File: | tools/clang/lib/CodeGen/CGBuiltin.cpp |
Warning: | line 1401, column 20 Called C++ object pointer is null |
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/Decl.h" | |||
24 | #include "clang/AST/OSLog.h" | |||
25 | #include "clang/Basic/TargetBuiltins.h" | |||
26 | #include "clang/Basic/TargetInfo.h" | |||
27 | #include "clang/CodeGen/CGFunctionInfo.h" | |||
28 | #include "llvm/ADT/SmallPtrSet.h" | |||
29 | #include "llvm/ADT/StringExtras.h" | |||
30 | #include "llvm/IR/DataLayout.h" | |||
31 | #include "llvm/IR/InlineAsm.h" | |||
32 | #include "llvm/IR/Intrinsics.h" | |||
33 | #include "llvm/IR/MDBuilder.h" | |||
34 | #include "llvm/Support/ConvertUTF.h" | |||
35 | #include "llvm/Support/ScopedPrinter.h" | |||
36 | #include "llvm/Support/TargetParser.h" | |||
37 | #include <sstream> | |||
38 | ||||
39 | using namespace clang; | |||
40 | using namespace CodeGen; | |||
41 | using namespace llvm; | |||
42 | ||||
43 | static | |||
44 | int64_t clamp(int64_t Value, int64_t Low, int64_t High) { | |||
45 | return std::min(High, std::max(Low, Value)); | |||
46 | } | |||
47 | ||||
48 | static void initializeAlloca(CodeGenFunction &CGF, AllocaInst *AI, Value *Size, unsigned AlignmentInBytes) { | |||
49 | ConstantInt *Byte; | |||
50 | switch (CGF.getLangOpts().getTrivialAutoVarInit()) { | |||
51 | case LangOptions::TrivialAutoVarInitKind::Uninitialized: | |||
52 | // Nothing to initialize. | |||
53 | return; | |||
54 | case LangOptions::TrivialAutoVarInitKind::Zero: | |||
55 | Byte = CGF.Builder.getInt8(0x00); | |||
56 | break; | |||
57 | case LangOptions::TrivialAutoVarInitKind::Pattern: { | |||
58 | llvm::Type *Int8 = llvm::IntegerType::getInt8Ty(CGF.CGM.getLLVMContext()); | |||
59 | Byte = llvm::dyn_cast<llvm::ConstantInt>( | |||
60 | initializationPatternFor(CGF.CGM, Int8)); | |||
61 | break; | |||
62 | } | |||
63 | } | |||
64 | CGF.Builder.CreateMemSet(AI, Byte, Size, AlignmentInBytes); | |||
65 | } | |||
66 | ||||
67 | /// getBuiltinLibFunction - Given a builtin id for a function like | |||
68 | /// "__builtin_fabsf", return a Function* for "fabsf". | |||
69 | llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD, | |||
70 | unsigned BuiltinID) { | |||
71 | assert(Context.BuiltinInfo.isLibFunction(BuiltinID))((Context.BuiltinInfo.isLibFunction(BuiltinID)) ? static_cast <void> (0) : __assert_fail ("Context.BuiltinInfo.isLibFunction(BuiltinID)" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 71, __PRETTY_FUNCTION__)); | |||
72 | ||||
73 | // Get the name, skip over the __builtin_ prefix (if necessary). | |||
74 | StringRef Name; | |||
75 | GlobalDecl D(FD); | |||
76 | ||||
77 | // If the builtin has been declared explicitly with an assembler label, | |||
78 | // use the mangled name. This differs from the plain label on platforms | |||
79 | // that prefix labels. | |||
80 | if (FD->hasAttr<AsmLabelAttr>()) | |||
81 | Name = getMangledName(D); | |||
82 | else | |||
83 | Name = Context.BuiltinInfo.getName(BuiltinID) + 10; | |||
84 | ||||
85 | llvm::FunctionType *Ty = | |||
86 | cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType())); | |||
87 | ||||
88 | return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false); | |||
89 | } | |||
90 | ||||
91 | /// Emit the conversions required to turn the given value into an | |||
92 | /// integer of the given size. | |||
93 | static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V, | |||
94 | QualType T, llvm::IntegerType *IntType) { | |||
95 | V = CGF.EmitToMemory(V, T); | |||
96 | ||||
97 | if (V->getType()->isPointerTy()) | |||
98 | return CGF.Builder.CreatePtrToInt(V, IntType); | |||
99 | ||||
100 | assert(V->getType() == IntType)((V->getType() == IntType) ? static_cast<void> (0) : __assert_fail ("V->getType() == IntType", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 100, __PRETTY_FUNCTION__)); | |||
101 | return V; | |||
102 | } | |||
103 | ||||
104 | static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V, | |||
105 | QualType T, llvm::Type *ResultType) { | |||
106 | V = CGF.EmitFromMemory(V, T); | |||
107 | ||||
108 | if (ResultType->isPointerTy()) | |||
109 | return CGF.Builder.CreateIntToPtr(V, ResultType); | |||
110 | ||||
111 | assert(V->getType() == ResultType)((V->getType() == ResultType) ? static_cast<void> (0 ) : __assert_fail ("V->getType() == ResultType", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 111, __PRETTY_FUNCTION__)); | |||
112 | return V; | |||
113 | } | |||
114 | ||||
115 | /// Utility to insert an atomic instruction based on Intrinsic::ID | |||
116 | /// and the expression node. | |||
117 | static Value *MakeBinaryAtomicValue( | |||
118 | CodeGenFunction &CGF, llvm::AtomicRMWInst::BinOp Kind, const CallExpr *E, | |||
119 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { | |||
120 | QualType T = E->getType(); | |||
121 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 121, __PRETTY_FUNCTION__)); | |||
122 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 123, __PRETTY_FUNCTION__)) | |||
123 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 123, __PRETTY_FUNCTION__)); | |||
124 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 124, __PRETTY_FUNCTION__)); | |||
125 | ||||
126 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); | |||
127 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); | |||
128 | ||||
129 | llvm::IntegerType *IntType = | |||
130 | llvm::IntegerType::get(CGF.getLLVMContext(), | |||
131 | CGF.getContext().getTypeSize(T)); | |||
132 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); | |||
133 | ||||
134 | llvm::Value *Args[2]; | |||
135 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); | |||
136 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); | |||
137 | llvm::Type *ValueType = Args[1]->getType(); | |||
138 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); | |||
139 | ||||
140 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( | |||
141 | Kind, Args[0], Args[1], Ordering); | |||
142 | return EmitFromInt(CGF, Result, T, ValueType); | |||
143 | } | |||
144 | ||||
145 | static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) { | |||
146 | Value *Val = CGF.EmitScalarExpr(E->getArg(0)); | |||
147 | Value *Address = CGF.EmitScalarExpr(E->getArg(1)); | |||
148 | ||||
149 | // Convert the type of the pointer to a pointer to the stored type. | |||
150 | Val = CGF.EmitToMemory(Val, E->getArg(0)->getType()); | |||
151 | Value *BC = CGF.Builder.CreateBitCast( | |||
152 | Address, llvm::PointerType::getUnqual(Val->getType()), "cast"); | |||
153 | LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType()); | |||
154 | LV.setNontemporal(true); | |||
155 | CGF.EmitStoreOfScalar(Val, LV, false); | |||
156 | return nullptr; | |||
157 | } | |||
158 | ||||
159 | static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) { | |||
160 | Value *Address = CGF.EmitScalarExpr(E->getArg(0)); | |||
161 | ||||
162 | LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType()); | |||
163 | LV.setNontemporal(true); | |||
164 | return CGF.EmitLoadOfScalar(LV, E->getExprLoc()); | |||
165 | } | |||
166 | ||||
167 | static RValue EmitBinaryAtomic(CodeGenFunction &CGF, | |||
168 | llvm::AtomicRMWInst::BinOp Kind, | |||
169 | const CallExpr *E) { | |||
170 | return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E)); | |||
171 | } | |||
172 | ||||
173 | /// Utility to insert an atomic instruction based Intrinsic::ID and | |||
174 | /// the expression node, where the return value is the result of the | |||
175 | /// operation. | |||
176 | static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF, | |||
177 | llvm::AtomicRMWInst::BinOp Kind, | |||
178 | const CallExpr *E, | |||
179 | Instruction::BinaryOps Op, | |||
180 | bool Invert = false) { | |||
181 | QualType T = E->getType(); | |||
182 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 182, __PRETTY_FUNCTION__)); | |||
183 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 184, __PRETTY_FUNCTION__)) | |||
184 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 184, __PRETTY_FUNCTION__)); | |||
185 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 185, __PRETTY_FUNCTION__)); | |||
186 | ||||
187 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); | |||
188 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); | |||
189 | ||||
190 | llvm::IntegerType *IntType = | |||
191 | llvm::IntegerType::get(CGF.getLLVMContext(), | |||
192 | CGF.getContext().getTypeSize(T)); | |||
193 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); | |||
194 | ||||
195 | llvm::Value *Args[2]; | |||
196 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); | |||
197 | llvm::Type *ValueType = Args[1]->getType(); | |||
198 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); | |||
199 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); | |||
200 | ||||
201 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( | |||
202 | Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent); | |||
203 | Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]); | |||
204 | if (Invert) | |||
205 | Result = CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result, | |||
206 | llvm::ConstantInt::get(IntType, -1)); | |||
207 | Result = EmitFromInt(CGF, Result, T, ValueType); | |||
208 | return RValue::get(Result); | |||
209 | } | |||
210 | ||||
211 | /// Utility to insert an atomic cmpxchg instruction. | |||
212 | /// | |||
213 | /// @param CGF The current codegen function. | |||
214 | /// @param E Builtin call expression to convert to cmpxchg. | |||
215 | /// arg0 - address to operate on | |||
216 | /// arg1 - value to compare with | |||
217 | /// arg2 - new value | |||
218 | /// @param ReturnBool Specifies whether to return success flag of | |||
219 | /// cmpxchg result or the old value. | |||
220 | /// | |||
221 | /// @returns result of cmpxchg, according to ReturnBool | |||
222 | /// | |||
223 | /// Note: In order to lower Microsoft's _InterlockedCompareExchange* intrinsics | |||
224 | /// invoke the function EmitAtomicCmpXchgForMSIntrin. | |||
225 | static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E, | |||
226 | bool ReturnBool) { | |||
227 | QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType(); | |||
228 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); | |||
229 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); | |||
230 | ||||
231 | llvm::IntegerType *IntType = llvm::IntegerType::get( | |||
232 | CGF.getLLVMContext(), CGF.getContext().getTypeSize(T)); | |||
233 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); | |||
234 | ||||
235 | Value *Args[3]; | |||
236 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); | |||
237 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); | |||
238 | llvm::Type *ValueType = Args[1]->getType(); | |||
239 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); | |||
240 | Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType); | |||
241 | ||||
242 | Value *Pair = CGF.Builder.CreateAtomicCmpXchg( | |||
243 | Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent, | |||
244 | llvm::AtomicOrdering::SequentiallyConsistent); | |||
245 | if (ReturnBool) | |||
246 | // Extract boolean success flag and zext it to int. | |||
247 | return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1), | |||
248 | CGF.ConvertType(E->getType())); | |||
249 | else | |||
250 | // Extract old value and emit it using the same type as compare value. | |||
251 | return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T, | |||
252 | ValueType); | |||
253 | } | |||
254 | ||||
255 | /// This function should be invoked to emit atomic cmpxchg for Microsoft's | |||
256 | /// _InterlockedCompareExchange* intrinsics which have the following signature: | |||
257 | /// T _InterlockedCompareExchange(T volatile *Destination, | |||
258 | /// T Exchange, | |||
259 | /// T Comparand); | |||
260 | /// | |||
261 | /// Whereas the llvm 'cmpxchg' instruction has the following syntax: | |||
262 | /// cmpxchg *Destination, Comparand, Exchange. | |||
263 | /// So we need to swap Comparand and Exchange when invoking | |||
264 | /// CreateAtomicCmpXchg. That is the reason we could not use the above utility | |||
265 | /// function MakeAtomicCmpXchgValue since it expects the arguments to be | |||
266 | /// already swapped. | |||
267 | ||||
268 | static | |||
269 | Value *EmitAtomicCmpXchgForMSIntrin(CodeGenFunction &CGF, const CallExpr *E, | |||
270 | AtomicOrdering SuccessOrdering = AtomicOrdering::SequentiallyConsistent) { | |||
271 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 271, __PRETTY_FUNCTION__)); | |||
272 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 273, __PRETTY_FUNCTION__)) | |||
273 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 273, __PRETTY_FUNCTION__)); | |||
274 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 275, __PRETTY_FUNCTION__)) | |||
275 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 275, __PRETTY_FUNCTION__)); | |||
276 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 277, __PRETTY_FUNCTION__)) | |||
277 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 277, __PRETTY_FUNCTION__)); | |||
278 | ||||
279 | auto *Destination = CGF.EmitScalarExpr(E->getArg(0)); | |||
280 | auto *Comparand = CGF.EmitScalarExpr(E->getArg(2)); | |||
281 | auto *Exchange = CGF.EmitScalarExpr(E->getArg(1)); | |||
282 | ||||
283 | // For Release ordering, the failure ordering should be Monotonic. | |||
284 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release ? | |||
285 | AtomicOrdering::Monotonic : | |||
286 | SuccessOrdering; | |||
287 | ||||
288 | auto *Result = CGF.Builder.CreateAtomicCmpXchg( | |||
289 | Destination, Comparand, Exchange, | |||
290 | SuccessOrdering, FailureOrdering); | |||
291 | Result->setVolatile(true); | |||
292 | return CGF.Builder.CreateExtractValue(Result, 0); | |||
293 | } | |||
294 | ||||
295 | static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E, | |||
296 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { | |||
297 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 297, __PRETTY_FUNCTION__)); | |||
298 | ||||
299 | auto *IntTy = CGF.ConvertType(E->getType()); | |||
300 | auto *Result = CGF.Builder.CreateAtomicRMW( | |||
301 | AtomicRMWInst::Add, | |||
302 | CGF.EmitScalarExpr(E->getArg(0)), | |||
303 | ConstantInt::get(IntTy, 1), | |||
304 | Ordering); | |||
305 | return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1)); | |||
306 | } | |||
307 | ||||
308 | static Value *EmitAtomicDecrementValue(CodeGenFunction &CGF, const CallExpr *E, | |||
309 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { | |||
310 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 310, __PRETTY_FUNCTION__)); | |||
311 | ||||
312 | auto *IntTy = CGF.ConvertType(E->getType()); | |||
313 | auto *Result = CGF.Builder.CreateAtomicRMW( | |||
314 | AtomicRMWInst::Sub, | |||
315 | CGF.EmitScalarExpr(E->getArg(0)), | |||
316 | ConstantInt::get(IntTy, 1), | |||
317 | Ordering); | |||
318 | return CGF.Builder.CreateSub(Result, ConstantInt::get(IntTy, 1)); | |||
319 | } | |||
320 | ||||
321 | // Build a plain volatile load. | |||
322 | static Value *EmitISOVolatileLoad(CodeGenFunction &CGF, const CallExpr *E) { | |||
323 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); | |||
324 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); | |||
325 | CharUnits LoadSize = CGF.getContext().getTypeSizeInChars(ElTy); | |||
326 | llvm::Type *ITy = | |||
327 | llvm::IntegerType::get(CGF.getLLVMContext(), LoadSize.getQuantity() * 8); | |||
328 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); | |||
329 | llvm::LoadInst *Load = CGF.Builder.CreateAlignedLoad(Ptr, LoadSize); | |||
330 | Load->setVolatile(true); | |||
331 | return Load; | |||
332 | } | |||
333 | ||||
334 | // Build a plain volatile store. | |||
335 | static Value *EmitISOVolatileStore(CodeGenFunction &CGF, const CallExpr *E) { | |||
336 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); | |||
337 | Value *Value = CGF.EmitScalarExpr(E->getArg(1)); | |||
338 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); | |||
339 | CharUnits StoreSize = CGF.getContext().getTypeSizeInChars(ElTy); | |||
340 | llvm::Type *ITy = | |||
341 | llvm::IntegerType::get(CGF.getLLVMContext(), StoreSize.getQuantity() * 8); | |||
342 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); | |||
343 | llvm::StoreInst *Store = | |||
344 | CGF.Builder.CreateAlignedStore(Value, Ptr, StoreSize); | |||
345 | Store->setVolatile(true); | |||
346 | return Store; | |||
347 | } | |||
348 | ||||
349 | // Emit a simple mangled intrinsic that has 1 argument and a return type | |||
350 | // matching the argument type. | |||
351 | static Value *emitUnaryBuiltin(CodeGenFunction &CGF, | |||
352 | const CallExpr *E, | |||
353 | unsigned IntrinsicID) { | |||
354 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
355 | ||||
356 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
357 | return CGF.Builder.CreateCall(F, Src0); | |||
358 | } | |||
359 | ||||
360 | // Emit an intrinsic that has 2 operands of the same type as its result. | |||
361 | static Value *emitBinaryBuiltin(CodeGenFunction &CGF, | |||
362 | const CallExpr *E, | |||
363 | unsigned IntrinsicID) { | |||
364 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
365 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); | |||
366 | ||||
367 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
368 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); | |||
369 | } | |||
370 | ||||
371 | // Emit an intrinsic that has 3 operands of the same type as its result. | |||
372 | static Value *emitTernaryBuiltin(CodeGenFunction &CGF, | |||
373 | const CallExpr *E, | |||
374 | unsigned IntrinsicID) { | |||
375 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
376 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); | |||
377 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); | |||
378 | ||||
379 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
380 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); | |||
381 | } | |||
382 | ||||
383 | // Emit an intrinsic that has 1 float or double operand, and 1 integer. | |||
384 | static Value *emitFPIntBuiltin(CodeGenFunction &CGF, | |||
385 | const CallExpr *E, | |||
386 | unsigned IntrinsicID) { | |||
387 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
388 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); | |||
389 | ||||
390 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
391 | return CGF.Builder.CreateCall(F, {Src0, Src1}); | |||
392 | } | |||
393 | ||||
394 | // Emit an intrinsic that has overloaded integer result and fp operand. | |||
395 | static Value *emitFPToIntRoundBuiltin(CodeGenFunction &CGF, | |||
396 | const CallExpr *E, | |||
397 | unsigned IntrinsicID) { | |||
398 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); | |||
399 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
400 | ||||
401 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, | |||
402 | {ResultType, Src0->getType()}); | |||
403 | return CGF.Builder.CreateCall(F, Src0); | |||
404 | } | |||
405 | ||||
406 | /// EmitFAbs - Emit a call to @llvm.fabs(). | |||
407 | static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) { | |||
408 | Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType()); | |||
409 | llvm::CallInst *Call = CGF.Builder.CreateCall(F, V); | |||
410 | Call->setDoesNotAccessMemory(); | |||
411 | return Call; | |||
412 | } | |||
413 | ||||
414 | /// Emit the computation of the sign bit for a floating point value. Returns | |||
415 | /// the i1 sign bit value. | |||
416 | static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) { | |||
417 | LLVMContext &C = CGF.CGM.getLLVMContext(); | |||
418 | ||||
419 | llvm::Type *Ty = V->getType(); | |||
420 | int Width = Ty->getPrimitiveSizeInBits(); | |||
421 | llvm::Type *IntTy = llvm::IntegerType::get(C, Width); | |||
422 | V = CGF.Builder.CreateBitCast(V, IntTy); | |||
423 | if (Ty->isPPC_FP128Ty()) { | |||
424 | // We want the sign bit of the higher-order double. The bitcast we just | |||
425 | // did works as if the double-double was stored to memory and then | |||
426 | // read as an i128. The "store" will put the higher-order double in the | |||
427 | // lower address in both little- and big-Endian modes, but the "load" | |||
428 | // will treat those bits as a different part of the i128: the low bits in | |||
429 | // little-Endian, the high bits in big-Endian. Therefore, on big-Endian | |||
430 | // we need to shift the high bits down to the low before truncating. | |||
431 | Width >>= 1; | |||
432 | if (CGF.getTarget().isBigEndian()) { | |||
433 | Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width); | |||
434 | V = CGF.Builder.CreateLShr(V, ShiftCst); | |||
435 | } | |||
436 | // We are truncating value in order to extract the higher-order | |||
437 | // double, which we will be using to extract the sign from. | |||
438 | IntTy = llvm::IntegerType::get(C, Width); | |||
439 | V = CGF.Builder.CreateTrunc(V, IntTy); | |||
440 | } | |||
441 | Value *Zero = llvm::Constant::getNullValue(IntTy); | |||
442 | return CGF.Builder.CreateICmpSLT(V, Zero); | |||
443 | } | |||
444 | ||||
445 | static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD, | |||
446 | const CallExpr *E, llvm::Constant *calleeValue) { | |||
447 | CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD)); | |||
448 | return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot()); | |||
449 | } | |||
450 | ||||
451 | /// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.* | |||
452 | /// depending on IntrinsicID. | |||
453 | /// | |||
454 | /// \arg CGF The current codegen function. | |||
455 | /// \arg IntrinsicID The ID for the Intrinsic we wish to generate. | |||
456 | /// \arg X The first argument to the llvm.*.with.overflow.*. | |||
457 | /// \arg Y The second argument to the llvm.*.with.overflow.*. | |||
458 | /// \arg Carry The carry returned by the llvm.*.with.overflow.*. | |||
459 | /// \returns The result (i.e. sum/product) returned by the intrinsic. | |||
460 | static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF, | |||
461 | const llvm::Intrinsic::ID IntrinsicID, | |||
462 | llvm::Value *X, llvm::Value *Y, | |||
463 | llvm::Value *&Carry) { | |||
464 | // Make sure we have integers of the same width. | |||
465 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 467, __PRETTY_FUNCTION__)) | |||
466 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 467, __PRETTY_FUNCTION__)) | |||
467 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 467, __PRETTY_FUNCTION__)); | |||
468 | ||||
469 | Function *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType()); | |||
470 | llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y}); | |||
471 | Carry = CGF.Builder.CreateExtractValue(Tmp, 1); | |||
472 | return CGF.Builder.CreateExtractValue(Tmp, 0); | |||
473 | } | |||
474 | ||||
475 | static Value *emitRangedBuiltin(CodeGenFunction &CGF, | |||
476 | unsigned IntrinsicID, | |||
477 | int low, int high) { | |||
478 | llvm::MDBuilder MDHelper(CGF.getLLVMContext()); | |||
479 | llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high)); | |||
480 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, {}); | |||
481 | llvm::Instruction *Call = CGF.Builder.CreateCall(F); | |||
482 | Call->setMetadata(llvm::LLVMContext::MD_range, RNode); | |||
483 | return Call; | |||
484 | } | |||
485 | ||||
486 | namespace { | |||
487 | struct WidthAndSignedness { | |||
488 | unsigned Width; | |||
489 | bool Signed; | |||
490 | }; | |||
491 | } | |||
492 | ||||
493 | static WidthAndSignedness | |||
494 | getIntegerWidthAndSignedness(const clang::ASTContext &context, | |||
495 | const clang::QualType Type) { | |||
496 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 496, __PRETTY_FUNCTION__)); | |||
497 | unsigned Width = Type->isBooleanType() ? 1 : context.getTypeInfo(Type).Width; | |||
498 | bool Signed = Type->isSignedIntegerType(); | |||
499 | return {Width, Signed}; | |||
500 | } | |||
501 | ||||
502 | // Given one or more integer types, this function produces an integer type that | |||
503 | // encompasses them: any value in one of the given types could be expressed in | |||
504 | // the encompassing type. | |||
505 | static struct WidthAndSignedness | |||
506 | EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) { | |||
507 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 507, __PRETTY_FUNCTION__)); | |||
508 | ||||
509 | // If any of the given types is signed, we must return a signed type. | |||
510 | bool Signed = false; | |||
511 | for (const auto &Type : Types) { | |||
512 | Signed |= Type.Signed; | |||
513 | } | |||
514 | ||||
515 | // The encompassing type must have a width greater than or equal to the width | |||
516 | // of the specified types. Additionally, if the encompassing type is signed, | |||
517 | // its width must be strictly greater than the width of any unsigned types | |||
518 | // given. | |||
519 | unsigned Width = 0; | |||
520 | for (const auto &Type : Types) { | |||
521 | unsigned MinWidth = Type.Width + (Signed && !Type.Signed); | |||
522 | if (Width < MinWidth) { | |||
523 | Width = MinWidth; | |||
524 | } | |||
525 | } | |||
526 | ||||
527 | return {Width, Signed}; | |||
528 | } | |||
529 | ||||
530 | Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) { | |||
531 | llvm::Type *DestType = Int8PtrTy; | |||
532 | if (ArgValue->getType() != DestType) | |||
533 | ArgValue = | |||
534 | Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data()); | |||
535 | ||||
536 | Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend; | |||
537 | return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue); | |||
538 | } | |||
539 | ||||
540 | /// Checks if using the result of __builtin_object_size(p, @p From) in place of | |||
541 | /// __builtin_object_size(p, @p To) is correct | |||
542 | static bool areBOSTypesCompatible(int From, int To) { | |||
543 | // Note: Our __builtin_object_size implementation currently treats Type=0 and | |||
544 | // Type=2 identically. Encoding this implementation detail here may make | |||
545 | // improving __builtin_object_size difficult in the future, so it's omitted. | |||
546 | return From == To || (From == 0 && To == 1) || (From == 3 && To == 2); | |||
547 | } | |||
548 | ||||
549 | static llvm::Value * | |||
550 | getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) { | |||
551 | return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true); | |||
552 | } | |||
553 | ||||
554 | llvm::Value * | |||
555 | CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type, | |||
556 | llvm::IntegerType *ResType, | |||
557 | llvm::Value *EmittedE, | |||
558 | bool IsDynamic) { | |||
559 | uint64_t ObjectSize; | |||
560 | if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type)) | |||
561 | return emitBuiltinObjectSize(E, Type, ResType, EmittedE, IsDynamic); | |||
562 | return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true); | |||
563 | } | |||
564 | ||||
565 | /// Returns a Value corresponding to the size of the given expression. | |||
566 | /// This Value may be either of the following: | |||
567 | /// - A llvm::Argument (if E is a param with the pass_object_size attribute on | |||
568 | /// it) | |||
569 | /// - A call to the @llvm.objectsize intrinsic | |||
570 | /// | |||
571 | /// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null | |||
572 | /// and we wouldn't otherwise try to reference a pass_object_size parameter, | |||
573 | /// we'll call @llvm.objectsize on EmittedE, rather than emitting E. | |||
574 | llvm::Value * | |||
575 | CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type, | |||
576 | llvm::IntegerType *ResType, | |||
577 | llvm::Value *EmittedE, bool IsDynamic) { | |||
578 | // We need to reference an argument if the pointer is a parameter with the | |||
579 | // pass_object_size attribute. | |||
580 | if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) { | |||
581 | auto *Param = dyn_cast<ParmVarDecl>(D->getDecl()); | |||
582 | auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>(); | |||
583 | if (Param != nullptr && PS != nullptr && | |||
584 | areBOSTypesCompatible(PS->getType(), Type)) { | |||
585 | auto Iter = SizeArguments.find(Param); | |||
586 | assert(Iter != SizeArguments.end())((Iter != SizeArguments.end()) ? static_cast<void> (0) : __assert_fail ("Iter != SizeArguments.end()", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 586, __PRETTY_FUNCTION__)); | |||
587 | ||||
588 | const ImplicitParamDecl *D = Iter->second; | |||
589 | auto DIter = LocalDeclMap.find(D); | |||
590 | assert(DIter != LocalDeclMap.end())((DIter != LocalDeclMap.end()) ? static_cast<void> (0) : __assert_fail ("DIter != LocalDeclMap.end()", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 590, __PRETTY_FUNCTION__)); | |||
591 | ||||
592 | return EmitLoadOfScalar(DIter->second, /*Volatile=*/false, | |||
593 | getContext().getSizeType(), E->getBeginLoc()); | |||
594 | } | |||
595 | } | |||
596 | ||||
597 | // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't | |||
598 | // evaluate E for side-effects. In either case, we shouldn't lower to | |||
599 | // @llvm.objectsize. | |||
600 | if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext()))) | |||
601 | return getDefaultBuiltinObjectSizeResult(Type, ResType); | |||
602 | ||||
603 | Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E); | |||
604 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 605, __PRETTY_FUNCTION__)) | |||
605 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 605, __PRETTY_FUNCTION__)); | |||
606 | ||||
607 | Function *F = | |||
608 | CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()}); | |||
609 | ||||
610 | // LLVM only supports 0 and 2, make sure that we pass along that as a boolean. | |||
611 | Value *Min = Builder.getInt1((Type & 2) != 0); | |||
612 | // For GCC compatibility, __builtin_object_size treat NULL as unknown size. | |||
613 | Value *NullIsUnknown = Builder.getTrue(); | |||
614 | Value *Dynamic = Builder.getInt1(IsDynamic); | |||
615 | return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown, Dynamic}); | |||
616 | } | |||
617 | ||||
618 | namespace { | |||
619 | /// A struct to generically describe a bit test intrinsic. | |||
620 | struct BitTest { | |||
621 | enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set }; | |||
622 | enum InterlockingKind : uint8_t { | |||
623 | Unlocked, | |||
624 | Sequential, | |||
625 | Acquire, | |||
626 | Release, | |||
627 | NoFence | |||
628 | }; | |||
629 | ||||
630 | ActionKind Action; | |||
631 | InterlockingKind Interlocking; | |||
632 | bool Is64Bit; | |||
633 | ||||
634 | static BitTest decodeBitTestBuiltin(unsigned BuiltinID); | |||
635 | }; | |||
636 | } // namespace | |||
637 | ||||
638 | BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) { | |||
639 | switch (BuiltinID) { | |||
640 | // Main portable variants. | |||
641 | case Builtin::BI_bittest: | |||
642 | return {TestOnly, Unlocked, false}; | |||
643 | case Builtin::BI_bittestandcomplement: | |||
644 | return {Complement, Unlocked, false}; | |||
645 | case Builtin::BI_bittestandreset: | |||
646 | return {Reset, Unlocked, false}; | |||
647 | case Builtin::BI_bittestandset: | |||
648 | return {Set, Unlocked, false}; | |||
649 | case Builtin::BI_interlockedbittestandreset: | |||
650 | return {Reset, Sequential, false}; | |||
651 | case Builtin::BI_interlockedbittestandset: | |||
652 | return {Set, Sequential, false}; | |||
653 | ||||
654 | // X86-specific 64-bit variants. | |||
655 | case Builtin::BI_bittest64: | |||
656 | return {TestOnly, Unlocked, true}; | |||
657 | case Builtin::BI_bittestandcomplement64: | |||
658 | return {Complement, Unlocked, true}; | |||
659 | case Builtin::BI_bittestandreset64: | |||
660 | return {Reset, Unlocked, true}; | |||
661 | case Builtin::BI_bittestandset64: | |||
662 | return {Set, Unlocked, true}; | |||
663 | case Builtin::BI_interlockedbittestandreset64: | |||
664 | return {Reset, Sequential, true}; | |||
665 | case Builtin::BI_interlockedbittestandset64: | |||
666 | return {Set, Sequential, true}; | |||
667 | ||||
668 | // ARM/AArch64-specific ordering variants. | |||
669 | case Builtin::BI_interlockedbittestandset_acq: | |||
670 | return {Set, Acquire, false}; | |||
671 | case Builtin::BI_interlockedbittestandset_rel: | |||
672 | return {Set, Release, false}; | |||
673 | case Builtin::BI_interlockedbittestandset_nf: | |||
674 | return {Set, NoFence, false}; | |||
675 | case Builtin::BI_interlockedbittestandreset_acq: | |||
676 | return {Reset, Acquire, false}; | |||
677 | case Builtin::BI_interlockedbittestandreset_rel: | |||
678 | return {Reset, Release, false}; | |||
679 | case Builtin::BI_interlockedbittestandreset_nf: | |||
680 | return {Reset, NoFence, false}; | |||
681 | } | |||
682 | llvm_unreachable("expected only bittest intrinsics")::llvm::llvm_unreachable_internal("expected only bittest intrinsics" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 682); | |||
683 | } | |||
684 | ||||
685 | static char bitActionToX86BTCode(BitTest::ActionKind A) { | |||
686 | switch (A) { | |||
687 | case BitTest::TestOnly: return '\0'; | |||
688 | case BitTest::Complement: return 'c'; | |||
689 | case BitTest::Reset: return 'r'; | |||
690 | case BitTest::Set: return 's'; | |||
691 | } | |||
692 | llvm_unreachable("invalid action")::llvm::llvm_unreachable_internal("invalid action", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 692); | |||
693 | } | |||
694 | ||||
695 | static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF, | |||
696 | BitTest BT, | |||
697 | const CallExpr *E, Value *BitBase, | |||
698 | Value *BitPos) { | |||
699 | char Action = bitActionToX86BTCode(BT.Action); | |||
700 | char SizeSuffix = BT.Is64Bit ? 'q' : 'l'; | |||
701 | ||||
702 | // Build the assembly. | |||
703 | SmallString<64> Asm; | |||
704 | raw_svector_ostream AsmOS(Asm); | |||
705 | if (BT.Interlocking != BitTest::Unlocked) | |||
706 | AsmOS << "lock "; | |||
707 | AsmOS << "bt"; | |||
708 | if (Action) | |||
709 | AsmOS << Action; | |||
710 | AsmOS << SizeSuffix << " $2, ($1)\n\tsetc ${0:b}"; | |||
711 | ||||
712 | // Build the constraints. FIXME: We should support immediates when possible. | |||
713 | std::string Constraints = "=r,r,r,~{cc},~{flags},~{fpsr}"; | |||
714 | llvm::IntegerType *IntType = llvm::IntegerType::get( | |||
715 | CGF.getLLVMContext(), | |||
716 | CGF.getContext().getTypeSize(E->getArg(1)->getType())); | |||
717 | llvm::Type *IntPtrType = IntType->getPointerTo(); | |||
718 | llvm::FunctionType *FTy = | |||
719 | llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false); | |||
720 | ||||
721 | llvm::InlineAsm *IA = | |||
722 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); | |||
723 | return CGF.Builder.CreateCall(IA, {BitBase, BitPos}); | |||
724 | } | |||
725 | ||||
726 | static llvm::AtomicOrdering | |||
727 | getBitTestAtomicOrdering(BitTest::InterlockingKind I) { | |||
728 | switch (I) { | |||
729 | case BitTest::Unlocked: return llvm::AtomicOrdering::NotAtomic; | |||
730 | case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent; | |||
731 | case BitTest::Acquire: return llvm::AtomicOrdering::Acquire; | |||
732 | case BitTest::Release: return llvm::AtomicOrdering::Release; | |||
733 | case BitTest::NoFence: return llvm::AtomicOrdering::Monotonic; | |||
734 | } | |||
735 | llvm_unreachable("invalid interlocking")::llvm::llvm_unreachable_internal("invalid interlocking", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 735); | |||
736 | } | |||
737 | ||||
738 | /// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of | |||
739 | /// bits and a bit position and read and optionally modify the bit at that | |||
740 | /// position. The position index can be arbitrarily large, i.e. it can be larger | |||
741 | /// than 31 or 63, so we need an indexed load in the general case. | |||
742 | static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF, | |||
743 | unsigned BuiltinID, | |||
744 | const CallExpr *E) { | |||
745 | Value *BitBase = CGF.EmitScalarExpr(E->getArg(0)); | |||
746 | Value *BitPos = CGF.EmitScalarExpr(E->getArg(1)); | |||
747 | ||||
748 | BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID); | |||
749 | ||||
750 | // X86 has special BT, BTC, BTR, and BTS instructions that handle the array | |||
751 | // indexing operation internally. Use them if possible. | |||
752 | llvm::Triple::ArchType Arch = CGF.getTarget().getTriple().getArch(); | |||
753 | if (Arch == llvm::Triple::x86 || Arch == llvm::Triple::x86_64) | |||
754 | return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos); | |||
755 | ||||
756 | // Otherwise, use generic code to load one byte and test the bit. Use all but | |||
757 | // the bottom three bits as the array index, and the bottom three bits to form | |||
758 | // a mask. | |||
759 | // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0; | |||
760 | Value *ByteIndex = CGF.Builder.CreateAShr( | |||
761 | BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx"); | |||
762 | Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy); | |||
763 | Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8, | |||
764 | ByteIndex, "bittest.byteaddr"), | |||
765 | CharUnits::One()); | |||
766 | Value *PosLow = | |||
767 | CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty), | |||
768 | llvm::ConstantInt::get(CGF.Int8Ty, 0x7)); | |||
769 | ||||
770 | // The updating instructions will need a mask. | |||
771 | Value *Mask = nullptr; | |||
772 | if (BT.Action != BitTest::TestOnly) { | |||
773 | Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow, | |||
774 | "bittest.mask"); | |||
775 | } | |||
776 | ||||
777 | // Check the action and ordering of the interlocked intrinsics. | |||
778 | llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking); | |||
779 | ||||
780 | Value *OldByte = nullptr; | |||
781 | if (Ordering != llvm::AtomicOrdering::NotAtomic) { | |||
782 | // Emit a combined atomicrmw load/store operation for the interlocked | |||
783 | // intrinsics. | |||
784 | llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or; | |||
785 | if (BT.Action == BitTest::Reset) { | |||
786 | Mask = CGF.Builder.CreateNot(Mask); | |||
787 | RMWOp = llvm::AtomicRMWInst::And; | |||
788 | } | |||
789 | OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask, | |||
790 | Ordering); | |||
791 | } else { | |||
792 | // Emit a plain load for the non-interlocked intrinsics. | |||
793 | OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte"); | |||
794 | Value *NewByte = nullptr; | |||
795 | switch (BT.Action) { | |||
796 | case BitTest::TestOnly: | |||
797 | // Don't store anything. | |||
798 | break; | |||
799 | case BitTest::Complement: | |||
800 | NewByte = CGF.Builder.CreateXor(OldByte, Mask); | |||
801 | break; | |||
802 | case BitTest::Reset: | |||
803 | NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask)); | |||
804 | break; | |||
805 | case BitTest::Set: | |||
806 | NewByte = CGF.Builder.CreateOr(OldByte, Mask); | |||
807 | break; | |||
808 | } | |||
809 | if (NewByte) | |||
810 | CGF.Builder.CreateStore(NewByte, ByteAddr); | |||
811 | } | |||
812 | ||||
813 | // However we loaded the old byte, either by plain load or atomicrmw, shift | |||
814 | // the bit into the low position and mask it to 0 or 1. | |||
815 | Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr"); | |||
816 | return CGF.Builder.CreateAnd( | |||
817 | ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res"); | |||
818 | } | |||
819 | ||||
820 | namespace { | |||
821 | enum class MSVCSetJmpKind { | |||
822 | _setjmpex, | |||
823 | _setjmp3, | |||
824 | _setjmp | |||
825 | }; | |||
826 | } | |||
827 | ||||
828 | /// MSVC handles setjmp a bit differently on different platforms. On every | |||
829 | /// architecture except 32-bit x86, the frame address is passed. On x86, extra | |||
830 | /// parameters can be passed as variadic arguments, but we always pass none. | |||
831 | static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind, | |||
832 | const CallExpr *E) { | |||
833 | llvm::Value *Arg1 = nullptr; | |||
834 | llvm::Type *Arg1Ty = nullptr; | |||
835 | StringRef Name; | |||
836 | bool IsVarArg = false; | |||
837 | if (SJKind == MSVCSetJmpKind::_setjmp3) { | |||
838 | Name = "_setjmp3"; | |||
839 | Arg1Ty = CGF.Int32Ty; | |||
840 | Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0); | |||
841 | IsVarArg = true; | |||
842 | } else { | |||
843 | Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex"; | |||
844 | Arg1Ty = CGF.Int8PtrTy; | |||
845 | if (CGF.getTarget().getTriple().getArch() == llvm::Triple::aarch64) { | |||
846 | Arg1 = CGF.Builder.CreateCall( | |||
847 | CGF.CGM.getIntrinsic(Intrinsic::sponentry, CGF.AllocaInt8PtrTy)); | |||
848 | } else | |||
849 | Arg1 = CGF.Builder.CreateCall( | |||
850 | CGF.CGM.getIntrinsic(Intrinsic::frameaddress, CGF.AllocaInt8PtrTy), | |||
851 | llvm::ConstantInt::get(CGF.Int32Ty, 0)); | |||
852 | } | |||
853 | ||||
854 | // Mark the call site and declaration with ReturnsTwice. | |||
855 | llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty}; | |||
856 | llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get( | |||
857 | CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex, | |||
858 | llvm::Attribute::ReturnsTwice); | |||
859 | llvm::FunctionCallee SetJmpFn = CGF.CGM.CreateRuntimeFunction( | |||
860 | llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name, | |||
861 | ReturnsTwiceAttr, /*Local=*/true); | |||
862 | ||||
863 | llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast( | |||
864 | CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy); | |||
865 | llvm::Value *Args[] = {Buf, Arg1}; | |||
866 | llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args); | |||
867 | CB->setAttributes(ReturnsTwiceAttr); | |||
868 | return RValue::get(CB); | |||
869 | } | |||
870 | ||||
871 | // Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code, | |||
872 | // we handle them here. | |||
873 | enum class CodeGenFunction::MSVCIntrin { | |||
874 | _BitScanForward, | |||
875 | _BitScanReverse, | |||
876 | _InterlockedAnd, | |||
877 | _InterlockedDecrement, | |||
878 | _InterlockedExchange, | |||
879 | _InterlockedExchangeAdd, | |||
880 | _InterlockedExchangeSub, | |||
881 | _InterlockedIncrement, | |||
882 | _InterlockedOr, | |||
883 | _InterlockedXor, | |||
884 | _InterlockedExchangeAdd_acq, | |||
885 | _InterlockedExchangeAdd_rel, | |||
886 | _InterlockedExchangeAdd_nf, | |||
887 | _InterlockedExchange_acq, | |||
888 | _InterlockedExchange_rel, | |||
889 | _InterlockedExchange_nf, | |||
890 | _InterlockedCompareExchange_acq, | |||
891 | _InterlockedCompareExchange_rel, | |||
892 | _InterlockedCompareExchange_nf, | |||
893 | _InterlockedOr_acq, | |||
894 | _InterlockedOr_rel, | |||
895 | _InterlockedOr_nf, | |||
896 | _InterlockedXor_acq, | |||
897 | _InterlockedXor_rel, | |||
898 | _InterlockedXor_nf, | |||
899 | _InterlockedAnd_acq, | |||
900 | _InterlockedAnd_rel, | |||
901 | _InterlockedAnd_nf, | |||
902 | _InterlockedIncrement_acq, | |||
903 | _InterlockedIncrement_rel, | |||
904 | _InterlockedIncrement_nf, | |||
905 | _InterlockedDecrement_acq, | |||
906 | _InterlockedDecrement_rel, | |||
907 | _InterlockedDecrement_nf, | |||
908 | __fastfail, | |||
909 | }; | |||
910 | ||||
911 | Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID, | |||
912 | const CallExpr *E) { | |||
913 | switch (BuiltinID) { | |||
914 | case MSVCIntrin::_BitScanForward: | |||
915 | case MSVCIntrin::_BitScanReverse: { | |||
916 | Value *ArgValue = EmitScalarExpr(E->getArg(1)); | |||
917 | ||||
918 | llvm::Type *ArgType = ArgValue->getType(); | |||
919 | llvm::Type *IndexType = | |||
920 | EmitScalarExpr(E->getArg(0))->getType()->getPointerElementType(); | |||
921 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
922 | ||||
923 | Value *ArgZero = llvm::Constant::getNullValue(ArgType); | |||
924 | Value *ResZero = llvm::Constant::getNullValue(ResultType); | |||
925 | Value *ResOne = llvm::ConstantInt::get(ResultType, 1); | |||
926 | ||||
