File: | clang/lib/CodeGen/CGBuiltin.cpp |
Warning: | line 15459, column 9 1st function call argument is an uninitialized value |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | //===---- CGBuiltin.cpp - Emit LLVM Code for builtins ---------------------===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | // |
9 | // This contains code to emit Builtin calls as LLVM code. |
10 | // |
11 | //===----------------------------------------------------------------------===// |
12 | |
13 | #include "CGCXXABI.h" |
14 | #include "CGObjCRuntime.h" |
15 | #include "CGOpenCLRuntime.h" |
16 | #include "CGRecordLayout.h" |
17 | #include "CodeGenFunction.h" |
18 | #include "CodeGenModule.h" |
19 | #include "ConstantEmitter.h" |
20 | #include "PatternInit.h" |
21 | #include "TargetInfo.h" |
22 | #include "clang/AST/ASTContext.h" |
23 | #include "clang/AST/Attr.h" |
24 | #include "clang/AST/Decl.h" |
25 | #include "clang/AST/OSLog.h" |
26 | #include "clang/Basic/TargetBuiltins.h" |
27 | #include "clang/Basic/TargetInfo.h" |
28 | #include "clang/CodeGen/CGFunctionInfo.h" |
29 | #include "llvm/ADT/SmallPtrSet.h" |
30 | #include "llvm/ADT/StringExtras.h" |
31 | #include "llvm/Analysis/ValueTracking.h" |
32 | #include "llvm/IR/DataLayout.h" |
33 | #include "llvm/IR/InlineAsm.h" |
34 | #include "llvm/IR/Intrinsics.h" |
35 | #include "llvm/IR/IntrinsicsAArch64.h" |
36 | #include "llvm/IR/IntrinsicsAMDGPU.h" |
37 | #include "llvm/IR/IntrinsicsARM.h" |
38 | #include "llvm/IR/IntrinsicsBPF.h" |
39 | #include "llvm/IR/IntrinsicsHexagon.h" |
40 | #include "llvm/IR/IntrinsicsNVPTX.h" |
41 | #include "llvm/IR/IntrinsicsPowerPC.h" |
42 | #include "llvm/IR/IntrinsicsR600.h" |
43 | #include "llvm/IR/IntrinsicsS390.h" |
44 | #include "llvm/IR/IntrinsicsWebAssembly.h" |
45 | #include "llvm/IR/IntrinsicsX86.h" |
46 | #include "llvm/IR/MDBuilder.h" |
47 | #include "llvm/IR/MatrixBuilder.h" |
48 | #include "llvm/Support/ConvertUTF.h" |
49 | #include "llvm/Support/ScopedPrinter.h" |
50 | #include "llvm/Support/X86TargetParser.h" |
51 | #include <sstream> |
52 | |
53 | using namespace clang; |
54 | using namespace CodeGen; |
55 | using namespace llvm; |
56 | |
57 | static |
58 | int64_t clamp(int64_t Value, int64_t Low, int64_t High) { |
59 | return std::min(High, std::max(Low, Value)); |
60 | } |
61 | |
62 | static void initializeAlloca(CodeGenFunction &CGF, AllocaInst *AI, Value *Size, |
63 | Align AlignmentInBytes) { |
64 | ConstantInt *Byte; |
65 | switch (CGF.getLangOpts().getTrivialAutoVarInit()) { |
66 | case LangOptions::TrivialAutoVarInitKind::Uninitialized: |
67 | // Nothing to initialize. |
68 | return; |
69 | case LangOptions::TrivialAutoVarInitKind::Zero: |
70 | Byte = CGF.Builder.getInt8(0x00); |
71 | break; |
72 | case LangOptions::TrivialAutoVarInitKind::Pattern: { |
73 | llvm::Type *Int8 = llvm::IntegerType::getInt8Ty(CGF.CGM.getLLVMContext()); |
74 | Byte = llvm::dyn_cast<llvm::ConstantInt>( |
75 | initializationPatternFor(CGF.CGM, Int8)); |
76 | break; |
77 | } |
78 | } |
79 | if (CGF.CGM.stopAutoInit()) |
80 | return; |
81 | auto *I = CGF.Builder.CreateMemSet(AI, Byte, Size, AlignmentInBytes); |
82 | I->addAnnotationMetadata("auto-init"); |
83 | } |
84 | |
85 | /// getBuiltinLibFunction - Given a builtin id for a function like |
86 | /// "__builtin_fabsf", return a Function* for "fabsf". |
87 | llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD, |
88 | unsigned BuiltinID) { |
89 | assert(Context.BuiltinInfo.isLibFunction(BuiltinID))((Context.BuiltinInfo.isLibFunction(BuiltinID)) ? static_cast <void> (0) : __assert_fail ("Context.BuiltinInfo.isLibFunction(BuiltinID)" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 89, __PRETTY_FUNCTION__)); |
90 | |
91 | // Get the name, skip over the __builtin_ prefix (if necessary). |
92 | StringRef Name; |
93 | GlobalDecl D(FD); |
94 | |
95 | // If the builtin has been declared explicitly with an assembler label, |
96 | // use the mangled name. This differs from the plain label on platforms |
97 | // that prefix labels. |
98 | if (FD->hasAttr<AsmLabelAttr>()) |
99 | Name = getMangledName(D); |
100 | else |
101 | Name = Context.BuiltinInfo.getName(BuiltinID) + 10; |
102 | |
103 | llvm::FunctionType *Ty = |
104 | cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType())); |
105 | |
106 | return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false); |
107 | } |
108 | |
109 | /// Emit the conversions required to turn the given value into an |
110 | /// integer of the given size. |
111 | static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V, |
112 | QualType T, llvm::IntegerType *IntType) { |
113 | V = CGF.EmitToMemory(V, T); |
114 | |
115 | if (V->getType()->isPointerTy()) |
116 | return CGF.Builder.CreatePtrToInt(V, IntType); |
117 | |
118 | assert(V->getType() == IntType)((V->getType() == IntType) ? static_cast<void> (0) : __assert_fail ("V->getType() == IntType", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 118, __PRETTY_FUNCTION__)); |
119 | return V; |
120 | } |
121 | |
122 | static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V, |
123 | QualType T, llvm::Type *ResultType) { |
124 | V = CGF.EmitFromMemory(V, T); |
125 | |
126 | if (ResultType->isPointerTy()) |
127 | return CGF.Builder.CreateIntToPtr(V, ResultType); |
128 | |
129 | assert(V->getType() == ResultType)((V->getType() == ResultType) ? static_cast<void> (0 ) : __assert_fail ("V->getType() == ResultType", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 129, __PRETTY_FUNCTION__)); |
130 | return V; |
131 | } |
132 | |
133 | /// Utility to insert an atomic instruction based on Intrinsic::ID |
134 | /// and the expression node. |
135 | static Value *MakeBinaryAtomicValue( |
136 | CodeGenFunction &CGF, llvm::AtomicRMWInst::BinOp Kind, const CallExpr *E, |
137 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
138 | QualType T = E->getType(); |
139 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 139, __PRETTY_FUNCTION__)); |
140 | assert(CGF.getContext().hasSameUnqualifiedType(T,((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)-> getType()->getPointeeType())) ? static_cast<void> (0 ) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 141, __PRETTY_FUNCTION__)) |
141 | E->getArg(0)->getType()->getPointeeType()))((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)-> getType()->getPointeeType())) ? static_cast<void> (0 ) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 141, __PRETTY_FUNCTION__)); |
142 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)-> getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 142, __PRETTY_FUNCTION__)); |
143 | |
144 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
145 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
146 | |
147 | llvm::IntegerType *IntType = |
148 | llvm::IntegerType::get(CGF.getLLVMContext(), |
149 | CGF.getContext().getTypeSize(T)); |
150 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
151 | |
152 | llvm::Value *Args[2]; |
153 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
154 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
155 | llvm::Type *ValueType = Args[1]->getType(); |
156 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
157 | |
158 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( |
159 | Kind, Args[0], Args[1], Ordering); |
160 | return EmitFromInt(CGF, Result, T, ValueType); |
161 | } |
162 | |
163 | static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) { |
164 | Value *Val = CGF.EmitScalarExpr(E->getArg(0)); |
165 | Value *Address = CGF.EmitScalarExpr(E->getArg(1)); |
166 | |
167 | // Convert the type of the pointer to a pointer to the stored type. |
168 | Val = CGF.EmitToMemory(Val, E->getArg(0)->getType()); |
169 | Value *BC = CGF.Builder.CreateBitCast( |
170 | Address, llvm::PointerType::getUnqual(Val->getType()), "cast"); |
171 | LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType()); |
172 | LV.setNontemporal(true); |
173 | CGF.EmitStoreOfScalar(Val, LV, false); |
174 | return nullptr; |
175 | } |
176 | |
177 | static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) { |
178 | Value *Address = CGF.EmitScalarExpr(E->getArg(0)); |
179 | |
180 | LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType()); |
181 | LV.setNontemporal(true); |
182 | return CGF.EmitLoadOfScalar(LV, E->getExprLoc()); |
183 | } |
184 | |
185 | static RValue EmitBinaryAtomic(CodeGenFunction &CGF, |
186 | llvm::AtomicRMWInst::BinOp Kind, |
187 | const CallExpr *E) { |
188 | return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E)); |
189 | } |
190 | |
191 | /// Utility to insert an atomic instruction based Intrinsic::ID and |
192 | /// the expression node, where the return value is the result of the |
193 | /// operation. |
194 | static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF, |
195 | llvm::AtomicRMWInst::BinOp Kind, |
196 | const CallExpr *E, |
197 | Instruction::BinaryOps Op, |
198 | bool Invert = false) { |
199 | QualType T = E->getType(); |
200 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 200, __PRETTY_FUNCTION__)); |
201 | assert(CGF.getContext().hasSameUnqualifiedType(T,((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)-> getType()->getPointeeType())) ? static_cast<void> (0 ) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 202, __PRETTY_FUNCTION__)) |
202 | E->getArg(0)->getType()->getPointeeType()))((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)-> getType()->getPointeeType())) ? static_cast<void> (0 ) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 202, __PRETTY_FUNCTION__)); |
203 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)-> getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 203, __PRETTY_FUNCTION__)); |
204 | |
205 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
206 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
207 | |
208 | llvm::IntegerType *IntType = |
209 | llvm::IntegerType::get(CGF.getLLVMContext(), |
210 | CGF.getContext().getTypeSize(T)); |
211 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
212 | |
213 | llvm::Value *Args[2]; |
214 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
215 | llvm::Type *ValueType = Args[1]->getType(); |
216 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
217 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
218 | |
219 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( |
220 | Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent); |
221 | Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]); |
222 | if (Invert) |
223 | Result = |
224 | CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result, |
225 | llvm::ConstantInt::getAllOnesValue(IntType)); |
226 | Result = EmitFromInt(CGF, Result, T, ValueType); |
227 | return RValue::get(Result); |
228 | } |
229 | |
230 | /// Utility to insert an atomic cmpxchg instruction. |
231 | /// |
232 | /// @param CGF The current codegen function. |
233 | /// @param E Builtin call expression to convert to cmpxchg. |
234 | /// arg0 - address to operate on |
235 | /// arg1 - value to compare with |
236 | /// arg2 - new value |
237 | /// @param ReturnBool Specifies whether to return success flag of |
238 | /// cmpxchg result or the old value. |
239 | /// |
240 | /// @returns result of cmpxchg, according to ReturnBool |
241 | /// |
242 | /// Note: In order to lower Microsoft's _InterlockedCompareExchange* intrinsics |
243 | /// invoke the function EmitAtomicCmpXchgForMSIntrin. |
244 | static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E, |
245 | bool ReturnBool) { |
246 | QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType(); |
247 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
248 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
249 | |
250 | llvm::IntegerType *IntType = llvm::IntegerType::get( |
251 | CGF.getLLVMContext(), CGF.getContext().getTypeSize(T)); |
252 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
253 | |
254 | Value *Args[3]; |
255 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
256 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
257 | llvm::Type *ValueType = Args[1]->getType(); |
258 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
259 | Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType); |
260 | |
261 | Value *Pair = CGF.Builder.CreateAtomicCmpXchg( |
262 | Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent, |
263 | llvm::AtomicOrdering::SequentiallyConsistent); |
264 | if (ReturnBool) |
265 | // Extract boolean success flag and zext it to int. |
266 | return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1), |
267 | CGF.ConvertType(E->getType())); |
268 | else |
269 | // Extract old value and emit it using the same type as compare value. |
270 | return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T, |
271 | ValueType); |
272 | } |
273 | |
274 | /// This function should be invoked to emit atomic cmpxchg for Microsoft's |
275 | /// _InterlockedCompareExchange* intrinsics which have the following signature: |
276 | /// T _InterlockedCompareExchange(T volatile *Destination, |
277 | /// T Exchange, |
278 | /// T Comparand); |
279 | /// |
280 | /// Whereas the llvm 'cmpxchg' instruction has the following syntax: |
281 | /// cmpxchg *Destination, Comparand, Exchange. |
282 | /// So we need to swap Comparand and Exchange when invoking |
283 | /// CreateAtomicCmpXchg. That is the reason we could not use the above utility |
284 | /// function MakeAtomicCmpXchgValue since it expects the arguments to be |
285 | /// already swapped. |
286 | |
287 | static |
288 | Value *EmitAtomicCmpXchgForMSIntrin(CodeGenFunction &CGF, const CallExpr *E, |
289 | AtomicOrdering SuccessOrdering = AtomicOrdering::SequentiallyConsistent) { |
290 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 290, __PRETTY_FUNCTION__)); |
291 | assert(CGF.getContext().hasSameUnqualifiedType(((CGF.getContext().hasSameUnqualifiedType( E->getType(), E ->getArg(0)->getType()->getPointeeType())) ? static_cast <void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 292, __PRETTY_FUNCTION__)) |
292 | E->getType(), E->getArg(0)->getType()->getPointeeType()))((CGF.getContext().hasSameUnqualifiedType( E->getType(), E ->getArg(0)->getType()->getPointeeType())) ? static_cast <void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 292, __PRETTY_FUNCTION__)); |
293 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),((CGF.getContext().hasSameUnqualifiedType(E->getType(), E-> getArg(1)->getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 294, __PRETTY_FUNCTION__)) |
294 | E->getArg(1)->getType()))((CGF.getContext().hasSameUnqualifiedType(E->getType(), E-> getArg(1)->getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 294, __PRETTY_FUNCTION__)); |
295 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),((CGF.getContext().hasSameUnqualifiedType(E->getType(), E-> getArg(2)->getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 296, __PRETTY_FUNCTION__)) |
296 | E->getArg(2)->getType()))((CGF.getContext().hasSameUnqualifiedType(E->getType(), E-> getArg(2)->getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 296, __PRETTY_FUNCTION__)); |
297 | |
298 | auto *Destination = CGF.EmitScalarExpr(E->getArg(0)); |
299 | auto *Comparand = CGF.EmitScalarExpr(E->getArg(2)); |
300 | auto *Exchange = CGF.EmitScalarExpr(E->getArg(1)); |
301 | |
302 | // For Release ordering, the failure ordering should be Monotonic. |
303 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release ? |
304 | AtomicOrdering::Monotonic : |
305 | SuccessOrdering; |
306 | |
307 | // The atomic instruction is marked volatile for consistency with MSVC. This |
308 | // blocks the few atomics optimizations that LLVM has. If we want to optimize |
309 | // _Interlocked* operations in the future, we will have to remove the volatile |
310 | // marker. |
311 | auto *Result = CGF.Builder.CreateAtomicCmpXchg( |
312 | Destination, Comparand, Exchange, |
313 | SuccessOrdering, FailureOrdering); |
314 | Result->setVolatile(true); |
315 | return CGF.Builder.CreateExtractValue(Result, 0); |
316 | } |
317 | |
318 | // 64-bit Microsoft platforms support 128 bit cmpxchg operations. They are |
319 | // prototyped like this: |
320 | // |
321 | // unsigned char _InterlockedCompareExchange128...( |
322 | // __int64 volatile * _Destination, |
323 | // __int64 _ExchangeHigh, |
324 | // __int64 _ExchangeLow, |
325 | // __int64 * _ComparandResult); |
326 | static Value *EmitAtomicCmpXchg128ForMSIntrin(CodeGenFunction &CGF, |
327 | const CallExpr *E, |
328 | AtomicOrdering SuccessOrdering) { |
329 | assert(E->getNumArgs() == 4)((E->getNumArgs() == 4) ? static_cast<void> (0) : __assert_fail ("E->getNumArgs() == 4", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 329, __PRETTY_FUNCTION__)); |
330 | llvm::Value *Destination = CGF.EmitScalarExpr(E->getArg(0)); |
331 | llvm::Value *ExchangeHigh = CGF.EmitScalarExpr(E->getArg(1)); |
332 | llvm::Value *ExchangeLow = CGF.EmitScalarExpr(E->getArg(2)); |
333 | llvm::Value *ComparandPtr = CGF.EmitScalarExpr(E->getArg(3)); |
334 | |
335 | assert(Destination->getType()->isPointerTy())((Destination->getType()->isPointerTy()) ? static_cast< void> (0) : __assert_fail ("Destination->getType()->isPointerTy()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 335, __PRETTY_FUNCTION__)); |
336 | assert(!ExchangeHigh->getType()->isPointerTy())((!ExchangeHigh->getType()->isPointerTy()) ? static_cast <void> (0) : __assert_fail ("!ExchangeHigh->getType()->isPointerTy()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 336, __PRETTY_FUNCTION__)); |
337 | assert(!ExchangeLow->getType()->isPointerTy())((!ExchangeLow->getType()->isPointerTy()) ? static_cast <void> (0) : __assert_fail ("!ExchangeLow->getType()->isPointerTy()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 337, __PRETTY_FUNCTION__)); |
338 | assert(ComparandPtr->getType()->isPointerTy())((ComparandPtr->getType()->isPointerTy()) ? static_cast <void> (0) : __assert_fail ("ComparandPtr->getType()->isPointerTy()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 338, __PRETTY_FUNCTION__)); |
339 | |
340 | // For Release ordering, the failure ordering should be Monotonic. |
341 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release |
342 | ? AtomicOrdering::Monotonic |
343 | : SuccessOrdering; |
344 | |
345 | // Convert to i128 pointers and values. |
346 | llvm::Type *Int128Ty = llvm::IntegerType::get(CGF.getLLVMContext(), 128); |
347 | llvm::Type *Int128PtrTy = Int128Ty->getPointerTo(); |
348 | Destination = CGF.Builder.CreateBitCast(Destination, Int128PtrTy); |
349 | Address ComparandResult(CGF.Builder.CreateBitCast(ComparandPtr, Int128PtrTy), |
350 | CGF.getContext().toCharUnitsFromBits(128)); |
351 | |
352 | // (((i128)hi) << 64) | ((i128)lo) |
353 | ExchangeHigh = CGF.Builder.CreateZExt(ExchangeHigh, Int128Ty); |
354 | ExchangeLow = CGF.Builder.CreateZExt(ExchangeLow, Int128Ty); |
355 | ExchangeHigh = |
356 | CGF.Builder.CreateShl(ExchangeHigh, llvm::ConstantInt::get(Int128Ty, 64)); |
357 | llvm::Value *Exchange = CGF.Builder.CreateOr(ExchangeHigh, ExchangeLow); |
358 | |
359 | // Load the comparand for the instruction. |
360 | llvm::Value *Comparand = CGF.Builder.CreateLoad(ComparandResult); |
361 | |
362 | auto *CXI = CGF.Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, |
363 | SuccessOrdering, FailureOrdering); |
364 | |
365 | // The atomic instruction is marked volatile for consistency with MSVC. This |
366 | // blocks the few atomics optimizations that LLVM has. If we want to optimize |
367 | // _Interlocked* operations in the future, we will have to remove the volatile |
368 | // marker. |
369 | CXI->setVolatile(true); |
370 | |
371 | // Store the result as an outparameter. |
372 | CGF.Builder.CreateStore(CGF.Builder.CreateExtractValue(CXI, 0), |
373 | ComparandResult); |
374 | |
375 | // Get the success boolean and zero extend it to i8. |
376 | Value *Success = CGF.Builder.CreateExtractValue(CXI, 1); |
377 | return CGF.Builder.CreateZExt(Success, CGF.Int8Ty); |
378 | } |
379 | |
380 | static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E, |
381 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
382 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 382, __PRETTY_FUNCTION__)); |
383 | |
384 | auto *IntTy = CGF.ConvertType(E->getType()); |
385 | auto *Result = CGF.Builder.CreateAtomicRMW( |
386 | AtomicRMWInst::Add, |
387 | CGF.EmitScalarExpr(E->getArg(0)), |
388 | ConstantInt::get(IntTy, 1), |
389 | Ordering); |
390 | return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1)); |
391 | } |
392 | |
393 | static Value *EmitAtomicDecrementValue(CodeGenFunction &CGF, const CallExpr *E, |
394 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
395 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 395, __PRETTY_FUNCTION__)); |
396 | |
397 | auto *IntTy = CGF.ConvertType(E->getType()); |
398 | auto *Result = CGF.Builder.CreateAtomicRMW( |
399 | AtomicRMWInst::Sub, |
400 | CGF.EmitScalarExpr(E->getArg(0)), |
401 | ConstantInt::get(IntTy, 1), |
402 | Ordering); |
403 | return CGF.Builder.CreateSub(Result, ConstantInt::get(IntTy, 1)); |
404 | } |
405 | |
406 | // Build a plain volatile load. |
407 | static Value *EmitISOVolatileLoad(CodeGenFunction &CGF, const CallExpr *E) { |
408 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); |
409 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
410 | CharUnits LoadSize = CGF.getContext().getTypeSizeInChars(ElTy); |
411 | llvm::Type *ITy = |
412 | llvm::IntegerType::get(CGF.getLLVMContext(), LoadSize.getQuantity() * 8); |
413 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
414 | llvm::LoadInst *Load = CGF.Builder.CreateAlignedLoad(Ptr, LoadSize); |
415 | Load->setVolatile(true); |
416 | return Load; |
417 | } |
418 | |
419 | // Build a plain volatile store. |
420 | static Value *EmitISOVolatileStore(CodeGenFunction &CGF, const CallExpr *E) { |
421 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); |
422 | Value *Value = CGF.EmitScalarExpr(E->getArg(1)); |
423 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
424 | CharUnits StoreSize = CGF.getContext().getTypeSizeInChars(ElTy); |
425 | llvm::Type *ITy = |
426 | llvm::IntegerType::get(CGF.getLLVMContext(), StoreSize.getQuantity() * 8); |
427 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
428 | llvm::StoreInst *Store = |
429 | CGF.Builder.CreateAlignedStore(Value, Ptr, StoreSize); |
430 | Store->setVolatile(true); |
431 | return Store; |
432 | } |
433 | |
434 | // Emit a simple mangled intrinsic that has 1 argument and a return type |
435 | // matching the argument type. Depending on mode, this may be a constrained |
436 | // floating-point intrinsic. |
437 | static Value *emitUnaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
438 | const CallExpr *E, unsigned IntrinsicID, |
439 | unsigned ConstrainedIntrinsicID) { |
440 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
441 | |
442 | if (CGF.Builder.getIsFPConstrained()) { |
443 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
444 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
445 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0 }); |
446 | } else { |
447 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
448 | return CGF.Builder.CreateCall(F, Src0); |
449 | } |
450 | } |
451 | |
452 | // Emit an intrinsic that has 2 operands of the same type as its result. |
453 | // Depending on mode, this may be a constrained floating-point intrinsic. |
454 | static Value *emitBinaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
455 | const CallExpr *E, unsigned IntrinsicID, |
456 | unsigned ConstrainedIntrinsicID) { |
457 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
458 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
459 | |
460 | if (CGF.Builder.getIsFPConstrained()) { |
461 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
462 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
463 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1 }); |
464 | } else { |
465 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
466 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); |
467 | } |
468 | } |
469 | |
470 | // Emit an intrinsic that has 3 operands of the same type as its result. |
471 | // Depending on mode, this may be a constrained floating-point intrinsic. |
472 | static Value *emitTernaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
473 | const CallExpr *E, unsigned IntrinsicID, |
474 | unsigned ConstrainedIntrinsicID) { |
475 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
476 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
477 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); |
478 | |
479 | if (CGF.Builder.getIsFPConstrained()) { |
480 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
481 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
482 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1, Src2 }); |
483 | } else { |
484 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
485 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); |
486 | } |
487 | } |
488 | |
489 | // Emit an intrinsic where all operands are of the same type as the result. |
490 | // Depending on mode, this may be a constrained floating-point intrinsic. |
491 | static Value *emitCallMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
492 | unsigned IntrinsicID, |
493 | unsigned ConstrainedIntrinsicID, |
494 | llvm::Type *Ty, |
495 | ArrayRef<Value *> Args) { |
496 | Function *F; |
497 | if (CGF.Builder.getIsFPConstrained()) |
498 | F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Ty); |
499 | else |
500 | F = CGF.CGM.getIntrinsic(IntrinsicID, Ty); |
501 | |
502 | if (CGF.Builder.getIsFPConstrained()) |
503 | return CGF.Builder.CreateConstrainedFPCall(F, Args); |
504 | else |
505 | return CGF.Builder.CreateCall(F, Args); |
506 | } |
507 | |
508 | // Emit a simple mangled intrinsic that has 1 argument and a return type |
509 | // matching the argument type. |
510 | static Value *emitUnaryBuiltin(CodeGenFunction &CGF, |
511 | const CallExpr *E, |
512 | unsigned IntrinsicID) { |
513 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
514 | |
515 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
516 | return CGF.Builder.CreateCall(F, Src0); |
517 | } |
518 | |
519 | // Emit an intrinsic that has 2 operands of the same type as its result. |
520 | static Value *emitBinaryBuiltin(CodeGenFunction &CGF, |
521 | const CallExpr *E, |
522 | unsigned IntrinsicID) { |
523 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
524 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
525 | |
526 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
527 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); |
528 | } |
529 | |
530 | // Emit an intrinsic that has 3 operands of the same type as its result. |
531 | static Value *emitTernaryBuiltin(CodeGenFunction &CGF, |
532 | const CallExpr *E, |
533 | unsigned IntrinsicID) { |
534 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
535 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
536 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); |
537 | |
538 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
539 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); |
540 | } |
541 | |
542 | // Emit an intrinsic that has 1 float or double operand, and 1 integer. |
543 | static Value *emitFPIntBuiltin(CodeGenFunction &CGF, |
544 | const CallExpr *E, |
545 | unsigned IntrinsicID) { |
546 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
547 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
548 | |
549 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
550 | return CGF.Builder.CreateCall(F, {Src0, Src1}); |
551 | } |
552 | |
553 | // Emit an intrinsic that has overloaded integer result and fp operand. |
554 | static Value * |
555 | emitMaybeConstrainedFPToIntRoundBuiltin(CodeGenFunction &CGF, const CallExpr *E, |
556 | unsigned IntrinsicID, |
557 | unsigned ConstrainedIntrinsicID) { |
558 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); |
559 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
560 | |
561 | if (CGF.Builder.getIsFPConstrained()) { |
562 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
563 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, |
564 | {ResultType, Src0->getType()}); |
565 | return CGF.Builder.CreateConstrainedFPCall(F, {Src0}); |
566 | } else { |
567 | Function *F = |
568 | CGF.CGM.getIntrinsic(IntrinsicID, {ResultType, Src0->getType()}); |
569 | return CGF.Builder.CreateCall(F, Src0); |
570 | } |
571 | } |
572 | |
573 | /// EmitFAbs - Emit a call to @llvm.fabs(). |
574 | static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) { |
575 | Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType()); |
576 | llvm::CallInst *Call = CGF.Builder.CreateCall(F, V); |
577 | Call->setDoesNotAccessMemory(); |
578 | return Call; |
579 | } |
580 | |
581 | /// Emit the computation of the sign bit for a floating point value. Returns |
582 | /// the i1 sign bit value. |
583 | static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) { |
584 | LLVMContext &C = CGF.CGM.getLLVMContext(); |
585 | |
586 | llvm::Type *Ty = V->getType(); |
587 | int Width = Ty->getPrimitiveSizeInBits(); |
588 | llvm::Type *IntTy = llvm::IntegerType::get(C, Width); |
589 | V = CGF.Builder.CreateBitCast(V, IntTy); |
590 | if (Ty->isPPC_FP128Ty()) { |
591 | // We want the sign bit of the higher-order double. The bitcast we just |
592 | // did works as if the double-double was stored to memory and then |
593 | // read as an i128. The "store" will put the higher-order double in the |
594 | // lower address in both little- and big-Endian modes, but the "load" |
595 | // will treat those bits as a different part of the i128: the low bits in |
596 | // little-Endian, the high bits in big-Endian. Therefore, on big-Endian |
597 | // we need to shift the high bits down to the low before truncating. |
598 | Width >>= 1; |
599 | if (CGF.getTarget().isBigEndian()) { |
600 | Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width); |
601 | V = CGF.Builder.CreateLShr(V, ShiftCst); |
602 | } |
603 | // We are truncating value in order to extract the higher-order |
604 | // double, which we will be using to extract the sign from. |
605 | IntTy = llvm::IntegerType::get(C, Width); |
606 | V = CGF.Builder.CreateTrunc(V, IntTy); |
607 | } |
608 | Value *Zero = llvm::Constant::getNullValue(IntTy); |
609 | return CGF.Builder.CreateICmpSLT(V, Zero); |
610 | } |
611 | |
612 | static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD, |
613 | const CallExpr *E, llvm::Constant *calleeValue) { |
614 | CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD)); |
615 | return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot()); |
616 | } |
617 | |
618 | /// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.* |
619 | /// depending on IntrinsicID. |
620 | /// |
621 | /// \arg CGF The current codegen function. |
622 | /// \arg IntrinsicID The ID for the Intrinsic we wish to generate. |
623 | /// \arg X The first argument to the llvm.*.with.overflow.*. |
624 | /// \arg Y The second argument to the llvm.*.with.overflow.*. |
625 | /// \arg Carry The carry returned by the llvm.*.with.overflow.*. |
626 | /// \returns The result (i.e. sum/product) returned by the intrinsic. |
627 | static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF, |
628 | const llvm::Intrinsic::ID IntrinsicID, |
629 | llvm::Value *X, llvm::Value *Y, |
630 | llvm::Value *&Carry) { |
631 | // Make sure we have integers of the same width. |
632 | assert(X->getType() == Y->getType() &&((X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? static_cast< 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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 634, __PRETTY_FUNCTION__)) |
633 | "Arguments must be the same type. (Did you forget to make sure both "((X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? static_cast< 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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 634, __PRETTY_FUNCTION__)) |
634 | "arguments have the same integer width?)")((X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? static_cast< 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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 634, __PRETTY_FUNCTION__)); |
635 | |
636 | Function *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType()); |
637 | llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y}); |
638 | Carry = CGF.Builder.CreateExtractValue(Tmp, 1); |
639 | return CGF.Builder.CreateExtractValue(Tmp, 0); |
640 | } |
641 | |
642 | static Value *emitRangedBuiltin(CodeGenFunction &CGF, |
643 | unsigned IntrinsicID, |
644 | int low, int high) { |
645 | llvm::MDBuilder MDHelper(CGF.getLLVMContext()); |
646 | llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high)); |
647 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, {}); |
648 | llvm::Instruction *Call = CGF.Builder.CreateCall(F); |
649 | Call->setMetadata(llvm::LLVMContext::MD_range, RNode); |
650 | return Call; |
651 | } |
652 | |
653 | namespace { |
654 | struct WidthAndSignedness { |
655 | unsigned Width; |
656 | bool Signed; |
657 | }; |
658 | } |
659 | |
660 | static WidthAndSignedness |
661 | getIntegerWidthAndSignedness(const clang::ASTContext &context, |
662 | const clang::QualType Type) { |
663 | assert(Type->isIntegerType() && "Given type is not an integer.")((Type->isIntegerType() && "Given type is not an integer." ) ? static_cast<void> (0) : __assert_fail ("Type->isIntegerType() && \"Given type is not an integer.\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 663, __PRETTY_FUNCTION__)); |
664 | unsigned Width = Type->isBooleanType() ? 1 |
665 | : Type->isExtIntType() ? context.getIntWidth(Type) |
666 | : context.getTypeInfo(Type).Width; |
667 | bool Signed = Type->isSignedIntegerType(); |
668 | return {Width, Signed}; |
669 | } |
670 | |
671 | // Given one or more integer types, this function produces an integer type that |
672 | // encompasses them: any value in one of the given types could be expressed in |
673 | // the encompassing type. |
674 | static struct WidthAndSignedness |
675 | EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) { |
676 | assert(Types.size() > 0 && "Empty list of types.")((Types.size() > 0 && "Empty list of types.") ? static_cast <void> (0) : __assert_fail ("Types.size() > 0 && \"Empty list of types.\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 676, __PRETTY_FUNCTION__)); |
677 | |
678 | // If any of the given types is signed, we must return a signed type. |
679 | bool Signed = false; |
680 | for (const auto &Type : Types) { |
681 | Signed |= Type.Signed; |
682 | } |
683 | |
684 | // The encompassing type must have a width greater than or equal to the width |
685 | // of the specified types. Additionally, if the encompassing type is signed, |
686 | // its width must be strictly greater than the width of any unsigned types |
687 | // given. |
688 | unsigned Width = 0; |
689 | for (const auto &Type : Types) { |
690 | unsigned MinWidth = Type.Width + (Signed && !Type.Signed); |
691 | if (Width < MinWidth) { |
692 | Width = MinWidth; |
693 | } |
694 | } |
695 | |
696 | return {Width, Signed}; |
697 | } |
698 | |
699 | Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) { |
700 | llvm::Type *DestType = Int8PtrTy; |
701 | if (ArgValue->getType() != DestType) |
702 | ArgValue = |
703 | Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data()); |
704 | |
705 | Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend; |
706 | return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue); |
707 | } |
708 | |
709 | /// Checks if using the result of __builtin_object_size(p, @p From) in place of |
710 | /// __builtin_object_size(p, @p To) is correct |
711 | static bool areBOSTypesCompatible(int From, int To) { |
712 | // Note: Our __builtin_object_size implementation currently treats Type=0 and |
713 | // Type=2 identically. Encoding this implementation detail here may make |
714 | // improving __builtin_object_size difficult in the future, so it's omitted. |
715 | return From == To || (From == 0 && To == 1) || (From == 3 && To == 2); |
