11536 Value *Val = EmitScalarExpr(E->getArg(1));
11537 // atomicrmw only deals with integer arguments, so we need to use
11538 // LLVM's nvvm_atomic_load_add_f64 intrinsic.
11539 Value *FnALAF64 =
11540 CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_add_f64, Ptr->getType());
11541 return Builder.CreateCall(FnALAF64, {Ptr, Val});
11542 }
11543
11544 case NVPTX::BI__nvvm_atom_inc_gen_ui: {
11545 Value *Ptr = EmitScalarExpr(E->getArg(0));
11546 Value *Val = EmitScalarExpr(E->getArg(1));
11547 Value *FnALI32 =
11548 CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_inc_32, Ptr->getType());
11549 return Builder.CreateCall(FnALI32, {Ptr, Val});
11550 }
11551
11552 case NVPTX::BI__nvvm_atom_dec_gen_ui: {
11553 Value *Ptr = EmitScalarExpr(E->getArg(0));
11554 Value *Val = EmitScalarExpr(E->getArg(1));
11555 Value *FnALD32 =
11556 CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_dec_32, Ptr->getType());
11557 return Builder.CreateCall(FnALD32, {Ptr, Val});
11558 }
11559
11560 case NVPTX::BI__nvvm_ldg_c:
11561 case NVPTX::BI__nvvm_ldg_c2:
11562 case NVPTX::BI__nvvm_ldg_c4:
11563 case NVPTX::BI__nvvm_ldg_s:
11564 case NVPTX::BI__nvvm_ldg_s2:
11565 case NVPTX::BI__nvvm_ldg_s4:
11566 case NVPTX::BI__nvvm_ldg_i:
11567 case NVPTX::BI__nvvm_ldg_i2:
11568 case NVPTX::BI__nvvm_ldg_i4:
11569 case NVPTX::BI__nvvm_ldg_l:
11570 case NVPTX::BI__nvvm_ldg_ll:
11571 case NVPTX::BI__nvvm_ldg_ll2:
11572 case NVPTX::BI__nvvm_ldg_uc:
11573 case NVPTX::BI__nvvm_ldg_uc2:
11574 case NVPTX::BI__nvvm_ldg_uc4:
11575 case NVPTX::BI__nvvm_ldg_us:
11576 case NVPTX::BI__nvvm_ldg_us2:
11577 case NVPTX::BI__nvvm_ldg_us4:
11578 case NVPTX::BI__nvvm_ldg_ui:
11579 case NVPTX::BI__nvvm_ldg_ui2:
11580 case NVPTX::BI__nvvm_ldg_ui4:
11581 case NVPTX::BI__nvvm_ldg_ul:
11582 case NVPTX::BI__nvvm_ldg_ull:
11583 case NVPTX::BI__nvvm_ldg_ull2:
11584 // PTX Interoperability section 2.2: "For a vector with an even number of
11585 // elements, its alignment is set to number of elements times the alignment
11586 // of its member: n*alignof(t)."
11587 return MakeLdg(Intrinsic::nvvm_ldg_global_i);
11588 case NVPTX::BI__nvvm_ldg_f:
11589 case NVPTX::BI__nvvm_ldg_f2:
11590 case NVPTX::BI__nvvm_ldg_f4:
11591 case NVPTX::BI__nvvm_ldg_d:
11592 case NVPTX::BI__nvvm_ldg_d2:
11593 return MakeLdg(Intrinsic::nvvm_ldg_global_f);
11594
11595 case NVPTX::BI__nvvm_atom_cta_add_gen_i:
11596 case NVPTX::BI__nvvm_atom_cta_add_gen_l:
11597 case NVPTX::BI__nvvm_atom_cta_add_gen_ll:
11598 return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_cta);
11599 case NVPTX::BI__nvvm_atom_sys_add_gen_i:
11600 case NVPTX::BI__nvvm_atom_sys_add_gen_l:
11601 case NVPTX::BI__nvvm_atom_sys_add_gen_ll:
11602 return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_sys);
11603 case NVPTX::BI__nvvm_atom_cta_add_gen_f:
11604 case NVPTX::BI__nvvm_atom_cta_add_gen_d:
11605 return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_cta);
11606 case NVPTX::BI__nvvm_atom_sys_add_gen_f:
11607 case NVPTX::BI__nvvm_atom_sys_add_gen_d:
11608 return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_sys);
11609 case NVPTX::BI__nvvm_atom_cta_xchg_gen_i:
11610 case NVPTX::BI__nvvm_atom_cta_xchg_gen_l:
11611 case NVPTX::BI__nvvm_atom_cta_xchg_gen_ll:
11612 return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_cta);
11613 case NVPTX::BI__nvvm_atom_sys_xchg_gen_i:
11614 case NVPTX::BI__nvvm_atom_sys_xchg_gen_l:
11615 case NVPTX::BI__nvvm_atom_sys_xchg_gen_ll:
11616 return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_sys);
11617 case NVPTX::BI__nvvm_atom_cta_max_gen_i:
11618 case NVPTX::BI__nvvm_atom_cta_max_gen_ui:
11619 case NVPTX::BI__nvvm_atom_cta_max_gen_l:
11620 case NVPTX::BI__nvvm_atom_cta_max_gen_ul:
11621 case NVPTX::BI__nvvm_atom_cta_max_gen_ll:
11622 case NVPTX::BI__nvvm_atom_cta_max_gen_ull:
11623 return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_cta);
11624 case NVPTX::BI__nvvm_atom_sys_max_gen_i:
11625 case NVPTX::BI__nvvm_atom_sys_max_gen_ui:
11626 case NVPTX::BI__nvvm_atom_sys_max_gen_l:
11627 case NVPTX::BI__nvvm_atom_sys_max_gen_ul:
11628 case NVPTX::BI__nvvm_atom_sys_max_gen_ll:
11629 case NVPTX::BI__nvvm_atom_sys_max_gen_ull:
11630 return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_sys);
11631 case NVPTX::BI__nvvm_atom_cta_min_gen_i:
11632 case NVPTX::BI__nvvm_atom_cta_min_gen_ui:
11633 case NVPTX::BI__nvvm_atom_cta_min_gen_l:
11634 case NVPTX::BI__nvvm_atom_cta_min_gen_ul:
11635 case NVPTX::BI__nvvm_atom_cta_min_gen_ll:
11636 case NVPTX::BI__nvvm_atom_cta_min_gen_ull:
11637 return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_cta);
11638 case NVPTX::BI__nvvm_atom_sys_min_gen_i:
11639 case NVPTX::BI__nvvm_atom_sys_min_gen_ui:
11640 case NVPTX::BI__nvvm_atom_sys_min_gen_l:
11641 case NVPTX::BI__nvvm_atom_sys_min_gen_ul:
11642 case NVPTX::BI__nvvm_atom_sys_min_gen_ll:
11643 case NVPTX::BI__nvvm_atom_sys_min_gen_ull:
11644 return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_sys);
11645 case NVPTX::BI__nvvm_atom_cta_inc_gen_ui:
11646 return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_cta);
11647 case NVPTX::BI__nvvm_atom_cta_dec_gen_ui:
11648 return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_cta);
11649 case NVPTX::BI__nvvm_atom_sys_inc_gen_ui:
11650 return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_sys);
11651 case NVPTX::BI__nvvm_atom_sys_dec_gen_ui:
11652 return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_sys);
11653 case NVPTX::BI__nvvm_atom_cta_and_gen_i:
11654 case NVPTX::BI__nvvm_atom_cta_and_gen_l:
11655 case NVPTX::BI__nvvm_atom_cta_and_gen_ll:
11656 return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_cta);
11657 case NVPTX::BI__nvvm_atom_sys_and_gen_i:
11658 case NVPTX::BI__nvvm_atom_sys_and_gen_l:
11659 case NVPTX::BI__nvvm_atom_sys_and_gen_ll:
11660 return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_sys);
11661 case NVPTX::BI__nvvm_atom_cta_or_gen_i:
11662 case NVPTX::BI__nvvm_atom_cta_or_gen_l:
11663 case NVPTX::BI__nvvm_atom_cta_or_gen_ll:
11664 return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_cta);
11665 case NVPTX::BI__nvvm_atom_sys_or_gen_i:
11666 case NVPTX::BI__nvvm_atom_sys_or_gen_l:
11667 case NVPTX::BI__nvvm_atom_sys_or_gen_ll:
11668 return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_sys);
11669 case NVPTX::BI__nvvm_atom_cta_xor_gen_i:
11670 case NVPTX::BI__nvvm_atom_cta_xor_gen_l:
11671 case NVPTX::BI__nvvm_atom_cta_xor_gen_ll:
11672 return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_cta);
11673 case NVPTX::BI__nvvm_atom_sys_xor_gen_i:
11674 case NVPTX::BI__nvvm_atom_sys_xor_gen_l:
11675 case NVPTX::BI__nvvm_atom_sys_xor_gen_ll:
11676 return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_sys);
11677 case NVPTX::BI__nvvm_atom_cta_cas_gen_i:
11678 case NVPTX::BI__nvvm_atom_cta_cas_gen_l:
11679 case NVPTX::BI__nvvm_atom_cta_cas_gen_ll: {
11680 Value *Ptr = EmitScalarExpr(E->getArg(0));
11681 return Builder.CreateCall(
11682 CGM.getIntrinsic(
11683 Intrinsic::nvvm_atomic_cas_gen_i_cta,
11684 {Ptr->getType()->getPointerElementType(), Ptr->getType()}),
11685 {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))});
11686 }
11687 case NVPTX::BI__nvvm_atom_sys_cas_gen_i:
11688 case NVPTX::BI__nvvm_atom_sys_cas_gen_l:
11689 case NVPTX::BI__nvvm_atom_sys_cas_gen_ll: {
11690 Value *Ptr = EmitScalarExpr(E->getArg(0));
11691 return Builder.CreateCall(
11692 CGM.getIntrinsic(
11693 Intrinsic::nvvm_atomic_cas_gen_i_sys,
11694 {Ptr->getType()->getPointerElementType(), Ptr->getType()}),
11695 {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))});
11696 }
11697 case NVPTX::BI__nvvm_match_all_sync_i32p:
11698 case NVPTX::BI__nvvm_match_all_sync_i64p: {
11699 Value *Mask = EmitScalarExpr(E->getArg(0));
11700 Value *Val = EmitScalarExpr(E->getArg(1));
11701 Address PredOutPtr = EmitPointerWithAlignment(E->getArg(2));
11702 Value *ResultPair = Builder.CreateCall(
11703 CGM.getIntrinsic(BuiltinID == NVPTX::BI__nvvm_match_all_sync_i32p
11704 ? Intrinsic::nvvm_match_all_sync_i32p
11705 : Intrinsic::nvvm_match_all_sync_i64p),
11706 {Mask, Val});
11707 Value *Pred = Builder.CreateZExt(Builder.CreateExtractValue(ResultPair, 1),
11708 PredOutPtr.getElementType());