927 | BasicBlock *Begin = Builder.GetInsertBlock(); | |||
928 | BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn); | |||
929 | Builder.SetInsertPoint(End); | |||
930 | PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result"); | |||
931 | ||||
932 | Builder.SetInsertPoint(Begin); | |||
933 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero); | |||
934 | BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn); | |||
935 | Builder.CreateCondBr(IsZero, End, NotZero); | |||
936 | Result->addIncoming(ResZero, Begin); | |||
937 | ||||
938 | Builder.SetInsertPoint(NotZero); | |||
939 | Address IndexAddress = EmitPointerWithAlignment(E->getArg(0)); | |||
940 | ||||
941 | if (BuiltinID == MSVCIntrin::_BitScanForward) { | |||
942 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); | |||
943 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); | |||
944 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); | |||
945 | Builder.CreateStore(ZeroCount, IndexAddress, false); | |||
946 | } else { | |||
947 | unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth(); | |||
948 | Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1); | |||
949 | ||||
950 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); | |||
951 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); | |||
952 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); | |||
953 | Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount); | |||
954 | Builder.CreateStore(Index, IndexAddress, false); | |||
955 | } | |||
956 | Builder.CreateBr(End); | |||
957 | Result->addIncoming(ResOne, NotZero); | |||
958 | ||||
959 | Builder.SetInsertPoint(End); | |||
960 | return Result; | |||
961 | } | |||
962 | case MSVCIntrin::_InterlockedAnd: | |||
963 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E); | |||
964 | case MSVCIntrin::_InterlockedExchange: | |||
965 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E); | |||
966 | case MSVCIntrin::_InterlockedExchangeAdd: | |||
967 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E); | |||
968 | case MSVCIntrin::_InterlockedExchangeSub: | |||
969 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E); | |||
970 | case MSVCIntrin::_InterlockedOr: | |||
971 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E); | |||
972 | case MSVCIntrin::_InterlockedXor: | |||
973 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E); | |||
974 | case MSVCIntrin::_InterlockedExchangeAdd_acq: | |||
975 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, | |||
976 | AtomicOrdering::Acquire); | |||
977 | case MSVCIntrin::_InterlockedExchangeAdd_rel: | |||
978 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, | |||
979 | AtomicOrdering::Release); | |||
980 | case MSVCIntrin::_InterlockedExchangeAdd_nf: | |||
981 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, | |||
982 | AtomicOrdering::Monotonic); | |||
983 | case MSVCIntrin::_InterlockedExchange_acq: | |||
984 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, | |||
985 | AtomicOrdering::Acquire); | |||
986 | case MSVCIntrin::_InterlockedExchange_rel: | |||
987 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, | |||
988 | AtomicOrdering::Release); | |||
989 | case MSVCIntrin::_InterlockedExchange_nf: | |||
990 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, | |||
991 | AtomicOrdering::Monotonic); | |||
992 | case MSVCIntrin::_InterlockedCompareExchange_acq: | |||
993 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Acquire); | |||
994 | case MSVCIntrin::_InterlockedCompareExchange_rel: | |||
995 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Release); | |||
996 | case MSVCIntrin::_InterlockedCompareExchange_nf: | |||
997 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Monotonic); | |||
998 | case MSVCIntrin::_InterlockedOr_acq: | |||
999 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, | |||
1000 | AtomicOrdering::Acquire); | |||
1001 | case MSVCIntrin::_InterlockedOr_rel: | |||
1002 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, | |||
1003 | AtomicOrdering::Release); | |||
1004 | case MSVCIntrin::_InterlockedOr_nf: | |||
1005 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, | |||
1006 | AtomicOrdering::Monotonic); | |||
1007 | case MSVCIntrin::_InterlockedXor_acq: | |||
1008 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, | |||
1009 | AtomicOrdering::Acquire); | |||
1010 | case MSVCIntrin::_InterlockedXor_rel: | |||
1011 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, | |||
1012 | AtomicOrdering::Release); | |||
1013 | case MSVCIntrin::_InterlockedXor_nf: | |||
1014 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, | |||
1015 | AtomicOrdering::Monotonic); | |||
1016 | case MSVCIntrin::_InterlockedAnd_acq: | |||
1017 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, | |||
1018 | AtomicOrdering::Acquire); | |||
1019 | case MSVCIntrin::_InterlockedAnd_rel: | |||
1020 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, | |||
1021 | AtomicOrdering::Release); | |||
1022 | case MSVCIntrin::_InterlockedAnd_nf: | |||
1023 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, | |||
1024 | AtomicOrdering::Monotonic); | |||
1025 | case MSVCIntrin::_InterlockedIncrement_acq: | |||
1026 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire); | |||
1027 | case MSVCIntrin::_InterlockedIncrement_rel: | |||
1028 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release); | |||
1029 | case MSVCIntrin::_InterlockedIncrement_nf: | |||
1030 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic); | |||
1031 | case MSVCIntrin::_InterlockedDecrement_acq: | |||
1032 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Acquire); | |||
1033 | case MSVCIntrin::_InterlockedDecrement_rel: | |||
1034 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Release); | |||
1035 | case MSVCIntrin::_InterlockedDecrement_nf: | |||
1036 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Monotonic); | |||
1037 | ||||
1038 | case MSVCIntrin::_InterlockedDecrement: | |||
1039 | return EmitAtomicDecrementValue(*this, E); | |||
1040 | case MSVCIntrin::_InterlockedIncrement: | |||
1041 | return EmitAtomicIncrementValue(*this, E); | |||
1042 | ||||
1043 | case MSVCIntrin::__fastfail: { | |||
1044 | // Request immediate process termination from the kernel. The instruction | |||
1045 | // sequences to do this are documented on MSDN: | |||
1046 | // https://msdn.microsoft.com/en-us/library/dn774154.aspx | |||
1047 | llvm::Triple::ArchType ISA = getTarget().getTriple().getArch(); | |||
1048 | StringRef Asm, Constraints; | |||
1049 | switch (ISA) { | |||
1050 | default: | |||
1051 | ErrorUnsupported(E, "__fastfail call for this architecture"); | |||
1052 | break; | |||
1053 | case llvm::Triple::x86: | |||
1054 | case llvm::Triple::x86_64: | |||
1055 | Asm = "int $$0x29"; | |||
1056 | Constraints = "{cx}"; | |||
1057 | break; | |||
1058 | case llvm::Triple::thumb: | |||
1059 | Asm = "udf #251"; | |||
1060 | Constraints = "{r0}"; | |||
1061 | break; | |||
1062 | case llvm::Triple::aarch64: | |||
1063 | Asm = "brk #0xF003"; | |||
1064 | Constraints = "{w0}"; | |||
1065 | } | |||
1066 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false); | |||
1067 | llvm::InlineAsm *IA = | |||
1068 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); | |||
1069 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( | |||
1070 | getLLVMContext(), llvm::AttributeList::FunctionIndex, | |||
1071 | llvm::Attribute::NoReturn); | |||
1072 | llvm::CallInst *CI = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0))); | |||
1073 | CI->setAttributes(NoReturnAttr); | |||
1074 | return CI; | |||
1075 | } | |||
1076 | } | |||
1077 | llvm_unreachable("Incorrect MSVC intrinsic!")::llvm::llvm_unreachable_internal("Incorrect MSVC intrinsic!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1077); | |||
1078 | } | |||
1079 | ||||
1080 | namespace { | |||
1081 | // ARC cleanup for __builtin_os_log_format | |||
1082 | struct CallObjCArcUse final : EHScopeStack::Cleanup { | |||
1083 | CallObjCArcUse(llvm::Value *object) : object(object) {} | |||
1084 | llvm::Value *object; | |||
1085 | ||||
1086 | void Emit(CodeGenFunction &CGF, Flags flags) override { | |||
1087 | CGF.EmitARCIntrinsicUse(object); | |||
1088 | } | |||
1089 | }; | |||
1090 | } | |||
1091 | ||||
1092 | Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E, | |||
1093 | BuiltinCheckKind Kind) { | |||
1094 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1095, __PRETTY_FUNCTION__)) | |||
1095 | && "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1095, __PRETTY_FUNCTION__)); | |||
1096 | ||||
1097 | Value *ArgValue = EmitScalarExpr(E); | |||
1098 | if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef()) | |||
1099 | return ArgValue; | |||
1100 | ||||
1101 | SanitizerScope SanScope(this); | |||
1102 | Value *Cond = Builder.CreateICmpNE( | |||
1103 | ArgValue, llvm::Constant::getNullValue(ArgValue->getType())); | |||
1104 | EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin), | |||
1105 | SanitizerHandler::InvalidBuiltin, | |||
1106 | {EmitCheckSourceLocation(E->getExprLoc()), | |||
1107 | llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)}, | |||
1108 | None); | |||
1109 | return ArgValue; | |||
1110 | } | |||
1111 | ||||
1112 | /// Get the argument type for arguments to os_log_helper. | |||
1113 | static CanQualType getOSLogArgType(ASTContext &C, int Size) { | |||
1114 | QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false); | |||
1115 | return C.getCanonicalType(UnsignedTy); | |||
1116 | } | |||
1117 | ||||
1118 | llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction( | |||
1119 | const analyze_os_log::OSLogBufferLayout &Layout, | |||
1120 | CharUnits BufferAlignment) { | |||
1121 | ASTContext &Ctx = getContext(); | |||
1122 | ||||
1123 | llvm::SmallString<64> Name; | |||
1124 | { | |||
1125 | raw_svector_ostream OS(Name); | |||
1126 | OS << "__os_log_helper"; | |||
1127 | OS << "_" << BufferAlignment.getQuantity(); | |||
1128 | OS << "_" << int(Layout.getSummaryByte()); | |||
1129 | OS << "_" << int(Layout.getNumArgsByte()); | |||
1130 | for (const auto &Item : Layout.Items) | |||
1131 | OS << "_" << int(Item.getSizeByte()) << "_" | |||
1132 | << int(Item.getDescriptorByte()); | |||
1133 | } | |||
1134 | ||||
1135 | if (llvm::Function *F = CGM.getModule().getFunction(Name)) | |||
1136 | return F; | |||
1137 | ||||
1138 | llvm::SmallVector<QualType, 4> ArgTys; | |||
1139 | FunctionArgList Args; | |||
1140 | Args.push_back(ImplicitParamDecl::Create( | |||
1141 | Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"), Ctx.VoidPtrTy, | |||
1142 | ImplicitParamDecl::Other)); | |||
1143 | ArgTys.emplace_back(Ctx.VoidPtrTy); | |||
1144 | ||||
1145 | for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) { | |||
1146 | char Size = Layout.Items[I].getSizeByte(); | |||
1147 | if (!Size) | |||
1148 | continue; | |||
1149 | ||||
1150 | QualType ArgTy = getOSLogArgType(Ctx, Size); | |||
1151 | Args.push_back(ImplicitParamDecl::Create( | |||
1152 | Ctx, nullptr, SourceLocation(), | |||
1153 | &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)), ArgTy, | |||
1154 | ImplicitParamDecl::Other)); | |||
1155 | ArgTys.emplace_back(ArgTy); | |||
1156 | } | |||
1157 | ||||
1158 | QualType ReturnTy = Ctx.VoidTy; | |||
1159 | QualType FuncionTy = Ctx.getFunctionType(ReturnTy, ArgTys, {}); | |||
1160 | ||||
1161 | // The helper function has linkonce_odr linkage to enable the linker to merge | |||
1162 | // identical functions. To ensure the merging always happens, 'noinline' is | |||
1163 | // attached to the function when compiling with -Oz. | |||
1164 | const CGFunctionInfo &FI = | |||
1165 | CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, Args); | |||
1166 | llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI); | |||
1167 | llvm::Function *Fn = llvm::Function::Create( | |||
1168 | FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule()); | |||
1169 | Fn->setVisibility(llvm::GlobalValue::HiddenVisibility); | |||
1170 | CGM.SetLLVMFunctionAttributes(GlobalDecl(), FI, Fn); | |||
1171 | CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn); | |||
1172 | Fn->setDoesNotThrow(); | |||
1173 | ||||
1174 | // Attach 'noinline' at -Oz. | |||
1175 | if (CGM.getCodeGenOpts().OptimizeSize == 2) | |||
1176 | Fn->addFnAttr(llvm::Attribute::NoInline); | |||
1177 | ||||
1178 | auto NL = ApplyDebugLocation::CreateEmpty(*this); | |||
1179 | IdentifierInfo *II = &Ctx.Idents.get(Name); | |||
1180 | FunctionDecl *FD = FunctionDecl::Create( | |||
1181 | Ctx, Ctx.getTranslationUnitDecl(), SourceLocation(), SourceLocation(), II, | |||
1182 | FuncionTy, nullptr, SC_PrivateExtern, false, false); | |||
1183 | ||||
1184 | StartFunction(FD, ReturnTy, Fn, FI, Args); | |||
1185 | ||||
1186 | // Create a scope with an artificial location for the body of this function. | |||
1187 | auto AL = ApplyDebugLocation::CreateArtificial(*this); | |||
1188 | ||||
1189 | CharUnits Offset; | |||
1190 | Address BufAddr(Builder.CreateLoad(GetAddrOfLocalVar(Args[0]), "buf"), | |||
1191 | BufferAlignment); | |||
1192 | Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()), | |||
1193 | Builder.CreateConstByteGEP(BufAddr, Offset++, "summary")); | |||
1194 | Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()), | |||
1195 | Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs")); | |||
1196 | ||||
1197 | unsigned I = 1; | |||
1198 | for (const auto &Item : Layout.Items) { | |||
1199 | Builder.CreateStore( | |||
1200 | Builder.getInt8(Item.getDescriptorByte()), | |||
1201 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor")); | |||
1202 | Builder.CreateStore( | |||
1203 | Builder.getInt8(Item.getSizeByte()), | |||
1204 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize")); | |||
1205 | ||||
1206 | CharUnits Size = Item.size(); | |||
1207 | if (!Size.getQuantity()) | |||
1208 | continue; | |||
1209 | ||||
1210 | Address Arg = GetAddrOfLocalVar(Args[I]); | |||
1211 | Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData"); | |||
1212 | Addr = Builder.CreateBitCast(Addr, Arg.getPointer()->getType(), | |||
1213 | "argDataCast"); | |||
1214 | Builder.CreateStore(Builder.CreateLoad(Arg), Addr); | |||
1215 | Offset += Size; | |||
1216 | ++I; | |||
1217 | } | |||
1218 | ||||
1219 | FinishFunction(); | |||
1220 | ||||
1221 | return Fn; | |||
1222 | } | |||
1223 | ||||
1224 | RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) { | |||
1225 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1226, __PRETTY_FUNCTION__)) | |||
1226 | "__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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1226, __PRETTY_FUNCTION__)); | |||
1227 | ASTContext &Ctx = getContext(); | |||
1228 | analyze_os_log::OSLogBufferLayout Layout; | |||
1229 | analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout); | |||
1230 | Address BufAddr = EmitPointerWithAlignment(E.getArg(0)); | |||
1231 | llvm::SmallVector<llvm::Value *, 4> RetainableOperands; | |||
1232 | ||||
1233 | // Ignore argument 1, the format string. It is not currently used. | |||
1234 | CallArgList Args; | |||
1235 | Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy); | |||
1236 | ||||
1237 | for (const auto &Item : Layout.Items) { | |||
1238 | int Size = Item.getSizeByte(); | |||
1239 | if (!Size) | |||
1240 | continue; | |||
1241 | ||||
1242 | llvm::Value *ArgVal; | |||
1243 | ||||
1244 | if (Item.getKind() == analyze_os_log::OSLogBufferItem::MaskKind) { | |||
1245 | uint64_t Val = 0; | |||
1246 | for (unsigned I = 0, E = Item.getMaskType().size(); I < E; ++I) | |||
1247 | Val |= ((uint64_t)Item.getMaskType()[I]) << I * 8; | |||
1248 | ArgVal = llvm::Constant::getIntegerValue(Int64Ty, llvm::APInt(64, Val)); | |||
1249 | } else if (const Expr *TheExpr = Item.getExpr()) { | |||
1250 | ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false); | |||
1251 | ||||
1252 | // Check if this is a retainable type. | |||
1253 | if (TheExpr->getType()->isObjCRetainableType()) { | |||
1254 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1255, __PRETTY_FUNCTION__)) | |||
1255 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1255, __PRETTY_FUNCTION__)); | |||
1256 | // Check if the object is constant, if not, save it in | |||
1257 | // RetainableOperands. | |||
1258 | if (!isa<Constant>(ArgVal)) | |||
1259 | RetainableOperands.push_back(ArgVal); | |||
1260 | } | |||
1261 | } else { | |||
1262 | ArgVal = Builder.getInt32(Item.getConstValue().getQuantity()); | |||
1263 | } | |||
1264 | ||||
1265 | unsigned ArgValSize = | |||
1266 | CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType()); | |||
1267 | llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(), | |||
1268 | ArgValSize); | |||
1269 | ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy); | |||
1270 | CanQualType ArgTy = getOSLogArgType(Ctx, Size); | |||
1271 | // If ArgVal has type x86_fp80, zero-extend ArgVal. | |||
1272 | ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy)); | |||
1273 | Args.add(RValue::get(ArgVal), ArgTy); | |||
1274 | } | |||
1275 | ||||
1276 | const CGFunctionInfo &FI = | |||
1277 | CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args); | |||
1278 | llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction( | |||
1279 | Layout, BufAddr.getAlignment()); | |||
1280 | EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args); | |||
1281 | ||||
1282 | // Push a clang.arc.use cleanup for each object in RetainableOperands. The | |||
1283 | // cleanup will cause the use to appear after the final log call, keeping | |||
1284 | // the object valid while it’s held in the log buffer. Note that if there’s | |||
1285 | // a release cleanup on the object, it will already be active; since | |||
1286 | // cleanups are emitted in reverse order, the use will occur before the | |||
1287 | // object is released. | |||
1288 | if (!RetainableOperands.empty() && getLangOpts().ObjCAutoRefCount && | |||
1289 | CGM.getCodeGenOpts().OptimizationLevel != 0) | |||
1290 | for (llvm::Value *Object : RetainableOperands) | |||
1291 | pushFullExprCleanup<CallObjCArcUse>(getARCCleanupKind(), Object); | |||
1292 | ||||
1293 | return RValue::get(BufAddr.getPointer()); | |||
1294 | } | |||
1295 | ||||
1296 | /// Determine if a binop is a checked mixed-sign multiply we can specialize. | |||
1297 | static bool isSpecialMixedSignMultiply(unsigned BuiltinID, | |||
1298 | WidthAndSignedness Op1Info, | |||
1299 | WidthAndSignedness Op2Info, | |||
1300 | WidthAndSignedness ResultInfo) { | |||
1301 | return BuiltinID == Builtin::BI__builtin_mul_overflow && | |||
1302 | std::max(Op1Info.Width, Op2Info.Width) >= ResultInfo.Width && | |||
1303 | Op1Info.Signed != Op2Info.Signed; | |||
1304 | } | |||
1305 | ||||
1306 | /// Emit a checked mixed-sign multiply. This is a cheaper specialization of | |||
1307 | /// the generic checked-binop irgen. | |||
1308 | static RValue | |||
1309 | EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1, | |||
1310 | WidthAndSignedness Op1Info, const clang::Expr *Op2, | |||
1311 | WidthAndSignedness Op2Info, | |||
1312 | const clang::Expr *ResultArg, QualType ResultQTy, | |||
1313 | WidthAndSignedness ResultInfo) { | |||
1314 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1316, __PRETTY_FUNCTION__)) | |||
1315 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1316, __PRETTY_FUNCTION__)) | |||
1316 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1316, __PRETTY_FUNCTION__)); | |||
1317 | ||||
1318 | // Emit the signed and unsigned operands. | |||
1319 | const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2; | |||
1320 | const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1; | |||
1321 | llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp); | |||
1322 | llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp); | |||
1323 | unsigned SignedOpWidth = Op1Info.Signed ? Op1Info.Width : Op2Info.Width; | |||
1324 | unsigned UnsignedOpWidth = Op1Info.Signed ? Op2Info.Width : Op1Info.Width; | |||
1325 | ||||
1326 | // One of the operands may be smaller than the other. If so, [s|z]ext it. | |||
1327 | if (SignedOpWidth < UnsignedOpWidth) | |||
1328 | Signed = CGF.Builder.CreateSExt(Signed, Unsigned->getType(), "op.sext"); | |||
1329 | if (UnsignedOpWidth < SignedOpWidth) | |||
1330 | Unsigned = CGF.Builder.CreateZExt(Unsigned, Signed->getType(), "op.zext"); | |||
1331 | ||||
1332 | llvm::Type *OpTy = Signed->getType(); | |||
1333 | llvm::Value *Zero = llvm::Constant::getNullValue(OpTy); | |||
1334 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); | |||
1335 | llvm::Type *ResTy = ResultPtr.getElementType(); | |||
1336 | unsigned OpWidth = std::max(Op1Info.Width, Op2Info.Width); | |||
1337 | ||||
1338 | // Take the absolute value of the signed operand. | |||
1339 | llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero); | |||
1340 | llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed); | |||
1341 | llvm::Value *AbsSigned = | |||
1342 | CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed); | |||
1343 | ||||
1344 | // Perform a checked unsigned multiplication. | |||
1345 | llvm::Value *UnsignedOverflow; | |||
1346 | llvm::Value *UnsignedResult = | |||
1347 | EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned, | |||
1348 | Unsigned, UnsignedOverflow); | |||
1349 | ||||
1350 | llvm::Value *Overflow, *Result; | |||
1351 | if (ResultInfo.Signed) { | |||
1352 | // Signed overflow occurs if the result is greater than INT_MAX or lesser | |||
1353 | // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative). | |||
1354 | auto IntMax = | |||
1355 | llvm::APInt::getSignedMaxValue(ResultInfo.Width).zextOrSelf(OpWidth); | |||
1356 | llvm::Value *MaxResult = | |||
1357 | CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax), | |||
1358 | CGF.Builder.CreateZExt(IsNegative, OpTy)); | |||
1359 | llvm::Value *SignedOverflow = | |||
1360 | CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult); | |||
1361 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow); | |||
1362 | ||||
1363 | // Prepare the signed result (possibly by negating it). | |||
1364 | llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult); | |||
1365 | llvm::Value *SignedResult = | |||
1366 | CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult); | |||
1367 | Result = CGF.Builder.CreateTrunc(SignedResult, ResTy); | |||
1368 | } else { | |||
1369 | // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX. | |||
1370 | llvm::Value *Underflow = CGF.Builder.CreateAnd( | |||
1371 | IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult)); | |||
1372 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow); | |||
1373 | if (ResultInfo.Width < OpWidth) { | |||
1374 | auto IntMax = | |||
1375 | llvm::APInt::getMaxValue(ResultInfo.Width).zext(OpWidth); | |||
1376 | llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT( | |||
1377 | UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax)); | |||
1378 | Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow); | |||
1379 | } | |||
1380 | ||||
1381 | // Negate the product if it would be negative in infinite precision. | |||
1382 | Result = CGF.Builder.CreateSelect( | |||
1383 | IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult); | |||
1384 | ||||
1385 | Result = CGF.Builder.CreateTrunc(Result, ResTy); | |||
1386 | } | |||
1387 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1387, __PRETTY_FUNCTION__)); | |||
1388 | ||||
1389 | bool isVolatile = | |||
1390 | ResultArg->getType()->getPointeeType().isVolatileQualified(); | |||
1391 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, | |||
1392 | isVolatile); | |||
1393 | return RValue::get(Overflow); | |||
1394 | } | |||
1395 | ||||
1396 | static llvm::Value *dumpRecord(CodeGenFunction &CGF, QualType RType, | |||
1397 | Value *&RecordPtr, CharUnits Align, | |||
1398 | llvm::FunctionCallee Func, int Lvl) { | |||
1399 | const auto *RT = RType->getAs<RecordType>(); | |||
| ||||
1400 | ASTContext &Context = CGF.getContext(); | |||
1401 | RecordDecl *RD = RT->getDecl()->getDefinition(); | |||
| ||||
1402 | std::string Pad = std::string(Lvl * 4, ' '); | |||
1403 | ||||
1404 | Value *GString = | |||
1405 | CGF.Builder.CreateGlobalStringPtr(RType.getAsString() + " {\n"); | |||
1406 | Value *Res = CGF.Builder.CreateCall(Func, {GString}); | |||
1407 | ||||
1408 | static llvm::DenseMap<QualType, const char *> Types; | |||
1409 | if (Types.empty()) { | |||
1410 | Types[Context.CharTy] = "%c"; | |||
1411 | Types[Context.BoolTy] = "%d"; | |||
1412 | Types[Context.SignedCharTy] = "%hhd"; | |||
1413 | Types[Context.UnsignedCharTy] = "%hhu"; | |||
1414 | Types[Context.IntTy] = "%d"; | |||
1415 | Types[Context.UnsignedIntTy] = "%u"; | |||
1416 | Types[Context.LongTy] = "%ld"; | |||
1417 | Types[Context.UnsignedLongTy] = "%lu"; | |||
1418 | Types[Context.LongLongTy] = "%lld"; | |||
1419 | Types[Context.UnsignedLongLongTy] = "%llu"; | |||
1420 | Types[Context.ShortTy] = "%hd"; | |||
1421 | Types[Context.UnsignedShortTy] = "%hu"; | |||
1422 | Types[Context.VoidPtrTy] = "%p"; | |||
1423 | Types[Context.FloatTy] = "%f"; | |||
1424 | Types[Context.DoubleTy] = "%f"; | |||
1425 | Types[Context.LongDoubleTy] = "%Lf"; | |||
1426 | Types[Context.getPointerType(Context.CharTy)] = "%s"; | |||
1427 | Types[Context.getPointerType(Context.getConstType(Context.CharTy))] = "%s"; | |||
1428 | } | |||
1429 | ||||
1430 | for (const auto *FD : RD->fields()) { | |||
1431 | Value *FieldPtr = RecordPtr; | |||
1432 | if (RD->isUnion()) | |||
1433 | FieldPtr = CGF.Builder.CreatePointerCast( | |||
1434 | FieldPtr, CGF.ConvertType(Context.getPointerType(FD->getType()))); | |||
1435 | else | |||
1436 | FieldPtr = CGF.Builder.CreateStructGEP(CGF.ConvertType(RType), FieldPtr, | |||
1437 | FD->getFieldIndex()); | |||
1438 | ||||
1439 | GString = CGF.Builder.CreateGlobalStringPtr( | |||
1440 | llvm::Twine(Pad) | |||
1441 | .concat(FD->getType().getAsString()) | |||
1442 | .concat(llvm::Twine(' ')) | |||
1443 | .concat(FD->getNameAsString()) | |||
1444 | .concat(" : ") | |||
1445 | .str()); | |||
1446 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); | |||
1447 | Res = CGF.Builder.CreateAdd(Res, TmpRes); | |||
1448 | ||||
1449 | QualType CanonicalType = | |||
1450 | FD->getType().getUnqualifiedType().getCanonicalType(); | |||
1451 | ||||
1452 | // We check whether we are in a recursive type | |||
1453 | if (CanonicalType->isRecordType()) { | |||
1454 | Value *TmpRes = | |||
1455 | dumpRecord(CGF, CanonicalType, FieldPtr, Align, Func, Lvl + 1); | |||
1456 | Res = CGF.Builder.CreateAdd(TmpRes, Res); | |||
1457 | continue; | |||
1458 | } | |||
1459 | ||||
1460 | // We try to determine the best format to print the current field | |||
1461 | llvm::Twine Format = Types.find(CanonicalType) == Types.end() | |||
1462 | ? Types[Context.VoidPtrTy] | |||
1463 | : Types[CanonicalType]; | |||
1464 | ||||
1465 | Address FieldAddress = Address(FieldPtr, Align); | |||
1466 | FieldPtr = CGF.Builder.CreateLoad(FieldAddress); | |||
1467 | ||||
1468 | // FIXME Need to handle bitfield here | |||
1469 | GString = CGF.Builder.CreateGlobalStringPtr( | |||
1470 | Format.concat(llvm::Twine('\n')).str()); | |||
1471 | TmpRes = CGF.Builder.CreateCall(Func, {GString, FieldPtr}); | |||
1472 | Res = CGF.Builder.CreateAdd(Res, TmpRes); | |||
1473 | } | |||
1474 | ||||
1475 | GString = CGF.Builder.CreateGlobalStringPtr(Pad + "}\n"); | |||
1476 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); | |||
1477 | Res = CGF.Builder.CreateAdd(Res, TmpRes); | |||
1478 | return Res; | |||
1479 | } | |||
1480 | ||||
1481 | static bool | |||
1482 | TypeRequiresBuiltinLaunderImp(const ASTContext &Ctx, QualType Ty, | |||
1483 | llvm::SmallPtrSetImpl<const Decl *> &Seen) { | |||
1484 | if (const auto *Arr = Ctx.getAsArrayType(Ty)) | |||
1485 | Ty = Ctx.getBaseElementType(Arr); | |||
1486 | ||||
1487 | const auto *Record = Ty->getAsCXXRecordDecl(); | |||
1488 | if (!Record) | |||
1489 | return false; | |||
1490 | ||||
1491 | // We've already checked this type, or are in the process of checking it. | |||
1492 | if (!Seen.insert(Record).second) | |||
1493 | return false; | |||
1494 | ||||
1495 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1496, __PRETTY_FUNCTION__)) | |||
1496 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1496, __PRETTY_FUNCTION__)); | |||
1497 | ||||
1498 | if (Record->isDynamicClass()) | |||
1499 | return true; | |||
1500 | ||||
1501 | for (FieldDecl *F : Record->fields()) { | |||
1502 | if (TypeRequiresBuiltinLaunderImp(Ctx, F->getType(), Seen)) | |||
1503 | return true; | |||
1504 | } | |||
1505 | return false; | |||
1506 | } | |||
1507 | ||||
1508 | /// Determine if the specified type requires laundering by checking if it is a | |||
1509 | /// dynamic class type or contains a subobject which is a dynamic class type. | |||
1510 | static bool TypeRequiresBuiltinLaunder(CodeGenModule &CGM, QualType Ty) { | |||
1511 | if (!CGM.getCodeGenOpts().StrictVTablePointers) | |||
1512 | return false; | |||
1513 | llvm::SmallPtrSet<const Decl *, 16> Seen; | |||
1514 | return TypeRequiresBuiltinLaunderImp(CGM.getContext(), Ty, Seen); | |||
1515 | } | |||
1516 | ||||
1517 | RValue CodeGenFunction::emitRotate(const CallExpr *E, bool IsRotateRight) { | |||
1518 | llvm::Value *Src = EmitScalarExpr(E->getArg(0)); | |||
1519 | llvm::Value *ShiftAmt = EmitScalarExpr(E->getArg(1)); | |||
1520 | ||||
1521 | // The builtin's shift arg may have a different type than the source arg and | |||
1522 | // result, but the LLVM intrinsic uses the same type for all values. | |||
1523 | llvm::Type *Ty = Src->getType(); | |||
1524 | ShiftAmt = Builder.CreateIntCast(ShiftAmt, Ty, false); | |||
1525 | ||||
1526 | // Rotate is a special case of LLVM funnel shift - 1st 2 args are the same. | |||
1527 | unsigned IID = IsRotateRight ? Intrinsic::fshr : Intrinsic::fshl; | |||
1528 | Function *F = CGM.getIntrinsic(IID, Ty); | |||
1529 | return RValue::get(Builder.CreateCall(F, { Src, Src, ShiftAmt })); | |||
1530 | } | |||
1531 | ||||
1532 | RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, | |||
1533 | const CallExpr *E, | |||
1534 | ReturnValueSlot ReturnValue) { | |||
1535 | const FunctionDecl *FD = GD.getDecl()->getAsFunction(); | |||
1536 | // See if we can constant fold this builtin. If so, don't emit it at all. | |||
1537 | Expr::EvalResult Result; | |||
1538 | if (E->EvaluateAsRValue(Result, CGM.getContext()) && | |||
1539 | !Result.hasSideEffects()) { | |||
1540 | if (Result.Val.isInt()) | |||
1541 | return RValue::get(llvm::ConstantInt::get(getLLVMContext(), | |||
1542 | Result.Val.getInt())); | |||
1543 | if (Result.Val.isFloat()) | |||
1544 | return RValue::get(llvm::ConstantFP::get(getLLVMContext(), | |||
1545 | Result.Val.getFloat())); | |||
1546 | } | |||
1547 | ||||
1548 | // There are LLVM math intrinsics/instructions corresponding to math library | |||
1549 | // functions except the LLVM op will never set errno while the math library | |||
1550 | // might. Also, math builtins have the same semantics as their math library | |||
1551 | // twins. Thus, we can transform math library and builtin calls to their | |||
1552 | // LLVM counterparts if the call is marked 'const' (known to never set errno). | |||
1553 | if (FD->hasAttr<ConstAttr>()) { | |||
1554 | switch (BuiltinID) { | |||
1555 | case Builtin::BIceil: | |||
1556 | case Builtin::BIceilf: | |||
1557 | case Builtin::BIceill: | |||
1558 | case Builtin::BI__builtin_ceil: | |||
1559 | case Builtin::BI__builtin_ceilf: | |||
1560 | case Builtin::BI__builtin_ceilf16: | |||
1561 | case Builtin::BI__builtin_ceill: | |||
1562 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::ceil)); | |||
1563 | ||||
1564 | case Builtin::BIcopysign: | |||
1565 | case Builtin::BIcopysignf: | |||
1566 | case Builtin::BIcopysignl: | |||
1567 | case Builtin::BI__builtin_copysign: | |||
1568 | case Builtin::BI__builtin_copysignf: | |||
1569 | case Builtin::BI__builtin_copysignf16: | |||
1570 | case Builtin::BI__builtin_copysignl: | |||
1571 | case Builtin::BI__builtin_copysignf128: | |||
1572 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign)); | |||
1573 | ||||
1574 | case Builtin::BIcos: | |||
1575 | case Builtin::BIcosf: | |||
1576 | case Builtin::BIcosl: | |||
1577 | case Builtin::BI__builtin_cos: | |||
1578 | case Builtin::BI__builtin_cosf: | |||
1579 | case Builtin::BI__builtin_cosf16: | |||
1580 | case Builtin::BI__builtin_cosl: | |||
1581 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::cos)); | |||
1582 | ||||
1583 | case Builtin::BIexp: | |||
1584 | case Builtin::BIexpf: | |||
1585 | case Builtin::BIexpl: | |||
1586 | case Builtin::BI__builtin_exp: | |||
1587 | case Builtin::BI__builtin_expf: | |||
1588 | case Builtin::BI__builtin_expf16: | |||
1589 | case Builtin::BI__builtin_expl: | |||
1590 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::exp)); | |||
1591 | ||||
1592 | case Builtin::BIexp2: | |||
1593 | case Builtin::BIexp2f: | |||
1594 | case Builtin::BIexp2l: | |||
1595 | case Builtin::BI__builtin_exp2: | |||
1596 | case Builtin::BI__builtin_exp2f: | |||
1597 | case Builtin::BI__builtin_exp2f16: | |||
1598 | case Builtin::BI__builtin_exp2l: | |||
1599 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::exp2)); | |||
1600 | ||||
1601 | case Builtin::BIfabs: | |||
1602 | case Builtin::BIfabsf: | |||
1603 | case Builtin::BIfabsl: | |||
1604 | case Builtin::BI__builtin_fabs: | |||
1605 | case Builtin::BI__builtin_fabsf: | |||
1606 | case Builtin::BI__builtin_fabsf16: | |||
1607 | case Builtin::BI__builtin_fabsl: | |||
1608 | case Builtin::BI__builtin_fabsf128: | |||
1609 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs)); | |||
1610 | ||||
1611 | case Builtin::BIfloor: | |||
1612 | case Builtin::BIfloorf: | |||
1613 | case Builtin::BIfloorl: | |||
1614 | case Builtin::BI__builtin_floor: | |||
1615 | case Builtin::BI__builtin_floorf: | |||
1616 | case Builtin::BI__builtin_floorf16: | |||
1617 | case Builtin::BI__builtin_floorl: | |||
1618 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::floor)); | |||
1619 | ||||
1620 | case Builtin::BIfma: | |||
1621 | case Builtin::BIfmaf: | |||
1622 | case Builtin::BIfmal: | |||
1623 | case Builtin::BI__builtin_fma: | |||
1624 | case Builtin::BI__builtin_fmaf: | |||
1625 | case Builtin::BI__builtin_fmaf16: | |||
1626 | case Builtin::BI__builtin_fmal: | |||
1627 | return RValue::get(emitTernaryBuiltin(*this, E, Intrinsic::fma)); | |||
1628 | ||||
1629 | case Builtin::BIfmax: | |||
1630 | case Builtin::BIfmaxf: | |||
1631 | case Builtin::BIfmaxl: | |||
1632 | case Builtin::BI__builtin_fmax: | |||
1633 | case Builtin::BI__builtin_fmaxf: | |||
1634 | case Builtin::BI__builtin_fmaxf16: | |||
1635 | case Builtin::BI__builtin_fmaxl: | |||
1636 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::maxnum)); | |||
1637 | ||||
1638 | case Builtin::BIfmin: | |||
1639 | case Builtin::BIfminf: | |||
1640 | case Builtin::BIfminl: | |||
1641 | case Builtin::BI__builtin_fmin: | |||
1642 | case Builtin::BI__builtin_fminf: | |||
1643 | case Builtin::BI__builtin_fminf16: | |||
1644 | case Builtin::BI__builtin_fminl: | |||
1645 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::minnum)); | |||
1646 | ||||
1647 | // fmod() is a special-case. It maps to the frem instruction rather than an | |||
1648 | // LLVM intrinsic. | |||
1649 | case Builtin::BIfmod: | |||
1650 | case Builtin::BIfmodf: | |||
1651 | case Builtin::BIfmodl: | |||
1652 | case Builtin::BI__builtin_fmod: | |||
1653 | case Builtin::BI__builtin_fmodf: | |||
1654 | case Builtin::BI__builtin_fmodf16: | |||
1655 | case Builtin::BI__builtin_fmodl: { | |||
1656 | Value *Arg1 = EmitScalarExpr(E->getArg(0)); | |||
1657 | Value *Arg2 = EmitScalarExpr(E->getArg(1)); | |||
1658 | return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod")); | |||
1659 | } | |||
1660 | ||||
1661 | case Builtin::BIlog: | |||
1662 | case Builtin::BIlogf: | |||
1663 | case Builtin::BIlogl: | |||
1664 | case Builtin::BI__builtin_log: | |||
1665 | case Builtin::BI__builtin_logf: | |||
1666 | case Builtin::BI__builtin_logf16: | |||
1667 | case Builtin::BI__builtin_logl: | |||
1668 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log)); | |||
1669 | ||||
1670 | case Builtin::BIlog10: | |||
1671 | case Builtin::BIlog10f: | |||
1672 | case Builtin::BIlog10l: | |||
1673 | case Builtin::BI__builtin_log10: | |||
1674 | case Builtin::BI__builtin_log10f: | |||
1675 | case Builtin::BI__builtin_log10f16: | |||
1676 | case Builtin::BI__builtin_log10l: | |||
1677 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log10)); | |||
1678 | ||||
1679 | case Builtin::BIlog2: | |||
1680 | case Builtin::BIlog2f: | |||
1681 | case Builtin::BIlog2l: | |||
1682 | case Builtin::BI__builtin_log2: | |||
1683 | case Builtin::BI__builtin_log2f: | |||
1684 | case Builtin::BI__builtin_log2f16: | |||
1685 | case Builtin::BI__builtin_log2l: | |||
1686 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log2)); | |||
1687 | ||||
1688 | case Builtin::BInearbyint: | |||
1689 | case Builtin::BInearbyintf: | |||
1690 | case Builtin::BInearbyintl: | |||
1691 | case Builtin::BI__builtin_nearbyint: | |||
1692 | case Builtin::BI__builtin_nearbyintf: | |||
1693 | case Builtin::BI__builtin_nearbyintl: | |||
1694 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::nearbyint)); | |||
1695 | ||||
1696 | case Builtin::BIpow: | |||
1697 | case Builtin::BIpowf: | |||
1698 | case Builtin::BIpowl: | |||
1699 | case Builtin::BI__builtin_pow: | |||
1700 | case Builtin::BI__builtin_powf: | |||
1701 | case Builtin::BI__builtin_powf16: | |||
1702 | case Builtin::BI__builtin_powl: | |||
1703 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::pow)); | |||
1704 | ||||
1705 | case Builtin::BIrint: | |||
1706 | case Builtin::BIrintf: | |||
1707 | case Builtin::BIrintl: | |||
1708 | case Builtin::BI__builtin_rint: | |||
1709 | case Builtin::BI__builtin_rintf: | |||
1710 | case Builtin::BI__builtin_rintf16: | |||
1711 | case Builtin::BI__builtin_rintl: | |||
1712 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::rint)); | |||
1713 | ||||
1714 | case Builtin::BIround: | |||
1715 | case Builtin::BIroundf: | |||
1716 | case Builtin::BIroundl: | |||
1717 | case Builtin::BI__builtin_round: | |||
1718 | case Builtin::BI__builtin_roundf: | |||
1719 | case Builtin::BI__builtin_roundf16: | |||
1720 | case Builtin::BI__builtin_roundl: | |||
1721 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::round)); | |||
1722 | ||||
1723 | case Builtin::BIsin: | |||
1724 | case Builtin::BIsinf: | |||
1725 | case Builtin::BIsinl: | |||
1726 | case Builtin::BI__builtin_sin: | |||
1727 | case Builtin::BI__builtin_sinf: | |||
1728 | case Builtin::BI__builtin_sinf16: | |||
1729 | case Builtin::BI__builtin_sinl: | |||
1730 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::sin)); | |||
1731 | ||||
1732 | case Builtin::BIsqrt: | |||
1733 | case Builtin::BIsqrtf: | |||
1734 | case Builtin::BIsqrtl: | |||
1735 | case Builtin::BI__builtin_sqrt: | |||
1736 | case Builtin::BI__builtin_sqrtf: | |||
1737 | case Builtin::BI__builtin_sqrtf16: | |||
1738 | case Builtin::BI__builtin_sqrtl: | |||
1739 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::sqrt)); | |||
1740 | ||||
1741 | case Builtin::BItrunc: | |||
1742 | case Builtin::BItruncf: | |||
1743 | case Builtin::BItruncl: | |||
1744 | case Builtin::BI__builtin_trunc: | |||
1745 | case Builtin::BI__builtin_truncf: | |||
1746 | case Builtin::BI__builtin_truncf16: | |||
1747 | case Builtin::BI__builtin_truncl: | |||
1748 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::trunc)); | |||
1749 | ||||
1750 | case Builtin::BIlround: | |||
1751 | case Builtin::BIlroundf: | |||
1752 | case Builtin::BIlroundl: | |||
1753 | case Builtin::BI__builtin_lround: | |||
1754 | case Builtin::BI__builtin_lroundf: | |||
1755 | case Builtin::BI__builtin_lroundl: | |||
1756 | return RValue::get(emitFPToIntRoundBuiltin(*this, E, Intrinsic::lround)); | |||
1757 | ||||
1758 | case Builtin::BIllround: | |||
1759 | case Builtin::BIllroundf: | |||
1760 | case Builtin::BIllroundl: | |||
1761 | case Builtin::BI__builtin_llround: | |||
1762 | case Builtin::BI__builtin_llroundf: | |||
1763 | case Builtin::BI__builtin_llroundl: | |||
1764 | return RValue::get(emitFPToIntRoundBuiltin(*this, E, Intrinsic::llround)); | |||
1765 | ||||
1766 | case Builtin::BIlrint: | |||
1767 | case Builtin::BIlrintf: | |||
1768 | case Builtin::BIlrintl: | |||
1769 | case Builtin::BI__builtin_lrint: | |||
1770 | case Builtin::BI__builtin_lrintf: | |||
1771 | case Builtin::BI__builtin_lrintl: | |||
1772 | return RValue::get(emitFPToIntRoundBuiltin(*this, E, Intrinsic::lrint)); | |||
1773 | ||||
1774 | case Builtin::BIllrint: | |||
1775 | case Builtin::BIllrintf: | |||
1776 | case Builtin::BIllrintl: | |||
1777 | case Builtin::BI__builtin_llrint: | |||
1778 | case Builtin::BI__builtin_llrintf: | |||
1779 | case Builtin::BI__builtin_llrintl: | |||
1780 | return RValue::get(emitFPToIntRoundBuiltin(*this, E, Intrinsic::llrint)); | |||
1781 | ||||
1782 | default: | |||
1783 | break; | |||
1784 | } | |||
1785 | } | |||
1786 | ||||
1787 | switch (BuiltinID) { | |||
1788 | default: break; | |||
1789 | case Builtin::BI__builtin___CFStringMakeConstantString: | |||
1790 | case Builtin::BI__builtin___NSStringMakeConstantString: | |||
1791 | return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType())); | |||
1792 | case Builtin::BI__builtin_stdarg_start: | |||
1793 | case Builtin::BI__builtin_va_start: | |||
1794 | case Builtin::BI__va_start: | |||
1795 | case Builtin::BI__builtin_va_end: | |||
1796 | return RValue::get( | |||
1797 | EmitVAStartEnd(BuiltinID == Builtin::BI__va_start | |||
1798 | ? EmitScalarExpr(E->getArg(0)) | |||
1799 | : EmitVAListRef(E->getArg(0)).getPointer(), | |||
1800 | BuiltinID != Builtin::BI__builtin_va_end)); | |||
1801 | case Builtin::BI__builtin_va_copy: { | |||
1802 | Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer(); | |||
1803 | Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer(); | |||
1804 | ||||
1805 | llvm::Type *Type = Int8PtrTy; | |||
1806 | ||||
1807 | DstPtr = Builder.CreateBitCast(DstPtr, Type); | |||
1808 | SrcPtr = Builder.CreateBitCast(SrcPtr, Type); | |||
1809 | return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy), | |||
1810 | {DstPtr, SrcPtr})); | |||
1811 | } | |||
1812 | case Builtin::BI__builtin_abs: | |||
1813 | case Builtin::BI__builtin_labs: | |||
1814 | case Builtin::BI__builtin_llabs: { | |||
1815 | // X < 0 ? -X : X | |||
1816 | // The negation has 'nsw' because abs of INT_MIN is undefined. | |||
1817 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
1818 | Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg"); | |||
1819 | Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType()); | |||
1820 | Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond"); | |||
1821 | Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs"); | |||
1822 | return RValue::get(Result); | |||
1823 | } | |||
1824 | case Builtin::BI__builtin_conj: | |||
1825 | case Builtin::BI__builtin_conjf: | |||
1826 | case Builtin::BI__builtin_conjl: { | |||
1827 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); | |||
1828 | Value *Real = ComplexVal.first; | |||
1829 | Value *Imag = ComplexVal.second; | |||
1830 | Value *Zero = | |||
1831 | Imag->getType()->isFPOrFPVectorTy() | |||
1832 | ? llvm::ConstantFP::getZeroValueForNegation(Imag->getType()) | |||
1833 | : llvm::Constant::getNullValue(Imag->getType()); | |||
1834 | ||||
1835 | Imag = Builder.CreateFSub(Zero, Imag, "sub"); | |||
1836 | return RValue::getComplex(std::make_pair(Real, Imag)); | |||
1837 | } | |||
1838 | case Builtin::BI__builtin_creal: | |||
1839 | case Builtin::BI__builtin_crealf: | |||
1840 | case Builtin::BI__builtin_creall: | |||
1841 | case Builtin::BIcreal: | |||
1842 | case Builtin::BIcrealf: | |||
1843 | case Builtin::BIcreall: { | |||
1844 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); | |||
1845 | return RValue::get(ComplexVal.first); | |||
1846 | } | |||
1847 | ||||
1848 | case Builtin::BI__builtin_dump_struct: { | |||
1849 | llvm::Type *LLVMIntTy = getTypes().ConvertType(getContext().IntTy); | |||