716 | } |
717 | |
718 | static llvm::Value * |
719 | getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) { |
720 | return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true); |
721 | } |
722 | |
723 | llvm::Value * |
724 | CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type, |
725 | llvm::IntegerType *ResType, |
726 | llvm::Value *EmittedE, |
727 | bool IsDynamic) { |
728 | uint64_t ObjectSize; |
729 | if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type)) |
730 | return emitBuiltinObjectSize(E, Type, ResType, EmittedE, IsDynamic); |
731 | return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true); |
732 | } |
733 | |
734 | /// Returns a Value corresponding to the size of the given expression. |
735 | /// This Value may be either of the following: |
736 | /// - A llvm::Argument (if E is a param with the pass_object_size attribute on |
737 | /// it) |
738 | /// - A call to the @llvm.objectsize intrinsic |
739 | /// |
740 | /// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null |
741 | /// and we wouldn't otherwise try to reference a pass_object_size parameter, |
742 | /// we'll call @llvm.objectsize on EmittedE, rather than emitting E. |
743 | llvm::Value * |
744 | CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type, |
745 | llvm::IntegerType *ResType, |
746 | llvm::Value *EmittedE, bool IsDynamic) { |
747 | // We need to reference an argument if the pointer is a parameter with the |
748 | // pass_object_size attribute. |
749 | if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) { |
750 | auto *Param = dyn_cast<ParmVarDecl>(D->getDecl()); |
751 | auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>(); |
752 | if (Param != nullptr && PS != nullptr && |
753 | areBOSTypesCompatible(PS->getType(), Type)) { |
754 | auto Iter = SizeArguments.find(Param); |
755 | assert(Iter != SizeArguments.end())((Iter != SizeArguments.end()) ? static_cast<void> (0) : __assert_fail ("Iter != SizeArguments.end()", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 755, __PRETTY_FUNCTION__)); |
756 | |
757 | const ImplicitParamDecl *D = Iter->second; |
758 | auto DIter = LocalDeclMap.find(D); |
759 | assert(DIter != LocalDeclMap.end())((DIter != LocalDeclMap.end()) ? static_cast<void> (0) : __assert_fail ("DIter != LocalDeclMap.end()", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 759, __PRETTY_FUNCTION__)); |
760 | |
761 | return EmitLoadOfScalar(DIter->second, /*Volatile=*/false, |
762 | getContext().getSizeType(), E->getBeginLoc()); |
763 | } |
764 | } |
765 | |
766 | // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't |
767 | // evaluate E for side-effects. In either case, we shouldn't lower to |
768 | // @llvm.objectsize. |
769 | if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext()))) |
770 | return getDefaultBuiltinObjectSizeResult(Type, ResType); |
771 | |
772 | Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E); |
773 | assert(Ptr->getType()->isPointerTy() &&((Ptr->getType()->isPointerTy() && "Non-pointer passed to __builtin_object_size?" ) ? static_cast<void> (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 774, __PRETTY_FUNCTION__)) |
774 | "Non-pointer passed to __builtin_object_size?")((Ptr->getType()->isPointerTy() && "Non-pointer passed to __builtin_object_size?" ) ? static_cast<void> (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 774, __PRETTY_FUNCTION__)); |
775 | |
776 | Function *F = |
777 | CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()}); |
778 | |
779 | // LLVM only supports 0 and 2, make sure that we pass along that as a boolean. |
780 | Value *Min = Builder.getInt1((Type & 2) != 0); |
781 | // For GCC compatibility, __builtin_object_size treat NULL as unknown size. |
782 | Value *NullIsUnknown = Builder.getTrue(); |
783 | Value *Dynamic = Builder.getInt1(IsDynamic); |
784 | return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown, Dynamic}); |
785 | } |
786 | |
787 | namespace { |
788 | /// A struct to generically describe a bit test intrinsic. |
789 | struct BitTest { |
790 | enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set }; |
791 | enum InterlockingKind : uint8_t { |
792 | Unlocked, |
793 | Sequential, |
794 | Acquire, |
795 | Release, |
796 | NoFence |
797 | }; |
798 | |
799 | ActionKind Action; |
800 | InterlockingKind Interlocking; |
801 | bool Is64Bit; |
802 | |
803 | static BitTest decodeBitTestBuiltin(unsigned BuiltinID); |
804 | }; |
805 | } // namespace |
806 | |
807 | BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) { |
808 | switch (BuiltinID) { |
809 | // Main portable variants. |
810 | case Builtin::BI_bittest: |
811 | return {TestOnly, Unlocked, false}; |
812 | case Builtin::BI_bittestandcomplement: |
813 | return {Complement, Unlocked, false}; |
814 | case Builtin::BI_bittestandreset: |
815 | return {Reset, Unlocked, false}; |
816 | case Builtin::BI_bittestandset: |
817 | return {Set, Unlocked, false}; |
818 | case Builtin::BI_interlockedbittestandreset: |
819 | return {Reset, Sequential, false}; |
820 | case Builtin::BI_interlockedbittestandset: |
821 | return {Set, Sequential, false}; |
822 | |
823 | // X86-specific 64-bit variants. |
824 | case Builtin::BI_bittest64: |
825 | return {TestOnly, Unlocked, true}; |
826 | case Builtin::BI_bittestandcomplement64: |
827 | return {Complement, Unlocked, true}; |
828 | case Builtin::BI_bittestandreset64: |
829 | return {Reset, Unlocked, true}; |
830 | case Builtin::BI_bittestandset64: |
831 | return {Set, Unlocked, true}; |
832 | case Builtin::BI_interlockedbittestandreset64: |
833 | return {Reset, Sequential, true}; |
834 | case Builtin::BI_interlockedbittestandset64: |
835 | return {Set, Sequential, true}; |
836 | |
837 | // ARM/AArch64-specific ordering variants. |
838 | case Builtin::BI_interlockedbittestandset_acq: |
839 | return {Set, Acquire, false}; |
840 | case Builtin::BI_interlockedbittestandset_rel: |
841 | return {Set, Release, false}; |
842 | case Builtin::BI_interlockedbittestandset_nf: |
843 | return {Set, NoFence, false}; |
844 | case Builtin::BI_interlockedbittestandreset_acq: |
845 | return {Reset, Acquire, false}; |
846 | case Builtin::BI_interlockedbittestandreset_rel: |
847 | return {Reset, Release, false}; |
848 | case Builtin::BI_interlockedbittestandreset_nf: |
849 | return {Reset, NoFence, false}; |
850 | } |
851 | llvm_unreachable("expected only bittest intrinsics")::llvm::llvm_unreachable_internal("expected only bittest intrinsics" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 851); |
852 | } |
853 | |
854 | static char bitActionToX86BTCode(BitTest::ActionKind A) { |
855 | switch (A) { |
856 | case BitTest::TestOnly: return '\0'; |
857 | case BitTest::Complement: return 'c'; |
858 | case BitTest::Reset: return 'r'; |
859 | case BitTest::Set: return 's'; |
860 | } |
861 | llvm_unreachable("invalid action")::llvm::llvm_unreachable_internal("invalid action", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 861); |
862 | } |
863 | |
864 | static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF, |
865 | BitTest BT, |
866 | const CallExpr *E, Value *BitBase, |
867 | Value *BitPos) { |
868 | char Action = bitActionToX86BTCode(BT.Action); |
869 | char SizeSuffix = BT.Is64Bit ? 'q' : 'l'; |
870 | |
871 | // Build the assembly. |
872 | SmallString<64> Asm; |
873 | raw_svector_ostream AsmOS(Asm); |
874 | if (BT.Interlocking != BitTest::Unlocked) |
875 | AsmOS << "lock "; |
876 | AsmOS << "bt"; |
877 | if (Action) |
878 | AsmOS << Action; |
879 | AsmOS << SizeSuffix << " $2, ($1)"; |
880 | |
881 | // Build the constraints. FIXME: We should support immediates when possible. |
882 | std::string Constraints = "={@ccc},r,r,~{cc},~{memory}"; |
883 | std::string MachineClobbers = CGF.getTarget().getClobbers(); |
884 | if (!MachineClobbers.empty()) { |
885 | Constraints += ','; |
886 | Constraints += MachineClobbers; |
887 | } |
888 | llvm::IntegerType *IntType = llvm::IntegerType::get( |
889 | CGF.getLLVMContext(), |
890 | CGF.getContext().getTypeSize(E->getArg(1)->getType())); |
891 | llvm::Type *IntPtrType = IntType->getPointerTo(); |
892 | llvm::FunctionType *FTy = |
893 | llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false); |
894 | |
895 | llvm::InlineAsm *IA = |
896 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); |
897 | return CGF.Builder.CreateCall(IA, {BitBase, BitPos}); |
898 | } |
899 | |
900 | static llvm::AtomicOrdering |
901 | getBitTestAtomicOrdering(BitTest::InterlockingKind I) { |
902 | switch (I) { |
903 | case BitTest::Unlocked: return llvm::AtomicOrdering::NotAtomic; |
904 | case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent; |
905 | case BitTest::Acquire: return llvm::AtomicOrdering::Acquire; |
906 | case BitTest::Release: return llvm::AtomicOrdering::Release; |
907 | case BitTest::NoFence: return llvm::AtomicOrdering::Monotonic; |
908 | } |
909 | llvm_unreachable("invalid interlocking")::llvm::llvm_unreachable_internal("invalid interlocking", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 909); |
910 | } |
911 | |
912 | /// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of |
913 | /// bits and a bit position and read and optionally modify the bit at that |
914 | /// position. The position index can be arbitrarily large, i.e. it can be larger |
915 | /// than 31 or 63, so we need an indexed load in the general case. |
916 | static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF, |
917 | unsigned BuiltinID, |
918 | const CallExpr *E) { |
919 | Value *BitBase = CGF.EmitScalarExpr(E->getArg(0)); |
920 | Value *BitPos = CGF.EmitScalarExpr(E->getArg(1)); |
921 | |
922 | BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID); |
923 | |
924 | // X86 has special BT, BTC, BTR, and BTS instructions that handle the array |
925 | // indexing operation internally. Use them if possible. |
926 | if (CGF.getTarget().getTriple().isX86()) |
927 | return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos); |
928 | |
929 | // Otherwise, use generic code to load one byte and test the bit. Use all but |
930 | // the bottom three bits as the array index, and the bottom three bits to form |
931 | // a mask. |
932 | // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0; |
933 | Value *ByteIndex = CGF.Builder.CreateAShr( |
934 | BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx"); |
935 | Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy); |
936 | Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8, |
937 | ByteIndex, "bittest.byteaddr"), |
938 | CharUnits::One()); |
939 | Value *PosLow = |
940 | CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty), |
941 | llvm::ConstantInt::get(CGF.Int8Ty, 0x7)); |
942 | |
943 | // The updating instructions will need a mask. |
944 | Value *Mask = nullptr; |
945 | if (BT.Action != BitTest::TestOnly) { |
946 | Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow, |
947 | "bittest.mask"); |
948 | } |
949 | |
950 | // Check the action and ordering of the interlocked intrinsics. |
951 | llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking); |
952 | |
953 | Value *OldByte = nullptr; |
954 | if (Ordering != llvm::AtomicOrdering::NotAtomic) { |
955 | // Emit a combined atomicrmw load/store operation for the interlocked |
956 | // intrinsics. |
957 | llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or; |
958 | if (BT.Action == BitTest::Reset) { |
959 | Mask = CGF.Builder.CreateNot(Mask); |
960 | RMWOp = llvm::AtomicRMWInst::And; |
961 | } |
962 | OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask, |
963 | Ordering); |
964 | } else { |
965 | // Emit a plain load for the non-interlocked intrinsics. |
966 | OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte"); |
967 | Value *NewByte = nullptr; |
968 | switch (BT.Action) { |
969 | case BitTest::TestOnly: |
970 | // Don't store anything. |
971 | break; |
972 | case BitTest::Complement: |
973 | NewByte = CGF.Builder.CreateXor(OldByte, Mask); |
974 | break; |
975 | case BitTest::Reset: |
976 | NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask)); |
977 | break; |
978 | case BitTest::Set: |
979 | NewByte = CGF.Builder.CreateOr(OldByte, Mask); |
980 | break; |
981 | } |
982 | if (NewByte) |
983 | CGF.Builder.CreateStore(NewByte, ByteAddr); |
984 | } |
985 | |
986 | // However we loaded the old byte, either by plain load or atomicrmw, shift |
987 | // the bit into the low position and mask it to 0 or 1. |
988 | Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr"); |
989 | return CGF.Builder.CreateAnd( |
990 | ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res"); |
991 | } |
992 | |
993 | namespace { |
994 | enum class MSVCSetJmpKind { |
995 | _setjmpex, |
996 | _setjmp3, |
997 | _setjmp |
998 | }; |
999 | } |
1000 | |
1001 | /// MSVC handles setjmp a bit differently on different platforms. On every |
1002 | /// architecture except 32-bit x86, the frame address is passed. On x86, extra |
1003 | /// parameters can be passed as variadic arguments, but we always pass none. |
1004 | static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind, |
1005 | const CallExpr *E) { |
1006 | llvm::Value *Arg1 = nullptr; |
1007 | llvm::Type *Arg1Ty = nullptr; |
1008 | StringRef Name; |
1009 | bool IsVarArg = false; |
1010 | if (SJKind == MSVCSetJmpKind::_setjmp3) { |
1011 | Name = "_setjmp3"; |
1012 | Arg1Ty = CGF.Int32Ty; |
1013 | Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0); |
1014 | IsVarArg = true; |
1015 | } else { |
1016 | Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex"; |
1017 | Arg1Ty = CGF.Int8PtrTy; |
1018 | if (CGF.getTarget().getTriple().getArch() == llvm::Triple::aarch64) { |
1019 | Arg1 = CGF.Builder.CreateCall( |
1020 | CGF.CGM.getIntrinsic(Intrinsic::sponentry, CGF.AllocaInt8PtrTy)); |
1021 | } else |
1022 | Arg1 = CGF.Builder.CreateCall( |
1023 | CGF.CGM.getIntrinsic(Intrinsic::frameaddress, CGF.AllocaInt8PtrTy), |
1024 | llvm::ConstantInt::get(CGF.Int32Ty, 0)); |
1025 | } |
1026 | |
1027 | // Mark the call site and declaration with ReturnsTwice. |
1028 | llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty}; |
1029 | llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get( |
1030 | CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex, |
1031 | llvm::Attribute::ReturnsTwice); |
1032 | llvm::FunctionCallee SetJmpFn = CGF.CGM.CreateRuntimeFunction( |
1033 | llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name, |
1034 | ReturnsTwiceAttr, /*Local=*/true); |
1035 | |
1036 | llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast( |
1037 | CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy); |
1038 | llvm::Value *Args[] = {Buf, Arg1}; |
1039 | llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args); |
1040 | CB->setAttributes(ReturnsTwiceAttr); |
1041 | return RValue::get(CB); |
1042 | } |
1043 | |
1044 | // Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code, |
1045 | // we handle them here. |
1046 | enum class CodeGenFunction::MSVCIntrin { |
1047 | _BitScanForward, |
1048 | _BitScanReverse, |
1049 | _InterlockedAnd, |
1050 | _InterlockedDecrement, |
1051 | _InterlockedExchange, |
1052 | _InterlockedExchangeAdd, |
1053 | _InterlockedExchangeSub, |
1054 | _InterlockedIncrement, |
1055 | _InterlockedOr, |
1056 | _InterlockedXor, |
1057 | _InterlockedExchangeAdd_acq, |
1058 | _InterlockedExchangeAdd_rel, |
1059 | _InterlockedExchangeAdd_nf, |
1060 | _InterlockedExchange_acq, |
1061 | _InterlockedExchange_rel, |
1062 | _InterlockedExchange_nf, |
1063 | _InterlockedCompareExchange_acq, |
1064 | _InterlockedCompareExchange_rel, |
1065 | _InterlockedCompareExchange_nf, |
1066 | _InterlockedCompareExchange128, |
1067 | _InterlockedCompareExchange128_acq, |
1068 | _InterlockedCompareExchange128_rel, |
1069 | _InterlockedCompareExchange128_nf, |
1070 | _InterlockedOr_acq, |
1071 | _InterlockedOr_rel, |
1072 | _InterlockedOr_nf, |
1073 | _InterlockedXor_acq, |
1074 | _InterlockedXor_rel, |
1075 | _InterlockedXor_nf, |
1076 | _InterlockedAnd_acq, |
1077 | _InterlockedAnd_rel, |
1078 | _InterlockedAnd_nf, |
1079 | _InterlockedIncrement_acq, |
1080 | _InterlockedIncrement_rel, |
1081 | _InterlockedIncrement_nf, |
1082 | _InterlockedDecrement_acq, |
1083 | _InterlockedDecrement_rel, |
1084 | _InterlockedDecrement_nf, |
1085 | __fastfail, |
1086 | }; |
1087 | |
1088 | static Optional<CodeGenFunction::MSVCIntrin> |
1089 | translateArmToMsvcIntrin(unsigned BuiltinID) { |
1090 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
1091 | switch (BuiltinID) { |
1092 | default: |
1093 | return None; |
1094 | case ARM::BI_BitScanForward: |
1095 | case ARM::BI_BitScanForward64: |
1096 | return MSVCIntrin::_BitScanForward; |
1097 | case ARM::BI_BitScanReverse: |
1098 | case ARM::BI_BitScanReverse64: |
1099 | return MSVCIntrin::_BitScanReverse; |
1100 | case ARM::BI_InterlockedAnd64: |
1101 | return MSVCIntrin::_InterlockedAnd; |
1102 | case ARM::BI_InterlockedExchange64: |
1103 | return MSVCIntrin::_InterlockedExchange; |
1104 | case ARM::BI_InterlockedExchangeAdd64: |
1105 | return MSVCIntrin::_InterlockedExchangeAdd; |
1106 | case ARM::BI_InterlockedExchangeSub64: |
1107 | return MSVCIntrin::_InterlockedExchangeSub; |
1108 | case ARM::BI_InterlockedOr64: |
1109 | return MSVCIntrin::_InterlockedOr; |
1110 | case ARM::BI_InterlockedXor64: |
1111 | return MSVCIntrin::_InterlockedXor; |
1112 | case ARM::BI_InterlockedDecrement64: |
1113 | return MSVCIntrin::_InterlockedDecrement; |
1114 | case ARM::BI_InterlockedIncrement64: |
1115 | return MSVCIntrin::_InterlockedIncrement; |
1116 | case ARM::BI_InterlockedExchangeAdd8_acq: |
1117 | case ARM::BI_InterlockedExchangeAdd16_acq: |
1118 | case ARM::BI_InterlockedExchangeAdd_acq: |
1119 | case ARM::BI_InterlockedExchangeAdd64_acq: |
1120 | return MSVCIntrin::_InterlockedExchangeAdd_acq; |
1121 | case ARM::BI_InterlockedExchangeAdd8_rel: |
1122 | case ARM::BI_InterlockedExchangeAdd16_rel: |
1123 | case ARM::BI_InterlockedExchangeAdd_rel: |
1124 | case ARM::BI_InterlockedExchangeAdd64_rel: |
1125 | return MSVCIntrin::_InterlockedExchangeAdd_rel; |
1126 | case ARM::BI_InterlockedExchangeAdd8_nf: |
1127 | case ARM::BI_InterlockedExchangeAdd16_nf: |
1128 | case ARM::BI_InterlockedExchangeAdd_nf: |
1129 | case ARM::BI_InterlockedExchangeAdd64_nf: |
1130 | return MSVCIntrin::_InterlockedExchangeAdd_nf; |
1131 | case ARM::BI_InterlockedExchange8_acq: |
1132 | case ARM::BI_InterlockedExchange16_acq: |
1133 | case ARM::BI_InterlockedExchange_acq: |
1134 | case ARM::BI_InterlockedExchange64_acq: |
1135 | return MSVCIntrin::_InterlockedExchange_acq; |
1136 | case ARM::BI_InterlockedExchange8_rel: |
1137 | case ARM::BI_InterlockedExchange16_rel: |
1138 | case ARM::BI_InterlockedExchange_rel: |
1139 | case ARM::BI_InterlockedExchange64_rel: |
1140 | return MSVCIntrin::_InterlockedExchange_rel; |
1141 | case ARM::BI_InterlockedExchange8_nf: |
1142 | case ARM::BI_InterlockedExchange16_nf: |
1143 | case ARM::BI_InterlockedExchange_nf: |
1144 | case ARM::BI_InterlockedExchange64_nf: |
1145 | return MSVCIntrin::_InterlockedExchange_nf; |
1146 | case ARM::BI_InterlockedCompareExchange8_acq: |
1147 | case ARM::BI_InterlockedCompareExchange16_acq: |
1148 | case ARM::BI_InterlockedCompareExchange_acq: |
1149 | case ARM::BI_InterlockedCompareExchange64_acq: |
1150 | return MSVCIntrin::_InterlockedCompareExchange_acq; |
1151 | case ARM::BI_InterlockedCompareExchange8_rel: |
1152 | case ARM::BI_InterlockedCompareExchange16_rel: |
1153 | case ARM::BI_InterlockedCompareExchange_rel: |
1154 | case ARM::BI_InterlockedCompareExchange64_rel: |
1155 | return MSVCIntrin::_InterlockedCompareExchange_rel; |
1156 | case ARM::BI_InterlockedCompareExchange8_nf: |
1157 | case ARM::BI_InterlockedCompareExchange16_nf: |
1158 | case ARM::BI_InterlockedCompareExchange_nf: |
1159 | case ARM::BI_InterlockedCompareExchange64_nf: |
1160 | return MSVCIntrin::_InterlockedCompareExchange_nf; |
1161 | case ARM::BI_InterlockedOr8_acq: |
1162 | case ARM::BI_InterlockedOr16_acq: |
1163 | case ARM::BI_InterlockedOr_acq: |
1164 | case ARM::BI_InterlockedOr64_acq: |
1165 | return MSVCIntrin::_InterlockedOr_acq; |
1166 | case ARM::BI_InterlockedOr8_rel: |
1167 | case ARM::BI_InterlockedOr16_rel: |
1168 | case ARM::BI_InterlockedOr_rel: |
1169 | case ARM::BI_InterlockedOr64_rel: |
1170 | return MSVCIntrin::_InterlockedOr_rel; |
1171 | case ARM::BI_InterlockedOr8_nf: |
1172 | case ARM::BI_InterlockedOr16_nf: |
1173 | case ARM::BI_InterlockedOr_nf: |
1174 | case ARM::BI_InterlockedOr64_nf: |
1175 | return MSVCIntrin::_InterlockedOr_nf; |
1176 | case ARM::BI_InterlockedXor8_acq: |
1177 | case ARM::BI_InterlockedXor16_acq: |
1178 | case ARM::BI_InterlockedXor_acq: |
1179 | case ARM::BI_InterlockedXor64_acq: |
1180 | return MSVCIntrin::_InterlockedXor_acq; |
1181 | case ARM::BI_InterlockedXor8_rel: |
1182 | case ARM::BI_InterlockedXor16_rel: |
1183 | case ARM::BI_InterlockedXor_rel: |
1184 | case ARM::BI_InterlockedXor64_rel: |
1185 | return MSVCIntrin::_InterlockedXor_rel; |
1186 | case ARM::BI_InterlockedXor8_nf: |
1187 | case ARM::BI_InterlockedXor16_nf: |
1188 | case ARM::BI_InterlockedXor_nf: |
1189 | case ARM::BI_InterlockedXor64_nf: |
1190 | return MSVCIntrin::_InterlockedXor_nf; |
1191 | case ARM::BI_InterlockedAnd8_acq: |
1192 | case ARM::BI_InterlockedAnd16_acq: |
1193 | case ARM::BI_InterlockedAnd_acq: |
1194 | case ARM::BI_InterlockedAnd64_acq: |
1195 | return MSVCIntrin::_InterlockedAnd_acq; |
1196 | case ARM::BI_InterlockedAnd8_rel: |
1197 | case ARM::BI_InterlockedAnd16_rel: |
1198 | case ARM::BI_InterlockedAnd_rel: |
1199 | case ARM::BI_InterlockedAnd64_rel: |
1200 | return MSVCIntrin::_InterlockedAnd_rel; |
1201 | case ARM::BI_InterlockedAnd8_nf: |
1202 | case ARM::BI_InterlockedAnd16_nf: |
1203 | case ARM::BI_InterlockedAnd_nf: |
1204 | case ARM::BI_InterlockedAnd64_nf: |
1205 | return MSVCIntrin::_InterlockedAnd_nf; |
1206 | case ARM::BI_InterlockedIncrement16_acq: |
1207 | case ARM::BI_InterlockedIncrement_acq: |
1208 | case ARM::BI_InterlockedIncrement64_acq: |
1209 | return MSVCIntrin::_InterlockedIncrement_acq; |
1210 | case ARM::BI_InterlockedIncrement16_rel: |
1211 | case ARM::BI_InterlockedIncrement_rel: |
1212 | case ARM::BI_InterlockedIncrement64_rel: |
1213 | return MSVCIntrin::_InterlockedIncrement_rel; |
1214 | case ARM::BI_InterlockedIncrement16_nf: |
1215 | case ARM::BI_InterlockedIncrement_nf: |
1216 | case ARM::BI_InterlockedIncrement64_nf: |
1217 | return MSVCIntrin::_InterlockedIncrement_nf; |
1218 | case ARM::BI_InterlockedDecrement16_acq: |
1219 | case ARM::BI_InterlockedDecrement_acq: |
1220 | case ARM::BI_InterlockedDecrement64_acq: |
1221 | return MSVCIntrin::_InterlockedDecrement_acq; |
1222 | case ARM::BI_InterlockedDecrement16_rel: |
1223 | case ARM::BI_InterlockedDecrement_rel: |
1224 | case ARM::BI_InterlockedDecrement64_rel: |
1225 | return MSVCIntrin::_InterlockedDecrement_rel; |
1226 | case ARM::BI_InterlockedDecrement16_nf: |
1227 | case ARM::BI_InterlockedDecrement_nf: |
1228 | case ARM::BI_InterlockedDecrement64_nf: |
1229 | return MSVCIntrin::_InterlockedDecrement_nf; |
1230 | } |
1231 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1231); |
1232 | } |
1233 | |
1234 | static Optional<CodeGenFunction::MSVCIntrin> |
1235 | translateAarch64ToMsvcIntrin(unsigned BuiltinID) { |
1236 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
1237 | switch (BuiltinID) { |
1238 | default: |
1239 | return None; |
1240 | case AArch64::BI_BitScanForward: |
1241 | case AArch64::BI_BitScanForward64: |
1242 | return MSVCIntrin::_BitScanForward; |
1243 | case AArch64::BI_BitScanReverse: |
1244 | case AArch64::BI_BitScanReverse64: |
1245 | return MSVCIntrin::_BitScanReverse; |
1246 | case AArch64::BI_InterlockedAnd64: |
1247 | return MSVCIntrin::_InterlockedAnd; |
1248 | case AArch64::BI_InterlockedExchange64: |
1249 | return MSVCIntrin::_InterlockedExchange; |
1250 | case AArch64::BI_InterlockedExchangeAdd64: |
1251 | return MSVCIntrin::_InterlockedExchangeAdd; |
1252 | case AArch64::BI_InterlockedExchangeSub64: |
1253 | return MSVCIntrin::_InterlockedExchangeSub; |
1254 | case AArch64::BI_InterlockedOr64: |
1255 | return MSVCIntrin::_InterlockedOr; |
1256 | case AArch64::BI_InterlockedXor64: |
1257 | return MSVCIntrin::_InterlockedXor; |
1258 | case AArch64::BI_InterlockedDecrement64: |
1259 | return MSVCIntrin::_InterlockedDecrement; |
1260 | case AArch64::BI_InterlockedIncrement64: |
1261 | return MSVCIntrin::_InterlockedIncrement; |
1262 | case AArch64::BI_InterlockedExchangeAdd8_acq: |
1263 | case AArch64::BI_InterlockedExchangeAdd16_acq: |
1264 | case AArch64::BI_InterlockedExchangeAdd_acq: |
1265 | case AArch64::BI_InterlockedExchangeAdd64_acq: |
1266 | return MSVCIntrin::_InterlockedExchangeAdd_acq; |
1267 | case AArch64::BI_InterlockedExchangeAdd8_rel: |
1268 | case AArch64::BI_InterlockedExchangeAdd16_rel: |
1269 | case AArch64::BI_InterlockedExchangeAdd_rel: |
1270 | case AArch64::BI_InterlockedExchangeAdd64_rel: |
1271 | return MSVCIntrin::_InterlockedExchangeAdd_rel; |
1272 | case AArch64::BI_InterlockedExchangeAdd8_nf: |
1273 | case AArch64::BI_InterlockedExchangeAdd16_nf: |
1274 | case AArch64::BI_InterlockedExchangeAdd_nf: |
1275 | case AArch64::BI_InterlockedExchangeAdd64_nf: |
1276 | return MSVCIntrin::_InterlockedExchangeAdd_nf; |
1277 | case AArch64::BI_InterlockedExchange8_acq: |
1278 | case AArch64::BI_InterlockedExchange16_acq: |
1279 | case AArch64::BI_InterlockedExchange_acq: |
1280 | case AArch64::BI_InterlockedExchange64_acq: |
1281 | return MSVCIntrin::_InterlockedExchange_acq; |
1282 | case AArch64::BI_InterlockedExchange8_rel: |
1283 | case AArch64::BI_InterlockedExchange16_rel: |
1284 | case AArch64::BI_InterlockedExchange_rel: |
1285 | case AArch64::BI_InterlockedExchange64_rel: |
1286 | return MSVCIntrin::_InterlockedExchange_rel; |
1287 | case AArch64::BI_InterlockedExchange8_nf: |
1288 | case AArch64::BI_InterlockedExchange16_nf: |
1289 | case AArch64::BI_InterlockedExchange_nf: |
1290 | case AArch64::BI_InterlockedExchange64_nf: |
1291 | return MSVCIntrin::_InterlockedExchange_nf; |
1292 | case AArch64::BI_InterlockedCompareExchange8_acq: |
1293 | case AArch64::BI_InterlockedCompareExchange16_acq: |
1294 | case AArch64::BI_InterlockedCompareExchange_acq: |
1295 | case AArch64::BI_InterlockedCompareExchange64_acq: |
1296 | return MSVCIntrin::_InterlockedCompareExchange_acq; |
1297 | case AArch64::BI_InterlockedCompareExchange8_rel: |
1298 | case AArch64::BI_InterlockedCompareExchange16_rel: |
1299 | case AArch64::BI_InterlockedCompareExchange_rel: |
1300 | case AArch64::BI_InterlockedCompareExchange64_rel: |
1301 | return MSVCIntrin::_InterlockedCompareExchange_rel; |
1302 | case AArch64::BI_InterlockedCompareExchange8_nf: |
1303 | case AArch64::BI_InterlockedCompareExchange16_nf: |
1304 | case AArch64::BI_InterlockedCompareExchange_nf: |
1305 | case AArch64::BI_InterlockedCompareExchange64_nf: |
1306 | return MSVCIntrin::_InterlockedCompareExchange_nf; |
1307 | case AArch64::BI_InterlockedCompareExchange128: |
1308 | return MSVCIntrin::_InterlockedCompareExchange128; |
1309 | case AArch64::BI_InterlockedCompareExchange128_acq: |
1310 | return MSVCIntrin::_InterlockedCompareExchange128_acq; |
1311 | case AArch64::BI_InterlockedCompareExchange128_nf: |
1312 | return MSVCIntrin::_InterlockedCompareExchange128_nf; |
1313 | case AArch64::BI_InterlockedCompareExchange128_rel: |
1314 | return MSVCIntrin::_InterlockedCompareExchange128_rel; |
1315 | case AArch64::BI_InterlockedOr8_acq: |
1316 | case AArch64::BI_InterlockedOr16_acq: |
1317 | case AArch64::BI_InterlockedOr_acq: |
1318 | case AArch64::BI_InterlockedOr64_acq: |
1319 | return MSVCIntrin::_InterlockedOr_acq; |
1320 | case AArch64::BI_InterlockedOr8_rel: |
1321 | case AArch64::BI_InterlockedOr16_rel: |
1322 | case AArch64::BI_InterlockedOr_rel: |
1323 | case AArch64::BI_InterlockedOr64_rel: |
1324 | return MSVCIntrin::_InterlockedOr_rel; |
1325 | case AArch64::BI_InterlockedOr8_nf: |
1326 | case AArch64::BI_InterlockedOr16_nf: |
1327 | case AArch64::BI_InterlockedOr_nf: |
1328 | case AArch64::BI_InterlockedOr64_nf: |
1329 | return MSVCIntrin::_InterlockedOr_nf; |
1330 | case AArch64::BI_InterlockedXor8_acq: |
1331 | case AArch64::BI_InterlockedXor16_acq: |
1332 | case AArch64::BI_InterlockedXor_acq: |
1333 | case AArch64::BI_InterlockedXor64_acq: |
1334 | return MSVCIntrin::_InterlockedXor_acq; |
1335 | case AArch64::BI_InterlockedXor8_rel: |
1336 | case AArch64::BI_InterlockedXor16_rel: |
1337 | case AArch64::BI_InterlockedXor_rel: |
1338 | case AArch64::BI_InterlockedXor64_rel: |
1339 | return MSVCIntrin::_InterlockedXor_rel; |
1340 | case AArch64::BI_InterlockedXor8_nf: |
1341 | case AArch64::BI_InterlockedXor16_nf: |
1342 | case AArch64::BI_InterlockedXor_nf: |
1343 | case AArch64::BI_InterlockedXor64_nf: |
1344 | return MSVCIntrin::_InterlockedXor_nf; |
1345 | case AArch64::BI_InterlockedAnd8_acq: |
1346 | case AArch64::BI_InterlockedAnd16_acq: |
1347 | case AArch64::BI_InterlockedAnd_acq: |
1348 | case AArch64::BI_InterlockedAnd64_acq: |
1349 | return MSVCIntrin::_InterlockedAnd_acq; |
1350 | case AArch64::BI_InterlockedAnd8_rel: |
1351 | case AArch64::BI_InterlockedAnd16_rel: |
1352 | case AArch64::BI_InterlockedAnd_rel: |
1353 | case AArch64::BI_InterlockedAnd64_rel: |
1354 | return MSVCIntrin::_InterlockedAnd_rel; |
1355 | case AArch64::BI_InterlockedAnd8_nf: |
1356 | case AArch64::BI_InterlockedAnd16_nf: |
1357 | case AArch64::BI_InterlockedAnd_nf: |
1358 | case AArch64::BI_InterlockedAnd64_nf: |
1359 | return MSVCIntrin::_InterlockedAnd_nf; |
1360 | case AArch64::BI_InterlockedIncrement16_acq: |
1361 | case AArch64::BI_InterlockedIncrement_acq: |
1362 | case AArch64::BI_InterlockedIncrement64_acq: |
1363 | return MSVCIntrin::_InterlockedIncrement_acq; |
1364 | case AArch64::BI_InterlockedIncrement16_rel: |
1365 | case AArch64::BI_InterlockedIncrement_rel: |
1366 | case AArch64::BI_InterlockedIncrement64_rel: |
1367 | return MSVCIntrin::_InterlockedIncrement_rel; |
1368 | case AArch64::BI_InterlockedIncrement16_nf: |
1369 | case AArch64::BI_InterlockedIncrement_nf: |
1370 | case AArch64::BI_InterlockedIncrement64_nf: |
1371 | return MSVCIntrin::_InterlockedIncrement_nf; |
1372 | case AArch64::BI_InterlockedDecrement16_acq: |
1373 | case AArch64::BI_InterlockedDecrement_acq: |
1374 | case AArch64::BI_InterlockedDecrement64_acq: |
1375 | return MSVCIntrin::_InterlockedDecrement_acq; |
1376 | case AArch64::BI_InterlockedDecrement16_rel: |
1377 | case AArch64::BI_InterlockedDecrement_rel: |
1378 | case AArch64::BI_InterlockedDecrement64_rel: |
1379 | return MSVCIntrin::_InterlockedDecrement_rel; |
1380 | case AArch64::BI_InterlockedDecrement16_nf: |
1381 | case AArch64::BI_InterlockedDecrement_nf: |
1382 | case AArch64::BI_InterlockedDecrement64_nf: |
1383 | return MSVCIntrin::_InterlockedDecrement_nf; |
1384 | } |
1385 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1385); |
1386 | } |
1387 | |
1388 | static Optional<CodeGenFunction::MSVCIntrin> |
1389 | translateX86ToMsvcIntrin(unsigned BuiltinID) { |
1390 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
1391 | switch (BuiltinID) { |
1392 | default: |
1393 | return None; |
1394 | case clang::X86::BI_BitScanForward: |
1395 | case clang::X86::BI_BitScanForward64: |
1396 | return MSVCIntrin::_BitScanForward; |
1397 | case clang::X86::BI_BitScanReverse: |
1398 | case clang::X86::BI_BitScanReverse64: |
1399 | return MSVCIntrin::_BitScanReverse; |
1400 | case clang::X86::BI_InterlockedAnd64: |
1401 | return MSVCIntrin::_InterlockedAnd; |
1402 | case clang::X86::BI_InterlockedCompareExchange128: |
1403 | return MSVCIntrin::_InterlockedCompareExchange128; |
1404 | case clang::X86::BI_InterlockedExchange64: |
1405 | return MSVCIntrin::_InterlockedExchange; |
1406 | case clang::X86::BI_InterlockedExchangeAdd64: |
1407 | return MSVCIntrin::_InterlockedExchangeAdd; |
1408 | case clang::X86::BI_InterlockedExchangeSub64: |
1409 | return MSVCIntrin::_InterlockedExchangeSub; |
1410 | case clang::X86::BI_InterlockedOr64: |
1411 | return MSVCIntrin::_InterlockedOr; |
1412 | case clang::X86::BI_InterlockedXor64: |
1413 | return MSVCIntrin::_InterlockedXor; |
1414 | case clang::X86::BI_InterlockedDecrement64: |
1415 | return MSVCIntrin::_InterlockedDecrement; |
1416 | case clang::X86::BI_InterlockedIncrement64: |
1417 | return MSVCIntrin::_InterlockedIncrement; |
1418 | } |
1419 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1419); |
1420 | } |
1421 | |
1422 | // Emit an MSVC intrinsic. Assumes that arguments have *not* been evaluated. |
1423 | Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID, |
1424 | const CallExpr *E) { |
1425 | switch (BuiltinID) { |
1426 | case MSVCIntrin::_BitScanForward: |
1427 | case MSVCIntrin::_BitScanReverse: { |
1428 | Address IndexAddress(EmitPointerWithAlignment(E->getArg(0))); |
1429 | Value *ArgValue = EmitScalarExpr(E->getArg(1)); |
1430 | |
1431 | llvm::Type *ArgType = ArgValue->getType(); |
1432 | llvm::Type *IndexType = |
1433 | IndexAddress.getPointer()->getType()->getPointerElementType(); |
1434 | llvm::Type *ResultType = ConvertType(E->getType()); |
1435 | |
1436 | Value *ArgZero = llvm::Constant::getNullValue(ArgType); |
1437 | Value *ResZero = llvm::Constant::getNullValue(ResultType); |
1438 | Value *ResOne = llvm::ConstantInt::get(ResultType, 1); |
1439 | |
1440 | BasicBlock *Begin = Builder.GetInsertBlock(); |
1441 | BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn); |
1442 | Builder.SetInsertPoint(End); |
1443 | PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result"); |
1444 | |
1445 | Builder.SetInsertPoint(Begin); |
1446 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero); |
1447 | BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn); |
1448 | Builder.CreateCondBr(IsZero, End, NotZero); |
1449 | Result->addIncoming(ResZero, Begin); |
1450 | |
1451 | Builder.SetInsertPoint(NotZero); |
1452 | |
1453 | if (BuiltinID == MSVCIntrin::_BitScanForward) { |
1454 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
1455 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); |
1456 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); |
1457 | Builder.CreateStore(ZeroCount, IndexAddress, false); |
1458 | } else { |
1459 | unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth(); |
1460 | Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1); |
1461 | |
1462 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
1463 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); |
1464 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); |
1465 | Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount); |
1466 | Builder.CreateStore(Index, IndexAddress, false); |
1467 | } |
1468 | Builder.CreateBr(End); |
1469 | Result->addIncoming(ResOne, NotZero); |
1470 | |
1471 | Builder.SetInsertPoint(End); |
1472 | return Result; |
1473 | } |
1474 | case MSVCIntrin::_InterlockedAnd: |
1475 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E); |
1476 | case MSVCIntrin::_InterlockedExchange: |
1477 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E); |
1478 | case MSVCIntrin::_InterlockedExchangeAdd: |
1479 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E); |
1480 | case MSVCIntrin::_InterlockedExchangeSub: |
1481 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E); |
1482 | case MSVCIntrin::_InterlockedOr: |
1483 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E); |
1484 | case MSVCIntrin::_InterlockedXor: |
1485 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E); |
1486 | case MSVCIntrin::_InterlockedExchangeAdd_acq: |
1487 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
1488 | AtomicOrdering::Acquire); |
1489 | case MSVCIntrin::_InterlockedExchangeAdd_rel: |
1490 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
1491 | AtomicOrdering::Release); |
1492 | case MSVCIntrin::_InterlockedExchangeAdd_nf: |
1493 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
1494 | AtomicOrdering::Monotonic); |
1495 | case MSVCIntrin::_InterlockedExchange_acq: |
1496 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
1497 | AtomicOrdering::Acquire); |
1498 | case MSVCIntrin::_InterlockedExchange_rel: |
1499 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
1500 | AtomicOrdering::Release); |