11709 Builder.CreateStore(Pred, PredOutPtr);
11710 return Builder.CreateExtractValue(ResultPair, 0);
11711 }
11712 case NVPTX::BI__hmma_m16n16k16_ld_a:
11713 case NVPTX::BI__hmma_m16n16k16_ld_b:
11714 case NVPTX::BI__hmma_m16n16k16_ld_c_f16:
11715 case NVPTX::BI__hmma_m16n16k16_ld_c_f32:
11716 case NVPTX::BI__hmma_m32n8k16_ld_a:
11717 case NVPTX::BI__hmma_m32n8k16_ld_b:
11718 case NVPTX::BI__hmma_m32n8k16_ld_c_f16:
11719 case NVPTX::BI__hmma_m32n8k16_ld_c_f32:
11720 case NVPTX::BI__hmma_m8n32k16_ld_a:
11721 case NVPTX::BI__hmma_m8n32k16_ld_b:
11722 case NVPTX::BI__hmma_m8n32k16_ld_c_f16:
11723 case NVPTX::BI__hmma_m8n32k16_ld_c_f32: {
11724 Address Dst = EmitPointerWithAlignment(E->getArg(0));
11725 Value *Src = EmitScalarExpr(E->getArg(1));
11726 Value *Ldm = EmitScalarExpr(E->getArg(2));
11727 llvm::APSInt isColMajorArg;
11728 if (!E->getArg(3)->isIntegerConstantExpr(isColMajorArg, getContext()))
11729 return nullptr;
11730 bool isColMajor = isColMajorArg.getSExtValue();
11731 unsigned IID;
11732 unsigned NumResults;
11733 switch (BuiltinID) {
11734 case NVPTX::BI__hmma_m16n16k16_ld_a:
11735 IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_load_a_f16_col_stride
11736 : Intrinsic::nvvm_wmma_m16n16k16_load_a_f16_row_stride;
11737 NumResults = 8;
11738 break;
11739 case NVPTX::BI__hmma_m16n16k16_ld_b:
11740 IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_load_b_f16_col_stride
11741 : Intrinsic::nvvm_wmma_m16n16k16_load_b_f16_row_stride;
11742 NumResults = 8;
11743 break;
11744 case NVPTX::BI__hmma_m16n16k16_ld_c_f16:
11745 IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_load_c_f16_col_stride
11746 : Intrinsic::nvvm_wmma_m16n16k16_load_c_f16_row_stride;
11747 NumResults = 4;
11748 break;
11749 case NVPTX::BI__hmma_m16n16k16_ld_c_f32:
11750 IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_load_c_f32_col_stride
11751 : Intrinsic::nvvm_wmma_m16n16k16_load_c_f32_row_stride;
11752 NumResults = 8;
11753 break;
11754 case NVPTX::BI__hmma_m32n8k16_ld_a:
11755 IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_load_a_f16_col_stride
11756 : Intrinsic::nvvm_wmma_m32n8k16_load_a_f16_row_stride;
11757 NumResults = 8;
11758 break;
11759 case NVPTX::BI__hmma_m32n8k16_ld_b:
11760 IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_load_b_f16_col_stride
11761 : Intrinsic::nvvm_wmma_m32n8k16_load_b_f16_row_stride;
11762 NumResults = 8;
11763 break;
11764 case NVPTX::BI__hmma_m32n8k16_ld_c_f16:
11765 IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_load_c_f16_col_stride
11766 : Intrinsic::nvvm_wmma_m32n8k16_load_c_f16_row_stride;
11767 NumResults = 4;
11768 break;
11769 case NVPTX::BI__hmma_m32n8k16_ld_c_f32:
11770 IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_load_c_f32_col_stride
11771 : Intrinsic::nvvm_wmma_m32n8k16_load_c_f32_row_stride;
11772 NumResults = 8;
11773 break;
11774 case NVPTX::BI__hmma_m8n32k16_ld_a:
11775 IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_load_a_f16_col_stride
11776 : Intrinsic::nvvm_wmma_m8n32k16_load_a_f16_row_stride;
11777 NumResults = 8;
11778 break;
11779 case NVPTX::BI__hmma_m8n32k16_ld_b:
11780 IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_load_b_f16_col_stride
11781 : Intrinsic::nvvm_wmma_m8n32k16_load_b_f16_row_stride;
11782 NumResults = 8;
11783 break;
11784 case NVPTX::BI__hmma_m8n32k16_ld_c_f16:
11785 IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_load_c_f16_col_stride
11786 : Intrinsic::nvvm_wmma_m8n32k16_load_c_f16_row_stride;
11787 NumResults = 4;
11788 break;
11789 case NVPTX::BI__hmma_m8n32k16_ld_c_f32:
11790 IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_load_c_f32_col_stride
11791 : Intrinsic::nvvm_wmma_m8n32k16_load_c_f32_row_stride;
11792 NumResults = 8;
11793 break;
11794 default:
11795 llvm_unreachable("Unexpected builtin ID.")::llvm::llvm_unreachable_internal("Unexpected builtin ID.", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11795)
;
11796 }
11797 Value *Result =
11798 Builder.CreateCall(CGM.getIntrinsic(IID, Src->getType()), {Src, Ldm});
11799
11800 // Save returned values.