1850 | llvm::FunctionType *LLVMFuncType = llvm::FunctionType::get( | |||
1851 | LLVMIntTy, {llvm::Type::getInt8PtrTy(getLLVMContext())}, true); | |||
1852 | ||||
1853 | Value *Func = EmitScalarExpr(E->getArg(1)->IgnoreImpCasts()); | |||
1854 | CharUnits Arg0Align = EmitPointerWithAlignment(E->getArg(0)).getAlignment(); | |||
1855 | ||||
1856 | const Expr *Arg0 = E->getArg(0)->IgnoreImpCasts(); | |||
1857 | QualType Arg0Type = Arg0->getType()->getPointeeType(); | |||
1858 | ||||
1859 | Value *RecordPtr = EmitScalarExpr(Arg0); | |||
1860 | Value *Res = dumpRecord(*this, Arg0Type, RecordPtr, Arg0Align, | |||
1861 | {LLVMFuncType, Func}, 0); | |||
1862 | return RValue::get(Res); | |||
1863 | } | |||
1864 | ||||
1865 | case Builtin::BI__builtin_preserve_access_index: { | |||
1866 | // Only enabled preserved access index region when debuginfo | |||
1867 | // is available as debuginfo is needed to preserve user-level | |||
1868 | // access pattern. | |||
1869 | if (!getDebugInfo()) { | |||
1870 | CGM.Error(E->getExprLoc(), "using builtin_preserve_access_index() without -g"); | |||
1871 | return RValue::get(EmitScalarExpr(E->getArg(0))); | |||
1872 | } | |||
1873 | ||||
1874 | // Nested builtin_preserve_access_index() not supported | |||
1875 | if (IsInPreservedAIRegion) { | |||
1876 | CGM.Error(E->getExprLoc(), "nested builtin_preserve_access_index() not supported"); | |||
1877 | return RValue::get(EmitScalarExpr(E->getArg(0))); | |||
1878 | } | |||
1879 | ||||
1880 | IsInPreservedAIRegion = true; | |||
1881 | Value *Res = EmitScalarExpr(E->getArg(0)); | |||
1882 | IsInPreservedAIRegion = false; | |||
1883 | return RValue::get(Res); | |||
1884 | } | |||
1885 | ||||
1886 | case Builtin::BI__builtin_cimag: | |||
1887 | case Builtin::BI__builtin_cimagf: | |||
1888 | case Builtin::BI__builtin_cimagl: | |||
1889 | case Builtin::BIcimag: | |||
1890 | case Builtin::BIcimagf: | |||
1891 | case Builtin::BIcimagl: { | |||
1892 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); | |||
1893 | return RValue::get(ComplexVal.second); | |||
1894 | } | |||
1895 | ||||
1896 | case Builtin::BI__builtin_clrsb: | |||
1897 | case Builtin::BI__builtin_clrsbl: | |||
1898 | case Builtin::BI__builtin_clrsbll: { | |||
1899 | // clrsb(x) -> clz(x < 0 ? ~x : x) - 1 or | |||
1900 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
1901 | ||||
1902 | llvm::Type *ArgType = ArgValue->getType(); | |||
1903 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); | |||
1904 | ||||
1905 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
1906 | Value *Zero = llvm::Constant::getNullValue(ArgType); | |||
1907 | Value *IsNeg = Builder.CreateICmpSLT(ArgValue, Zero, "isneg"); | |||
1908 | Value *Inverse = Builder.CreateNot(ArgValue, "not"); | |||
1909 | Value *Tmp = Builder.CreateSelect(IsNeg, Inverse, ArgValue); | |||
1910 | Value *Ctlz = Builder.CreateCall(F, {Tmp, Builder.getFalse()}); | |||
1911 | Value *Result = Builder.CreateSub(Ctlz, llvm::ConstantInt::get(ArgType, 1)); | |||
1912 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
1913 | "cast"); | |||
1914 | return RValue::get(Result); | |||
1915 | } | |||
1916 | case Builtin::BI__builtin_ctzs: | |||
1917 | case Builtin::BI__builtin_ctz: | |||
1918 | case Builtin::BI__builtin_ctzl: | |||
1919 | case Builtin::BI__builtin_ctzll: { | |||
1920 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero); | |||
1921 | ||||
1922 | llvm::Type *ArgType = ArgValue->getType(); | |||
1923 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); | |||
1924 | ||||
1925 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
1926 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); | |||
1927 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); | |||
1928 | if (Result->getType() != ResultType) | |||
1929 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
1930 | "cast"); | |||
1931 | return RValue::get(Result); | |||
1932 | } | |||
1933 | case Builtin::BI__builtin_clzs: | |||
1934 | case Builtin::BI__builtin_clz: | |||
1935 | case Builtin::BI__builtin_clzl: | |||
1936 | case Builtin::BI__builtin_clzll: { | |||
1937 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero); | |||
1938 | ||||
1939 | llvm::Type *ArgType = ArgValue->getType(); | |||
1940 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); | |||
1941 | ||||
1942 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
1943 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); | |||
1944 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); | |||
1945 | if (Result->getType() != ResultType) | |||
1946 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
1947 | "cast"); | |||
1948 | return RValue::get(Result); | |||
1949 | } | |||
1950 | case Builtin::BI__builtin_ffs: | |||
1951 | case Builtin::BI__builtin_ffsl: | |||
1952 | case Builtin::BI__builtin_ffsll: { | |||
1953 | // ffs(x) -> x ? cttz(x) + 1 : 0 | |||
1954 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
1955 | ||||
1956 | llvm::Type *ArgType = ArgValue->getType(); | |||
1957 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); | |||
1958 | ||||
1959 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
1960 | Value *Tmp = | |||
1961 | Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}), | |||
1962 | llvm::ConstantInt::get(ArgType, 1)); | |||
1963 | Value *Zero = llvm::Constant::getNullValue(ArgType); | |||
1964 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero"); | |||
1965 | Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs"); | |||
1966 | if (Result->getType() != ResultType) | |||
1967 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
1968 | "cast"); | |||
1969 | return RValue::get(Result); | |||
1970 | } | |||
1971 | case Builtin::BI__builtin_parity: | |||
1972 | case Builtin::BI__builtin_parityl: | |||
1973 | case Builtin::BI__builtin_parityll: { | |||
1974 | // parity(x) -> ctpop(x) & 1 | |||
1975 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
1976 | ||||
1977 | llvm::Type *ArgType = ArgValue->getType(); | |||
1978 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); | |||
1979 | ||||
1980 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
1981 | Value *Tmp = Builder.CreateCall(F, ArgValue); | |||
1982 | Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1)); | |||
1983 | if (Result->getType() != ResultType) | |||
1984 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
1985 | "cast"); | |||
1986 | return RValue::get(Result); | |||
1987 | } | |||
1988 | case Builtin::BI__lzcnt16: | |||
1989 | case Builtin::BI__lzcnt: | |||
1990 | case Builtin::BI__lzcnt64: { | |||
1991 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
1992 | ||||
1993 | llvm::Type *ArgType = ArgValue->getType(); | |||
1994 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); | |||
1995 | ||||
1996 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
1997 | Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()}); | |||
1998 | if (Result->getType() != ResultType) | |||
1999 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
2000 | "cast"); | |||
2001 | return RValue::get(Result); | |||
2002 | } | |||
2003 | case Builtin::BI__popcnt16: | |||
2004 | case Builtin::BI__popcnt: | |||
2005 | case Builtin::BI__popcnt64: | |||
2006 | case Builtin::BI__builtin_popcount: | |||
2007 | case Builtin::BI__builtin_popcountl: | |||
2008 | case Builtin::BI__builtin_popcountll: { | |||
2009 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
2010 | ||||
2011 | llvm::Type *ArgType = ArgValue->getType(); | |||
2012 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); | |||
2013 | ||||
2014 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
2015 | Value *Result = Builder.CreateCall(F, ArgValue); | |||
2016 | if (Result->getType() != ResultType) | |||
2017 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
2018 | "cast"); | |||
2019 | return RValue::get(Result); | |||
2020 | } | |||
2021 | case Builtin::BI__builtin_unpredictable: { | |||
2022 | // Always return the argument of __builtin_unpredictable. LLVM does not | |||
2023 | // handle this builtin. Metadata for this builtin should be added directly | |||
2024 | // to instructions such as branches or switches that use it. | |||
2025 | return RValue::get(EmitScalarExpr(E->getArg(0))); | |||
2026 | } | |||
2027 | case Builtin::BI__builtin_expect: { | |||
2028 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
2029 | llvm::Type *ArgType = ArgValue->getType(); | |||
2030 | ||||
2031 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); | |||
2032 | // Don't generate llvm.expect on -O0 as the backend won't use it for | |||
2033 | // anything. | |||
2034 | // Note, we still IRGen ExpectedValue because it could have side-effects. | |||
2035 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) | |||
2036 | return RValue::get(ArgValue); | |||
2037 | ||||
2038 | Function *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType); | |||
2039 | Value *Result = | |||
2040 | Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval"); | |||
2041 | return RValue::get(Result); | |||
2042 | } | |||
2043 | case Builtin::BI__builtin_assume_aligned: { | |||
2044 | const Expr *Ptr = E->getArg(0); | |||
2045 | Value *PtrValue = EmitScalarExpr(Ptr); | |||
2046 | Value *OffsetValue = | |||
2047 | (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr; | |||
2048 | ||||
2049 | Value *AlignmentValue = EmitScalarExpr(E->getArg(1)); | |||
2050 | ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue); | |||
2051 | if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment)) | |||
2052 | AlignmentCI = ConstantInt::get(AlignmentCI->getType(), | |||
2053 | llvm::Value::MaximumAlignment); | |||
2054 | ||||
2055 | EmitAlignmentAssumption(PtrValue, Ptr, | |||
2056 | /*The expr loc is sufficient.*/ SourceLocation(), | |||
2057 | AlignmentCI, OffsetValue); | |||
2058 | return RValue::get(PtrValue); | |||
2059 | } | |||
2060 | case Builtin::BI__assume: | |||
2061 | case Builtin::BI__builtin_assume: { | |||
2062 | if (E->getArg(0)->HasSideEffects(getContext())) | |||
2063 | return RValue::get(nullptr); | |||
2064 | ||||
2065 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
2066 | Function *FnAssume = CGM.getIntrinsic(Intrinsic::assume); | |||
2067 | return RValue::get(Builder.CreateCall(FnAssume, ArgValue)); | |||
2068 | } | |||
2069 | case Builtin::BI__builtin_bswap16: | |||
2070 | case Builtin::BI__builtin_bswap32: | |||
2071 | case Builtin::BI__builtin_bswap64: { | |||
2072 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap)); | |||
2073 | } | |||
2074 | case Builtin::BI__builtin_bitreverse8: | |||
2075 | case Builtin::BI__builtin_bitreverse16: | |||
2076 | case Builtin::BI__builtin_bitreverse32: | |||
2077 | case Builtin::BI__builtin_bitreverse64: { | |||
2078 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse)); | |||
2079 | } | |||
2080 | case Builtin::BI__builtin_rotateleft8: | |||
2081 | case Builtin::BI__builtin_rotateleft16: | |||
2082 | case Builtin::BI__builtin_rotateleft32: | |||
2083 | case Builtin::BI__builtin_rotateleft64: | |||
2084 | case Builtin::BI_rotl8: // Microsoft variants of rotate left | |||
2085 | case Builtin::BI_rotl16: | |||
2086 | case Builtin::BI_rotl: | |||
2087 | case Builtin::BI_lrotl: | |||
2088 | case Builtin::BI_rotl64: | |||
2089 | return emitRotate(E, false); | |||
2090 | ||||
2091 | case Builtin::BI__builtin_rotateright8: | |||
2092 | case Builtin::BI__builtin_rotateright16: | |||
2093 | case Builtin::BI__builtin_rotateright32: | |||
2094 | case Builtin::BI__builtin_rotateright64: | |||
2095 | case Builtin::BI_rotr8: // Microsoft variants of rotate right | |||
2096 | case Builtin::BI_rotr16: | |||
2097 | case Builtin::BI_rotr: | |||
2098 | case Builtin::BI_lrotr: | |||
2099 | case Builtin::BI_rotr64: | |||
2100 | return emitRotate(E, true); | |||
2101 | ||||
2102 | case Builtin::BI__builtin_constant_p: { | |||
2103 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
2104 | ||||
2105 | const Expr *Arg = E->getArg(0); | |||
2106 | QualType ArgType = Arg->getType(); | |||
2107 | // FIXME: The allowance for Obj-C pointers and block pointers is historical | |||
2108 | // and likely a mistake. | |||
2109 | if (!ArgType->isIntegralOrEnumerationType() && !ArgType->isFloatingType() && | |||
2110 | !ArgType->isObjCObjectPointerType() && !ArgType->isBlockPointerType()) | |||
2111 | // Per the GCC documentation, only numeric constants are recognized after | |||
2112 | // inlining. | |||
2113 | return RValue::get(ConstantInt::get(ResultType, 0)); | |||
2114 | ||||
2115 | if (Arg->HasSideEffects(getContext())) | |||
2116 | // The argument is unevaluated, so be conservative if it might have | |||
2117 | // side-effects. | |||
2118 | return RValue::get(ConstantInt::get(ResultType, 0)); | |||
2119 | ||||
2120 | Value *ArgValue = EmitScalarExpr(Arg); | |||
2121 | if (ArgType->isObjCObjectPointerType()) { | |||
2122 | // Convert Objective-C objects to id because we cannot distinguish between | |||
2123 | // LLVM types for Obj-C classes as they are opaque. | |||
2124 | ArgType = CGM.getContext().getObjCIdType(); | |||
2125 | ArgValue = Builder.CreateBitCast(ArgValue, ConvertType(ArgType)); | |||
2126 | } | |||
2127 | Function *F = | |||
2128 | CGM.getIntrinsic(Intrinsic::is_constant, ConvertType(ArgType)); | |||
2129 | Value *Result = Builder.CreateCall(F, ArgValue); | |||
2130 | if (Result->getType() != ResultType) | |||
2131 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/false); | |||
2132 | return RValue::get(Result); | |||
2133 | } | |||
2134 | case Builtin::BI__builtin_dynamic_object_size: | |||
2135 | case Builtin::BI__builtin_object_size: { | |||
2136 | unsigned Type = | |||
2137 | E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue(); | |||
2138 | auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType())); | |||
2139 | ||||
2140 | // We pass this builtin onto the optimizer so that it can figure out the | |||
2141 | // object size in more complex cases. | |||
2142 | bool IsDynamic = BuiltinID == Builtin::BI__builtin_dynamic_object_size; | |||
2143 | return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType, | |||
2144 | /*EmittedE=*/nullptr, IsDynamic)); | |||
2145 | } | |||
2146 | case Builtin::BI__builtin_prefetch: { | |||
2147 | Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0)); | |||
2148 | // FIXME: Technically these constants should of type 'int', yes? | |||
2149 | RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) : | |||
2150 | llvm::ConstantInt::get(Int32Ty, 0); | |||
2151 | Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : | |||
2152 | llvm::ConstantInt::get(Int32Ty, 3); | |||
2153 | Value *Data = llvm::ConstantInt::get(Int32Ty, 1); | |||
2154 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); | |||
2155 | return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data})); | |||
2156 | } | |||
2157 | case Builtin::BI__builtin_readcyclecounter: { | |||
2158 | Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter); | |||
2159 | return RValue::get(Builder.CreateCall(F)); | |||
2160 | } | |||
2161 | case Builtin::BI__builtin___clear_cache: { | |||
2162 | Value *Begin = EmitScalarExpr(E->getArg(0)); | |||
2163 | Value *End = EmitScalarExpr(E->getArg(1)); | |||
2164 | Function *F = CGM.getIntrinsic(Intrinsic::clear_cache); | |||
2165 | return RValue::get(Builder.CreateCall(F, {Begin, End})); | |||
2166 | } | |||
2167 | case Builtin::BI__builtin_trap: | |||
2168 | return RValue::get(EmitTrapCall(Intrinsic::trap)); | |||
2169 | case Builtin::BI__debugbreak: | |||
2170 | return RValue::get(EmitTrapCall(Intrinsic::debugtrap)); | |||
2171 | case Builtin::BI__builtin_unreachable: { | |||
2172 | EmitUnreachable(E->getExprLoc()); | |||
2173 | ||||
2174 | // We do need to preserve an insertion point. | |||
2175 | EmitBlock(createBasicBlock("unreachable.cont")); | |||
2176 | ||||
2177 | return RValue::get(nullptr); | |||
2178 | } | |||
2179 | ||||
2180 | case Builtin::BI__builtin_powi: | |||
2181 | case Builtin::BI__builtin_powif: | |||
2182 | case Builtin::BI__builtin_powil: { | |||
2183 | Value *Base = EmitScalarExpr(E->getArg(0)); | |||
2184 | Value *Exponent = EmitScalarExpr(E->getArg(1)); | |||
2185 | llvm::Type *ArgType = Base->getType(); | |||
2186 | Function *F = CGM.getIntrinsic(Intrinsic::powi, ArgType); | |||
2187 | return RValue::get(Builder.CreateCall(F, {Base, Exponent})); | |||
2188 | } | |||
2189 | ||||
2190 | case Builtin::BI__builtin_isgreater: | |||
2191 | case Builtin::BI__builtin_isgreaterequal: | |||
2192 | case Builtin::BI__builtin_isless: | |||
2193 | case Builtin::BI__builtin_islessequal: | |||
2194 | case Builtin::BI__builtin_islessgreater: | |||
2195 | case Builtin::BI__builtin_isunordered: { | |||
2196 | // Ordered comparisons: we know the arguments to these are matching scalar | |||
2197 | // floating point values. | |||
2198 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
2199 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
2200 | ||||
2201 | switch (BuiltinID) { | |||
2202 | default: llvm_unreachable("Unknown ordered comparison")::llvm::llvm_unreachable_internal("Unknown ordered comparison" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 2202); | |||
2203 | case Builtin::BI__builtin_isgreater: | |||
2204 | LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp"); | |||
2205 | break; | |||
2206 | case Builtin::BI__builtin_isgreaterequal: | |||
2207 | LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp"); | |||
2208 | break; | |||
2209 | case Builtin::BI__builtin_isless: | |||
2210 | LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp"); | |||
2211 | break; | |||
2212 | case Builtin::BI__builtin_islessequal: | |||
2213 | LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp"); | |||
2214 | break; | |||
2215 | case Builtin::BI__builtin_islessgreater: | |||
2216 | LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp"); | |||
2217 | break; | |||
2218 | case Builtin::BI__builtin_isunordered: | |||
2219 | LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp"); | |||
2220 | break; | |||
2221 | } | |||
2222 | // ZExt bool to int type. | |||
2223 | return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType()))); | |||
2224 | } | |||
2225 | case Builtin::BI__builtin_isnan: { | |||
2226 | Value *V = EmitScalarExpr(E->getArg(0)); | |||
2227 | V = Builder.CreateFCmpUNO(V, V, "cmp"); | |||
2228 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); | |||
2229 | } | |||
2230 | ||||
2231 | case Builtin::BIfinite: | |||
2232 | case Builtin::BI__finite: | |||
2233 | case Builtin::BIfinitef: | |||
2234 | case Builtin::BI__finitef: | |||
2235 | case Builtin::BIfinitel: | |||
2236 | case Builtin::BI__finitel: | |||
2237 | case Builtin::BI__builtin_isinf: | |||
2238 | case Builtin::BI__builtin_isfinite: { | |||
2239 | // isinf(x) --> fabs(x) == infinity | |||
2240 | // isfinite(x) --> fabs(x) != infinity | |||
2241 | // x != NaN via the ordered compare in either case. | |||
2242 | Value *V = EmitScalarExpr(E->getArg(0)); | |||
2243 | Value *Fabs = EmitFAbs(*this, V); | |||
2244 | Constant *Infinity = ConstantFP::getInfinity(V->getType()); | |||
2245 | CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf) | |||
2246 | ? CmpInst::FCMP_OEQ | |||
2247 | : CmpInst::FCMP_ONE; | |||
2248 | Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf"); | |||
2249 | return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType()))); | |||
2250 | } | |||
2251 | ||||
2252 | case Builtin::BI__builtin_isinf_sign: { | |||
2253 | // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0 | |||
2254 | Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
2255 | Value *AbsArg = EmitFAbs(*this, Arg); | |||
2256 | Value *IsInf = Builder.CreateFCmpOEQ( | |||
2257 | AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf"); | |||
2258 | Value *IsNeg = EmitSignBit(*this, Arg); | |||
2259 | ||||
2260 | llvm::Type *IntTy = ConvertType(E->getType()); | |||
2261 | Value *Zero = Constant::getNullValue(IntTy); | |||
2262 | Value *One = ConstantInt::get(IntTy, 1); | |||
2263 | Value *NegativeOne = ConstantInt::get(IntTy, -1); | |||
2264 | Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One); | |||
2265 | Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero); | |||
2266 | return RValue::get(Result); | |||
2267 | } | |||
2268 | ||||
2269 | case Builtin::BI__builtin_isnormal: { | |||
2270 | // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min | |||
2271 | Value *V = EmitScalarExpr(E->getArg(0)); | |||
2272 | Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq"); | |||
2273 | ||||
2274 | Value *Abs = EmitFAbs(*this, V); | |||
2275 | Value *IsLessThanInf = | |||
2276 | Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf"); | |||
2277 | APFloat Smallest = APFloat::getSmallestNormalized( | |||
2278 | getContext().getFloatTypeSemantics(E->getArg(0)->getType())); | |||
2279 | Value *IsNormal = | |||
2280 | Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest), | |||
2281 | "isnormal"); | |||
2282 | V = Builder.CreateAnd(Eq, IsLessThanInf, "and"); | |||
2283 | V = Builder.CreateAnd(V, IsNormal, "and"); | |||
2284 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); | |||
2285 | } | |||
2286 | ||||
2287 | case Builtin::BI__builtin_flt_rounds: { | |||
2288 | Function *F = CGM.getIntrinsic(Intrinsic::flt_rounds); | |||
2289 | ||||
2290 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
2291 | Value *Result = Builder.CreateCall(F); | |||
2292 | if (Result->getType() != ResultType) | |||
2293 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
2294 | "cast"); | |||
2295 | return RValue::get(Result); | |||
2296 | } | |||
2297 | ||||
2298 | case Builtin::BI__builtin_fpclassify: { | |||
2299 | Value *V = EmitScalarExpr(E->getArg(5)); | |||
2300 | llvm::Type *Ty = ConvertType(E->getArg(5)->getType()); | |||
2301 | ||||
2302 | // Create Result | |||
2303 | BasicBlock *Begin = Builder.GetInsertBlock(); | |||
2304 | BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn); | |||
2305 | Builder.SetInsertPoint(End); | |||
2306 | PHINode *Result = | |||
2307 | Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4, | |||
2308 | "fpclassify_result"); | |||
2309 | ||||
2310 | // if (V==0) return FP_ZERO | |||
2311 | Builder.SetInsertPoint(Begin); | |||
2312 | Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty), | |||
2313 | "iszero"); | |||
2314 | Value *ZeroLiteral = EmitScalarExpr(E->getArg(4)); | |||
2315 | BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn); | |||
2316 | Builder.CreateCondBr(IsZero, End, NotZero); | |||
2317 | Result->addIncoming(ZeroLiteral, Begin); | |||
2318 | ||||
2319 | // if (V != V) return FP_NAN | |||
2320 | Builder.SetInsertPoint(NotZero); | |||
2321 | Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp"); | |||
2322 | Value *NanLiteral = EmitScalarExpr(E->getArg(0)); | |||
2323 | BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn); | |||
2324 | Builder.CreateCondBr(IsNan, End, NotNan); | |||
2325 | Result->addIncoming(NanLiteral, NotZero); | |||
2326 | ||||
2327 | // if (fabs(V) == infinity) return FP_INFINITY | |||
2328 | Builder.SetInsertPoint(NotNan); | |||
2329 | Value *VAbs = EmitFAbs(*this, V); | |||
2330 | Value *IsInf = | |||
2331 | Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()), | |||
2332 | "isinf"); | |||
2333 | Value *InfLiteral = EmitScalarExpr(E->getArg(1)); | |||
2334 | BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn); | |||
2335 | Builder.CreateCondBr(IsInf, End, NotInf); | |||
2336 | Result->addIncoming(InfLiteral, NotNan); | |||
2337 | ||||
2338 | // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL | |||
2339 | Builder.SetInsertPoint(NotInf); | |||
2340 | APFloat Smallest = APFloat::getSmallestNormalized( | |||
2341 | getContext().getFloatTypeSemantics(E->getArg(5)->getType())); | |||
2342 | Value *IsNormal = | |||
2343 | Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest), | |||
2344 | "isnormal"); | |||
2345 | Value *NormalResult = | |||
2346 | Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)), | |||
2347 | EmitScalarExpr(E->getArg(3))); | |||
2348 | Builder.CreateBr(End); | |||
2349 | Result->addIncoming(NormalResult, NotInf); | |||
2350 | ||||
2351 | // return Result | |||
2352 | Builder.SetInsertPoint(End); | |||
2353 | return RValue::get(Result); | |||
2354 | } | |||
2355 | ||||
2356 | case Builtin::BIalloca: | |||
2357 | case Builtin::BI_alloca: | |||
2358 | case Builtin::BI__builtin_alloca: { | |||
2359 | Value *Size = EmitScalarExpr(E->getArg(0)); | |||
2360 | const TargetInfo &TI = getContext().getTargetInfo(); | |||
2361 | // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__. | |||
2362 | unsigned SuitableAlignmentInBytes = | |||
2363 | CGM.getContext() | |||
2364 | .toCharUnitsFromBits(TI.getSuitableAlign()) | |||
2365 | .getQuantity(); | |||
2366 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); | |||
2367 | AI->setAlignment(MaybeAlign(SuitableAlignmentInBytes)); | |||
2368 | initializeAlloca(*this, AI, Size, SuitableAlignmentInBytes); | |||
2369 | return RValue::get(AI); | |||
2370 | } | |||
2371 | ||||
2372 | case Builtin::BI__builtin_alloca_with_align: { | |||
2373 | Value *Size = EmitScalarExpr(E->getArg(0)); | |||
2374 | Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1)); | |||
2375 | auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue); | |||
2376 | unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue(); | |||
2377 | unsigned AlignmentInBytes = | |||
2378 | CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getQuantity(); | |||
2379 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); | |||
2380 | AI->setAlignment(MaybeAlign(AlignmentInBytes)); | |||
2381 | initializeAlloca(*this, AI, Size, AlignmentInBytes); | |||
2382 | return RValue::get(AI); | |||
2383 | } | |||
2384 | ||||
2385 | case Builtin::BIbzero: | |||
2386 | case Builtin::BI__builtin_bzero: { | |||
2387 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
2388 | Value *SizeVal = EmitScalarExpr(E->getArg(1)); | |||
2389 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
2390 | E->getArg(0)->getExprLoc(), FD, 0); | |||
2391 | Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false); | |||
2392 | return RValue::get(nullptr); | |||
2393 | } | |||
2394 | case Builtin::BImemcpy: | |||
2395 | case Builtin::BI__builtin_memcpy: { | |||
2396 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
2397 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
2398 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); | |||
2399 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
2400 | E->getArg(0)->getExprLoc(), FD, 0); | |||
2401 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), | |||
2402 | E->getArg(1)->getExprLoc(), FD, 1); | |||
2403 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); | |||
2404 | return RValue::get(Dest.getPointer()); | |||
2405 | } | |||
2406 | ||||
2407 | case Builtin::BI__builtin_char_memchr: | |||
2408 | BuiltinID = Builtin::BI__builtin_memchr; | |||
2409 | break; | |||
2410 | ||||
2411 | case Builtin::BI__builtin___memcpy_chk: { | |||
2412 | // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2. | |||
2413 | Expr::EvalResult SizeResult, DstSizeResult; | |||
2414 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || | |||
2415 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) | |||
2416 | break; | |||
2417 | llvm::APSInt Size = SizeResult.Val.getInt(); | |||
2418 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); | |||
2419 | if (Size.ugt(DstSize)) | |||
2420 | break; | |||
2421 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
2422 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
2423 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); | |||
2424 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); | |||
2425 | return RValue::get(Dest.getPointer()); | |||
2426 | } | |||
2427 | ||||
2428 | case Builtin::BI__builtin_objc_memmove_collectable: { | |||
2429 | Address DestAddr = EmitPointerWithAlignment(E->getArg(0)); | |||
2430 | Address SrcAddr = EmitPointerWithAlignment(E->getArg(1)); | |||
2431 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); | |||
2432 | CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, | |||
2433 | DestAddr, SrcAddr, SizeVal); | |||
2434 | return RValue::get(DestAddr.getPointer()); | |||
2435 | } | |||
2436 | ||||
2437 | case Builtin::BI__builtin___memmove_chk: { | |||
2438 | // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2. | |||
2439 | Expr::EvalResult SizeResult, DstSizeResult; | |||
2440 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || | |||
2441 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) | |||
2442 | break; | |||
2443 | llvm::APSInt Size = SizeResult.Val.getInt(); | |||
2444 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); | |||
2445 | if (Size.ugt(DstSize)) | |||
2446 | break; | |||
2447 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
2448 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
2449 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); | |||
2450 | Builder.CreateMemMove(Dest, Src, SizeVal, false); | |||
2451 | return RValue::get(Dest.getPointer()); | |||
2452 | } | |||
2453 | ||||
2454 | case Builtin::BImemmove: | |||
2455 | case Builtin::BI__builtin_memmove: { | |||
2456 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
2457 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
2458 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); | |||
2459 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
2460 | E->getArg(0)->getExprLoc(), FD, 0); | |||
2461 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), | |||
2462 | E->getArg(1)->getExprLoc(), FD, 1); | |||
2463 | Builder.CreateMemMove(Dest, Src, SizeVal, false); | |||
2464 | return RValue::get(Dest.getPointer()); | |||
2465 | } | |||
2466 | case Builtin::BImemset: | |||
2467 | case Builtin::BI__builtin_memset: { | |||
2468 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
2469 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), | |||
2470 | Builder.getInt8Ty()); | |||
2471 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); | |||
2472 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
2473 | E->getArg(0)->getExprLoc(), FD, 0); | |||
2474 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); | |||
2475 | return RValue::get(Dest.getPointer()); | |||
2476 | } | |||
2477 | case Builtin::BI__builtin___memset_chk: { | |||
2478 | // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2. | |||
2479 | Expr::EvalResult SizeResult, DstSizeResult; | |||
2480 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || | |||
2481 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) | |||
2482 | break; | |||
2483 | llvm::APSInt Size = SizeResult.Val.getInt(); | |||
2484 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); | |||
2485 | if (Size.ugt(DstSize)) | |||
2486 | break; | |||
2487 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
2488 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), | |||
2489 | Builder.getInt8Ty()); | |||
2490 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); | |||
2491 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); | |||
2492 | return RValue::get(Dest.getPointer()); | |||
2493 | } | |||
2494 | case Builtin::BI__builtin_wmemcmp: { | |||
2495 | // The MSVC runtime library does not provide a definition of wmemcmp, so we | |||
2496 | // need an inline implementation. | |||
2497 | if (!getTarget().getTriple().isOSMSVCRT()) | |||
2498 | break; | |||
2499 | ||||
2500 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); | |||
2501 | ||||
2502 | Value *Dst = EmitScalarExpr(E->getArg(0)); | |||
2503 | Value *Src = EmitScalarExpr(E->getArg(1)); | |||
2504 | Value *Size = EmitScalarExpr(E->getArg(2)); | |||
2505 | ||||
2506 | BasicBlock *Entry = Builder.GetInsertBlock(); | |||
2507 | BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt"); | |||
2508 | BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt"); | |||
2509 | BasicBlock *Next = createBasicBlock("wmemcmp.next"); | |||
2510 | BasicBlock *Exit = createBasicBlock("wmemcmp.exit"); | |||
2511 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); | |||
2512 | Builder.CreateCondBr(SizeEq0, Exit, CmpGT); | |||
2513 | ||||
2514 | EmitBlock(CmpGT); | |||
2515 | PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2); | |||
2516 | DstPhi->addIncoming(Dst, Entry); | |||
2517 | PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2); | |||
2518 | SrcPhi->addIncoming(Src, Entry); | |||
2519 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); | |||
2520 | SizePhi->addIncoming(Size, Entry); | |||
2521 | CharUnits WCharAlign = | |||
2522 | getContext().getTypeAlignInChars(getContext().WCharTy); | |||
2523 | Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign); | |||
2524 | Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign); | |||
2525 | Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh); | |||
2526 | Builder.CreateCondBr(DstGtSrc, Exit, CmpLT); | |||
2527 | ||||
2528 | EmitBlock(CmpLT); | |||
2529 | Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh); | |||
2530 | Builder.CreateCondBr(DstLtSrc, Exit, Next); | |||
2531 | ||||
2532 | EmitBlock(Next); | |||
2533 | Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1); | |||
2534 | Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1); | |||
2535 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); | |||
2536 | Value *NextSizeEq0 = | |||
2537 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); | |||
2538 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT); | |||
2539 | DstPhi->addIncoming(NextDst, Next); | |||
2540 | SrcPhi->addIncoming(NextSrc, Next); | |||
2541 | SizePhi->addIncoming(NextSize, Next); | |||
2542 | ||||
2543 | EmitBlock(Exit); | |||
2544 | PHINode *Ret = Builder.CreatePHI(IntTy, 4); | |||
2545 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry); | |||
2546 | Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT); | |||
2547 | Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT); | |||
2548 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Next); | |||
2549 | return RValue::get(Ret); | |||
2550 | } | |||
2551 | case Builtin::BI__builtin_dwarf_cfa: { | |||
2552 | // The offset in bytes from the first argument to the CFA. | |||
2553 | // | |||
2554 | // Why on earth is this in the frontend? Is there any reason at | |||
2555 | // all that the backend can't reasonably determine this while | |||
2556 | // lowering llvm.eh.dwarf.cfa()? | |||
2557 | // | |||
2558 | // TODO: If there's a satisfactory reason, add a target hook for | |||
2559 | // this instead of hard-coding 0, which is correct for most targets. | |||
2560 | int32_t Offset = 0; | |||
2561 | ||||
2562 | Function *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa); | |||
2563 | return RValue::get(Builder.CreateCall(F, | |||
2564 | llvm::ConstantInt::get(Int32Ty, Offset))); | |||
2565 | } | |||
2566 | case Builtin::BI__builtin_return_address: { | |||
2567 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), | |||
2568 | getContext().UnsignedIntTy); | |||
2569 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); | |||
2570 | return RValue::get(Builder.CreateCall(F, Depth)); | |||
2571 | } | |||
2572 | case Builtin::BI_ReturnAddress: { | |||
2573 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); | |||
2574 | return RValue::get(Builder.CreateCall(F, Builder.getInt32(0))); | |||
2575 | } | |||
2576 | case Builtin::BI__builtin_frame_address: { | |||
2577 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), | |||
2578 | getContext().UnsignedIntTy); | |||
2579 | Function *F = CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy); | |||
2580 | return RValue::get(Builder.CreateCall(F, Depth)); | |||
2581 | } | |||
2582 | case Builtin::BI__builtin_extract_return_addr: { | |||
2583 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
2584 | Value *Result = getTargetHooks().decodeReturnAddress(*this, Address); | |||
2585 | return RValue::get(Result); | |||
2586 | } | |||
2587 | case Builtin::BI__builtin_frob_return_addr: { | |||
2588 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
2589 | Value *Result = getTargetHooks().encodeReturnAddress(*this, Address); | |||
2590 | return RValue::get(Result); | |||
2591 | } | |||
2592 | case Builtin::BI__builtin_dwarf_sp_column: { | |||
2593 | llvm::IntegerType *Ty | |||
2594 | = cast<llvm::IntegerType>(ConvertType(E->getType())); | |||
2595 | int Column = getTargetHooks().getDwarfEHStackPointer(CGM); | |||
2596 | if (Column == -1) { | |||
2597 | CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column"); | |||
2598 | return RValue::get(llvm::UndefValue::get(Ty)); | |||
2599 | } | |||
2600 | return RValue::get(llvm::ConstantInt::get(Ty, Column, true)); | |||
2601 | } | |||
2602 | case Builtin::BI__builtin_init_dwarf_reg_size_table: { | |||
2603 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
2604 | if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address)) | |||
2605 | CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table"); | |||
2606 | return RValue::get(llvm::UndefValue::get(ConvertType(E->getType()))); | |||
2607 | } | |||
2608 | case Builtin::BI__builtin_eh_return: { | |||
2609 | Value *Int = EmitScalarExpr(E->getArg(0)); | |||
2610 | Value *Ptr = EmitScalarExpr(E->getArg(1)); | |||
2611 | ||||
2612 | llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType()); | |||
2613 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 2614, __PRETTY_FUNCTION__)) | |||
2614 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 2614, __PRETTY_FUNCTION__)); | |||
2615 | Function *F = | |||
2616 | CGM.getIntrinsic(IntTy->getBitWidth() == 32 ? Intrinsic::eh_return_i32 | |||
2617 | : Intrinsic::eh_return_i64); | |||
2618 | Builder.CreateCall(F, {Int, Ptr}); | |||
2619 | Builder.CreateUnreachable(); | |||
2620 | ||||
2621 | // We do need to preserve an insertion point. | |||
2622 | EmitBlock(createBasicBlock("builtin_eh_return.cont")); | |||
2623 | ||||
2624 | return RValue::get(nullptr); | |||
2625 | } | |||
2626 | case Builtin::BI__builtin_unwind_init: { | |||
2627 | Function *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init); | |||
2628 | return RValue::get(Builder.CreateCall(F)); | |||
2629 | } | |||
2630 | case Builtin::BI__builtin_extend_pointer: { | |||
2631 | // Extends a pointer to the size of an _Unwind_Word, which is | |||
2632 | // uint64_t on all platforms. Generally this gets poked into a | |||
2633 | // register and eventually used as an address, so if the | |||
2634 | // addressing registers are wider than pointers and the platform | |||
2635 | // doesn't implicitly ignore high-order bits when doing | |||
2636 | // addressing, we need to make sure we zext / sext based on | |||
2637 | // the platform's expectations. | |||
2638 | // | |||
2639 | // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html | |||
2640 | ||||
2641 | // Cast the pointer to intptr_t. | |||
2642 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
2643 | Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast"); | |||
2644 | ||||
2645 | // If that's 64 bits, we're done. | |||
2646 | if (IntPtrTy->getBitWidth() == 64) | |||
2647 | return RValue::get(Result); | |||
2648 | ||||
2649 | // Otherwise, ask the codegen data what to do. | |||
2650 | if (getTargetHooks().extendPointerWithSExt()) | |||
2651 | return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext")); | |||
2652 | else | |||
2653 | return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext")); | |||
2654 | } | |||
2655 | case Builtin::BI__builtin_setjmp: { | |||
2656 | // Buffer is a void**. | |||
2657 | Address Buf = EmitPointerWithAlignment(E->getArg(0)); | |||
2658 | ||||
2659 | // Store the frame pointer to the setjmp buffer. | |||
2660 | Value *FrameAddr = Builder.CreateCall( | |||
2661 | CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy), | |||
2662 | ConstantInt::get(Int32Ty, 0)); | |||
2663 | Builder.CreateStore(FrameAddr, Buf); | |||
2664 | ||||
2665 | // Store the stack pointer to the setjmp buffer. | |||
2666 | Value *StackAddr = | |||
2667 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave)); | |||
2668 | Address StackSaveSlot = Builder.CreateConstInBoundsGEP(Buf, 2); | |||
2669 | Builder.CreateStore(StackAddr, StackSaveSlot); | |||
2670 | ||||
2671 | // Call LLVM's EH setjmp, which is lightweight. | |||
2672 | Function *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp); | |||
2673 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); | |||
2674 | return RValue::get(Builder.CreateCall(F, Buf.getPointer())); | |||
2675 | } | |||
2676 | case Builtin::BI__builtin_longjmp: { | |||
2677 | Value *Buf = EmitScalarExpr(E->getArg(0)); | |||
2678 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); | |||
2679 | ||||
2680 | // Call LLVM's EH longjmp, which is lightweight. | |||
2681 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf); | |||
2682 | ||||
2683 | // longjmp doesn't return; mark this as unreachable. | |||
2684 | Builder.CreateUnreachable(); | |||
2685 | ||||
2686 | // We do need to preserve an insertion point. | |||
2687 | EmitBlock(createBasicBlock("longjmp.cont")); | |||
2688 | ||||
2689 | return RValue::get(nullptr); | |||
2690 | } | |||
2691 | case Builtin::BI__builtin_launder: { | |||
2692 | const Expr *Arg = E->getArg(0); | |||
2693 | QualType ArgTy = Arg->getType()->getPointeeType(); | |||
2694 | Value *Ptr = EmitScalarExpr(Arg); | |||
2695 | if (TypeRequiresBuiltinLaunder(CGM, ArgTy)) | |||
2696 | Ptr = Builder.CreateLaunderInvariantGroup(Ptr); | |||
2697 | ||||
2698 | return RValue::get(Ptr); | |||
2699 | } | |||
2700 | case Builtin::BI__sync_fetch_and_add: | |||
2701 | case Builtin::BI__sync_fetch_and_sub: | |||
2702 | case Builtin::BI__sync_fetch_and_or: | |||
2703 | case Builtin::BI__sync_fetch_and_and: | |||
2704 | case Builtin::BI__sync_fetch_and_xor: | |||
2705 | case Builtin::BI__sync_fetch_and_nand: | |||
2706 | case Builtin::BI__sync_add_and_fetch: | |||
2707 | case Builtin::BI__sync_sub_and_fetch: | |||
2708 | case Builtin::BI__sync_and_and_fetch: | |||
2709 | case Builtin::BI__sync_or_and_fetch: | |||
2710 | case Builtin::BI__sync_xor_and_fetch: | |||
2711 | case Builtin::BI__sync_nand_and_fetch: | |||
2712 | case Builtin::BI__sync_val_compare_and_swap: | |||
2713 | case Builtin::BI__sync_bool_compare_and_swap: | |||
2714 | case Builtin::BI__sync_lock_test_and_set: | |||
2715 | case Builtin::BI__sync_lock_release: | |||
2716 | case Builtin::BI__sync_swap: | |||
2717 | llvm_unreachable("Shouldn't make it through sema")::llvm::llvm_unreachable_internal("Shouldn't make it through sema" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 2717); | |||
2718 | case Builtin::BI__sync_fetch_and_add_1: | |||
2719 | case Builtin::BI__sync_fetch_and_add_2: | |||
2720 | case Builtin::BI__sync_fetch_and_add_4: | |||
2721 | case Builtin::BI__sync_fetch_and_add_8: | |||
2722 | case Builtin::BI__sync_fetch_and_add_16: | |||
2723 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E); | |||
2724 | case Builtin::BI__sync_fetch_and_sub_1: | |||
2725 | case Builtin::BI__sync_fetch_and_sub_2: | |||
2726 | case Builtin::BI__sync_fetch_and_sub_4: | |||
2727 | case Builtin::BI__sync_fetch_and_sub_8: | |||
2728 | case Builtin::BI__sync_fetch_and_sub_16: | |||
2729 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E); | |||
2730 | case Builtin::BI__sync_fetch_and_or_1: | |||
2731 | case Builtin::BI__sync_fetch_and_or_2: | |||
2732 | case Builtin::BI__sync_fetch_and_or_4: | |||
2733 | case Builtin::BI__sync_fetch_and_or_8: | |||
2734 | case Builtin::BI__sync_fetch_and_or_16: | |||
2735 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E); | |||
2736 | case Builtin::BI__sync_fetch_and_and_1: | |||
2737 | case Builtin::BI__sync_fetch_and_and_2: | |||
2738 | case Builtin::BI__sync_fetch_and_and_4: | |||
2739 | case Builtin::BI__sync_fetch_and_and_8: | |||
2740 | case Builtin::BI__sync_fetch_and_and_16: | |||
2741 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E); | |||
2742 | case Builtin::BI__sync_fetch_and_xor_1: | |||
2743 | case Builtin::BI__sync_fetch_and_xor_2: | |||
2744 | case Builtin::BI__sync_fetch_and_xor_4: | |||