1501 | case MSVCIntrin::_InterlockedExchange_nf: |
1502 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
1503 | AtomicOrdering::Monotonic); |
1504 | case MSVCIntrin::_InterlockedCompareExchange_acq: |
1505 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Acquire); |
1506 | case MSVCIntrin::_InterlockedCompareExchange_rel: |
1507 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Release); |
1508 | case MSVCIntrin::_InterlockedCompareExchange_nf: |
1509 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Monotonic); |
1510 | case MSVCIntrin::_InterlockedCompareExchange128: |
1511 | return EmitAtomicCmpXchg128ForMSIntrin( |
1512 | *this, E, AtomicOrdering::SequentiallyConsistent); |
1513 | case MSVCIntrin::_InterlockedCompareExchange128_acq: |
1514 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Acquire); |
1515 | case MSVCIntrin::_InterlockedCompareExchange128_rel: |
1516 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Release); |
1517 | case MSVCIntrin::_InterlockedCompareExchange128_nf: |
1518 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Monotonic); |
1519 | case MSVCIntrin::_InterlockedOr_acq: |
1520 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
1521 | AtomicOrdering::Acquire); |
1522 | case MSVCIntrin::_InterlockedOr_rel: |
1523 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
1524 | AtomicOrdering::Release); |
1525 | case MSVCIntrin::_InterlockedOr_nf: |
1526 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
1527 | AtomicOrdering::Monotonic); |
1528 | case MSVCIntrin::_InterlockedXor_acq: |
1529 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
1530 | AtomicOrdering::Acquire); |
1531 | case MSVCIntrin::_InterlockedXor_rel: |
1532 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
1533 | AtomicOrdering::Release); |
1534 | case MSVCIntrin::_InterlockedXor_nf: |
1535 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
1536 | AtomicOrdering::Monotonic); |
1537 | case MSVCIntrin::_InterlockedAnd_acq: |
1538 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
1539 | AtomicOrdering::Acquire); |
1540 | case MSVCIntrin::_InterlockedAnd_rel: |
1541 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
1542 | AtomicOrdering::Release); |
1543 | case MSVCIntrin::_InterlockedAnd_nf: |
1544 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
1545 | AtomicOrdering::Monotonic); |
1546 | case MSVCIntrin::_InterlockedIncrement_acq: |
1547 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire); |
1548 | case MSVCIntrin::_InterlockedIncrement_rel: |
1549 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release); |
1550 | case MSVCIntrin::_InterlockedIncrement_nf: |
1551 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic); |
1552 | case MSVCIntrin::_InterlockedDecrement_acq: |
1553 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Acquire); |
1554 | case MSVCIntrin::_InterlockedDecrement_rel: |
1555 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Release); |
1556 | case MSVCIntrin::_InterlockedDecrement_nf: |
1557 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Monotonic); |
1558 | |
1559 | case MSVCIntrin::_InterlockedDecrement: |
1560 | return EmitAtomicDecrementValue(*this, E); |
1561 | case MSVCIntrin::_InterlockedIncrement: |
1562 | return EmitAtomicIncrementValue(*this, E); |
1563 | |
1564 | case MSVCIntrin::__fastfail: { |
1565 | // Request immediate process termination from the kernel. The instruction |
1566 | // sequences to do this are documented on MSDN: |
1567 | // https://msdn.microsoft.com/en-us/library/dn774154.aspx |
1568 | llvm::Triple::ArchType ISA = getTarget().getTriple().getArch(); |
1569 | StringRef Asm, Constraints; |
1570 | switch (ISA) { |
1571 | default: |
1572 | ErrorUnsupported(E, "__fastfail call for this architecture"); |
1573 | break; |
1574 | case llvm::Triple::x86: |
1575 | case llvm::Triple::x86_64: |
1576 | Asm = "int $$0x29"; |
1577 | Constraints = "{cx}"; |
1578 | break; |
1579 | case llvm::Triple::thumb: |
1580 | Asm = "udf #251"; |
1581 | Constraints = "{r0}"; |
1582 | break; |
1583 | case llvm::Triple::aarch64: |
1584 | Asm = "brk #0xF003"; |
1585 | Constraints = "{w0}"; |
1586 | } |
1587 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false); |
1588 | llvm::InlineAsm *IA = |
1589 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); |
1590 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( |
1591 | getLLVMContext(), llvm::AttributeList::FunctionIndex, |
1592 | llvm::Attribute::NoReturn); |
1593 | llvm::CallInst *CI = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0))); |
1594 | CI->setAttributes(NoReturnAttr); |
1595 | return CI; |
1596 | } |
1597 | } |
1598 | llvm_unreachable("Incorrect MSVC intrinsic!")::llvm::llvm_unreachable_internal("Incorrect MSVC intrinsic!" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1598); |
1599 | } |
1600 | |
1601 | namespace { |
1602 | // ARC cleanup for __builtin_os_log_format |
1603 | struct CallObjCArcUse final : EHScopeStack::Cleanup { |
1604 | CallObjCArcUse(llvm::Value *object) : object(object) {} |
1605 | llvm::Value *object; |
1606 | |
1607 | void Emit(CodeGenFunction &CGF, Flags flags) override { |
1608 | CGF.EmitARCIntrinsicUse(object); |
1609 | } |
1610 | }; |
1611 | } |
1612 | |
1613 | Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E, |
1614 | BuiltinCheckKind Kind) { |
1615 | assert((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero)(((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && "Unsupported builtin check kind") ? static_cast<void> ( 0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1616, __PRETTY_FUNCTION__)) |
1616 | && "Unsupported builtin check kind")(((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && "Unsupported builtin check kind") ? static_cast<void> ( 0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1616, __PRETTY_FUNCTION__)); |
1617 | |
1618 | Value *ArgValue = EmitScalarExpr(E); |
1619 | if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef()) |
1620 | return ArgValue; |
1621 | |
1622 | SanitizerScope SanScope(this); |
1623 | Value *Cond = Builder.CreateICmpNE( |
1624 | ArgValue, llvm::Constant::getNullValue(ArgValue->getType())); |
1625 | EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin), |
1626 | SanitizerHandler::InvalidBuiltin, |
1627 | {EmitCheckSourceLocation(E->getExprLoc()), |
1628 | llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)}, |
1629 | None); |
1630 | return ArgValue; |
1631 | } |
1632 | |
1633 | /// Get the argument type for arguments to os_log_helper. |
1634 | static CanQualType getOSLogArgType(ASTContext &C, int Size) { |
1635 | QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false); |
1636 | return C.getCanonicalType(UnsignedTy); |
1637 | } |
1638 | |
1639 | llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction( |
1640 | const analyze_os_log::OSLogBufferLayout &Layout, |
1641 | CharUnits BufferAlignment) { |
1642 | ASTContext &Ctx = getContext(); |
1643 | |
1644 | llvm::SmallString<64> Name; |
1645 | { |
1646 | raw_svector_ostream OS(Name); |
1647 | OS << "__os_log_helper"; |
1648 | OS << "_" << BufferAlignment.getQuantity(); |
1649 | OS << "_" << int(Layout.getSummaryByte()); |
1650 | OS << "_" << int(Layout.getNumArgsByte()); |
1651 | for (const auto &Item : Layout.Items) |
1652 | OS << "_" << int(Item.getSizeByte()) << "_" |
1653 | << int(Item.getDescriptorByte()); |
1654 | } |
1655 | |
1656 | if (llvm::Function *F = CGM.getModule().getFunction(Name)) |
1657 | return F; |
1658 | |
1659 | llvm::SmallVector<QualType, 4> ArgTys; |
1660 | FunctionArgList Args; |
1661 | Args.push_back(ImplicitParamDecl::Create( |
1662 | Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"), Ctx.VoidPtrTy, |
1663 | ImplicitParamDecl::Other)); |
1664 | ArgTys.emplace_back(Ctx.VoidPtrTy); |
1665 | |
1666 | for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) { |
1667 | char Size = Layout.Items[I].getSizeByte(); |
1668 | if (!Size) |
1669 | continue; |
1670 | |
1671 | QualType ArgTy = getOSLogArgType(Ctx, Size); |
1672 | Args.push_back(ImplicitParamDecl::Create( |
1673 | Ctx, nullptr, SourceLocation(), |
1674 | &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)), ArgTy, |
1675 | ImplicitParamDecl::Other)); |
1676 | ArgTys.emplace_back(ArgTy); |
1677 | } |
1678 | |
1679 | QualType ReturnTy = Ctx.VoidTy; |
1680 | QualType FuncionTy = Ctx.getFunctionType(ReturnTy, ArgTys, {}); |
1681 | |
1682 | // The helper function has linkonce_odr linkage to enable the linker to merge |
1683 | // identical functions. To ensure the merging always happens, 'noinline' is |
1684 | // attached to the function when compiling with -Oz. |
1685 | const CGFunctionInfo &FI = |
1686 | CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, Args); |
1687 | llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI); |
1688 | llvm::Function *Fn = llvm::Function::Create( |
1689 | FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule()); |
1690 | Fn->setVisibility(llvm::GlobalValue::HiddenVisibility); |
1691 | CGM.SetLLVMFunctionAttributes(GlobalDecl(), FI, Fn); |
1692 | CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn); |
1693 | Fn->setDoesNotThrow(); |
1694 | |
1695 | // Attach 'noinline' at -Oz. |
1696 | if (CGM.getCodeGenOpts().OptimizeSize == 2) |
1697 | Fn->addFnAttr(llvm::Attribute::NoInline); |
1698 | |
1699 | auto NL = ApplyDebugLocation::CreateEmpty(*this); |
1700 | IdentifierInfo *II = &Ctx.Idents.get(Name); |
1701 | FunctionDecl *FD = FunctionDecl::Create( |
1702 | Ctx, Ctx.getTranslationUnitDecl(), SourceLocation(), SourceLocation(), II, |
1703 | FuncionTy, nullptr, SC_PrivateExtern, false, false); |
1704 | // Avoid generating debug location info for the function. |
1705 | FD->setImplicit(); |
1706 | |
1707 | StartFunction(FD, ReturnTy, Fn, FI, Args); |
1708 | |
1709 | // Create a scope with an artificial location for the body of this function. |
1710 | auto AL = ApplyDebugLocation::CreateArtificial(*this); |
1711 | |
1712 | CharUnits Offset; |
1713 | Address BufAddr(Builder.CreateLoad(GetAddrOfLocalVar(Args[0]), "buf"), |
1714 | BufferAlignment); |
1715 | Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()), |
1716 | Builder.CreateConstByteGEP(BufAddr, Offset++, "summary")); |
1717 | Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()), |
1718 | Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs")); |
1719 | |
1720 | unsigned I = 1; |
1721 | for (const auto &Item : Layout.Items) { |
1722 | Builder.CreateStore( |
1723 | Builder.getInt8(Item.getDescriptorByte()), |
1724 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor")); |
1725 | Builder.CreateStore( |
1726 | Builder.getInt8(Item.getSizeByte()), |
1727 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize")); |
1728 | |
1729 | CharUnits Size = Item.size(); |
1730 | if (!Size.getQuantity()) |
1731 | continue; |
1732 | |
1733 | Address Arg = GetAddrOfLocalVar(Args[I]); |
1734 | Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData"); |
1735 | Addr = Builder.CreateBitCast(Addr, Arg.getPointer()->getType(), |
1736 | "argDataCast"); |
1737 | Builder.CreateStore(Builder.CreateLoad(Arg), Addr); |
1738 | Offset += Size; |
1739 | ++I; |
1740 | } |
1741 | |
1742 | FinishFunction(); |
1743 | |
1744 | return Fn; |
1745 | } |
1746 | |
1747 | RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) { |
1748 | assert(E.getNumArgs() >= 2 &&((E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments" ) ? static_cast<void> (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1749, __PRETTY_FUNCTION__)) |
1749 | "__builtin_os_log_format takes at least 2 arguments")((E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments" ) ? static_cast<void> (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1749, __PRETTY_FUNCTION__)); |
1750 | ASTContext &Ctx = getContext(); |
1751 | analyze_os_log::OSLogBufferLayout Layout; |
1752 | analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout); |
1753 | Address BufAddr = EmitPointerWithAlignment(E.getArg(0)); |
1754 | llvm::SmallVector<llvm::Value *, 4> RetainableOperands; |
1755 | |
1756 | // Ignore argument 1, the format string. It is not currently used. |
1757 | CallArgList Args; |
1758 | Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy); |
1759 | |
1760 | for (const auto &Item : Layout.Items) { |
1761 | int Size = Item.getSizeByte(); |
1762 | if (!Size) |
1763 | continue; |
1764 | |
1765 | llvm::Value *ArgVal; |
1766 | |
1767 | if (Item.getKind() == analyze_os_log::OSLogBufferItem::MaskKind) { |
1768 | uint64_t Val = 0; |
1769 | for (unsigned I = 0, E = Item.getMaskType().size(); I < E; ++I) |
1770 | Val |= ((uint64_t)Item.getMaskType()[I]) << I * 8; |
1771 | ArgVal = llvm::Constant::getIntegerValue(Int64Ty, llvm::APInt(64, Val)); |
1772 | } else if (const Expr *TheExpr = Item.getExpr()) { |
1773 | ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false); |
1774 | |
1775 | // If a temporary object that requires destruction after the full |
1776 | // expression is passed, push a lifetime-extended cleanup to extend its |
1777 | // lifetime to the end of the enclosing block scope. |
1778 | auto LifetimeExtendObject = [&](const Expr *E) { |
1779 | E = E->IgnoreParenCasts(); |
1780 | // Extend lifetimes of objects returned by function calls and message |
1781 | // sends. |
1782 | |
1783 | // FIXME: We should do this in other cases in which temporaries are |
1784 | // created including arguments of non-ARC types (e.g., C++ |
1785 | // temporaries). |
1786 | if (isa<CallExpr>(E) || isa<ObjCMessageExpr>(E)) |
1787 | return true; |
1788 | return false; |
1789 | }; |
1790 | |
1791 | if (TheExpr->getType()->isObjCRetainableType() && |
1792 | getLangOpts().ObjCAutoRefCount && LifetimeExtendObject(TheExpr)) { |
1793 | assert(getEvaluationKind(TheExpr->getType()) == TEK_Scalar &&((getEvaluationKind(TheExpr->getType()) == TEK_Scalar && "Only scalar can be a ObjC retainable type") ? static_cast< void> (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1794, __PRETTY_FUNCTION__)) |
1794 | "Only scalar can be a ObjC retainable type")((getEvaluationKind(TheExpr->getType()) == TEK_Scalar && "Only scalar can be a ObjC retainable type") ? static_cast< void> (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1794, __PRETTY_FUNCTION__)); |
1795 | if (!isa<Constant>(ArgVal)) { |
1796 | CleanupKind Cleanup = getARCCleanupKind(); |
1797 | QualType Ty = TheExpr->getType(); |
1798 | Address Alloca = Address::invalid(); |
1799 | Address Addr = CreateMemTemp(Ty, "os.log.arg", &Alloca); |
1800 | ArgVal = EmitARCRetain(Ty, ArgVal); |
1801 | Builder.CreateStore(ArgVal, Addr); |
1802 | pushLifetimeExtendedDestroy(Cleanup, Alloca, Ty, |
1803 | CodeGenFunction::destroyARCStrongPrecise, |
1804 | Cleanup & EHCleanup); |
1805 | |
1806 | // Push a clang.arc.use call to ensure ARC optimizer knows that the |
1807 | // argument has to be alive. |
1808 | if (CGM.getCodeGenOpts().OptimizationLevel != 0) |
1809 | pushCleanupAfterFullExpr<CallObjCArcUse>(Cleanup, ArgVal); |
1810 | } |
1811 | } |
1812 | } else { |
1813 | ArgVal = Builder.getInt32(Item.getConstValue().getQuantity()); |
1814 | } |
1815 | |
1816 | unsigned ArgValSize = |
1817 | CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType()); |
1818 | llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(), |
1819 | ArgValSize); |
1820 | ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy); |
1821 | CanQualType ArgTy = getOSLogArgType(Ctx, Size); |
1822 | // If ArgVal has type x86_fp80, zero-extend ArgVal. |
1823 | ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy)); |
1824 | Args.add(RValue::get(ArgVal), ArgTy); |
1825 | } |
1826 | |
1827 | const CGFunctionInfo &FI = |
1828 | CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args); |
1829 | llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction( |
1830 | Layout, BufAddr.getAlignment()); |
1831 | EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args); |
1832 | return RValue::get(BufAddr.getPointer()); |
1833 | } |
1834 | |
1835 | static bool isSpecialUnsignedMultiplySignedResult( |
1836 | unsigned BuiltinID, WidthAndSignedness Op1Info, WidthAndSignedness Op2Info, |
1837 | WidthAndSignedness ResultInfo) { |
1838 | return BuiltinID == Builtin::BI__builtin_mul_overflow && |
1839 | Op1Info.Width == Op2Info.Width && Op2Info.Width == ResultInfo.Width && |
1840 | !Op1Info.Signed && !Op2Info.Signed && ResultInfo.Signed; |
1841 | } |
1842 | |
1843 | static RValue EmitCheckedUnsignedMultiplySignedResult( |
1844 | CodeGenFunction &CGF, const clang::Expr *Op1, WidthAndSignedness Op1Info, |
1845 | const clang::Expr *Op2, WidthAndSignedness Op2Info, |
1846 | const clang::Expr *ResultArg, QualType ResultQTy, |
1847 | WidthAndSignedness ResultInfo) { |
1848 | assert(isSpecialUnsignedMultiplySignedResult(((isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Cannot specialize this multiply" ) ? static_cast<void> (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1850, __PRETTY_FUNCTION__)) |
1849 | Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) &&((isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Cannot specialize this multiply" ) ? static_cast<void> (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1850, __PRETTY_FUNCTION__)) |
1850 | "Cannot specialize this multiply")((isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Cannot specialize this multiply" ) ? static_cast<void> (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1850, __PRETTY_FUNCTION__)); |
1851 | |
1852 | llvm::Value *V1 = CGF.EmitScalarExpr(Op1); |
1853 | llvm::Value *V2 = CGF.EmitScalarExpr(Op2); |
1854 | |
1855 | llvm::Value *HasOverflow; |
1856 | llvm::Value *Result = EmitOverflowIntrinsic( |
1857 | CGF, llvm::Intrinsic::umul_with_overflow, V1, V2, HasOverflow); |
1858 | |
1859 | // The intrinsic call will detect overflow when the value is > UINT_MAX, |
1860 | // however, since the original builtin had a signed result, we need to report |
1861 | // an overflow when the result is greater than INT_MAX. |
1862 | auto IntMax = llvm::APInt::getSignedMaxValue(ResultInfo.Width); |
1863 | llvm::Value *IntMaxValue = llvm::ConstantInt::get(Result->getType(), IntMax); |
1864 | |
1865 | llvm::Value *IntMaxOverflow = CGF.Builder.CreateICmpUGT(Result, IntMaxValue); |
1866 | HasOverflow = CGF.Builder.CreateOr(HasOverflow, IntMaxOverflow); |
1867 | |
1868 | bool isVolatile = |
1869 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
1870 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); |
1871 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, |
1872 | isVolatile); |
1873 | return RValue::get(HasOverflow); |
1874 | } |
1875 | |
1876 | /// Determine if a binop is a checked mixed-sign multiply we can specialize. |
1877 | static bool isSpecialMixedSignMultiply(unsigned BuiltinID, |
1878 | WidthAndSignedness Op1Info, |
1879 | WidthAndSignedness Op2Info, |
1880 | WidthAndSignedness ResultInfo) { |
1881 | return BuiltinID == Builtin::BI__builtin_mul_overflow && |
1882 | std::max(Op1Info.Width, Op2Info.Width) >= ResultInfo.Width && |
1883 | Op1Info.Signed != Op2Info.Signed; |
1884 | } |
1885 | |
1886 | /// Emit a checked mixed-sign multiply. This is a cheaper specialization of |
1887 | /// the generic checked-binop irgen. |
1888 | static RValue |
1889 | EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1, |
1890 | WidthAndSignedness Op1Info, const clang::Expr *Op2, |
1891 | WidthAndSignedness Op2Info, |
1892 | const clang::Expr *ResultArg, QualType ResultQTy, |
1893 | WidthAndSignedness ResultInfo) { |
1894 | assert(isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info,((isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize" ) ? static_cast<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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1896, __PRETTY_FUNCTION__)) |
1895 | Op2Info, ResultInfo) &&((isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize" ) ? static_cast<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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1896, __PRETTY_FUNCTION__)) |
1896 | "Not a mixed-sign multipliction we can specialize")((isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize" ) ? static_cast<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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1896, __PRETTY_FUNCTION__)); |
1897 | |
1898 | // Emit the signed and unsigned operands. |
1899 | const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2; |
1900 | const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1; |
1901 | llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp); |
1902 | llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp); |
1903 | unsigned SignedOpWidth = Op1Info.Signed ? Op1Info.Width : Op2Info.Width; |
1904 | unsigned UnsignedOpWidth = Op1Info.Signed ? Op2Info.Width : Op1Info.Width; |
1905 | |
1906 | // One of the operands may be smaller than the other. If so, [s|z]ext it. |
1907 | if (SignedOpWidth < UnsignedOpWidth) |
1908 | Signed = CGF.Builder.CreateSExt(Signed, Unsigned->getType(), "op.sext"); |
1909 | if (UnsignedOpWidth < SignedOpWidth) |
1910 | Unsigned = CGF.Builder.CreateZExt(Unsigned, Signed->getType(), "op.zext"); |
1911 | |
1912 | llvm::Type *OpTy = Signed->getType(); |
1913 | llvm::Value *Zero = llvm::Constant::getNullValue(OpTy); |
1914 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); |
1915 | llvm::Type *ResTy = ResultPtr.getElementType(); |
1916 | unsigned OpWidth = std::max(Op1Info.Width, Op2Info.Width); |
1917 | |
1918 | // Take the absolute value of the signed operand. |
1919 | llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero); |
1920 | llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed); |
1921 | llvm::Value *AbsSigned = |
1922 | CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed); |
1923 | |
1924 | // Perform a checked unsigned multiplication. |
1925 | llvm::Value *UnsignedOverflow; |
1926 | llvm::Value *UnsignedResult = |
1927 | EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned, |
1928 | Unsigned, UnsignedOverflow); |
1929 | |
1930 | llvm::Value *Overflow, *Result; |
1931 | if (ResultInfo.Signed) { |
1932 | // Signed overflow occurs if the result is greater than INT_MAX or lesser |
1933 | // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative). |
1934 | auto IntMax = |
1935 | llvm::APInt::getSignedMaxValue(ResultInfo.Width).zextOrSelf(OpWidth); |
1936 | llvm::Value *MaxResult = |
1937 | CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax), |
1938 | CGF.Builder.CreateZExt(IsNegative, OpTy)); |
1939 | llvm::Value *SignedOverflow = |
1940 | CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult); |
1941 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow); |
1942 | |
1943 | // Prepare the signed result (possibly by negating it). |
1944 | llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult); |
1945 | llvm::Value *SignedResult = |
1946 | CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult); |
1947 | Result = CGF.Builder.CreateTrunc(SignedResult, ResTy); |
1948 | } else { |
1949 | // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX. |
1950 | llvm::Value *Underflow = CGF.Builder.CreateAnd( |
1951 | IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult)); |
1952 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow); |
1953 | if (ResultInfo.Width < OpWidth) { |
1954 | auto IntMax = |
1955 | llvm::APInt::getMaxValue(ResultInfo.Width).zext(OpWidth); |
1956 | llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT( |
1957 | UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax)); |
1958 | Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow); |
1959 | } |
1960 | |
1961 | // Negate the product if it would be negative in infinite precision. |
1962 | Result = CGF.Builder.CreateSelect( |
1963 | IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult); |
1964 | |
1965 | Result = CGF.Builder.CreateTrunc(Result, ResTy); |
1966 | } |
1967 | assert(Overflow && Result && "Missing overflow or result")((Overflow && Result && "Missing overflow or result" ) ? static_cast<void> (0) : __assert_fail ("Overflow && Result && \"Missing overflow or result\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 1967, __PRETTY_FUNCTION__)); |
1968 | |
1969 | bool isVolatile = |
1970 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
1971 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, |
1972 | isVolatile); |
1973 | return RValue::get(Overflow); |
1974 | } |
1975 | |
1976 | static llvm::Value *dumpRecord(CodeGenFunction &CGF, QualType RType, |
1977 | Value *&RecordPtr, CharUnits Align, |
1978 | llvm::FunctionCallee Func, int Lvl) { |
1979 | ASTContext &Context = CGF.getContext(); |
1980 | RecordDecl *RD = RType->castAs<RecordType>()->getDecl()->getDefinition(); |
1981 | std::string Pad = std::string(Lvl * 4, ' '); |
1982 | |
1983 | Value *GString = |
1984 | CGF.Builder.CreateGlobalStringPtr(RType.getAsString() + " {\n"); |
1985 | Value *Res = CGF.Builder.CreateCall(Func, {GString}); |
1986 | |
1987 | static llvm::DenseMap<QualType, const char *> Types; |
1988 | if (Types.empty()) { |
1989 | Types[Context.CharTy] = "%c"; |
1990 | Types[Context.BoolTy] = "%d"; |
1991 | Types[Context.SignedCharTy] = "%hhd"; |
1992 | Types[Context.UnsignedCharTy] = "%hhu"; |
1993 | Types[Context.IntTy] = "%d"; |
1994 | Types[Context.UnsignedIntTy] = "%u"; |
1995 | Types[Context.LongTy] = "%ld"; |
1996 | Types[Context.UnsignedLongTy] = "%lu"; |
1997 | Types[Context.LongLongTy] = "%lld"; |
1998 | Types[Context.UnsignedLongLongTy] = "%llu"; |
1999 | Types[Context.ShortTy] = "%hd"; |
2000 | Types[Context.UnsignedShortTy] = "%hu"; |
2001 | Types[Context.VoidPtrTy] = "%p"; |
2002 | Types[Context.FloatTy] = "%f"; |
2003 | Types[Context.DoubleTy] = "%f"; |
2004 | Types[Context.LongDoubleTy] = "%Lf"; |
2005 | Types[Context.getPointerType(Context.CharTy)] = "%s"; |
2006 | Types[Context.getPointerType(Context.getConstType(Context.CharTy))] = "%s"; |
2007 | } |
2008 | |
2009 | for (const auto *FD : RD->fields()) { |
2010 | Value *FieldPtr = RecordPtr; |
2011 | if (RD->isUnion()) |
2012 | FieldPtr = CGF.Builder.CreatePointerCast( |
2013 | FieldPtr, CGF.ConvertType(Context.getPointerType(FD->getType()))); |
2014 | else |
2015 | FieldPtr = CGF.Builder.CreateStructGEP(CGF.ConvertType(RType), FieldPtr, |
2016 | FD->getFieldIndex()); |
2017 | |
2018 | GString = CGF.Builder.CreateGlobalStringPtr( |
2019 | llvm::Twine(Pad) |
2020 | .concat(FD->getType().getAsString()) |
2021 | .concat(llvm::Twine(' ')) |
2022 | .concat(FD->getNameAsString()) |
2023 | .concat(" : ") |
2024 | .str()); |
2025 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); |
2026 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
2027 | |
2028 | QualType CanonicalType = |
2029 | FD->getType().getUnqualifiedType().getCanonicalType(); |
2030 | |
2031 | // We check whether we are in a recursive type |
2032 | if (CanonicalType->isRecordType()) { |
2033 | TmpRes = dumpRecord(CGF, CanonicalType, FieldPtr, Align, Func, Lvl + 1); |
2034 | Res = CGF.Builder.CreateAdd(TmpRes, Res); |
2035 | continue; |
2036 | } |
2037 | |
2038 | // We try to determine the best format to print the current field |
2039 | llvm::Twine Format = Types.find(CanonicalType) == Types.end() |
2040 | ? Types[Context.VoidPtrTy] |
2041 | : Types[CanonicalType]; |
2042 | |
2043 | Address FieldAddress = Address(FieldPtr, Align); |
2044 | FieldPtr = CGF.Builder.CreateLoad(FieldAddress); |
2045 | |
2046 | // FIXME Need to handle bitfield here |
2047 | GString = CGF.Builder.CreateGlobalStringPtr( |
2048 | Format.concat(llvm::Twine('\n')).str()); |
2049 | TmpRes = CGF.Builder.CreateCall(Func, {GString, FieldPtr}); |
2050 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
2051 | } |
2052 | |
2053 | GString = CGF.Builder.CreateGlobalStringPtr(Pad + "}\n"); |
2054 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); |
2055 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
2056 | return Res; |
2057 | } |
2058 | |
2059 | static bool |
2060 | TypeRequiresBuiltinLaunderImp(const ASTContext &Ctx, QualType Ty, |
2061 | llvm::SmallPtrSetImpl<const Decl *> &Seen) { |
2062 | if (const auto *Arr = Ctx.getAsArrayType(Ty)) |
2063 | Ty = Ctx.getBaseElementType(Arr); |
2064 | |
2065 | const auto *Record = Ty->getAsCXXRecordDecl(); |
2066 | if (!Record) |
2067 | return false; |
2068 | |
2069 | // We've already checked this type, or are in the process of checking it. |
2070 | if (!Seen.insert(Record).second) |
2071 | return false; |
2072 | |
2073 | assert(Record->hasDefinition() &&((Record->hasDefinition() && "Incomplete types should already be diagnosed" ) ? static_cast<void> (0) : __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 2074, __PRETTY_FUNCTION__)) |
2074 | "Incomplete types should already be diagnosed")((Record->hasDefinition() && "Incomplete types should already be diagnosed" ) ? static_cast<void> (0) : __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 2074, __PRETTY_FUNCTION__)); |
2075 | |
2076 | if (Record->isDynamicClass()) |
2077 | return true; |
2078 | |
2079 | for (FieldDecl *F : Record->fields()) { |
2080 | if (TypeRequiresBuiltinLaunderImp(Ctx, F->getType(), Seen)) |
2081 | return true; |
2082 | } |
2083 | return false; |
2084 | } |
2085 | |
2086 | /// Determine if the specified type requires laundering by checking if it is a |
2087 | /// dynamic class type or contains a subobject which is a dynamic class type. |
2088 | static bool TypeRequiresBuiltinLaunder(CodeGenModule &CGM, QualType Ty) { |
2089 | if (!CGM.getCodeGenOpts().StrictVTablePointers) |
2090 | return false; |
2091 | llvm::SmallPtrSet<const Decl *, 16> Seen; |
2092 | return TypeRequiresBuiltinLaunderImp(CGM.getContext(), Ty, Seen); |
2093 | } |
2094 | |
2095 | RValue CodeGenFunction::emitRotate(const CallExpr *E, bool IsRotateRight) { |
2096 | llvm::Value *Src = EmitScalarExpr(E->getArg(0)); |
2097 | llvm::Value *ShiftAmt = EmitScalarExpr(E->getArg(1)); |
2098 | |
2099 | // The builtin's shift arg may have a different type than the source arg and |
2100 | // result, but the LLVM intrinsic uses the same type for all values. |
2101 | llvm::Type *Ty = Src->getType(); |
2102 | ShiftAmt = Builder.CreateIntCast(ShiftAmt, Ty, false); |
2103 | |
2104 | // Rotate is a special case of LLVM funnel shift - 1st 2 args are the same. |
2105 | unsigned IID = IsRotateRight ? Intrinsic::fshr : Intrinsic::fshl; |
2106 | Function *F = CGM.getIntrinsic(IID, Ty); |
2107 | return RValue::get(Builder.CreateCall(F, { Src, Src, ShiftAmt })); |
2108 | } |
2109 | |
2110 | // Map math builtins for long-double to f128 version. |
2111 | static unsigned mutateLongDoubleBuiltin(unsigned BuiltinID) { |
2112 | switch (BuiltinID) { |
2113 | #define MUTATE_LDBL(func) \ |
2114 | case Builtin::BI__builtin_##func##l: \ |
2115 | return Builtin::BI__builtin_##func##f128; |
2116 | MUTATE_LDBL(sqrt) |
2117 | MUTATE_LDBL(cbrt) |
2118 | MUTATE_LDBL(fabs) |
2119 | MUTATE_LDBL(log) |
2120 | MUTATE_LDBL(log2) |
2121 | MUTATE_LDBL(log10) |
2122 | MUTATE_LDBL(log1p) |
2123 | MUTATE_LDBL(logb) |
2124 | MUTATE_LDBL(exp) |
2125 | MUTATE_LDBL(exp2) |
2126 | MUTATE_LDBL(expm1) |
2127 | MUTATE_LDBL(fdim) |
2128 | MUTATE_LDBL(hypot) |
2129 | MUTATE_LDBL(ilogb) |
2130 | MUTATE_LDBL(pow) |
2131 | MUTATE_LDBL(fmin) |
2132 | MUTATE_LDBL(fmax) |
2133 | MUTATE_LDBL(ceil) |
2134 | MUTATE_LDBL(trunc) |
2135 | MUTATE_LDBL(rint) |
2136 | MUTATE_LDBL(nearbyint) |
2137 | MUTATE_LDBL(round) |
2138 | MUTATE_LDBL(floor) |
2139 | MUTATE_LDBL(lround) |
2140 | MUTATE_LDBL(llround) |
2141 | MUTATE_LDBL(lrint) |
2142 | MUTATE_LDBL(llrint) |
2143 | MUTATE_LDBL(fmod) |
2144 | MUTATE_LDBL(modf) |
2145 | MUTATE_LDBL(nan) |
2146 | MUTATE_LDBL(nans) |
2147 | MUTATE_LDBL(inf) |
2148 | MUTATE_LDBL(fma) |
2149 | MUTATE_LDBL(sin) |
2150 | MUTATE_LDBL(cos) |
2151 | MUTATE_LDBL(tan) |
2152 | MUTATE_LDBL(sinh) |
2153 | MUTATE_LDBL(cosh) |
2154 | MUTATE_LDBL(tanh) |
2155 | MUTATE_LDBL(asin) |
2156 | MUTATE_LDBL(acos) |
2157 | MUTATE_LDBL(atan) |
2158 | MUTATE_LDBL(asinh) |
2159 | MUTATE_LDBL(acosh) |
2160 | MUTATE_LDBL(atanh) |
2161 | MUTATE_LDBL(atan2) |
2162 | MUTATE_LDBL(erf) |
2163 | MUTATE_LDBL(erfc) |
2164 | MUTATE_LDBL(ldexp) |
2165 | MUTATE_LDBL(frexp) |
2166 | MUTATE_LDBL(huge_val) |
2167 | MUTATE_LDBL(copysign) |
2168 | MUTATE_LDBL(nextafter) |
2169 | MUTATE_LDBL(nexttoward) |
2170 | MUTATE_LDBL(remainder) |
2171 | MUTATE_LDBL(remquo) |
2172 | MUTATE_LDBL(scalbln) |
2173 | MUTATE_LDBL(scalbn) |
2174 | MUTATE_LDBL(tgamma) |
2175 | MUTATE_LDBL(lgamma) |
2176 | #undef MUTATE_LDBL |
2177 | default: |
2178 | return BuiltinID; |
2179 | } |
2180 | } |
2181 | |
2182 | RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, |
2183 | const CallExpr *E, |
2184 | ReturnValueSlot ReturnValue) { |
2185 | const FunctionDecl *FD = GD.getDecl()->getAsFunction(); |
2186 | // See if we can constant fold this builtin. If so, don't emit it at all. |
2187 | Expr::EvalResult Result; |
2188 | if (E->EvaluateAsRValue(Result, CGM.getContext()) && |
2189 | !Result.hasSideEffects()) { |
2190 | if (Result.Val.isInt()) |
2191 | return RValue::get(llvm::ConstantInt::get(getLLVMContext(), |
2192 | Result.Val.getInt())); |
2193 | if (Result.Val.isFloat()) |
2194 | return RValue::get(llvm::ConstantFP::get(getLLVMContext(), |
2195 | Result.Val.getFloat())); |
2196 | } |
2197 | |
2198 | // If current long-double semantics is IEEE 128-bit, replace math builtins |
2199 | // of long-double with f128 equivalent. |
2200 | // TODO: This mutation should also be applied to other targets other than PPC, |
2201 | // after backend supports IEEE 128-bit style libcalls. |
2202 | if (getTarget().getTriple().isPPC64() && |
2203 | &getTarget().getLongDoubleFormat() == &llvm::APFloat::IEEEquad()) |
2204 | BuiltinID = mutateLongDoubleBuiltin(BuiltinID); |
2205 | |
2206 | // If the builtin has been declared explicitly with an assembler label, |
2207 | // disable the specialized emitting below. Ideally we should communicate the |
2208 | // rename in IR, or at least avoid generating the intrinsic calls that are |
2209 | // likely to get lowered to the renamed library functions. |
2210 | const unsigned BuiltinIDIfNoAsmLabel = |
2211 | FD->hasAttr<AsmLabelAttr>() ? 0 : BuiltinID; |
2212 | |
2213 | // There are LLVM math intrinsics/instructions corresponding to math library |
2214 | // functions except the LLVM op will never set errno while the math library |
2215 | // might. Also, math builtins have the same semantics as their math library |
2216 | // twins. Thus, we can transform math library and builtin calls to their |
2217 | // LLVM counterparts if the call is marked 'const' (known to never set errno). |
2218 | if (FD->hasAttr<ConstAttr>()) { |
2219 | switch (BuiltinIDIfNoAsmLabel) { |
2220 | case Builtin::BIceil: |
2221 | case Builtin::BIceilf: |
2222 | case Builtin::BIceill: |
2223 | case Builtin::BI__builtin_ceil: |
2224 | case Builtin::BI__builtin_ceilf: |
2225 | case Builtin::BI__builtin_ceilf16: |
2226 | case Builtin::BI__builtin_ceill: |
2227 | case Builtin::BI__builtin_ceilf128: |
2228 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2229 | Intrinsic::ceil, |
2230 | Intrinsic::experimental_constrained_ceil)); |
2231 | |
2232 | case Builtin::BIcopysign: |
2233 | case Builtin::BIcopysignf: |
2234 | case Builtin::BIcopysignl: |
2235 | case Builtin::BI__builtin_copysign: |
2236 | case Builtin::BI__builtin_copysignf: |
2237 | case Builtin::BI__builtin_copysignf16: |
2238 | case Builtin::BI__builtin_copysignl: |
2239 | case Builtin::BI__builtin_copysignf128: |
2240 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign)); |
2241 | |
2242 | case Builtin::BIcos: |
2243 | case Builtin::BIcosf: |
2244 | case Builtin::BIcosl: |
2245 | case Builtin::BI__builtin_cos: |
2246 | case Builtin::BI__builtin_cosf: |
2247 | case Builtin::BI__builtin_cosf16: |
2248 | case Builtin::BI__builtin_cosl: |
2249 | case Builtin::BI__builtin_cosf128: |
2250 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2251 | Intrinsic::cos, |
2252 | Intrinsic::experimental_constrained_cos)); |
2253 | |
2254 | case Builtin::BIexp: |
2255 | case Builtin::BIexpf: |
2256 | case Builtin::BIexpl: |
2257 | case Builtin::BI__builtin_exp: |
2258 | case Builtin::BI__builtin_expf: |
2259 | case Builtin::BI__builtin_expf16: |