11801 for (unsigned i = 0; i < NumResults; ++i) {
11802 Builder.CreateAlignedStore(
11803 Builder.CreateBitCast(Builder.CreateExtractValue(Result, i),
11804 Dst.getElementType()),
11805 Builder.CreateGEP(Dst.getPointer(), llvm::ConstantInt::get(IntTy, i)),
11806 CharUnits::fromQuantity(4));
11807 }
11808 return Result;
11809 }
11810
11811 case NVPTX::BI__hmma_m16n16k16_st_c_f16:
11812 case NVPTX::BI__hmma_m16n16k16_st_c_f32:
11813 case NVPTX::BI__hmma_m32n8k16_st_c_f16:
11814 case NVPTX::BI__hmma_m32n8k16_st_c_f32:
11815 case NVPTX::BI__hmma_m8n32k16_st_c_f16:
11816 case NVPTX::BI__hmma_m8n32k16_st_c_f32: {
11817 Value *Dst = EmitScalarExpr(E->getArg(0));
11818 Address Src = EmitPointerWithAlignment(E->getArg(1));
11819 Value *Ldm = EmitScalarExpr(E->getArg(2));
11820 llvm::APSInt isColMajorArg;
11821 if (!E->getArg(3)->isIntegerConstantExpr(isColMajorArg, getContext()))
11822 return nullptr;
11823 bool isColMajor = isColMajorArg.getSExtValue();
11824 unsigned IID;
11825 unsigned NumResults = 8;
11826 // PTX Instructions (and LLVM instrinsics) are defined for slice _d_, yet
11827 // for some reason nvcc builtins use _c_.
11828 switch (BuiltinID) {
11829 case NVPTX::BI__hmma_m16n16k16_st_c_f16:
11830 IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_store_d_f16_col_stride
11831 : Intrinsic::nvvm_wmma_m16n16k16_store_d_f16_row_stride;
11832 NumResults = 4;
11833 break;
11834 case NVPTX::BI__hmma_m16n16k16_st_c_f32:
11835 IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_store_d_f32_col_stride
11836 : Intrinsic::nvvm_wmma_m16n16k16_store_d_f32_row_stride;
11837 break;
11838 case NVPTX::BI__hmma_m32n8k16_st_c_f16:
11839 IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_store_d_f16_col_stride
11840 : Intrinsic::nvvm_wmma_m32n8k16_store_d_f16_row_stride;
11841 NumResults = 4;
11842 break;
11843 case NVPTX::BI__hmma_m32n8k16_st_c_f32:
11844 IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_store_d_f32_col_stride
11845 : Intrinsic::nvvm_wmma_m32n8k16_store_d_f32_row_stride;
11846 break;
11847 case NVPTX::BI__hmma_m8n32k16_st_c_f16:
11848 IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_store_d_f16_col_stride
11849 : Intrinsic::nvvm_wmma_m8n32k16_store_d_f16_row_stride;
11850 NumResults = 4;
11851 break;
11852 case NVPTX::BI__hmma_m8n32k16_st_c_f32:
11853 IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_store_d_f32_col_stride
11854 : Intrinsic::nvvm_wmma_m8n32k16_store_d_f32_row_stride;
11855 break;
11856 default:
11857 llvm_unreachable("Unexpected builtin ID.")::llvm::llvm_unreachable_internal("Unexpected builtin ID.", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11857)
;
11858 }
11859 Function *Intrinsic = CGM.getIntrinsic(IID, Dst->getType());
11860 llvm::Type *ParamType = Intrinsic->getFunctionType()->getParamType(1);
11861 SmallVector<Value *, 10> Values = {Dst};
11862 for (unsigned i = 0; i < NumResults; ++i) {
11863 Value *V = Builder.CreateAlignedLoad(
11864 Builder.CreateGEP(Src.getPointer(), llvm::ConstantInt::get(IntTy, i)),
11865 CharUnits::fromQuantity(4));
11866 Values.push_back(Builder.CreateBitCast(V, ParamType));
11867 }
11868 Values.push_back(Ldm);
11869 Value *Result = Builder.CreateCall(Intrinsic, Values);
11870 return Result;
11871 }
11872
11873 // BI__hmma_m16n16k16_mma_<Dtype><CType>(d, a, b, c, layout, satf) -->
11874 // Intrinsic::nvvm_wmma_m16n16k16_mma_sync<layout A,B><DType><CType><Satf>
11875 case NVPTX::BI__hmma_m16n16k16_mma_f16f16:
11876 case NVPTX::BI__hmma_m16n16k16_mma_f32f16:
11877 case NVPTX::BI__hmma_m16n16k16_mma_f32f32:
11878 case NVPTX::BI__hmma_m16n16k16_mma_f16f32:
11879 case NVPTX::BI__hmma_m32n8k16_mma_f16f16:
11880 case NVPTX::BI__hmma_m32n8k16_mma_f32f16:
11881 case NVPTX::BI__hmma_m32n8k16_mma_f32f32:
11882 case NVPTX::BI__hmma_m32n8k16_mma_f16f32:
11883 case NVPTX::BI__hmma_m8n32k16_mma_f16f16:
11884 case NVPTX::BI__hmma_m8n32k16_mma_f32f16:
11885 case NVPTX::BI__hmma_m8n32k16_mma_f32f32:
11886 case NVPTX::BI__hmma_m8n32k16_mma_f16f32: {
11887 Address Dst = EmitPointerWithAlignment(E->getArg(0));
11888 Address SrcA = EmitPointerWithAlignment(E->getArg(1));
11889 Address SrcB = EmitPointerWithAlignment(E->getArg(2));
11890 Address SrcC = EmitPointerWithAlignment(E->getArg(3));
11891 llvm::APSInt LayoutArg;
11892 if (!E->getArg(4)->isIntegerConstantExpr(LayoutArg, getContext()))
11893 return nullptr;
11894 int Layout = LayoutArg.getSExtValue();