2745 | case Builtin::BI__sync_fetch_and_xor_8: | |||
2746 | case Builtin::BI__sync_fetch_and_xor_16: | |||
2747 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E); | |||
2748 | case Builtin::BI__sync_fetch_and_nand_1: | |||
2749 | case Builtin::BI__sync_fetch_and_nand_2: | |||
2750 | case Builtin::BI__sync_fetch_and_nand_4: | |||
2751 | case Builtin::BI__sync_fetch_and_nand_8: | |||
2752 | case Builtin::BI__sync_fetch_and_nand_16: | |||
2753 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E); | |||
2754 | ||||
2755 | // Clang extensions: not overloaded yet. | |||
2756 | case Builtin::BI__sync_fetch_and_min: | |||
2757 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E); | |||
2758 | case Builtin::BI__sync_fetch_and_max: | |||
2759 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E); | |||
2760 | case Builtin::BI__sync_fetch_and_umin: | |||
2761 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E); | |||
2762 | case Builtin::BI__sync_fetch_and_umax: | |||
2763 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E); | |||
2764 | ||||
2765 | case Builtin::BI__sync_add_and_fetch_1: | |||
2766 | case Builtin::BI__sync_add_and_fetch_2: | |||
2767 | case Builtin::BI__sync_add_and_fetch_4: | |||
2768 | case Builtin::BI__sync_add_and_fetch_8: | |||
2769 | case Builtin::BI__sync_add_and_fetch_16: | |||
2770 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E, | |||
2771 | llvm::Instruction::Add); | |||
2772 | case Builtin::BI__sync_sub_and_fetch_1: | |||
2773 | case Builtin::BI__sync_sub_and_fetch_2: | |||
2774 | case Builtin::BI__sync_sub_and_fetch_4: | |||
2775 | case Builtin::BI__sync_sub_and_fetch_8: | |||
2776 | case Builtin::BI__sync_sub_and_fetch_16: | |||
2777 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E, | |||
2778 | llvm::Instruction::Sub); | |||
2779 | case Builtin::BI__sync_and_and_fetch_1: | |||
2780 | case Builtin::BI__sync_and_and_fetch_2: | |||
2781 | case Builtin::BI__sync_and_and_fetch_4: | |||
2782 | case Builtin::BI__sync_and_and_fetch_8: | |||
2783 | case Builtin::BI__sync_and_and_fetch_16: | |||
2784 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E, | |||
2785 | llvm::Instruction::And); | |||
2786 | case Builtin::BI__sync_or_and_fetch_1: | |||
2787 | case Builtin::BI__sync_or_and_fetch_2: | |||
2788 | case Builtin::BI__sync_or_and_fetch_4: | |||
2789 | case Builtin::BI__sync_or_and_fetch_8: | |||
2790 | case Builtin::BI__sync_or_and_fetch_16: | |||
2791 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E, | |||
2792 | llvm::Instruction::Or); | |||
2793 | case Builtin::BI__sync_xor_and_fetch_1: | |||
2794 | case Builtin::BI__sync_xor_and_fetch_2: | |||
2795 | case Builtin::BI__sync_xor_and_fetch_4: | |||
2796 | case Builtin::BI__sync_xor_and_fetch_8: | |||
2797 | case Builtin::BI__sync_xor_and_fetch_16: | |||
2798 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E, | |||
2799 | llvm::Instruction::Xor); | |||
2800 | case Builtin::BI__sync_nand_and_fetch_1: | |||
2801 | case Builtin::BI__sync_nand_and_fetch_2: | |||
2802 | case Builtin::BI__sync_nand_and_fetch_4: | |||
2803 | case Builtin::BI__sync_nand_and_fetch_8: | |||
2804 | case Builtin::BI__sync_nand_and_fetch_16: | |||
2805 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E, | |||
2806 | llvm::Instruction::And, true); | |||
2807 | ||||
2808 | case Builtin::BI__sync_val_compare_and_swap_1: | |||
2809 | case Builtin::BI__sync_val_compare_and_swap_2: | |||
2810 | case Builtin::BI__sync_val_compare_and_swap_4: | |||
2811 | case Builtin::BI__sync_val_compare_and_swap_8: | |||
2812 | case Builtin::BI__sync_val_compare_and_swap_16: | |||
2813 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, false)); | |||
2814 | ||||
2815 | case Builtin::BI__sync_bool_compare_and_swap_1: | |||
2816 | case Builtin::BI__sync_bool_compare_and_swap_2: | |||
2817 | case Builtin::BI__sync_bool_compare_and_swap_4: | |||
2818 | case Builtin::BI__sync_bool_compare_and_swap_8: | |||
2819 | case Builtin::BI__sync_bool_compare_and_swap_16: | |||
2820 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, true)); | |||
2821 | ||||
2822 | case Builtin::BI__sync_swap_1: | |||
2823 | case Builtin::BI__sync_swap_2: | |||
2824 | case Builtin::BI__sync_swap_4: | |||
2825 | case Builtin::BI__sync_swap_8: | |||
2826 | case Builtin::BI__sync_swap_16: | |||
2827 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); | |||
2828 | ||||
2829 | case Builtin::BI__sync_lock_test_and_set_1: | |||
2830 | case Builtin::BI__sync_lock_test_and_set_2: | |||
2831 | case Builtin::BI__sync_lock_test_and_set_4: | |||
2832 | case Builtin::BI__sync_lock_test_and_set_8: | |||
2833 | case Builtin::BI__sync_lock_test_and_set_16: | |||
2834 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); | |||
2835 | ||||
2836 | case Builtin::BI__sync_lock_release_1: | |||
2837 | case Builtin::BI__sync_lock_release_2: | |||
2838 | case Builtin::BI__sync_lock_release_4: | |||
2839 | case Builtin::BI__sync_lock_release_8: | |||
2840 | case Builtin::BI__sync_lock_release_16: { | |||
2841 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
2842 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); | |||
2843 | CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy); | |||
2844 | llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(), | |||
2845 | StoreSize.getQuantity() * 8); | |||
2846 | Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo()); | |||
2847 | llvm::StoreInst *Store = | |||
2848 | Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr, | |||
2849 | StoreSize); | |||
2850 | Store->setAtomic(llvm::AtomicOrdering::Release); | |||
2851 | return RValue::get(nullptr); | |||
2852 | } | |||
2853 | ||||
2854 | case Builtin::BI__sync_synchronize: { | |||
2855 | // We assume this is supposed to correspond to a C++0x-style | |||
2856 | // sequentially-consistent fence (i.e. this is only usable for | |||
2857 | // synchronization, not device I/O or anything like that). This intrinsic | |||
2858 | // is really badly designed in the sense that in theory, there isn't | |||
2859 | // any way to safely use it... but in practice, it mostly works | |||
2860 | // to use it with non-atomic loads and stores to get acquire/release | |||
2861 | // semantics. | |||
2862 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent); | |||
2863 | return RValue::get(nullptr); | |||
2864 | } | |||
2865 | ||||
2866 | case Builtin::BI__builtin_nontemporal_load: | |||
2867 | return RValue::get(EmitNontemporalLoad(*this, E)); | |||
2868 | case Builtin::BI__builtin_nontemporal_store: | |||
2869 | return RValue::get(EmitNontemporalStore(*this, E)); | |||
2870 | case Builtin::BI__c11_atomic_is_lock_free: | |||
2871 | case Builtin::BI__atomic_is_lock_free: { | |||
2872 | // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the | |||
2873 | // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since | |||
2874 | // _Atomic(T) is always properly-aligned. | |||
2875 | const char *LibCallName = "__atomic_is_lock_free"; | |||
2876 | CallArgList Args; | |||
2877 | Args.add(RValue::get(EmitScalarExpr(E->getArg(0))), | |||
2878 | getContext().getSizeType()); | |||
2879 | if (BuiltinID == Builtin::BI__atomic_is_lock_free) | |||
2880 | Args.add(RValue::get(EmitScalarExpr(E->getArg(1))), | |||
2881 | getContext().VoidPtrTy); | |||
2882 | else | |||
2883 | Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)), | |||
2884 | getContext().VoidPtrTy); | |||
2885 | const CGFunctionInfo &FuncInfo = | |||
2886 | CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args); | |||
2887 | llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo); | |||
2888 | llvm::FunctionCallee Func = CGM.CreateRuntimeFunction(FTy, LibCallName); | |||
2889 | return EmitCall(FuncInfo, CGCallee::forDirect(Func), | |||
2890 | ReturnValueSlot(), Args); | |||
2891 | } | |||
2892 | ||||
2893 | case Builtin::BI__atomic_test_and_set: { | |||
2894 | // Look at the argument type to determine whether this is a volatile | |||
2895 | // operation. The parameter type is always volatile. | |||
2896 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); | |||
2897 | bool Volatile = | |||
2898 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); | |||
2899 | ||||
2900 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
2901 | unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace(); | |||
2902 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); | |||
2903 | Value *NewVal = Builder.getInt8(1); | |||
2904 | Value *Order = EmitScalarExpr(E->getArg(1)); | |||
2905 | if (isa<llvm::ConstantInt>(Order)) { | |||
2906 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); | |||
2907 | AtomicRMWInst *Result = nullptr; | |||
2908 | switch (ord) { | |||
2909 | case 0: // memory_order_relaxed | |||
2910 | default: // invalid order | |||
2911 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
2912 | llvm::AtomicOrdering::Monotonic); | |||
2913 | break; | |||
2914 | case 1: // memory_order_consume | |||
2915 | case 2: // memory_order_acquire | |||
2916 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
2917 | llvm::AtomicOrdering::Acquire); | |||
2918 | break; | |||
2919 | case 3: // memory_order_release | |||
2920 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
2921 | llvm::AtomicOrdering::Release); | |||
2922 | break; | |||
2923 | case 4: // memory_order_acq_rel | |||
2924 | ||||
2925 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
2926 | llvm::AtomicOrdering::AcquireRelease); | |||
2927 | break; | |||
2928 | case 5: // memory_order_seq_cst | |||
2929 | Result = Builder.CreateAtomicRMW( | |||
2930 | llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
2931 | llvm::AtomicOrdering::SequentiallyConsistent); | |||
2932 | break; | |||
2933 | } | |||
2934 | Result->setVolatile(Volatile); | |||
2935 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); | |||
2936 | } | |||
2937 | ||||
2938 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); | |||
2939 | ||||
2940 | llvm::BasicBlock *BBs[5] = { | |||
2941 | createBasicBlock("monotonic", CurFn), | |||
2942 | createBasicBlock("acquire", CurFn), | |||
2943 | createBasicBlock("release", CurFn), | |||
2944 | createBasicBlock("acqrel", CurFn), | |||
2945 | createBasicBlock("seqcst", CurFn) | |||
2946 | }; | |||
2947 | llvm::AtomicOrdering Orders[5] = { | |||
2948 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire, | |||
2949 | llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease, | |||
2950 | llvm::AtomicOrdering::SequentiallyConsistent}; | |||
2951 | ||||
2952 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); | |||
2953 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); | |||
2954 | ||||
2955 | Builder.SetInsertPoint(ContBB); | |||
2956 | PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set"); | |||
2957 | ||||
2958 | for (unsigned i = 0; i < 5; ++i) { | |||
2959 | Builder.SetInsertPoint(BBs[i]); | |||
2960 | AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, | |||
2961 | Ptr, NewVal, Orders[i]); | |||
2962 | RMW->setVolatile(Volatile); | |||
2963 | Result->addIncoming(RMW, BBs[i]); | |||
2964 | Builder.CreateBr(ContBB); | |||
2965 | } | |||
2966 | ||||
2967 | SI->addCase(Builder.getInt32(0), BBs[0]); | |||
2968 | SI->addCase(Builder.getInt32(1), BBs[1]); | |||
2969 | SI->addCase(Builder.getInt32(2), BBs[1]); | |||
2970 | SI->addCase(Builder.getInt32(3), BBs[2]); | |||
2971 | SI->addCase(Builder.getInt32(4), BBs[3]); | |||
2972 | SI->addCase(Builder.getInt32(5), BBs[4]); | |||
2973 | ||||
2974 | Builder.SetInsertPoint(ContBB); | |||
2975 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); | |||
2976 | } | |||
2977 | ||||
2978 | case Builtin::BI__atomic_clear: { | |||
2979 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); | |||
2980 | bool Volatile = | |||
2981 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); | |||
2982 | ||||
2983 | Address Ptr = EmitPointerWithAlignment(E->getArg(0)); | |||
2984 | unsigned AddrSpace = Ptr.getPointer()->getType()->getPointerAddressSpace(); | |||
2985 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); | |||
2986 | Value *NewVal = Builder.getInt8(0); | |||
2987 | Value *Order = EmitScalarExpr(E->getArg(1)); | |||
2988 | if (isa<llvm::ConstantInt>(Order)) { | |||
2989 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); | |||
2990 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); | |||
2991 | switch (ord) { | |||
2992 | case 0: // memory_order_relaxed | |||
2993 | default: // invalid order | |||
2994 | Store->setOrdering(llvm::AtomicOrdering::Monotonic); | |||
2995 | break; | |||
2996 | case 3: // memory_order_release | |||
2997 | Store->setOrdering(llvm::AtomicOrdering::Release); | |||
2998 | break; | |||
2999 | case 5: // memory_order_seq_cst | |||
3000 | Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent); | |||
3001 | break; | |||
3002 | } | |||
3003 | return RValue::get(nullptr); | |||
3004 | } | |||
3005 | ||||
3006 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); | |||
3007 | ||||
3008 | llvm::BasicBlock *BBs[3] = { | |||
3009 | createBasicBlock("monotonic", CurFn), | |||
3010 | createBasicBlock("release", CurFn), | |||
3011 | createBasicBlock("seqcst", CurFn) | |||
3012 | }; | |||
3013 | llvm::AtomicOrdering Orders[3] = { | |||
3014 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release, | |||
3015 | llvm::AtomicOrdering::SequentiallyConsistent}; | |||
3016 | ||||
3017 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); | |||
3018 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); | |||
3019 | ||||
3020 | for (unsigned i = 0; i < 3; ++i) { | |||
3021 | Builder.SetInsertPoint(BBs[i]); | |||
3022 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); | |||
3023 | Store->setOrdering(Orders[i]); | |||
3024 | Builder.CreateBr(ContBB); | |||
3025 | } | |||
3026 | ||||
3027 | SI->addCase(Builder.getInt32(0), BBs[0]); | |||
3028 | SI->addCase(Builder.getInt32(3), BBs[1]); | |||
3029 | SI->addCase(Builder.getInt32(5), BBs[2]); | |||
3030 | ||||
3031 | Builder.SetInsertPoint(ContBB); | |||
3032 | return RValue::get(nullptr); | |||
3033 | } | |||
3034 | ||||
3035 | case Builtin::BI__atomic_thread_fence: | |||
3036 | case Builtin::BI__atomic_signal_fence: | |||
3037 | case Builtin::BI__c11_atomic_thread_fence: | |||
3038 | case Builtin::BI__c11_atomic_signal_fence: { | |||
3039 | llvm::SyncScope::ID SSID; | |||
3040 | if (BuiltinID == Builtin::BI__atomic_signal_fence || | |||
3041 | BuiltinID == Builtin::BI__c11_atomic_signal_fence) | |||
3042 | SSID = llvm::SyncScope::SingleThread; | |||
3043 | else | |||
3044 | SSID = llvm::SyncScope::System; | |||
3045 | Value *Order = EmitScalarExpr(E->getArg(0)); | |||
3046 | if (isa<llvm::ConstantInt>(Order)) { | |||
3047 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); | |||
3048 | switch (ord) { | |||
3049 | case 0: // memory_order_relaxed | |||
3050 | default: // invalid order | |||
3051 | break; | |||
3052 | case 1: // memory_order_consume | |||
3053 | case 2: // memory_order_acquire | |||
3054 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); | |||
3055 | break; | |||
3056 | case 3: // memory_order_release | |||
3057 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); | |||
3058 | break; | |||
3059 | case 4: // memory_order_acq_rel | |||
3060 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); | |||
3061 | break; | |||
3062 | case 5: // memory_order_seq_cst | |||
3063 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); | |||
3064 | break; | |||
3065 | } | |||
3066 | return RValue::get(nullptr); | |||
3067 | } | |||
3068 | ||||
3069 | llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB; | |||
3070 | AcquireBB = createBasicBlock("acquire", CurFn); | |||
3071 | ReleaseBB = createBasicBlock("release", CurFn); | |||
3072 | AcqRelBB = createBasicBlock("acqrel", CurFn); | |||
3073 | SeqCstBB = createBasicBlock("seqcst", CurFn); | |||
3074 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); | |||
3075 | ||||
3076 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); | |||
3077 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB); | |||
3078 | ||||
3079 | Builder.SetInsertPoint(AcquireBB); | |||
3080 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); | |||
3081 | Builder.CreateBr(ContBB); | |||
3082 | SI->addCase(Builder.getInt32(1), AcquireBB); | |||
3083 | SI->addCase(Builder.getInt32(2), AcquireBB); | |||
3084 | ||||
3085 | Builder.SetInsertPoint(ReleaseBB); | |||
3086 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); | |||
3087 | Builder.CreateBr(ContBB); | |||
3088 | SI->addCase(Builder.getInt32(3), ReleaseBB); | |||
3089 | ||||
3090 | Builder.SetInsertPoint(AcqRelBB); | |||
3091 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); | |||
3092 | Builder.CreateBr(ContBB); | |||
3093 | SI->addCase(Builder.getInt32(4), AcqRelBB); | |||
3094 | ||||
3095 | Builder.SetInsertPoint(SeqCstBB); | |||
3096 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); | |||
3097 | Builder.CreateBr(ContBB); | |||
3098 | SI->addCase(Builder.getInt32(5), SeqCstBB); | |||
3099 | ||||
3100 | Builder.SetInsertPoint(ContBB); | |||
3101 | return RValue::get(nullptr); | |||
3102 | } | |||
3103 | ||||
3104 | case Builtin::BI__builtin_signbit: | |||
3105 | case Builtin::BI__builtin_signbitf: | |||
3106 | case Builtin::BI__builtin_signbitl: { | |||
3107 | return RValue::get( | |||
3108 | Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))), | |||
3109 | ConvertType(E->getType()))); | |||
3110 | } | |||
3111 | case Builtin::BI__annotation: { | |||
3112 | // Re-encode each wide string to UTF8 and make an MDString. | |||
3113 | SmallVector<Metadata *, 1> Strings; | |||
3114 | for (const Expr *Arg : E->arguments()) { | |||
3115 | const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts()); | |||
3116 | assert(Str->getCharByteWidth() == 2)((Str->getCharByteWidth() == 2) ? static_cast<void> ( 0) : __assert_fail ("Str->getCharByteWidth() == 2", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3116, __PRETTY_FUNCTION__)); | |||
3117 | StringRef WideBytes = Str->getBytes(); | |||
3118 | std::string StrUtf8; | |||
3119 | if (!convertUTF16ToUTF8String( | |||
3120 | makeArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) { | |||
3121 | CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument"); | |||
3122 | continue; | |||
3123 | } | |||
3124 | Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8)); | |||
3125 | } | |||
3126 | ||||
3127 | // Build and MDTuple of MDStrings and emit the intrinsic call. | |||
3128 | llvm::Function *F = | |||
3129 | CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {}); | |||
3130 | MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings); | |||
3131 | Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple)); | |||
3132 | return RValue::getIgnored(); | |||
3133 | } | |||
3134 | case Builtin::BI__builtin_annotation: { | |||
3135 | llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0)); | |||
3136 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::annotation, | |||
3137 | AnnVal->getType()); | |||
3138 | ||||
3139 | // Get the annotation string, go through casts. Sema requires this to be a | |||
3140 | // non-wide string literal, potentially casted, so the cast<> is safe. | |||
3141 | const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts(); | |||
3142 | StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString(); | |||
3143 | return RValue::get(EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc())); | |||
3144 | } | |||
3145 | case Builtin::BI__builtin_addcb: | |||
3146 | case Builtin::BI__builtin_addcs: | |||
3147 | case Builtin::BI__builtin_addc: | |||
3148 | case Builtin::BI__builtin_addcl: | |||
3149 | case Builtin::BI__builtin_addcll: | |||
3150 | case Builtin::BI__builtin_subcb: | |||
3151 | case Builtin::BI__builtin_subcs: | |||
3152 | case Builtin::BI__builtin_subc: | |||
3153 | case Builtin::BI__builtin_subcl: | |||
3154 | case Builtin::BI__builtin_subcll: { | |||
3155 | ||||
3156 | // We translate all of these builtins from expressions of the form: | |||
3157 | // int x = ..., y = ..., carryin = ..., carryout, result; | |||
3158 | // result = __builtin_addc(x, y, carryin, &carryout); | |||
3159 | // | |||
3160 | // to LLVM IR of the form: | |||
3161 | // | |||
3162 | // %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y) | |||
3163 | // %tmpsum1 = extractvalue {i32, i1} %tmp1, 0 | |||
3164 | // %carry1 = extractvalue {i32, i1} %tmp1, 1 | |||
3165 | // %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1, | |||
3166 | // i32 %carryin) | |||
3167 | // %result = extractvalue {i32, i1} %tmp2, 0 | |||
3168 | // %carry2 = extractvalue {i32, i1} %tmp2, 1 | |||
3169 | // %tmp3 = or i1 %carry1, %carry2 | |||
3170 | // %tmp4 = zext i1 %tmp3 to i32 | |||
3171 | // store i32 %tmp4, i32* %carryout | |||
3172 | ||||
3173 | // Scalarize our inputs. | |||
3174 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); | |||
3175 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); | |||
3176 | llvm::Value *Carryin = EmitScalarExpr(E->getArg(2)); | |||
3177 | Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3)); | |||
3178 | ||||
3179 | // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow. | |||
3180 | llvm::Intrinsic::ID IntrinsicId; | |||
3181 | switch (BuiltinID) { | |||
3182 | default: llvm_unreachable("Unknown multiprecision builtin id.")::llvm::llvm_unreachable_internal("Unknown multiprecision builtin id." , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3182); | |||
3183 | case Builtin::BI__builtin_addcb: | |||
3184 | case Builtin::BI__builtin_addcs: | |||
3185 | case Builtin::BI__builtin_addc: | |||
3186 | case Builtin::BI__builtin_addcl: | |||
3187 | case Builtin::BI__builtin_addcll: | |||
3188 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; | |||
3189 | break; | |||
3190 | case Builtin::BI__builtin_subcb: | |||
3191 | case Builtin::BI__builtin_subcs: | |||
3192 | case Builtin::BI__builtin_subc: | |||
3193 | case Builtin::BI__builtin_subcl: | |||
3194 | case Builtin::BI__builtin_subcll: | |||
3195 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; | |||
3196 | break; | |||
3197 | } | |||
3198 | ||||
3199 | // Construct our resulting LLVM IR expression. | |||
3200 | llvm::Value *Carry1; | |||
3201 | llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId, | |||
3202 | X, Y, Carry1); | |||
3203 | llvm::Value *Carry2; | |||
3204 | llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId, | |||
3205 | Sum1, Carryin, Carry2); | |||
3206 | llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2), | |||
3207 | X->getType()); | |||
3208 | Builder.CreateStore(CarryOut, CarryOutPtr); | |||
3209 | return RValue::get(Sum2); | |||
3210 | } | |||
3211 | ||||
3212 | case Builtin::BI__builtin_add_overflow: | |||
3213 | case Builtin::BI__builtin_sub_overflow: | |||
3214 | case Builtin::BI__builtin_mul_overflow: { | |||
3215 | const clang::Expr *LeftArg = E->getArg(0); | |||
3216 | const clang::Expr *RightArg = E->getArg(1); | |||
3217 | const clang::Expr *ResultArg = E->getArg(2); | |||
3218 | ||||
3219 | clang::QualType ResultQTy = | |||
3220 | ResultArg->getType()->castAs<PointerType>()->getPointeeType(); | |||
3221 | ||||
3222 | WidthAndSignedness LeftInfo = | |||
3223 | getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType()); | |||
3224 | WidthAndSignedness RightInfo = | |||
3225 | getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType()); | |||
3226 | WidthAndSignedness ResultInfo = | |||
3227 | getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy); | |||
3228 | ||||
3229 | // Handle mixed-sign multiplication as a special case, because adding | |||
3230 | // runtime or backend support for our generic irgen would be too expensive. | |||
3231 | if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo)) | |||
3232 | return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg, | |||
3233 | RightInfo, ResultArg, ResultQTy, | |||
3234 | ResultInfo); | |||
3235 | ||||
3236 | WidthAndSignedness EncompassingInfo = | |||
3237 | EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo}); | |||
3238 | ||||
3239 | llvm::Type *EncompassingLLVMTy = | |||
3240 | llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width); | |||
3241 | ||||
3242 | llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy); | |||
3243 | ||||
3244 | llvm::Intrinsic::ID IntrinsicId; | |||
3245 | switch (BuiltinID) { | |||
3246 | default: | |||
3247 | llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3247); | |||
3248 | case Builtin::BI__builtin_add_overflow: | |||
3249 | IntrinsicId = EncompassingInfo.Signed | |||
3250 | ? llvm::Intrinsic::sadd_with_overflow | |||
3251 | : llvm::Intrinsic::uadd_with_overflow; | |||
3252 | break; | |||
3253 | case Builtin::BI__builtin_sub_overflow: | |||
3254 | IntrinsicId = EncompassingInfo.Signed | |||
3255 | ? llvm::Intrinsic::ssub_with_overflow | |||
3256 | : llvm::Intrinsic::usub_with_overflow; | |||
3257 | break; | |||
3258 | case Builtin::BI__builtin_mul_overflow: | |||
3259 | IntrinsicId = EncompassingInfo.Signed | |||
3260 | ? llvm::Intrinsic::smul_with_overflow | |||
3261 | : llvm::Intrinsic::umul_with_overflow; | |||
3262 | break; | |||
3263 | } | |||
3264 | ||||
3265 | llvm::Value *Left = EmitScalarExpr(LeftArg); | |||
3266 | llvm::Value *Right = EmitScalarExpr(RightArg); | |||
3267 | Address ResultPtr = EmitPointerWithAlignment(ResultArg); | |||
3268 | ||||
3269 | // Extend each operand to the encompassing type. | |||
3270 | Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed); | |||
3271 | Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed); | |||
3272 | ||||
3273 | // Perform the operation on the extended values. | |||
3274 | llvm::Value *Overflow, *Result; | |||
3275 | Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow); | |||
3276 | ||||
3277 | if (EncompassingInfo.Width > ResultInfo.Width) { | |||
3278 | // The encompassing type is wider than the result type, so we need to | |||
3279 | // truncate it. | |||
3280 | llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy); | |||
3281 | ||||
3282 | // To see if the truncation caused an overflow, we will extend | |||
3283 | // the result and then compare it to the original result. | |||
3284 | llvm::Value *ResultTruncExt = Builder.CreateIntCast( | |||
3285 | ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed); | |||
3286 | llvm::Value *TruncationOverflow = | |||
3287 | Builder.CreateICmpNE(Result, ResultTruncExt); | |||
3288 | ||||
3289 | Overflow = Builder.CreateOr(Overflow, TruncationOverflow); | |||
3290 | Result = ResultTrunc; | |||
3291 | } | |||
3292 | ||||
3293 | // Finally, store the result using the pointer. | |||
3294 | bool isVolatile = | |||
3295 | ResultArg->getType()->getPointeeType().isVolatileQualified(); | |||
3296 | Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile); | |||
3297 | ||||
3298 | return RValue::get(Overflow); | |||
3299 | } | |||
3300 | ||||
3301 | case Builtin::BI__builtin_uadd_overflow: | |||
3302 | case Builtin::BI__builtin_uaddl_overflow: | |||
3303 | case Builtin::BI__builtin_uaddll_overflow: | |||
3304 | case Builtin::BI__builtin_usub_overflow: | |||
3305 | case Builtin::BI__builtin_usubl_overflow: | |||
3306 | case Builtin::BI__builtin_usubll_overflow: | |||
3307 | case Builtin::BI__builtin_umul_overflow: | |||
3308 | case Builtin::BI__builtin_umull_overflow: | |||
3309 | case Builtin::BI__builtin_umulll_overflow: | |||
3310 | case Builtin::BI__builtin_sadd_overflow: | |||
3311 | case Builtin::BI__builtin_saddl_overflow: | |||
3312 | case Builtin::BI__builtin_saddll_overflow: | |||
3313 | case Builtin::BI__builtin_ssub_overflow: | |||
3314 | case Builtin::BI__builtin_ssubl_overflow: | |||
3315 | case Builtin::BI__builtin_ssubll_overflow: | |||
3316 | case Builtin::BI__builtin_smul_overflow: | |||
3317 | case Builtin::BI__builtin_smull_overflow: | |||
3318 | case Builtin::BI__builtin_smulll_overflow: { | |||
3319 | ||||
3320 | // We translate all of these builtins directly to the relevant llvm IR node. | |||
3321 | ||||
3322 | // Scalarize our inputs. | |||
3323 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); | |||
3324 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); | |||
3325 | Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2)); | |||
3326 | ||||
3327 | // Decide which of the overflow intrinsics we are lowering to: | |||
3328 | llvm::Intrinsic::ID IntrinsicId; | |||
3329 | switch (BuiltinID) { | |||
3330 | default: llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3330); | |||
3331 | case Builtin::BI__builtin_uadd_overflow: | |||
3332 | case Builtin::BI__builtin_uaddl_overflow: | |||
3333 | case Builtin::BI__builtin_uaddll_overflow: | |||
3334 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; | |||
3335 | break; | |||
3336 | case Builtin::BI__builtin_usub_overflow: | |||
3337 | case Builtin::BI__builtin_usubl_overflow: | |||
3338 | case Builtin::BI__builtin_usubll_overflow: | |||
3339 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; | |||
3340 | break; | |||
3341 | case Builtin::BI__builtin_umul_overflow: | |||
3342 | case Builtin::BI__builtin_umull_overflow: | |||
3343 | case Builtin::BI__builtin_umulll_overflow: | |||
3344 | IntrinsicId = llvm::Intrinsic::umul_with_overflow; | |||
3345 | break; | |||
3346 | case Builtin::BI__builtin_sadd_overflow: | |||
3347 | case Builtin::BI__builtin_saddl_overflow: | |||
3348 | case Builtin::BI__builtin_saddll_overflow: | |||
3349 | IntrinsicId = llvm::Intrinsic::sadd_with_overflow; | |||
3350 | break; | |||
3351 | case Builtin::BI__builtin_ssub_overflow: | |||
3352 | case Builtin::BI__builtin_ssubl_overflow: | |||
3353 | case Builtin::BI__builtin_ssubll_overflow: | |||
3354 | IntrinsicId = llvm::Intrinsic::ssub_with_overflow; | |||
3355 | break; | |||
3356 | case Builtin::BI__builtin_smul_overflow: | |||
3357 | case Builtin::BI__builtin_smull_overflow: | |||
3358 | case Builtin::BI__builtin_smulll_overflow: | |||
3359 | IntrinsicId = llvm::Intrinsic::smul_with_overflow; | |||
3360 | break; | |||
3361 | } | |||
3362 | ||||
3363 | ||||
3364 | llvm::Value *Carry; | |||
3365 | llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry); | |||
3366 | Builder.CreateStore(Sum, SumOutPtr); | |||
3367 | ||||
3368 | return RValue::get(Carry); | |||
3369 | } | |||
3370 | case Builtin::BI__builtin_addressof: | |||
3371 | return RValue::get(EmitLValue(E->getArg(0)).getPointer()); | |||
3372 | case Builtin::BI__builtin_operator_new: | |||
3373 | return EmitBuiltinNewDeleteCall( | |||
3374 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false); | |||
3375 | case Builtin::BI__builtin_operator_delete: | |||
3376 | return EmitBuiltinNewDeleteCall( | |||
3377 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true); | |||
3378 | ||||
3379 | case Builtin::BI__noop: | |||
3380 | // __noop always evaluates to an integer literal zero. | |||
3381 | return RValue::get(ConstantInt::get(IntTy, 0)); | |||
3382 | case Builtin::BI__builtin_call_with_static_chain: { | |||
3383 | const CallExpr *Call = cast<CallExpr>(E->getArg(0)); | |||
3384 | const Expr *Chain = E->getArg(1); | |||
3385 | return EmitCall(Call->getCallee()->getType(), | |||
3386 | EmitCallee(Call->getCallee()), Call, ReturnValue, | |||
3387 | EmitScalarExpr(Chain)); | |||
3388 | } | |||
3389 | case Builtin::BI_InterlockedExchange8: | |||
3390 | case Builtin::BI_InterlockedExchange16: | |||
3391 | case Builtin::BI_InterlockedExchange: | |||
3392 | case Builtin::BI_InterlockedExchangePointer: | |||
3393 | return RValue::get( | |||
3394 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E)); | |||
3395 | case Builtin::BI_InterlockedCompareExchangePointer: | |||
3396 | case Builtin::BI_InterlockedCompareExchangePointer_nf: { | |||
3397 | llvm::Type *RTy; | |||
3398 | llvm::IntegerType *IntType = | |||
3399 | IntegerType::get(getLLVMContext(), | |||
3400 | getContext().getTypeSize(E->getType())); | |||
3401 | llvm::Type *IntPtrType = IntType->getPointerTo(); | |||
3402 | ||||
3403 | llvm::Value *Destination = | |||
3404 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType); | |||
3405 | ||||
3406 | llvm::Value *Exchange = EmitScalarExpr(E->getArg(1)); | |||
3407 | RTy = Exchange->getType(); | |||
3408 | Exchange = Builder.CreatePtrToInt(Exchange, IntType); | |||
3409 | ||||
3410 | llvm::Value *Comparand = | |||
3411 | Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType); | |||
3412 | ||||
3413 | auto Ordering = | |||
3414 | BuiltinID == Builtin::BI_InterlockedCompareExchangePointer_nf ? | |||
3415 | AtomicOrdering::Monotonic : AtomicOrdering::SequentiallyConsistent; | |||
3416 | ||||
3417 | auto Result = Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, | |||
3418 | Ordering, Ordering); | |||
3419 | Result->setVolatile(true); | |||
3420 | ||||
3421 | return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result, | |||
3422 | 0), | |||
3423 | RTy)); | |||
3424 | } | |||
3425 | case Builtin::BI_InterlockedCompareExchange8: | |||
3426 | case Builtin::BI_InterlockedCompareExchange16: | |||
3427 | case Builtin::BI_InterlockedCompareExchange: | |||
3428 | case Builtin::BI_InterlockedCompareExchange64: | |||
3429 | return RValue::get(EmitAtomicCmpXchgForMSIntrin(*this, E)); | |||
3430 | case Builtin::BI_InterlockedIncrement16: | |||
3431 | case Builtin::BI_InterlockedIncrement: | |||
3432 | return RValue::get( | |||
3433 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E)); | |||
3434 | case Builtin::BI_InterlockedDecrement16: | |||
3435 | case Builtin::BI_InterlockedDecrement: | |||
3436 | return RValue::get( | |||
3437 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E)); | |||
3438 | case Builtin::BI_InterlockedAnd8: | |||
3439 | case Builtin::BI_InterlockedAnd16: | |||
3440 | case Builtin::BI_InterlockedAnd: | |||
3441 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E)); | |||
3442 | case Builtin::BI_InterlockedExchangeAdd8: | |||
3443 | case Builtin::BI_InterlockedExchangeAdd16: | |||
3444 | case Builtin::BI_InterlockedExchangeAdd: | |||
3445 | return RValue::get( | |||
3446 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E)); | |||
3447 | case Builtin::BI_InterlockedExchangeSub8: | |||
3448 | case Builtin::BI_InterlockedExchangeSub16: | |||
3449 | case Builtin::BI_InterlockedExchangeSub: | |||
3450 | return RValue::get( | |||
3451 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E)); | |||
3452 | case Builtin::BI_InterlockedOr8: | |||
3453 | case Builtin::BI_InterlockedOr16: | |||
3454 | case Builtin::BI_InterlockedOr: | |||
3455 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E)); | |||
3456 | case Builtin::BI_InterlockedXor8: | |||
3457 | case Builtin::BI_InterlockedXor16: | |||
3458 | case Builtin::BI_InterlockedXor: | |||
3459 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E)); | |||
3460 | ||||
3461 | case Builtin::BI_bittest64: | |||
3462 | case Builtin::BI_bittest: | |||
3463 | case Builtin::BI_bittestandcomplement64: | |||
3464 | case Builtin::BI_bittestandcomplement: | |||
3465 | case Builtin::BI_bittestandreset64: | |||
3466 | case Builtin::BI_bittestandreset: | |||
3467 | case Builtin::BI_bittestandset64: | |||
3468 | case Builtin::BI_bittestandset: | |||
3469 | case Builtin::BI_interlockedbittestandreset: | |||
3470 | case Builtin::BI_interlockedbittestandreset64: | |||
3471 | case Builtin::BI_interlockedbittestandset64: | |||
3472 | case Builtin::BI_interlockedbittestandset: | |||
3473 | case Builtin::BI_interlockedbittestandset_acq: | |||
3474 | case Builtin::BI_interlockedbittestandset_rel: | |||
3475 | case Builtin::BI_interlockedbittestandset_nf: | |||
3476 | case Builtin::BI_interlockedbittestandreset_acq: | |||
3477 | case Builtin::BI_interlockedbittestandreset_rel: | |||
3478 | case Builtin::BI_interlockedbittestandreset_nf: | |||
3479 | return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E)); | |||
3480 | ||||
3481 | // These builtins exist to emit regular volatile loads and stores not | |||
3482 | // affected by the -fms-volatile setting. | |||
3483 | case Builtin::BI__iso_volatile_load8: | |||
3484 | case Builtin::BI__iso_volatile_load16: | |||
3485 | case Builtin::BI__iso_volatile_load32: | |||
3486 | case Builtin::BI__iso_volatile_load64: | |||
3487 | return RValue::get(EmitISOVolatileLoad(*this, E)); | |||
3488 | case Builtin::BI__iso_volatile_store8: | |||
3489 | case Builtin::BI__iso_volatile_store16: | |||
3490 | case Builtin::BI__iso_volatile_store32: | |||
3491 | case Builtin::BI__iso_volatile_store64: | |||
3492 | return RValue::get(EmitISOVolatileStore(*this, E)); | |||
3493 | ||||
3494 | case Builtin::BI__exception_code: | |||
3495 | case Builtin::BI_exception_code: | |||
3496 | return RValue::get(EmitSEHExceptionCode()); | |||
3497 | case Builtin::BI__exception_info: | |||
3498 | case Builtin::BI_exception_info: | |||
3499 | return RValue::get(EmitSEHExceptionInfo()); | |||
3500 | case Builtin::BI__abnormal_termination: | |||
3501 | case Builtin::BI_abnormal_termination: | |||
3502 | return RValue::get(EmitSEHAbnormalTermination()); | |||
3503 | case Builtin::BI_setjmpex: | |||
3504 | if (getTarget().getTriple().isOSMSVCRT()) | |||
3505 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); | |||
3506 | break; | |||
3507 | case Builtin::BI_setjmp: | |||
3508 | if (getTarget().getTriple().isOSMSVCRT()) { | |||
3509 | if (getTarget().getTriple().getArch() == llvm::Triple::x86) | |||
3510 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E); | |||
3511 | else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64) | |||
3512 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); | |||
3513 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E); | |||
3514 | } | |||
3515 | break; | |||
3516 | ||||
3517 | case Builtin::BI__GetExceptionInfo: { | |||
3518 | if (llvm::GlobalVariable *GV = | |||
3519 | CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType())) | |||
3520 | return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy)); | |||
3521 | break; | |||
3522 | } | |||
3523 | ||||
3524 | case Builtin::BI__fastfail: | |||
3525 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E)); | |||
3526 | ||||
3527 | case Builtin::BI__builtin_coro_size: { | |||
3528 | auto & Context = getContext(); | |||
3529 | auto SizeTy = Context.getSizeType(); | |||
3530 | auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy)); | |||
3531 | Function *F = CGM.getIntrinsic(Intrinsic::coro_size, T); | |||
3532 | return RValue::get(Builder.CreateCall(F)); | |||
3533 | } | |||
3534 | ||||
3535 | case Builtin::BI__builtin_coro_id: | |||
3536 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_id); | |||
3537 | case Builtin::BI__builtin_coro_promise: | |||
3538 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise); | |||
3539 | case Builtin::BI__builtin_coro_resume: | |||
3540 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume); | |||
3541 | case Builtin::BI__builtin_coro_frame: | |||
3542 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame); | |||
3543 | case Builtin::BI__builtin_coro_noop: | |||
3544 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop); | |||
3545 | case Builtin::BI__builtin_coro_free: | |||
3546 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_free); | |||
3547 | case Builtin::BI__builtin_coro_destroy: | |||
3548 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy); | |||
3549 | case Builtin::BI__builtin_coro_done: | |||
3550 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_done); | |||
3551 | case Builtin::BI__builtin_coro_alloc: | |||
3552 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc); | |||
3553 | case Builtin::BI__builtin_coro_begin: | |||
3554 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin); | |||
3555 | case Builtin::BI__builtin_coro_end: | |||
3556 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_end); | |||
3557 | case Builtin::BI__builtin_coro_suspend: | |||
3558 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend); | |||
3559 | case Builtin::BI__builtin_coro_param: | |||
3560 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_param); | |||
3561 | ||||
3562 | // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions | |||
3563 | case Builtin::BIread_pipe: | |||
3564 | case Builtin::BIwrite_pipe: { | |||
3565 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), | |||
3566 | *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
3567 | CGOpenCLRuntime OpenCLRT(CGM); | |||
3568 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); | |||
3569 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); | |||
3570 | ||||
3571 | // Type of the generic packet parameter. | |||
3572 | unsigned GenericAS = | |||
3573 | getContext().getTargetAddressSpace(LangAS::opencl_generic); | |||
3574 | llvm::Type *I8PTy = llvm::PointerType::get( | |||
3575 | llvm::Type::getInt8Ty(getLLVMContext()), GenericAS); | |||
3576 | ||||
3577 | // Testing which overloaded version we should generate the call for. | |||
3578 | if (2U == E->getNumArgs()) { | |||
3579 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2" | |||
3580 | : "__write_pipe_2"; | |||
3581 | // Creating a generic function type to be able to call with any builtin or | |||
3582 | // user defined type. | |||
3583 | llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty}; | |||
3584 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
3585 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3586 | Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy); | |||
3587 | return RValue::get( | |||
3588 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
3589 | {Arg0, BCast, PacketSize, PacketAlign})); | |||
3590 | } else { | |||
3591 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3592, __PRETTY_FUNCTION__)) | |||
3592 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3592, __PRETTY_FUNCTION__)); | |||
3593 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4" | |||
3594 | : "__write_pipe_4"; | |||
3595 | ||||
3596 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy, | |||
3597 | Int32Ty, Int32Ty}; | |||
3598 | Value *Arg2 = EmitScalarExpr(E->getArg(2)), | |||
3599 | *Arg3 = EmitScalarExpr(E->getArg(3)); | |||
3600 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
3601 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3602 | Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy); | |||
3603 | // We know the third argument is an integer type, but we may need to cast | |||
3604 | // it to i32. | |||
3605 | if (Arg2->getType() != Int32Ty) | |||
3606 | Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty); | |||
3607 | return RValue::get(Builder.CreateCall( | |||
3608 | CGM.CreateRuntimeFunction(FTy, Name), | |||
3609 | {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign})); | |||
3610 | } | |||
3611 | } | |||
3612 | // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write | |||
3613 | // functions | |||
3614 | case Builtin::BIreserve_read_pipe: | |||
3615 | case Builtin::BIreserve_write_pipe: | |||
3616 | case Builtin::BIwork_group_reserve_read_pipe: | |||
3617 | case Builtin::BIwork_group_reserve_write_pipe: | |||
3618 | case Builtin::BIsub_group_reserve_read_pipe: | |||
3619 | case Builtin::BIsub_group_reserve_write_pipe: { | |||
3620 | // Composing the mangled name for the function. | |||
3621 | const char *Name; | |||
3622 | if (BuiltinID == Builtin::BIreserve_read_pipe) | |||
3623 | Name = "__reserve_read_pipe"; | |||
3624 | else if (BuiltinID == Builtin::BIreserve_write_pipe) | |||
3625 | Name = "__reserve_write_pipe"; | |||
3626 | else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe) | |||
3627 | Name = "__work_group_reserve_read_pipe"; | |||
3628 | else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe) | |||
3629 | Name = "__work_group_reserve_write_pipe"; | |||
3630 | else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe) | |||
3631 | Name = "__sub_group_reserve_read_pipe"; | |||
3632 | else | |||
3633 | Name = "__sub_group_reserve_write_pipe"; | |||
3634 | ||||
3635 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), | |||
3636 | *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
3637 | llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy); | |||
3638 | CGOpenCLRuntime OpenCLRT(CGM); | |||
3639 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); | |||