2260 | case Builtin::BI__builtin_expl: |
2261 | case Builtin::BI__builtin_expf128: |
2262 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2263 | Intrinsic::exp, |
2264 | Intrinsic::experimental_constrained_exp)); |
2265 | |
2266 | case Builtin::BIexp2: |
2267 | case Builtin::BIexp2f: |
2268 | case Builtin::BIexp2l: |
2269 | case Builtin::BI__builtin_exp2: |
2270 | case Builtin::BI__builtin_exp2f: |
2271 | case Builtin::BI__builtin_exp2f16: |
2272 | case Builtin::BI__builtin_exp2l: |
2273 | case Builtin::BI__builtin_exp2f128: |
2274 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2275 | Intrinsic::exp2, |
2276 | Intrinsic::experimental_constrained_exp2)); |
2277 | |
2278 | case Builtin::BIfabs: |
2279 | case Builtin::BIfabsf: |
2280 | case Builtin::BIfabsl: |
2281 | case Builtin::BI__builtin_fabs: |
2282 | case Builtin::BI__builtin_fabsf: |
2283 | case Builtin::BI__builtin_fabsf16: |
2284 | case Builtin::BI__builtin_fabsl: |
2285 | case Builtin::BI__builtin_fabsf128: |
2286 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs)); |
2287 | |
2288 | case Builtin::BIfloor: |
2289 | case Builtin::BIfloorf: |
2290 | case Builtin::BIfloorl: |
2291 | case Builtin::BI__builtin_floor: |
2292 | case Builtin::BI__builtin_floorf: |
2293 | case Builtin::BI__builtin_floorf16: |
2294 | case Builtin::BI__builtin_floorl: |
2295 | case Builtin::BI__builtin_floorf128: |
2296 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2297 | Intrinsic::floor, |
2298 | Intrinsic::experimental_constrained_floor)); |
2299 | |
2300 | case Builtin::BIfma: |
2301 | case Builtin::BIfmaf: |
2302 | case Builtin::BIfmal: |
2303 | case Builtin::BI__builtin_fma: |
2304 | case Builtin::BI__builtin_fmaf: |
2305 | case Builtin::BI__builtin_fmaf16: |
2306 | case Builtin::BI__builtin_fmal: |
2307 | case Builtin::BI__builtin_fmaf128: |
2308 | return RValue::get(emitTernaryMaybeConstrainedFPBuiltin(*this, E, |
2309 | Intrinsic::fma, |
2310 | Intrinsic::experimental_constrained_fma)); |
2311 | |
2312 | case Builtin::BIfmax: |
2313 | case Builtin::BIfmaxf: |
2314 | case Builtin::BIfmaxl: |
2315 | case Builtin::BI__builtin_fmax: |
2316 | case Builtin::BI__builtin_fmaxf: |
2317 | case Builtin::BI__builtin_fmaxf16: |
2318 | case Builtin::BI__builtin_fmaxl: |
2319 | case Builtin::BI__builtin_fmaxf128: |
2320 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
2321 | Intrinsic::maxnum, |
2322 | Intrinsic::experimental_constrained_maxnum)); |
2323 | |
2324 | case Builtin::BIfmin: |
2325 | case Builtin::BIfminf: |
2326 | case Builtin::BIfminl: |
2327 | case Builtin::BI__builtin_fmin: |
2328 | case Builtin::BI__builtin_fminf: |
2329 | case Builtin::BI__builtin_fminf16: |
2330 | case Builtin::BI__builtin_fminl: |
2331 | case Builtin::BI__builtin_fminf128: |
2332 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
2333 | Intrinsic::minnum, |
2334 | Intrinsic::experimental_constrained_minnum)); |
2335 | |
2336 | // fmod() is a special-case. It maps to the frem instruction rather than an |
2337 | // LLVM intrinsic. |
2338 | case Builtin::BIfmod: |
2339 | case Builtin::BIfmodf: |
2340 | case Builtin::BIfmodl: |
2341 | case Builtin::BI__builtin_fmod: |
2342 | case Builtin::BI__builtin_fmodf: |
2343 | case Builtin::BI__builtin_fmodf16: |
2344 | case Builtin::BI__builtin_fmodl: |
2345 | case Builtin::BI__builtin_fmodf128: { |
2346 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
2347 | Value *Arg1 = EmitScalarExpr(E->getArg(0)); |
2348 | Value *Arg2 = EmitScalarExpr(E->getArg(1)); |
2349 | return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod")); |
2350 | } |
2351 | |
2352 | case Builtin::BIlog: |
2353 | case Builtin::BIlogf: |
2354 | case Builtin::BIlogl: |
2355 | case Builtin::BI__builtin_log: |
2356 | case Builtin::BI__builtin_logf: |
2357 | case Builtin::BI__builtin_logf16: |
2358 | case Builtin::BI__builtin_logl: |
2359 | case Builtin::BI__builtin_logf128: |
2360 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2361 | Intrinsic::log, |
2362 | Intrinsic::experimental_constrained_log)); |
2363 | |
2364 | case Builtin::BIlog10: |
2365 | case Builtin::BIlog10f: |
2366 | case Builtin::BIlog10l: |
2367 | case Builtin::BI__builtin_log10: |
2368 | case Builtin::BI__builtin_log10f: |
2369 | case Builtin::BI__builtin_log10f16: |
2370 | case Builtin::BI__builtin_log10l: |
2371 | case Builtin::BI__builtin_log10f128: |
2372 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2373 | Intrinsic::log10, |
2374 | Intrinsic::experimental_constrained_log10)); |
2375 | |
2376 | case Builtin::BIlog2: |
2377 | case Builtin::BIlog2f: |
2378 | case Builtin::BIlog2l: |
2379 | case Builtin::BI__builtin_log2: |
2380 | case Builtin::BI__builtin_log2f: |
2381 | case Builtin::BI__builtin_log2f16: |
2382 | case Builtin::BI__builtin_log2l: |
2383 | case Builtin::BI__builtin_log2f128: |
2384 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2385 | Intrinsic::log2, |
2386 | Intrinsic::experimental_constrained_log2)); |
2387 | |
2388 | case Builtin::BInearbyint: |
2389 | case Builtin::BInearbyintf: |
2390 | case Builtin::BInearbyintl: |
2391 | case Builtin::BI__builtin_nearbyint: |
2392 | case Builtin::BI__builtin_nearbyintf: |
2393 | case Builtin::BI__builtin_nearbyintl: |
2394 | case Builtin::BI__builtin_nearbyintf128: |
2395 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2396 | Intrinsic::nearbyint, |
2397 | Intrinsic::experimental_constrained_nearbyint)); |
2398 | |
2399 | case Builtin::BIpow: |
2400 | case Builtin::BIpowf: |
2401 | case Builtin::BIpowl: |
2402 | case Builtin::BI__builtin_pow: |
2403 | case Builtin::BI__builtin_powf: |
2404 | case Builtin::BI__builtin_powf16: |
2405 | case Builtin::BI__builtin_powl: |
2406 | case Builtin::BI__builtin_powf128: |
2407 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
2408 | Intrinsic::pow, |
2409 | Intrinsic::experimental_constrained_pow)); |
2410 | |
2411 | case Builtin::BIrint: |
2412 | case Builtin::BIrintf: |
2413 | case Builtin::BIrintl: |
2414 | case Builtin::BI__builtin_rint: |
2415 | case Builtin::BI__builtin_rintf: |
2416 | case Builtin::BI__builtin_rintf16: |
2417 | case Builtin::BI__builtin_rintl: |
2418 | case Builtin::BI__builtin_rintf128: |
2419 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2420 | Intrinsic::rint, |
2421 | Intrinsic::experimental_constrained_rint)); |
2422 | |
2423 | case Builtin::BIround: |
2424 | case Builtin::BIroundf: |
2425 | case Builtin::BIroundl: |
2426 | case Builtin::BI__builtin_round: |
2427 | case Builtin::BI__builtin_roundf: |
2428 | case Builtin::BI__builtin_roundf16: |
2429 | case Builtin::BI__builtin_roundl: |
2430 | case Builtin::BI__builtin_roundf128: |
2431 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2432 | Intrinsic::round, |
2433 | Intrinsic::experimental_constrained_round)); |
2434 | |
2435 | case Builtin::BIsin: |
2436 | case Builtin::BIsinf: |
2437 | case Builtin::BIsinl: |
2438 | case Builtin::BI__builtin_sin: |
2439 | case Builtin::BI__builtin_sinf: |
2440 | case Builtin::BI__builtin_sinf16: |
2441 | case Builtin::BI__builtin_sinl: |
2442 | case Builtin::BI__builtin_sinf128: |
2443 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2444 | Intrinsic::sin, |
2445 | Intrinsic::experimental_constrained_sin)); |
2446 | |
2447 | case Builtin::BIsqrt: |
2448 | case Builtin::BIsqrtf: |
2449 | case Builtin::BIsqrtl: |
2450 | case Builtin::BI__builtin_sqrt: |
2451 | case Builtin::BI__builtin_sqrtf: |
2452 | case Builtin::BI__builtin_sqrtf16: |
2453 | case Builtin::BI__builtin_sqrtl: |
2454 | case Builtin::BI__builtin_sqrtf128: |
2455 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2456 | Intrinsic::sqrt, |
2457 | Intrinsic::experimental_constrained_sqrt)); |
2458 | |
2459 | case Builtin::BItrunc: |
2460 | case Builtin::BItruncf: |
2461 | case Builtin::BItruncl: |
2462 | case Builtin::BI__builtin_trunc: |
2463 | case Builtin::BI__builtin_truncf: |
2464 | case Builtin::BI__builtin_truncf16: |
2465 | case Builtin::BI__builtin_truncl: |
2466 | case Builtin::BI__builtin_truncf128: |
2467 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
2468 | Intrinsic::trunc, |
2469 | Intrinsic::experimental_constrained_trunc)); |
2470 | |
2471 | case Builtin::BIlround: |
2472 | case Builtin::BIlroundf: |
2473 | case Builtin::BIlroundl: |
2474 | case Builtin::BI__builtin_lround: |
2475 | case Builtin::BI__builtin_lroundf: |
2476 | case Builtin::BI__builtin_lroundl: |
2477 | case Builtin::BI__builtin_lroundf128: |
2478 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
2479 | *this, E, Intrinsic::lround, |
2480 | Intrinsic::experimental_constrained_lround)); |
2481 | |
2482 | case Builtin::BIllround: |
2483 | case Builtin::BIllroundf: |
2484 | case Builtin::BIllroundl: |
2485 | case Builtin::BI__builtin_llround: |
2486 | case Builtin::BI__builtin_llroundf: |
2487 | case Builtin::BI__builtin_llroundl: |
2488 | case Builtin::BI__builtin_llroundf128: |
2489 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
2490 | *this, E, Intrinsic::llround, |
2491 | Intrinsic::experimental_constrained_llround)); |
2492 | |
2493 | case Builtin::BIlrint: |
2494 | case Builtin::BIlrintf: |
2495 | case Builtin::BIlrintl: |
2496 | case Builtin::BI__builtin_lrint: |
2497 | case Builtin::BI__builtin_lrintf: |
2498 | case Builtin::BI__builtin_lrintl: |
2499 | case Builtin::BI__builtin_lrintf128: |
2500 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
2501 | *this, E, Intrinsic::lrint, |
2502 | Intrinsic::experimental_constrained_lrint)); |
2503 | |
2504 | case Builtin::BIllrint: |
2505 | case Builtin::BIllrintf: |
2506 | case Builtin::BIllrintl: |
2507 | case Builtin::BI__builtin_llrint: |
2508 | case Builtin::BI__builtin_llrintf: |
2509 | case Builtin::BI__builtin_llrintl: |
2510 | case Builtin::BI__builtin_llrintf128: |
2511 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
2512 | *this, E, Intrinsic::llrint, |
2513 | Intrinsic::experimental_constrained_llrint)); |
2514 | |
2515 | default: |
2516 | break; |
2517 | } |
2518 | } |
2519 | |
2520 | switch (BuiltinIDIfNoAsmLabel) { |
2521 | default: break; |
2522 | case Builtin::BI__builtin___CFStringMakeConstantString: |
2523 | case Builtin::BI__builtin___NSStringMakeConstantString: |
2524 | return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType())); |
2525 | case Builtin::BI__builtin_stdarg_start: |
2526 | case Builtin::BI__builtin_va_start: |
2527 | case Builtin::BI__va_start: |
2528 | case Builtin::BI__builtin_va_end: |
2529 | return RValue::get( |
2530 | EmitVAStartEnd(BuiltinID == Builtin::BI__va_start |
2531 | ? EmitScalarExpr(E->getArg(0)) |
2532 | : EmitVAListRef(E->getArg(0)).getPointer(), |
2533 | BuiltinID != Builtin::BI__builtin_va_end)); |
2534 | case Builtin::BI__builtin_va_copy: { |
2535 | Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer(); |
2536 | Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer(); |
2537 | |
2538 | llvm::Type *Type = Int8PtrTy; |
2539 | |
2540 | DstPtr = Builder.CreateBitCast(DstPtr, Type); |
2541 | SrcPtr = Builder.CreateBitCast(SrcPtr, Type); |
2542 | return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy), |
2543 | {DstPtr, SrcPtr})); |
2544 | } |
2545 | case Builtin::BI__builtin_abs: |
2546 | case Builtin::BI__builtin_labs: |
2547 | case Builtin::BI__builtin_llabs: { |
2548 | // X < 0 ? -X : X |
2549 | // The negation has 'nsw' because abs of INT_MIN is undefined. |
2550 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2551 | Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg"); |
2552 | Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType()); |
2553 | Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond"); |
2554 | Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs"); |
2555 | return RValue::get(Result); |
2556 | } |
2557 | case Builtin::BI__builtin_complex: { |
2558 | Value *Real = EmitScalarExpr(E->getArg(0)); |
2559 | Value *Imag = EmitScalarExpr(E->getArg(1)); |
2560 | return RValue::getComplex({Real, Imag}); |
2561 | } |
2562 | case Builtin::BI__builtin_conj: |
2563 | case Builtin::BI__builtin_conjf: |
2564 | case Builtin::BI__builtin_conjl: |
2565 | case Builtin::BIconj: |
2566 | case Builtin::BIconjf: |
2567 | case Builtin::BIconjl: { |
2568 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
2569 | Value *Real = ComplexVal.first; |
2570 | Value *Imag = ComplexVal.second; |
2571 | Imag = Builder.CreateFNeg(Imag, "neg"); |
2572 | return RValue::getComplex(std::make_pair(Real, Imag)); |
2573 | } |
2574 | case Builtin::BI__builtin_creal: |
2575 | case Builtin::BI__builtin_crealf: |
2576 | case Builtin::BI__builtin_creall: |
2577 | case Builtin::BIcreal: |
2578 | case Builtin::BIcrealf: |
2579 | case Builtin::BIcreall: { |
2580 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
2581 | return RValue::get(ComplexVal.first); |
2582 | } |
2583 | |
2584 | case Builtin::BI__builtin_dump_struct: { |
2585 | llvm::Type *LLVMIntTy = getTypes().ConvertType(getContext().IntTy); |
2586 | llvm::FunctionType *LLVMFuncType = llvm::FunctionType::get( |
2587 | LLVMIntTy, {llvm::Type::getInt8PtrTy(getLLVMContext())}, true); |
2588 | |
2589 | Value *Func = EmitScalarExpr(E->getArg(1)->IgnoreImpCasts()); |
2590 | CharUnits Arg0Align = EmitPointerWithAlignment(E->getArg(0)).getAlignment(); |
2591 | |
2592 | const Expr *Arg0 = E->getArg(0)->IgnoreImpCasts(); |
2593 | QualType Arg0Type = Arg0->getType()->getPointeeType(); |
2594 | |
2595 | Value *RecordPtr = EmitScalarExpr(Arg0); |
2596 | Value *Res = dumpRecord(*this, Arg0Type, RecordPtr, Arg0Align, |
2597 | {LLVMFuncType, Func}, 0); |
2598 | return RValue::get(Res); |
2599 | } |
2600 | |
2601 | case Builtin::BI__builtin_preserve_access_index: { |
2602 | // Only enabled preserved access index region when debuginfo |
2603 | // is available as debuginfo is needed to preserve user-level |
2604 | // access pattern. |
2605 | if (!getDebugInfo()) { |
2606 | CGM.Error(E->getExprLoc(), "using builtin_preserve_access_index() without -g"); |
2607 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
2608 | } |
2609 | |
2610 | // Nested builtin_preserve_access_index() not supported |
2611 | if (IsInPreservedAIRegion) { |
2612 | CGM.Error(E->getExprLoc(), "nested builtin_preserve_access_index() not supported"); |
2613 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
2614 | } |
2615 | |
2616 | IsInPreservedAIRegion = true; |
2617 | Value *Res = EmitScalarExpr(E->getArg(0)); |
2618 | IsInPreservedAIRegion = false; |
2619 | return RValue::get(Res); |
2620 | } |
2621 | |
2622 | case Builtin::BI__builtin_cimag: |
2623 | case Builtin::BI__builtin_cimagf: |
2624 | case Builtin::BI__builtin_cimagl: |
2625 | case Builtin::BIcimag: |
2626 | case Builtin::BIcimagf: |
2627 | case Builtin::BIcimagl: { |
2628 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
2629 | return RValue::get(ComplexVal.second); |
2630 | } |
2631 | |
2632 | case Builtin::BI__builtin_clrsb: |
2633 | case Builtin::BI__builtin_clrsbl: |
2634 | case Builtin::BI__builtin_clrsbll: { |
2635 | // clrsb(x) -> clz(x < 0 ? ~x : x) - 1 or |
2636 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2637 | |
2638 | llvm::Type *ArgType = ArgValue->getType(); |
2639 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
2640 | |
2641 | llvm::Type *ResultType = ConvertType(E->getType()); |
2642 | Value *Zero = llvm::Constant::getNullValue(ArgType); |
2643 | Value *IsNeg = Builder.CreateICmpSLT(ArgValue, Zero, "isneg"); |
2644 | Value *Inverse = Builder.CreateNot(ArgValue, "not"); |
2645 | Value *Tmp = Builder.CreateSelect(IsNeg, Inverse, ArgValue); |
2646 | Value *Ctlz = Builder.CreateCall(F, {Tmp, Builder.getFalse()}); |
2647 | Value *Result = Builder.CreateSub(Ctlz, llvm::ConstantInt::get(ArgType, 1)); |
2648 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
2649 | "cast"); |
2650 | return RValue::get(Result); |
2651 | } |
2652 | case Builtin::BI__builtin_ctzs: |
2653 | case Builtin::BI__builtin_ctz: |
2654 | case Builtin::BI__builtin_ctzl: |
2655 | case Builtin::BI__builtin_ctzll: { |
2656 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero); |
2657 | |
2658 | llvm::Type *ArgType = ArgValue->getType(); |
2659 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
2660 | |
2661 | llvm::Type *ResultType = ConvertType(E->getType()); |
2662 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); |
2663 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); |
2664 | if (Result->getType() != ResultType) |
2665 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
2666 | "cast"); |
2667 | return RValue::get(Result); |
2668 | } |
2669 | case Builtin::BI__builtin_clzs: |
2670 | case Builtin::BI__builtin_clz: |
2671 | case Builtin::BI__builtin_clzl: |
2672 | case Builtin::BI__builtin_clzll: { |
2673 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero); |
2674 | |
2675 | llvm::Type *ArgType = ArgValue->getType(); |
2676 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
2677 | |
2678 | llvm::Type *ResultType = ConvertType(E->getType()); |
2679 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); |
2680 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); |
2681 | if (Result->getType() != ResultType) |
2682 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
2683 | "cast"); |
2684 | return RValue::get(Result); |
2685 | } |
2686 | case Builtin::BI__builtin_ffs: |
2687 | case Builtin::BI__builtin_ffsl: |
2688 | case Builtin::BI__builtin_ffsll: { |
2689 | // ffs(x) -> x ? cttz(x) + 1 : 0 |
2690 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2691 | |
2692 | llvm::Type *ArgType = ArgValue->getType(); |
2693 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
2694 | |
2695 | llvm::Type *ResultType = ConvertType(E->getType()); |
2696 | Value *Tmp = |
2697 | Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}), |
2698 | llvm::ConstantInt::get(ArgType, 1)); |
2699 | Value *Zero = llvm::Constant::getNullValue(ArgType); |
2700 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero"); |
2701 | Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs"); |
2702 | if (Result->getType() != ResultType) |
2703 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
2704 | "cast"); |
2705 | return RValue::get(Result); |
2706 | } |
2707 | case Builtin::BI__builtin_parity: |
2708 | case Builtin::BI__builtin_parityl: |
2709 | case Builtin::BI__builtin_parityll: { |
2710 | // parity(x) -> ctpop(x) & 1 |
2711 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2712 | |
2713 | llvm::Type *ArgType = ArgValue->getType(); |
2714 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); |
2715 | |
2716 | llvm::Type *ResultType = ConvertType(E->getType()); |
2717 | Value *Tmp = Builder.CreateCall(F, ArgValue); |
2718 | Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1)); |
2719 | if (Result->getType() != ResultType) |
2720 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
2721 | "cast"); |
2722 | return RValue::get(Result); |
2723 | } |
2724 | case Builtin::BI__lzcnt16: |
2725 | case Builtin::BI__lzcnt: |
2726 | case Builtin::BI__lzcnt64: { |
2727 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2728 | |
2729 | llvm::Type *ArgType = ArgValue->getType(); |
2730 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
2731 | |
2732 | llvm::Type *ResultType = ConvertType(E->getType()); |
2733 | Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()}); |
2734 | if (Result->getType() != ResultType) |
2735 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
2736 | "cast"); |
2737 | return RValue::get(Result); |
2738 | } |
2739 | case Builtin::BI__popcnt16: |
2740 | case Builtin::BI__popcnt: |
2741 | case Builtin::BI__popcnt64: |
2742 | case Builtin::BI__builtin_popcount: |
2743 | case Builtin::BI__builtin_popcountl: |
2744 | case Builtin::BI__builtin_popcountll: { |
2745 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2746 | |
2747 | llvm::Type *ArgType = ArgValue->getType(); |
2748 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); |
2749 | |
2750 | llvm::Type *ResultType = ConvertType(E->getType()); |
2751 | Value *Result = Builder.CreateCall(F, ArgValue); |
2752 | if (Result->getType() != ResultType) |
2753 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
2754 | "cast"); |
2755 | return RValue::get(Result); |
2756 | } |
2757 | case Builtin::BI__builtin_unpredictable: { |
2758 | // Always return the argument of __builtin_unpredictable. LLVM does not |
2759 | // handle this builtin. Metadata for this builtin should be added directly |
2760 | // to instructions such as branches or switches that use it. |
2761 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
2762 | } |
2763 | case Builtin::BI__builtin_expect: { |
2764 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2765 | llvm::Type *ArgType = ArgValue->getType(); |
2766 | |
2767 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); |
2768 | // Don't generate llvm.expect on -O0 as the backend won't use it for |
2769 | // anything. |
2770 | // Note, we still IRGen ExpectedValue because it could have side-effects. |
2771 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) |
2772 | return RValue::get(ArgValue); |
2773 | |
2774 | Function *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType); |
2775 | Value *Result = |
2776 | Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval"); |
2777 | return RValue::get(Result); |
2778 | } |
2779 | case Builtin::BI__builtin_expect_with_probability: { |
2780 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2781 | llvm::Type *ArgType = ArgValue->getType(); |
2782 | |
2783 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); |
2784 | llvm::APFloat Probability(0.0); |
2785 | const Expr *ProbArg = E->getArg(2); |
2786 | bool EvalSucceed = ProbArg->EvaluateAsFloat(Probability, CGM.getContext()); |
2787 | assert(EvalSucceed && "probability should be able to evaluate as float")((EvalSucceed && "probability should be able to evaluate as float" ) ? static_cast<void> (0) : __assert_fail ("EvalSucceed && \"probability should be able to evaluate as float\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 2787, __PRETTY_FUNCTION__)); |
2788 | (void)EvalSucceed; |
2789 | bool LoseInfo = false; |
2790 | Probability.convert(llvm::APFloat::IEEEdouble(), |
2791 | llvm::RoundingMode::Dynamic, &LoseInfo); |
2792 | llvm::Type *Ty = ConvertType(ProbArg->getType()); |
2793 | Constant *Confidence = ConstantFP::get(Ty, Probability); |
2794 | // Don't generate llvm.expect.with.probability on -O0 as the backend |
2795 | // won't use it for anything. |
2796 | // Note, we still IRGen ExpectedValue because it could have side-effects. |
2797 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) |
2798 | return RValue::get(ArgValue); |
2799 | |
2800 | Function *FnExpect = |
2801 | CGM.getIntrinsic(Intrinsic::expect_with_probability, ArgType); |
2802 | Value *Result = Builder.CreateCall( |
2803 | FnExpect, {ArgValue, ExpectedValue, Confidence}, "expval"); |
2804 | return RValue::get(Result); |
2805 | } |
2806 | case Builtin::BI__builtin_assume_aligned: { |
2807 | const Expr *Ptr = E->getArg(0); |
2808 | Value *PtrValue = EmitScalarExpr(Ptr); |
2809 | Value *OffsetValue = |
2810 | (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr; |
2811 | |
2812 | Value *AlignmentValue = EmitScalarExpr(E->getArg(1)); |
2813 | ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue); |
2814 | if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment)) |
2815 | AlignmentCI = ConstantInt::get(AlignmentCI->getType(), |
2816 | llvm::Value::MaximumAlignment); |
2817 | |
2818 | emitAlignmentAssumption(PtrValue, Ptr, |
2819 | /*The expr loc is sufficient.*/ SourceLocation(), |
2820 | AlignmentCI, OffsetValue); |
2821 | return RValue::get(PtrValue); |
2822 | } |
2823 | case Builtin::BI__assume: |
2824 | case Builtin::BI__builtin_assume: { |
2825 | if (E->getArg(0)->HasSideEffects(getContext())) |
2826 | return RValue::get(nullptr); |
2827 | |
2828 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
2829 | Function *FnAssume = CGM.getIntrinsic(Intrinsic::assume); |
2830 | return RValue::get(Builder.CreateCall(FnAssume, ArgValue)); |
2831 | } |
2832 | case Builtin::BI__builtin_bswap16: |
2833 | case Builtin::BI__builtin_bswap32: |
2834 | case Builtin::BI__builtin_bswap64: { |
2835 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap)); |
2836 | } |
2837 | case Builtin::BI__builtin_bitreverse8: |
2838 | case Builtin::BI__builtin_bitreverse16: |
2839 | case Builtin::BI__builtin_bitreverse32: |
2840 | case Builtin::BI__builtin_bitreverse64: { |
2841 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse)); |
2842 | } |
2843 | case Builtin::BI__builtin_rotateleft8: |
2844 | case Builtin::BI__builtin_rotateleft16: |
2845 | case Builtin::BI__builtin_rotateleft32: |
2846 | case Builtin::BI__builtin_rotateleft64: |
2847 | case Builtin::BI_rotl8: // Microsoft variants of rotate left |
2848 | case Builtin::BI_rotl16: |
2849 | case Builtin::BI_rotl: |
2850 | case Builtin::BI_lrotl: |
2851 | case Builtin::BI_rotl64: |
2852 | return emitRotate(E, false); |
2853 | |
2854 | case Builtin::BI__builtin_rotateright8: |
2855 | case Builtin::BI__builtin_rotateright16: |
2856 | case Builtin::BI__builtin_rotateright32: |
2857 | case Builtin::BI__builtin_rotateright64: |
2858 | case Builtin::BI_rotr8: // Microsoft variants of rotate right |
2859 | case Builtin::BI_rotr16: |
2860 | case Builtin::BI_rotr: |
2861 | case Builtin::BI_lrotr: |
2862 | case Builtin::BI_rotr64: |
2863 | return emitRotate(E, true); |
2864 | |
2865 | case Builtin::BI__builtin_constant_p: { |
2866 | llvm::Type *ResultType = ConvertType(E->getType()); |
2867 | |
2868 | const Expr *Arg = E->getArg(0); |
2869 | QualType ArgType = Arg->getType(); |
2870 | // FIXME: The allowance for Obj-C pointers and block pointers is historical |
2871 | // and likely a mistake. |
2872 | if (!ArgType->isIntegralOrEnumerationType() && !ArgType->isFloatingType() && |
2873 | !ArgType->isObjCObjectPointerType() && !ArgType->isBlockPointerType()) |
2874 | // Per the GCC documentation, only numeric constants are recognized after |
2875 | // inlining. |
2876 | return RValue::get(ConstantInt::get(ResultType, 0)); |
2877 | |
2878 | if (Arg->HasSideEffects(getContext())) |
2879 | // The argument is unevaluated, so be conservative if it might have |
2880 | // side-effects. |
2881 | return RValue::get(ConstantInt::get(ResultType, 0)); |
2882 | |
2883 | Value *ArgValue = EmitScalarExpr(Arg); |
2884 | if (ArgType->isObjCObjectPointerType()) { |
2885 | // Convert Objective-C objects to id because we cannot distinguish between |
2886 | // LLVM types for Obj-C classes as they are opaque. |
2887 | ArgType = CGM.getContext().getObjCIdType(); |
2888 | ArgValue = Builder.CreateBitCast(ArgValue, ConvertType(ArgType)); |
2889 | } |
2890 | Function *F = |
2891 | CGM.getIntrinsic(Intrinsic::is_constant, ConvertType(ArgType)); |
2892 | Value *Result = Builder.CreateCall(F, ArgValue); |
2893 | if (Result->getType() != ResultType) |
2894 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/false); |
2895 | return RValue::get(Result); |
2896 | } |
2897 | case Builtin::BI__builtin_dynamic_object_size: |
2898 | case Builtin::BI__builtin_object_size: { |
2899 | unsigned Type = |
2900 | E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
2901 | auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType())); |
2902 | |
2903 | // We pass this builtin onto the optimizer so that it can figure out the |
2904 | // object size in more complex cases. |
2905 | bool IsDynamic = BuiltinID == Builtin::BI__builtin_dynamic_object_size; |
2906 | return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType, |
2907 | /*EmittedE=*/nullptr, IsDynamic)); |
2908 | } |
2909 | case Builtin::BI__builtin_prefetch: { |
2910 | Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0)); |
2911 | // FIXME: Technically these constants should of type 'int', yes? |
2912 | RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) : |
2913 | llvm::ConstantInt::get(Int32Ty, 0); |
2914 | Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : |
2915 | llvm::ConstantInt::get(Int32Ty, 3); |
2916 | Value *Data = llvm::ConstantInt::get(Int32Ty, 1); |
2917 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); |
2918 | return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data})); |
2919 | } |
2920 | case Builtin::BI__builtin_readcyclecounter: { |
2921 | Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter); |
2922 | return RValue::get(Builder.CreateCall(F)); |
2923 | } |
2924 | case Builtin::BI__builtin___clear_cache: { |
2925 | Value *Begin = EmitScalarExpr(E->getArg(0)); |
2926 | Value *End = EmitScalarExpr(E->getArg(1)); |
2927 | Function *F = CGM.getIntrinsic(Intrinsic::clear_cache); |
2928 | return RValue::get(Builder.CreateCall(F, {Begin, End})); |
2929 | } |
2930 | case Builtin::BI__builtin_trap: |
2931 | return RValue::get(EmitTrapCall(Intrinsic::trap)); |
2932 | case Builtin::BI__debugbreak: |
2933 | return RValue::get(EmitTrapCall(Intrinsic::debugtrap)); |
2934 | case Builtin::BI__builtin_unreachable: { |
2935 | EmitUnreachable(E->getExprLoc()); |
2936 | |
2937 | // We do need to preserve an insertion point. |
2938 | EmitBlock(createBasicBlock("unreachable.cont")); |
2939 | |
2940 | return RValue::get(nullptr); |
2941 | } |
2942 | |
2943 | case Builtin::BI__builtin_powi: |
2944 | case Builtin::BI__builtin_powif: |
2945 | case Builtin::BI__builtin_powil: |
2946 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin( |
2947 | *this, E, Intrinsic::powi, Intrinsic::experimental_constrained_powi)); |
2948 | |
2949 | case Builtin::BI__builtin_isgreater: |
2950 | case Builtin::BI__builtin_isgreaterequal: |
2951 | case Builtin::BI__builtin_isless: |
2952 | case Builtin::BI__builtin_islessequal: |
2953 | case Builtin::BI__builtin_islessgreater: |
2954 | case Builtin::BI__builtin_isunordered: { |
2955 | // Ordered comparisons: we know the arguments to these are matching scalar |
2956 | // floating point values. |
2957 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
2958 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
2959 | Value *LHS = EmitScalarExpr(E->getArg(0)); |
2960 | Value *RHS = EmitScalarExpr(E->getArg(1)); |
2961 | |
2962 | switch (BuiltinID) { |
2963 | default: llvm_unreachable("Unknown ordered comparison")::llvm::llvm_unreachable_internal("Unknown ordered comparison" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 2963); |
2964 | case Builtin::BI__builtin_isgreater: |
2965 | LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp"); |
2966 | break; |
2967 | case Builtin::BI__builtin_isgreaterequal: |
2968 | LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp"); |
2969 | break; |
2970 | case Builtin::BI__builtin_isless: |
2971 | LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp"); |
2972 | break; |
2973 | case Builtin::BI__builtin_islessequal: |
2974 | LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp"); |
2975 | break; |
2976 | case Builtin::BI__builtin_islessgreater: |
2977 | LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp"); |
2978 | break; |
2979 | case Builtin::BI__builtin_isunordered: |
2980 | LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp"); |
2981 | break; |
2982 | } |
2983 | // ZExt bool to int type. |
2984 | return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType()))); |
2985 | } |
2986 | case Builtin::BI__builtin_isnan: { |
2987 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
2988 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
2989 | Value *V = EmitScalarExpr(E->getArg(0)); |
2990 | V = Builder.CreateFCmpUNO(V, V, "cmp"); |
2991 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
2992 | } |
2993 | |
2994 | case Builtin::BI__builtin_matrix_transpose: { |
2995 | const auto *MatrixTy = E->getArg(0)->getType()->getAs<ConstantMatrixType>(); |
2996 | Value *MatValue = EmitScalarExpr(E->getArg(0)); |
2997 | MatrixBuilder<CGBuilderTy> MB(Builder); |
2998 | Value *Result = MB.CreateMatrixTranspose(MatValue, MatrixTy->getNumRows(), |
2999 | MatrixTy->getNumColumns()); |
3000 | return RValue::get(Result); |
3001 | } |
3002 | |
3003 | case Builtin::BI__builtin_matrix_column_major_load: { |
3004 | MatrixBuilder<CGBuilderTy> MB(Builder); |
3005 | // Emit everything that isn't dependent on the first parameter type |
3006 | Value *Stride = EmitScalarExpr(E->getArg(3)); |
3007 | const auto *ResultTy = E->getType()->getAs<ConstantMatrixType>(); |
3008 | auto *PtrTy = E->getArg(0)->getType()->getAs<PointerType>(); |
3009 | assert(PtrTy && "arg0 must be of pointer type")((PtrTy && "arg0 must be of pointer type") ? static_cast <void> (0) : __assert_fail ("PtrTy && \"arg0 must be of pointer type\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 3009, __PRETTY_FUNCTION__)); |
3010 | bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified(); |
3011 | |
3012 | Address Src = EmitPointerWithAlignment(E->getArg(0)); |
3013 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(0)->getType(), |
3014 | E->getArg(0)->getExprLoc(), FD, 0); |
3015 | Value *Result = MB.CreateColumnMajorLoad( |
3016 | Src.getPointer(), Align(Src.getAlignment().getQuantity()), Stride, |
3017 | IsVolatile, ResultTy->getNumRows(), ResultTy->getNumColumns(), |
3018 | "matrix"); |
3019 | return RValue::get(Result); |
3020 | } |
3021 | |
3022 | case Builtin::BI__builtin_matrix_column_major_store: { |
3023 | MatrixBuilder<CGBuilderTy> MB(Builder); |
3024 | Value *Matrix = EmitScalarExpr(E->getArg(0)); |
3025 | Address Dst = EmitPointerWithAlignment(E->getArg(1)); |
3026 | Value *Stride = EmitScalarExpr(E->getArg(2)); |
3027 | |
3028 | const auto *MatrixTy = E->getArg(0)->getType()->getAs<ConstantMatrixType>(); |
3029 | auto *PtrTy = E->getArg(1)->getType()->getAs<PointerType>(); |
3030 | assert(PtrTy && "arg1 must be of pointer type")((PtrTy && "arg1 must be of pointer type") ? static_cast <void> (0) : __assert_fail ("PtrTy && \"arg1 must be of pointer type\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 3030, __PRETTY_FUNCTION__)); |
3031 | bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified(); |
3032 | |
3033 | EmitNonNullArgCheck(RValue::get(Dst.getPointer()), E->getArg(1)->getType(), |
3034 | E->getArg(1)->getExprLoc(), FD, 0); |
3035 | Value *Result = MB.CreateColumnMajorStore( |
3036 | Matrix, Dst.getPointer(), Align(Dst.getAlignment().getQuantity()), |
3037 | Stride, IsVolatile, MatrixTy->getNumRows(), MatrixTy->getNumColumns()); |
3038 | return RValue::get(Result); |
3039 | } |
3040 | |
3041 | case Builtin::BIfinite: |
3042 | case Builtin::BI__finite: |
3043 | case Builtin::BIfinitef: |
3044 | case Builtin::BI__finitef: |
3045 | case Builtin::BIfinitel: |
3046 | case Builtin::BI__finitel: |
3047 | case Builtin::BI__builtin_isinf: |
3048 | case Builtin::BI__builtin_isfinite: { |
3049 | // isinf(x) --> fabs(x) == infinity |
3050 | // isfinite(x) --> fabs(x) != infinity |
3051 | // x != NaN via the ordered compare in either case. |
3052 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
3053 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
3054 | Value *V = EmitScalarExpr(E->getArg(0)); |
3055 | Value *Fabs = EmitFAbs(*this, V); |
3056 | Constant *Infinity = ConstantFP::getInfinity(V->getType()); |
3057 | CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf) |
3058 | ? CmpInst::FCMP_OEQ |
3059 | : CmpInst::FCMP_ONE; |
3060 | Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf"); |
3061 | return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType()))); |
3062 | } |
3063 | |
3064 | case Builtin::BI__builtin_isinf_sign: { |
3065 | // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0 |
3066 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
3067 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
3068 | Value *Arg = EmitScalarExpr(E->getArg(0)); |
3069 | Value *AbsArg = EmitFAbs(*this, Arg); |
3070 | Value *IsInf = Builder.CreateFCmpOEQ( |
3071 | AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf"); |
3072 | Value *IsNeg = EmitSignBit(*this, Arg); |
3073 | |
3074 | llvm::Type *IntTy = ConvertType(E->getType()); |
3075 | Value *Zero = Constant::getNullValue(IntTy); |
3076 | Value *One = ConstantInt::get(IntTy, 1); |
3077 | Value *NegativeOne = ConstantInt::get(IntTy, -1); |
3078 | Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One); |