11895 if (Layout < 0 || Layout > 3)
11896 return nullptr;
11897 llvm::APSInt SatfArg;
11898 if (!E->getArg(5)->isIntegerConstantExpr(SatfArg, getContext()))
11899 return nullptr;
11900 bool Satf = SatfArg.getSExtValue();
11901
11902 // clang-format off
11903#define MMA_VARIANTS(geom, type) {{ \
11904 Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type, \
11905 Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type##_satfinite, \
11906 Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \
11907 Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \
11908 Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type, \
11909 Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type##_satfinite, \
11910 Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type, \
11911 Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type##_satfinite \
11912 }}
11913 // clang-format on
11914
11915 auto getMMAIntrinsic = [Layout, Satf](std::array<unsigned, 8> Variants) {
11916 unsigned Index = Layout * 2 + Satf;
11917 assert(Index < 8)(static_cast <bool> (Index < 8) ? void (0) : __assert_fail
("Index < 8", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11917, __extension__ __PRETTY_FUNCTION__))
;
11918 return Variants[Index];
11919 };
11920 unsigned IID;
11921 unsigned NumEltsC;
11922 unsigned NumEltsD;
11923 switch (BuiltinID) {
11924 case NVPTX::BI__hmma_m16n16k16_mma_f16f16:
11925 IID = getMMAIntrinsic(MMA_VARIANTS(m16n16k16, f16_f16));
11926 NumEltsC = 4;
11927 NumEltsD = 4;
11928 break;
11929 case NVPTX::BI__hmma_m16n16k16_mma_f32f16:
11930 IID = getMMAIntrinsic(MMA_VARIANTS(m16n16k16, f32_f16));
11931 NumEltsC = 4;
11932 NumEltsD = 8;
11933 break;
11934 case NVPTX::BI__hmma_m16n16k16_mma_f16f32:
11935 IID = getMMAIntrinsic(MMA_VARIANTS(m16n16k16, f16_f32));
11936 NumEltsC = 8;
11937 NumEltsD = 4;
11938 break;
11939 case NVPTX::BI__hmma_m16n16k16_mma_f32f32:
11940 IID = getMMAIntrinsic(MMA_VARIANTS(m16n16k16, f32_f32));
11941 NumEltsC = 8;
11942 NumEltsD = 8;
11943 break;
11944 case NVPTX::BI__hmma_m32n8k16_mma_f16f16:
11945 IID = getMMAIntrinsic(MMA_VARIANTS(m32n8k16, f16_f16));
11946 NumEltsC = 4;
11947 NumEltsD = 4;
11948 break;
11949 case NVPTX::BI__hmma_m32n8k16_mma_f32f16:
11950 IID = getMMAIntrinsic(MMA_VARIANTS(m32n8k16, f32_f16));
11951 NumEltsC = 4;
11952 NumEltsD = 8;
11953 break;
11954 case NVPTX::BI__hmma_m32n8k16_mma_f16f32:
11955 IID = getMMAIntrinsic(MMA_VARIANTS(m32n8k16, f16_f32));
11956 NumEltsC = 8;
11957 NumEltsD = 4;
11958 break;
11959 case NVPTX::BI__hmma_m32n8k16_mma_f32f32:
11960 IID = getMMAIntrinsic(MMA_VARIANTS(m32n8k16, f32_f32));
11961 NumEltsC = 8;
11962 NumEltsD = 8;
11963 break;
11964 case NVPTX::BI__hmma_m8n32k16_mma_f16f16:
11965 IID = getMMAIntrinsic(MMA_VARIANTS(m8n32k16, f16_f16));
11966 NumEltsC = 4;
11967 NumEltsD = 4;
11968 break;
11969 case NVPTX::BI__hmma_m8n32k16_mma_f32f16:
11970 IID = getMMAIntrinsic(MMA_VARIANTS(m8n32k16, f32_f16));
11971 NumEltsC = 4;
11972 NumEltsD = 8;
11973 break;
11974 case NVPTX::BI__hmma_m8n32k16_mma_f16f32:
11975 IID = getMMAIntrinsic(MMA_VARIANTS(m8n32k16, f16_f32));
11976 NumEltsC = 8;
11977 NumEltsD = 4;
11978 break;
11979 case NVPTX::BI__hmma_m8n32k16_mma_f32f32:
11980 IID = getMMAIntrinsic(MMA_VARIANTS(m8n32k16, f32_f32));
11981 NumEltsC = 8;
11982 NumEltsD = 8;
11983 break;
11984 default:
11985 llvm_unreachable("Unexpected builtin ID.")::llvm::llvm_unreachable_internal("Unexpected builtin ID.", "/build/llvm-toolchain-snapshot-7~svn338205/tools/clang/lib/CodeGen/CGBuiltin.cpp"
, 11985)
;
11986 }
11987#undef MMA_VARIANTS
11988
11989 SmallVector<Value *, 24> Values;
11990 Function *Intrinsic = CGM.getIntrinsic(IID);
11991 llvm::Type *ABType = Intrinsic->getFunctionType()->getParamType(0);
11992 // Load A
11993 for (unsigned i = 0; i < 8; ++i) {
11994 Value *V = Builder.CreateAlignedLoad(
11995 Builder.CreateGEP(SrcA.getPointer(),
11996 llvm::ConstantInt::get(IntTy, i)),
11997 CharUnits::fromQuantity(4));
11998 Values.push_back(Builder.CreateBitCast(V, ABType));
11999 }
12000 // Load B
12001 for (unsigned i = 0; i < 8; ++i) {
12002 Value *V = Builder.CreateAlignedLoad(
12003 Builder.CreateGEP(SrcB.getPointer(),
12004 llvm::ConstantInt::get(IntTy, i)),
12005 CharUnits::fromQuantity(4));
12006 Values.push_back(Builder.CreateBitCast(V, ABType));
12007 }
12008 // Load C
12009 llvm::Type *CType = Intrinsic->getFunctionType()->getParamType(16);