3640 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); | |||
3641 | ||||
3642 | // Building the generic function prototype. | |||
3643 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty}; | |||
3644 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
3645 | ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3646 | // We know the second argument is an integer type, but we may need to cast | |||
3647 | // it to i32. | |||
3648 | if (Arg1->getType() != Int32Ty) | |||
3649 | Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty); | |||
3650 | return RValue::get( | |||
3651 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
3652 | {Arg0, Arg1, PacketSize, PacketAlign})); | |||
3653 | } | |||
3654 | // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write | |||
3655 | // functions | |||
3656 | case Builtin::BIcommit_read_pipe: | |||
3657 | case Builtin::BIcommit_write_pipe: | |||
3658 | case Builtin::BIwork_group_commit_read_pipe: | |||
3659 | case Builtin::BIwork_group_commit_write_pipe: | |||
3660 | case Builtin::BIsub_group_commit_read_pipe: | |||
3661 | case Builtin::BIsub_group_commit_write_pipe: { | |||
3662 | const char *Name; | |||
3663 | if (BuiltinID == Builtin::BIcommit_read_pipe) | |||
3664 | Name = "__commit_read_pipe"; | |||
3665 | else if (BuiltinID == Builtin::BIcommit_write_pipe) | |||
3666 | Name = "__commit_write_pipe"; | |||
3667 | else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe) | |||
3668 | Name = "__work_group_commit_read_pipe"; | |||
3669 | else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe) | |||
3670 | Name = "__work_group_commit_write_pipe"; | |||
3671 | else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe) | |||
3672 | Name = "__sub_group_commit_read_pipe"; | |||
3673 | else | |||
3674 | Name = "__sub_group_commit_write_pipe"; | |||
3675 | ||||
3676 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), | |||
3677 | *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
3678 | CGOpenCLRuntime OpenCLRT(CGM); | |||
3679 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); | |||
3680 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); | |||
3681 | ||||
3682 | // Building the generic function prototype. | |||
3683 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty}; | |||
3684 | llvm::FunctionType *FTy = | |||
3685 | llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()), | |||
3686 | llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3687 | ||||
3688 | return RValue::get( | |||
3689 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
3690 | {Arg0, Arg1, PacketSize, PacketAlign})); | |||
3691 | } | |||
3692 | // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions | |||
3693 | case Builtin::BIget_pipe_num_packets: | |||
3694 | case Builtin::BIget_pipe_max_packets: { | |||
3695 | const char *BaseName; | |||
3696 | const PipeType *PipeTy = E->getArg(0)->getType()->getAs<PipeType>(); | |||
3697 | if (BuiltinID == Builtin::BIget_pipe_num_packets) | |||
3698 | BaseName = "__get_pipe_num_packets"; | |||
3699 | else | |||
3700 | BaseName = "__get_pipe_max_packets"; | |||
3701 | auto Name = std::string(BaseName) + | |||
3702 | std::string(PipeTy->isReadOnly() ? "_ro" : "_wo"); | |||
3703 | ||||
3704 | // Building the generic function prototype. | |||
3705 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); | |||
3706 | CGOpenCLRuntime OpenCLRT(CGM); | |||
3707 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); | |||
3708 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); | |||
3709 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty}; | |||
3710 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
3711 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3712 | ||||
3713 | return RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
3714 | {Arg0, PacketSize, PacketAlign})); | |||
3715 | } | |||
3716 | ||||
3717 | // OpenCL v2.0 s6.13.9 - Address space qualifier functions. | |||
3718 | case Builtin::BIto_global: | |||
3719 | case Builtin::BIto_local: | |||
3720 | case Builtin::BIto_private: { | |||
3721 | auto Arg0 = EmitScalarExpr(E->getArg(0)); | |||
3722 | auto NewArgT = llvm::PointerType::get(Int8Ty, | |||
3723 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
3724 | auto NewRetT = llvm::PointerType::get(Int8Ty, | |||
3725 | CGM.getContext().getTargetAddressSpace( | |||
3726 | E->getType()->getPointeeType().getAddressSpace())); | |||
3727 | auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false); | |||
3728 | llvm::Value *NewArg; | |||
3729 | if (Arg0->getType()->getPointerAddressSpace() != | |||
3730 | NewArgT->getPointerAddressSpace()) | |||
3731 | NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT); | |||
3732 | else | |||
3733 | NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT); | |||
3734 | auto NewName = std::string("__") + E->getDirectCallee()->getName().str(); | |||
3735 | auto NewCall = | |||
3736 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg}); | |||
3737 | return RValue::get(Builder.CreateBitOrPointerCast(NewCall, | |||
3738 | ConvertType(E->getType()))); | |||
3739 | } | |||
3740 | ||||
3741 | // OpenCL v2.0, s6.13.17 - Enqueue kernel function. | |||
3742 | // It contains four different overload formats specified in Table 6.13.17.1. | |||
3743 | case Builtin::BIenqueue_kernel: { | |||
3744 | StringRef Name; // Generated function call name | |||
3745 | unsigned NumArgs = E->getNumArgs(); | |||
3746 | ||||
3747 | llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy); | |||
3748 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( | |||
3749 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
3750 | ||||
3751 | llvm::Value *Queue = EmitScalarExpr(E->getArg(0)); | |||
3752 | llvm::Value *Flags = EmitScalarExpr(E->getArg(1)); | |||
3753 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(2)); | |||
3754 | llvm::Value *Range = NDRangeL.getAddress().getPointer(); | |||
3755 | llvm::Type *RangeTy = NDRangeL.getAddress().getType(); | |||
3756 | ||||
3757 | if (NumArgs == 4) { | |||
3758 | // The most basic form of the call with parameters: | |||
3759 | // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void) | |||
3760 | Name = "__enqueue_kernel_basic"; | |||
3761 | llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy, | |||
3762 | GenericVoidPtrTy}; | |||
3763 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
3764 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3765 | ||||
3766 | auto Info = | |||
3767 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); | |||
3768 | llvm::Value *Kernel = | |||
3769 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
3770 | llvm::Value *Block = | |||
3771 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
3772 | ||||
3773 | AttrBuilder B; | |||
3774 | B.addByValAttr(NDRangeL.getAddress().getElementType()); | |||
3775 | llvm::AttributeList ByValAttrSet = | |||
3776 | llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B); | |||
3777 | ||||
3778 | auto RTCall = | |||
3779 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet), | |||
3780 | {Queue, Flags, Range, Kernel, Block}); | |||
3781 | RTCall->setAttributes(ByValAttrSet); | |||
3782 | return RValue::get(RTCall); | |||
3783 | } | |||
3784 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3784, __PRETTY_FUNCTION__)); | |||
3785 | ||||
3786 | // Create a temporary array to hold the sizes of local pointer arguments | |||
3787 | // for the block. \p First is the position of the first size argument. | |||
3788 | auto CreateArrayForSizeVar = [=](unsigned First) | |||
3789 | -> std::tuple<llvm::Value *, llvm::Value *, llvm::Value *> { | |||
3790 | llvm::APInt ArraySize(32, NumArgs - First); | |||
3791 | QualType SizeArrayTy = getContext().getConstantArrayType( | |||
3792 | getContext().getSizeType(), ArraySize, nullptr, ArrayType::Normal, | |||
3793 | /*IndexTypeQuals=*/0); | |||
3794 | auto Tmp = CreateMemTemp(SizeArrayTy, "block_sizes"); | |||
3795 | llvm::Value *TmpPtr = Tmp.getPointer(); | |||
3796 | llvm::Value *TmpSize = EmitLifetimeStart( | |||
3797 | CGM.getDataLayout().getTypeAllocSize(Tmp.getElementType()), TmpPtr); | |||
3798 | llvm::Value *ElemPtr; | |||
3799 | // Each of the following arguments specifies the size of the corresponding | |||
3800 | // argument passed to the enqueued block. | |||
3801 | auto *Zero = llvm::ConstantInt::get(IntTy, 0); | |||
3802 | for (unsigned I = First; I < NumArgs; ++I) { | |||
3803 | auto *Index = llvm::ConstantInt::get(IntTy, I - First); | |||
3804 | auto *GEP = Builder.CreateGEP(TmpPtr, {Zero, Index}); | |||
3805 | if (I == First) | |||
3806 | ElemPtr = GEP; | |||
3807 | auto *V = | |||
3808 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy); | |||
3809 | Builder.CreateAlignedStore( | |||
3810 | V, GEP, CGM.getDataLayout().getPrefTypeAlignment(SizeTy)); | |||
3811 | } | |||
3812 | return std::tie(ElemPtr, TmpSize, TmpPtr); | |||
3813 | }; | |||
3814 | ||||
3815 | // Could have events and/or varargs. | |||
3816 | if (E->getArg(3)->getType()->isBlockPointerType()) { | |||
3817 | // No events passed, but has variadic arguments. | |||
3818 | Name = "__enqueue_kernel_varargs"; | |||
3819 | auto Info = | |||
3820 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); | |||
3821 | llvm::Value *Kernel = | |||
3822 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
3823 | auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
3824 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; | |||
3825 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(4); | |||
3826 | ||||
3827 | // Create a vector of the arguments, as well as a constant value to | |||
3828 | // express to the runtime the number of variadic arguments. | |||
3829 | std::vector<llvm::Value *> Args = { | |||
3830 | Queue, Flags, Range, | |||
3831 | Kernel, Block, ConstantInt::get(IntTy, NumArgs - 4), | |||
3832 | ElemPtr}; | |||
3833 | std::vector<llvm::Type *> ArgTys = { | |||
3834 | QueueTy, IntTy, RangeTy, GenericVoidPtrTy, | |||
3835 | GenericVoidPtrTy, IntTy, ElemPtr->getType()}; | |||
3836 | ||||
3837 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
3838 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3839 | auto Call = | |||
3840 | RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
3841 | llvm::ArrayRef<llvm::Value *>(Args))); | |||
3842 | if (TmpSize) | |||
3843 | EmitLifetimeEnd(TmpSize, TmpPtr); | |||
3844 | return Call; | |||
3845 | } | |||
3846 | // Any calls now have event arguments passed. | |||
3847 | if (NumArgs >= 7) { | |||
3848 | llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy); | |||
3849 | llvm::PointerType *EventPtrTy = EventTy->getPointerTo( | |||
3850 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
3851 | ||||
3852 | llvm::Value *NumEvents = | |||
3853 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty); | |||
3854 | ||||
3855 | // Since SemaOpenCLBuiltinEnqueueKernel allows fifth and sixth arguments | |||
3856 | // to be a null pointer constant (including `0` literal), we can take it | |||
3857 | // into account and emit null pointer directly. | |||
3858 | llvm::Value *EventWaitList = nullptr; | |||
3859 | if (E->getArg(4)->isNullPointerConstant( | |||
3860 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { | |||
3861 | EventWaitList = llvm::ConstantPointerNull::get(EventPtrTy); | |||
3862 | } else { | |||
3863 | EventWaitList = E->getArg(4)->getType()->isArrayType() | |||
3864 | ? EmitArrayToPointerDecay(E->getArg(4)).getPointer() | |||
3865 | : EmitScalarExpr(E->getArg(4)); | |||
3866 | // Convert to generic address space. | |||
3867 | EventWaitList = Builder.CreatePointerCast(EventWaitList, EventPtrTy); | |||
3868 | } | |||
3869 | llvm::Value *EventRet = nullptr; | |||
3870 | if (E->getArg(5)->isNullPointerConstant( | |||
3871 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { | |||
3872 | EventRet = llvm::ConstantPointerNull::get(EventPtrTy); | |||
3873 | } else { | |||
3874 | EventRet = | |||
3875 | Builder.CreatePointerCast(EmitScalarExpr(E->getArg(5)), EventPtrTy); | |||
3876 | } | |||
3877 | ||||
3878 | auto Info = | |||
3879 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6)); | |||
3880 | llvm::Value *Kernel = | |||
3881 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
3882 | llvm::Value *Block = | |||
3883 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
3884 | ||||
3885 | std::vector<llvm::Type *> ArgTys = { | |||
3886 | QueueTy, Int32Ty, RangeTy, Int32Ty, | |||
3887 | EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy}; | |||
3888 | ||||
3889 | std::vector<llvm::Value *> Args = {Queue, Flags, Range, | |||
3890 | NumEvents, EventWaitList, EventRet, | |||
3891 | Kernel, Block}; | |||
3892 | ||||
3893 | if (NumArgs == 7) { | |||
3894 | // Has events but no variadics. | |||
3895 | Name = "__enqueue_kernel_basic_events"; | |||
3896 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
3897 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3898 | return RValue::get( | |||
3899 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
3900 | llvm::ArrayRef<llvm::Value *>(Args))); | |||
3901 | } | |||
3902 | // Has event info and variadics | |||
3903 | // Pass the number of variadics to the runtime function too. | |||
3904 | Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7)); | |||
3905 | ArgTys.push_back(Int32Ty); | |||
3906 | Name = "__enqueue_kernel_events_varargs"; | |||
3907 | ||||
3908 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; | |||
3909 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(7); | |||
3910 | Args.push_back(ElemPtr); | |||
3911 | ArgTys.push_back(ElemPtr->getType()); | |||
3912 | ||||
3913 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
3914 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
3915 | auto Call = | |||
3916 | RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
3917 | llvm::ArrayRef<llvm::Value *>(Args))); | |||
3918 | if (TmpSize) | |||
3919 | EmitLifetimeEnd(TmpSize, TmpPtr); | |||
3920 | return Call; | |||
3921 | } | |||
3922 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
3923 | } | |||
3924 | // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block | |||
3925 | // parameter. | |||
3926 | case Builtin::BIget_kernel_work_group_size: { | |||
3927 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( | |||
3928 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
3929 | auto Info = | |||
3930 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); | |||
3931 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
3932 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
3933 | return RValue::get(Builder.CreateCall( | |||
3934 | CGM.CreateRuntimeFunction( | |||
3935 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, | |||
3936 | false), | |||
3937 | "__get_kernel_work_group_size_impl"), | |||
3938 | {Kernel, Arg})); | |||
3939 | } | |||
3940 | case Builtin::BIget_kernel_preferred_work_group_size_multiple: { | |||
3941 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( | |||
3942 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
3943 | auto Info = | |||
3944 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); | |||
3945 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
3946 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
3947 | return RValue::get(Builder.CreateCall( | |||
3948 | CGM.CreateRuntimeFunction( | |||
3949 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, | |||
3950 | false), | |||
3951 | "__get_kernel_preferred_work_group_size_multiple_impl"), | |||
3952 | {Kernel, Arg})); | |||
3953 | } | |||
3954 | case Builtin::BIget_kernel_max_sub_group_size_for_ndrange: | |||
3955 | case Builtin::BIget_kernel_sub_group_count_for_ndrange: { | |||
3956 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( | |||
3957 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
3958 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(0)); | |||
3959 | llvm::Value *NDRange = NDRangeL.getAddress().getPointer(); | |||
3960 | auto Info = | |||
3961 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1)); | |||
3962 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
3963 | Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
3964 | const char *Name = | |||
3965 | BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange | |||
3966 | ? "__get_kernel_max_sub_group_size_for_ndrange_impl" | |||
3967 | : "__get_kernel_sub_group_count_for_ndrange_impl"; | |||
3968 | return RValue::get(Builder.CreateCall( | |||
3969 | CGM.CreateRuntimeFunction( | |||
3970 | llvm::FunctionType::get( | |||
3971 | IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy}, | |||
3972 | false), | |||
3973 | Name), | |||
3974 | {NDRange, Kernel, Block})); | |||
3975 | } | |||
3976 | ||||
3977 | case Builtin::BI__builtin_store_half: | |||
3978 | case Builtin::BI__builtin_store_halff: { | |||
3979 | Value *Val = EmitScalarExpr(E->getArg(0)); | |||
3980 | Address Address = EmitPointerWithAlignment(E->getArg(1)); | |||
3981 | Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy()); | |||
3982 | return RValue::get(Builder.CreateStore(HalfVal, Address)); | |||
3983 | } | |||
3984 | case Builtin::BI__builtin_load_half: { | |||
3985 | Address Address = EmitPointerWithAlignment(E->getArg(0)); | |||
3986 | Value *HalfVal = Builder.CreateLoad(Address); | |||
3987 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy())); | |||
3988 | } | |||
3989 | case Builtin::BI__builtin_load_halff: { | |||
3990 | Address Address = EmitPointerWithAlignment(E->getArg(0)); | |||
3991 | Value *HalfVal = Builder.CreateLoad(Address); | |||
3992 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy())); | |||
3993 | } | |||
3994 | case Builtin::BIprintf: | |||
3995 | if (getTarget().getTriple().isNVPTX()) | |||
3996 | return EmitNVPTXDevicePrintfCallExpr(E, ReturnValue); | |||
3997 | break; | |||
3998 | case Builtin::BI__builtin_canonicalize: | |||
3999 | case Builtin::BI__builtin_canonicalizef: | |||
4000 | case Builtin::BI__builtin_canonicalizef16: | |||
4001 | case Builtin::BI__builtin_canonicalizel: | |||
4002 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize)); | |||
4003 | ||||
4004 | case Builtin::BI__builtin_thread_pointer: { | |||
4005 | if (!getContext().getTargetInfo().isTLSSupported()) | |||
4006 | CGM.ErrorUnsupported(E, "__builtin_thread_pointer"); | |||
4007 | // Fall through - it's already mapped to the intrinsic by GCCBuiltin. | |||
4008 | break; | |||
4009 | } | |||
4010 | case Builtin::BI__builtin_os_log_format: | |||
4011 | return emitBuiltinOSLogFormat(*E); | |||
4012 | ||||
4013 | case Builtin::BI__xray_customevent: { | |||
4014 | if (!ShouldXRayInstrumentFunction()) | |||
4015 | return RValue::getIgnored(); | |||
4016 | ||||
4017 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( | |||
4018 | XRayInstrKind::Custom)) | |||
4019 | return RValue::getIgnored(); | |||
4020 | ||||
4021 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) | |||
4022 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents()) | |||
4023 | return RValue::getIgnored(); | |||
4024 | ||||
4025 | Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent); | |||
4026 | auto FTy = F->getFunctionType(); | |||
4027 | auto Arg0 = E->getArg(0); | |||
4028 | auto Arg0Val = EmitScalarExpr(Arg0); | |||
4029 | auto Arg0Ty = Arg0->getType(); | |||
4030 | auto PTy0 = FTy->getParamType(0); | |||
4031 | if (PTy0 != Arg0Val->getType()) { | |||
4032 | if (Arg0Ty->isArrayType()) | |||
4033 | Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer(); | |||
4034 | else | |||
4035 | Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0); | |||
4036 | } | |||
4037 | auto Arg1 = EmitScalarExpr(E->getArg(1)); | |||
4038 | auto PTy1 = FTy->getParamType(1); | |||
4039 | if (PTy1 != Arg1->getType()) | |||
4040 | Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1); | |||
4041 | return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1})); | |||
4042 | } | |||
4043 | ||||
4044 | case Builtin::BI__xray_typedevent: { | |||
4045 | // TODO: There should be a way to always emit events even if the current | |||
4046 | // function is not instrumented. Losing events in a stream can cripple | |||
4047 | // a trace. | |||
4048 | if (!ShouldXRayInstrumentFunction()) | |||
4049 | return RValue::getIgnored(); | |||
4050 | ||||
4051 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( | |||
4052 | XRayInstrKind::Typed)) | |||
4053 | return RValue::getIgnored(); | |||
4054 | ||||
4055 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) | |||
4056 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents()) | |||
4057 | return RValue::getIgnored(); | |||
4058 | ||||
4059 | Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent); | |||
4060 | auto FTy = F->getFunctionType(); | |||
4061 | auto Arg0 = EmitScalarExpr(E->getArg(0)); | |||
4062 | auto PTy0 = FTy->getParamType(0); | |||
4063 | if (PTy0 != Arg0->getType()) | |||
4064 | Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0); | |||
4065 | auto Arg1 = E->getArg(1); | |||
4066 | auto Arg1Val = EmitScalarExpr(Arg1); | |||
4067 | auto Arg1Ty = Arg1->getType(); | |||
4068 | auto PTy1 = FTy->getParamType(1); | |||
4069 | if (PTy1 != Arg1Val->getType()) { | |||
4070 | if (Arg1Ty->isArrayType()) | |||
4071 | Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer(); | |||
4072 | else | |||
4073 | Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1); | |||
4074 | } | |||
4075 | auto Arg2 = EmitScalarExpr(E->getArg(2)); | |||
4076 | auto PTy2 = FTy->getParamType(2); | |||
4077 | if (PTy2 != Arg2->getType()) | |||
4078 | Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2); | |||
4079 | return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2})); | |||
4080 | } | |||
4081 | ||||
4082 | case Builtin::BI__builtin_ms_va_start: | |||
4083 | case Builtin::BI__builtin_ms_va_end: | |||
4084 | return RValue::get( | |||
4085 | EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(), | |||
4086 | BuiltinID == Builtin::BI__builtin_ms_va_start)); | |||
4087 | ||||
4088 | case Builtin::BI__builtin_ms_va_copy: { | |||
4089 | // Lower this manually. We can't reliably determine whether or not any | |||
4090 | // given va_copy() is for a Win64 va_list from the calling convention | |||
4091 | // alone, because it's legal to do this from a System V ABI function. | |||
4092 | // With opaque pointer types, we won't have enough information in LLVM | |||
4093 | // IR to determine this from the argument types, either. Best to do it | |||
4094 | // now, while we have enough information. | |||
4095 | Address DestAddr = EmitMSVAListRef(E->getArg(0)); | |||
4096 | Address SrcAddr = EmitMSVAListRef(E->getArg(1)); | |||
4097 | ||||
4098 | llvm::Type *BPP = Int8PtrPtrTy; | |||
4099 | ||||
4100 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"), | |||
4101 | DestAddr.getAlignment()); | |||
4102 | SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"), | |||
4103 | SrcAddr.getAlignment()); | |||
4104 | ||||
4105 | Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val"); | |||
4106 | return RValue::get(Builder.CreateStore(ArgPtr, DestAddr)); | |||
4107 | } | |||
4108 | } | |||
4109 | ||||
4110 | // If this is an alias for a lib function (e.g. __builtin_sin), emit | |||
4111 | // the call using the normal call path, but using the unmangled | |||
4112 | // version of the function name. | |||
4113 | if (getContext().BuiltinInfo.isLibFunction(BuiltinID)) | |||
4114 | return emitLibraryCall(*this, FD, E, | |||
4115 | CGM.getBuiltinLibFunction(FD, BuiltinID)); | |||
4116 | ||||
4117 | // If this is a predefined lib function (e.g. malloc), emit the call | |||
4118 | // using exactly the normal call path. | |||
4119 | if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID)) | |||
4120 | return emitLibraryCall(*this, FD, E, | |||
4121 | cast<llvm::Constant>(EmitScalarExpr(E->getCallee()))); | |||
4122 | ||||
4123 | // Check that a call to a target specific builtin has the correct target | |||
4124 | // features. | |||
4125 | // This is down here to avoid non-target specific builtins, however, if | |||
4126 | // generic builtins start to require generic target features then we | |||
4127 | // can move this up to the beginning of the function. | |||
4128 | checkTargetFeatures(E, FD); | |||
4129 | ||||
4130 | if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID)) | |||
4131 | LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth); | |||
4132 | ||||
4133 | // See if we have a target specific intrinsic. | |||
4134 | const char *Name = getContext().BuiltinInfo.getName(BuiltinID); | |||
4135 | Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic; | |||
4136 | StringRef Prefix = | |||
4137 | llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch()); | |||
4138 | if (!Prefix.empty()) { | |||
4139 | IntrinsicID = Intrinsic::getIntrinsicForGCCBuiltin(Prefix.data(), Name); | |||
4140 | // NOTE we don't need to perform a compatibility flag check here since the | |||
4141 | // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the | |||
4142 | // MS builtins via ALL_MS_LANGUAGES and are filtered earlier. | |||
4143 | if (IntrinsicID == Intrinsic::not_intrinsic) | |||
4144 | IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name); | |||
4145 | } | |||
4146 | ||||
4147 | if (IntrinsicID != Intrinsic::not_intrinsic) { | |||
4148 | SmallVector<Value*, 16> Args; | |||
4149 | ||||
4150 | // Find out if any arguments are required to be integer constant | |||
4151 | // expressions. | |||
4152 | unsigned ICEArguments = 0; | |||
4153 | ASTContext::GetBuiltinTypeError Error; | |||
4154 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
4155 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4155, __PRETTY_FUNCTION__)); | |||
4156 | ||||
4157 | Function *F = CGM.getIntrinsic(IntrinsicID); | |||
4158 | llvm::FunctionType *FTy = F->getFunctionType(); | |||
4159 | ||||
4160 | for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) { | |||
4161 | Value *ArgValue; | |||
4162 | // If this is a normal argument, just emit it as a scalar. | |||
4163 | if ((ICEArguments & (1 << i)) == 0) { | |||
4164 | ArgValue = EmitScalarExpr(E->getArg(i)); | |||
4165 | } else { | |||
4166 | // If this is required to be a constant, constant fold it so that we | |||
4167 | // know that the generated intrinsic gets a ConstantInt. | |||
4168 | llvm::APSInt Result; | |||
4169 | bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result,getContext()); | |||
4170 | assert(IsConst && "Constant arg isn't actually constant?")((IsConst && "Constant arg isn't actually constant?") ? static_cast<void> (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4170, __PRETTY_FUNCTION__)); | |||
4171 | (void)IsConst; | |||
4172 | ArgValue = llvm::ConstantInt::get(getLLVMContext(), Result); | |||
4173 | } | |||
4174 | ||||
4175 | // If the intrinsic arg type is different from the builtin arg type | |||
4176 | // we need to do a bit cast. | |||
4177 | llvm::Type *PTy = FTy->getParamType(i); | |||
4178 | if (PTy != ArgValue->getType()) { | |||
4179 | // XXX - vector of pointers? | |||
4180 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(PTy)) { | |||
4181 | if (PtrTy->getAddressSpace() != | |||
4182 | ArgValue->getType()->getPointerAddressSpace()) { | |||
4183 | ArgValue = Builder.CreateAddrSpaceCast( | |||
4184 | ArgValue, | |||
4185 | ArgValue->getType()->getPointerTo(PtrTy->getAddressSpace())); | |||
4186 | } | |||
4187 | } | |||
4188 | ||||
4189 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4190, __PRETTY_FUNCTION__)) | |||
4190 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4190, __PRETTY_FUNCTION__)); | |||
4191 | ArgValue = Builder.CreateBitCast(ArgValue, PTy); | |||
4192 | } | |||
4193 | ||||
4194 | Args.push_back(ArgValue); | |||
4195 | } | |||
4196 | ||||
4197 | Value *V = Builder.CreateCall(F, Args); | |||
4198 | QualType BuiltinRetType = E->getType(); | |||
4199 | ||||
4200 | llvm::Type *RetTy = VoidTy; | |||
4201 | if (!BuiltinRetType->isVoidType()) | |||
4202 | RetTy = ConvertType(BuiltinRetType); | |||
4203 | ||||
4204 | if (RetTy != V->getType()) { | |||
4205 | // XXX - vector of pointers? | |||
4206 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(RetTy)) { | |||
4207 | if (PtrTy->getAddressSpace() != V->getType()->getPointerAddressSpace()) { | |||
4208 | V = Builder.CreateAddrSpaceCast( | |||
4209 | V, V->getType()->getPointerTo(PtrTy->getAddressSpace())); | |||
4210 | } | |||
4211 | } | |||
4212 | ||||
4213 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4214, __PRETTY_FUNCTION__)) | |||
4214 | "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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4214, __PRETTY_FUNCTION__)); | |||
4215 | V = Builder.CreateBitCast(V, RetTy); | |||
4216 | } | |||
4217 | ||||
4218 | return RValue::get(V); | |||
4219 | } | |||
4220 | ||||
4221 | // See if we have a target specific builtin that needs to be lowered. | |||
4222 | if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E)) | |||
4223 | return RValue::get(V); | |||
4224 | ||||
4225 | ErrorUnsupported(E, "builtin function"); | |||
4226 | ||||
4227 | // Unknown builtin, for now just dump it out and return undef. | |||
4228 | return GetUndefRValue(E->getType()); | |||
4229 | } | |||
4230 | ||||
4231 | static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF, | |||
4232 | unsigned BuiltinID, const CallExpr *E, | |||
4233 | llvm::Triple::ArchType Arch) { | |||
4234 | switch (Arch) { | |||
4235 | case llvm::Triple::arm: | |||
4236 | case llvm::Triple::armeb: | |||
4237 | case llvm::Triple::thumb: | |||
4238 | case llvm::Triple::thumbeb: | |||
4239 | return CGF->EmitARMBuiltinExpr(BuiltinID, E, Arch); | |||
4240 | case llvm::Triple::aarch64: | |||
4241 | case llvm::Triple::aarch64_be: | |||
4242 | return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch); | |||
4243 | case llvm::Triple::bpfeb: | |||
4244 | case llvm::Triple::bpfel: | |||
4245 | return CGF->EmitBPFBuiltinExpr(BuiltinID, E); | |||
4246 | case llvm::Triple::x86: | |||
4247 | case llvm::Triple::x86_64: | |||
4248 | return CGF->EmitX86BuiltinExpr(BuiltinID, E); | |||
4249 | case llvm::Triple::ppc: | |||
4250 | case llvm::Triple::ppc64: | |||
4251 | case llvm::Triple::ppc64le: | |||
4252 | return CGF->EmitPPCBuiltinExpr(BuiltinID, E); | |||
4253 | case llvm::Triple::r600: | |||
4254 | case llvm::Triple::amdgcn: | |||
4255 | return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E); | |||
4256 | case llvm::Triple::systemz: | |||
4257 | return CGF->EmitSystemZBuiltinExpr(BuiltinID, E); | |||
4258 | case llvm::Triple::nvptx: | |||
4259 | case llvm::Triple::nvptx64: | |||
4260 | return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E); | |||
4261 | case llvm::Triple::wasm32: | |||
4262 | case llvm::Triple::wasm64: | |||
4263 | return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E); | |||
4264 | case llvm::Triple::hexagon: | |||
4265 | return CGF->EmitHexagonBuiltinExpr(BuiltinID, E); | |||
4266 | default: | |||
4267 | return nullptr; | |||
4268 | } | |||
4269 | } | |||
4270 | ||||
4271 | Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID, | |||
4272 | const CallExpr *E) { | |||
4273 | if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) { | |||
4274 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4274, __PRETTY_FUNCTION__)); | |||
4275 | return EmitTargetArchBuiltinExpr( | |||
4276 | this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E, | |||
4277 | getContext().getAuxTargetInfo()->getTriple().getArch()); | |||
4278 | } | |||
4279 | ||||
4280 | return EmitTargetArchBuiltinExpr(this, BuiltinID, E, | |||
4281 | getTarget().getTriple().getArch()); | |||
4282 | } | |||
4283 | ||||
4284 | static llvm::VectorType *GetNeonType(CodeGenFunction *CGF, | |||
4285 | NeonTypeFlags TypeFlags, | |||
4286 | bool HasLegalHalfType=true, | |||
4287 | bool V1Ty=false) { | |||
4288 | int IsQuad = TypeFlags.isQuad(); | |||
4289 | switch (TypeFlags.getEltType()) { | |||
4290 | case NeonTypeFlags::Int8: | |||
4291 | case NeonTypeFlags::Poly8: | |||
4292 | return llvm::VectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad)); | |||
4293 | case NeonTypeFlags::Int16: | |||
4294 | case NeonTypeFlags::Poly16: | |||
4295 | return llvm::VectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); | |||
4296 | case NeonTypeFlags::Float16: | |||
4297 | if (HasLegalHalfType) | |||
4298 | return llvm::VectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad)); | |||
4299 | else | |||
4300 | return llvm::VectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); | |||
4301 | case NeonTypeFlags::Int32: | |||
4302 | return llvm::VectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad)); | |||
4303 | case NeonTypeFlags::Int64: | |||
4304 | case NeonTypeFlags::Poly64: | |||
4305 | return llvm::VectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad)); | |||
4306 | case NeonTypeFlags::Poly128: | |||
4307 | // FIXME: i128 and f128 doesn't get fully support in Clang and llvm. | |||
4308 | // There is a lot of i128 and f128 API missing. | |||
4309 | // so we use v16i8 to represent poly128 and get pattern matched. | |||
4310 | return llvm::VectorType::get(CGF->Int8Ty, 16); | |||
4311 | case NeonTypeFlags::Float32: | |||
4312 | return llvm::VectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad)); | |||
4313 | case NeonTypeFlags::Float64: | |||
4314 | return llvm::VectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad)); | |||
4315 | } | |||
4316 | llvm_unreachable("Unknown vector element type!")::llvm::llvm_unreachable_internal("Unknown vector element type!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4316); | |||
4317 | } | |||
4318 | ||||
4319 | static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF, | |||
4320 | NeonTypeFlags IntTypeFlags) { | |||
4321 | int IsQuad = IntTypeFlags.isQuad(); | |||
4322 | switch (IntTypeFlags.getEltType()) { | |||
4323 | case NeonTypeFlags::Int16: | |||
4324 | return llvm::VectorType::get(CGF->HalfTy, (4 << IsQuad)); | |||
4325 | case NeonTypeFlags::Int32: | |||
4326 | return llvm::VectorType::get(CGF->FloatTy, (2 << IsQuad)); | |||
4327 | case NeonTypeFlags::Int64: | |||
4328 | return llvm::VectorType::get(CGF->DoubleTy, (1 << IsQuad)); | |||
4329 | default: | |||
4330 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4330); | |||
4331 | } | |||
4332 | } | |||
4333 | ||||
4334 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) { | |||
4335 | unsigned nElts = V->getType()->getVectorNumElements(); | |||
4336 | Value* SV = llvm::ConstantVector::getSplat(nElts, C); | |||
4337 | return Builder.CreateShuffleVector(V, V, SV, "lane"); | |||
4338 | } | |||
4339 | ||||
4340 | Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops, | |||
4341 | const char *name, | |||
4342 | unsigned shift, bool rightshift) { | |||
4343 | unsigned j = 0; | |||
4344 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); | |||
4345 | ai != ae; ++ai, ++j) | |||
4346 | if (shift > 0 && shift == j) | |||
4347 | Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift); | |||
4348 | else | |||
4349 | Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name); | |||
4350 | ||||
4351 | return Builder.CreateCall(F, Ops, name); | |||
4352 | } | |||
4353 | ||||
4354 | Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty, | |||
4355 | bool neg) { | |||
4356 | int SV = cast<ConstantInt>(V)->getSExtValue(); | |||
4357 | return ConstantInt::get(Ty, neg ? -SV : SV); | |||
4358 | } | |||
4359 | ||||
4360 | // Right-shift a vector by a constant. | |||
4361 | Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift, | |||
4362 | llvm::Type *Ty, bool usgn, | |||
4363 | const char *name) { | |||
4364 | llvm::VectorType *VTy = cast<llvm::VectorType>(Ty); | |||
4365 | ||||
4366 | int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue(); | |||
4367 | int EltSize = VTy->getScalarSizeInBits(); | |||
4368 | ||||
4369 | Vec = Builder.CreateBitCast(Vec, Ty); | |||
4370 | ||||
4371 | // lshr/ashr are undefined when the shift amount is equal to the vector | |||
4372 | // element size. | |||
4373 | if (ShiftAmt == EltSize) { | |||
4374 | if (usgn) { | |||
4375 | // Right-shifting an unsigned value by its size yields 0. | |||
4376 | return llvm::ConstantAggregateZero::get(VTy); | |||
4377 | } else { | |||
4378 | // Right-shifting a signed value by its size is equivalent | |||
4379 | // to a shift of size-1. | |||
4380 | --ShiftAmt; | |||
4381 | Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt); | |||
4382 | } | |||
4383 | } | |||
4384 | ||||
4385 | Shift = EmitNeonShiftVector(Shift, Ty, false); | |||
4386 | if (usgn) | |||
4387 | return Builder.CreateLShr(Vec, Shift, name); | |||
4388 | else | |||
4389 | return Builder.CreateAShr(Vec, Shift, name); | |||
4390 | } | |||
4391 | ||||
4392 | enum { | |||
4393 | AddRetType = (1 << 0), | |||
4394 | Add1ArgType = (1 << 1), | |||
4395 | Add2ArgTypes = (1 << 2), | |||
4396 | ||||
4397 | VectorizeRetType = (1 << 3), | |||
4398 | VectorizeArgTypes = (1 << 4), | |||
4399 | ||||
4400 | InventFloatType = (1 << 5), | |||
4401 | UnsignedAlts = (1 << 6), | |||
4402 | ||||
4403 | Use64BitVectors = (1 << 7), | |||
4404 | Use128BitVectors = (1 << 8), | |||
4405 | ||||
4406 | Vectorize1ArgType = Add1ArgType | VectorizeArgTypes, | |||
4407 | VectorRet = AddRetType | VectorizeRetType, | |||
4408 | VectorRetGetArgs01 = | |||
4409 | AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes, | |||
4410 | FpCmpzModifiers = | |||
4411 | AddRetType | VectorizeRetType | Add1ArgType | InventFloatType | |||
4412 | }; | |||
4413 | ||||
4414 | namespace { | |||
4415 | struct NeonIntrinsicInfo { | |||
4416 | const char *NameHint; | |||
4417 | unsigned BuiltinID; | |||
4418 | unsigned LLVMIntrinsic; | |||
4419 | unsigned AltLLVMIntrinsic; | |||
4420 | unsigned TypeModifier; | |||
4421 | ||||
4422 | bool operator<(unsigned RHSBuiltinID) const { | |||
4423 | return BuiltinID < RHSBuiltinID; | |||
4424 | } | |||
4425 | bool operator<(const NeonIntrinsicInfo &TE) const { | |||
4426 | return BuiltinID < TE.BuiltinID; | |||
4427 | } | |||
4428 | }; | |||
4429 | } // end anonymous namespace | |||
4430 | ||||
4431 | #define NEONMAP0(NameBase) \ | |||
4432 | { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 } | |||
4433 | ||||
4434 | #define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \ | |||
4435 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ | |||
4436 | Intrinsic::LLVMIntrinsic, 0, TypeModifier } | |||
4437 | ||||
4438 | #define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \ | |||
4439 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ | |||
4440 | Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \ | |||
4441 | TypeModifier } | |||
4442 | ||||
4443 | static const NeonIntrinsicInfo ARMSIMDIntrinsicMap [] = { | |||
4444 | NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), | |||
4445 | NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), | |||
4446 | NEONMAP1(vabs_v, arm_neon_vabs, 0), | |||
4447 | NEONMAP1(vabsq_v, arm_neon_vabs, 0), | |||
4448 | NEONMAP0(vaddhn_v), | |||
4449 | NEONMAP1(vaesdq_v, arm_neon_aesd, 0), | |||
4450 | NEONMAP1(vaeseq_v, arm_neon_aese, 0), | |||
4451 | NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0), | |||
4452 | NEONMAP1(vaesmcq_v, arm_neon_aesmc, 0), | |||
4453 | NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType), | |||
4454 | NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType), | |||
4455 | NEONMAP1(vcage_v, arm_neon_vacge, 0), | |||
4456 | NEONMAP1(vcageq_v, arm_neon_vacge, 0), | |||
4457 | NEONMAP1(vcagt_v, arm_neon_vacgt, 0), | |||
4458 | NEONMAP1(vcagtq_v, arm_neon_vacgt, 0), | |||
4459 | NEONMAP1(vcale_v, arm_neon_vacge, 0), | |||
4460 | NEONMAP1(vcaleq_v, arm_neon_vacge, 0), | |||
4461 | NEONMAP1(vcalt_v, arm_neon_vacgt, 0), | |||
4462 | NEONMAP1(vcaltq_v, arm_neon_vacgt, 0), | |||
4463 | NEONMAP0(vceqz_v), | |||
4464 | NEONMAP0(vceqzq_v), | |||
4465 | NEONMAP0(vcgez_v), | |||
4466 | NEONMAP0(vcgezq_v), | |||
4467 | NEONMAP0(vcgtz_v), | |||
4468 | NEONMAP0(vcgtzq_v), | |||
4469 | NEONMAP0(vclez_v), | |||
4470 | NEONMAP0(vclezq_v), | |||
4471 | NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType), | |||
4472 | NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType), | |||
4473 | NEONMAP0(vcltz_v), | |||
4474 | NEONMAP0(vcltzq_v), | |||
4475 | NEONMAP1(vclz_v, ctlz, Add1ArgType), | |||
4476 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), | |||
4477 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), | |||
4478 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), | |||
4479 | NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0), | |||
4480 | NEONMAP0(vcvt_f16_v), | |||
4481 | NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0), | |||
4482 | NEONMAP0(vcvt_f32_v), | |||
4483 | NEONMAP2(vcvt_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), | |||
4484 | NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), | |||
4485 | NEONMAP1(vcvt_n_s16_v, arm_neon_vcvtfp2fxs, 0), | |||
4486 | NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0), | |||
4487 | NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0), | |||
4488 | NEONMAP1(vcvt_n_u16_v, arm_neon_vcvtfp2fxu, 0), | |||
4489 | NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0), | |||
4490 | NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0), | |||
4491 | NEONMAP0(vcvt_s16_v), | |||
4492 | NEONMAP0(vcvt_s32_v), | |||
4493 | NEONMAP0(vcvt_s64_v), | |||
4494 | NEONMAP0(vcvt_u16_v), | |||
4495 | NEONMAP0(vcvt_u32_v), | |||
4496 | NEONMAP0(vcvt_u64_v), | |||
4497 | NEONMAP1(vcvta_s16_v, arm_neon_vcvtas, 0), | |||
4498 | NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0), | |||
4499 | NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0), | |||
4500 | NEONMAP1(vcvta_u16_v, arm_neon_vcvtau, 0), | |||
4501 | NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0), | |||
4502 | NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0), | |||
4503 | NEONMAP1(vcvtaq_s16_v, arm_neon_vcvtas, 0), | |||
4504 | NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0), | |||
4505 | NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0), | |||
4506 | NEONMAP1(vcvtaq_u16_v, arm_neon_vcvtau, 0), | |||
4507 | NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0), | |||
4508 | NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0), | |||
4509 | NEONMAP1(vcvtm_s16_v, arm_neon_vcvtms, 0), | |||
4510 | NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0), | |||
4511 | NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0), | |||
4512 | NEONMAP1(vcvtm_u16_v, arm_neon_vcvtmu, 0), | |||
4513 | NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0), | |||
4514 | NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0), | |||
4515 | NEONMAP1(vcvtmq_s16_v, arm_neon_vcvtms, 0), | |||
4516 | NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0), | |||
4517 | NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0), | |||
4518 | NEONMAP1(vcvtmq_u16_v, arm_neon_vcvtmu, 0), | |||
4519 | NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0), | |||
4520 | NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0), | |||
4521 | NEONMAP1(vcvtn_s16_v, arm_neon_vcvtns, 0), | |||
4522 | NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0), | |||
4523 | NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0), | |||
4524 | NEONMAP1(vcvtn_u16_v, arm_neon_vcvtnu, 0), | |||
4525 | NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0), | |||
4526 | NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0), | |||
4527 | NEONMAP1(vcvtnq_s16_v, arm_neon_vcvtns, 0), | |||
4528 | NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0), | |||
4529 | NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0), | |||
4530 | NEONMAP1(vcvtnq_u16_v, arm_neon_vcvtnu, 0), | |||
4531 | NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0), | |||
4532 | NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0), | |||
4533 | NEONMAP1(vcvtp_s16_v, arm_neon_vcvtps, 0), | |||
4534 | NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0), | |||
4535 | NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0), | |||
4536 | NEONMAP1(vcvtp_u16_v, arm_neon_vcvtpu, 0), | |||