3079 | Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero); |
3080 | return RValue::get(Result); |
3081 | } |
3082 | |
3083 | case Builtin::BI__builtin_isnormal: { |
3084 | // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min |
3085 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
3086 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
3087 | Value *V = EmitScalarExpr(E->getArg(0)); |
3088 | Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq"); |
3089 | |
3090 | Value *Abs = EmitFAbs(*this, V); |
3091 | Value *IsLessThanInf = |
3092 | Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf"); |
3093 | APFloat Smallest = APFloat::getSmallestNormalized( |
3094 | getContext().getFloatTypeSemantics(E->getArg(0)->getType())); |
3095 | Value *IsNormal = |
3096 | Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest), |
3097 | "isnormal"); |
3098 | V = Builder.CreateAnd(Eq, IsLessThanInf, "and"); |
3099 | V = Builder.CreateAnd(V, IsNormal, "and"); |
3100 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
3101 | } |
3102 | |
3103 | case Builtin::BI__builtin_flt_rounds: { |
3104 | Function *F = CGM.getIntrinsic(Intrinsic::flt_rounds); |
3105 | |
3106 | llvm::Type *ResultType = ConvertType(E->getType()); |
3107 | Value *Result = Builder.CreateCall(F); |
3108 | if (Result->getType() != ResultType) |
3109 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
3110 | "cast"); |
3111 | return RValue::get(Result); |
3112 | } |
3113 | |
3114 | case Builtin::BI__builtin_fpclassify: { |
3115 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
3116 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
3117 | Value *V = EmitScalarExpr(E->getArg(5)); |
3118 | llvm::Type *Ty = ConvertType(E->getArg(5)->getType()); |
3119 | |
3120 | // Create Result |
3121 | BasicBlock *Begin = Builder.GetInsertBlock(); |
3122 | BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn); |
3123 | Builder.SetInsertPoint(End); |
3124 | PHINode *Result = |
3125 | Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4, |
3126 | "fpclassify_result"); |
3127 | |
3128 | // if (V==0) return FP_ZERO |
3129 | Builder.SetInsertPoint(Begin); |
3130 | Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty), |
3131 | "iszero"); |
3132 | Value *ZeroLiteral = EmitScalarExpr(E->getArg(4)); |
3133 | BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn); |
3134 | Builder.CreateCondBr(IsZero, End, NotZero); |
3135 | Result->addIncoming(ZeroLiteral, Begin); |
3136 | |
3137 | // if (V != V) return FP_NAN |
3138 | Builder.SetInsertPoint(NotZero); |
3139 | Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp"); |
3140 | Value *NanLiteral = EmitScalarExpr(E->getArg(0)); |
3141 | BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn); |
3142 | Builder.CreateCondBr(IsNan, End, NotNan); |
3143 | Result->addIncoming(NanLiteral, NotZero); |
3144 | |
3145 | // if (fabs(V) == infinity) return FP_INFINITY |
3146 | Builder.SetInsertPoint(NotNan); |
3147 | Value *VAbs = EmitFAbs(*this, V); |
3148 | Value *IsInf = |
3149 | Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()), |
3150 | "isinf"); |
3151 | Value *InfLiteral = EmitScalarExpr(E->getArg(1)); |
3152 | BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn); |
3153 | Builder.CreateCondBr(IsInf, End, NotInf); |
3154 | Result->addIncoming(InfLiteral, NotNan); |
3155 | |
3156 | // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL |
3157 | Builder.SetInsertPoint(NotInf); |
3158 | APFloat Smallest = APFloat::getSmallestNormalized( |
3159 | getContext().getFloatTypeSemantics(E->getArg(5)->getType())); |
3160 | Value *IsNormal = |
3161 | Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest), |
3162 | "isnormal"); |
3163 | Value *NormalResult = |
3164 | Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)), |
3165 | EmitScalarExpr(E->getArg(3))); |
3166 | Builder.CreateBr(End); |
3167 | Result->addIncoming(NormalResult, NotInf); |
3168 | |
3169 | // return Result |
3170 | Builder.SetInsertPoint(End); |
3171 | return RValue::get(Result); |
3172 | } |
3173 | |
3174 | case Builtin::BIalloca: |
3175 | case Builtin::BI_alloca: |
3176 | case Builtin::BI__builtin_alloca: { |
3177 | Value *Size = EmitScalarExpr(E->getArg(0)); |
3178 | const TargetInfo &TI = getContext().getTargetInfo(); |
3179 | // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__. |
3180 | const Align SuitableAlignmentInBytes = |
3181 | CGM.getContext() |
3182 | .toCharUnitsFromBits(TI.getSuitableAlign()) |
3183 | .getAsAlign(); |
3184 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); |
3185 | AI->setAlignment(SuitableAlignmentInBytes); |
3186 | initializeAlloca(*this, AI, Size, SuitableAlignmentInBytes); |
3187 | return RValue::get(AI); |
3188 | } |
3189 | |
3190 | case Builtin::BI__builtin_alloca_with_align: { |
3191 | Value *Size = EmitScalarExpr(E->getArg(0)); |
3192 | Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1)); |
3193 | auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue); |
3194 | unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue(); |
3195 | const Align AlignmentInBytes = |
3196 | CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getAsAlign(); |
3197 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); |
3198 | AI->setAlignment(AlignmentInBytes); |
3199 | initializeAlloca(*this, AI, Size, AlignmentInBytes); |
3200 | return RValue::get(AI); |
3201 | } |
3202 | |
3203 | case Builtin::BIbzero: |
3204 | case Builtin::BI__builtin_bzero: { |
3205 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
3206 | Value *SizeVal = EmitScalarExpr(E->getArg(1)); |
3207 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
3208 | E->getArg(0)->getExprLoc(), FD, 0); |
3209 | Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false); |
3210 | return RValue::get(nullptr); |
3211 | } |
3212 | case Builtin::BImemcpy: |
3213 | case Builtin::BI__builtin_memcpy: |
3214 | case Builtin::BImempcpy: |
3215 | case Builtin::BI__builtin_mempcpy: { |
3216 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
3217 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
3218 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
3219 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
3220 | E->getArg(0)->getExprLoc(), FD, 0); |
3221 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
3222 | E->getArg(1)->getExprLoc(), FD, 1); |
3223 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); |
3224 | if (BuiltinID == Builtin::BImempcpy || |
3225 | BuiltinID == Builtin::BI__builtin_mempcpy) |
3226 | return RValue::get(Builder.CreateInBoundsGEP(Dest.getPointer(), SizeVal)); |
3227 | else |
3228 | return RValue::get(Dest.getPointer()); |
3229 | } |
3230 | |
3231 | case Builtin::BI__builtin_memcpy_inline: { |
3232 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
3233 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
3234 | uint64_t Size = |
3235 | E->getArg(2)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
3236 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
3237 | E->getArg(0)->getExprLoc(), FD, 0); |
3238 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
3239 | E->getArg(1)->getExprLoc(), FD, 1); |
3240 | Builder.CreateMemCpyInline(Dest, Src, Size); |
3241 | return RValue::get(nullptr); |
3242 | } |
3243 | |
3244 | case Builtin::BI__builtin_char_memchr: |
3245 | BuiltinID = Builtin::BI__builtin_memchr; |
3246 | break; |
3247 | |
3248 | case Builtin::BI__builtin___memcpy_chk: { |
3249 | // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2. |
3250 | Expr::EvalResult SizeResult, DstSizeResult; |
3251 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
3252 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
3253 | break; |
3254 | llvm::APSInt Size = SizeResult.Val.getInt(); |
3255 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
3256 | if (Size.ugt(DstSize)) |
3257 | break; |
3258 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
3259 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
3260 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
3261 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); |
3262 | return RValue::get(Dest.getPointer()); |
3263 | } |
3264 | |
3265 | case Builtin::BI__builtin_objc_memmove_collectable: { |
3266 | Address DestAddr = EmitPointerWithAlignment(E->getArg(0)); |
3267 | Address SrcAddr = EmitPointerWithAlignment(E->getArg(1)); |
3268 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
3269 | CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, |
3270 | DestAddr, SrcAddr, SizeVal); |
3271 | return RValue::get(DestAddr.getPointer()); |
3272 | } |
3273 | |
3274 | case Builtin::BI__builtin___memmove_chk: { |
3275 | // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2. |
3276 | Expr::EvalResult SizeResult, DstSizeResult; |
3277 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
3278 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
3279 | break; |
3280 | llvm::APSInt Size = SizeResult.Val.getInt(); |
3281 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
3282 | if (Size.ugt(DstSize)) |
3283 | break; |
3284 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
3285 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
3286 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
3287 | Builder.CreateMemMove(Dest, Src, SizeVal, false); |
3288 | return RValue::get(Dest.getPointer()); |
3289 | } |
3290 | |
3291 | case Builtin::BImemmove: |
3292 | case Builtin::BI__builtin_memmove: { |
3293 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
3294 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
3295 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
3296 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
3297 | E->getArg(0)->getExprLoc(), FD, 0); |
3298 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
3299 | E->getArg(1)->getExprLoc(), FD, 1); |
3300 | Builder.CreateMemMove(Dest, Src, SizeVal, false); |
3301 | return RValue::get(Dest.getPointer()); |
3302 | } |
3303 | case Builtin::BImemset: |
3304 | case Builtin::BI__builtin_memset: { |
3305 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
3306 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), |
3307 | Builder.getInt8Ty()); |
3308 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
3309 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
3310 | E->getArg(0)->getExprLoc(), FD, 0); |
3311 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); |
3312 | return RValue::get(Dest.getPointer()); |
3313 | } |
3314 | case Builtin::BI__builtin___memset_chk: { |
3315 | // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2. |
3316 | Expr::EvalResult SizeResult, DstSizeResult; |
3317 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
3318 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
3319 | break; |
3320 | llvm::APSInt Size = SizeResult.Val.getInt(); |
3321 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
3322 | if (Size.ugt(DstSize)) |
3323 | break; |
3324 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
3325 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), |
3326 | Builder.getInt8Ty()); |
3327 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
3328 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); |
3329 | return RValue::get(Dest.getPointer()); |
3330 | } |
3331 | case Builtin::BI__builtin_wmemcmp: { |
3332 | // The MSVC runtime library does not provide a definition of wmemcmp, so we |
3333 | // need an inline implementation. |
3334 | if (!getTarget().getTriple().isOSMSVCRT()) |
3335 | break; |
3336 | |
3337 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); |
3338 | |
3339 | Value *Dst = EmitScalarExpr(E->getArg(0)); |
3340 | Value *Src = EmitScalarExpr(E->getArg(1)); |
3341 | Value *Size = EmitScalarExpr(E->getArg(2)); |
3342 | |
3343 | BasicBlock *Entry = Builder.GetInsertBlock(); |
3344 | BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt"); |
3345 | BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt"); |
3346 | BasicBlock *Next = createBasicBlock("wmemcmp.next"); |
3347 | BasicBlock *Exit = createBasicBlock("wmemcmp.exit"); |
3348 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); |
3349 | Builder.CreateCondBr(SizeEq0, Exit, CmpGT); |
3350 | |
3351 | EmitBlock(CmpGT); |
3352 | PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2); |
3353 | DstPhi->addIncoming(Dst, Entry); |
3354 | PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2); |
3355 | SrcPhi->addIncoming(Src, Entry); |
3356 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); |
3357 | SizePhi->addIncoming(Size, Entry); |
3358 | CharUnits WCharAlign = |
3359 | getContext().getTypeAlignInChars(getContext().WCharTy); |
3360 | Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign); |
3361 | Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign); |
3362 | Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh); |
3363 | Builder.CreateCondBr(DstGtSrc, Exit, CmpLT); |
3364 | |
3365 | EmitBlock(CmpLT); |
3366 | Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh); |
3367 | Builder.CreateCondBr(DstLtSrc, Exit, Next); |
3368 | |
3369 | EmitBlock(Next); |
3370 | Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1); |
3371 | Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1); |
3372 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); |
3373 | Value *NextSizeEq0 = |
3374 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); |
3375 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT); |
3376 | DstPhi->addIncoming(NextDst, Next); |
3377 | SrcPhi->addIncoming(NextSrc, Next); |
3378 | SizePhi->addIncoming(NextSize, Next); |
3379 | |
3380 | EmitBlock(Exit); |
3381 | PHINode *Ret = Builder.CreatePHI(IntTy, 4); |
3382 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry); |
3383 | Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT); |
3384 | Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT); |
3385 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Next); |
3386 | return RValue::get(Ret); |
3387 | } |
3388 | case Builtin::BI__builtin_dwarf_cfa: { |
3389 | // The offset in bytes from the first argument to the CFA. |
3390 | // |
3391 | // Why on earth is this in the frontend? Is there any reason at |
3392 | // all that the backend can't reasonably determine this while |
3393 | // lowering llvm.eh.dwarf.cfa()? |
3394 | // |
3395 | // TODO: If there's a satisfactory reason, add a target hook for |
3396 | // this instead of hard-coding 0, which is correct for most targets. |
3397 | int32_t Offset = 0; |
3398 | |
3399 | Function *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa); |
3400 | return RValue::get(Builder.CreateCall(F, |
3401 | llvm::ConstantInt::get(Int32Ty, Offset))); |
3402 | } |
3403 | case Builtin::BI__builtin_return_address: { |
3404 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), |
3405 | getContext().UnsignedIntTy); |
3406 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); |
3407 | return RValue::get(Builder.CreateCall(F, Depth)); |
3408 | } |
3409 | case Builtin::BI_ReturnAddress: { |
3410 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); |
3411 | return RValue::get(Builder.CreateCall(F, Builder.getInt32(0))); |
3412 | } |
3413 | case Builtin::BI__builtin_frame_address: { |
3414 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), |
3415 | getContext().UnsignedIntTy); |
3416 | Function *F = CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy); |
3417 | return RValue::get(Builder.CreateCall(F, Depth)); |
3418 | } |
3419 | case Builtin::BI__builtin_extract_return_addr: { |
3420 | Value *Address = EmitScalarExpr(E->getArg(0)); |
3421 | Value *Result = getTargetHooks().decodeReturnAddress(*this, Address); |
3422 | return RValue::get(Result); |
3423 | } |
3424 | case Builtin::BI__builtin_frob_return_addr: { |
3425 | Value *Address = EmitScalarExpr(E->getArg(0)); |
3426 | Value *Result = getTargetHooks().encodeReturnAddress(*this, Address); |
3427 | return RValue::get(Result); |
3428 | } |
3429 | case Builtin::BI__builtin_dwarf_sp_column: { |
3430 | llvm::IntegerType *Ty |
3431 | = cast<llvm::IntegerType>(ConvertType(E->getType())); |
3432 | int Column = getTargetHooks().getDwarfEHStackPointer(CGM); |
3433 | if (Column == -1) { |
3434 | CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column"); |
3435 | return RValue::get(llvm::UndefValue::get(Ty)); |
3436 | } |
3437 | return RValue::get(llvm::ConstantInt::get(Ty, Column, true)); |
3438 | } |
3439 | case Builtin::BI__builtin_init_dwarf_reg_size_table: { |
3440 | Value *Address = EmitScalarExpr(E->getArg(0)); |
3441 | if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address)) |
3442 | CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table"); |
3443 | return RValue::get(llvm::UndefValue::get(ConvertType(E->getType()))); |
3444 | } |
3445 | case Builtin::BI__builtin_eh_return: { |
3446 | Value *Int = EmitScalarExpr(E->getArg(0)); |
3447 | Value *Ptr = EmitScalarExpr(E->getArg(1)); |
3448 | |
3449 | llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType()); |
3450 | assert((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) &&(((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants" ) ? static_cast<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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 3451, __PRETTY_FUNCTION__)) |
3451 | "LLVM's __builtin_eh_return only supports 32- and 64-bit variants")(((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants" ) ? static_cast<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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 3451, __PRETTY_FUNCTION__)); |
3452 | Function *F = |
3453 | CGM.getIntrinsic(IntTy->getBitWidth() == 32 ? Intrinsic::eh_return_i32 |
3454 | : Intrinsic::eh_return_i64); |
3455 | Builder.CreateCall(F, {Int, Ptr}); |
3456 | Builder.CreateUnreachable(); |
3457 | |
3458 | // We do need to preserve an insertion point. |
3459 | EmitBlock(createBasicBlock("builtin_eh_return.cont")); |
3460 | |
3461 | return RValue::get(nullptr); |
3462 | } |
3463 | case Builtin::BI__builtin_unwind_init: { |
3464 | Function *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init); |
3465 | return RValue::get(Builder.CreateCall(F)); |
3466 | } |
3467 | case Builtin::BI__builtin_extend_pointer: { |
3468 | // Extends a pointer to the size of an _Unwind_Word, which is |
3469 | // uint64_t on all platforms. Generally this gets poked into a |
3470 | // register and eventually used as an address, so if the |
3471 | // addressing registers are wider than pointers and the platform |
3472 | // doesn't implicitly ignore high-order bits when doing |
3473 | // addressing, we need to make sure we zext / sext based on |
3474 | // the platform's expectations. |
3475 | // |
3476 | // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html |
3477 | |
3478 | // Cast the pointer to intptr_t. |
3479 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
3480 | Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast"); |
3481 | |
3482 | // If that's 64 bits, we're done. |
3483 | if (IntPtrTy->getBitWidth() == 64) |
3484 | return RValue::get(Result); |
3485 | |
3486 | // Otherwise, ask the codegen data what to do. |
3487 | if (getTargetHooks().extendPointerWithSExt()) |
3488 | return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext")); |
3489 | else |
3490 | return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext")); |
3491 | } |
3492 | case Builtin::BI__builtin_setjmp: { |
3493 | // Buffer is a void**. |
3494 | Address Buf = EmitPointerWithAlignment(E->getArg(0)); |
3495 | |
3496 | // Store the frame pointer to the setjmp buffer. |
3497 | Value *FrameAddr = Builder.CreateCall( |
3498 | CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy), |
3499 | ConstantInt::get(Int32Ty, 0)); |
3500 | Builder.CreateStore(FrameAddr, Buf); |
3501 | |
3502 | // Store the stack pointer to the setjmp buffer. |
3503 | Value *StackAddr = |
3504 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave)); |
3505 | Address StackSaveSlot = Builder.CreateConstInBoundsGEP(Buf, 2); |
3506 | Builder.CreateStore(StackAddr, StackSaveSlot); |
3507 | |
3508 | // Call LLVM's EH setjmp, which is lightweight. |
3509 | Function *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp); |
3510 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); |
3511 | return RValue::get(Builder.CreateCall(F, Buf.getPointer())); |
3512 | } |
3513 | case Builtin::BI__builtin_longjmp: { |
3514 | Value *Buf = EmitScalarExpr(E->getArg(0)); |
3515 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); |
3516 | |
3517 | // Call LLVM's EH longjmp, which is lightweight. |
3518 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf); |
3519 | |
3520 | // longjmp doesn't return; mark this as unreachable. |
3521 | Builder.CreateUnreachable(); |
3522 | |
3523 | // We do need to preserve an insertion point. |
3524 | EmitBlock(createBasicBlock("longjmp.cont")); |
3525 | |
3526 | return RValue::get(nullptr); |
3527 | } |
3528 | case Builtin::BI__builtin_launder: { |
3529 | const Expr *Arg = E->getArg(0); |
3530 | QualType ArgTy = Arg->getType()->getPointeeType(); |
3531 | Value *Ptr = EmitScalarExpr(Arg); |
3532 | if (TypeRequiresBuiltinLaunder(CGM, ArgTy)) |
3533 | Ptr = Builder.CreateLaunderInvariantGroup(Ptr); |
3534 | |
3535 | return RValue::get(Ptr); |
3536 | } |
3537 | case Builtin::BI__sync_fetch_and_add: |
3538 | case Builtin::BI__sync_fetch_and_sub: |
3539 | case Builtin::BI__sync_fetch_and_or: |
3540 | case Builtin::BI__sync_fetch_and_and: |
3541 | case Builtin::BI__sync_fetch_and_xor: |
3542 | case Builtin::BI__sync_fetch_and_nand: |
3543 | case Builtin::BI__sync_add_and_fetch: |
3544 | case Builtin::BI__sync_sub_and_fetch: |
3545 | case Builtin::BI__sync_and_and_fetch: |
3546 | case Builtin::BI__sync_or_and_fetch: |
3547 | case Builtin::BI__sync_xor_and_fetch: |
3548 | case Builtin::BI__sync_nand_and_fetch: |
3549 | case Builtin::BI__sync_val_compare_and_swap: |
3550 | case Builtin::BI__sync_bool_compare_and_swap: |
3551 | case Builtin::BI__sync_lock_test_and_set: |
3552 | case Builtin::BI__sync_lock_release: |
3553 | case Builtin::BI__sync_swap: |
3554 | llvm_unreachable("Shouldn't make it through sema")::llvm::llvm_unreachable_internal("Shouldn't make it through sema" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 3554); |
3555 | case Builtin::BI__sync_fetch_and_add_1: |
3556 | case Builtin::BI__sync_fetch_and_add_2: |
3557 | case Builtin::BI__sync_fetch_and_add_4: |
3558 | case Builtin::BI__sync_fetch_and_add_8: |
3559 | case Builtin::BI__sync_fetch_and_add_16: |
3560 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E); |
3561 | case Builtin::BI__sync_fetch_and_sub_1: |
3562 | case Builtin::BI__sync_fetch_and_sub_2: |
3563 | case Builtin::BI__sync_fetch_and_sub_4: |
3564 | case Builtin::BI__sync_fetch_and_sub_8: |
3565 | case Builtin::BI__sync_fetch_and_sub_16: |
3566 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E); |
3567 | case Builtin::BI__sync_fetch_and_or_1: |
3568 | case Builtin::BI__sync_fetch_and_or_2: |
3569 | case Builtin::BI__sync_fetch_and_or_4: |
3570 | case Builtin::BI__sync_fetch_and_or_8: |
3571 | case Builtin::BI__sync_fetch_and_or_16: |
3572 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E); |
3573 | case Builtin::BI__sync_fetch_and_and_1: |
3574 | case Builtin::BI__sync_fetch_and_and_2: |
3575 | case Builtin::BI__sync_fetch_and_and_4: |
3576 | case Builtin::BI__sync_fetch_and_and_8: |
3577 | case Builtin::BI__sync_fetch_and_and_16: |
3578 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E); |
3579 | case Builtin::BI__sync_fetch_and_xor_1: |
3580 | case Builtin::BI__sync_fetch_and_xor_2: |
3581 | case Builtin::BI__sync_fetch_and_xor_4: |
3582 | case Builtin::BI__sync_fetch_and_xor_8: |
3583 | case Builtin::BI__sync_fetch_and_xor_16: |
3584 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E); |
3585 | case Builtin::BI__sync_fetch_and_nand_1: |
3586 | case Builtin::BI__sync_fetch_and_nand_2: |
3587 | case Builtin::BI__sync_fetch_and_nand_4: |
3588 | case Builtin::BI__sync_fetch_and_nand_8: |
3589 | case Builtin::BI__sync_fetch_and_nand_16: |
3590 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E); |
3591 | |
3592 | // Clang extensions: not overloaded yet. |
3593 | case Builtin::BI__sync_fetch_and_min: |
3594 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E); |
3595 | case Builtin::BI__sync_fetch_and_max: |
3596 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E); |
3597 | case Builtin::BI__sync_fetch_and_umin: |
3598 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E); |
3599 | case Builtin::BI__sync_fetch_and_umax: |
3600 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E); |
3601 | |
3602 | case Builtin::BI__sync_add_and_fetch_1: |
3603 | case Builtin::BI__sync_add_and_fetch_2: |
3604 | case Builtin::BI__sync_add_and_fetch_4: |
3605 | case Builtin::BI__sync_add_and_fetch_8: |
3606 | case Builtin::BI__sync_add_and_fetch_16: |
3607 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E, |
3608 | llvm::Instruction::Add); |
3609 | case Builtin::BI__sync_sub_and_fetch_1: |
3610 | case Builtin::BI__sync_sub_and_fetch_2: |
3611 | case Builtin::BI__sync_sub_and_fetch_4: |
3612 | case Builtin::BI__sync_sub_and_fetch_8: |
3613 | case Builtin::BI__sync_sub_and_fetch_16: |
3614 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E, |
3615 | llvm::Instruction::Sub); |
3616 | case Builtin::BI__sync_and_and_fetch_1: |
3617 | case Builtin::BI__sync_and_and_fetch_2: |
3618 | case Builtin::BI__sync_and_and_fetch_4: |
3619 | case Builtin::BI__sync_and_and_fetch_8: |
3620 | case Builtin::BI__sync_and_and_fetch_16: |
3621 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E, |
3622 | llvm::Instruction::And); |
3623 | case Builtin::BI__sync_or_and_fetch_1: |
3624 | case Builtin::BI__sync_or_and_fetch_2: |
3625 | case Builtin::BI__sync_or_and_fetch_4: |
3626 | case Builtin::BI__sync_or_and_fetch_8: |
3627 | case Builtin::BI__sync_or_and_fetch_16: |
3628 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E, |
3629 | llvm::Instruction::Or); |
3630 | case Builtin::BI__sync_xor_and_fetch_1: |
3631 | case Builtin::BI__sync_xor_and_fetch_2: |
3632 | case Builtin::BI__sync_xor_and_fetch_4: |
3633 | case Builtin::BI__sync_xor_and_fetch_8: |
3634 | case Builtin::BI__sync_xor_and_fetch_16: |
3635 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E, |
3636 | llvm::Instruction::Xor); |
3637 | case Builtin::BI__sync_nand_and_fetch_1: |
3638 | case Builtin::BI__sync_nand_and_fetch_2: |
3639 | case Builtin::BI__sync_nand_and_fetch_4: |
3640 | case Builtin::BI__sync_nand_and_fetch_8: |
3641 | case Builtin::BI__sync_nand_and_fetch_16: |
3642 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E, |
3643 | llvm::Instruction::And, true); |
3644 | |
3645 | case Builtin::BI__sync_val_compare_and_swap_1: |
3646 | case Builtin::BI__sync_val_compare_and_swap_2: |
3647 | case Builtin::BI__sync_val_compare_and_swap_4: |
3648 | case Builtin::BI__sync_val_compare_and_swap_8: |
3649 | case Builtin::BI__sync_val_compare_and_swap_16: |
3650 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, false)); |
3651 | |
3652 | case Builtin::BI__sync_bool_compare_and_swap_1: |
3653 | case Builtin::BI__sync_bool_compare_and_swap_2: |
3654 | case Builtin::BI__sync_bool_compare_and_swap_4: |
3655 | case Builtin::BI__sync_bool_compare_and_swap_8: |
3656 | case Builtin::BI__sync_bool_compare_and_swap_16: |
3657 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, true)); |
3658 | |
3659 | case Builtin::BI__sync_swap_1: |
3660 | case Builtin::BI__sync_swap_2: |
3661 | case Builtin::BI__sync_swap_4: |
3662 | case Builtin::BI__sync_swap_8: |
3663 | case Builtin::BI__sync_swap_16: |
3664 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); |
3665 | |
3666 | case Builtin::BI__sync_lock_test_and_set_1: |
3667 | case Builtin::BI__sync_lock_test_and_set_2: |
3668 | case Builtin::BI__sync_lock_test_and_set_4: |
3669 | case Builtin::BI__sync_lock_test_and_set_8: |
3670 | case Builtin::BI__sync_lock_test_and_set_16: |
3671 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); |
3672 | |
3673 | case Builtin::BI__sync_lock_release_1: |
3674 | case Builtin::BI__sync_lock_release_2: |
3675 | case Builtin::BI__sync_lock_release_4: |
3676 | case Builtin::BI__sync_lock_release_8: |
3677 | case Builtin::BI__sync_lock_release_16: { |
3678 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
3679 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
3680 | CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy); |
3681 | llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(), |
3682 | StoreSize.getQuantity() * 8); |
3683 | Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
3684 | llvm::StoreInst *Store = |
3685 | Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr, |
3686 | StoreSize); |
3687 | Store->setAtomic(llvm::AtomicOrdering::Release); |
3688 | return RValue::get(nullptr); |
3689 | } |
3690 | |
3691 | case Builtin::BI__sync_synchronize: { |
3692 | // We assume this is supposed to correspond to a C++0x-style |
3693 | // sequentially-consistent fence (i.e. this is only usable for |
3694 | // synchronization, not device I/O or anything like that). This intrinsic |
3695 | // is really badly designed in the sense that in theory, there isn't |
3696 | // any way to safely use it... but in practice, it mostly works |
3697 | // to use it with non-atomic loads and stores to get acquire/release |
3698 | // semantics. |
3699 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent); |
3700 | return RValue::get(nullptr); |
3701 | } |
3702 | |
3703 | case Builtin::BI__builtin_nontemporal_load: |
3704 | return RValue::get(EmitNontemporalLoad(*this, E)); |
3705 | case Builtin::BI__builtin_nontemporal_store: |
3706 | return RValue::get(EmitNontemporalStore(*this, E)); |
3707 | case Builtin::BI__c11_atomic_is_lock_free: |
3708 | case Builtin::BI__atomic_is_lock_free: { |
3709 | // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the |
3710 | // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since |
3711 | // _Atomic(T) is always properly-aligned. |
3712 | const char *LibCallName = "__atomic_is_lock_free"; |
3713 | CallArgList Args; |
3714 | Args.add(RValue::get(EmitScalarExpr(E->getArg(0))), |
3715 | getContext().getSizeType()); |
3716 | if (BuiltinID == Builtin::BI__atomic_is_lock_free) |
3717 | Args.add(RValue::get(EmitScalarExpr(E->getArg(1))), |
3718 | getContext().VoidPtrTy); |
3719 | else |
3720 | Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)), |
3721 | getContext().VoidPtrTy); |
3722 | const CGFunctionInfo &FuncInfo = |
3723 | CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args); |
3724 | llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo); |
3725 | llvm::FunctionCallee Func = CGM.CreateRuntimeFunction(FTy, LibCallName); |
3726 | return EmitCall(FuncInfo, CGCallee::forDirect(Func), |
3727 | ReturnValueSlot(), Args); |
3728 | } |
3729 | |
3730 | case Builtin::BI__atomic_test_and_set: { |
3731 | // Look at the argument type to determine whether this is a volatile |
3732 | // operation. The parameter type is always volatile. |
3733 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); |
3734 | bool Volatile = |
3735 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); |
3736 | |
3737 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
3738 | unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace(); |
3739 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); |
3740 | Value *NewVal = Builder.getInt8(1); |
3741 | Value *Order = EmitScalarExpr(E->getArg(1)); |
3742 | if (isa<llvm::ConstantInt>(Order)) { |
3743 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
3744 | AtomicRMWInst *Result = nullptr; |
3745 | switch (ord) { |
3746 | case 0: // memory_order_relaxed |
3747 | default: // invalid order |
3748 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
3749 | llvm::AtomicOrdering::Monotonic); |
3750 | break; |
3751 | case 1: // memory_order_consume |
3752 | case 2: // memory_order_acquire |
3753 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
3754 | llvm::AtomicOrdering::Acquire); |
3755 | break; |
3756 | case 3: // memory_order_release |
3757 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
3758 | llvm::AtomicOrdering::Release); |
3759 | break; |
3760 | case 4: // memory_order_acq_rel |
3761 | |
3762 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
3763 | llvm::AtomicOrdering::AcquireRelease); |
3764 | break; |
3765 | case 5: // memory_order_seq_cst |
3766 | Result = Builder.CreateAtomicRMW( |
3767 | llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
3768 | llvm::AtomicOrdering::SequentiallyConsistent); |
3769 | break; |
3770 | } |
3771 | Result->setVolatile(Volatile); |
3772 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); |
3773 | } |
3774 | |
3775 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
3776 | |
3777 | llvm::BasicBlock *BBs[5] = { |
3778 | createBasicBlock("monotonic", CurFn), |
3779 | createBasicBlock("acquire", CurFn), |
3780 | createBasicBlock("release", CurFn), |
3781 | createBasicBlock("acqrel", CurFn), |
3782 | createBasicBlock("seqcst", CurFn) |
3783 | }; |
3784 | llvm::AtomicOrdering Orders[5] = { |
3785 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire, |
3786 | llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease, |
3787 | llvm::AtomicOrdering::SequentiallyConsistent}; |
3788 | |
3789 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
3790 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); |
3791 | |
3792 | Builder.SetInsertPoint(ContBB); |
3793 | PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set"); |
3794 | |
3795 | for (unsigned i = 0; i < 5; ++i) { |
3796 | Builder.SetInsertPoint(BBs[i]); |
3797 | AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, |
3798 | Ptr, NewVal, Orders[i]); |
3799 | RMW->setVolatile(Volatile); |
3800 | Result->addIncoming(RMW, BBs[i]); |
3801 | Builder.CreateBr(ContBB); |
3802 | } |
3803 | |
3804 | SI->addCase(Builder.getInt32(0), BBs[0]); |
3805 | SI->addCase(Builder.getInt32(1), BBs[1]); |
3806 | SI->addCase(Builder.getInt32(2), BBs[1]); |
3807 | SI->addCase(Builder.getInt32(3), BBs[2]); |
3808 | SI->addCase(Builder.getInt32(4), BBs[3]); |
3809 | SI->addCase(Builder.getInt32(5), BBs[4]); |
3810 | |
3811 | Builder.SetInsertPoint(ContBB); |
3812 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); |
3813 | } |
3814 | |
3815 | case Builtin::BI__atomic_clear: { |
3816 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); |
3817 | bool Volatile = |
3818 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); |
3819 | |
3820 | Address Ptr = EmitPointerWithAlignment(E->getArg(0)); |
3821 | unsigned AddrSpace = Ptr.getPointer()->getType()->getPointerAddressSpace(); |
3822 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); |
3823 | Value *NewVal = Builder.getInt8(0); |
3824 | Value *Order = EmitScalarExpr(E->getArg(1)); |
3825 | if (isa<llvm::ConstantInt>(Order)) { |
3826 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
3827 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); |
3828 | switch (ord) { |
3829 | case 0: // memory_order_relaxed |
3830 | default: // invalid order |
3831 | Store->setOrdering(llvm::AtomicOrdering::Monotonic); |
3832 | break; |
3833 | case 3: // memory_order_release |
3834 | Store->setOrdering(llvm::AtomicOrdering::Release); |
3835 | break; |
3836 | case 5: // memory_order_seq_cst |
3837 | Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent); |
3838 | break; |
3839 | } |
3840 | return RValue::get(nullptr); |
3841 | } |
3842 | |
3843 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
3844 | |
3845 | llvm::BasicBlock *BBs[3] = { |
3846 | createBasicBlock("monotonic", CurFn), |
3847 | createBasicBlock("release", CurFn), |
3848 | createBasicBlock("seqcst", CurFn) |
3849 | }; |
3850 | llvm::AtomicOrdering Orders[3] = { |
3851 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release, |
3852 | llvm::AtomicOrdering::SequentiallyConsistent}; |
3853 | |
3854 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
3855 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); |
3856 | |
3857 | for (unsigned i = 0; i < 3; ++i) { |
3858 | Builder.SetInsertPoint(BBs[i]); |
3859 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); |