12010 for (unsigned i = 0; i < NumEltsC; ++i) {
12011 Value *V = Builder.CreateAlignedLoad(
12012 Builder.CreateGEP(SrcC.getPointer(),
12013 llvm::ConstantInt::get(IntTy, i)),
12014 CharUnits::fromQuantity(4));
12015 Values.push_back(Builder.CreateBitCast(V, CType));
12016 }
12017 Value *Result = Builder.CreateCall(Intrinsic, Values);
12018 llvm::Type *DType = Dst.getElementType();
12019 for (unsigned i = 0; i < NumEltsD; ++i)
12020 Builder.CreateAlignedStore(
12021 Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), DType),
12022 Builder.CreateGEP(Dst.getPointer(), llvm::ConstantInt::get(IntTy, i)),
12023 CharUnits::fromQuantity(4));
12024 return Result;
12025 }
12026 default:
12027 return nullptr;
12028 }
12029}
12030
12031Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
12032 const CallExpr *E) {
12033 switch (BuiltinID) {
12034 case WebAssembly::BI__builtin_wasm_memory_size: {
12035 llvm::Type *ResultType = ConvertType(E->getType());
12036 Value *I = EmitScalarExpr(E->getArg(0));
12037 Value *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_size, ResultType);
12038 return Builder.CreateCall(Callee, I);
12039 }
12040 case WebAssembly::BI__builtin_wasm_memory_grow: {
12041 llvm::Type *ResultType = ConvertType(E->getType());
12042 Value *Args[] = {
12043 EmitScalarExpr(E->getArg(0)),
12044 EmitScalarExpr(E->getArg(1))
12045 };
12046 Value *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_grow, ResultType);
12047 return Builder.CreateCall(Callee, Args);
12048 }
12049 case WebAssembly::BI__builtin_wasm_mem_size: {
12050 llvm::Type *ResultType = ConvertType(E->getType());
12051 Value *I = EmitScalarExpr(E->getArg(0));
12052 Value *Callee = CGM.getIntrinsic(Intrinsic::wasm_mem_size, ResultType);
12053 return Builder.CreateCall(Callee, I);
12054 }
12055 case WebAssembly::BI__builtin_wasm_mem_grow: {
12056 llvm::Type *ResultType = ConvertType(E->getType());
12057 Value *Args[] = {
12058 EmitScalarExpr(E->getArg(0)),
12059 EmitScalarExpr(E->getArg(1))
12060 };
12061 Value *Callee = CGM.getIntrinsic(Intrinsic::wasm_mem_grow, ResultType);
12062 return Builder.CreateCall(Callee, Args);
12063 }
12064 case WebAssembly::BI__builtin_wasm_current_memory: {
12065 llvm::Type *ResultType = ConvertType(E->getType());
12066 Value *Callee = CGM.getIntrinsic(Intrinsic::wasm_current_memory, ResultType);
12067 return Builder.CreateCall(Callee);
12068 }
12069 case WebAssembly::BI__builtin_wasm_grow_memory: {
12070 Value *X = EmitScalarExpr(E->getArg(0));
12071 Value *Callee = CGM.getIntrinsic(Intrinsic::wasm_grow_memory, X->getType());
12072 return Builder.CreateCall(Callee, X);
12073 }
12074 case WebAssembly::BI__builtin_wasm_throw: {
12075 Value *Tag = EmitScalarExpr(E->getArg(0));
12076 Value *Obj = EmitScalarExpr(E->getArg(1));
12077 Value *Callee = CGM.getIntrinsic(Intrinsic::wasm_throw);
12078 return Builder.CreateCall(Callee, {Tag, Obj});
12079 }
12080 case WebAssembly::BI__builtin_wasm_rethrow: {
12081 Value *Callee = CGM.getIntrinsic(Intrinsic::wasm_rethrow);
12082 return Builder.CreateCall(Callee);
12083 }
12084
12085 default:
12086 return nullptr;
12087 }
12088}
12089
12090Value *CodeGenFunction::EmitHexagonBuiltinExpr(unsigned BuiltinID,
12091 const CallExpr *E) {
12092 SmallVector<llvm::Value *, 4> Ops;
12093 Intrinsic::ID ID = Intrinsic::not_intrinsic;
12094
12095 auto MakeCircLd = [&](unsigned IntID, bool HasImm) {
12096 // The base pointer is passed by address, so it needs to be loaded.
12097 Address BP = EmitPointerWithAlignment(E->getArg(0));
12098 BP = Address(Builder.CreateBitCast(BP.getPointer(), Int8PtrPtrTy),
12099 BP.getAlignment());
12100 llvm::Value *Base = Builder.CreateLoad(BP);
12101 // Operands are Base, Increment, Modifier, Start.
12102 if (HasImm)
12103 Ops = { Base, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)),
12104 EmitScalarExpr(E->getArg(3)) };
12105 else
12106 Ops = { Base, EmitScalarExpr(E->getArg(1)),
12107 EmitScalarExpr(E->getArg(2)) };
12108
12109 llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops);
12110 llvm::Value *NewBase = Builder.CreateExtractValue(Result, 1);
12111 llvm::Value *LV = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)),
12112 NewBase->getType()->getPointerTo());
12113 Address Dest = EmitPointerWithAlignment(E->getArg(0));
12114 // The intrinsic generates two results. The new value for the base pointer
12115 // needs to be stored.