4537 | NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0), | |||
4538 | NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0), | |||
4539 | NEONMAP1(vcvtpq_s16_v, arm_neon_vcvtps, 0), | |||
4540 | NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0), | |||
4541 | NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0), | |||
4542 | NEONMAP1(vcvtpq_u16_v, arm_neon_vcvtpu, 0), | |||
4543 | NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0), | |||
4544 | NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0), | |||
4545 | NEONMAP0(vcvtq_f16_v), | |||
4546 | NEONMAP0(vcvtq_f32_v), | |||
4547 | NEONMAP2(vcvtq_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), | |||
4548 | NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), | |||
4549 | NEONMAP1(vcvtq_n_s16_v, arm_neon_vcvtfp2fxs, 0), | |||
4550 | NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0), | |||
4551 | NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0), | |||
4552 | NEONMAP1(vcvtq_n_u16_v, arm_neon_vcvtfp2fxu, 0), | |||
4553 | NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0), | |||
4554 | NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0), | |||
4555 | NEONMAP0(vcvtq_s16_v), | |||
4556 | NEONMAP0(vcvtq_s32_v), | |||
4557 | NEONMAP0(vcvtq_s64_v), | |||
4558 | NEONMAP0(vcvtq_u16_v), | |||
4559 | NEONMAP0(vcvtq_u32_v), | |||
4560 | NEONMAP0(vcvtq_u64_v), | |||
4561 | NEONMAP2(vdot_v, arm_neon_udot, arm_neon_sdot, 0), | |||
4562 | NEONMAP2(vdotq_v, arm_neon_udot, arm_neon_sdot, 0), | |||
4563 | NEONMAP0(vext_v), | |||
4564 | NEONMAP0(vextq_v), | |||
4565 | NEONMAP0(vfma_v), | |||
4566 | NEONMAP0(vfmaq_v), | |||
4567 | NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), | |||
4568 | NEONMAP2(vhaddq_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), | |||
4569 | NEONMAP2(vhsub_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), | |||
4570 | NEONMAP2(vhsubq_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), | |||
4571 | NEONMAP0(vld1_dup_v), | |||
4572 | NEONMAP1(vld1_v, arm_neon_vld1, 0), | |||
4573 | NEONMAP1(vld1_x2_v, arm_neon_vld1x2, 0), | |||
4574 | NEONMAP1(vld1_x3_v, arm_neon_vld1x3, 0), | |||
4575 | NEONMAP1(vld1_x4_v, arm_neon_vld1x4, 0), | |||
4576 | NEONMAP0(vld1q_dup_v), | |||
4577 | NEONMAP1(vld1q_v, arm_neon_vld1, 0), | |||
4578 | NEONMAP1(vld1q_x2_v, arm_neon_vld1x2, 0), | |||
4579 | NEONMAP1(vld1q_x3_v, arm_neon_vld1x3, 0), | |||
4580 | NEONMAP1(vld1q_x4_v, arm_neon_vld1x4, 0), | |||
4581 | NEONMAP1(vld2_dup_v, arm_neon_vld2dup, 0), | |||
4582 | NEONMAP1(vld2_lane_v, arm_neon_vld2lane, 0), | |||
4583 | NEONMAP1(vld2_v, arm_neon_vld2, 0), | |||
4584 | NEONMAP1(vld2q_dup_v, arm_neon_vld2dup, 0), | |||
4585 | NEONMAP1(vld2q_lane_v, arm_neon_vld2lane, 0), | |||
4586 | NEONMAP1(vld2q_v, arm_neon_vld2, 0), | |||
4587 | NEONMAP1(vld3_dup_v, arm_neon_vld3dup, 0), | |||
4588 | NEONMAP1(vld3_lane_v, arm_neon_vld3lane, 0), | |||
4589 | NEONMAP1(vld3_v, arm_neon_vld3, 0), | |||
4590 | NEONMAP1(vld3q_dup_v, arm_neon_vld3dup, 0), | |||
4591 | NEONMAP1(vld3q_lane_v, arm_neon_vld3lane, 0), | |||
4592 | NEONMAP1(vld3q_v, arm_neon_vld3, 0), | |||
4593 | NEONMAP1(vld4_dup_v, arm_neon_vld4dup, 0), | |||
4594 | NEONMAP1(vld4_lane_v, arm_neon_vld4lane, 0), | |||
4595 | NEONMAP1(vld4_v, arm_neon_vld4, 0), | |||
4596 | NEONMAP1(vld4q_dup_v, arm_neon_vld4dup, 0), | |||
4597 | NEONMAP1(vld4q_lane_v, arm_neon_vld4lane, 0), | |||
4598 | NEONMAP1(vld4q_v, arm_neon_vld4, 0), | |||
4599 | NEONMAP2(vmax_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), | |||
4600 | NEONMAP1(vmaxnm_v, arm_neon_vmaxnm, Add1ArgType), | |||
4601 | NEONMAP1(vmaxnmq_v, arm_neon_vmaxnm, Add1ArgType), | |||
4602 | NEONMAP2(vmaxq_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), | |||
4603 | NEONMAP2(vmin_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), | |||
4604 | NEONMAP1(vminnm_v, arm_neon_vminnm, Add1ArgType), | |||
4605 | NEONMAP1(vminnmq_v, arm_neon_vminnm, Add1ArgType), | |||
4606 | NEONMAP2(vminq_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), | |||
4607 | NEONMAP0(vmovl_v), | |||
4608 | NEONMAP0(vmovn_v), | |||
4609 | NEONMAP1(vmul_v, arm_neon_vmulp, Add1ArgType), | |||
4610 | NEONMAP0(vmull_v), | |||
4611 | NEONMAP1(vmulq_v, arm_neon_vmulp, Add1ArgType), | |||
4612 | NEONMAP2(vpadal_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), | |||
4613 | NEONMAP2(vpadalq_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), | |||
4614 | NEONMAP1(vpadd_v, arm_neon_vpadd, Add1ArgType), | |||
4615 | NEONMAP2(vpaddl_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), | |||
4616 | NEONMAP2(vpaddlq_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), | |||
4617 | NEONMAP1(vpaddq_v, arm_neon_vpadd, Add1ArgType), | |||
4618 | NEONMAP2(vpmax_v, arm_neon_vpmaxu, arm_neon_vpmaxs, Add1ArgType | UnsignedAlts), | |||
4619 | NEONMAP2(vpmin_v, arm_neon_vpminu, arm_neon_vpmins, Add1ArgType | UnsignedAlts), | |||
4620 | NEONMAP1(vqabs_v, arm_neon_vqabs, Add1ArgType), | |||
4621 | NEONMAP1(vqabsq_v, arm_neon_vqabs, Add1ArgType), | |||
4622 | NEONMAP2(vqadd_v, arm_neon_vqaddu, arm_neon_vqadds, Add1ArgType | UnsignedAlts), | |||
4623 | NEONMAP2(vqaddq_v, arm_neon_vqaddu, arm_neon_vqadds, Add1ArgType | UnsignedAlts), | |||
4624 | NEONMAP2(vqdmlal_v, arm_neon_vqdmull, arm_neon_vqadds, 0), | |||
4625 | NEONMAP2(vqdmlsl_v, arm_neon_vqdmull, arm_neon_vqsubs, 0), | |||
4626 | NEONMAP1(vqdmulh_v, arm_neon_vqdmulh, Add1ArgType), | |||
4627 | NEONMAP1(vqdmulhq_v, arm_neon_vqdmulh, Add1ArgType), | |||
4628 | NEONMAP1(vqdmull_v, arm_neon_vqdmull, Add1ArgType), | |||
4629 | NEONMAP2(vqmovn_v, arm_neon_vqmovnu, arm_neon_vqmovns, Add1ArgType | UnsignedAlts), | |||
4630 | NEONMAP1(vqmovun_v, arm_neon_vqmovnsu, Add1ArgType), | |||
4631 | NEONMAP1(vqneg_v, arm_neon_vqneg, Add1ArgType), | |||
4632 | NEONMAP1(vqnegq_v, arm_neon_vqneg, Add1ArgType), | |||
4633 | NEONMAP1(vqrdmulh_v, arm_neon_vqrdmulh, Add1ArgType), | |||
4634 | NEONMAP1(vqrdmulhq_v, arm_neon_vqrdmulh, Add1ArgType), | |||
4635 | NEONMAP2(vqrshl_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), | |||
4636 | NEONMAP2(vqrshlq_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), | |||
4637 | NEONMAP2(vqshl_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), | |||
4638 | NEONMAP2(vqshl_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), | |||
4639 | NEONMAP2(vqshlq_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), | |||
4640 | NEONMAP2(vqshlq_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), | |||
4641 | NEONMAP1(vqshlu_n_v, arm_neon_vqshiftsu, 0), | |||
4642 | NEONMAP1(vqshluq_n_v, arm_neon_vqshiftsu, 0), | |||
4643 | NEONMAP2(vqsub_v, arm_neon_vqsubu, arm_neon_vqsubs, Add1ArgType | UnsignedAlts), | |||
4644 | NEONMAP2(vqsubq_v, arm_neon_vqsubu, arm_neon_vqsubs, Add1ArgType | UnsignedAlts), | |||
4645 | NEONMAP1(vraddhn_v, arm_neon_vraddhn, Add1ArgType), | |||
4646 | NEONMAP2(vrecpe_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), | |||
4647 | NEONMAP2(vrecpeq_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), | |||
4648 | NEONMAP1(vrecps_v, arm_neon_vrecps, Add1ArgType), | |||
4649 | NEONMAP1(vrecpsq_v, arm_neon_vrecps, Add1ArgType), | |||
4650 | NEONMAP2(vrhadd_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), | |||
4651 | NEONMAP2(vrhaddq_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), | |||
4652 | NEONMAP1(vrnd_v, arm_neon_vrintz, Add1ArgType), | |||
4653 | NEONMAP1(vrnda_v, arm_neon_vrinta, Add1ArgType), | |||
4654 | NEONMAP1(vrndaq_v, arm_neon_vrinta, Add1ArgType), | |||
4655 | NEONMAP0(vrndi_v), | |||
4656 | NEONMAP0(vrndiq_v), | |||
4657 | NEONMAP1(vrndm_v, arm_neon_vrintm, Add1ArgType), | |||
4658 | NEONMAP1(vrndmq_v, arm_neon_vrintm, Add1ArgType), | |||
4659 | NEONMAP1(vrndn_v, arm_neon_vrintn, Add1ArgType), | |||
4660 | NEONMAP1(vrndnq_v, arm_neon_vrintn, Add1ArgType), | |||
4661 | NEONMAP1(vrndp_v, arm_neon_vrintp, Add1ArgType), | |||
4662 | NEONMAP1(vrndpq_v, arm_neon_vrintp, Add1ArgType), | |||
4663 | NEONMAP1(vrndq_v, arm_neon_vrintz, Add1ArgType), | |||
4664 | NEONMAP1(vrndx_v, arm_neon_vrintx, Add1ArgType), | |||
4665 | NEONMAP1(vrndxq_v, arm_neon_vrintx, Add1ArgType), | |||
4666 | NEONMAP2(vrshl_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), | |||
4667 | NEONMAP2(vrshlq_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), | |||
4668 | NEONMAP2(vrshr_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), | |||
4669 | NEONMAP2(vrshrq_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), | |||
4670 | NEONMAP2(vrsqrte_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), | |||
4671 | NEONMAP2(vrsqrteq_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), | |||
4672 | NEONMAP1(vrsqrts_v, arm_neon_vrsqrts, Add1ArgType), | |||
4673 | NEONMAP1(vrsqrtsq_v, arm_neon_vrsqrts, Add1ArgType), | |||
4674 | NEONMAP1(vrsubhn_v, arm_neon_vrsubhn, Add1ArgType), | |||
4675 | NEONMAP1(vsha1su0q_v, arm_neon_sha1su0, 0), | |||
4676 | NEONMAP1(vsha1su1q_v, arm_neon_sha1su1, 0), | |||
4677 | NEONMAP1(vsha256h2q_v, arm_neon_sha256h2, 0), | |||
4678 | NEONMAP1(vsha256hq_v, arm_neon_sha256h, 0), | |||
4679 | NEONMAP1(vsha256su0q_v, arm_neon_sha256su0, 0), | |||
4680 | NEONMAP1(vsha256su1q_v, arm_neon_sha256su1, 0), | |||
4681 | NEONMAP0(vshl_n_v), | |||
4682 | NEONMAP2(vshl_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), | |||
4683 | NEONMAP0(vshll_n_v), | |||
4684 | NEONMAP0(vshlq_n_v), | |||
4685 | NEONMAP2(vshlq_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), | |||
4686 | NEONMAP0(vshr_n_v), | |||
4687 | NEONMAP0(vshrn_n_v), | |||
4688 | NEONMAP0(vshrq_n_v), | |||
4689 | NEONMAP1(vst1_v, arm_neon_vst1, 0), | |||
4690 | NEONMAP1(vst1_x2_v, arm_neon_vst1x2, 0), | |||
4691 | NEONMAP1(vst1_x3_v, arm_neon_vst1x3, 0), | |||
4692 | NEONMAP1(vst1_x4_v, arm_neon_vst1x4, 0), | |||
4693 | NEONMAP1(vst1q_v, arm_neon_vst1, 0), | |||
4694 | NEONMAP1(vst1q_x2_v, arm_neon_vst1x2, 0), | |||
4695 | NEONMAP1(vst1q_x3_v, arm_neon_vst1x3, 0), | |||
4696 | NEONMAP1(vst1q_x4_v, arm_neon_vst1x4, 0), | |||
4697 | NEONMAP1(vst2_lane_v, arm_neon_vst2lane, 0), | |||
4698 | NEONMAP1(vst2_v, arm_neon_vst2, 0), | |||
4699 | NEONMAP1(vst2q_lane_v, arm_neon_vst2lane, 0), | |||
4700 | NEONMAP1(vst2q_v, arm_neon_vst2, 0), | |||
4701 | NEONMAP1(vst3_lane_v, arm_neon_vst3lane, 0), | |||
4702 | NEONMAP1(vst3_v, arm_neon_vst3, 0), | |||
4703 | NEONMAP1(vst3q_lane_v, arm_neon_vst3lane, 0), | |||
4704 | NEONMAP1(vst3q_v, arm_neon_vst3, 0), | |||
4705 | NEONMAP1(vst4_lane_v, arm_neon_vst4lane, 0), | |||
4706 | NEONMAP1(vst4_v, arm_neon_vst4, 0), | |||
4707 | NEONMAP1(vst4q_lane_v, arm_neon_vst4lane, 0), | |||
4708 | NEONMAP1(vst4q_v, arm_neon_vst4, 0), | |||
4709 | NEONMAP0(vsubhn_v), | |||
4710 | NEONMAP0(vtrn_v), | |||
4711 | NEONMAP0(vtrnq_v), | |||
4712 | NEONMAP0(vtst_v), | |||
4713 | NEONMAP0(vtstq_v), | |||
4714 | NEONMAP0(vuzp_v), | |||
4715 | NEONMAP0(vuzpq_v), | |||
4716 | NEONMAP0(vzip_v), | |||
4717 | NEONMAP0(vzipq_v) | |||
4718 | }; | |||
4719 | ||||
4720 | static const NeonIntrinsicInfo AArch64SIMDIntrinsicMap[] = { | |||
4721 | NEONMAP1(vabs_v, aarch64_neon_abs, 0), | |||
4722 | NEONMAP1(vabsq_v, aarch64_neon_abs, 0), | |||
4723 | NEONMAP0(vaddhn_v), | |||
4724 | NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0), | |||
4725 | NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0), | |||
4726 | NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0), | |||
4727 | NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0), | |||
4728 | NEONMAP1(vcage_v, aarch64_neon_facge, 0), | |||
4729 | NEONMAP1(vcageq_v, aarch64_neon_facge, 0), | |||
4730 | NEONMAP1(vcagt_v, aarch64_neon_facgt, 0), | |||
4731 | NEONMAP1(vcagtq_v, aarch64_neon_facgt, 0), | |||
4732 | NEONMAP1(vcale_v, aarch64_neon_facge, 0), | |||
4733 | NEONMAP1(vcaleq_v, aarch64_neon_facge, 0), | |||
4734 | NEONMAP1(vcalt_v, aarch64_neon_facgt, 0), | |||
4735 | NEONMAP1(vcaltq_v, aarch64_neon_facgt, 0), | |||
4736 | NEONMAP0(vceqz_v), | |||
4737 | NEONMAP0(vceqzq_v), | |||
4738 | NEONMAP0(vcgez_v), | |||
4739 | NEONMAP0(vcgezq_v), | |||
4740 | NEONMAP0(vcgtz_v), | |||
4741 | NEONMAP0(vcgtzq_v), | |||
4742 | NEONMAP0(vclez_v), | |||
4743 | NEONMAP0(vclezq_v), | |||
4744 | NEONMAP1(vcls_v, aarch64_neon_cls, Add1ArgType), | |||
4745 | NEONMAP1(vclsq_v, aarch64_neon_cls, Add1ArgType), | |||
4746 | NEONMAP0(vcltz_v), | |||
4747 | NEONMAP0(vcltzq_v), | |||
4748 | NEONMAP1(vclz_v, ctlz, Add1ArgType), | |||
4749 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), | |||
4750 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), | |||
4751 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), | |||
4752 | NEONMAP1(vcvt_f16_f32, aarch64_neon_vcvtfp2hf, 0), | |||
4753 | NEONMAP0(vcvt_f16_v), | |||
4754 | NEONMAP1(vcvt_f32_f16, aarch64_neon_vcvthf2fp, 0), | |||
4755 | NEONMAP0(vcvt_f32_v), | |||
4756 | NEONMAP2(vcvt_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
4757 | NEONMAP2(vcvt_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
4758 | NEONMAP2(vcvt_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
4759 | NEONMAP1(vcvt_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), | |||
4760 | NEONMAP1(vcvt_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), | |||
4761 | NEONMAP1(vcvt_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), | |||
4762 | NEONMAP1(vcvt_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), | |||
4763 | NEONMAP1(vcvt_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), | |||
4764 | NEONMAP1(vcvt_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), | |||
4765 | NEONMAP0(vcvtq_f16_v), | |||
4766 | NEONMAP0(vcvtq_f32_v), | |||
4767 | NEONMAP2(vcvtq_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
4768 | NEONMAP2(vcvtq_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
4769 | NEONMAP2(vcvtq_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
4770 | NEONMAP1(vcvtq_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), | |||
4771 | NEONMAP1(vcvtq_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), | |||
4772 | NEONMAP1(vcvtq_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), | |||
4773 | NEONMAP1(vcvtq_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), | |||
4774 | NEONMAP1(vcvtq_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), | |||
4775 | NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), | |||
4776 | NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType), | |||
4777 | NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0), | |||
4778 | NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0), | |||
4779 | NEONMAP0(vext_v), | |||
4780 | NEONMAP0(vextq_v), | |||
4781 | NEONMAP0(vfma_v), | |||
4782 | NEONMAP0(vfmaq_v), | |||
4783 | NEONMAP1(vfmlal_high_v, aarch64_neon_fmlal2, 0), | |||
4784 | NEONMAP1(vfmlal_low_v, aarch64_neon_fmlal, 0), | |||
4785 | NEONMAP1(vfmlalq_high_v, aarch64_neon_fmlal2, 0), | |||
4786 | NEONMAP1(vfmlalq_low_v, aarch64_neon_fmlal, 0), | |||
4787 | NEONMAP1(vfmlsl_high_v, aarch64_neon_fmlsl2, 0), | |||
4788 | NEONMAP1(vfmlsl_low_v, aarch64_neon_fmlsl, 0), | |||
4789 | NEONMAP1(vfmlslq_high_v, aarch64_neon_fmlsl2, 0), | |||
4790 | NEONMAP1(vfmlslq_low_v, aarch64_neon_fmlsl, 0), | |||
4791 | NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), | |||
4792 | NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), | |||
4793 | NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), | |||
4794 | NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), | |||
4795 | NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0), | |||
4796 | NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0), | |||
4797 | NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0), | |||
4798 | NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0), | |||
4799 | NEONMAP1(vld1q_x3_v, aarch64_neon_ld1x3, 0), | |||
4800 | NEONMAP1(vld1q_x4_v, aarch64_neon_ld1x4, 0), | |||
4801 | NEONMAP0(vmovl_v), | |||
4802 | NEONMAP0(vmovn_v), | |||
4803 | NEONMAP1(vmul_v, aarch64_neon_pmul, Add1ArgType), | |||
4804 | NEONMAP1(vmulq_v, aarch64_neon_pmul, Add1ArgType), | |||
4805 | NEONMAP1(vpadd_v, aarch64_neon_addp, Add1ArgType), | |||
4806 | NEONMAP2(vpaddl_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), | |||
4807 | NEONMAP2(vpaddlq_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), | |||
4808 | NEONMAP1(vpaddq_v, aarch64_neon_addp, Add1ArgType), | |||
4809 | NEONMAP1(vqabs_v, aarch64_neon_sqabs, Add1ArgType), | |||
4810 | NEONMAP1(vqabsq_v, aarch64_neon_sqabs, Add1ArgType), | |||
4811 | NEONMAP2(vqadd_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), | |||
4812 | NEONMAP2(vqaddq_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), | |||
4813 | NEONMAP2(vqdmlal_v, aarch64_neon_sqdmull, aarch64_neon_sqadd, 0), | |||
4814 | NEONMAP2(vqdmlsl_v, aarch64_neon_sqdmull, aarch64_neon_sqsub, 0), | |||
4815 | NEONMAP1(vqdmulh_v, aarch64_neon_sqdmulh, Add1ArgType), | |||
4816 | NEONMAP1(vqdmulhq_v, aarch64_neon_sqdmulh, Add1ArgType), | |||
4817 | NEONMAP1(vqdmull_v, aarch64_neon_sqdmull, Add1ArgType), | |||
4818 | NEONMAP2(vqmovn_v, aarch64_neon_uqxtn, aarch64_neon_sqxtn, Add1ArgType | UnsignedAlts), | |||
4819 | NEONMAP1(vqmovun_v, aarch64_neon_sqxtun, Add1ArgType), | |||
4820 | NEONMAP1(vqneg_v, aarch64_neon_sqneg, Add1ArgType), | |||
4821 | NEONMAP1(vqnegq_v, aarch64_neon_sqneg, Add1ArgType), | |||
4822 | NEONMAP1(vqrdmulh_v, aarch64_neon_sqrdmulh, Add1ArgType), | |||
4823 | NEONMAP1(vqrdmulhq_v, aarch64_neon_sqrdmulh, Add1ArgType), | |||
4824 | NEONMAP2(vqrshl_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), | |||
4825 | NEONMAP2(vqrshlq_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), | |||
4826 | NEONMAP2(vqshl_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl, UnsignedAlts), | |||
4827 | NEONMAP2(vqshl_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), | |||
4828 | NEONMAP2(vqshlq_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl,UnsignedAlts), | |||
4829 | NEONMAP2(vqshlq_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), | |||
4830 | NEONMAP1(vqshlu_n_v, aarch64_neon_sqshlu, 0), | |||
4831 | NEONMAP1(vqshluq_n_v, aarch64_neon_sqshlu, 0), | |||
4832 | NEONMAP2(vqsub_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), | |||
4833 | NEONMAP2(vqsubq_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), | |||
4834 | NEONMAP1(vraddhn_v, aarch64_neon_raddhn, Add1ArgType), | |||
4835 | NEONMAP2(vrecpe_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), | |||
4836 | NEONMAP2(vrecpeq_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), | |||
4837 | NEONMAP1(vrecps_v, aarch64_neon_frecps, Add1ArgType), | |||
4838 | NEONMAP1(vrecpsq_v, aarch64_neon_frecps, Add1ArgType), | |||
4839 | NEONMAP2(vrhadd_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), | |||
4840 | NEONMAP2(vrhaddq_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), | |||
4841 | NEONMAP0(vrndi_v), | |||
4842 | NEONMAP0(vrndiq_v), | |||
4843 | NEONMAP2(vrshl_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), | |||
4844 | NEONMAP2(vrshlq_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), | |||
4845 | NEONMAP2(vrshr_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), | |||
4846 | NEONMAP2(vrshrq_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), | |||
4847 | NEONMAP2(vrsqrte_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), | |||
4848 | NEONMAP2(vrsqrteq_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), | |||
4849 | NEONMAP1(vrsqrts_v, aarch64_neon_frsqrts, Add1ArgType), | |||
4850 | NEONMAP1(vrsqrtsq_v, aarch64_neon_frsqrts, Add1ArgType), | |||
4851 | NEONMAP1(vrsubhn_v, aarch64_neon_rsubhn, Add1ArgType), | |||
4852 | NEONMAP1(vsha1su0q_v, aarch64_crypto_sha1su0, 0), | |||
4853 | NEONMAP1(vsha1su1q_v, aarch64_crypto_sha1su1, 0), | |||
4854 | NEONMAP1(vsha256h2q_v, aarch64_crypto_sha256h2, 0), | |||
4855 | NEONMAP1(vsha256hq_v, aarch64_crypto_sha256h, 0), | |||
4856 | NEONMAP1(vsha256su0q_v, aarch64_crypto_sha256su0, 0), | |||
4857 | NEONMAP1(vsha256su1q_v, aarch64_crypto_sha256su1, 0), | |||
4858 | NEONMAP0(vshl_n_v), | |||
4859 | NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), | |||
4860 | NEONMAP0(vshll_n_v), | |||
4861 | NEONMAP0(vshlq_n_v), | |||
4862 | NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), | |||
4863 | NEONMAP0(vshr_n_v), | |||
4864 | NEONMAP0(vshrn_n_v), | |||
4865 | NEONMAP0(vshrq_n_v), | |||
4866 | NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0), | |||
4867 | NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0), | |||
4868 | NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0), | |||
4869 | NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0), | |||
4870 | NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0), | |||
4871 | NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0), | |||
4872 | NEONMAP0(vsubhn_v), | |||
4873 | NEONMAP0(vtst_v), | |||
4874 | NEONMAP0(vtstq_v), | |||
4875 | }; | |||
4876 | ||||
4877 | static const NeonIntrinsicInfo AArch64SISDIntrinsicMap[] = { | |||
4878 | NEONMAP1(vabdd_f64, aarch64_sisd_fabd, Add1ArgType), | |||
4879 | NEONMAP1(vabds_f32, aarch64_sisd_fabd, Add1ArgType), | |||
4880 | NEONMAP1(vabsd_s64, aarch64_neon_abs, Add1ArgType), | |||
4881 | NEONMAP1(vaddlv_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), | |||
4882 | NEONMAP1(vaddlv_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), | |||
4883 | NEONMAP1(vaddlvq_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), | |||
4884 | NEONMAP1(vaddlvq_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), | |||
4885 | NEONMAP1(vaddv_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), | |||
4886 | NEONMAP1(vaddv_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), | |||
4887 | NEONMAP1(vaddv_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
4888 | NEONMAP1(vaddvq_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), | |||
4889 | NEONMAP1(vaddvq_f64, aarch64_neon_faddv, AddRetType | Add1ArgType), | |||
4890 | NEONMAP1(vaddvq_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), | |||
4891 | NEONMAP1(vaddvq_s64, aarch64_neon_saddv, AddRetType | Add1ArgType), | |||
4892 | NEONMAP1(vaddvq_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
4893 | NEONMAP1(vaddvq_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
4894 | NEONMAP1(vcaged_f64, aarch64_neon_facge, AddRetType | Add1ArgType), | |||
4895 | NEONMAP1(vcages_f32, aarch64_neon_facge, AddRetType | Add1ArgType), | |||
4896 | NEONMAP1(vcagtd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), | |||
4897 | NEONMAP1(vcagts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), | |||
4898 | NEONMAP1(vcaled_f64, aarch64_neon_facge, AddRetType | Add1ArgType), | |||
4899 | NEONMAP1(vcales_f32, aarch64_neon_facge, AddRetType | Add1ArgType), | |||
4900 | NEONMAP1(vcaltd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), | |||
4901 | NEONMAP1(vcalts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), | |||
4902 | NEONMAP1(vcvtad_s64_f64, aarch64_neon_fcvtas, AddRetType | Add1ArgType), | |||
4903 | NEONMAP1(vcvtad_u64_f64, aarch64_neon_fcvtau, AddRetType | Add1ArgType), | |||
4904 | NEONMAP1(vcvtas_s32_f32, aarch64_neon_fcvtas, AddRetType | Add1ArgType), | |||
4905 | NEONMAP1(vcvtas_u32_f32, aarch64_neon_fcvtau, AddRetType | Add1ArgType), | |||
4906 | NEONMAP1(vcvtd_n_f64_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), | |||
4907 | NEONMAP1(vcvtd_n_f64_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), | |||
4908 | NEONMAP1(vcvtd_n_s64_f64, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), | |||
4909 | NEONMAP1(vcvtd_n_u64_f64, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), | |||
4910 | NEONMAP1(vcvtmd_s64_f64, aarch64_neon_fcvtms, AddRetType | Add1ArgType), | |||
4911 | NEONMAP1(vcvtmd_u64_f64, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), | |||
4912 | NEONMAP1(vcvtms_s32_f32, aarch64_neon_fcvtms, AddRetType | Add1ArgType), | |||
4913 | NEONMAP1(vcvtms_u32_f32, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), | |||
4914 | NEONMAP1(vcvtnd_s64_f64, aarch64_neon_fcvtns, AddRetType | Add1ArgType), | |||
4915 | NEONMAP1(vcvtnd_u64_f64, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), | |||
4916 | NEONMAP1(vcvtns_s32_f32, aarch64_neon_fcvtns, AddRetType | Add1ArgType), | |||
4917 | NEONMAP1(vcvtns_u32_f32, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), | |||
4918 | NEONMAP1(vcvtpd_s64_f64, aarch64_neon_fcvtps, AddRetType | Add1ArgType), | |||
4919 | NEONMAP1(vcvtpd_u64_f64, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), | |||
4920 | NEONMAP1(vcvtps_s32_f32, aarch64_neon_fcvtps, AddRetType | Add1ArgType), | |||
4921 | NEONMAP1(vcvtps_u32_f32, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), | |||
4922 | NEONMAP1(vcvts_n_f32_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), | |||
4923 | NEONMAP1(vcvts_n_f32_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), | |||
4924 | NEONMAP1(vcvts_n_s32_f32, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), | |||
4925 | NEONMAP1(vcvts_n_u32_f32, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), | |||
4926 | NEONMAP1(vcvtxd_f32_f64, aarch64_sisd_fcvtxn, 0), | |||
4927 | NEONMAP1(vmaxnmv_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
4928 | NEONMAP1(vmaxnmvq_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
4929 | NEONMAP1(vmaxnmvq_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
4930 | NEONMAP1(vmaxv_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
4931 | NEONMAP1(vmaxv_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), | |||
4932 | NEONMAP1(vmaxv_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), | |||
4933 | NEONMAP1(vmaxvq_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
4934 | NEONMAP1(vmaxvq_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
4935 | NEONMAP1(vmaxvq_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), | |||
4936 | NEONMAP1(vmaxvq_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), | |||
4937 | NEONMAP1(vminnmv_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
4938 | NEONMAP1(vminnmvq_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
4939 | NEONMAP1(vminnmvq_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
4940 | NEONMAP1(vminv_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
4941 | NEONMAP1(vminv_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), | |||
4942 | NEONMAP1(vminv_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), | |||
4943 | NEONMAP1(vminvq_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
4944 | NEONMAP1(vminvq_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
4945 | NEONMAP1(vminvq_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), | |||
4946 | NEONMAP1(vminvq_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), | |||
4947 | NEONMAP1(vmull_p64, aarch64_neon_pmull64, 0), | |||
4948 | NEONMAP1(vmulxd_f64, aarch64_neon_fmulx, Add1ArgType), | |||
4949 | NEONMAP1(vmulxs_f32, aarch64_neon_fmulx, Add1ArgType), | |||
4950 | NEONMAP1(vpaddd_s64, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
4951 | NEONMAP1(vpaddd_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
4952 | NEONMAP1(vpmaxnmqd_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
4953 | NEONMAP1(vpmaxnms_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
4954 | NEONMAP1(vpmaxqd_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
4955 | NEONMAP1(vpmaxs_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
4956 | NEONMAP1(vpminnmqd_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
4957 | NEONMAP1(vpminnms_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
4958 | NEONMAP1(vpminqd_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
4959 | NEONMAP1(vpmins_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
4960 | NEONMAP1(vqabsb_s8, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), | |||
4961 | NEONMAP1(vqabsd_s64, aarch64_neon_sqabs, Add1ArgType), | |||
4962 | NEONMAP1(vqabsh_s16, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), | |||
4963 | NEONMAP1(vqabss_s32, aarch64_neon_sqabs, Add1ArgType), | |||
4964 | NEONMAP1(vqaddb_s8, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), | |||
4965 | NEONMAP1(vqaddb_u8, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), | |||
4966 | NEONMAP1(vqaddd_s64, aarch64_neon_sqadd, Add1ArgType), | |||
4967 | NEONMAP1(vqaddd_u64, aarch64_neon_uqadd, Add1ArgType), | |||
4968 | NEONMAP1(vqaddh_s16, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), | |||
4969 | NEONMAP1(vqaddh_u16, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), | |||
4970 | NEONMAP1(vqadds_s32, aarch64_neon_sqadd, Add1ArgType), | |||
4971 | NEONMAP1(vqadds_u32, aarch64_neon_uqadd, Add1ArgType), | |||
4972 | NEONMAP1(vqdmulhh_s16, aarch64_neon_sqdmulh, Vectorize1ArgType | Use64BitVectors), | |||
4973 | NEONMAP1(vqdmulhs_s32, aarch64_neon_sqdmulh, Add1ArgType), | |||
4974 | NEONMAP1(vqdmullh_s16, aarch64_neon_sqdmull, VectorRet | Use128BitVectors), | |||
4975 | NEONMAP1(vqdmulls_s32, aarch64_neon_sqdmulls_scalar, 0), | |||
4976 | NEONMAP1(vqmovnd_s64, aarch64_neon_scalar_sqxtn, AddRetType | Add1ArgType), | |||
4977 | NEONMAP1(vqmovnd_u64, aarch64_neon_scalar_uqxtn, AddRetType | Add1ArgType), | |||
4978 | NEONMAP1(vqmovnh_s16, aarch64_neon_sqxtn, VectorRet | Use64BitVectors), | |||
4979 | NEONMAP1(vqmovnh_u16, aarch64_neon_uqxtn, VectorRet | Use64BitVectors), | |||
4980 | NEONMAP1(vqmovns_s32, aarch64_neon_sqxtn, VectorRet | Use64BitVectors), | |||
4981 | NEONMAP1(vqmovns_u32, aarch64_neon_uqxtn, VectorRet | Use64BitVectors), | |||
4982 | NEONMAP1(vqmovund_s64, aarch64_neon_scalar_sqxtun, AddRetType | Add1ArgType), | |||
4983 | NEONMAP1(vqmovunh_s16, aarch64_neon_sqxtun, VectorRet | Use64BitVectors), | |||
4984 | NEONMAP1(vqmovuns_s32, aarch64_neon_sqxtun, VectorRet | Use64BitVectors), | |||
4985 | NEONMAP1(vqnegb_s8, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors), | |||
4986 | NEONMAP1(vqnegd_s64, aarch64_neon_sqneg, Add1ArgType), | |||
4987 | NEONMAP1(vqnegh_s16, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors), | |||
4988 | NEONMAP1(vqnegs_s32, aarch64_neon_sqneg, Add1ArgType), | |||
4989 | NEONMAP1(vqrdmulhh_s16, aarch64_neon_sqrdmulh, Vectorize1ArgType | Use64BitVectors), | |||
4990 | NEONMAP1(vqrdmulhs_s32, aarch64_neon_sqrdmulh, Add1ArgType), | |||
4991 | NEONMAP1(vqrshlb_s8, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors), | |||
4992 | NEONMAP1(vqrshlb_u8, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors), | |||
4993 | NEONMAP1(vqrshld_s64, aarch64_neon_sqrshl, Add1ArgType), | |||
4994 | NEONMAP1(vqrshld_u64, aarch64_neon_uqrshl, Add1ArgType), | |||
4995 | NEONMAP1(vqrshlh_s16, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors), | |||
4996 | NEONMAP1(vqrshlh_u16, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors), | |||
4997 | NEONMAP1(vqrshls_s32, aarch64_neon_sqrshl, Add1ArgType), | |||
4998 | NEONMAP1(vqrshls_u32, aarch64_neon_uqrshl, Add1ArgType), | |||
4999 | NEONMAP1(vqrshrnd_n_s64, aarch64_neon_sqrshrn, AddRetType), | |||
5000 | NEONMAP1(vqrshrnd_n_u64, aarch64_neon_uqrshrn, AddRetType), | |||
5001 | NEONMAP1(vqrshrnh_n_s16, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors), | |||
5002 | NEONMAP1(vqrshrnh_n_u16, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors), | |||
5003 | NEONMAP1(vqrshrns_n_s32, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors), | |||
5004 | NEONMAP1(vqrshrns_n_u32, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors), | |||
5005 | NEONMAP1(vqrshrund_n_s64, aarch64_neon_sqrshrun, AddRetType), | |||
5006 | NEONMAP1(vqrshrunh_n_s16, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors), | |||
5007 | NEONMAP1(vqrshruns_n_s32, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors), | |||
5008 | NEONMAP1(vqshlb_n_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), | |||
5009 | NEONMAP1(vqshlb_n_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), | |||
5010 | NEONMAP1(vqshlb_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), | |||
5011 | NEONMAP1(vqshlb_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), | |||
5012 | NEONMAP1(vqshld_s64, aarch64_neon_sqshl, Add1ArgType), | |||
5013 | NEONMAP1(vqshld_u64, aarch64_neon_uqshl, Add1ArgType), | |||
5014 | NEONMAP1(vqshlh_n_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), | |||
5015 | NEONMAP1(vqshlh_n_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), | |||
5016 | NEONMAP1(vqshlh_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), | |||
5017 | NEONMAP1(vqshlh_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), | |||
5018 | NEONMAP1(vqshls_n_s32, aarch64_neon_sqshl, Add1ArgType), | |||
5019 | NEONMAP1(vqshls_n_u32, aarch64_neon_uqshl, Add1ArgType), | |||
5020 | NEONMAP1(vqshls_s32, aarch64_neon_sqshl, Add1ArgType), | |||
5021 | NEONMAP1(vqshls_u32, aarch64_neon_uqshl, Add1ArgType), | |||
5022 | NEONMAP1(vqshlub_n_s8, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors), | |||
5023 | NEONMAP1(vqshluh_n_s16, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors), | |||
5024 | NEONMAP1(vqshlus_n_s32, aarch64_neon_sqshlu, Add1ArgType), | |||
5025 | NEONMAP1(vqshrnd_n_s64, aarch64_neon_sqshrn, AddRetType), | |||
5026 | NEONMAP1(vqshrnd_n_u64, aarch64_neon_uqshrn, AddRetType), | |||
5027 | NEONMAP1(vqshrnh_n_s16, aarch64_neon_sqshrn, VectorRet | Use64BitVectors), | |||
5028 | NEONMAP1(vqshrnh_n_u16, aarch64_neon_uqshrn, VectorRet | Use64BitVectors), | |||
5029 | NEONMAP1(vqshrns_n_s32, aarch64_neon_sqshrn, VectorRet | Use64BitVectors), | |||
5030 | NEONMAP1(vqshrns_n_u32, aarch64_neon_uqshrn, VectorRet | Use64BitVectors), | |||
5031 | NEONMAP1(vqshrund_n_s64, aarch64_neon_sqshrun, AddRetType), | |||
5032 | NEONMAP1(vqshrunh_n_s16, aarch64_neon_sqshrun, VectorRet | Use64BitVectors), | |||
5033 | NEONMAP1(vqshruns_n_s32, aarch64_neon_sqshrun, VectorRet | Use64BitVectors), | |||
5034 | NEONMAP1(vqsubb_s8, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors), | |||
5035 | NEONMAP1(vqsubb_u8, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors), | |||
5036 | NEONMAP1(vqsubd_s64, aarch64_neon_sqsub, Add1ArgType), | |||
5037 | NEONMAP1(vqsubd_u64, aarch64_neon_uqsub, Add1ArgType), | |||
5038 | NEONMAP1(vqsubh_s16, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors), | |||
5039 | NEONMAP1(vqsubh_u16, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors), | |||
5040 | NEONMAP1(vqsubs_s32, aarch64_neon_sqsub, Add1ArgType), | |||
5041 | NEONMAP1(vqsubs_u32, aarch64_neon_uqsub, Add1ArgType), | |||
5042 | NEONMAP1(vrecped_f64, aarch64_neon_frecpe, Add1ArgType), | |||
5043 | NEONMAP1(vrecpes_f32, aarch64_neon_frecpe, Add1ArgType), | |||
5044 | NEONMAP1(vrecpxd_f64, aarch64_neon_frecpx, Add1ArgType), | |||
5045 | NEONMAP1(vrecpxs_f32, aarch64_neon_frecpx, Add1ArgType), | |||
5046 | NEONMAP1(vrshld_s64, aarch64_neon_srshl, Add1ArgType), | |||
5047 | NEONMAP1(vrshld_u64, aarch64_neon_urshl, Add1ArgType), | |||
5048 | NEONMAP1(vrsqrted_f64, aarch64_neon_frsqrte, Add1ArgType), | |||
5049 | NEONMAP1(vrsqrtes_f32, aarch64_neon_frsqrte, Add1ArgType), | |||
5050 | NEONMAP1(vrsqrtsd_f64, aarch64_neon_frsqrts, Add1ArgType), | |||
5051 | NEONMAP1(vrsqrtss_f32, aarch64_neon_frsqrts, Add1ArgType), | |||
5052 | NEONMAP1(vsha1cq_u32, aarch64_crypto_sha1c, 0), | |||
5053 | NEONMAP1(vsha1h_u32, aarch64_crypto_sha1h, 0), | |||
5054 | NEONMAP1(vsha1mq_u32, aarch64_crypto_sha1m, 0), | |||
5055 | NEONMAP1(vsha1pq_u32, aarch64_crypto_sha1p, 0), | |||
5056 | NEONMAP1(vshld_s64, aarch64_neon_sshl, Add1ArgType), | |||
5057 | NEONMAP1(vshld_u64, aarch64_neon_ushl, Add1ArgType), | |||
5058 | NEONMAP1(vslid_n_s64, aarch64_neon_vsli, Vectorize1ArgType), | |||
5059 | NEONMAP1(vslid_n_u64, aarch64_neon_vsli, Vectorize1ArgType), | |||
5060 | NEONMAP1(vsqaddb_u8, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors), | |||
5061 | NEONMAP1(vsqaddd_u64, aarch64_neon_usqadd, Add1ArgType), | |||
5062 | NEONMAP1(vsqaddh_u16, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors), | |||
5063 | NEONMAP1(vsqadds_u32, aarch64_neon_usqadd, Add1ArgType), | |||
5064 | NEONMAP1(vsrid_n_s64, aarch64_neon_vsri, Vectorize1ArgType), | |||
5065 | NEONMAP1(vsrid_n_u64, aarch64_neon_vsri, Vectorize1ArgType), | |||
5066 | NEONMAP1(vuqaddb_s8, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors), | |||
5067 | NEONMAP1(vuqaddd_s64, aarch64_neon_suqadd, Add1ArgType), | |||
5068 | NEONMAP1(vuqaddh_s16, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors), | |||
5069 | NEONMAP1(vuqadds_s32, aarch64_neon_suqadd, Add1ArgType), | |||
5070 | // FP16 scalar intrinisics go here. | |||
5071 | NEONMAP1(vabdh_f16, aarch64_sisd_fabd, Add1ArgType), | |||
5072 | NEONMAP1(vcvtah_s32_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType), | |||
5073 | NEONMAP1(vcvtah_s64_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType), | |||
5074 | NEONMAP1(vcvtah_u32_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType), | |||
5075 | NEONMAP1(vcvtah_u64_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType), | |||
5076 | NEONMAP1(vcvth_n_f16_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), | |||
5077 | NEONMAP1(vcvth_n_f16_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), | |||
5078 | NEONMAP1(vcvth_n_f16_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), | |||
5079 | NEONMAP1(vcvth_n_f16_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), | |||
5080 | NEONMAP1(vcvth_n_s32_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), | |||
5081 | NEONMAP1(vcvth_n_s64_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), | |||
5082 | NEONMAP1(vcvth_n_u32_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), | |||
5083 | NEONMAP1(vcvth_n_u64_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), | |||
5084 | NEONMAP1(vcvtmh_s32_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType), | |||
5085 | NEONMAP1(vcvtmh_s64_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType), | |||
5086 | NEONMAP1(vcvtmh_u32_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), | |||
5087 | NEONMAP1(vcvtmh_u64_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), | |||
5088 | NEONMAP1(vcvtnh_s32_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType), | |||
5089 | NEONMAP1(vcvtnh_s64_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType), | |||
5090 | NEONMAP1(vcvtnh_u32_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), | |||
5091 | NEONMAP1(vcvtnh_u64_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), | |||
5092 | NEONMAP1(vcvtph_s32_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType), | |||
5093 | NEONMAP1(vcvtph_s64_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType), | |||
5094 | NEONMAP1(vcvtph_u32_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), | |||
5095 | NEONMAP1(vcvtph_u64_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), | |||
5096 | NEONMAP1(vmulxh_f16, aarch64_neon_fmulx, Add1ArgType), | |||
5097 | NEONMAP1(vrecpeh_f16, aarch64_neon_frecpe, Add1ArgType), | |||
5098 | NEONMAP1(vrecpxh_f16, aarch64_neon_frecpx, Add1ArgType), | |||
5099 | NEONMAP1(vrsqrteh_f16, aarch64_neon_frsqrte, Add1ArgType), | |||
5100 | NEONMAP1(vrsqrtsh_f16, aarch64_neon_frsqrts, Add1ArgType), | |||
5101 | }; | |||
5102 | ||||
5103 | #undef NEONMAP0 | |||
5104 | #undef NEONMAP1 | |||
5105 | #undef NEONMAP2 | |||
5106 | ||||
5107 | static bool NEONSIMDIntrinsicsProvenSorted = false; | |||
5108 | ||||
5109 | static bool AArch64SIMDIntrinsicsProvenSorted = false; | |||
5110 | static bool AArch64SISDIntrinsicsProvenSorted = false; | |||
5111 | ||||
5112 | ||||
5113 | static const NeonIntrinsicInfo * | |||
5114 | findNeonIntrinsicInMap(ArrayRef<NeonIntrinsicInfo> IntrinsicMap, | |||
5115 | unsigned BuiltinID, bool &MapProvenSorted) { | |||
5116 | ||||
5117 | #ifndef NDEBUG | |||
5118 | if (!MapProvenSorted) { | |||
5119 | assert(std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap)))((std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap ))) ? static_cast<void> (0) : __assert_fail ("std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap))" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5119, __PRETTY_FUNCTION__)); | |||
5120 | MapProvenSorted = true; | |||
5121 | } | |||
5122 | #endif | |||
5123 | ||||
5124 | const NeonIntrinsicInfo *Builtin = llvm::lower_bound(IntrinsicMap, BuiltinID); | |||
5125 | ||||
5126 | if (Builtin != IntrinsicMap.end() && Builtin->BuiltinID == BuiltinID) | |||
5127 | return Builtin; | |||
5128 | ||||
5129 | return nullptr; | |||
5130 | } | |||
5131 | ||||
5132 | Function *CodeGenFunction::LookupNeonLLVMIntrinsic(unsigned IntrinsicID, | |||
5133 | unsigned Modifier, | |||
5134 | llvm::Type *ArgType, | |||
5135 | const CallExpr *E) { | |||
5136 | int VectorSize = 0; | |||
5137 | if (Modifier & Use64BitVectors) | |||
5138 | VectorSize = 64; | |||
5139 | else if (Modifier & Use128BitVectors) | |||
5140 | VectorSize = 128; | |||
5141 | ||||
5142 | // Return type. | |||
5143 | SmallVector<llvm::Type *, 3> Tys; | |||
5144 | if (Modifier & AddRetType) { | |||
5145 | llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext())); | |||
5146 | if (Modifier & VectorizeRetType) | |||
5147 | Ty = llvm::VectorType::get( | |||
5148 | Ty, VectorSize ? VectorSize / Ty->getPrimitiveSizeInBits() : 1); | |||
5149 | ||||
5150 | Tys.push_back(Ty); | |||
5151 | } | |||
5152 | ||||
5153 | // Arguments. | |||
5154 | if (Modifier & VectorizeArgTypes) { | |||
5155 | int Elts = VectorSize ? VectorSize / ArgType->getPrimitiveSizeInBits() : 1; | |||
5156 | ArgType = llvm::VectorType::get(ArgType, Elts); | |||
5157 | } | |||
5158 | ||||
5159 | if (Modifier & (Add1ArgType | Add2ArgTypes)) | |||
5160 | Tys.push_back(ArgType); | |||
5161 | ||||
5162 | if (Modifier & Add2ArgTypes) | |||
5163 | Tys.push_back(ArgType); | |||
5164 | ||||
5165 | if (Modifier & InventFloatType) | |||
5166 | Tys.push_back(FloatTy); | |||
5167 | ||||
5168 | return CGM.getIntrinsic(IntrinsicID, Tys); | |||
5169 | } | |||
5170 | ||||
5171 | static Value *EmitCommonNeonSISDBuiltinExpr(CodeGenFunction &CGF, | |||
5172 | const NeonIntrinsicInfo &SISDInfo, | |||
5173 | SmallVectorImpl<Value *> &Ops, | |||
5174 | const CallExpr *E) { | |||
5175 | unsigned BuiltinID = SISDInfo.BuiltinID; | |||
5176 | unsigned int Int = SISDInfo.LLVMIntrinsic; | |||
5177 | unsigned Modifier = SISDInfo.TypeModifier; | |||
5178 | const char *s = SISDInfo.NameHint; | |||
5179 | ||||
5180 | switch (BuiltinID) { | |||
5181 | case NEON::BI__builtin_neon_vcled_s64: | |||
5182 | case NEON::BI__builtin_neon_vcled_u64: | |||
5183 | case NEON::BI__builtin_neon_vcles_f32: | |||
5184 | case NEON::BI__builtin_neon_vcled_f64: | |||
5185 | case NEON::BI__builtin_neon_vcltd_s64: | |||
5186 | case NEON::BI__builtin_neon_vcltd_u64: | |||
5187 | case NEON::BI__builtin_neon_vclts_f32: | |||
5188 | case NEON::BI__builtin_neon_vcltd_f64: | |||
5189 | case NEON::BI__builtin_neon_vcales_f32: | |||
5190 | case NEON::BI__builtin_neon_vcaled_f64: | |||
5191 | case NEON::BI__builtin_neon_vcalts_f32: | |||
5192 | case NEON::BI__builtin_neon_vcaltd_f64: | |||
5193 | // Only one direction of comparisons actually exist, cmle is actually a cmge | |||
5194 | // with swapped operands. The table gives us the right intrinsic but we | |||
5195 | // still need to do the swap. | |||
5196 | std::swap(Ops[0], Ops[1]); | |||
5197 | break; | |||
5198 | } | |||
5199 | ||||
5200 | assert(Int && "Generic code assumes a valid intrinsic")((Int && "Generic code assumes a valid intrinsic") ? static_cast <void> (0) : __assert_fail ("Int && \"Generic code assumes a valid intrinsic\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5200, __PRETTY_FUNCTION__)); | |||
5201 | ||||
5202 | // Determine the type(s) of this overloaded AArch64 intrinsic. | |||
5203 | const Expr *Arg = E->getArg(0); | |||
5204 | llvm::Type *ArgTy = CGF.ConvertType(Arg->getType()); | |||
5205 | Function *F = CGF.LookupNeonLLVMIntrinsic(Int, Modifier, ArgTy, E); | |||
5206 | ||||
5207 | int j = 0; | |||
5208 | ConstantInt *C0 = ConstantInt::get(CGF.SizeTy, 0); | |||
5209 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); | |||
5210 | ai != ae; ++ai, ++j) { | |||
5211 | llvm::Type *ArgTy = ai->getType(); | |||
5212 | if (Ops[j]->getType()->getPrimitiveSizeInBits() == | |||
5213 | ArgTy->getPrimitiveSizeInBits()) | |||
5214 | continue; | |||
5215 | ||||