3860 | Store->setOrdering(Orders[i]); |
3861 | Builder.CreateBr(ContBB); |
3862 | } |
3863 | |
3864 | SI->addCase(Builder.getInt32(0), BBs[0]); |
3865 | SI->addCase(Builder.getInt32(3), BBs[1]); |
3866 | SI->addCase(Builder.getInt32(5), BBs[2]); |
3867 | |
3868 | Builder.SetInsertPoint(ContBB); |
3869 | return RValue::get(nullptr); |
3870 | } |
3871 | |
3872 | case Builtin::BI__atomic_thread_fence: |
3873 | case Builtin::BI__atomic_signal_fence: |
3874 | case Builtin::BI__c11_atomic_thread_fence: |
3875 | case Builtin::BI__c11_atomic_signal_fence: { |
3876 | llvm::SyncScope::ID SSID; |
3877 | if (BuiltinID == Builtin::BI__atomic_signal_fence || |
3878 | BuiltinID == Builtin::BI__c11_atomic_signal_fence) |
3879 | SSID = llvm::SyncScope::SingleThread; |
3880 | else |
3881 | SSID = llvm::SyncScope::System; |
3882 | Value *Order = EmitScalarExpr(E->getArg(0)); |
3883 | if (isa<llvm::ConstantInt>(Order)) { |
3884 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
3885 | switch (ord) { |
3886 | case 0: // memory_order_relaxed |
3887 | default: // invalid order |
3888 | break; |
3889 | case 1: // memory_order_consume |
3890 | case 2: // memory_order_acquire |
3891 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); |
3892 | break; |
3893 | case 3: // memory_order_release |
3894 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); |
3895 | break; |
3896 | case 4: // memory_order_acq_rel |
3897 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); |
3898 | break; |
3899 | case 5: // memory_order_seq_cst |
3900 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); |
3901 | break; |
3902 | } |
3903 | return RValue::get(nullptr); |
3904 | } |
3905 | |
3906 | llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB; |
3907 | AcquireBB = createBasicBlock("acquire", CurFn); |
3908 | ReleaseBB = createBasicBlock("release", CurFn); |
3909 | AcqRelBB = createBasicBlock("acqrel", CurFn); |
3910 | SeqCstBB = createBasicBlock("seqcst", CurFn); |
3911 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
3912 | |
3913 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
3914 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB); |
3915 | |
3916 | Builder.SetInsertPoint(AcquireBB); |
3917 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); |
3918 | Builder.CreateBr(ContBB); |
3919 | SI->addCase(Builder.getInt32(1), AcquireBB); |
3920 | SI->addCase(Builder.getInt32(2), AcquireBB); |
3921 | |
3922 | Builder.SetInsertPoint(ReleaseBB); |
3923 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); |
3924 | Builder.CreateBr(ContBB); |
3925 | SI->addCase(Builder.getInt32(3), ReleaseBB); |
3926 | |
3927 | Builder.SetInsertPoint(AcqRelBB); |
3928 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); |
3929 | Builder.CreateBr(ContBB); |
3930 | SI->addCase(Builder.getInt32(4), AcqRelBB); |
3931 | |
3932 | Builder.SetInsertPoint(SeqCstBB); |
3933 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); |
3934 | Builder.CreateBr(ContBB); |
3935 | SI->addCase(Builder.getInt32(5), SeqCstBB); |
3936 | |
3937 | Builder.SetInsertPoint(ContBB); |
3938 | return RValue::get(nullptr); |
3939 | } |
3940 | |
3941 | case Builtin::BI__builtin_signbit: |
3942 | case Builtin::BI__builtin_signbitf: |
3943 | case Builtin::BI__builtin_signbitl: { |
3944 | return RValue::get( |
3945 | Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))), |
3946 | ConvertType(E->getType()))); |
3947 | } |
3948 | case Builtin::BI__warn_memset_zero_len: |
3949 | return RValue::getIgnored(); |
3950 | case Builtin::BI__annotation: { |
3951 | // Re-encode each wide string to UTF8 and make an MDString. |
3952 | SmallVector<Metadata *, 1> Strings; |
3953 | for (const Expr *Arg : E->arguments()) { |
3954 | const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts()); |
3955 | assert(Str->getCharByteWidth() == 2)((Str->getCharByteWidth() == 2) ? static_cast<void> ( 0) : __assert_fail ("Str->getCharByteWidth() == 2", "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 3955, __PRETTY_FUNCTION__)); |
3956 | StringRef WideBytes = Str->getBytes(); |
3957 | std::string StrUtf8; |
3958 | if (!convertUTF16ToUTF8String( |
3959 | makeArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) { |
3960 | CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument"); |
3961 | continue; |
3962 | } |
3963 | Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8)); |
3964 | } |
3965 | |
3966 | // Build and MDTuple of MDStrings and emit the intrinsic call. |
3967 | llvm::Function *F = |
3968 | CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {}); |
3969 | MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings); |
3970 | Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple)); |
3971 | return RValue::getIgnored(); |
3972 | } |
3973 | case Builtin::BI__builtin_annotation: { |
3974 | llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0)); |
3975 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::annotation, |
3976 | AnnVal->getType()); |
3977 | |
3978 | // Get the annotation string, go through casts. Sema requires this to be a |
3979 | // non-wide string literal, potentially casted, so the cast<> is safe. |
3980 | const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts(); |
3981 | StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString(); |
3982 | return RValue::get( |
3983 | EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc(), nullptr)); |
3984 | } |
3985 | case Builtin::BI__builtin_addcb: |
3986 | case Builtin::BI__builtin_addcs: |
3987 | case Builtin::BI__builtin_addc: |
3988 | case Builtin::BI__builtin_addcl: |
3989 | case Builtin::BI__builtin_addcll: |
3990 | case Builtin::BI__builtin_subcb: |
3991 | case Builtin::BI__builtin_subcs: |
3992 | case Builtin::BI__builtin_subc: |
3993 | case Builtin::BI__builtin_subcl: |
3994 | case Builtin::BI__builtin_subcll: { |
3995 | |
3996 | // We translate all of these builtins from expressions of the form: |
3997 | // int x = ..., y = ..., carryin = ..., carryout, result; |
3998 | // result = __builtin_addc(x, y, carryin, &carryout); |
3999 | // |
4000 | // to LLVM IR of the form: |
4001 | // |
4002 | // %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y) |
4003 | // %tmpsum1 = extractvalue {i32, i1} %tmp1, 0 |
4004 | // %carry1 = extractvalue {i32, i1} %tmp1, 1 |
4005 | // %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1, |
4006 | // i32 %carryin) |
4007 | // %result = extractvalue {i32, i1} %tmp2, 0 |
4008 | // %carry2 = extractvalue {i32, i1} %tmp2, 1 |
4009 | // %tmp3 = or i1 %carry1, %carry2 |
4010 | // %tmp4 = zext i1 %tmp3 to i32 |
4011 | // store i32 %tmp4, i32* %carryout |
4012 | |
4013 | // Scalarize our inputs. |
4014 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
4015 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
4016 | llvm::Value *Carryin = EmitScalarExpr(E->getArg(2)); |
4017 | Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3)); |
4018 | |
4019 | // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow. |
4020 | llvm::Intrinsic::ID IntrinsicId; |
4021 | switch (BuiltinID) { |
4022 | default: llvm_unreachable("Unknown multiprecision builtin id.")::llvm::llvm_unreachable_internal("Unknown multiprecision builtin id." , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 4022); |
4023 | case Builtin::BI__builtin_addcb: |
4024 | case Builtin::BI__builtin_addcs: |
4025 | case Builtin::BI__builtin_addc: |
4026 | case Builtin::BI__builtin_addcl: |
4027 | case Builtin::BI__builtin_addcll: |
4028 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; |
4029 | break; |
4030 | case Builtin::BI__builtin_subcb: |
4031 | case Builtin::BI__builtin_subcs: |
4032 | case Builtin::BI__builtin_subc: |
4033 | case Builtin::BI__builtin_subcl: |
4034 | case Builtin::BI__builtin_subcll: |
4035 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; |
4036 | break; |
4037 | } |
4038 | |
4039 | // Construct our resulting LLVM IR expression. |
4040 | llvm::Value *Carry1; |
4041 | llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId, |
4042 | X, Y, Carry1); |
4043 | llvm::Value *Carry2; |
4044 | llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId, |
4045 | Sum1, Carryin, Carry2); |
4046 | llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2), |
4047 | X->getType()); |
4048 | Builder.CreateStore(CarryOut, CarryOutPtr); |
4049 | return RValue::get(Sum2); |
4050 | } |
4051 | |
4052 | case Builtin::BI__builtin_add_overflow: |
4053 | case Builtin::BI__builtin_sub_overflow: |
4054 | case Builtin::BI__builtin_mul_overflow: { |
4055 | const clang::Expr *LeftArg = E->getArg(0); |
4056 | const clang::Expr *RightArg = E->getArg(1); |
4057 | const clang::Expr *ResultArg = E->getArg(2); |
4058 | |
4059 | clang::QualType ResultQTy = |
4060 | ResultArg->getType()->castAs<PointerType>()->getPointeeType(); |
4061 | |
4062 | WidthAndSignedness LeftInfo = |
4063 | getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType()); |
4064 | WidthAndSignedness RightInfo = |
4065 | getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType()); |
4066 | WidthAndSignedness ResultInfo = |
4067 | getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy); |
4068 | |
4069 | // Handle mixed-sign multiplication as a special case, because adding |
4070 | // runtime or backend support for our generic irgen would be too expensive. |
4071 | if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo)) |
4072 | return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg, |
4073 | RightInfo, ResultArg, ResultQTy, |
4074 | ResultInfo); |
4075 | |
4076 | if (isSpecialUnsignedMultiplySignedResult(BuiltinID, LeftInfo, RightInfo, |
4077 | ResultInfo)) |
4078 | return EmitCheckedUnsignedMultiplySignedResult( |
4079 | *this, LeftArg, LeftInfo, RightArg, RightInfo, ResultArg, ResultQTy, |
4080 | ResultInfo); |
4081 | |
4082 | WidthAndSignedness EncompassingInfo = |
4083 | EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo}); |
4084 | |
4085 | llvm::Type *EncompassingLLVMTy = |
4086 | llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width); |
4087 | |
4088 | llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy); |
4089 | |
4090 | llvm::Intrinsic::ID IntrinsicId; |
4091 | switch (BuiltinID) { |
4092 | default: |
4093 | llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 4093); |
4094 | case Builtin::BI__builtin_add_overflow: |
4095 | IntrinsicId = EncompassingInfo.Signed |
4096 | ? llvm::Intrinsic::sadd_with_overflow |
4097 | : llvm::Intrinsic::uadd_with_overflow; |
4098 | break; |
4099 | case Builtin::BI__builtin_sub_overflow: |
4100 | IntrinsicId = EncompassingInfo.Signed |
4101 | ? llvm::Intrinsic::ssub_with_overflow |
4102 | : llvm::Intrinsic::usub_with_overflow; |
4103 | break; |
4104 | case Builtin::BI__builtin_mul_overflow: |
4105 | IntrinsicId = EncompassingInfo.Signed |
4106 | ? llvm::Intrinsic::smul_with_overflow |
4107 | : llvm::Intrinsic::umul_with_overflow; |
4108 | break; |
4109 | } |
4110 | |
4111 | llvm::Value *Left = EmitScalarExpr(LeftArg); |
4112 | llvm::Value *Right = EmitScalarExpr(RightArg); |
4113 | Address ResultPtr = EmitPointerWithAlignment(ResultArg); |
4114 | |
4115 | // Extend each operand to the encompassing type. |
4116 | Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed); |
4117 | Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed); |
4118 | |
4119 | // Perform the operation on the extended values. |
4120 | llvm::Value *Overflow, *Result; |
4121 | Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow); |
4122 | |
4123 | if (EncompassingInfo.Width > ResultInfo.Width) { |
4124 | // The encompassing type is wider than the result type, so we need to |
4125 | // truncate it. |
4126 | llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy); |
4127 | |
4128 | // To see if the truncation caused an overflow, we will extend |
4129 | // the result and then compare it to the original result. |
4130 | llvm::Value *ResultTruncExt = Builder.CreateIntCast( |
4131 | ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed); |
4132 | llvm::Value *TruncationOverflow = |
4133 | Builder.CreateICmpNE(Result, ResultTruncExt); |
4134 | |
4135 | Overflow = Builder.CreateOr(Overflow, TruncationOverflow); |
4136 | Result = ResultTrunc; |
4137 | } |
4138 | |
4139 | // Finally, store the result using the pointer. |
4140 | bool isVolatile = |
4141 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
4142 | Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile); |
4143 | |
4144 | return RValue::get(Overflow); |
4145 | } |
4146 | |
4147 | case Builtin::BI__builtin_uadd_overflow: |
4148 | case Builtin::BI__builtin_uaddl_overflow: |
4149 | case Builtin::BI__builtin_uaddll_overflow: |
4150 | case Builtin::BI__builtin_usub_overflow: |
4151 | case Builtin::BI__builtin_usubl_overflow: |
4152 | case Builtin::BI__builtin_usubll_overflow: |
4153 | case Builtin::BI__builtin_umul_overflow: |
4154 | case Builtin::BI__builtin_umull_overflow: |
4155 | case Builtin::BI__builtin_umulll_overflow: |
4156 | case Builtin::BI__builtin_sadd_overflow: |
4157 | case Builtin::BI__builtin_saddl_overflow: |
4158 | case Builtin::BI__builtin_saddll_overflow: |
4159 | case Builtin::BI__builtin_ssub_overflow: |
4160 | case Builtin::BI__builtin_ssubl_overflow: |
4161 | case Builtin::BI__builtin_ssubll_overflow: |
4162 | case Builtin::BI__builtin_smul_overflow: |
4163 | case Builtin::BI__builtin_smull_overflow: |
4164 | case Builtin::BI__builtin_smulll_overflow: { |
4165 | |
4166 | // We translate all of these builtins directly to the relevant llvm IR node. |
4167 | |
4168 | // Scalarize our inputs. |
4169 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
4170 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
4171 | Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2)); |
4172 | |
4173 | // Decide which of the overflow intrinsics we are lowering to: |
4174 | llvm::Intrinsic::ID IntrinsicId; |
4175 | switch (BuiltinID) { |
4176 | default: llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 4176); |
4177 | case Builtin::BI__builtin_uadd_overflow: |
4178 | case Builtin::BI__builtin_uaddl_overflow: |
4179 | case Builtin::BI__builtin_uaddll_overflow: |
4180 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; |
4181 | break; |
4182 | case Builtin::BI__builtin_usub_overflow: |
4183 | case Builtin::BI__builtin_usubl_overflow: |
4184 | case Builtin::BI__builtin_usubll_overflow: |
4185 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; |
4186 | break; |
4187 | case Builtin::BI__builtin_umul_overflow: |
4188 | case Builtin::BI__builtin_umull_overflow: |
4189 | case Builtin::BI__builtin_umulll_overflow: |
4190 | IntrinsicId = llvm::Intrinsic::umul_with_overflow; |
4191 | break; |
4192 | case Builtin::BI__builtin_sadd_overflow: |
4193 | case Builtin::BI__builtin_saddl_overflow: |
4194 | case Builtin::BI__builtin_saddll_overflow: |
4195 | IntrinsicId = llvm::Intrinsic::sadd_with_overflow; |
4196 | break; |
4197 | case Builtin::BI__builtin_ssub_overflow: |
4198 | case Builtin::BI__builtin_ssubl_overflow: |
4199 | case Builtin::BI__builtin_ssubll_overflow: |
4200 | IntrinsicId = llvm::Intrinsic::ssub_with_overflow; |
4201 | break; |
4202 | case Builtin::BI__builtin_smul_overflow: |
4203 | case Builtin::BI__builtin_smull_overflow: |
4204 | case Builtin::BI__builtin_smulll_overflow: |
4205 | IntrinsicId = llvm::Intrinsic::smul_with_overflow; |
4206 | break; |
4207 | } |
4208 | |
4209 | |
4210 | llvm::Value *Carry; |
4211 | llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry); |
4212 | Builder.CreateStore(Sum, SumOutPtr); |
4213 | |
4214 | return RValue::get(Carry); |
4215 | } |
4216 | case Builtin::BI__builtin_addressof: |
4217 | return RValue::get(EmitLValue(E->getArg(0)).getPointer(*this)); |
4218 | case Builtin::BI__builtin_operator_new: |
4219 | return EmitBuiltinNewDeleteCall( |
4220 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false); |
4221 | case Builtin::BI__builtin_operator_delete: |
4222 | return EmitBuiltinNewDeleteCall( |
4223 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true); |
4224 | |
4225 | case Builtin::BI__builtin_is_aligned: |
4226 | return EmitBuiltinIsAligned(E); |
4227 | case Builtin::BI__builtin_align_up: |
4228 | return EmitBuiltinAlignTo(E, true); |
4229 | case Builtin::BI__builtin_align_down: |
4230 | return EmitBuiltinAlignTo(E, false); |
4231 | |
4232 | case Builtin::BI__noop: |
4233 | // __noop always evaluates to an integer literal zero. |
4234 | return RValue::get(ConstantInt::get(IntTy, 0)); |
4235 | case Builtin::BI__builtin_call_with_static_chain: { |
4236 | const CallExpr *Call = cast<CallExpr>(E->getArg(0)); |
4237 | const Expr *Chain = E->getArg(1); |
4238 | return EmitCall(Call->getCallee()->getType(), |
4239 | EmitCallee(Call->getCallee()), Call, ReturnValue, |
4240 | EmitScalarExpr(Chain)); |
4241 | } |
4242 | case Builtin::BI_InterlockedExchange8: |
4243 | case Builtin::BI_InterlockedExchange16: |
4244 | case Builtin::BI_InterlockedExchange: |
4245 | case Builtin::BI_InterlockedExchangePointer: |
4246 | return RValue::get( |
4247 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E)); |
4248 | case Builtin::BI_InterlockedCompareExchangePointer: |
4249 | case Builtin::BI_InterlockedCompareExchangePointer_nf: { |
4250 | llvm::Type *RTy; |
4251 | llvm::IntegerType *IntType = |
4252 | IntegerType::get(getLLVMContext(), |
4253 | getContext().getTypeSize(E->getType())); |
4254 | llvm::Type *IntPtrType = IntType->getPointerTo(); |
4255 | |
4256 | llvm::Value *Destination = |
4257 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType); |
4258 | |
4259 | llvm::Value *Exchange = EmitScalarExpr(E->getArg(1)); |
4260 | RTy = Exchange->getType(); |
4261 | Exchange = Builder.CreatePtrToInt(Exchange, IntType); |
4262 | |
4263 | llvm::Value *Comparand = |
4264 | Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType); |
4265 | |
4266 | auto Ordering = |
4267 | BuiltinID == Builtin::BI_InterlockedCompareExchangePointer_nf ? |
4268 | AtomicOrdering::Monotonic : AtomicOrdering::SequentiallyConsistent; |
4269 | |
4270 | auto Result = Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, |
4271 | Ordering, Ordering); |
4272 | Result->setVolatile(true); |
4273 | |
4274 | return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result, |
4275 | 0), |
4276 | RTy)); |
4277 | } |
4278 | case Builtin::BI_InterlockedCompareExchange8: |
4279 | case Builtin::BI_InterlockedCompareExchange16: |
4280 | case Builtin::BI_InterlockedCompareExchange: |
4281 | case Builtin::BI_InterlockedCompareExchange64: |
4282 | return RValue::get(EmitAtomicCmpXchgForMSIntrin(*this, E)); |
4283 | case Builtin::BI_InterlockedIncrement16: |
4284 | case Builtin::BI_InterlockedIncrement: |
4285 | return RValue::get( |
4286 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E)); |
4287 | case Builtin::BI_InterlockedDecrement16: |
4288 | case Builtin::BI_InterlockedDecrement: |
4289 | return RValue::get( |
4290 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E)); |
4291 | case Builtin::BI_InterlockedAnd8: |
4292 | case Builtin::BI_InterlockedAnd16: |
4293 | case Builtin::BI_InterlockedAnd: |
4294 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E)); |
4295 | case Builtin::BI_InterlockedExchangeAdd8: |
4296 | case Builtin::BI_InterlockedExchangeAdd16: |
4297 | case Builtin::BI_InterlockedExchangeAdd: |
4298 | return RValue::get( |
4299 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E)); |
4300 | case Builtin::BI_InterlockedExchangeSub8: |
4301 | case Builtin::BI_InterlockedExchangeSub16: |
4302 | case Builtin::BI_InterlockedExchangeSub: |
4303 | return RValue::get( |
4304 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E)); |
4305 | case Builtin::BI_InterlockedOr8: |
4306 | case Builtin::BI_InterlockedOr16: |
4307 | case Builtin::BI_InterlockedOr: |
4308 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E)); |
4309 | case Builtin::BI_InterlockedXor8: |
4310 | case Builtin::BI_InterlockedXor16: |
4311 | case Builtin::BI_InterlockedXor: |
4312 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E)); |
4313 | |
4314 | case Builtin::BI_bittest64: |
4315 | case Builtin::BI_bittest: |
4316 | case Builtin::BI_bittestandcomplement64: |
4317 | case Builtin::BI_bittestandcomplement: |
4318 | case Builtin::BI_bittestandreset64: |
4319 | case Builtin::BI_bittestandreset: |
4320 | case Builtin::BI_bittestandset64: |
4321 | case Builtin::BI_bittestandset: |
4322 | case Builtin::BI_interlockedbittestandreset: |
4323 | case Builtin::BI_interlockedbittestandreset64: |
4324 | case Builtin::BI_interlockedbittestandset64: |
4325 | case Builtin::BI_interlockedbittestandset: |
4326 | case Builtin::BI_interlockedbittestandset_acq: |
4327 | case Builtin::BI_interlockedbittestandset_rel: |
4328 | case Builtin::BI_interlockedbittestandset_nf: |
4329 | case Builtin::BI_interlockedbittestandreset_acq: |
4330 | case Builtin::BI_interlockedbittestandreset_rel: |
4331 | case Builtin::BI_interlockedbittestandreset_nf: |
4332 | return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E)); |
4333 | |
4334 | // These builtins exist to emit regular volatile loads and stores not |
4335 | // affected by the -fms-volatile setting. |
4336 | case Builtin::BI__iso_volatile_load8: |
4337 | case Builtin::BI__iso_volatile_load16: |
4338 | case Builtin::BI__iso_volatile_load32: |
4339 | case Builtin::BI__iso_volatile_load64: |
4340 | return RValue::get(EmitISOVolatileLoad(*this, E)); |
4341 | case Builtin::BI__iso_volatile_store8: |
4342 | case Builtin::BI__iso_volatile_store16: |
4343 | case Builtin::BI__iso_volatile_store32: |
4344 | case Builtin::BI__iso_volatile_store64: |
4345 | return RValue::get(EmitISOVolatileStore(*this, E)); |
4346 | |
4347 | case Builtin::BI__exception_code: |
4348 | case Builtin::BI_exception_code: |
4349 | return RValue::get(EmitSEHExceptionCode()); |
4350 | case Builtin::BI__exception_info: |
4351 | case Builtin::BI_exception_info: |
4352 | return RValue::get(EmitSEHExceptionInfo()); |
4353 | case Builtin::BI__abnormal_termination: |
4354 | case Builtin::BI_abnormal_termination: |
4355 | return RValue::get(EmitSEHAbnormalTermination()); |
4356 | case Builtin::BI_setjmpex: |
4357 | if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 && |
4358 | E->getArg(0)->getType()->isPointerType()) |
4359 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); |
4360 | break; |
4361 | case Builtin::BI_setjmp: |
4362 | if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 && |
4363 | E->getArg(0)->getType()->isPointerType()) { |
4364 | if (getTarget().getTriple().getArch() == llvm::Triple::x86) |
4365 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E); |
4366 | else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64) |
4367 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); |
4368 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E); |
4369 | } |
4370 | break; |
4371 | |
4372 | case Builtin::BI__GetExceptionInfo: { |
4373 | if (llvm::GlobalVariable *GV = |
4374 | CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType())) |
4375 | return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy)); |
4376 | break; |
4377 | } |
4378 | |
4379 | case Builtin::BI__fastfail: |
4380 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E)); |
4381 | |
4382 | case Builtin::BI__builtin_coro_size: { |
4383 | auto & Context = getContext(); |
4384 | auto SizeTy = Context.getSizeType(); |
4385 | auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy)); |
4386 | Function *F = CGM.getIntrinsic(Intrinsic::coro_size, T); |
4387 | return RValue::get(Builder.CreateCall(F)); |
4388 | } |
4389 | |
4390 | case Builtin::BI__builtin_coro_id: |
4391 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_id); |
4392 | case Builtin::BI__builtin_coro_promise: |
4393 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise); |
4394 | case Builtin::BI__builtin_coro_resume: |
4395 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume); |
4396 | case Builtin::BI__builtin_coro_frame: |
4397 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame); |
4398 | case Builtin::BI__builtin_coro_noop: |
4399 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop); |
4400 | case Builtin::BI__builtin_coro_free: |
4401 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_free); |
4402 | case Builtin::BI__builtin_coro_destroy: |
4403 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy); |
4404 | case Builtin::BI__builtin_coro_done: |
4405 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_done); |
4406 | case Builtin::BI__builtin_coro_alloc: |
4407 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc); |
4408 | case Builtin::BI__builtin_coro_begin: |
4409 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin); |
4410 | case Builtin::BI__builtin_coro_end: |
4411 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_end); |
4412 | case Builtin::BI__builtin_coro_suspend: |
4413 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend); |
4414 | case Builtin::BI__builtin_coro_param: |
4415 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_param); |
4416 | |
4417 | // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions |
4418 | case Builtin::BIread_pipe: |
4419 | case Builtin::BIwrite_pipe: { |
4420 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
4421 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
4422 | CGOpenCLRuntime OpenCLRT(CGM); |
4423 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
4424 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
4425 | |
4426 | // Type of the generic packet parameter. |
4427 | unsigned GenericAS = |
4428 | getContext().getTargetAddressSpace(LangAS::opencl_generic); |
4429 | llvm::Type *I8PTy = llvm::PointerType::get( |
4430 | llvm::Type::getInt8Ty(getLLVMContext()), GenericAS); |
4431 | |
4432 | // Testing which overloaded version we should generate the call for. |
4433 | if (2U == E->getNumArgs()) { |
4434 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2" |
4435 | : "__write_pipe_2"; |
4436 | // Creating a generic function type to be able to call with any builtin or |
4437 | // user defined type. |
4438 | llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty}; |
4439 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
4440 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
4441 | Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy); |
4442 | return RValue::get( |
4443 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
4444 | {Arg0, BCast, PacketSize, PacketAlign})); |
4445 | } else { |
4446 | assert(4 == E->getNumArgs() &&((4 == E->getNumArgs() && "Illegal number of parameters to pipe function" ) ? static_cast<void> (0) : __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 4447, __PRETTY_FUNCTION__)) |
4447 | "Illegal number of parameters to pipe function")((4 == E->getNumArgs() && "Illegal number of parameters to pipe function" ) ? static_cast<void> (0) : __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 4447, __PRETTY_FUNCTION__)); |
4448 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4" |
4449 | : "__write_pipe_4"; |
4450 | |
4451 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy, |
4452 | Int32Ty, Int32Ty}; |
4453 | Value *Arg2 = EmitScalarExpr(E->getArg(2)), |
4454 | *Arg3 = EmitScalarExpr(E->getArg(3)); |
4455 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
4456 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
4457 | Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy); |
4458 | // We know the third argument is an integer type, but we may need to cast |
4459 | // it to i32. |
4460 | if (Arg2->getType() != Int32Ty) |
4461 | Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty); |
4462 | return RValue::get( |
4463 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
4464 | {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign})); |
4465 | } |
4466 | } |
4467 | // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write |
4468 | // functions |
4469 | case Builtin::BIreserve_read_pipe: |
4470 | case Builtin::BIreserve_write_pipe: |
4471 | case Builtin::BIwork_group_reserve_read_pipe: |
4472 | case Builtin::BIwork_group_reserve_write_pipe: |
4473 | case Builtin::BIsub_group_reserve_read_pipe: |
4474 | case Builtin::BIsub_group_reserve_write_pipe: { |
4475 | // Composing the mangled name for the function. |
4476 | const char *Name; |
4477 | if (BuiltinID == Builtin::BIreserve_read_pipe) |
4478 | Name = "__reserve_read_pipe"; |
4479 | else if (BuiltinID == Builtin::BIreserve_write_pipe) |
4480 | Name = "__reserve_write_pipe"; |
4481 | else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe) |
4482 | Name = "__work_group_reserve_read_pipe"; |
4483 | else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe) |
4484 | Name = "__work_group_reserve_write_pipe"; |
4485 | else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe) |
4486 | Name = "__sub_group_reserve_read_pipe"; |
4487 | else |
4488 | Name = "__sub_group_reserve_write_pipe"; |
4489 | |
4490 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
4491 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
4492 | llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy); |
4493 | CGOpenCLRuntime OpenCLRT(CGM); |
4494 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
4495 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
4496 | |
4497 | // Building the generic function prototype. |
4498 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty}; |
4499 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
4500 | ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
4501 | // We know the second argument is an integer type, but we may need to cast |
4502 | // it to i32. |
4503 | if (Arg1->getType() != Int32Ty) |
4504 | Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty); |
4505 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
4506 | {Arg0, Arg1, PacketSize, PacketAlign})); |
4507 | } |
4508 | // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write |
4509 | // functions |
4510 | case Builtin::BIcommit_read_pipe: |
4511 | case Builtin::BIcommit_write_pipe: |
4512 | case Builtin::BIwork_group_commit_read_pipe: |
4513 | case Builtin::BIwork_group_commit_write_pipe: |
4514 | case Builtin::BIsub_group_commit_read_pipe: |
4515 | case Builtin::BIsub_group_commit_write_pipe: { |
4516 | const char *Name; |
4517 | if (BuiltinID == Builtin::BIcommit_read_pipe) |
4518 | Name = "__commit_read_pipe"; |
4519 | else if (BuiltinID == Builtin::BIcommit_write_pipe) |
4520 | Name = "__commit_write_pipe"; |
4521 | else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe) |
4522 | Name = "__work_group_commit_read_pipe"; |
4523 | else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe) |
4524 | Name = "__work_group_commit_write_pipe"; |
4525 | else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe) |
4526 | Name = "__sub_group_commit_read_pipe"; |
4527 | else |
4528 | Name = "__sub_group_commit_write_pipe"; |
4529 | |
4530 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
4531 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
4532 | CGOpenCLRuntime OpenCLRT(CGM); |
4533 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
4534 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
4535 | |
4536 | // Building the generic function prototype. |
4537 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty}; |
4538 | llvm::FunctionType *FTy = |
4539 | llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()), |
4540 | llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
4541 | |
4542 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
4543 | {Arg0, Arg1, PacketSize, PacketAlign})); |
4544 | } |
4545 | // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions |
4546 | case Builtin::BIget_pipe_num_packets: |
4547 | case Builtin::BIget_pipe_max_packets: { |
4548 | const char *BaseName; |
4549 | const auto *PipeTy = E->getArg(0)->getType()->castAs<PipeType>(); |
4550 | if (BuiltinID == Builtin::BIget_pipe_num_packets) |
4551 | BaseName = "__get_pipe_num_packets"; |
4552 | else |
4553 | BaseName = "__get_pipe_max_packets"; |
4554 | std::string Name = std::string(BaseName) + |
4555 | std::string(PipeTy->isReadOnly() ? "_ro" : "_wo"); |
4556 | |
4557 | // Building the generic function prototype. |
4558 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); |
4559 | CGOpenCLRuntime OpenCLRT(CGM); |
4560 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
4561 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
4562 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty}; |
4563 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
4564 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
4565 | |
4566 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
4567 | {Arg0, PacketSize, PacketAlign})); |
4568 | } |
4569 | |
4570 | // OpenCL v2.0 s6.13.9 - Address space qualifier functions. |
4571 | case Builtin::BIto_global: |
4572 | case Builtin::BIto_local: |
4573 | case Builtin::BIto_private: { |
4574 | auto Arg0 = EmitScalarExpr(E->getArg(0)); |
4575 | auto NewArgT = llvm::PointerType::get(Int8Ty, |
4576 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
4577 | auto NewRetT = llvm::PointerType::get(Int8Ty, |
4578 | CGM.getContext().getTargetAddressSpace( |
4579 | E->getType()->getPointeeType().getAddressSpace())); |
4580 | auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false); |
4581 | llvm::Value *NewArg; |
4582 | if (Arg0->getType()->getPointerAddressSpace() != |
4583 | NewArgT->getPointerAddressSpace()) |
4584 | NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT); |
4585 | else |
4586 | NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT); |
4587 | auto NewName = std::string("__") + E->getDirectCallee()->getName().str(); |
4588 | auto NewCall = |
4589 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg}); |
4590 | return RValue::get(Builder.CreateBitOrPointerCast(NewCall, |
4591 | ConvertType(E->getType()))); |
4592 | } |
4593 | |
4594 | // OpenCL v2.0, s6.13.17 - Enqueue kernel function. |
4595 | // It contains four different overload formats specified in Table 6.13.17.1. |
4596 | case Builtin::BIenqueue_kernel: { |
4597 | StringRef Name; // Generated function call name |
4598 | unsigned NumArgs = E->getNumArgs(); |
4599 | |
4600 | llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy); |
4601 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
4602 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
4603 | |
4604 | llvm::Value *Queue = EmitScalarExpr(E->getArg(0)); |
4605 | llvm::Value *Flags = EmitScalarExpr(E->getArg(1)); |
4606 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(2)); |
4607 | llvm::Value *Range = NDRangeL.getAddress(*this).getPointer(); |
4608 | llvm::Type *RangeTy = NDRangeL.getAddress(*this).getType(); |
4609 | |
4610 | if (NumArgs == 4) { |
4611 | // The most basic form of the call with parameters: |
4612 | // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void) |
4613 | Name = "__enqueue_kernel_basic"; |
4614 | llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy, |
4615 | GenericVoidPtrTy}; |
4616 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
4617 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
4618 | |
4619 | auto Info = |
4620 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); |
4621 | llvm::Value *Kernel = |
4622 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
4623 | llvm::Value *Block = |
4624 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
4625 | |
4626 | AttrBuilder B; |
4627 | B.addByValAttr(NDRangeL.getAddress(*this).getElementType()); |
4628 | llvm::AttributeList ByValAttrSet = |
4629 | llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B); |
4630 | |
4631 | auto RTCall = |
4632 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet), |
4633 | {Queue, Flags, Range, Kernel, Block}); |
4634 | RTCall->setAttributes(ByValAttrSet); |
4635 | return RValue::get(RTCall); |
4636 | } |
4637 | assert(NumArgs >= 5 && "Invalid enqueue_kernel signature")((NumArgs >= 5 && "Invalid enqueue_kernel signature" ) ? static_cast<void> (0) : __assert_fail ("NumArgs >= 5 && \"Invalid enqueue_kernel signature\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 4637, __PRETTY_FUNCTION__)); |
4638 | |
4639 | // Create a temporary array to hold the sizes of local pointer arguments |