12116 Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment());
12117 return Builder.CreateExtractValue(Result, 0);
12118 };
12119
12120 auto MakeCircSt = [&](unsigned IntID, bool HasImm) {
12121 // The base pointer is passed by address, so it needs to be loaded.
12122 Address BP = EmitPointerWithAlignment(E->getArg(0));
12123 BP = Address(Builder.CreateBitCast(BP.getPointer(), Int8PtrPtrTy),
12124 BP.getAlignment());
12125 llvm::Value *Base = Builder.CreateLoad(BP);
12126 // Operands are Base, Increment, Modifier, Value, Start.
12127 if (HasImm)
12128 Ops = { Base, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)),
12129 EmitScalarExpr(E->getArg(3)), EmitScalarExpr(E->getArg(4)) };
12130 else
12131 Ops = { Base, EmitScalarExpr(E->getArg(1)),
12132 EmitScalarExpr(E->getArg(2)), EmitScalarExpr(E->getArg(3)) };
12133
12134 llvm::Value *NewBase = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops);
12135 llvm::Value *LV = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)),
12136 NewBase->getType()->getPointerTo());
12137 Address Dest = EmitPointerWithAlignment(E->getArg(0));
12138 // The intrinsic generates one result, which is the new value for the base
12139 // pointer. It needs to be stored.
12140 return Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment());
12141 };
12142
12143 // Handle the conversion of bit-reverse load intrinsics to bit code.
12144 // The intrinsic call after this function only reads from memory and the
12145 // write to memory is dealt by the store instruction.
12146 auto MakeBrevLd = [&](unsigned IntID, llvm::Type *DestTy) {
12147 // The intrinsic generates one result, which is the new value for the base
12148 // pointer. It needs to be returned. The result of the load instruction is
12149 // passed to intrinsic by address, so the value needs to be stored.
12150 llvm::Value *BaseAddress =
12151 Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int8PtrTy);
12152
12153 // Expressions like &(*pt++) will be incremented per evaluation.
12154 // EmitPointerWithAlignment and EmitScalarExpr evaluates the expression
12155 // per call.
12156 Address DestAddr = EmitPointerWithAlignment(E->getArg(1));
12157 DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), Int8PtrTy),
12158 DestAddr.getAlignment());
12159 llvm::Value *DestAddress = DestAddr.getPointer();
12160
12161 // Operands are Base, Dest, Modifier.
12162 // The intrinsic format in LLVM IR is defined as
12163 // { ValueType, i8* } (i8*, i32).
12164 Ops = {BaseAddress, EmitScalarExpr(E->getArg(2))};
12165
12166 llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops);
12167 // The value needs to be stored as the variable is passed by reference.
12168 llvm::Value *DestVal = Builder.CreateExtractValue(Result, 0);
12169
12170 // The store needs to be truncated to fit the destination type.
12171 // While i32 and i64 are natively supported on Hexagon, i8 and i16 needs
12172 // to be handled with stores of respective destination type.
12173 DestVal = Builder.CreateTrunc(DestVal, DestTy);
12174
12175 llvm::Value *DestForStore =
12176 Builder.CreateBitCast(DestAddress, DestVal->getType()->getPointerTo());
12177 Builder.CreateAlignedStore(DestVal, DestForStore, DestAddr.getAlignment());
12178 // The updated value of the base pointer is returned.
12179 return Builder.CreateExtractValue(Result, 1);
12180 };
12181
12182 switch (BuiltinID) {
12183 case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry:
12184 case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry_128B: {
12185 Address Dest = EmitPointerWithAlignment(E->getArg(2));
12186 unsigned Size;
12187 if (BuiltinID == Hexagon::BI__builtin_HEXAGON_V6_vaddcarry) {
12188 Size = 512;
12189 ID = Intrinsic::hexagon_V6_vaddcarry;
12190 } else {
12191 Size = 1024;
12192 ID = Intrinsic::hexagon_V6_vaddcarry_128B;
12193 }
12194 Dest = Builder.CreateBitCast(Dest,
12195 llvm::VectorType::get(Builder.getInt1Ty(), Size)->getPointerTo(0));
12196 LoadInst *QLd = Builder.CreateLoad(Dest);
12197 Ops = { EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), QLd };
12198 llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
12199 llvm::Value *Vprd = Builder.CreateExtractValue(Result, 1);
12200 llvm::Value *Base = Builder.CreateBitCast(EmitScalarExpr(E->getArg(2)),
12201 Vprd->getType()->getPointerTo(0));
12202 Builder.CreateAlignedStore(Vprd, Base, Dest.getAlignment());
12203 return Builder.CreateExtractValue(Result, 0);
12204 }
12205 case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry:
12206 case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry_128B: {
12207 Address Dest = EmitPointerWithAlignment(E->getArg(2));