5216 | assert(ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy())((ArgTy->isVectorTy() && !Ops[j]->getType()-> isVectorTy()) ? static_cast<void> (0) : __assert_fail ( "ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy()" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5216, __PRETTY_FUNCTION__)); | |||
5217 | // The constant argument to an _n_ intrinsic always has Int32Ty, so truncate | |||
5218 | // it before inserting. | |||
5219 | Ops[j] = | |||
5220 | CGF.Builder.CreateTruncOrBitCast(Ops[j], ArgTy->getVectorElementType()); | |||
5221 | Ops[j] = | |||
5222 | CGF.Builder.CreateInsertElement(UndefValue::get(ArgTy), Ops[j], C0); | |||
5223 | } | |||
5224 | ||||
5225 | Value *Result = CGF.EmitNeonCall(F, Ops, s); | |||
5226 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); | |||
5227 | if (ResultType->getPrimitiveSizeInBits() < | |||
5228 | Result->getType()->getPrimitiveSizeInBits()) | |||
5229 | return CGF.Builder.CreateExtractElement(Result, C0); | |||
5230 | ||||
5231 | return CGF.Builder.CreateBitCast(Result, ResultType, s); | |||
5232 | } | |||
5233 | ||||
5234 | Value *CodeGenFunction::EmitCommonNeonBuiltinExpr( | |||
5235 | unsigned BuiltinID, unsigned LLVMIntrinsic, unsigned AltLLVMIntrinsic, | |||
5236 | const char *NameHint, unsigned Modifier, const CallExpr *E, | |||
5237 | SmallVectorImpl<llvm::Value *> &Ops, Address PtrOp0, Address PtrOp1, | |||
5238 | llvm::Triple::ArchType Arch) { | |||
5239 | // Get the last argument, which specifies the vector type. | |||
5240 | llvm::APSInt NeonTypeConst; | |||
5241 | const Expr *Arg = E->getArg(E->getNumArgs() - 1); | |||
5242 | if (!Arg->isIntegerConstantExpr(NeonTypeConst, getContext())) | |||
5243 | return nullptr; | |||
5244 | ||||
5245 | // Determine the type of this overloaded NEON intrinsic. | |||
5246 | NeonTypeFlags Type(NeonTypeConst.getZExtValue()); | |||
5247 | bool Usgn = Type.isUnsigned(); | |||
5248 | bool Quad = Type.isQuad(); | |||
5249 | const bool HasLegalHalfType = getTarget().hasLegalHalfType(); | |||
5250 | ||||
5251 | llvm::VectorType *VTy = GetNeonType(this, Type, HasLegalHalfType); | |||
5252 | llvm::Type *Ty = VTy; | |||
5253 | if (!Ty) | |||
5254 | return nullptr; | |||
5255 | ||||
5256 | auto getAlignmentValue32 = [&](Address addr) -> Value* { | |||
5257 | return Builder.getInt32(addr.getAlignment().getQuantity()); | |||
5258 | }; | |||
5259 | ||||
5260 | unsigned Int = LLVMIntrinsic; | |||
5261 | if ((Modifier & UnsignedAlts) && !Usgn) | |||
5262 | Int = AltLLVMIntrinsic; | |||
5263 | ||||
5264 | switch (BuiltinID) { | |||
5265 | default: break; | |||
5266 | case NEON::BI__builtin_neon_vpadd_v: | |||
5267 | case NEON::BI__builtin_neon_vpaddq_v: | |||
5268 | // We don't allow fp/int overloading of intrinsics. | |||
5269 | if (VTy->getElementType()->isFloatingPointTy() && | |||
5270 | Int == Intrinsic::aarch64_neon_addp) | |||
5271 | Int = Intrinsic::aarch64_neon_faddp; | |||
5272 | break; | |||
5273 | case NEON::BI__builtin_neon_vabs_v: | |||
5274 | case NEON::BI__builtin_neon_vabsq_v: | |||
5275 | if (VTy->getElementType()->isFloatingPointTy()) | |||
5276 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, Ty), Ops, "vabs"); | |||
5277 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), Ops, "vabs"); | |||
5278 | case NEON::BI__builtin_neon_vaddhn_v: { | |||
5279 | llvm::VectorType *SrcTy = | |||
5280 | llvm::VectorType::getExtendedElementVectorType(VTy); | |||
5281 | ||||
5282 | // %sum = add <4 x i32> %lhs, %rhs | |||
5283 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); | |||
5284 | Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy); | |||
5285 | Ops[0] = Builder.CreateAdd(Ops[0], Ops[1], "vaddhn"); | |||
5286 | ||||
5287 | // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16> | |||
5288 | Constant *ShiftAmt = | |||
5289 | ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2); | |||
5290 | Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vaddhn"); | |||
5291 | ||||
5292 | // %res = trunc <4 x i32> %high to <4 x i16> | |||
5293 | return Builder.CreateTrunc(Ops[0], VTy, "vaddhn"); | |||
5294 | } | |||
5295 | case NEON::BI__builtin_neon_vcale_v: | |||
5296 | case NEON::BI__builtin_neon_vcaleq_v: | |||
5297 | case NEON::BI__builtin_neon_vcalt_v: | |||
5298 | case NEON::BI__builtin_neon_vcaltq_v: | |||
5299 | std::swap(Ops[0], Ops[1]); | |||
5300 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
5301 | case NEON::BI__builtin_neon_vcage_v: | |||
5302 | case NEON::BI__builtin_neon_vcageq_v: | |||
5303 | case NEON::BI__builtin_neon_vcagt_v: | |||
5304 | case NEON::BI__builtin_neon_vcagtq_v: { | |||
5305 | llvm::Type *Ty; | |||
5306 | switch (VTy->getScalarSizeInBits()) { | |||
5307 | default: llvm_unreachable("unexpected type")::llvm::llvm_unreachable_internal("unexpected type", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5307); | |||
5308 | case 32: | |||
5309 | Ty = FloatTy; | |||
5310 | break; | |||
5311 | case 64: | |||
5312 | Ty = DoubleTy; | |||
5313 | break; | |||
5314 | case 16: | |||
5315 | Ty = HalfTy; | |||
5316 | break; | |||
5317 | } | |||
5318 | llvm::Type *VecFlt = llvm::VectorType::get(Ty, VTy->getNumElements()); | |||
5319 | llvm::Type *Tys[] = { VTy, VecFlt }; | |||
5320 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
5321 | return EmitNeonCall(F, Ops, NameHint); | |||
5322 | } | |||
5323 | case NEON::BI__builtin_neon_vceqz_v: | |||
5324 | case NEON::BI__builtin_neon_vceqzq_v: | |||
5325 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OEQ, | |||
5326 | ICmpInst::ICMP_EQ, "vceqz"); | |||
5327 | case NEON::BI__builtin_neon_vcgez_v: | |||
5328 | case NEON::BI__builtin_neon_vcgezq_v: | |||
5329 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGE, | |||
5330 | ICmpInst::ICMP_SGE, "vcgez"); | |||
5331 | case NEON::BI__builtin_neon_vclez_v: | |||
5332 | case NEON::BI__builtin_neon_vclezq_v: | |||
5333 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLE, | |||
5334 | ICmpInst::ICMP_SLE, "vclez"); | |||
5335 | case NEON::BI__builtin_neon_vcgtz_v: | |||
5336 | case NEON::BI__builtin_neon_vcgtzq_v: | |||
5337 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGT, | |||
5338 | ICmpInst::ICMP_SGT, "vcgtz"); | |||
5339 | case NEON::BI__builtin_neon_vcltz_v: | |||
5340 | case NEON::BI__builtin_neon_vcltzq_v: | |||
5341 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLT, | |||
5342 | ICmpInst::ICMP_SLT, "vcltz"); | |||
5343 | case NEON::BI__builtin_neon_vclz_v: | |||
5344 | case NEON::BI__builtin_neon_vclzq_v: | |||
5345 | // We generate target-independent intrinsic, which needs a second argument | |||
5346 | // for whether or not clz of zero is undefined; on ARM it isn't. | |||
5347 | Ops.push_back(Builder.getInt1(getTarget().isCLZForZeroUndef())); | |||
5348 | break; | |||
5349 | case NEON::BI__builtin_neon_vcvt_f32_v: | |||
5350 | case NEON::BI__builtin_neon_vcvtq_f32_v: | |||
5351 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
5352 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float32, false, Quad), | |||
5353 | HasLegalHalfType); | |||
5354 | return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") | |||
5355 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); | |||
5356 | case NEON::BI__builtin_neon_vcvt_f16_v: | |||
5357 | case NEON::BI__builtin_neon_vcvtq_f16_v: | |||
5358 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
5359 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float16, false, Quad), | |||
5360 | HasLegalHalfType); | |||
5361 | return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") | |||
5362 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); | |||
5363 | case NEON::BI__builtin_neon_vcvt_n_f16_v: | |||
5364 | case NEON::BI__builtin_neon_vcvt_n_f32_v: | |||
5365 | case NEON::BI__builtin_neon_vcvt_n_f64_v: | |||
5366 | case NEON::BI__builtin_neon_vcvtq_n_f16_v: | |||
5367 | case NEON::BI__builtin_neon_vcvtq_n_f32_v: | |||
5368 | case NEON::BI__builtin_neon_vcvtq_n_f64_v: { | |||
5369 | llvm::Type *Tys[2] = { GetFloatNeonType(this, Type), Ty }; | |||
5370 | Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic; | |||
5371 | Function *F = CGM.getIntrinsic(Int, Tys); | |||
5372 | return EmitNeonCall(F, Ops, "vcvt_n"); | |||
5373 | } | |||
5374 | case NEON::BI__builtin_neon_vcvt_n_s16_v: | |||
5375 | case NEON::BI__builtin_neon_vcvt_n_s32_v: | |||
5376 | case NEON::BI__builtin_neon_vcvt_n_u16_v: | |||
5377 | case NEON::BI__builtin_neon_vcvt_n_u32_v: | |||
5378 | case NEON::BI__builtin_neon_vcvt_n_s64_v: | |||
5379 | case NEON::BI__builtin_neon_vcvt_n_u64_v: | |||
5380 | case NEON::BI__builtin_neon_vcvtq_n_s16_v: | |||
5381 | case NEON::BI__builtin_neon_vcvtq_n_s32_v: | |||
5382 | case NEON::BI__builtin_neon_vcvtq_n_u16_v: | |||
5383 | case NEON::BI__builtin_neon_vcvtq_n_u32_v: | |||
5384 | case NEON::BI__builtin_neon_vcvtq_n_s64_v: | |||
5385 | case NEON::BI__builtin_neon_vcvtq_n_u64_v: { | |||
5386 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
5387 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
5388 | return EmitNeonCall(F, Ops, "vcvt_n"); | |||
5389 | } | |||
5390 | case NEON::BI__builtin_neon_vcvt_s32_v: | |||
5391 | case NEON::BI__builtin_neon_vcvt_u32_v: | |||
5392 | case NEON::BI__builtin_neon_vcvt_s64_v: | |||
5393 | case NEON::BI__builtin_neon_vcvt_u64_v: | |||
5394 | case NEON::BI__builtin_neon_vcvt_s16_v: | |||
5395 | case NEON::BI__builtin_neon_vcvt_u16_v: | |||
5396 | case NEON::BI__builtin_neon_vcvtq_s32_v: | |||
5397 | case NEON::BI__builtin_neon_vcvtq_u32_v: | |||
5398 | case NEON::BI__builtin_neon_vcvtq_s64_v: | |||
5399 | case NEON::BI__builtin_neon_vcvtq_u64_v: | |||
5400 | case NEON::BI__builtin_neon_vcvtq_s16_v: | |||
5401 | case NEON::BI__builtin_neon_vcvtq_u16_v: { | |||
5402 | Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type)); | |||
5403 | return Usgn ? Builder.CreateFPToUI(Ops[0], Ty, "vcvt") | |||
5404 | : Builder.CreateFPToSI(Ops[0], Ty, "vcvt"); | |||
5405 | } | |||
5406 | case NEON::BI__builtin_neon_vcvta_s16_v: | |||
5407 | case NEON::BI__builtin_neon_vcvta_s32_v: | |||
5408 | case NEON::BI__builtin_neon_vcvta_s64_v: | |||
5409 | case NEON::BI__builtin_neon_vcvta_u16_v: | |||
5410 | case NEON::BI__builtin_neon_vcvta_u32_v: | |||
5411 | case NEON::BI__builtin_neon_vcvta_u64_v: | |||
5412 | case NEON::BI__builtin_neon_vcvtaq_s16_v: | |||
5413 | case NEON::BI__builtin_neon_vcvtaq_s32_v: | |||
5414 | case NEON::BI__builtin_neon_vcvtaq_s64_v: | |||
5415 | case NEON::BI__builtin_neon_vcvtaq_u16_v: | |||
5416 | case NEON::BI__builtin_neon_vcvtaq_u32_v: | |||
5417 | case NEON::BI__builtin_neon_vcvtaq_u64_v: | |||
5418 | case NEON::BI__builtin_neon_vcvtn_s16_v: | |||
5419 | case NEON::BI__builtin_neon_vcvtn_s32_v: | |||
5420 | case NEON::BI__builtin_neon_vcvtn_s64_v: | |||
5421 | case NEON::BI__builtin_neon_vcvtn_u16_v: | |||
5422 | case NEON::BI__builtin_neon_vcvtn_u32_v: | |||
5423 | case NEON::BI__builtin_neon_vcvtn_u64_v: | |||
5424 | case NEON::BI__builtin_neon_vcvtnq_s16_v: | |||
5425 | case NEON::BI__builtin_neon_vcvtnq_s32_v: | |||
5426 | case NEON::BI__builtin_neon_vcvtnq_s64_v: | |||
5427 | case NEON::BI__builtin_neon_vcvtnq_u16_v: | |||
5428 | case NEON::BI__builtin_neon_vcvtnq_u32_v: | |||
5429 | case NEON::BI__builtin_neon_vcvtnq_u64_v: | |||
5430 | case NEON::BI__builtin_neon_vcvtp_s16_v: | |||
5431 | case NEON::BI__builtin_neon_vcvtp_s32_v: | |||
5432 | case NEON::BI__builtin_neon_vcvtp_s64_v: | |||
5433 | case NEON::BI__builtin_neon_vcvtp_u16_v: | |||
5434 | case NEON::BI__builtin_neon_vcvtp_u32_v: | |||
5435 | case NEON::BI__builtin_neon_vcvtp_u64_v: | |||
5436 | case NEON::BI__builtin_neon_vcvtpq_s16_v: | |||
5437 | case NEON::BI__builtin_neon_vcvtpq_s32_v: | |||
5438 | case NEON::BI__builtin_neon_vcvtpq_s64_v: | |||
5439 | case NEON::BI__builtin_neon_vcvtpq_u16_v: | |||
5440 | case NEON::BI__builtin_neon_vcvtpq_u32_v: | |||
5441 | case NEON::BI__builtin_neon_vcvtpq_u64_v: | |||
5442 | case NEON::BI__builtin_neon_vcvtm_s16_v: | |||
5443 | case NEON::BI__builtin_neon_vcvtm_s32_v: | |||
5444 | case NEON::BI__builtin_neon_vcvtm_s64_v: | |||
5445 | case NEON::BI__builtin_neon_vcvtm_u16_v: | |||
5446 | case NEON::BI__builtin_neon_vcvtm_u32_v: | |||
5447 | case NEON::BI__builtin_neon_vcvtm_u64_v: | |||
5448 | case NEON::BI__builtin_neon_vcvtmq_s16_v: | |||
5449 | case NEON::BI__builtin_neon_vcvtmq_s32_v: | |||
5450 | case NEON::BI__builtin_neon_vcvtmq_s64_v: | |||
5451 | case NEON::BI__builtin_neon_vcvtmq_u16_v: | |||
5452 | case NEON::BI__builtin_neon_vcvtmq_u32_v: | |||
5453 | case NEON::BI__builtin_neon_vcvtmq_u64_v: { | |||
5454 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
5455 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint); | |||
5456 | } | |||
5457 | case NEON::BI__builtin_neon_vext_v: | |||
5458 | case NEON::BI__builtin_neon_vextq_v: { | |||
5459 | int CV = cast<ConstantInt>(Ops[2])->getSExtValue(); | |||
5460 | SmallVector<uint32_t, 16> Indices; | |||
5461 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) | |||
5462 | Indices.push_back(i+CV); | |||
5463 | ||||
5464 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
5465 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
5466 | return Builder.CreateShuffleVector(Ops[0], Ops[1], Indices, "vext"); | |||
5467 | } | |||
5468 | case NEON::BI__builtin_neon_vfma_v: | |||
5469 | case NEON::BI__builtin_neon_vfmaq_v: { | |||
5470 | Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty); | |||
5471 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
5472 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
5473 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
5474 | ||||
5475 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. | |||
5476 | return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]}); | |||
5477 | } | |||
5478 | case NEON::BI__builtin_neon_vld1_v: | |||
5479 | case NEON::BI__builtin_neon_vld1q_v: { | |||
5480 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; | |||
5481 | Ops.push_back(getAlignmentValue32(PtrOp0)); | |||
5482 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "vld1"); | |||
5483 | } | |||
5484 | case NEON::BI__builtin_neon_vld1_x2_v: | |||
5485 | case NEON::BI__builtin_neon_vld1q_x2_v: | |||
5486 | case NEON::BI__builtin_neon_vld1_x3_v: | |||
5487 | case NEON::BI__builtin_neon_vld1q_x3_v: | |||
5488 | case NEON::BI__builtin_neon_vld1_x4_v: | |||
5489 | case NEON::BI__builtin_neon_vld1q_x4_v: { | |||
5490 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getVectorElementType()); | |||
5491 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
5492 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
5493 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
5494 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld1xN"); | |||
5495 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
5496 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
5497 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
5498 | } | |||
5499 | case NEON::BI__builtin_neon_vld2_v: | |||
5500 | case NEON::BI__builtin_neon_vld2q_v: | |||
5501 | case NEON::BI__builtin_neon_vld3_v: | |||
5502 | case NEON::BI__builtin_neon_vld3q_v: | |||
5503 | case NEON::BI__builtin_neon_vld4_v: | |||
5504 | case NEON::BI__builtin_neon_vld4q_v: | |||
5505 | case NEON::BI__builtin_neon_vld2_dup_v: | |||
5506 | case NEON::BI__builtin_neon_vld2q_dup_v: | |||
5507 | case NEON::BI__builtin_neon_vld3_dup_v: | |||
5508 | case NEON::BI__builtin_neon_vld3q_dup_v: | |||
5509 | case NEON::BI__builtin_neon_vld4_dup_v: | |||
5510 | case NEON::BI__builtin_neon_vld4q_dup_v: { | |||
5511 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; | |||
5512 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
5513 | Value *Align = getAlignmentValue32(PtrOp1); | |||
5514 | Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, NameHint); | |||
5515 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
5516 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
5517 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
5518 | } | |||
5519 | case NEON::BI__builtin_neon_vld1_dup_v: | |||
5520 | case NEON::BI__builtin_neon_vld1q_dup_v: { | |||
5521 | Value *V = UndefValue::get(Ty); | |||
5522 | Ty = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
5523 | PtrOp0 = Builder.CreateBitCast(PtrOp0, Ty); | |||
5524 | LoadInst *Ld = Builder.CreateLoad(PtrOp0); | |||
5525 | llvm::Constant *CI = ConstantInt::get(SizeTy, 0); | |||
5526 | Ops[0] = Builder.CreateInsertElement(V, Ld, CI); | |||
5527 | return EmitNeonSplat(Ops[0], CI); | |||
5528 | } | |||
5529 | case NEON::BI__builtin_neon_vld2_lane_v: | |||
5530 | case NEON::BI__builtin_neon_vld2q_lane_v: | |||
5531 | case NEON::BI__builtin_neon_vld3_lane_v: | |||
5532 | case NEON::BI__builtin_neon_vld3q_lane_v: | |||
5533 | case NEON::BI__builtin_neon_vld4_lane_v: | |||
5534 | case NEON::BI__builtin_neon_vld4q_lane_v: { | |||
5535 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; | |||
5536 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
5537 | for (unsigned I = 2; I < Ops.size() - 1; ++I) | |||
5538 | Ops[I] = Builder.CreateBitCast(Ops[I], Ty); | |||
5539 | Ops.push_back(getAlignmentValue32(PtrOp1)); | |||
5540 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), NameHint); | |||
5541 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
5542 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
5543 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
5544 | } | |||
5545 | case NEON::BI__builtin_neon_vmovl_v: { | |||
5546 | llvm::Type *DTy =llvm::VectorType::getTruncatedElementVectorType(VTy); | |||
5547 | Ops[0] = Builder.CreateBitCast(Ops[0], DTy); | |||
5548 | if (Usgn) | |||
5549 | return Builder.CreateZExt(Ops[0], Ty, "vmovl"); | |||
5550 | return Builder.CreateSExt(Ops[0], Ty, "vmovl"); | |||
5551 | } | |||
5552 | case NEON::BI__builtin_neon_vmovn_v: { | |||
5553 | llvm::Type *QTy = llvm::VectorType::getExtendedElementVectorType(VTy); | |||
5554 | Ops[0] = Builder.CreateBitCast(Ops[0], QTy); | |||
5555 | return Builder.CreateTrunc(Ops[0], Ty, "vmovn"); | |||
5556 | } | |||
5557 | case NEON::BI__builtin_neon_vmull_v: | |||
5558 | // FIXME: the integer vmull operations could be emitted in terms of pure | |||
5559 | // LLVM IR (2 exts followed by a mul). Unfortunately LLVM has a habit of | |||
5560 | // hoisting the exts outside loops. Until global ISel comes along that can | |||
5561 | // see through such movement this leads to bad CodeGen. So we need an | |||
5562 | // intrinsic for now. | |||
5563 | Int = Usgn ? Intrinsic::arm_neon_vmullu : Intrinsic::arm_neon_vmulls; | |||
5564 | Int = Type.isPoly() ? (unsigned)Intrinsic::arm_neon_vmullp : Int; | |||
5565 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull"); | |||
5566 | case NEON::BI__builtin_neon_vpadal_v: | |||
5567 | case NEON::BI__builtin_neon_vpadalq_v: { | |||
5568 | // The source operand type has twice as many elements of half the size. | |||
5569 | unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits(); | |||
5570 | llvm::Type *EltTy = | |||
5571 | llvm::IntegerType::get(getLLVMContext(), EltBits / 2); | |||
5572 | llvm::Type *NarrowTy = | |||
5573 | llvm::VectorType::get(EltTy, VTy->getNumElements() * 2); | |||
5574 | llvm::Type *Tys[2] = { Ty, NarrowTy }; | |||
5575 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint); | |||
5576 | } | |||
5577 | case NEON::BI__builtin_neon_vpaddl_v: | |||
5578 | case NEON::BI__builtin_neon_vpaddlq_v: { | |||
5579 | // The source operand type has twice as many elements of half the size. | |||
5580 | unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits(); | |||
5581 | llvm::Type *EltTy = llvm::IntegerType::get(getLLVMContext(), EltBits / 2); | |||
5582 | llvm::Type *NarrowTy = | |||
5583 | llvm::VectorType::get(EltTy, VTy->getNumElements() * 2); | |||
5584 | llvm::Type *Tys[2] = { Ty, NarrowTy }; | |||
5585 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vpaddl"); | |||
5586 | } | |||
5587 | case NEON::BI__builtin_neon_vqdmlal_v: | |||
5588 | case NEON::BI__builtin_neon_vqdmlsl_v: { | |||
5589 | SmallVector<Value *, 2> MulOps(Ops.begin() + 1, Ops.end()); | |||
5590 | Ops[1] = | |||
5591 | EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), MulOps, "vqdmlal"); | |||
5592 | Ops.resize(2); | |||
5593 | return EmitNeonCall(CGM.getIntrinsic(AltLLVMIntrinsic, Ty), Ops, NameHint); | |||
5594 | } | |||
5595 | case NEON::BI__builtin_neon_vqshl_n_v: | |||
5596 | case NEON::BI__builtin_neon_vqshlq_n_v: | |||
5597 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshl_n", | |||
5598 | 1, false); | |||
5599 | case NEON::BI__builtin_neon_vqshlu_n_v: | |||
5600 | case NEON::BI__builtin_neon_vqshluq_n_v: | |||
5601 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshlu_n", | |||
5602 | 1, false); | |||
5603 | case NEON::BI__builtin_neon_vrecpe_v: | |||
5604 | case NEON::BI__builtin_neon_vrecpeq_v: | |||
5605 | case NEON::BI__builtin_neon_vrsqrte_v: | |||
5606 | case NEON::BI__builtin_neon_vrsqrteq_v: | |||
5607 | Int = Ty->isFPOrFPVectorTy() ? LLVMIntrinsic : AltLLVMIntrinsic; | |||
5608 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint); | |||
5609 | case NEON::BI__builtin_neon_vrndi_v: | |||
5610 | case NEON::BI__builtin_neon_vrndiq_v: | |||
5611 | Int = Intrinsic::nearbyint; | |||
5612 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint); | |||
5613 | case NEON::BI__builtin_neon_vrshr_n_v: | |||
5614 | case NEON::BI__builtin_neon_vrshrq_n_v: | |||
5615 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshr_n", | |||
5616 | 1, true); | |||
5617 | case NEON::BI__builtin_neon_vshl_n_v: | |||
5618 | case NEON::BI__builtin_neon_vshlq_n_v: | |||
5619 | Ops[1] = EmitNeonShiftVector(Ops[1], Ty, false); | |||
5620 | return Builder.CreateShl(Builder.CreateBitCast(Ops[0],Ty), Ops[1], | |||
5621 | "vshl_n"); | |||
5622 | case NEON::BI__builtin_neon_vshll_n_v: { | |||
5623 | llvm::Type *SrcTy = llvm::VectorType::getTruncatedElementVectorType(VTy); | |||
5624 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); | |||
5625 | if (Usgn) | |||
5626 | Ops[0] = Builder.CreateZExt(Ops[0], VTy); | |||
5627 | else | |||
5628 | Ops[0] = Builder.CreateSExt(Ops[0], VTy); | |||
5629 | Ops[1] = EmitNeonShiftVector(Ops[1], VTy, false); | |||
5630 | return Builder.CreateShl(Ops[0], Ops[1], "vshll_n"); | |||
5631 | } | |||
5632 | case NEON::BI__builtin_neon_vshrn_n_v: { | |||
5633 | llvm::Type *SrcTy = llvm::VectorType::getExtendedElementVectorType(VTy); | |||
5634 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); | |||
5635 | Ops[1] = EmitNeonShiftVector(Ops[1], SrcTy, false); | |||
5636 | if (Usgn) | |||
5637 | Ops[0] = Builder.CreateLShr(Ops[0], Ops[1]); | |||
5638 | else | |||
5639 | Ops[0] = Builder.CreateAShr(Ops[0], Ops[1]); | |||
5640 | return Builder.CreateTrunc(Ops[0], Ty, "vshrn_n"); | |||
5641 | } | |||
5642 | case NEON::BI__builtin_neon_vshr_n_v: | |||
5643 | case NEON::BI__builtin_neon_vshrq_n_v: | |||
5644 | return EmitNeonRShiftImm(Ops[0], Ops[1], Ty, Usgn, "vshr_n"); | |||
5645 | case NEON::BI__builtin_neon_vst1_v: | |||
5646 | case NEON::BI__builtin_neon_vst1q_v: | |||
5647 | case NEON::BI__builtin_neon_vst2_v: | |||
5648 | case NEON::BI__builtin_neon_vst2q_v: | |||
5649 | case NEON::BI__builtin_neon_vst3_v: | |||
5650 | case NEON::BI__builtin_neon_vst3q_v: | |||
5651 | case NEON::BI__builtin_neon_vst4_v: | |||
5652 | case NEON::BI__builtin_neon_vst4q_v: | |||
5653 | case NEON::BI__builtin_neon_vst2_lane_v: | |||
5654 | case NEON::BI__builtin_neon_vst2q_lane_v: | |||
5655 | case NEON::BI__builtin_neon_vst3_lane_v: | |||
5656 | case NEON::BI__builtin_neon_vst3q_lane_v: | |||
5657 | case NEON::BI__builtin_neon_vst4_lane_v: | |||
5658 | case NEON::BI__builtin_neon_vst4q_lane_v: { | |||
5659 | llvm::Type *Tys[] = {Int8PtrTy, Ty}; | |||
5660 | Ops.push_back(getAlignmentValue32(PtrOp0)); | |||
5661 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, ""); | |||
5662 | } | |||
5663 | case NEON::BI__builtin_neon_vst1_x2_v: | |||
5664 | case NEON::BI__builtin_neon_vst1q_x2_v: | |||
5665 | case NEON::BI__builtin_neon_vst1_x3_v: | |||
5666 | case NEON::BI__builtin_neon_vst1q_x3_v: | |||
5667 | case NEON::BI__builtin_neon_vst1_x4_v: | |||
5668 | case NEON::BI__builtin_neon_vst1q_x4_v: { | |||
5669 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getVectorElementType()); | |||
5670 | // TODO: Currently in AArch32 mode the pointer operand comes first, whereas | |||
5671 | // in AArch64 it comes last. We may want to stick to one or another. | |||
5672 | if (Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_be) { | |||
5673 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
5674 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); | |||
5675 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, ""); | |||
5676 | } | |||
5677 | llvm::Type *Tys[2] = { PTy, VTy }; | |||
5678 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, ""); | |||
5679 | } | |||
5680 | case NEON::BI__builtin_neon_vsubhn_v: { | |||
5681 | llvm::VectorType *SrcTy = | |||
5682 | llvm::VectorType::getExtendedElementVectorType(VTy); | |||
5683 | ||||
5684 | // %sum = add <4 x i32> %lhs, %rhs | |||
5685 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); | |||
5686 | Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy); | |||
5687 | Ops[0] = Builder.CreateSub(Ops[0], Ops[1], "vsubhn"); | |||
5688 | ||||
5689 | // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16> | |||
5690 | Constant *ShiftAmt = | |||
5691 | ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2); | |||
5692 | Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vsubhn"); | |||
5693 | ||||
5694 | // %res = trunc <4 x i32> %high to <4 x i16> | |||
5695 | return Builder.CreateTrunc(Ops[0], VTy, "vsubhn"); | |||
5696 | } | |||
5697 | case NEON::BI__builtin_neon_vtrn_v: | |||
5698 | case NEON::BI__builtin_neon_vtrnq_v: { | |||
5699 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
5700 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
5701 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
5702 | Value *SV = nullptr; | |||
5703 | ||||
5704 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
5705 | SmallVector<uint32_t, 16> Indices; | |||
5706 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { | |||
5707 | Indices.push_back(i+vi); | |||
5708 | Indices.push_back(i+e+vi); | |||
5709 | } | |||
5710 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
5711 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn"); | |||
5712 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
5713 | } | |||
5714 | return SV; | |||
5715 | } | |||
5716 | case NEON::BI__builtin_neon_vtst_v: | |||
5717 | case NEON::BI__builtin_neon_vtstq_v: { | |||
5718 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
5719 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
5720 | Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]); | |||
5721 | Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0], | |||
5722 | ConstantAggregateZero::get(Ty)); | |||
5723 | return Builder.CreateSExt(Ops[0], Ty, "vtst"); | |||
5724 | } | |||
5725 | case NEON::BI__builtin_neon_vuzp_v: | |||
5726 | case NEON::BI__builtin_neon_vuzpq_v: { | |||
5727 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
5728 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
5729 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
5730 | Value *SV = nullptr; | |||
5731 | ||||
5732 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
5733 | SmallVector<uint32_t, 16> Indices; | |||
5734 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) | |||
5735 | Indices.push_back(2*i+vi); | |||
5736 | ||||
5737 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
5738 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp"); | |||
5739 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
5740 | } | |||
5741 | return SV; | |||
5742 | } | |||
5743 | case NEON::BI__builtin_neon_vzip_v: | |||
5744 | case NEON::BI__builtin_neon_vzipq_v: { | |||
5745 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
5746 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
5747 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
5748 | Value *SV = nullptr; | |||
5749 | ||||
5750 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
5751 | SmallVector<uint32_t, 16> Indices; | |||
5752 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { | |||
5753 | Indices.push_back((i + vi*e) >> 1); | |||
5754 | Indices.push_back(((i + vi*e) >> 1)+e); | |||
5755 | } | |||
5756 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
5757 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip"); | |||
5758 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
5759 | } | |||
5760 | return SV; | |||
5761 | } | |||
5762 | case NEON::BI__builtin_neon_vdot_v: | |||
5763 | case NEON::BI__builtin_neon_vdotq_v: { | |||
5764 | llvm::Type *InputTy = | |||
5765 | llvm::VectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8); | |||
5766 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
5767 | Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic; | |||
5768 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vdot"); | |||
5769 | } | |||
5770 | case NEON::BI__builtin_neon_vfmlal_low_v: | |||
5771 | case NEON::BI__builtin_neon_vfmlalq_low_v: { | |||
5772 | llvm::Type *InputTy = | |||
5773 | llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); | |||
5774 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
5775 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_low"); | |||
5776 | } | |||
5777 | case NEON::BI__builtin_neon_vfmlsl_low_v: | |||
5778 | case NEON::BI__builtin_neon_vfmlslq_low_v: { | |||
5779 | llvm::Type *InputTy = | |||
5780 | llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); | |||
5781 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
5782 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_low"); | |||
5783 | } | |||
5784 | case NEON::BI__builtin_neon_vfmlal_high_v: | |||
5785 | case NEON::BI__builtin_neon_vfmlalq_high_v: { | |||
5786 | llvm::Type *InputTy = | |||
5787 | llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); | |||
5788 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
5789 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_high"); | |||
5790 | } | |||
5791 | case NEON::BI__builtin_neon_vfmlsl_high_v: | |||
5792 | case NEON::BI__builtin_neon_vfmlslq_high_v: { | |||
5793 | llvm::Type *InputTy = | |||
5794 | llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); | |||
5795 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
5796 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_high"); | |||
5797 | } | |||
5798 | } | |||
5799 | ||||
5800 | assert(Int && "Expected valid intrinsic number")((Int && "Expected valid intrinsic number") ? static_cast <void> (0) : __assert_fail ("Int && \"Expected valid intrinsic number\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5800, __PRETTY_FUNCTION__)); | |||
5801 | ||||
5802 | // Determine the type(s) of this overloaded AArch64 intrinsic. | |||
5803 | Function *F = LookupNeonLLVMIntrinsic(Int, Modifier, Ty, E); | |||
5804 | ||||
5805 | Value *Result = EmitNeonCall(F, Ops, NameHint); | |||
5806 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
5807 | // AArch64 intrinsic one-element vector type cast to | |||
5808 | // scalar type expected by the builtin | |||
5809 | return Builder.CreateBitCast(Result, ResultType, NameHint); | |||
5810 | } | |||
5811 | ||||
5812 | Value *CodeGenFunction::EmitAArch64CompareBuiltinExpr( | |||
5813 | Value *Op, llvm::Type *Ty, const CmpInst::Predicate Fp, | |||
5814 | const CmpInst::Predicate Ip, const Twine &Name) { | |||
5815 | llvm::Type *OTy = Op->getType(); | |||
5816 | ||||
5817 | // FIXME: this is utterly horrific. We should not be looking at previous | |||
5818 | // codegen context to find out what needs doing. Unfortunately TableGen | |||
5819 | // currently gives us exactly the same calls for vceqz_f32 and vceqz_s32 | |||
5820 | // (etc). | |||
5821 | if (BitCastInst *BI = dyn_cast<BitCastInst>(Op)) | |||
5822 | OTy = BI->getOperand(0)->getType(); | |||
5823 | ||||
5824 | Op = Builder.CreateBitCast(Op, OTy); | |||
5825 | if (OTy->getScalarType()->isFloatingPointTy()) { | |||
5826 | Op = Builder.CreateFCmp(Fp, Op, Constant::getNullValue(OTy)); | |||
5827 | } else { | |||
5828 | Op = Builder.CreateICmp(Ip, Op, Constant::getNullValue(OTy)); | |||
5829 | } | |||
5830 | return Builder.CreateSExt(Op, Ty, Name); | |||
5831 | } | |||
5832 | ||||
5833 | static Value *packTBLDVectorList(CodeGenFunction &CGF, ArrayRef<Value *> Ops, | |||
5834 | Value *ExtOp, Value *IndexOp, | |||
5835 | llvm::Type *ResTy, unsigned IntID, | |||
5836 | const char *Name) { | |||
5837 | SmallVector<Value *, 2> TblOps; | |||
5838 | if (ExtOp) | |||
5839 | TblOps.push_back(ExtOp); | |||
5840 | ||||
5841 | // Build a vector containing sequential number like (0, 1, 2, ..., 15) | |||
5842 | SmallVector<uint32_t, 16> Indices; | |||
5843 | llvm::VectorType *TblTy = cast<llvm::VectorType>(Ops[0]->getType()); | |||
5844 | for (unsigned i = 0, e = TblTy->getNumElements(); i != e; ++i) { | |||
5845 | Indices.push_back(2*i); | |||
5846 | Indices.push_back(2*i+1); | |||
5847 | } | |||
5848 | ||||
5849 | int PairPos = 0, End = Ops.size() - 1; | |||
5850 | while (PairPos < End) { | |||
5851 | TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos], | |||
5852 | Ops[PairPos+1], Indices, | |||
5853 | Name)); | |||
5854 | PairPos += 2; | |||
5855 | } | |||
5856 | ||||
5857 | // If there's an odd number of 64-bit lookup table, fill the high 64-bit | |||
5858 | // of the 128-bit lookup table with zero. | |||
5859 | if (PairPos == End) { | |||
5860 | Value *ZeroTbl = ConstantAggregateZero::get(TblTy); | |||
5861 | TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos], | |||
5862 | ZeroTbl, Indices, Name)); | |||
5863 | } | |||
5864 | ||||
5865 | Function *TblF; | |||
5866 | TblOps.push_back(IndexOp); | |||
5867 | TblF = CGF.CGM.getIntrinsic(IntID, ResTy); | |||
5868 | ||||
5869 | return CGF.EmitNeonCall(TblF, TblOps, Name); | |||
5870 | } | |||
5871 | ||||
5872 | Value *CodeGenFunction::GetValueForARMHint(unsigned BuiltinID) { | |||
5873 | unsigned Value; | |||
5874 | switch (BuiltinID) { | |||
5875 | default: | |||
5876 | return nullptr; | |||
5877 | case ARM::BI__builtin_arm_nop: | |||
5878 | Value = 0; | |||
5879 | break; | |||
5880 | case ARM::BI__builtin_arm_yield: | |||
5881 | case ARM::BI__yield: | |||
5882 | Value = 1; | |||
5883 | break; | |||
5884 | case ARM::BI__builtin_arm_wfe: | |||
5885 | case ARM::BI__wfe: | |||
5886 | Value = 2; | |||
5887 | break; | |||
5888 | case ARM::BI__builtin_arm_wfi: | |||
5889 | case ARM::BI__wfi: | |||
5890 | Value = 3; | |||
5891 | break; | |||
5892 | case ARM::BI__builtin_arm_sev: | |||
5893 | case ARM::BI__sev: | |||
5894 | Value = 4; | |||
5895 | break; | |||
5896 | case ARM::BI__builtin_arm_sevl: | |||
5897 | case ARM::BI__sevl: | |||
5898 | Value = 5; | |||
5899 | break; | |||
5900 | } | |||
5901 | ||||
5902 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_hint), | |||
5903 | llvm::ConstantInt::get(Int32Ty, Value)); | |||
5904 | } | |||
5905 | ||||
5906 | // Generates the IR for the read/write special register builtin, | |||
5907 | // ValueType is the type of the value that is to be written or read, | |||
5908 | // RegisterType is the type of the register being written to or read from. | |||
5909 | static Value *EmitSpecialRegisterBuiltin(CodeGenFunction &CGF, | |||
5910 | const CallExpr *E, | |||
5911 | llvm::Type *RegisterType, | |||
5912 | llvm::Type *ValueType, | |||
5913 | bool IsRead, | |||
5914 | StringRef SysReg = "") { | |||
5915 | // write and register intrinsics only support 32 and 64 bit operations. | |||
5916 | assert((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64))(((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy (64)) && "Unsupported size for register.") ? static_cast <void> (0) : __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5917, __PRETTY_FUNCTION__)) | |||
5917 | && "Unsupported size for register.")(((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy (64)) && "Unsupported size for register.") ? static_cast <void> (0) : __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5917, __PRETTY_FUNCTION__)); | |||
5918 | ||||
5919 | CodeGen::CGBuilderTy &Builder = CGF.Builder; | |||
5920 | CodeGen::CodeGenModule &CGM = CGF.CGM; | |||
5921 | LLVMContext &Context = CGM.getLLVMContext(); | |||
5922 | ||||
5923 | if (SysReg.empty()) { | |||
5924 | const Expr *SysRegStrExpr = E->getArg(0)->IgnoreParenCasts(); | |||
5925 | SysReg = cast<clang::StringLiteral>(SysRegStrExpr)->getString(); | |||
5926 | } | |||
5927 | ||||
5928 | llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysReg) }; | |||
5929 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); | |||
5930 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); | |||
5931 | ||||
5932 | llvm::Type *Types[] = { RegisterType }; | |||
5933 | ||||
5934 | bool MixedTypes = RegisterType->isIntegerTy(64) && ValueType->isIntegerTy(32); | |||
5935 | assert(!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64))((!(RegisterType->isIntegerTy(32) && ValueType-> isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register" ) ? static_cast<void> (0) : __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5936, __PRETTY_FUNCTION__)) | |||
5936 | && "Can't fit 64-bit value in 32-bit register")((!(RegisterType->isIntegerTy(32) && ValueType-> isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register" ) ? static_cast<void> (0) : __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5936, __PRETTY_FUNCTION__)); | |||
5937 | ||||
5938 | if (IsRead) { | |||
5939 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types); | |||
5940 | llvm::Value *Call = Builder.CreateCall(F, Metadata); | |||
5941 | ||||
5942 | if (MixedTypes) | |||
5943 | // Read into 64 bit register and then truncate result to 32 bit. | |||
5944 | return Builder.CreateTrunc(Call, ValueType); | |||
5945 | ||||
5946 | if (ValueType->isPointerTy()) | |||
5947 | // Have i32/i64 result (Call) but want to return a VoidPtrTy (i8*). | |||
5948 | return Builder.CreateIntToPtr(Call, ValueType); | |||
5949 | ||||
5950 | return Call; | |||
5951 | } | |||
5952 | ||||
5953 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types); | |||
5954 | llvm::Value *ArgValue = CGF.EmitScalarExpr(E->getArg(1)); | |||
5955 | if (MixedTypes) { | |||
5956 | // Extend 32 bit write value to 64 bit to pass to write. | |||
5957 | ArgValue = Builder.CreateZExt(ArgValue, RegisterType); | |||
5958 | return Builder.CreateCall(F, { Metadata, ArgValue }); | |||
5959 | } | |||
5960 | ||||
5961 | if (ValueType->isPointerTy()) { | |||
5962 | // Have VoidPtrTy ArgValue but want to return an i32/i64. | |||
5963 | ArgValue = Builder.CreatePtrToInt(ArgValue, RegisterType); | |||
5964 | return Builder.CreateCall(F, { Metadata, ArgValue }); | |||
5965 | } | |||
5966 | ||||
5967 | return Builder.CreateCall(F, { Metadata, ArgValue }); | |||
5968 | } | |||
5969 | ||||
5970 | /// Return true if BuiltinID is an overloaded Neon intrinsic with an extra | |||
5971 | /// argument that specifies the vector type. | |||
5972 | static bool HasExtraNeonArgument(unsigned BuiltinID) { | |||
5973 | switch (BuiltinID) { | |||
5974 | default: break; | |||
5975 | case NEON::BI__builtin_neon_vget_lane_i8: | |||
5976 | case NEON::BI__builtin_neon_vget_lane_i16: | |||
5977 | case NEON::BI__builtin_neon_vget_lane_i32: | |||
5978 | case NEON::BI__builtin_neon_vget_lane_i64: | |||
5979 | case NEON::BI__builtin_neon_vget_lane_f32: | |||
5980 | case NEON::BI__builtin_neon_vgetq_lane_i8: | |||
5981 | case NEON::BI__builtin_neon_vgetq_lane_i16: | |||
5982 | case NEON::BI__builtin_neon_vgetq_lane_i32: | |||
5983 | case NEON::BI__builtin_neon_vgetq_lane_i64: | |||
5984 | case NEON::BI__builtin_neon_vgetq_lane_f32: | |||
5985 | case NEON::BI__builtin_neon_vset_lane_i8: | |||
5986 | case NEON::BI__builtin_neon_vset_lane_i16: | |||
5987 | case NEON::BI__builtin_neon_vset_lane_i32: | |||
5988 | case NEON::BI__builtin_neon_vset_lane_i64: | |||
5989 | case NEON::BI__builtin_neon_vset_lane_f32: | |||
5990 | case NEON::BI__builtin_neon_vsetq_lane_i8: | |||
5991 | case NEON::BI__builtin_neon_vsetq_lane_i16: | |||
5992 | case NEON::BI__builtin_neon_vsetq_lane_i32: | |||
5993 | case NEON::BI__builtin_neon_vsetq_lane_i64: | |||
5994 | case NEON::BI__builtin_neon_vsetq_lane_f32: | |||
5995 | case NEON::BI__builtin_neon_vsha1h_u32: | |||
5996 | case NEON::BI__builtin_neon_vsha1cq_u32: | |||
5997 | case NEON::BI__builtin_neon_vsha1pq_u32: | |||
5998 | case NEON::BI__builtin_neon_vsha1mq_u32: | |||
5999 | case clang::ARM::BI_MoveToCoprocessor: | |||
6000 | case clang::ARM::BI_MoveToCoprocessor2: | |||
6001 | return false; | |||
6002 | } | |||
6003 | return true; | |||
6004 | } | |||
6005 | ||||
6006 | Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID, | |||
6007 | const CallExpr *E, | |||
6008 | llvm::Triple::ArchType Arch) { | |||
6009 | if (auto Hint = GetValueForARMHint(BuiltinID)) | |||
6010 | return Hint; | |||
6011 | ||||
6012 | if (BuiltinID == ARM::BI__emit) { | |||
6013 | bool IsThumb = getTarget().getTriple().getArch() == llvm::Triple::thumb; | |||
6014 | llvm::FunctionType *FTy = | |||
6015 | llvm::FunctionType::get(VoidTy, /*Variadic=*/false); | |||
6016 | ||||
6017 | Expr::EvalResult Result; | |||
6018 | if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext())) | |||
6019 | llvm_unreachable("Sema will ensure that the parameter is constant")::llvm::llvm_unreachable_internal("Sema will ensure that the parameter is constant" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6019); | |||
6020 | ||||
6021 | llvm::APSInt Value = Result.Val.getInt(); | |||
6022 | uint64_t ZExtValue = Value.zextOrTrunc(IsThumb ? 16 : 32).getZExtValue(); | |||
6023 | ||||
6024 | llvm::InlineAsm *Emit = | |||
6025 | IsThumb ? InlineAsm::get(FTy, ".inst.n 0x" + utohexstr(ZExtValue), "", | |||
6026 | /*hasSideEffects=*/true) | |||
6027 | : InlineAsm::get(FTy, ".inst 0x" + utohexstr(ZExtValue), "", | |||
6028 | /*hasSideEffects=*/true); | |||
6029 | ||||
6030 | return Builder.CreateCall(Emit); | |||
6031 | } | |||
6032 | ||||
6033 | if (BuiltinID == ARM::BI__builtin_arm_dbg) { | |||
6034 | Value *Option = EmitScalarExpr(E->getArg(0)); | |||
6035 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_dbg), Option); | |||
6036 | } | |||
6037 | ||||
6038 | if (BuiltinID == ARM::BI__builtin_arm_prefetch) { | |||
6039 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
6040 | Value *RW = EmitScalarExpr(E->getArg(1)); | |||
6041 | Value *IsData = EmitScalarExpr(E->getArg(2)); | |||
6042 | ||||
6043 | // Locality is not supported on ARM target | |||
6044 | Value *Locality = llvm::ConstantInt::get(Int32Ty, 3); | |||
6045 | ||||
6046 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); | |||
6047 | return Builder.CreateCall(F, {Address, RW, Locality, IsData}); | |||
6048 | } | |||
6049 | ||||
6050 | if (BuiltinID == ARM::BI__builtin_arm_rbit) { | |||
6051 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
6052 | return Builder.CreateCall( | |||
6053 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); | |||
6054 | } | |||
6055 | ||||
6056 | if (BuiltinID == ARM::BI__clear_cache) { | |||
6057 | assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments")((E->getNumArgs() == 2 && "__clear_cache takes 2 arguments" ) ? static_cast<void> (0) : __assert_fail ("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6057, __PRETTY_FUNCTION__)); | |||
6058 | const FunctionDecl *FD = E->getDirectCallee(); | |||
6059 | Value *Ops[2]; | |||
6060 | for (unsigned i = 0; i < 2; i++) | |||
6061 | Ops[i] = EmitScalarExpr(E->getArg(i)); | |||
6062 | llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType()); | |||
6063 | llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty); | |||
6064 | StringRef Name = FD->getName(); | |||
6065 | return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops); | |||
6066 | } | |||
6067 | ||||
6068 | if (BuiltinID == ARM::BI__builtin_arm_mcrr || | |||
6069 | BuiltinID == ARM::BI__builtin_arm_mcrr2) { | |||
6070 | Function *F; | |||
6071 | ||||
6072 | switch (BuiltinID) { | |||
6073 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6073); | |||