4640 | // for the block. \p First is the position of the first size argument. |
4641 | auto CreateArrayForSizeVar = [=](unsigned First) |
4642 | -> std::tuple<llvm::Value *, llvm::Value *, llvm::Value *> { |
4643 | llvm::APInt ArraySize(32, NumArgs - First); |
4644 | QualType SizeArrayTy = getContext().getConstantArrayType( |
4645 | getContext().getSizeType(), ArraySize, nullptr, ArrayType::Normal, |
4646 | /*IndexTypeQuals=*/0); |
4647 | auto Tmp = CreateMemTemp(SizeArrayTy, "block_sizes"); |
4648 | llvm::Value *TmpPtr = Tmp.getPointer(); |
4649 | llvm::Value *TmpSize = EmitLifetimeStart( |
4650 | CGM.getDataLayout().getTypeAllocSize(Tmp.getElementType()), TmpPtr); |
4651 | llvm::Value *ElemPtr; |
4652 | // Each of the following arguments specifies the size of the corresponding |
4653 | // argument passed to the enqueued block. |
4654 | auto *Zero = llvm::ConstantInt::get(IntTy, 0); |
4655 | for (unsigned I = First; I < NumArgs; ++I) { |
4656 | auto *Index = llvm::ConstantInt::get(IntTy, I - First); |
4657 | auto *GEP = Builder.CreateGEP(TmpPtr, {Zero, Index}); |
4658 | if (I == First) |
4659 | ElemPtr = GEP; |
4660 | auto *V = |
4661 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy); |
4662 | Builder.CreateAlignedStore( |
4663 | V, GEP, CGM.getDataLayout().getPrefTypeAlign(SizeTy)); |
4664 | } |
4665 | return std::tie(ElemPtr, TmpSize, TmpPtr); |
4666 | }; |
4667 | |
4668 | // Could have events and/or varargs. |
4669 | if (E->getArg(3)->getType()->isBlockPointerType()) { |
4670 | // No events passed, but has variadic arguments. |
4671 | Name = "__enqueue_kernel_varargs"; |
4672 | auto Info = |
4673 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); |
4674 | llvm::Value *Kernel = |
4675 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
4676 | auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
4677 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; |
4678 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(4); |
4679 | |
4680 | // Create a vector of the arguments, as well as a constant value to |
4681 | // express to the runtime the number of variadic arguments. |
4682 | llvm::Value *const Args[] = {Queue, Flags, |
4683 | Range, Kernel, |
4684 | Block, ConstantInt::get(IntTy, NumArgs - 4), |
4685 | ElemPtr}; |
4686 | llvm::Type *const ArgTys[] = { |
4687 | QueueTy, IntTy, RangeTy, GenericVoidPtrTy, |
4688 | GenericVoidPtrTy, IntTy, ElemPtr->getType()}; |
4689 | |
4690 | llvm::FunctionType *FTy = llvm::FunctionType::get(Int32Ty, ArgTys, false); |
4691 | auto Call = RValue::get( |
4692 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Args)); |
4693 | if (TmpSize) |
4694 | EmitLifetimeEnd(TmpSize, TmpPtr); |
4695 | return Call; |
4696 | } |
4697 | // Any calls now have event arguments passed. |
4698 | if (NumArgs >= 7) { |
4699 | llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy); |
4700 | llvm::PointerType *EventPtrTy = EventTy->getPointerTo( |
4701 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
4702 | |
4703 | llvm::Value *NumEvents = |
4704 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty); |
4705 | |
4706 | // Since SemaOpenCLBuiltinEnqueueKernel allows fifth and sixth arguments |
4707 | // to be a null pointer constant (including `0` literal), we can take it |
4708 | // into account and emit null pointer directly. |
4709 | llvm::Value *EventWaitList = nullptr; |
4710 | if (E->getArg(4)->isNullPointerConstant( |
4711 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { |
4712 | EventWaitList = llvm::ConstantPointerNull::get(EventPtrTy); |
4713 | } else { |
4714 | EventWaitList = E->getArg(4)->getType()->isArrayType() |
4715 | ? EmitArrayToPointerDecay(E->getArg(4)).getPointer() |
4716 | : EmitScalarExpr(E->getArg(4)); |
4717 | // Convert to generic address space. |
4718 | EventWaitList = Builder.CreatePointerCast(EventWaitList, EventPtrTy); |
4719 | } |
4720 | llvm::Value *EventRet = nullptr; |
4721 | if (E->getArg(5)->isNullPointerConstant( |
4722 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { |
4723 | EventRet = llvm::ConstantPointerNull::get(EventPtrTy); |
4724 | } else { |
4725 | EventRet = |
4726 | Builder.CreatePointerCast(EmitScalarExpr(E->getArg(5)), EventPtrTy); |
4727 | } |
4728 | |
4729 | auto Info = |
4730 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6)); |
4731 | llvm::Value *Kernel = |
4732 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
4733 | llvm::Value *Block = |
4734 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
4735 | |
4736 | std::vector<llvm::Type *> ArgTys = { |
4737 | QueueTy, Int32Ty, RangeTy, Int32Ty, |
4738 | EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy}; |
4739 | |
4740 | std::vector<llvm::Value *> Args = {Queue, Flags, Range, |
4741 | NumEvents, EventWaitList, EventRet, |
4742 | Kernel, Block}; |
4743 | |
4744 | if (NumArgs == 7) { |
4745 | // Has events but no variadics. |
4746 | Name = "__enqueue_kernel_basic_events"; |
4747 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
4748 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
4749 | return RValue::get( |
4750 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
4751 | llvm::ArrayRef<llvm::Value *>(Args))); |
4752 | } |
4753 | // Has event info and variadics |
4754 | // Pass the number of variadics to the runtime function too. |
4755 | Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7)); |
4756 | ArgTys.push_back(Int32Ty); |
4757 | Name = "__enqueue_kernel_events_varargs"; |
4758 | |
4759 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; |
4760 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(7); |
4761 | Args.push_back(ElemPtr); |
4762 | ArgTys.push_back(ElemPtr->getType()); |
4763 | |
4764 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
4765 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
4766 | auto Call = |
4767 | RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
4768 | llvm::ArrayRef<llvm::Value *>(Args))); |
4769 | if (TmpSize) |
4770 | EmitLifetimeEnd(TmpSize, TmpPtr); |
4771 | return Call; |
4772 | } |
4773 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; |
4774 | } |
4775 | // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block |
4776 | // parameter. |
4777 | case Builtin::BIget_kernel_work_group_size: { |
4778 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
4779 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
4780 | auto Info = |
4781 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); |
4782 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
4783 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
4784 | return RValue::get(EmitRuntimeCall( |
4785 | CGM.CreateRuntimeFunction( |
4786 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, |
4787 | false), |
4788 | "__get_kernel_work_group_size_impl"), |
4789 | {Kernel, Arg})); |
4790 | } |
4791 | case Builtin::BIget_kernel_preferred_work_group_size_multiple: { |
4792 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
4793 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
4794 | auto Info = |
4795 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); |
4796 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
4797 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
4798 | return RValue::get(EmitRuntimeCall( |
4799 | CGM.CreateRuntimeFunction( |
4800 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, |
4801 | false), |
4802 | "__get_kernel_preferred_work_group_size_multiple_impl"), |
4803 | {Kernel, Arg})); |
4804 | } |
4805 | case Builtin::BIget_kernel_max_sub_group_size_for_ndrange: |
4806 | case Builtin::BIget_kernel_sub_group_count_for_ndrange: { |
4807 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
4808 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
4809 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(0)); |
4810 | llvm::Value *NDRange = NDRangeL.getAddress(*this).getPointer(); |
4811 | auto Info = |
4812 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1)); |
4813 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
4814 | Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
4815 | const char *Name = |
4816 | BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange |
4817 | ? "__get_kernel_max_sub_group_size_for_ndrange_impl" |
4818 | : "__get_kernel_sub_group_count_for_ndrange_impl"; |
4819 | return RValue::get(EmitRuntimeCall( |
4820 | CGM.CreateRuntimeFunction( |
4821 | llvm::FunctionType::get( |
4822 | IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy}, |
4823 | false), |
4824 | Name), |
4825 | {NDRange, Kernel, Block})); |
4826 | } |
4827 | |
4828 | case Builtin::BI__builtin_store_half: |
4829 | case Builtin::BI__builtin_store_halff: { |
4830 | Value *Val = EmitScalarExpr(E->getArg(0)); |
4831 | Address Address = EmitPointerWithAlignment(E->getArg(1)); |
4832 | Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy()); |
4833 | return RValue::get(Builder.CreateStore(HalfVal, Address)); |
4834 | } |
4835 | case Builtin::BI__builtin_load_half: { |
4836 | Address Address = EmitPointerWithAlignment(E->getArg(0)); |
4837 | Value *HalfVal = Builder.CreateLoad(Address); |
4838 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy())); |
4839 | } |
4840 | case Builtin::BI__builtin_load_halff: { |
4841 | Address Address = EmitPointerWithAlignment(E->getArg(0)); |
4842 | Value *HalfVal = Builder.CreateLoad(Address); |
4843 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy())); |
4844 | } |
4845 | case Builtin::BIprintf: |
4846 | if (getTarget().getTriple().isNVPTX()) |
4847 | return EmitNVPTXDevicePrintfCallExpr(E, ReturnValue); |
4848 | if (getTarget().getTriple().getArch() == Triple::amdgcn && |
4849 | getLangOpts().HIP) |
4850 | return EmitAMDGPUDevicePrintfCallExpr(E, ReturnValue); |
4851 | break; |
4852 | case Builtin::BI__builtin_canonicalize: |
4853 | case Builtin::BI__builtin_canonicalizef: |
4854 | case Builtin::BI__builtin_canonicalizef16: |
4855 | case Builtin::BI__builtin_canonicalizel: |
4856 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize)); |
4857 | |
4858 | case Builtin::BI__builtin_thread_pointer: { |
4859 | if (!getContext().getTargetInfo().isTLSSupported()) |
4860 | CGM.ErrorUnsupported(E, "__builtin_thread_pointer"); |
4861 | // Fall through - it's already mapped to the intrinsic by GCCBuiltin. |
4862 | break; |
4863 | } |
4864 | case Builtin::BI__builtin_os_log_format: |
4865 | return emitBuiltinOSLogFormat(*E); |
4866 | |
4867 | case Builtin::BI__xray_customevent: { |
4868 | if (!ShouldXRayInstrumentFunction()) |
4869 | return RValue::getIgnored(); |
4870 | |
4871 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( |
4872 | XRayInstrKind::Custom)) |
4873 | return RValue::getIgnored(); |
4874 | |
4875 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) |
4876 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents()) |
4877 | return RValue::getIgnored(); |
4878 | |
4879 | Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent); |
4880 | auto FTy = F->getFunctionType(); |
4881 | auto Arg0 = E->getArg(0); |
4882 | auto Arg0Val = EmitScalarExpr(Arg0); |
4883 | auto Arg0Ty = Arg0->getType(); |
4884 | auto PTy0 = FTy->getParamType(0); |
4885 | if (PTy0 != Arg0Val->getType()) { |
4886 | if (Arg0Ty->isArrayType()) |
4887 | Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer(); |
4888 | else |
4889 | Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0); |
4890 | } |
4891 | auto Arg1 = EmitScalarExpr(E->getArg(1)); |
4892 | auto PTy1 = FTy->getParamType(1); |
4893 | if (PTy1 != Arg1->getType()) |
4894 | Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1); |
4895 | return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1})); |
4896 | } |
4897 | |
4898 | case Builtin::BI__xray_typedevent: { |
4899 | // TODO: There should be a way to always emit events even if the current |
4900 | // function is not instrumented. Losing events in a stream can cripple |
4901 | // a trace. |
4902 | if (!ShouldXRayInstrumentFunction()) |
4903 | return RValue::getIgnored(); |
4904 | |
4905 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( |
4906 | XRayInstrKind::Typed)) |
4907 | return RValue::getIgnored(); |
4908 | |
4909 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) |
4910 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents()) |
4911 | return RValue::getIgnored(); |
4912 | |
4913 | Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent); |
4914 | auto FTy = F->getFunctionType(); |
4915 | auto Arg0 = EmitScalarExpr(E->getArg(0)); |
4916 | auto PTy0 = FTy->getParamType(0); |
4917 | if (PTy0 != Arg0->getType()) |
4918 | Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0); |
4919 | auto Arg1 = E->getArg(1); |
4920 | auto Arg1Val = EmitScalarExpr(Arg1); |
4921 | auto Arg1Ty = Arg1->getType(); |
4922 | auto PTy1 = FTy->getParamType(1); |
4923 | if (PTy1 != Arg1Val->getType()) { |
4924 | if (Arg1Ty->isArrayType()) |
4925 | Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer(); |
4926 | else |
4927 | Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1); |
4928 | } |
4929 | auto Arg2 = EmitScalarExpr(E->getArg(2)); |
4930 | auto PTy2 = FTy->getParamType(2); |
4931 | if (PTy2 != Arg2->getType()) |
4932 | Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2); |
4933 | return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2})); |
4934 | } |
4935 | |
4936 | case Builtin::BI__builtin_ms_va_start: |
4937 | case Builtin::BI__builtin_ms_va_end: |
4938 | return RValue::get( |
4939 | EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(), |
4940 | BuiltinID == Builtin::BI__builtin_ms_va_start)); |
4941 | |
4942 | case Builtin::BI__builtin_ms_va_copy: { |
4943 | // Lower this manually. We can't reliably determine whether or not any |
4944 | // given va_copy() is for a Win64 va_list from the calling convention |
4945 | // alone, because it's legal to do this from a System V ABI function. |
4946 | // With opaque pointer types, we won't have enough information in LLVM |
4947 | // IR to determine this from the argument types, either. Best to do it |
4948 | // now, while we have enough information. |
4949 | Address DestAddr = EmitMSVAListRef(E->getArg(0)); |
4950 | Address SrcAddr = EmitMSVAListRef(E->getArg(1)); |
4951 | |
4952 | llvm::Type *BPP = Int8PtrPtrTy; |
4953 | |
4954 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"), |
4955 | DestAddr.getAlignment()); |
4956 | SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"), |
4957 | SrcAddr.getAlignment()); |
4958 | |
4959 | Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val"); |
4960 | return RValue::get(Builder.CreateStore(ArgPtr, DestAddr)); |
4961 | } |
4962 | } |
4963 | |
4964 | // If this is an alias for a lib function (e.g. __builtin_sin), emit |
4965 | // the call using the normal call path, but using the unmangled |
4966 | // version of the function name. |
4967 | if (getContext().BuiltinInfo.isLibFunction(BuiltinID)) |
4968 | return emitLibraryCall(*this, FD, E, |
4969 | CGM.getBuiltinLibFunction(FD, BuiltinID)); |
4970 | |
4971 | // If this is a predefined lib function (e.g. malloc), emit the call |
4972 | // using exactly the normal call path. |
4973 | if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID)) |
4974 | return emitLibraryCall(*this, FD, E, |
4975 | cast<llvm::Constant>(EmitScalarExpr(E->getCallee()))); |
4976 | |
4977 | // Check that a call to a target specific builtin has the correct target |
4978 | // features. |
4979 | // This is down here to avoid non-target specific builtins, however, if |
4980 | // generic builtins start to require generic target features then we |
4981 | // can move this up to the beginning of the function. |
4982 | checkTargetFeatures(E, FD); |
4983 | |
4984 | if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID)) |
4985 | LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth); |
4986 | |
4987 | // See if we have a target specific intrinsic. |
4988 | const char *Name = getContext().BuiltinInfo.getName(BuiltinID); |
4989 | Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic; |
4990 | StringRef Prefix = |
4991 | llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch()); |
4992 | if (!Prefix.empty()) { |
4993 | IntrinsicID = Intrinsic::getIntrinsicForGCCBuiltin(Prefix.data(), Name); |
4994 | // NOTE we don't need to perform a compatibility flag check here since the |
4995 | // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the |
4996 | // MS builtins via ALL_MS_LANGUAGES and are filtered earlier. |
4997 | if (IntrinsicID == Intrinsic::not_intrinsic) |
4998 | IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name); |
4999 | } |
5000 | |
5001 | if (IntrinsicID != Intrinsic::not_intrinsic) { |
5002 | SmallVector<Value*, 16> Args; |
5003 | |
5004 | // Find out if any arguments are required to be integer constant |
5005 | // expressions. |
5006 | unsigned ICEArguments = 0; |
5007 | ASTContext::GetBuiltinTypeError Error; |
5008 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); |
5009 | assert(Error == ASTContext::GE_None && "Should not codegen an error")((Error == ASTContext::GE_None && "Should not codegen an error" ) ? static_cast<void> (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5009, __PRETTY_FUNCTION__)); |
5010 | |
5011 | Function *F = CGM.getIntrinsic(IntrinsicID); |
5012 | llvm::FunctionType *FTy = F->getFunctionType(); |
5013 | |
5014 | for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) { |
5015 | Value *ArgValue; |
5016 | // If this is a normal argument, just emit it as a scalar. |
5017 | if ((ICEArguments & (1 << i)) == 0) { |
5018 | ArgValue = EmitScalarExpr(E->getArg(i)); |
5019 | } else { |
5020 | // If this is required to be a constant, constant fold it so that we |
5021 | // know that the generated intrinsic gets a ConstantInt. |
5022 | ArgValue = llvm::ConstantInt::get( |
5023 | getLLVMContext(), |
5024 | *E->getArg(i)->getIntegerConstantExpr(getContext())); |
5025 | } |
5026 | |
5027 | // If the intrinsic arg type is different from the builtin arg type |
5028 | // we need to do a bit cast. |
5029 | llvm::Type *PTy = FTy->getParamType(i); |
5030 | if (PTy != ArgValue->getType()) { |
5031 | // XXX - vector of pointers? |
5032 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(PTy)) { |
5033 | if (PtrTy->getAddressSpace() != |
5034 | ArgValue->getType()->getPointerAddressSpace()) { |
5035 | ArgValue = Builder.CreateAddrSpaceCast( |
5036 | ArgValue, |
5037 | ArgValue->getType()->getPointerTo(PtrTy->getAddressSpace())); |
5038 | } |
5039 | } |
5040 | |
5041 | assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&((PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && "Must be able to losslessly bit cast to param") ? static_cast <void> (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5042, __PRETTY_FUNCTION__)) |
5042 | "Must be able to losslessly bit cast to param")((PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && "Must be able to losslessly bit cast to param") ? static_cast <void> (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5042, __PRETTY_FUNCTION__)); |
5043 | ArgValue = Builder.CreateBitCast(ArgValue, PTy); |
5044 | } |
5045 | |
5046 | Args.push_back(ArgValue); |
5047 | } |
5048 | |
5049 | Value *V = Builder.CreateCall(F, Args); |
5050 | QualType BuiltinRetType = E->getType(); |
5051 | |
5052 | llvm::Type *RetTy = VoidTy; |
5053 | if (!BuiltinRetType->isVoidType()) |
5054 | RetTy = ConvertType(BuiltinRetType); |
5055 | |
5056 | if (RetTy != V->getType()) { |
5057 | // XXX - vector of pointers? |
5058 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(RetTy)) { |
5059 | if (PtrTy->getAddressSpace() != V->getType()->getPointerAddressSpace()) { |
5060 | V = Builder.CreateAddrSpaceCast( |
5061 | V, V->getType()->getPointerTo(PtrTy->getAddressSpace())); |
5062 | } |
5063 | } |
5064 | |
5065 | assert(V->getType()->canLosslesslyBitCastTo(RetTy) &&((V->getType()->canLosslesslyBitCastTo(RetTy) && "Must be able to losslessly bit cast result type") ? static_cast <void> (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5066, __PRETTY_FUNCTION__)) |
5066 | "Must be able to losslessly bit cast result type")((V->getType()->canLosslesslyBitCastTo(RetTy) && "Must be able to losslessly bit cast result type") ? static_cast <void> (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5066, __PRETTY_FUNCTION__)); |
5067 | V = Builder.CreateBitCast(V, RetTy); |
5068 | } |
5069 | |
5070 | return RValue::get(V); |
5071 | } |
5072 | |
5073 | // Some target-specific builtins can have aggregate return values, e.g. |
5074 | // __builtin_arm_mve_vld2q_u32. So if the result is an aggregate, force |
5075 | // ReturnValue to be non-null, so that the target-specific emission code can |
5076 | // always just emit into it. |
5077 | TypeEvaluationKind EvalKind = getEvaluationKind(E->getType()); |
5078 | if (EvalKind == TEK_Aggregate && ReturnValue.isNull()) { |
5079 | Address DestPtr = CreateMemTemp(E->getType(), "agg.tmp"); |
5080 | ReturnValue = ReturnValueSlot(DestPtr, false); |
5081 | } |
5082 | |
5083 | // Now see if we can emit a target-specific builtin. |
5084 | if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E, ReturnValue)) { |
5085 | switch (EvalKind) { |
5086 | case TEK_Scalar: |
5087 | return RValue::get(V); |
5088 | case TEK_Aggregate: |
5089 | return RValue::getAggregate(ReturnValue.getValue(), |
5090 | ReturnValue.isVolatile()); |
5091 | case TEK_Complex: |
5092 | llvm_unreachable("No current target builtin returns complex")::llvm::llvm_unreachable_internal("No current target builtin returns complex" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5092); |
5093 | } |
5094 | llvm_unreachable("Bad evaluation kind in EmitBuiltinExpr")::llvm::llvm_unreachable_internal("Bad evaluation kind in EmitBuiltinExpr" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5094); |
5095 | } |
5096 | |
5097 | ErrorUnsupported(E, "builtin function"); |
5098 | |
5099 | // Unknown builtin, for now just dump it out and return undef. |
5100 | return GetUndefRValue(E->getType()); |
5101 | } |
5102 | |
5103 | static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF, |
5104 | unsigned BuiltinID, const CallExpr *E, |
5105 | ReturnValueSlot ReturnValue, |
5106 | llvm::Triple::ArchType Arch) { |
5107 | switch (Arch) { |
5108 | case llvm::Triple::arm: |
5109 | case llvm::Triple::armeb: |
5110 | case llvm::Triple::thumb: |
5111 | case llvm::Triple::thumbeb: |
5112 | return CGF->EmitARMBuiltinExpr(BuiltinID, E, ReturnValue, Arch); |
5113 | case llvm::Triple::aarch64: |
5114 | case llvm::Triple::aarch64_32: |
5115 | case llvm::Triple::aarch64_be: |
5116 | return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch); |
5117 | case llvm::Triple::bpfeb: |
5118 | case llvm::Triple::bpfel: |
5119 | return CGF->EmitBPFBuiltinExpr(BuiltinID, E); |
5120 | case llvm::Triple::x86: |
5121 | case llvm::Triple::x86_64: |
5122 | return CGF->EmitX86BuiltinExpr(BuiltinID, E); |
5123 | case llvm::Triple::ppc: |
5124 | case llvm::Triple::ppcle: |
5125 | case llvm::Triple::ppc64: |
5126 | case llvm::Triple::ppc64le: |
5127 | return CGF->EmitPPCBuiltinExpr(BuiltinID, E); |
5128 | case llvm::Triple::r600: |
5129 | case llvm::Triple::amdgcn: |
5130 | return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E); |
5131 | case llvm::Triple::systemz: |
5132 | return CGF->EmitSystemZBuiltinExpr(BuiltinID, E); |
5133 | case llvm::Triple::nvptx: |
5134 | case llvm::Triple::nvptx64: |
5135 | return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E); |
5136 | case llvm::Triple::wasm32: |
5137 | case llvm::Triple::wasm64: |
5138 | return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E); |
5139 | case llvm::Triple::hexagon: |
5140 | return CGF->EmitHexagonBuiltinExpr(BuiltinID, E); |
5141 | default: |
5142 | return nullptr; |
5143 | } |
5144 | } |
5145 | |
5146 | Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID, |
5147 | const CallExpr *E, |
5148 | ReturnValueSlot ReturnValue) { |
5149 | if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) { |
5150 | assert(getContext().getAuxTargetInfo() && "Missing aux target info")((getContext().getAuxTargetInfo() && "Missing aux target info" ) ? static_cast<void> (0) : __assert_fail ("getContext().getAuxTargetInfo() && \"Missing aux target info\"" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5150, __PRETTY_FUNCTION__)); |
5151 | return EmitTargetArchBuiltinExpr( |
5152 | this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E, |
5153 | ReturnValue, getContext().getAuxTargetInfo()->getTriple().getArch()); |
5154 | } |
5155 | |
5156 | return EmitTargetArchBuiltinExpr(this, BuiltinID, E, ReturnValue, |
5157 | getTarget().getTriple().getArch()); |
5158 | } |
5159 | |
5160 | static llvm::FixedVectorType *GetNeonType(CodeGenFunction *CGF, |
5161 | NeonTypeFlags TypeFlags, |
5162 | bool HasLegalHalfType = true, |
5163 | bool V1Ty = false, |
5164 | bool AllowBFloatArgsAndRet = true) { |
5165 | int IsQuad = TypeFlags.isQuad(); |
5166 | switch (TypeFlags.getEltType()) { |
5167 | case NeonTypeFlags::Int8: |
5168 | case NeonTypeFlags::Poly8: |
5169 | return llvm::FixedVectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad)); |
5170 | case NeonTypeFlags::Int16: |
5171 | case NeonTypeFlags::Poly16: |
5172 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
5173 | case NeonTypeFlags::BFloat16: |
5174 | if (AllowBFloatArgsAndRet) |
5175 | return llvm::FixedVectorType::get(CGF->BFloatTy, V1Ty ? 1 : (4 << IsQuad)); |
5176 | else |
5177 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
5178 | case NeonTypeFlags::Float16: |
5179 | if (HasLegalHalfType) |
5180 | return llvm::FixedVectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad)); |
5181 | else |
5182 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
5183 | case NeonTypeFlags::Int32: |
5184 | return llvm::FixedVectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad)); |
5185 | case NeonTypeFlags::Int64: |
5186 | case NeonTypeFlags::Poly64: |
5187 | return llvm::FixedVectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad)); |
5188 | case NeonTypeFlags::Poly128: |
5189 | // FIXME: i128 and f128 doesn't get fully support in Clang and llvm. |
5190 | // There is a lot of i128 and f128 API missing. |
5191 | // so we use v16i8 to represent poly128 and get pattern matched. |
5192 | return llvm::FixedVectorType::get(CGF->Int8Ty, 16); |
5193 | case NeonTypeFlags::Float32: |
5194 | return llvm::FixedVectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad)); |
5195 | case NeonTypeFlags::Float64: |
5196 | return llvm::FixedVectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad)); |
5197 | } |
5198 | llvm_unreachable("Unknown vector element type!")::llvm::llvm_unreachable_internal("Unknown vector element type!" , "/build/llvm-toolchain-snapshot-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5198); |
5199 | } |
5200 | |
5201 | static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF, |
5202 | NeonTypeFlags IntTypeFlags) { |
5203 | int IsQuad = IntTypeFlags.isQuad(); |
5204 | switch (IntTypeFlags.getEltType()) { |
5205 | case NeonTypeFlags::Int16: |
5206 | return llvm::FixedVectorType::get(CGF->HalfTy, (4 << IsQuad)); |
5207 | case NeonTypeFlags::Int32: |
5208 | return llvm::FixedVectorType::get(CGF->FloatTy, (2 << IsQuad)); |
5209 | case NeonTypeFlags::Int64: |
5210 | return llvm::FixedVectorType::get(CGF->DoubleTy, (1 << IsQuad)); |
5211 | default: |
5212 | 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-12~++20210124100612+2afaf072f5c1/clang/lib/CodeGen/CGBuiltin.cpp" , 5212); |
5213 | } |
5214 | } |
5215 | |
5216 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C, |
5217 | const ElementCount &Count) { |
5218 | Value *SV = llvm::ConstantVector::getSplat(Count, C); |
5219 | return Builder.CreateShuffleVector(V, V, SV, "lane"); |
5220 | } |
5221 | |
5222 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) { |
5223 | ElementCount EC = cast<llvm::VectorType>(V->getType())->getElementCount(); |
5224 | return EmitNeonSplat(V, C, EC); |
5225 | } |
5226 | |
5227 | Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops, |
5228 | const char *name, |
5229 | unsigned shift, bool rightshift) { |
5230 | unsigned j = 0; |
5231 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); |
5232 | ai != ae; ++ai, ++j) { |
5233 | if (F->isConstrainedFPIntrinsic()) |
5234 | if (ai->getType()->isMetadataTy()) |
5235 | continue; |
5236 | if (shift > 0 && shift == j) |
5237 | Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift); |
5238 | else |
5239 | Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name); |
5240 | } |
5241 | |
5242 | if (F->isConstrainedFPIntrinsic()) |
5243 | return Builder.CreateConstrainedFPCall(F, Ops, name); |
5244 | else |
5245 | return Builder.CreateCall(F, Ops, name); |
5246 | } |
5247 | |
5248 | Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty, |
5249 | bool neg) { |
5250 | int SV = cast<ConstantInt>(V)->getSExtValue(); |
5251 | return ConstantInt::get(Ty, neg ? -SV : SV); |
5252 | } |
5253 | |
5254 | // Right-shift a vector by a constant. |
5255 | Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift, |
5256 | llvm::Type *Ty, bool usgn, |
5257 | const char *name) { |
5258 | llvm::VectorType *VTy = cast<llvm::VectorType>(Ty); |
5259 | |
5260 | int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue(); |
5261 | int EltSize = VTy->getScalarSizeInBits(); |
5262 | |
5263 | Vec = Builder.CreateBitCast(Vec, Ty); |
5264 | |
5265 | // lshr/ashr are undefined when the shift amount is equal to the vector |
5266 | // element size. |
5267 | if (ShiftAmt == EltSize) { |
5268 | if (usgn) { |
5269 | // Right-shifting an unsigned value by its size yields 0. |
5270 | return llvm::ConstantAggregateZero::get(VTy); |
5271 | } else { |
5272 | // Right-shifting a signed value by its size is equivalent |
5273 | // to a shift of size-1. |
5274 | --ShiftAmt; |
5275 | Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt); |
5276 | } |
5277 | } |
5278 | |
5279 | Shift = EmitNeonShiftVector(Shift, Ty, false); |
5280 | if (usgn) |
5281 | return Builder.CreateLShr(Vec, Shift, name); |
5282 | else |
5283 | return Builder.CreateAShr(Vec, Shift, name); |
5284 | } |
5285 | |
5286 | enum { |
5287 | AddRetType = (1 << 0), |
5288 | Add1ArgType = (1 << 1), |
5289 | Add2ArgTypes = (1 << 2), |
5290 | |
5291 | VectorizeRetType = (1 << 3), |
5292 | VectorizeArgTypes = (1 << 4), |
5293 | |
5294 | InventFloatType = (1 << 5), |
5295 | UnsignedAlts = (1 << 6), |
5296 | |
5297 | Use64BitVectors = (1 << 7), |
5298 | Use128BitVectors = (1 << 8), |
5299 | |
5300 | Vectorize1ArgType = Add1ArgType | VectorizeArgTypes, |
5301 | VectorRet = AddRetType | VectorizeRetType, |
5302 | VectorRetGetArgs01 = |
5303 | AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes, |
5304 | FpCmpzModifiers = |
5305 | AddRetType | VectorizeRetType | Add1ArgType | InventFloatType |
5306 | }; |
5307 | |
5308 | namespace { |
5309 | struct ARMVectorIntrinsicInfo { |
5310 | const char *NameHint; |
5311 | unsigned BuiltinID; |
5312 | unsigned LLVMIntrinsic; |
5313 | unsigned AltLLVMIntrinsic; |
5314 | uint64_t TypeModifier; |
5315 | |
5316 | bool operator<(unsigned RHSBuiltinID) const { |
5317 | return BuiltinID < RHSBuiltinID; |
5318 | } |
5319 | bool operator<(const ARMVectorIntrinsicInfo &TE) const { |
5320 | return BuiltinID < TE.BuiltinID; |
5321 | } |
5322 | }; |
5323 | } // end anonymous namespace |
5324 | |
5325 | #define NEONMAP0(NameBase) \ |
5326 | { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 } |
5327 | |
5328 | #define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \ |
5329 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ |
5330 | Intrinsic::LLVMIntrinsic, 0, TypeModifier } |
5331 | |
5332 | #define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \ |
5333 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ |
5334 | Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \ |
5335 | TypeModifier } |
5336 | |
5337 | static const ARMVectorIntrinsicInfo ARMSIMDIntrinsicMap [] = { |
5338 | NEONMAP1(__a32_vcvt_bf16_v, arm_neon_vcvtfp2bf, 0), |
5339 | NEONMAP0(splat_lane_v), |
5340 | NEONMAP0(splat_laneq_v), |
5341 | NEONMAP0(splatq_lane_v), |
5342 | NEONMAP0(splatq_laneq_v), |
5343 | NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), |
5344 | NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), |
5345 | NEONMAP1(vabs_v, arm_neon_vabs, 0), |
5346 | NEONMAP1(vabsq_v, arm_neon_vabs, 0), |
5347 | NEONMAP0(vaddhn_v), |
5348 | NEONMAP1(vaesdq_v, arm_neon_aesd, 0), |
5349 | NEONMAP1(vaeseq_v, arm_neon_aese, 0), |
5350 | NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0), |
5351 | NEONMAP1(vaesmcq_v, arm_neon_aesmc, 0), |
5352 | NEONMAP1(vbfdot_v, arm_neon_bfdot, 0), |
5353 | NEONMAP1(vbfdotq_v, arm_neon_bfdot, 0), |
5354 | NEONMAP1(vbfmlalbq_v, arm_neon_bfmlalb, 0), |
5355 | NEONMAP1(vbfmlaltq_v, arm_neon_bfmlalt, 0), |
5356 | NEONMAP1(vbfmmlaq_v, arm_neon_bfmmla, 0), |
5357 | NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType), |
5358 | NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType), |
5359 | NEONMAP1(vcadd_rot270_v, arm_neon_vcadd_rot270, Add1ArgType), |
5360 | NEONMAP1(vcadd_rot90_v, arm_neon_vcadd_rot90, Add1ArgType), |
5361 | NEONMAP1(vcaddq_rot270_v, arm_neon_vcadd_rot270, Add1ArgType), |
5362 | NEONMAP1(vcaddq_rot90_v, arm_neon_vcadd_rot90, Add1ArgType), |
5363 | NEONMAP1(vcage_v, arm_neon_vacge, 0), |
5364 | NEONMAP1(vcageq_v, arm_neon_vacge, 0), |
5365 | NEONMAP1(vcagt_v, arm_neon_vacgt, 0), |
5366 | NEONMAP1(vcagtq_v, arm_neon_vacgt, 0), |
5367 | NEONMAP1(vcale_v, arm_neon_vacge, 0), |
5368 | NEONMAP1(vcaleq_v, arm_neon_vacge, 0), |
5369 | NEONMAP1(vcalt_v, arm_neon_vacgt, 0), |
5370 | NEONMAP1(vcaltq_v, arm_neon_vacgt, 0), |
5371 | NEONMAP0(vceqz_v), |
5372 | NEONMAP0(vceqzq_v), |
5373 | NEONMAP0(vcgez_v), |
5374 | NEONMAP0(vcgezq_v), |
5375 | NEONMAP0(vcgtz_v), |
5376 | NEONMAP0(vcgtzq_v), |
5377 | NEONMAP0(vclez_v), |
5378 | NEONMAP0(vclezq_v), |
5379 | NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType), |
5380 | NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType), |
5381 | NEONMAP0(vcltz_v), |
5382 | NEONMAP0(vcltzq_v), |
5383 | NEONMAP1(vclz_v, ctlz, Add1ArgType), |
5384 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), |
5385 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), |
5386 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), |
5387 | NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0), |
5388 | NEONMAP0(vcvt_f16_v), |
5389 | NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0), |
5390 | NEONMAP0(vcvt_f32_v), |
5391 | NEONMAP2(vcvt_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
5392 | NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
5393 | NEONMAP1(vcvt_n_s16_v, arm_neon_vcvtfp2fxs, 0), |
5394 | NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0), |
5395 | NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0), |
5396 | NEONMAP1(vcvt_n_u16_v, arm_neon_vcvtfp2fxu, 0), |
5397 | NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0), |
5398 | NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0), |
5399 | NEONMAP0(vcvt_s16_v), |
5400 | NEONMAP0(vcvt_s32_v), |
5401 | NEONMAP0(vcvt_s64_v), |
5402 | NEONMAP0(vcvt_u16_v), |
5403 | NEONMAP0(vcvt_u32_v), |
5404 | NEONMAP0(vcvt_u64_v), |
5405 | NEONMAP1(vcvta_s16_v, arm_neon_vcvtas, 0), |
5406 | NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0), |
5407 | NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0), |
5408 | NEONMAP1(vcvta_u16_v, arm_neon_vcvtau, 0), |
5409 | NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0), |
5410 | NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0), |
5411 | NEONMAP1(vcvtaq_s16_v, arm_neon_vcvtas, 0), |
5412 | NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0), |
5413 | NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0), |
5414 | NEONMAP1(vcvtaq_u16_v, arm_neon_vcvtau, 0), |
5415 | NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0), |
5416 | NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0), |
5417 | NEONMAP1(vcvth_bf16_f32, arm_neon_vcvtbfp2bf, 0), |
5418 | NEONMAP1(vcvtm_s16_v, arm_neon_vcvtms, 0), |
5419 | NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0), |
5420 | NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0), |