12208 unsigned Size;
12209 if (BuiltinID == Hexagon::BI__builtin_HEXAGON_V6_vsubcarry) {
12210 Size = 512;
12211 ID = Intrinsic::hexagon_V6_vsubcarry;
12212 } else {
12213 Size = 1024;
12214 ID = Intrinsic::hexagon_V6_vsubcarry_128B;
12215 }
12216 Dest = Builder.CreateBitCast(Dest,
12217 llvm::VectorType::get(Builder.getInt1Ty(), Size)->getPointerTo(0));
12218 LoadInst *QLd = Builder.CreateLoad(Dest);
12219 Ops = { EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), QLd };
12220 llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
12221 llvm::Value *Vprd = Builder.CreateExtractValue(Result, 1);
12222 llvm::Value *Base = Builder.CreateBitCast(EmitScalarExpr(E->getArg(2)),
12223 Vprd->getType()->getPointerTo(0));
12224 Builder.CreateAlignedStore(Vprd, Base, Dest.getAlignment());
12225 return Builder.CreateExtractValue(Result, 0);
12226 }
12227 case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pci:
12228 return MakeCircLd(Intrinsic::hexagon_L2_loadrub_pci, /*HasImm*/true);
12229 case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pci:
12230 return MakeCircLd(Intrinsic::hexagon_L2_loadrb_pci, /*HasImm*/true);
12231 case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pci:
12232 return MakeCircLd(Intrinsic::hexagon_L2_loadruh_pci, /*HasImm*/true);
12233 case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pci:
12234 return MakeCircLd(Intrinsic::hexagon_L2_loadrh_pci, /*HasImm*/true);
12235 case Hexagon::BI__builtin_HEXAGON_L2_loadri_pci:
12236 return MakeCircLd(Intrinsic::hexagon_L2_loadri_pci, /*HasImm*/true);
12237 case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pci:
12238 return MakeCircLd(Intrinsic::hexagon_L2_loadrd_pci, /*HasImm*/true);
12239 case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pcr:
12240 return MakeCircLd(Intrinsic::hexagon_L2_loadrub_pcr, /*HasImm*/false);
12241 case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pcr:
12242 return MakeCircLd(Intrinsic::hexagon_L2_loadrb_pcr, /*HasImm*/false);
12243 case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pcr:
12244 return MakeCircLd(Intrinsic::hexagon_L2_loadruh_pcr, /*HasImm*/false);
12245 case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pcr:
12246 return MakeCircLd(Intrinsic::hexagon_L2_loadrh_pcr, /*HasImm*/false);
12247 case Hexagon::BI__builtin_HEXAGON_L2_loadri_pcr:
12248 return MakeCircLd(Intrinsic::hexagon_L2_loadri_pcr, /*HasImm*/false);
12249 case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pcr:
12250 return MakeCircLd(Intrinsic::hexagon_L2_loadrd_pcr, /*HasImm*/false);
12251 case Hexagon::BI__builtin_HEXAGON_S2_storerb_pci:
12252 return MakeCircSt(Intrinsic::hexagon_S2_storerb_pci, /*HasImm*/true);
12253 case Hexagon::BI__builtin_HEXAGON_S2_storerh_pci:
12254 return MakeCircSt(Intrinsic::hexagon_S2_storerh_pci, /*HasImm*/true);
12255 case Hexagon::BI__builtin_HEXAGON_S2_storerf_pci:
12256 return MakeCircSt(Intrinsic::hexagon_S2_storerf_pci, /*HasImm*/true);
12257 case Hexagon::BI__builtin_HEXAGON_S2_storeri_pci:
12258 return MakeCircSt(Intrinsic::hexagon_S2_storeri_pci, /*HasImm*/true);
12259 case Hexagon::BI__builtin_HEXAGON_S2_storerd_pci:
12260 return MakeCircSt(Intrinsic::hexagon_S2_storerd_pci, /*HasImm*/true);
12261 case Hexagon::BI__builtin_HEXAGON_S2_storerb_pcr:
12262 return MakeCircSt(Intrinsic::hexagon_S2_storerb_pcr, /*HasImm*/false);
12263 case Hexagon::BI__builtin_HEXAGON_S2_storerh_pcr:
12264 return MakeCircSt(Intrinsic::hexagon_S2_storerh_pcr, /*HasImm*/false);
12265 case Hexagon::BI__builtin_HEXAGON_S2_storerf_pcr:
12266 return MakeCircSt(Intrinsic::hexagon_S2_storerf_pcr, /*HasImm*/false);
12267 case Hexagon::BI__builtin_HEXAGON_S2_storeri_pcr:
12268 return MakeCircSt(Intrinsic::hexagon_S2_storeri_pcr, /*HasImm*/false);
12269 case Hexagon::BI__builtin_HEXAGON_S2_storerd_pcr:
12270 return MakeCircSt(Intrinsic::hexagon_S2_storerd_pcr, /*HasImm*/false);
12271 case Hexagon::BI__builtin_brev_ldub:
12272 return MakeBrevLd(Intrinsic::hexagon_L2_loadrub_pbr, Int8Ty);
12273 case Hexagon::BI__builtin_brev_ldb:
12274 return MakeBrevLd(Intrinsic::hexagon_L2_loadrb_pbr, Int8Ty);
12275 case Hexagon::BI__builtin_brev_lduh:
12276 return MakeBrevLd(Intrinsic::hexagon_L2_loadruh_pbr, Int16Ty);
12277 case Hexagon::BI__builtin_brev_ldh:
12278 return MakeBrevLd(Intrinsic::hexagon_L2_loadrh_pbr, Int16Ty);
12279 case Hexagon::BI__builtin_brev_ldw:
12280 return MakeBrevLd(Intrinsic::hexagon_L2_loadri_pbr, Int32Ty);
12281 case Hexagon::BI__builtin_brev_ldd:
12282 return MakeBrevLd(Intrinsic::hexagon_L2_loadrd_pbr, Int64Ty);
12283 default:
12284 break;
12285 } // switch
12286
12287 return nullptr;
12288}