5421 | NEONMAP1(vcvtm_u16_v, arm_neon_vcvtmu, 0), |
5422 | NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0), |
5423 | NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0), |
5424 | NEONMAP1(vcvtmq_s16_v, arm_neon_vcvtms, 0), |
5425 | NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0), |
5426 | NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0), |
5427 | NEONMAP1(vcvtmq_u16_v, arm_neon_vcvtmu, 0), |
5428 | NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0), |
5429 | NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0), |
5430 | NEONMAP1(vcvtn_s16_v, arm_neon_vcvtns, 0), |
5431 | NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0), |
5432 | NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0), |
5433 | NEONMAP1(vcvtn_u16_v, arm_neon_vcvtnu, 0), |
5434 | NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0), |
5435 | NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0), |
5436 | NEONMAP1(vcvtnq_s16_v, arm_neon_vcvtns, 0), |
5437 | NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0), |
5438 | NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0), |
5439 | NEONMAP1(vcvtnq_u16_v, arm_neon_vcvtnu, 0), |
5440 | NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0), |
5441 | NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0), |
5442 | NEONMAP1(vcvtp_s16_v, arm_neon_vcvtps, 0), |
5443 | NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0), |
5444 | NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0), |
5445 | NEONMAP1(vcvtp_u16_v, arm_neon_vcvtpu, 0), |
5446 | NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0), |
5447 | NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0), |
5448 | NEONMAP1(vcvtpq_s16_v, arm_neon_vcvtps, 0), |
5449 | NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0), |
5450 | NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0), |
5451 | NEONMAP1(vcvtpq_u16_v, arm_neon_vcvtpu, 0), |
5452 | NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0), |
5453 | NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0), |
5454 | NEONMAP0(vcvtq_f16_v), |
5455 | NEONMAP0(vcvtq_f32_v), |
5456 | NEONMAP2(vcvtq_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
5457 | NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
5458 | NEONMAP1(vcvtq_n_s16_v, arm_neon_vcvtfp2fxs, 0), |
5459 | NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0), |
5460 | NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0), |
5461 | NEONMAP1(vcvtq_n_u16_v, arm_neon_vcvtfp2fxu, 0), |
5462 | NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0), |
5463 | NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0), |
5464 | NEONMAP0(vcvtq_s16_v), |
5465 | NEONMAP0(vcvtq_s32_v), |
5466 | NEONMAP0(vcvtq_s64_v), |
5467 | NEONMAP0(vcvtq_u16_v), |
5468 | NEONMAP0(vcvtq_u32_v), |
5469 | NEONMAP0(vcvtq_u64_v), |
5470 | NEONMAP2(vdot_v, arm_neon_udot, arm_neon_sdot, 0), |
5471 | NEONMAP2(vdotq_v, arm_neon_udot, arm_neon_sdot, 0), |
5472 | NEONMAP0(vext_v), |
5473 | NEONMAP0(vextq_v), |
5474 | NEONMAP0(vfma_v), |
5475 | NEONMAP0(vfmaq_v), |
5476 | NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), |
5477 | NEONMAP2(vhaddq_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), |
5478 | NEONMAP2(vhsub_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), |
5479 | NEONMAP2(vhsubq_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), |
5480 | NEONMAP0(vld1_dup_v), |
5481 | NEONMAP1(vld1_v, arm_neon_vld1, 0), |
5482 | NEONMAP1(vld1_x2_v, arm_neon_vld1x2, 0), |
5483 | NEONMAP1(vld1_x3_v, arm_neon_vld1x3, 0), |
5484 | NEONMAP1(vld1_x4_v, arm_neon_vld1x4, 0), |
5485 | NEONMAP0(vld1q_dup_v), |
5486 | NEONMAP1(vld1q_v, arm_neon_vld1, 0), |
5487 | NEONMAP1(vld1q_x2_v, arm_neon_vld1x2, 0), |
5488 | NEONMAP1(vld1q_x3_v, arm_neon_vld1x3, 0), |
5489 | NEONMAP1(vld1q_x4_v, arm_neon_vld1x4, 0), |
5490 | NEONMAP1(vld2_dup_v, arm_neon_vld2dup, 0), |
5491 | NEONMAP1(vld2_lane_v, arm_neon_vld2lane, 0), |
5492 | NEONMAP1(vld2_v, arm_neon_vld2, 0), |
5493 | NEONMAP1(vld2q_dup_v, arm_neon_vld2dup, 0), |
5494 | NEONMAP1(vld2q_lane_v, arm_neon_vld2lane, 0), |
5495 | NEONMAP1(vld2q_v, arm_neon_vld2, 0), |
5496 | NEONMAP1(vld3_dup_v, arm_neon_vld3dup, 0), |
5497 | NEONMAP1(vld3_lane_v, arm_neon_vld3lane, 0), |
5498 | NEONMAP1(vld3_v, arm_neon_vld3, 0), |
5499 | NEONMAP1(vld3q_dup_v, arm_neon_vld3dup, 0), |
5500 | NEONMAP1(vld3q_lane_v, arm_neon_vld3lane, 0), |
5501 | NEONMAP1(vld3q_v, arm_neon_vld3, 0), |
5502 | NEONMAP1(vld4_dup_v, arm_neon_vld4dup, 0), |
5503 | NEONMAP1(vld4_lane_v, arm_neon_vld4lane, 0), |
5504 | NEONMAP1(vld4_v, arm_neon_vld4, 0), |
5505 | NEONMAP1(vld4q_dup_v, arm_neon_vld4dup, 0), |
5506 | NEONMAP1(vld4q_lane_v, arm_neon_vld4lane, 0), |
5507 | NEONMAP1(vld4q_v, arm_neon_vld4, 0), |
5508 | NEONMAP2(vmax_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), |
5509 | NEONMAP1(vmaxnm_v, arm_neon_vmaxnm, Add1ArgType), |
5510 | NEONMAP1(vmaxnmq_v, arm_neon_vmaxnm, Add1ArgType), |
5511 | NEONMAP2(vmaxq_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), |
5512 | NEONMAP2(vmin_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), |
5513 | NEONMAP1(vminnm_v, arm_neon_vminnm, Add1ArgType), |
5514 | NEONMAP1(vminnmq_v, arm_neon_vminnm, Add1ArgType), |
5515 | NEONMAP2(vminq_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), |
5516 | NEONMAP2(vmmlaq_v, arm_neon_ummla, arm_neon_smmla, 0), |
5517 | NEONMAP0(vmovl_v), |
5518 | NEONMAP0(vmovn_v), |
5519 | NEONMAP1(vmul_v, arm_neon_vmulp, Add1ArgType), |
5520 | NEONMAP0(vmull_v), |
5521 | NEONMAP1(vmulq_v, arm_neon_vmulp, Add1ArgType), |
5522 | NEONMAP2(vpadal_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), |
5523 | NEONMAP2(vpadalq_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), |
5524 | NEONMAP1(vpadd_v, arm_neon_vpadd, Add1ArgType), |
5525 | NEONMAP2(vpaddl_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), |
5526 | NEONMAP2(vpaddlq_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), |
5527 | NEONMAP1(vpaddq_v, arm_neon_vpadd, Add1ArgType), |
5528 | NEONMAP2(vpmax_v, arm_neon_vpmaxu, arm_neon_vpmaxs, Add1ArgType | UnsignedAlts), |
5529 | NEONMAP2(vpmin_v, arm_neon_vpminu, arm_neon_vpmins, Add1ArgType | UnsignedAlts), |
5530 | NEONMAP1(vqabs_v, arm_neon_vqabs, Add1ArgType), |
5531 | NEONMAP1(vqabsq_v, arm_neon_vqabs, Add1ArgType), |
5532 | NEONMAP2(vqadd_v, uadd_sat, sadd_sat, Add1ArgType | UnsignedAlts), |
5533 | NEONMAP2(vqaddq_v, uadd_sat, sadd_sat, Add1ArgType | UnsignedAlts), |
5534 | NEONMAP2(vqdmlal_v, arm_neon_vqdmull, sadd_sat, 0), |
5535 | NEONMAP2(vqdmlsl_v, arm_neon_vqdmull, ssub_sat, 0), |
5536 | NEONMAP1(vqdmulh_v, arm_neon_vqdmulh, Add1ArgType), |
5537 | NEONMAP1(vqdmulhq_v, arm_neon_vqdmulh, Add1ArgType), |
5538 | NEONMAP1(vqdmull_v, arm_neon_vqdmull, Add1ArgType), |
5539 | NEONMAP2(vqmovn_v, arm_neon_vqmovnu, arm_neon_vqmovns, Add1ArgType | UnsignedAlts), |
5540 | NEONMAP1(vqmovun_v, arm_neon_vqmovnsu, Add1ArgType), |
5541 | NEONMAP1(vqneg_v, arm_neon_vqneg, Add1ArgType), |
5542 | NEONMAP1(vqnegq_v, arm_neon_vqneg, Add1ArgType), |
5543 | NEONMAP1(vqrdmulh_v, arm_neon_vqrdmulh, Add1ArgType), |
5544 | NEONMAP1(vqrdmulhq_v, arm_neon_vqrdmulh, Add1ArgType), |
5545 | NEONMAP2(vqrshl_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), |
5546 | NEONMAP2(vqrshlq_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), |
5547 | NEONMAP2(vqshl_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), |
5548 | NEONMAP2(vqshl_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), |
5549 | NEONMAP2(vqshlq_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), |
5550 | NEONMAP2(vqshlq_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), |
5551 | NEONMAP1(vqshlu_n_v, arm_neon_vqshiftsu, 0), |
5552 | NEONMAP1(vqshluq_n_v, arm_neon_vqshiftsu, 0), |
5553 | NEONMAP2(vqsub_v, usub_sat, ssub_sat, Add1ArgType | UnsignedAlts), |
5554 | NEONMAP2(vqsubq_v, usub_sat, ssub_sat, Add1ArgType | UnsignedAlts), |
5555 | NEONMAP1(vraddhn_v, arm_neon_vraddhn, Add1ArgType), |
5556 | NEONMAP2(vrecpe_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), |
5557 | NEONMAP2(vrecpeq_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), |
5558 | NEONMAP1(vrecps_v, arm_neon_vrecps, Add1ArgType), |
5559 | NEONMAP1(vrecpsq_v, arm_neon_vrecps, Add1ArgType), |
5560 | NEONMAP2(vrhadd_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), |
5561 | NEONMAP2(vrhaddq_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), |
5562 | NEONMAP1(vrnd_v, arm_neon_vrintz, Add1ArgType), |
5563 | NEONMAP1(vrnda_v, arm_neon_vrinta, Add1ArgType), |
5564 | NEONMAP1(vrndaq_v, arm_neon_vrinta, Add1ArgType), |
5565 | NEONMAP0(vrndi_v), |
5566 | NEONMAP0(vrndiq_v), |
5567 | NEONMAP1(vrndm_v, arm_neon_vrintm, Add1ArgType), |
5568 | NEONMAP1(vrndmq_v, arm_neon_vrintm, Add1ArgType), |
5569 | NEONMAP1(vrndn_v, arm_neon_vrintn, Add1ArgType), |
5570 | NEONMAP1(vrndnq_v, arm_neon_vrintn, Add1ArgType), |
5571 | NEONMAP1(vrndp_v, arm_neon_vrintp, Add1ArgType), |
5572 | NEONMAP1(vrndpq_v, arm_neon_vrintp, Add1ArgType), |
5573 | NEONMAP1(vrndq_v, arm_neon_vrintz, Add1ArgType), |
5574 | NEONMAP1(vrndx_v, arm_neon_vrintx, Add1ArgType), |
5575 | NEONMAP1(vrndxq_v, arm_neon_vrintx, Add1ArgType), |
5576 | NEONMAP2(vrshl_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), |
5577 | NEONMAP2(vrshlq_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), |
5578 | NEONMAP2(vrshr_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), |
5579 | NEONMAP2(vrshrq_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), |
5580 | NEONMAP2(vrsqrte_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), |
5581 | NEONMAP2(vrsqrteq_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), |
5582 | NEONMAP1(vrsqrts_v, arm_neon_vrsqrts, Add1ArgType), |
5583 | NEONMAP1(vrsqrtsq_v, arm_neon_vrsqrts, Add1ArgType), |
5584 | NEONMAP1(vrsubhn_v, arm_neon_vrsubhn, Add1ArgType), |
5585 | NEONMAP1(vsha1su0q_v, arm_neon_sha1su0, 0), |
5586 | NEONMAP1(vsha1su1q_v, arm_neon_sha1su1, 0), |
5587 | NEONMAP1(vsha256h2q_v, arm_neon_sha256h2, 0), |
5588 | NEONMAP1(vsha256hq_v, arm_neon_sha256h, 0), |
5589 | NEONMAP1(vsha256su0q_v, arm_neon_sha256su0, 0), |
5590 | NEONMAP1(vsha256su1q_v, arm_neon_sha256su1, 0), |
5591 | NEONMAP0(vshl_n_v), |
5592 | NEONMAP2(vshl_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), |
5593 | NEONMAP0(vshll_n_v), |
5594 | NEONMAP0(vshlq_n_v), |
5595 | NEONMAP2(vshlq_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), |
5596 | NEONMAP0(vshr_n_v), |
5597 | NEONMAP0(vshrn_n_v), |
5598 | NEONMAP0(vshrq_n_v), |
5599 | NEONMAP1(vst1_v, arm_neon_vst1, 0), |
5600 | NEONMAP1(vst1_x2_v, arm_neon_vst1x2, 0), |
5601 | NEONMAP1(vst1_x3_v, arm_neon_vst1x3, 0), |
5602 | NEONMAP1(vst1_x4_v, arm_neon_vst1x4, 0), |
5603 | NEONMAP1(vst1q_v, arm_neon_vst1, 0), |
5604 | NEONMAP1(vst1q_x2_v, arm_neon_vst1x2, 0), |
5605 | NEONMAP1(vst1q_x3_v, arm_neon_vst1x3, 0), |
5606 | NEONMAP1(vst1q_x4_v, arm_neon_vst1x4, 0), |
5607 | NEONMAP1(vst2_lane_v, arm_neon_vst2lane, 0), |
5608 | NEONMAP1(vst2_v, arm_neon_vst2, 0), |
5609 | NEONMAP1(vst2q_lane_v, arm_neon_vst2lane, 0), |
5610 | NEONMAP1(vst2q_v, arm_neon_vst2, 0), |
5611 | NEONMAP1(vst3_lane_v, arm_neon_vst3lane, 0), |
5612 | NEONMAP1(vst3_v, arm_neon_vst3, 0), |
5613 | NEONMAP1(vst3q_lane_v, arm_neon_vst3lane, 0), |
5614 | NEONMAP1(vst3q_v, arm_neon_vst3, 0), |
5615 | NEONMAP1(vst4_lane_v, arm_neon_vst4lane, 0), |
5616 | NEONMAP1(vst4_v, arm_neon_vst4, 0), |
5617 | NEONMAP1(vst4q_lane_v, arm_neon_vst4lane, 0), |
5618 | NEONMAP1(vst4q_v, arm_neon_vst4, 0), |
5619 | NEONMAP0(vsubhn_v), |
5620 | NEONMAP0(vtrn_v), |
5621 | NEONMAP0(vtrnq_v), |
5622 | NEONMAP0(vtst_v), |
5623 | NEONMAP0(vtstq_v), |
5624 | NEONMAP1(vusdot_v, arm_neon_usdot, 0), |
5625 | NEONMAP1(vusdotq_v, arm_neon_usdot, 0), |
5626 | NEONMAP1(vusmmlaq_v, arm_neon_usmmla, 0), |
5627 | NEONMAP0(vuzp_v), |
5628 | NEONMAP0(vuzpq_v), |
5629 | NEONMAP0(vzip_v), |
5630 | NEONMAP0(vzipq_v) |
5631 | }; |
5632 | |
5633 | static const ARMVectorIntrinsicInfo AArch64SIMDIntrinsicMap[] = { |
5634 | NEONMAP1(__a64_vcvtq_low_bf16_v, aarch64_neon_bfcvtn, 0), |
5635 | NEONMAP0(splat_lane_v), |
5636 | NEONMAP0(splat_laneq_v), |
5637 | NEONMAP0(splatq_lane_v), |
5638 | NEONMAP0(splatq_laneq_v), |
5639 | NEONMAP1(vabs_v, aarch64_neon_abs, 0), |
5640 | NEONMAP1(vabsq_v, aarch64_neon_abs, 0), |
5641 | NEONMAP0(vaddhn_v), |
5642 | NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0), |
5643 | NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0), |
5644 | NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0), |
5645 | NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0), |
5646 | NEONMAP1(vbfdot_v, aarch64_neon_bfdot, 0), |
5647 | NEONMAP1(vbfdotq_v, aarch64_neon_bfdot, 0), |
5648 | NEONMAP1(vbfmlalbq_v, aarch64_neon_bfmlalb, 0), |
5649 | NEONMAP1(vbfmlaltq_v, aarch64_neon_bfmlalt, 0), |
5650 | NEONMAP1(vbfmmlaq_v, aarch64_neon_bfmmla, 0), |
5651 | NEONMAP1(vcadd_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType), |
5652 | NEONMAP1(vcadd_rot90_v, aarch64_neon_vcadd_rot90, Add1ArgType), |
5653 | NEONMAP1(vcaddq_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType), |
5654 | NEONMAP1(vcaddq_rot90_v, aarch64_neon_vcadd_rot90, Add1ArgType), |
5655 | NEONMAP1(vcage_v, aarch64_neon_facge, 0), |
5656 | NEONMAP1(vcageq_v, aarch64_neon_facge, 0), |
5657 | NEONMAP1(vcagt_v, aarch64_neon_facgt, 0), |
5658 | NEONMAP1(vcagtq_v, aarch64_neon_facgt, 0), |
5659 | NEONMAP1(vcale_v, aarch64_neon_facge, 0), |
5660 | NEONMAP1(vcaleq_v, aarch64_neon_facge, 0), |
5661 | NEONMAP1(vcalt_v, aarch64_neon_facgt, 0), |
5662 | NEONMAP1(vcaltq_v, aarch64_neon_facgt, 0), |
5663 | NEONMAP0(vceqz_v), |
5664 | NEONMAP0(vceqzq_v), |
5665 | NEONMAP0(vcgez_v), |
5666 | NEONMAP0(vcgezq_v), |
5667 | NEONMAP0(vcgtz_v), |
5668 | NEONMAP0(vcgtzq_v), |
5669 | NEONMAP0(vclez_v), |
5670 | NEONMAP0(vclezq_v), |
5671 | NEONMAP1(vcls_v, aarch64_neon_cls, Add1ArgType), |
5672 | NEONMAP1(vclsq_v, aarch64_neon_cls, Add1ArgType), |
5673 | NEONMAP0(vcltz_v), |
5674 | NEONMAP0(vcltzq_v), |
5675 | NEONMAP1(vclz_v, ctlz, Add1ArgType), |
5676 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), |
5677 | NEONMAP1(vcmla_rot180_v, aarch64_neon_vcmla_rot180, Add1ArgType), |
5678 | NEONMAP1(vcmla_rot270_v, aarch64_neon_vcmla_rot270, Add1ArgType), |
5679 | NEONMAP1(vcmla_rot90_v, aarch64_neon_vcmla_rot90, Add1ArgType), |
5680 | NEONMAP1(vcmla_v, aarch64_neon_vcmla_rot0, Add1ArgType), |
5681 | NEONMAP1(vcmlaq_rot180_v, aarch64_neon_vcmla_rot180, Add1ArgType), |
5682 | NEONMAP1(vcmlaq_rot270_v, aarch64_neon_vcmla_rot270, Add1ArgType), |
5683 | NEONMAP1(vcmlaq_rot90_v, aarch64_neon_vcmla_rot90, Add1ArgType), |
5684 | NEONMAP1(vcmlaq_v, aarch64_neon_vcmla_rot0, Add1ArgType), |
5685 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), |
5686 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), |
5687 | NEONMAP1(vcvt_f16_f32, aarch64_neon_vcvtfp2hf, 0), |
5688 | NEONMAP0(vcvt_f16_v), |
5689 | NEONMAP1(vcvt_f32_f16, aarch64_neon_vcvthf2fp, 0), |
5690 | NEONMAP0(vcvt_f32_v), |
5691 | NEONMAP2(vcvt_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
5692 | NEONMAP2(vcvt_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
5693 | NEONMAP2(vcvt_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
5694 | NEONMAP1(vcvt_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), |
5695 | NEONMAP1(vcvt_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), |
5696 | NEONMAP1(vcvt_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), |
5697 | NEONMAP1(vcvt_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), |
5698 | NEONMAP1(vcvt_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), |
5699 | NEONMAP1(vcvt_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), |
5700 | NEONMAP0(vcvtq_f16_v), |
5701 | NEONMAP0(vcvtq_f32_v), |
5702 | NEONMAP1(vcvtq_high_bf16_v, aarch64_neon_bfcvtn2, 0), |
5703 | NEONMAP2(vcvtq_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
5704 | NEONMAP2(vcvtq_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
5705 | NEONMAP2(vcvtq_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
5706 | NEONMAP1(vcvtq_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), |
5707 | NEONMAP1(vcvtq_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), |
5708 | NEONMAP1(vcvtq_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), |
5709 | NEONMAP1(vcvtq_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), |
5710 | NEONMAP1(vcvtq_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), |
5711 | NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), |
5712 | NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType), |
5713 | NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0), |
5714 | NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0), |
5715 | NEONMAP0(vext_v), |
5716 | NEONMAP0(vextq_v), |
5717 | NEONMAP0(vfma_v), |
5718 | NEONMAP0(vfmaq_v), |
5719 | NEONMAP1(vfmlal_high_v, aarch64_neon_fmlal2, 0), |
5720 | NEONMAP1(vfmlal_low_v, aarch64_neon_fmlal, 0), |
5721 | NEONMAP1(vfmlalq_high_v, aarch64_neon_fmlal2, 0), |
5722 | NEONMAP1(vfmlalq_low_v, aarch64_neon_fmlal, 0), |
5723 | NEONMAP1(vfmlsl_high_v, aarch64_neon_fmlsl2, 0), |
5724 | NEONMAP1(vfmlsl_low_v, aarch64_neon_fmlsl, 0), |
5725 | NEONMAP1(vfmlslq_high_v, aarch64_neon_fmlsl2, 0), |
5726 | NEONMAP1(vfmlslq_low_v, aarch64_neon_fmlsl, 0), |
5727 | NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), |
5728 | NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), |
5729 | NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), |
5730 | NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), |
5731 | NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0), |
5732 | NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0), |
5733 | NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0), |
5734 | NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0), |
5735 | NEONMAP1(vld1q_x3_v, aarch64_neon_ld1x3, 0), |
5736 | NEONMAP1(vld1q_x4_v, aarch64_neon_ld1x4, 0), |
5737 | NEONMAP2(vmmlaq_v, aarch64_neon_ummla, aarch64_neon_smmla, 0), |
5738 | NEONMAP0(vmovl_v), |
5739 | NEONMAP0(vmovn_v), |
5740 | NEONMAP1(vmul_v, aarch64_neon_pmul, Add1ArgType), |
5741 | NEONMAP1(vmulq_v, aarch64_neon_pmul, Add1ArgType), |
5742 | NEONMAP1(vpadd_v, aarch64_neon_addp, Add1ArgType), |
5743 | NEONMAP2(vpaddl_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), |
5744 | NEONMAP2(vpaddlq_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), |
5745 | NEONMAP1(vpaddq_v, aarch64_neon_addp, Add1ArgType), |
5746 | NEONMAP1(vqabs_v, aarch64_neon_sqabs, Add1ArgType), |
5747 | NEONMAP1(vqabsq_v, aarch64_neon_sqabs, Add1ArgType), |
5748 | NEONMAP2(vqadd_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), |
5749 | NEONMAP2(vqaddq_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), |
5750 | NEONMAP2(vqdmlal_v, aarch64_neon_sqdmull, aarch64_neon_sqadd, 0), |
5751 | NEONMAP2(vqdmlsl_v, aarch64_neon_sqdmull, aarch64_neon_sqsub, 0), |
5752 | NEONMAP1(vqdmulh_lane_v, aarch64_neon_sqdmulh_lane, 0), |
5753 | NEONMAP1(vqdmulh_laneq_v, aarch64_neon_sqdmulh_laneq, 0), |
5754 | NEONMAP1(vqdmulh_v, aarch64_neon_sqdmulh, Add1ArgType), |
5755 | NEONMAP1(vqdmulhq_lane_v, aarch64_neon_sqdmulh_lane, 0), |
5756 | NEONMAP1(vqdmulhq_laneq_v, aarch64_neon_sqdmulh_laneq, 0), |
5757 | NEONMAP1(vqdmulhq_v, aarch64_neon_sqdmulh, Add1ArgType), |
5758 | NEONMAP1(vqdmull_v, aarch64_neon_sqdmull, Add1ArgType), |
5759 | NEONMAP2(vqmovn_v, aarch64_neon_uqxtn, aarch64_neon_sqxtn, Add1ArgType | UnsignedAlts), |
5760 | NEONMAP1(vqmovun_v, aarch64_neon_sqxtun, Add1ArgType), |
5761 | NEONMAP1(vqneg_v, aarch64_neon_sqneg, Add1ArgType), |
5762 | NEONMAP1(vqnegq_v, aarch64_neon_sqneg, Add1ArgType), |
5763 | NEONMAP1(vqrdmulh_lane_v, aarch64_neon_sqrdmulh_lane, 0), |
5764 | NEONMAP1(vqrdmulh_laneq_v, aarch64_neon_sqrdmulh_laneq, 0), |
5765 | NEONMAP1(vqrdmulh_v, aarch64_neon_sqrdmulh, Add1ArgType), |
5766 | NEONMAP1(vqrdmulhq_lane_v, aarch64_neon_sqrdmulh_lane, 0), |
5767 | NEONMAP1(vqrdmulhq_laneq_v, aarch64_neon_sqrdmulh_laneq, 0), |
5768 | NEONMAP1(vqrdmulhq_v, aarch64_neon_sqrdmulh, Add1ArgType), |
5769 | NEONMAP2(vqrshl_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), |
5770 | NEONMAP2(vqrshlq_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), |
5771 | NEONMAP2(vqshl_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl, UnsignedAlts), |
5772 | NEONMAP2(vqshl_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), |
5773 | NEONMAP2(vqshlq_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl,UnsignedAlts), |
5774 | NEONMAP2(vqshlq_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), |
5775 | NEONMAP1(vqshlu_n_v, aarch64_neon_sqshlu, 0), |
5776 | NEONMAP1(vqshluq_n_v, aarch64_neon_sqshlu, 0), |
5777 | NEONMAP2(vqsub_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), |
5778 | NEONMAP2(vqsubq_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), |
5779 | NEONMAP1(vraddhn_v, aarch64_neon_raddhn, Add1ArgType), |
5780 | NEONMAP2(vrecpe_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), |
5781 | NEONMAP2(vrecpeq_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), |
5782 | NEONMAP1(vrecps_v, aarch64_neon_frecps, Add1ArgType), |
5783 | NEONMAP1(vrecpsq_v, aarch64_neon_frecps, Add1ArgType), |
5784 | NEONMAP2(vrhadd_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), |
5785 | NEONMAP2(vrhaddq_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), |
5786 | NEONMAP0(vrndi_v), |
5787 | NEONMAP0(vrndiq_v), |
5788 | NEONMAP2(vrshl_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), |
5789 | NEONMAP2(vrshlq_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), |
5790 | NEONMAP2(vrshr_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), |
5791 | NEONMAP2(vrshrq_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), |
5792 | NEONMAP2(vrsqrte_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), |
5793 | NEONMAP2(vrsqrteq_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), |
5794 | NEONMAP1(vrsqrts_v, aarch64_neon_frsqrts, Add1ArgType), |
5795 | NEONMAP1(vrsqrtsq_v, aarch64_neon_frsqrts, Add1ArgType), |
5796 | NEONMAP1(vrsubhn_v, aarch64_neon_rsubhn, Add1ArgType), |
5797 | NEONMAP1(vsha1su0q_v, aarch64_crypto_sha1su0, 0), |
5798 | NEONMAP1(vsha1su1q_v, aarch64_crypto_sha1su1, 0), |
5799 | NEONMAP1(vsha256h2q_v, aarch64_crypto_sha256h2, 0), |
5800 | NEONMAP1(vsha256hq_v, aarch64_crypto_sha256h, 0), |
5801 | NEONMAP1(vsha256su0q_v, aarch64_crypto_sha256su0, 0), |
5802 | NEONMAP1(vsha256su1q_v, aarch64_crypto_sha256su1, 0), |
5803 | NEONMAP0(vshl_n_v), |
5804 | NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), |
5805 | NEONMAP0(vshll_n_v), |
5806 | NEONMAP0(vshlq_n_v), |
5807 | NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), |
5808 | NEONMAP0(vshr_n_v), |
5809 | NEONMAP0(vshrn_n_v), |
5810 | NEONMAP0(vshrq_n_v), |
5811 | NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0), |
5812 | NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0), |
5813 | NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0), |
5814 | NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0), |
5815 | NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0), |
5816 | NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0), |
5817 | NEONMAP0(vsubhn_v), |
5818 | NEONMAP0(vtst_v), |
5819 | NEONMAP0(vtstq_v), |
5820 | NEONMAP1(vusdot_v, aarch64_neon_usdot, 0), |
5821 | NEONMAP1(vusdotq_v, aarch64_neon_usdot, 0), |
5822 | NEONMAP1(vusmmlaq_v, aarch64_neon_usmmla, 0), |
5823 | }; |
5824 | |
5825 | static const ARMVectorIntrinsicInfo AArch64SISDIntrinsicMap[] = { |
5826 | NEONMAP1(vabdd_f64, aarch64_sisd_fabd, Add1ArgType), |
5827 | NEONMAP1(vabds_f32, aarch64_sisd_fabd, Add1ArgType), |
5828 | NEONMAP1(vabsd_s64, aarch64_neon_abs, Add1ArgType), |
5829 | NEONMAP1(vaddlv_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), |
5830 | NEONMAP1(vaddlv_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), |
5831 | NEONMAP1(vaddlvq_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), |
5832 | NEONMAP1(vaddlvq_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), |
5833 | NEONMAP1(vaddv_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), |
5834 | NEONMAP1(vaddv_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), |
5835 | NEONMAP1(vaddv_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
5836 | NEONMAP1(vaddvq_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), |
5837 | NEONMAP1(vaddvq_f64, aarch64_neon_faddv, AddRetType | Add1ArgType), |
5838 | NEONMAP1(vaddvq_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), |
5839 | NEONMAP1(vaddvq_s64, aarch64_neon_saddv, AddRetType | Add1ArgType), |
5840 | NEONMAP1(vaddvq_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
5841 | NEONMAP1(vaddvq_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
5842 | NEONMAP1(vcaged_f64, aarch64_neon_facge, AddRetType | Add1ArgType), |
5843 | NEONMAP1(vcages_f32, aarch64_neon_facge, AddRetType | Add1ArgType), |
5844 | NEONMAP1(vcagtd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), |
5845 | NEONMAP1(vcagts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), |
5846 | NEONMAP1(vcaled_f64, aarch64_neon_facge, AddRetType | Add1ArgType), |
5847 | NEONMAP1(vcales_f32, aarch64_neon_facge, AddRetType | Add1ArgType), |
5848 | NEONMAP1(vcaltd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), |
5849 | NEONMAP1(vcalts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), |
5850 | NEONMAP1(vcvtad_s64_f64, aarch64_neon_fcvtas, AddRetType | Add1ArgType), |
5851 | NEONMAP1(vcvtad_u64_f64, aarch64_neon_fcvtau, AddRetType | Add1ArgType), |
5852 | NEONMAP1(vcvtas_s32_f32, aarch64_neon_fcvtas, AddRetType | Add1ArgType), |
5853 | NEONMAP1(vcvtas_u32_f32, aarch64_neon_fcvtau, AddRetType | Add1ArgType), |
5854 | NEONMAP1(vcvtd_n_f64_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), |
5855 | NEONMAP1(vcvtd_n_f64_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), |
5856 | NEONMAP1(vcvtd_n_s64_f64, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), |
5857 | NEONMAP1(vcvtd_n_u64_f64, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), |
5858 | NEONMAP1(vcvtd_s64_f64, aarch64_neon_fcvtzs, AddRetType | Add1ArgType), |
5859 | NEONMAP1(vcvtd_u64_f64, aarch64_neon_fcvtzu, AddRetType | Add1ArgType), |
5860 | NEONMAP1(vcvth_bf16_f32, aarch64_neon_bfcvt, 0), |
5861 | NEONMAP1(vcvtmd_s64_f64, aarch64_neon_fcvtms, AddRetType | Add1ArgType), |
5862 | NEONMAP1(vcvtmd_u64_f64, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), |
5863 | NEONMAP1(vcvtms_s32_f32, aarch64_neon_fcvtms, AddRetType | Add1ArgType), |
5864 | NEONMAP1(vcvtms_u32_f32, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), |
5865 | NEONMAP1(vcvtnd_s64_f64, aarch64_neon_fcvtns, AddRetType | Add1ArgType), |
5866 | NEONMAP1(vcvtnd_u64_f64, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), |
5867 | NEONMAP1(vcvtns_s32_f32, aarch64_neon_fcvtns, AddRetType | Add1ArgType), |
5868 | NEONMAP1(vcvtns_u32_f32, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), |
5869 | NEONMAP1(vcvtpd_s64_f64, aarch64_neon_fcvtps, AddRetType | Add1ArgType), |
5870 | NEONMAP1(vcvtpd_u64_f64, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), |
5871 | NEONMAP1(vcvtps_s32_f32, aarch64_neon_fcvtps, AddRetType | Add1ArgType), |
5872 | NEONMAP1(vcvtps_u32_f32, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), |
5873 | NEONMAP1(vcvts_n_f32_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), |
5874 | NEONMAP1(vcvts_n_f32_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), |
5875 | NEONMAP1(vcvts_n_s32_f32, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), |
5876 | NEONMAP1(vcvts_n_u32_f32, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), |
5877 | NEONMAP1(vcvts_s32_f32, aarch64_neon_fcvtzs, AddRetType | Add1ArgType), |
5878 | NEONMAP1(vcvts_u32_f32, aarch64_neon_fcvtzu, AddRetType | Add1ArgType), |
5879 | NEONMAP1(vcvtxd_f32_f64, aarch64_sisd_fcvtxn, 0), |
5880 | NEONMAP1(vmaxnmv_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
5881 | NEONMAP1(vmaxnmvq_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
5882 | NEONMAP1(vmaxnmvq_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
5883 | NEONMAP1(vmaxv_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
5884 | NEONMAP1(vmaxv_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), |
5885 | NEONMAP1(vmaxv_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), |
5886 | NEONMAP1(vmaxvq_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
5887 | NEONMAP1(vmaxvq_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
5888 | NEONMAP1(vmaxvq_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), |
5889 | NEONMAP1(vmaxvq_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), |
5890 | NEONMAP1(vminnmv_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
5891 | NEONMAP1(vminnmvq_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
5892 | NEONMAP1(vminnmvq_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
5893 | NEONMAP1(vminv_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
5894 | NEONMAP1(vminv_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), |
5895 | NEONMAP1(vminv_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), |
5896 | NEONMAP1(vminvq_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
5897 | NEONMAP1(vminvq_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), |
5898 | NEONMAP1(vminvq_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), |
5899 | NEONMAP1(vminvq_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), |
5900 | NEONMAP1(vmull_p64, aarch64_neon_pmull64, 0), |
5901 | NEONMAP1(vmulxd_f64, aarch64_neon_fmulx, Add1ArgType), |
5902 | NEONMAP1(vmulxs_f32, aarch64_neon_fmulx, Add1ArgType), |
5903 | NEONMAP1(vpaddd_s64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
5904 | NEONMAP1(vpaddd_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
5905 | NEONMAP1(vpmaxnmqd_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
5906 | NEONMAP1(vpmaxnms_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
5907 | NEONMAP1(vpmaxqd_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
5908 | NEONMAP1(vpmaxs_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
5909 | NEONMAP1(vpminnmqd_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
5910 | NEONMAP1(vpminnms_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
5911 | NEONMAP1(vpminqd_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), |
5912 | NEONMAP1(vpmins_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
5913 | NEONMAP1(vqabsb_s8, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), |
5914 | NEONMAP1(vqabsd_s64, aarch64_neon_sqabs, Add1ArgType), |
5915 | NEONMAP1(vqabsh_s16, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), |
5916 | NEONMAP1(vqabss_s32, aarch64_neon_sqabs, Add1ArgType), |
5917 | NEONMAP1(vqaddb_s8, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), |
5918 | NEONMAP1(vqaddb_u8, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), |
5919 | NEONMAP1(vqaddd_s64, aarch64_neon_sqadd, Add1ArgType), |
5920 | NEONMAP1(vqaddd_u64, aarch64_neon_uqadd, Add1ArgType), |
5921 | NEONMAP1(vqaddh_s16, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), |
5922 | NEONMAP1(vqaddh_u16, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), |
5923 | NEONMAP1(vqadds_s32, aarch64_neon_sqadd, Add1ArgType), |
5924 | NEONMAP1(vqadds_u32, aarch64_neon_uqadd, Add1ArgType), |
5925 | NEONMAP1(vqdmulhh_s16, aarch64_neon_sqdmulh, Vectorize1ArgType | Use64BitVectors), |
5926 | NEONMAP1(vqdmulhs_s32, aarch64_neon_sqdmulh, Add1ArgType), |
5927 | NEONMAP1(vqdmullh_s16, aarch64_neon_sqdmull, VectorRet | Use128BitVectors), |
5928 | NEONMAP1(vqdmulls_s32, aarch64_neon_sqdmulls_scalar, 0), |
5929 | NEONMAP1(vqmovnd_s64, aarch64_neon_scalar_sqxtn, AddRetType | Add1ArgType), |
5930 | NEONMAP1(vqmovnd_u64, aarch64_neon_scalar_uqxtn, AddRetType | Add1ArgType), |
5931 | NEONMAP1(vqmovnh_s16, aarch64_neon_sqxtn, VectorRet | Use64BitVectors), |
5932 | NEONMAP1(vqmovnh_u16, aarch64_neon_uqxtn, VectorRet | Use64BitVectors), |
5933 | NEONMAP1(vqmovns_s32, aarch64_neon_sqxtn, VectorRet | Use64BitVectors), |
5934 | NEONMAP1(vqmovns_u32, aarch64_neon_uqxtn, VectorRet | Use64BitVectors), |
5935 | NEONMAP1(vqmovund_s64, aarch64_neon_scalar_sqxtun, AddRetType | Add1ArgType), |
5936 | NEONMAP1(vqmovunh_s16, aarch64_neon_sqxtun, VectorRet | Use64BitVectors), |
5937 | NEONMAP1(vqmovuns_s32, aarch64_neon_sqxtun, VectorRet | Use64BitVectors), |
5938 | NEONMAP1(vqnegb_s8, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors), |
5939 | NEONMAP1(vqnegd_s64, aarch64_neon_sqneg, Add1ArgType), |
5940 | NEONMAP1(vqnegh_s16, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors), |
5941 | NEONMAP1(vqnegs_s32, aarch64_neon_sqneg, Add1ArgType), |
5942 | NEONMAP1(vqrdmulhh_s16, aarch64_neon_sqrdmulh, Vectorize1ArgType | Use64BitVectors), |
5943 | NEONMAP1(vqrdmulhs_s32, aarch64_neon_sqrdmulh, Add1ArgType), |
5944 | NEONMAP1(vqrshlb_s8, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors), |
5945 | NEONMAP1(vqrshlb_u8, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors), |
5946 | NEONMAP1(vqrshld_s64, aarch64_neon_sqrshl, Add1ArgType), |
5947 | NEONMAP1(vqrshld_u64, aarch64_neon_uqrshl, Add1ArgType), |
5948 | NEONMAP1(vqrshlh_s16, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors), |
5949 | NEONMAP1(vqrshlh_u16, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors), |
5950 | NEONMAP1(vqrshls_s32, aarch64_neon_sqrshl, Add1ArgType), |
5951 | NEONMAP1(vqrshls_u32, aarch64_neon_uqrshl, Add1ArgType), |
5952 | NEONMAP1(vqrshrnd_n_s64, aarch64_neon_sqrshrn, AddRetType), |
5953 | NEONMAP1(vqrshrnd_n_u64, aarch64_neon_uqrshrn, AddRetType), |
5954 | NEONMAP1(vqrshrnh_n_s16, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors), |
5955 | NEONMAP1(vqrshrnh_n_u16, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors), |
5956 | NEONMAP1(vqrshrns_n_s32, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors), |
5957 | NEONMAP1(vqrshrns_n_u32, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors), |
5958 | NEONMAP1(vqrshrund_n_s64, aarch64_neon_sqrshrun, AddRetType), |
5959 | NEONMAP1(vqrshrunh_n_s16, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors), |
5960 | NEONMAP1(vqrshruns_n_s32, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors), |
5961 | NEONMAP1(vqshlb_n_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), |
5962 | NEONMAP1(vqshlb_n_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), |
5963 | NEONMAP1(vqshlb_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), |
5964 | NEONMAP1(vqshlb_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), |
5965 | NEONMAP1(vqshld_s64, aarch64_neon_sqshl, Add1ArgType), |
5966 | NEONMAP1(vqshld_u64, aarch64_neon_uqshl, Add1ArgType), |
5967 | NEONMAP1(vqshlh_n_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), |
5968 | NEONMAP1(vqshlh_n_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), |
5969 | NEONMAP1(vqshlh_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), |
5970 | NEONMAP1(vqshlh_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), |
5971 | NEONMAP1(vqshls_n_s32, aarch64_neon_sqshl, Add1ArgType), |
5972 | NEONMAP1(vqshls_n_u32, aarch64_neon_uqshl, Add1ArgType), |
5973 | NEONMAP1(vqshls_s32, aarch64_neon_sqshl, Add1ArgType), |
5974 | NEONMAP1(vqshls_u32, aarch64_neon_uqshl, Add1ArgType), |
5975 | NEONMAP1(vqshlub_n_s8, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors), |
5976 | NEONMAP1(vqshluh_n_s16, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors), |
5977 | NEONMAP1(vqshlus_n_s32, aarch64_neon_sqshlu, Add1ArgType), |
5978 | NEONMAP1(vqshrnd_n_s64, aarch64_neon_sqshrn, AddRetType), |
5979 | NEONMAP1(vqshrnd_n_u64, aarch64_neon_uqshrn, AddRetType), |
5980 | NEONMAP1(vqshrnh_n_s16, aarch64_neon_sqshrn, VectorRet | Use64BitVectors), |
5981 | NEONMAP1(vqshrnh_n_u16, aarch64_neon_uqshrn, VectorRet | Use64BitVectors), |
5982 | NEONMAP1(vqshrns_n_s32, aarch64_neon_sqshrn, VectorRet | Use64BitVectors), |
5983 | NEONMAP1(vqshrns_n_u32, aarch64_neon_uqshrn, VectorRet | Use64BitVectors), |
5984 | NEONMAP1(vqshrund_n_s64, aarch64_neon_sqshrun, AddRetType), |
5985 | NEONMAP1(vqshrunh_n_s16, aarch64_neon_sqshrun, VectorRet | Use64BitVectors), |
5986 | NEONMAP1(vqshruns_n_s32, aarch64_neon_sqshrun, VectorRet | Use64BitVectors), |
5987 | NEONMAP1(vqsubb_s8, aarch64_neon_sqs |