6074 | case ARM::BI__builtin_arm_mcrr: | |||
6075 | F = CGM.getIntrinsic(Intrinsic::arm_mcrr); | |||
6076 | break; | |||
6077 | case ARM::BI__builtin_arm_mcrr2: | |||
6078 | F = CGM.getIntrinsic(Intrinsic::arm_mcrr2); | |||
6079 | break; | |||
6080 | } | |||
6081 | ||||
6082 | // MCRR{2} instruction has 5 operands but | |||
6083 | // the intrinsic has 4 because Rt and Rt2 | |||
6084 | // are represented as a single unsigned 64 | |||
6085 | // bit integer in the intrinsic definition | |||
6086 | // but internally it's represented as 2 32 | |||
6087 | // bit integers. | |||
6088 | ||||
6089 | Value *Coproc = EmitScalarExpr(E->getArg(0)); | |||
6090 | Value *Opc1 = EmitScalarExpr(E->getArg(1)); | |||
6091 | Value *RtAndRt2 = EmitScalarExpr(E->getArg(2)); | |||
6092 | Value *CRm = EmitScalarExpr(E->getArg(3)); | |||
6093 | ||||
6094 | Value *C1 = llvm::ConstantInt::get(Int64Ty, 32); | |||
6095 | Value *Rt = Builder.CreateTruncOrBitCast(RtAndRt2, Int32Ty); | |||
6096 | Value *Rt2 = Builder.CreateLShr(RtAndRt2, C1); | |||
6097 | Rt2 = Builder.CreateTruncOrBitCast(Rt2, Int32Ty); | |||
6098 | ||||
6099 | return Builder.CreateCall(F, {Coproc, Opc1, Rt, Rt2, CRm}); | |||
6100 | } | |||
6101 | ||||
6102 | if (BuiltinID == ARM::BI__builtin_arm_mrrc || | |||
6103 | BuiltinID == ARM::BI__builtin_arm_mrrc2) { | |||
6104 | Function *F; | |||
6105 | ||||
6106 | switch (BuiltinID) { | |||
6107 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6107); | |||
6108 | case ARM::BI__builtin_arm_mrrc: | |||
6109 | F = CGM.getIntrinsic(Intrinsic::arm_mrrc); | |||
6110 | break; | |||
6111 | case ARM::BI__builtin_arm_mrrc2: | |||
6112 | F = CGM.getIntrinsic(Intrinsic::arm_mrrc2); | |||
6113 | break; | |||
6114 | } | |||
6115 | ||||
6116 | Value *Coproc = EmitScalarExpr(E->getArg(0)); | |||
6117 | Value *Opc1 = EmitScalarExpr(E->getArg(1)); | |||
6118 | Value *CRm = EmitScalarExpr(E->getArg(2)); | |||
6119 | Value *RtAndRt2 = Builder.CreateCall(F, {Coproc, Opc1, CRm}); | |||
6120 | ||||
6121 | // Returns an unsigned 64 bit integer, represented | |||
6122 | // as two 32 bit integers. | |||
6123 | ||||
6124 | Value *Rt = Builder.CreateExtractValue(RtAndRt2, 1); | |||
6125 | Value *Rt1 = Builder.CreateExtractValue(RtAndRt2, 0); | |||
6126 | Rt = Builder.CreateZExt(Rt, Int64Ty); | |||
6127 | Rt1 = Builder.CreateZExt(Rt1, Int64Ty); | |||
6128 | ||||
6129 | Value *ShiftCast = llvm::ConstantInt::get(Int64Ty, 32); | |||
6130 | RtAndRt2 = Builder.CreateShl(Rt, ShiftCast, "shl", true); | |||
6131 | RtAndRt2 = Builder.CreateOr(RtAndRt2, Rt1); | |||
6132 | ||||
6133 | return Builder.CreateBitCast(RtAndRt2, ConvertType(E->getType())); | |||
6134 | } | |||
6135 | ||||
6136 | if (BuiltinID == ARM::BI__builtin_arm_ldrexd || | |||
6137 | ((BuiltinID == ARM::BI__builtin_arm_ldrex || | |||
6138 | BuiltinID == ARM::BI__builtin_arm_ldaex) && | |||
6139 | getContext().getTypeSize(E->getType()) == 64) || | |||
6140 | BuiltinID == ARM::BI__ldrexd) { | |||
6141 | Function *F; | |||
6142 | ||||
6143 | switch (BuiltinID) { | |||
6144 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6144); | |||
6145 | case ARM::BI__builtin_arm_ldaex: | |||
6146 | F = CGM.getIntrinsic(Intrinsic::arm_ldaexd); | |||
6147 | break; | |||
6148 | case ARM::BI__builtin_arm_ldrexd: | |||
6149 | case ARM::BI__builtin_arm_ldrex: | |||
6150 | case ARM::BI__ldrexd: | |||
6151 | F = CGM.getIntrinsic(Intrinsic::arm_ldrexd); | |||
6152 | break; | |||
6153 | } | |||
6154 | ||||
6155 | Value *LdPtr = EmitScalarExpr(E->getArg(0)); | |||
6156 | Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy), | |||
6157 | "ldrexd"); | |||
6158 | ||||
6159 | Value *Val0 = Builder.CreateExtractValue(Val, 1); | |||
6160 | Value *Val1 = Builder.CreateExtractValue(Val, 0); | |||
6161 | Val0 = Builder.CreateZExt(Val0, Int64Ty); | |||
6162 | Val1 = Builder.CreateZExt(Val1, Int64Ty); | |||
6163 | ||||
6164 | Value *ShiftCst = llvm::ConstantInt::get(Int64Ty, 32); | |||
6165 | Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */); | |||
6166 | Val = Builder.CreateOr(Val, Val1); | |||
6167 | return Builder.CreateBitCast(Val, ConvertType(E->getType())); | |||
6168 | } | |||
6169 | ||||
6170 | if (BuiltinID == ARM::BI__builtin_arm_ldrex || | |||
6171 | BuiltinID == ARM::BI__builtin_arm_ldaex) { | |||
6172 | Value *LoadAddr = EmitScalarExpr(E->getArg(0)); | |||
6173 | ||||
6174 | QualType Ty = E->getType(); | |||
6175 | llvm::Type *RealResTy = ConvertType(Ty); | |||
6176 | llvm::Type *PtrTy = llvm::IntegerType::get( | |||
6177 | getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo(); | |||
6178 | LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy); | |||
6179 | ||||
6180 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_ldaex | |||
6181 | ? Intrinsic::arm_ldaex | |||
6182 | : Intrinsic::arm_ldrex, | |||
6183 | PtrTy); | |||
6184 | Value *Val = Builder.CreateCall(F, LoadAddr, "ldrex"); | |||
6185 | ||||
6186 | if (RealResTy->isPointerTy()) | |||
6187 | return Builder.CreateIntToPtr(Val, RealResTy); | |||
6188 | else { | |||
6189 | llvm::Type *IntResTy = llvm::IntegerType::get( | |||
6190 | getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy)); | |||
6191 | Val = Builder.CreateTruncOrBitCast(Val, IntResTy); | |||
6192 | return Builder.CreateBitCast(Val, RealResTy); | |||
6193 | } | |||
6194 | } | |||
6195 | ||||
6196 | if (BuiltinID == ARM::BI__builtin_arm_strexd || | |||
6197 | ((BuiltinID == ARM::BI__builtin_arm_stlex || | |||
6198 | BuiltinID == ARM::BI__builtin_arm_strex) && | |||
6199 | getContext().getTypeSize(E->getArg(0)->getType()) == 64)) { | |||
6200 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex | |||
6201 | ? Intrinsic::arm_stlexd | |||
6202 | : Intrinsic::arm_strexd); | |||
6203 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty); | |||
6204 | ||||
6205 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); | |||
6206 | Value *Val = EmitScalarExpr(E->getArg(0)); | |||
6207 | Builder.CreateStore(Val, Tmp); | |||
6208 | ||||
6209 | Address LdPtr = Builder.CreateBitCast(Tmp,llvm::PointerType::getUnqual(STy)); | |||
6210 | Val = Builder.CreateLoad(LdPtr); | |||
6211 | ||||
6212 | Value *Arg0 = Builder.CreateExtractValue(Val, 0); | |||
6213 | Value *Arg1 = Builder.CreateExtractValue(Val, 1); | |||
6214 | Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), Int8PtrTy); | |||
6215 | return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "strexd"); | |||
6216 | } | |||
6217 | ||||
6218 | if (BuiltinID == ARM::BI__builtin_arm_strex || | |||
6219 | BuiltinID == ARM::BI__builtin_arm_stlex) { | |||
6220 | Value *StoreVal = EmitScalarExpr(E->getArg(0)); | |||
6221 | Value *StoreAddr = EmitScalarExpr(E->getArg(1)); | |||
6222 | ||||
6223 | QualType Ty = E->getArg(0)->getType(); | |||
6224 | llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(), | |||
6225 | getContext().getTypeSize(Ty)); | |||
6226 | StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo()); | |||
6227 | ||||
6228 | if (StoreVal->getType()->isPointerTy()) | |||
6229 | StoreVal = Builder.CreatePtrToInt(StoreVal, Int32Ty); | |||
6230 | else { | |||
6231 | llvm::Type *IntTy = llvm::IntegerType::get( | |||
6232 | getLLVMContext(), | |||
6233 | CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType())); | |||
6234 | StoreVal = Builder.CreateBitCast(StoreVal, IntTy); | |||
6235 | StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int32Ty); | |||
6236 | } | |||
6237 | ||||
6238 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex | |||
6239 | ? Intrinsic::arm_stlex | |||
6240 | : Intrinsic::arm_strex, | |||
6241 | StoreAddr->getType()); | |||
6242 | return Builder.CreateCall(F, {StoreVal, StoreAddr}, "strex"); | |||
6243 | } | |||
6244 | ||||
6245 | if (BuiltinID == ARM::BI__builtin_arm_clrex) { | |||
6246 | Function *F = CGM.getIntrinsic(Intrinsic::arm_clrex); | |||
6247 | return Builder.CreateCall(F); | |||
6248 | } | |||
6249 | ||||
6250 | // CRC32 | |||
6251 | Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic; | |||
6252 | switch (BuiltinID) { | |||
6253 | case ARM::BI__builtin_arm_crc32b: | |||
6254 | CRCIntrinsicID = Intrinsic::arm_crc32b; break; | |||
6255 | case ARM::BI__builtin_arm_crc32cb: | |||
6256 | CRCIntrinsicID = Intrinsic::arm_crc32cb; break; | |||
6257 | case ARM::BI__builtin_arm_crc32h: | |||
6258 | CRCIntrinsicID = Intrinsic::arm_crc32h; break; | |||
6259 | case ARM::BI__builtin_arm_crc32ch: | |||
6260 | CRCIntrinsicID = Intrinsic::arm_crc32ch; break; | |||
6261 | case ARM::BI__builtin_arm_crc32w: | |||
6262 | case ARM::BI__builtin_arm_crc32d: | |||
6263 | CRCIntrinsicID = Intrinsic::arm_crc32w; break; | |||
6264 | case ARM::BI__builtin_arm_crc32cw: | |||
6265 | case ARM::BI__builtin_arm_crc32cd: | |||
6266 | CRCIntrinsicID = Intrinsic::arm_crc32cw; break; | |||
6267 | } | |||
6268 | ||||
6269 | if (CRCIntrinsicID != Intrinsic::not_intrinsic) { | |||
6270 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); | |||
6271 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
6272 | ||||
6273 | // crc32{c,}d intrinsics are implemnted as two calls to crc32{c,}w | |||
6274 | // intrinsics, hence we need different codegen for these cases. | |||
6275 | if (BuiltinID == ARM::BI__builtin_arm_crc32d || | |||
6276 | BuiltinID == ARM::BI__builtin_arm_crc32cd) { | |||
6277 | Value *C1 = llvm::ConstantInt::get(Int64Ty, 32); | |||
6278 | Value *Arg1a = Builder.CreateTruncOrBitCast(Arg1, Int32Ty); | |||
6279 | Value *Arg1b = Builder.CreateLShr(Arg1, C1); | |||
6280 | Arg1b = Builder.CreateTruncOrBitCast(Arg1b, Int32Ty); | |||
6281 | ||||
6282 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); | |||
6283 | Value *Res = Builder.CreateCall(F, {Arg0, Arg1a}); | |||
6284 | return Builder.CreateCall(F, {Res, Arg1b}); | |||
6285 | } else { | |||
6286 | Arg1 = Builder.CreateZExtOrBitCast(Arg1, Int32Ty); | |||
6287 | ||||
6288 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); | |||
6289 | return Builder.CreateCall(F, {Arg0, Arg1}); | |||
6290 | } | |||
6291 | } | |||
6292 | ||||
6293 | if (BuiltinID == ARM::BI__builtin_arm_rsr || | |||
6294 | BuiltinID == ARM::BI__builtin_arm_rsr64 || | |||
6295 | BuiltinID == ARM::BI__builtin_arm_rsrp || | |||
6296 | BuiltinID == ARM::BI__builtin_arm_wsr || | |||
6297 | BuiltinID == ARM::BI__builtin_arm_wsr64 || | |||
6298 | BuiltinID == ARM::BI__builtin_arm_wsrp) { | |||
6299 | ||||
6300 | bool IsRead = BuiltinID == ARM::BI__builtin_arm_rsr || | |||
6301 | BuiltinID == ARM::BI__builtin_arm_rsr64 || | |||
6302 | BuiltinID == ARM::BI__builtin_arm_rsrp; | |||
6303 | ||||
6304 | bool IsPointerBuiltin = BuiltinID == ARM::BI__builtin_arm_rsrp || | |||
6305 | BuiltinID == ARM::BI__builtin_arm_wsrp; | |||
6306 | ||||
6307 | bool Is64Bit = BuiltinID == ARM::BI__builtin_arm_rsr64 || | |||
6308 | BuiltinID == ARM::BI__builtin_arm_wsr64; | |||
6309 | ||||
6310 | llvm::Type *ValueType; | |||
6311 | llvm::Type *RegisterType; | |||
6312 | if (IsPointerBuiltin) { | |||
6313 | ValueType = VoidPtrTy; | |||
6314 | RegisterType = Int32Ty; | |||
6315 | } else if (Is64Bit) { | |||
6316 | ValueType = RegisterType = Int64Ty; | |||
6317 | } else { | |||
6318 | ValueType = RegisterType = Int32Ty; | |||
6319 | } | |||
6320 | ||||
6321 | return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, IsRead); | |||
6322 | } | |||
6323 | ||||
6324 | // Find out if any arguments are required to be integer constant | |||
6325 | // expressions. | |||
6326 | unsigned ICEArguments = 0; | |||
6327 | ASTContext::GetBuiltinTypeError Error; | |||
6328 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
6329 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6329, __PRETTY_FUNCTION__)); | |||
6330 | ||||
6331 | auto getAlignmentValue32 = [&](Address addr) -> Value* { | |||
6332 | return Builder.getInt32(addr.getAlignment().getQuantity()); | |||
6333 | }; | |||
6334 | ||||
6335 | Address PtrOp0 = Address::invalid(); | |||
6336 | Address PtrOp1 = Address::invalid(); | |||
6337 | SmallVector<Value*, 4> Ops; | |||
6338 | bool HasExtraArg = HasExtraNeonArgument(BuiltinID); | |||
6339 | unsigned NumArgs = E->getNumArgs() - (HasExtraArg ? 1 : 0); | |||
6340 | for (unsigned i = 0, e = NumArgs; i != e; i++) { | |||
6341 | if (i == 0) { | |||
6342 | switch (BuiltinID) { | |||
6343 | case NEON::BI__builtin_neon_vld1_v: | |||
6344 | case NEON::BI__builtin_neon_vld1q_v: | |||
6345 | case NEON::BI__builtin_neon_vld1q_lane_v: | |||
6346 | case NEON::BI__builtin_neon_vld1_lane_v: | |||
6347 | case NEON::BI__builtin_neon_vld1_dup_v: | |||
6348 | case NEON::BI__builtin_neon_vld1q_dup_v: | |||
6349 | case NEON::BI__builtin_neon_vst1_v: | |||
6350 | case NEON::BI__builtin_neon_vst1q_v: | |||
6351 | case NEON::BI__builtin_neon_vst1q_lane_v: | |||
6352 | case NEON::BI__builtin_neon_vst1_lane_v: | |||
6353 | case NEON::BI__builtin_neon_vst2_v: | |||
6354 | case NEON::BI__builtin_neon_vst2q_v: | |||
6355 | case NEON::BI__builtin_neon_vst2_lane_v: | |||
6356 | case NEON::BI__builtin_neon_vst2q_lane_v: | |||
6357 | case NEON::BI__builtin_neon_vst3_v: | |||
6358 | case NEON::BI__builtin_neon_vst3q_v: | |||
6359 | case NEON::BI__builtin_neon_vst3_lane_v: | |||
6360 | case NEON::BI__builtin_neon_vst3q_lane_v: | |||
6361 | case NEON::BI__builtin_neon_vst4_v: | |||
6362 | case NEON::BI__builtin_neon_vst4q_v: | |||
6363 | case NEON::BI__builtin_neon_vst4_lane_v: | |||
6364 | case NEON::BI__builtin_neon_vst4q_lane_v: | |||
6365 | // Get the alignment for the argument in addition to the value; | |||
6366 | // we'll use it later. | |||
6367 | PtrOp0 = EmitPointerWithAlignment(E->getArg(0)); | |||
6368 | Ops.push_back(PtrOp0.getPointer()); | |||
6369 | continue; | |||
6370 | } | |||
6371 | } | |||
6372 | if (i == 1) { | |||
6373 | switch (BuiltinID) { | |||
6374 | case NEON::BI__builtin_neon_vld2_v: | |||
6375 | case NEON::BI__builtin_neon_vld2q_v: | |||
6376 | case NEON::BI__builtin_neon_vld3_v: | |||
6377 | case NEON::BI__builtin_neon_vld3q_v: | |||
6378 | case NEON::BI__builtin_neon_vld4_v: | |||
6379 | case NEON::BI__builtin_neon_vld4q_v: | |||
6380 | case NEON::BI__builtin_neon_vld2_lane_v: | |||
6381 | case NEON::BI__builtin_neon_vld2q_lane_v: | |||
6382 | case NEON::BI__builtin_neon_vld3_lane_v: | |||
6383 | case NEON::BI__builtin_neon_vld3q_lane_v: | |||
6384 | case NEON::BI__builtin_neon_vld4_lane_v: | |||
6385 | case NEON::BI__builtin_neon_vld4q_lane_v: | |||
6386 | case NEON::BI__builtin_neon_vld2_dup_v: | |||
6387 | case NEON::BI__builtin_neon_vld2q_dup_v: | |||
6388 | case NEON::BI__builtin_neon_vld3_dup_v: | |||
6389 | case NEON::BI__builtin_neon_vld3q_dup_v: | |||
6390 | case NEON::BI__builtin_neon_vld4_dup_v: | |||
6391 | case NEON::BI__builtin_neon_vld4q_dup_v: | |||
6392 | // Get the alignment for the argument in addition to the value; | |||
6393 | // we'll use it later. | |||
6394 | PtrOp1 = EmitPointerWithAlignment(E->getArg(1)); | |||
6395 | Ops.push_back(PtrOp1.getPointer()); | |||
6396 | continue; | |||
6397 | } | |||
6398 | } | |||
6399 | ||||
6400 | if ((ICEArguments & (1 << i)) == 0) { | |||
6401 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
6402 | } else { | |||
6403 | // If this is required to be a constant, constant fold it so that we know | |||
6404 | // that the generated intrinsic gets a ConstantInt. | |||
6405 | llvm::APSInt Result; | |||
6406 | bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext()); | |||
6407 | assert(IsConst && "Constant arg isn't actually constant?")((IsConst && "Constant arg isn't actually constant?") ? static_cast<void> (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6407, __PRETTY_FUNCTION__)); (void)IsConst; | |||
6408 | Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result)); | |||
6409 | } | |||
6410 | } | |||
6411 | ||||
6412 | switch (BuiltinID) { | |||
6413 | default: break; | |||
6414 | ||||
6415 | case NEON::BI__builtin_neon_vget_lane_i8: | |||
6416 | case NEON::BI__builtin_neon_vget_lane_i16: | |||
6417 | case NEON::BI__builtin_neon_vget_lane_i32: | |||
6418 | case NEON::BI__builtin_neon_vget_lane_i64: | |||
6419 | case NEON::BI__builtin_neon_vget_lane_f32: | |||
6420 | case NEON::BI__builtin_neon_vgetq_lane_i8: | |||
6421 | case NEON::BI__builtin_neon_vgetq_lane_i16: | |||
6422 | case NEON::BI__builtin_neon_vgetq_lane_i32: | |||
6423 | case NEON::BI__builtin_neon_vgetq_lane_i64: | |||
6424 | case NEON::BI__builtin_neon_vgetq_lane_f32: | |||
6425 | return Builder.CreateExtractElement(Ops[0], Ops[1], "vget_lane"); | |||
6426 | ||||
6427 | case NEON::BI__builtin_neon_vrndns_f32: { | |||
6428 | Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
6429 | llvm::Type *Tys[] = {Arg->getType()}; | |||
6430 | Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vrintn, Tys); | |||
6431 | return Builder.CreateCall(F, {Arg}, "vrndn"); } | |||
6432 | ||||
6433 | case NEON::BI__builtin_neon_vset_lane_i8: | |||
6434 | case NEON::BI__builtin_neon_vset_lane_i16: | |||
6435 | case NEON::BI__builtin_neon_vset_lane_i32: | |||
6436 | case NEON::BI__builtin_neon_vset_lane_i64: | |||
6437 | case NEON::BI__builtin_neon_vset_lane_f32: | |||
6438 | case NEON::BI__builtin_neon_vsetq_lane_i8: | |||
6439 | case NEON::BI__builtin_neon_vsetq_lane_i16: | |||
6440 | case NEON::BI__builtin_neon_vsetq_lane_i32: | |||
6441 | case NEON::BI__builtin_neon_vsetq_lane_i64: | |||
6442 | case NEON::BI__builtin_neon_vsetq_lane_f32: | |||
6443 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); | |||
6444 | ||||
6445 | case NEON::BI__builtin_neon_vsha1h_u32: | |||
6446 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1h), Ops, | |||
6447 | "vsha1h"); | |||
6448 | case NEON::BI__builtin_neon_vsha1cq_u32: | |||
6449 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1c), Ops, | |||
6450 | "vsha1h"); | |||
6451 | case NEON::BI__builtin_neon_vsha1pq_u32: | |||
6452 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1p), Ops, | |||
6453 | "vsha1h"); | |||
6454 | case NEON::BI__builtin_neon_vsha1mq_u32: | |||
6455 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1m), Ops, | |||
6456 | "vsha1h"); | |||
6457 | ||||
6458 | // The ARM _MoveToCoprocessor builtins put the input register value as | |||
6459 | // the first argument, but the LLVM intrinsic expects it as the third one. | |||
6460 | case ARM::BI_MoveToCoprocessor: | |||
6461 | case ARM::BI_MoveToCoprocessor2: { | |||
6462 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI_MoveToCoprocessor ? | |||
6463 | Intrinsic::arm_mcr : Intrinsic::arm_mcr2); | |||
6464 | return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0], | |||
6465 | Ops[3], Ops[4], Ops[5]}); | |||
6466 | } | |||
6467 | case ARM::BI_BitScanForward: | |||
6468 | case ARM::BI_BitScanForward64: | |||
6469 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E); | |||
6470 | case ARM::BI_BitScanReverse: | |||
6471 | case ARM::BI_BitScanReverse64: | |||
6472 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E); | |||
6473 | ||||
6474 | case ARM::BI_InterlockedAnd64: | |||
6475 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E); | |||
6476 | case ARM::BI_InterlockedExchange64: | |||
6477 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E); | |||
6478 | case ARM::BI_InterlockedExchangeAdd64: | |||
6479 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E); | |||
6480 | case ARM::BI_InterlockedExchangeSub64: | |||
6481 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E); | |||
6482 | case ARM::BI_InterlockedOr64: | |||
6483 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E); | |||
6484 | case ARM::BI_InterlockedXor64: | |||
6485 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E); | |||
6486 | case ARM::BI_InterlockedDecrement64: | |||
6487 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E); | |||
6488 | case ARM::BI_InterlockedIncrement64: | |||
6489 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E); | |||
6490 | case ARM::BI_InterlockedExchangeAdd8_acq: | |||
6491 | case ARM::BI_InterlockedExchangeAdd16_acq: | |||
6492 | case ARM::BI_InterlockedExchangeAdd_acq: | |||
6493 | case ARM::BI_InterlockedExchangeAdd64_acq: | |||
6494 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_acq, E); | |||
6495 | case ARM::BI_InterlockedExchangeAdd8_rel: | |||
6496 | case ARM::BI_InterlockedExchangeAdd16_rel: | |||
6497 | case ARM::BI_InterlockedExchangeAdd_rel: | |||
6498 | case ARM::BI_InterlockedExchangeAdd64_rel: | |||
6499 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_rel, E); | |||
6500 | case ARM::BI_InterlockedExchangeAdd8_nf: | |||
6501 | case ARM::BI_InterlockedExchangeAdd16_nf: | |||
6502 | case ARM::BI_InterlockedExchangeAdd_nf: | |||
6503 | case ARM::BI_InterlockedExchangeAdd64_nf: | |||
6504 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_nf, E); | |||
6505 | case ARM::BI_InterlockedExchange8_acq: | |||
6506 | case ARM::BI_InterlockedExchange16_acq: | |||
6507 | case ARM::BI_InterlockedExchange_acq: | |||
6508 | case ARM::BI_InterlockedExchange64_acq: | |||
6509 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_acq, E); | |||
6510 | case ARM::BI_InterlockedExchange8_rel: | |||
6511 | case ARM::BI_InterlockedExchange16_rel: | |||
6512 | case ARM::BI_InterlockedExchange_rel: | |||
6513 | case ARM::BI_InterlockedExchange64_rel: | |||
6514 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_rel, E); | |||
6515 | case ARM::BI_InterlockedExchange8_nf: | |||
6516 | case ARM::BI_InterlockedExchange16_nf: | |||
6517 | case ARM::BI_InterlockedExchange_nf: | |||
6518 | case ARM::BI_InterlockedExchange64_nf: | |||
6519 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_nf, E); | |||
6520 | case ARM::BI_InterlockedCompareExchange8_acq: | |||
6521 | case ARM::BI_InterlockedCompareExchange16_acq: | |||
6522 | case ARM::BI_InterlockedCompareExchange_acq: | |||
6523 | case ARM::BI_InterlockedCompareExchange64_acq: | |||
6524 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_acq, E); | |||
6525 | case ARM::BI_InterlockedCompareExchange8_rel: | |||
6526 | case ARM::BI_InterlockedCompareExchange16_rel: | |||
6527 | case ARM::BI_InterlockedCompareExchange_rel: | |||
6528 | case ARM::BI_InterlockedCompareExchange64_rel: | |||
6529 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_rel, E); | |||
6530 | case ARM::BI_InterlockedCompareExchange8_nf: | |||
6531 | case ARM::BI_InterlockedCompareExchange16_nf: | |||
6532 | case ARM::BI_InterlockedCompareExchange_nf: | |||
6533 | case ARM::BI_InterlockedCompareExchange64_nf: | |||
6534 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_nf, E); | |||
6535 | case ARM::BI_InterlockedOr8_acq: | |||
6536 | case ARM::BI_InterlockedOr16_acq: | |||
6537 | case ARM::BI_InterlockedOr_acq: | |||
6538 | case ARM::BI_InterlockedOr64_acq: | |||
6539 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_acq, E); | |||
6540 | case ARM::BI_InterlockedOr8_rel: | |||
6541 | case ARM::BI_InterlockedOr16_rel: | |||
6542 | case ARM::BI_InterlockedOr_rel: | |||
6543 | case ARM::BI_InterlockedOr64_rel: | |||
6544 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_rel, E); | |||
6545 | case ARM::BI_InterlockedOr8_nf: | |||
6546 | case ARM::BI_InterlockedOr16_nf: | |||
6547 | case ARM::BI_InterlockedOr_nf: | |||
6548 | case ARM::BI_InterlockedOr64_nf: | |||
6549 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_nf, E); | |||
6550 | case ARM::BI_InterlockedXor8_acq: | |||
6551 | case ARM::BI_InterlockedXor16_acq: | |||
6552 | case ARM::BI_InterlockedXor_acq: | |||
6553 | case ARM::BI_InterlockedXor64_acq: | |||
6554 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_acq, E); | |||
6555 | case ARM::BI_InterlockedXor8_rel: | |||
6556 | case ARM::BI_InterlockedXor16_rel: | |||
6557 | case ARM::BI_InterlockedXor_rel: | |||
6558 | case ARM::BI_InterlockedXor64_rel: | |||
6559 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_rel, E); | |||
6560 | case ARM::BI_InterlockedXor8_nf: | |||
6561 | case ARM::BI_InterlockedXor16_nf: | |||
6562 | case ARM::BI_InterlockedXor_nf: | |||
6563 | case ARM::BI_InterlockedXor64_nf: | |||
6564 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_nf, E); | |||
6565 | case ARM::BI_InterlockedAnd8_acq: | |||
6566 | case ARM::BI_InterlockedAnd16_acq: | |||
6567 | case ARM::BI_InterlockedAnd_acq: | |||
6568 | case ARM::BI_InterlockedAnd64_acq: | |||
6569 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_acq, E); | |||
6570 | case ARM::BI_InterlockedAnd8_rel: | |||
6571 | case ARM::BI_InterlockedAnd16_rel: | |||
6572 | case ARM::BI_InterlockedAnd_rel: | |||
6573 | case ARM::BI_InterlockedAnd64_rel: | |||
6574 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_rel, E); | |||
6575 | case ARM::BI_InterlockedAnd8_nf: | |||
6576 | case ARM::BI_InterlockedAnd16_nf: | |||
6577 | case ARM::BI_InterlockedAnd_nf: | |||
6578 | case ARM::BI_InterlockedAnd64_nf: | |||
6579 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_nf, E); | |||
6580 | case ARM::BI_InterlockedIncrement16_acq: | |||
6581 | case ARM::BI_InterlockedIncrement_acq: | |||
6582 | case ARM::BI_InterlockedIncrement64_acq: | |||
6583 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_acq, E); | |||
6584 | case ARM::BI_InterlockedIncrement16_rel: | |||
6585 | case ARM::BI_InterlockedIncrement_rel: | |||
6586 | case ARM::BI_InterlockedIncrement64_rel: | |||
6587 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_rel, E); | |||
6588 | case ARM::BI_InterlockedIncrement16_nf: | |||
6589 | case ARM::BI_InterlockedIncrement_nf: | |||
6590 | case ARM::BI_InterlockedIncrement64_nf: | |||
6591 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_nf, E); | |||
6592 | case ARM::BI_InterlockedDecrement16_acq: | |||
6593 | case ARM::BI_InterlockedDecrement_acq: | |||
6594 | case ARM::BI_InterlockedDecrement64_acq: | |||
6595 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_acq, E); | |||
6596 | case ARM::BI_InterlockedDecrement16_rel: | |||
6597 | case ARM::BI_InterlockedDecrement_rel: | |||
6598 | case ARM::BI_InterlockedDecrement64_rel: | |||
6599 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_rel, E); | |||
6600 | case ARM::BI_InterlockedDecrement16_nf: | |||
6601 | case ARM::BI_InterlockedDecrement_nf: | |||
6602 | case ARM::BI_InterlockedDecrement64_nf: | |||
6603 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_nf, E); | |||
6604 | } | |||
6605 | ||||
6606 | // Get the last argument, which specifies the vector type. | |||
6607 | assert(HasExtraArg)((HasExtraArg) ? static_cast<void> (0) : __assert_fail ( "HasExtraArg", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6607, __PRETTY_FUNCTION__)); | |||
6608 | llvm::APSInt Result; | |||
6609 | const Expr *Arg = E->getArg(E->getNumArgs()-1); | |||
6610 | if (!Arg->isIntegerConstantExpr(Result, getContext())) | |||
6611 | return nullptr; | |||
6612 | ||||
6613 | if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f || | |||
6614 | BuiltinID == ARM::BI__builtin_arm_vcvtr_d) { | |||
6615 | // Determine the overloaded type of this builtin. | |||
6616 | llvm::Type *Ty; | |||
6617 | if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f) | |||
6618 | Ty = FloatTy; | |||
6619 | else | |||
6620 | Ty = DoubleTy; | |||
6621 | ||||
6622 | // Determine whether this is an unsigned conversion or not. | |||
6623 | bool usgn = Result.getZExtValue() == 1; | |||
6624 | unsigned Int = usgn ? Intrinsic::arm_vcvtru : Intrinsic::arm_vcvtr; | |||
6625 | ||||
6626 | // Call the appropriate intrinsic. | |||
6627 | Function *F = CGM.getIntrinsic(Int, Ty); | |||
6628 | return Builder.CreateCall(F, Ops, "vcvtr"); | |||
6629 | } | |||
6630 | ||||
6631 | // Determine the type of this overloaded NEON intrinsic. | |||
6632 | NeonTypeFlags Type(Result.getZExtValue()); | |||
6633 | bool usgn = Type.isUnsigned(); | |||
6634 | bool rightShift = false; | |||
6635 | ||||
6636 | llvm::VectorType *VTy = GetNeonType(this, Type, | |||
6637 | getTarget().hasLegalHalfType()); | |||
6638 | llvm::Type *Ty = VTy; | |||
6639 | if (!Ty) | |||
6640 | return nullptr; | |||
6641 | ||||
6642 | // Many NEON builtins have identical semantics and uses in ARM and | |||
6643 | // AArch64. Emit these in a single function. | |||
6644 | auto IntrinsicMap = makeArrayRef(ARMSIMDIntrinsicMap); | |||
6645 | const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap( | |||
6646 | IntrinsicMap, BuiltinID, NEONSIMDIntrinsicsProvenSorted); | |||
6647 | if (Builtin) | |||
6648 | return EmitCommonNeonBuiltinExpr( | |||
6649 | Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic, | |||
6650 | Builtin->NameHint, Builtin->TypeModifier, E, Ops, PtrOp0, PtrOp1, Arch); | |||
6651 | ||||
6652 | unsigned Int; | |||
6653 | switch (BuiltinID) { | |||
6654 | default: return nullptr; | |||
6655 | case NEON::BI__builtin_neon_vld1q_lane_v: | |||
6656 | // Handle 64-bit integer elements as a special case. Use shuffles of | |||
6657 | // one-element vectors to avoid poor code for i64 in the backend. | |||
6658 | if (VTy->getElementType()->isIntegerTy(64)) { | |||
6659 | // Extract the other lane. | |||
6660 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
6661 | uint32_t Lane = cast<ConstantInt>(Ops[2])->getZExtValue(); | |||
6662 | Value *SV = llvm::ConstantVector::get(ConstantInt::get(Int32Ty, 1-Lane)); | |||
6663 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV); | |||
6664 | // Load the value as a one-element vector. | |||
6665 | Ty = llvm::VectorType::get(VTy->getElementType(), 1); | |||
6666 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; | |||
6667 | Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vld1, Tys); | |||
6668 | Value *Align = getAlignmentValue32(PtrOp0); | |||
6669 | Value *Ld = Builder.CreateCall(F, {Ops[0], Align}); | |||
6670 | // Combine them. | |||
6671 | uint32_t Indices[] = {1 - Lane, Lane}; | |||
6672 | SV = llvm::ConstantDataVector::get(getLLVMContext(), Indices); | |||
6673 | return Builder.CreateShuffleVector(Ops[1], Ld, SV, "vld1q_lane"); | |||
6674 | } | |||
6675 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
6676 | case NEON::BI__builtin_neon_vld1_lane_v: { | |||
6677 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
6678 | PtrOp0 = Builder.CreateElementBitCast(PtrOp0, VTy->getElementType()); | |||
6679 | Value *Ld = Builder.CreateLoad(PtrOp0); | |||
6680 | return Builder.CreateInsertElement(Ops[1], Ld, Ops[2], "vld1_lane"); | |||
6681 | } | |||
6682 | case NEON::BI__builtin_neon_vqrshrn_n_v: | |||
6683 | Int = | |||
6684 | usgn ? Intrinsic::arm_neon_vqrshiftnu : Intrinsic::arm_neon_vqrshiftns; | |||
6685 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n", | |||
6686 | 1, true); | |||
6687 | case NEON::BI__builtin_neon_vqrshrun_n_v: | |||
6688 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqrshiftnsu, Ty), | |||
6689 | Ops, "vqrshrun_n", 1, true); | |||
6690 | case NEON::BI__builtin_neon_vqshrn_n_v: | |||
6691 | Int = usgn ? Intrinsic::arm_neon_vqshiftnu : Intrinsic::arm_neon_vqshiftns; | |||
6692 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n", | |||
6693 | 1, true); | |||
6694 | case NEON::BI__builtin_neon_vqshrun_n_v: | |||
6695 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqshiftnsu, Ty), | |||
6696 | Ops, "vqshrun_n", 1, true); | |||
6697 | case NEON::BI__builtin_neon_vrecpe_v: | |||
6698 | case NEON::BI__builtin_neon_vrecpeq_v: | |||
6699 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrecpe, Ty), | |||
6700 | Ops, "vrecpe"); | |||
6701 | case NEON::BI__builtin_neon_vrshrn_n_v: | |||
6702 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrshiftn, Ty), | |||
6703 | Ops, "vrshrn_n", 1, true); | |||
6704 | case NEON::BI__builtin_neon_vrsra_n_v: | |||
6705 | case NEON::BI__builtin_neon_vrsraq_n_v: | |||
6706 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
6707 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
6708 | Ops[2] = EmitNeonShiftVector(Ops[2], Ty, true); | |||
6709 | Int = usgn ? Intrinsic::arm_neon_vrshiftu : Intrinsic::arm_neon_vrshifts; | |||
6710 | Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Ty), {Ops[1], Ops[2]}); | |||
6711 | return Builder.CreateAdd(Ops[0], Ops[1], "vrsra_n"); | |||
6712 | case NEON::BI__builtin_neon_vsri_n_v: | |||
6713 | case NEON::BI__builtin_neon_vsriq_n_v: | |||
6714 | rightShift = true; | |||
6715 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
6716 | case NEON::BI__builtin_neon_vsli_n_v: | |||
6717 | case NEON::BI__builtin_neon_vsliq_n_v: | |||
6718 | Ops[2] = EmitNeonShiftVector(Ops[2], Ty, rightShift); | |||
6719 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vshiftins, Ty), | |||
6720 | Ops, "vsli_n"); | |||
6721 | case NEON::BI__builtin_neon_vsra_n_v: | |||
6722 | case NEON::BI__builtin_neon_vsraq_n_v: | |||
6723 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
6724 | Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n"); | |||
6725 | return Builder.CreateAdd(Ops[0], Ops[1]); | |||
6726 | case NEON::BI__builtin_neon_vst1q_lane_v: | |||
6727 | // Handle 64-bit integer elements as a special case. Use a shuffle to get | |||
6728 | // a one-element vector and avoid poor code for i64 in the backend. | |||
6729 | if (VTy->getElementType()->isIntegerTy(64)) { | |||
6730 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
6731 | Value *SV = llvm::ConstantVector::get(cast<llvm::Constant>(Ops[2])); | |||
6732 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV); | |||
6733 | Ops[2] = getAlignmentValue32(PtrOp0); | |||
6734 | llvm::Type *Tys[] = {Int8PtrTy, Ops[1]->getType()}; | |||
6735 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_neon_vst1, | |||
6736 | Tys), Ops); | |||
6737 | } | |||
6738 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
6739 | case NEON::BI__builtin_neon_vst1_lane_v: { | |||
6740 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
6741 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]); | |||
6742 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
6743 | auto St = Builder.CreateStore(Ops[1], Builder.CreateBitCast(PtrOp0, Ty)); | |||
6744 | return St; | |||
6745 | } | |||
6746 | case NEON::BI__builtin_neon_vtbl1_v: | |||
6747 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl1), | |||
6748 | Ops, "vtbl1"); | |||
6749 | case NEON::BI__builtin_neon_vtbl2_v: | |||
6750 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl2), | |||
6751 | Ops, "vtbl2"); | |||
6752 | case NEON::BI__builtin_neon_vtbl3_v: | |||
6753 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl3), | |||
6754 | Ops, "vtbl3"); | |||
6755 | case NEON::BI__builtin_neon_vtbl4_v: | |||
6756 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl4), | |||
6757 | Ops, "vtbl4"); | |||
6758 | case NEON::BI__builtin_neon_vtbx1_v: | |||
6759 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx1), | |||
6760 | Ops, "vtbx1"); | |||
6761 | case NEON::BI__builtin_neon_vtbx2_v: | |||
6762 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx2), | |||
6763 | Ops, "vtbx2"); | |||
6764 | case NEON::BI__builtin_neon_vtbx3_v: | |||
6765 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx3), | |||
6766 | Ops, "vtbx3"); | |||
6767 | case NEON::BI__builtin_neon_vtbx4_v: | |||
6768 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx4), | |||
6769 | Ops, "vtbx4"); | |||
6770 | } | |||
6771 | } | |||
6772 | ||||
6773 | static Value *EmitAArch64TblBuiltinExpr(CodeGenFunction &CGF, unsigned BuiltinID, | |||
6774 | const CallExpr *E, | |||
6775 | SmallVectorImpl<Value *> &Ops, | |||
6776 | llvm::Triple::ArchType Arch) { | |||
6777 | unsigned int Int = 0; | |||
6778 | const char *s = nullptr; | |||
6779 | ||||
6780 | switch (BuiltinID) { | |||
6781 | default: | |||
6782 | return nullptr; | |||
6783 | case NEON::BI__builtin_neon_vtbl1_v: | |||
6784 | case NEON::BI__builtin_neon_vqtbl1_v: | |||
6785 | case NEON::BI__builtin_neon_vqtbl1q_v: | |||
6786 | case NEON::BI__builtin_neon_vtbl2_v: | |||
6787 | case NEON::BI__builtin_neon_vqtbl2_v: | |||
6788 | case NEON::BI__builtin_neon_vqtbl2q_v: | |||
6789 | case NEON::BI__builtin_neon_vtbl3_v: | |||
6790 | case NEON::BI__builtin_neon_vqtbl3_v: | |||
6791 | case NEON::BI__builtin_neon_vqtbl3q_v: | |||
6792 | case NEON::BI__builtin_neon_vtbl4_v: | |||
6793 | case NEON::BI__builtin_neon_vqtbl4_v: | |||
6794 | case NEON::BI__builtin_neon_vqtbl4q_v: | |||
6795 | break; | |||
6796 | case NEON::BI__builtin_neon_vtbx1_v: | |||
6797 | case NEON::BI__builtin_neon_vqtbx1_v: | |||
6798 | case NEON::BI__builtin_neon_vqtbx1q_v: | |||
6799 | case NEON::BI__builtin_neon_vtbx2_v: | |||
6800 | case NEON::BI__builtin_neon_vqtbx2_v: | |||
6801 | case NEON::BI__builtin_neon_vqtbx2q_v: | |||
6802 | case NEON::BI__builtin_neon_vtbx3_v: | |||
6803 | case NEON::BI__builtin_neon_vqtbx3_v: | |||
6804 | case NEON::BI__builtin_neon_vqtbx3q_v: | |||
6805 | case NEON::BI__builtin_neon_vtbx4_v: | |||
6806 | case NEON::BI__builtin_neon_vqtbx4_v: | |||
6807 | case NEON::BI__builtin_neon_vqtbx4q_v: | |||
6808 | break; | |||
6809 | } | |||
6810 | ||||
6811 | assert(E->getNumArgs() >= 3)((E->getNumArgs() >= 3) ? static_cast<void> (0) : __assert_fail ("E->getNumArgs() >= 3", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6811, __PRETTY_FUNCTION__)); | |||
6812 | ||||
6813 | // Get the last argument, which specifies the vector type. | |||
6814 | llvm::APSInt Result; | |||
6815 | const Expr *Arg = E->getArg(E->getNumArgs() - 1); | |||
6816 | if (!Arg->isIntegerConstantExpr(Result, CGF.getContext())) | |||
6817 | return nullptr; | |||
6818 | ||||
6819 | // Determine the type of this overloaded NEON intrinsic. | |||
6820 | NeonTypeFlags Type(Result.getZExtValue()); | |||
6821 | llvm::VectorType *Ty = GetNeonType(&CGF, Type); | |||
6822 | if (!Ty) | |||
6823 | return nullptr; | |||
6824 | ||||
6825 | CodeGen::CGBuilderTy &Builder = CGF.Builder; | |||
6826 | ||||
6827 | // AArch64 scalar builtins are not overloaded, they do not have an extra | |||
6828 | // argument that specifies the vector type, need to handle each case. | |||
6829 | switch (BuiltinID) { | |||
6830 | case NEON::BI__builtin_neon_vtbl1_v: { | |||
6831 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 1), nullptr, | |||
6832 | Ops[1], Ty, Intrinsic::aarch64_neon_tbl1, | |||
6833 | "vtbl1"); | |||
6834 | } | |||
6835 | case NEON::BI__builtin_neon_vtbl2_v: { | |||
6836 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 2), nullptr, | |||
6837 | Ops[2], Ty, Intrinsic::aarch64_neon_tbl1, | |||
6838 | "vtbl1"); | |||
6839 | } | |||
6840 | case NEON::BI__builtin_neon_vtbl3_v: { | |||
6841 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 3), nullptr, | |||
6842 | Ops[3], Ty, Intrinsic::aarch64_neon_tbl2, | |||
6843 | "vtbl2"); | |||
6844 | } | |||
6845 | case NEON::BI__builtin_neon_vtbl4_v: { | |||
6846 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 4), nullptr, | |||
6847 | Ops[4], Ty, Intrinsic::aarch64_neon_tbl2, | |||
6848 | "vtbl2"); | |||
6849 | } | |||
6850 | case NEON::BI__builtin_neon_vtbx1_v: { | |||
6851 | Value *TblRes = | |||
6852 | packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 1), nullptr, Ops[2], | |||
6853 | Ty, Intrinsic::aarch64_neon_tbl1, "vtbl1"); | |||
6854 | ||||
6855 | llvm::Constant *EightV = ConstantInt::get(Ty, 8); | |||
6856 | Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[2], EightV); | |||
6857 | CmpRes = Builder.CreateSExt(CmpRes, Ty); | |||
6858 | ||||
6859 | Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]); | |||
6860 | Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes); | |||
6861 | return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx"); | |||
6862 | } | |||
6863 | case NEON::BI__builtin_neon_vtbx2_v: { | |||
6864 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 2), Ops[0], | |||
6865 | Ops[3], Ty, Intrinsic::aarch64_neon_tbx1, | |||
6866 | "vtbx1"); | |||
6867 | } | |||
6868 | case NEON::BI__builtin_neon_vtbx3_v: { | |||
6869 | Value *TblRes = | |||
6870 | packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 3), nullptr, Ops[4], | |||
6871 | Ty, Intrinsic::aarch64_neon_tbl2, "vtbl2"); | |||
6872 | ||||
6873 | llvm::Constant *TwentyFourV = ConstantInt::get(Ty, 24); | |||
6874 | Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[4], | |||
6875 | TwentyFourV); | |||
6876 | CmpRes = Builder.CreateSExt(CmpRes, Ty); | |||
6877 | ||||
6878 | Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]); | |||
6879 | Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes); | |||
6880 | return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx"); | |||
6881 | } | |||
6882 | case NEON::BI__builtin_neon_vtbx4_v: { | |||
6883 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 4), Ops[0], | |||
6884 | Ops[5], Ty, Intrinsic::aarch64_neon_tbx2, | |||
6885 | "vtbx2"); | |||
6886 | } | |||
6887 | case NEON::BI__builtin_neon_vqtbl1_v: | |||
6888 | case NEON::BI__builtin_neon_vqtbl1q_v: | |||
6889 | Int = Intrinsic::aarch64_neon_tbl1; s = "vtbl1"; break; | |||
6890 | case NEON::BI__builtin_neon_vqtbl2_v: | |||
6891 | case NEON::BI__builtin_neon_vqtbl2q_v: { | |||
6892 | Int = Intrinsic::aarch64_neon_tbl2; s = "vtbl2"; break; | |||
6893 | case NEON::BI__builtin_neon_vqtbl3_v: | |||
6894 | case NEON::BI__builtin_neon_vqtbl3q_v: | |||
6895 | Int = Intrinsic::aarch64_neon_tbl3; s = "vtbl3"; break; | |||
6896 | case NEON::BI__builtin_neon_vqtbl4_v: | |||
6897 | case NEON::BI__builtin_neon_vqtbl4q_v: | |||
6898 | Int = Intrinsic::aarch64_neon_tbl4; s = "vtbl4"; break; | |||
6899 | case NEON::BI__builtin_neon_vqtbx1_v: | |||
6900 | case NEON::BI__builtin_neon_vqtbx1q_v: | |||
6901 | Int = Intrinsic::aarch64_neon_tbx1; s = "vtbx1"; break; | |||
6902 | case NEON::BI__builtin_neon_vqtbx2_v: | |||
6903 | case NEON::BI__builtin_neon_vqtbx2q_v: | |||
6904 | Int = Intrinsic::aarch64_neon_tbx2; s = "vtbx2"; break; | |||
6905 | case NEON::BI__builtin_neon_vqtbx3_v: | |||
6906 | case NEON::BI__builtin_neon_vqtbx3q_v: | |||
6907 | Int = Intrinsic::aarch64_neon_tbx3; s = "vtbx3"; break; | |||
6908 | case NEON::BI__builtin_neon_vqtbx4_v: | |||
6909 | case NEON::BI__builtin_neon_vqtbx4q_v: | |||
6910 | Int = Intrinsic::aarch64_neon_tbx4; s = "vtbx4"; break; | |||
6911 | } | |||
6912 | } | |||
6913 | ||||
6914 | if (!Int) | |||
6915 | return nullptr; | |||
6916 | ||||
6917 | Function *F = CGF.CGM.getIntrinsic(Int, Ty); | |||
6918 | return CGF.EmitNeonCall(F, Ops, s); | |||
6919 | } | |||
6920 | ||||
6921 | Value *CodeGenFunction::vectorWrapScalar16(Value *Op) { | |||
6922 | llvm::Type *VTy = llvm::VectorType::get(Int16Ty, 4); | |||
6923 | Op = Builder.CreateBitCast(Op, Int16Ty); | |||
6924 | Value *V = UndefValue::get(VTy); | |||
6925 | llvm::Constant *CI = ConstantInt::get(SizeTy, 0); | |||
6926 | Op = Builder.CreateInsertElement(V, Op, CI); | |||
6927 | return Op; | |||
6928 | } | |||
6929 | ||||
6930 | Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID, | |||
6931 | const CallExpr *E, | |||
6932 | llvm::Triple::ArchType Arch) { | |||
6933 | unsigned HintID = static_cast<unsigned>(-1); | |||
6934 | switch (BuiltinID) { | |||
6935 | default: break; | |||
6936 | case AArch64::BI__builtin_arm_nop: | |||
6937 | HintID = 0; | |||
6938 | break; | |||
6939 | case AArch64::BI__builtin_arm_yield: | |||
6940 | case AArch64::BI__yield: | |||
6941 | HintID = 1; | |||
6942 | break; | |||
6943 | case AArch64::BI__builtin_arm_wfe: | |||
6944 | case AArch64::BI__wfe: | |||
6945 | HintID = 2; | |||
6946 | break; | |||
6947 | case AArch64::BI__builtin_arm_wfi: | |||
6948 | case AArch64::BI__wfi: | |||
6949 | HintID = 3; | |||
6950 | break; | |||
6951 | case AArch64::BI__builtin_arm_sev: | |||
6952 | case AArch64::BI__sev: | |||
6953 | HintID = 4; | |||
6954 | break; | |||
6955 | case AArch64::BI__builtin_arm_sevl: | |||
6956 | case AArch64::BI__sevl: | |||
6957 | HintID = 5; | |||
6958 | break; | |||
6959 | } | |||
6960 | ||||
6961 | if (HintID != static_cast<unsigned>(-1)) { | |||
6962 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_hint); | |||
6963 | return Builder.CreateCall(F, llvm::ConstantInt::get(Int32Ty, HintID)); | |||
6964 | } | |||
6965 | ||||
6966 | if (BuiltinID == AArch64::BI__builtin_arm_prefetch) { | |||
6967 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
6968 | Value *RW = EmitScalarExpr(E->getArg(1)); | |||
6969 | Value *CacheLevel = EmitScalarExpr(E->getArg(2)); | |||
6970 | Value *RetentionPolicy = EmitScalarExpr(E->getArg(3)); | |||
6971 | Value *IsData = EmitScalarExpr(E->getArg(4)); | |||
6972 | ||||
6973 | Value *Locality = nullptr; | |||
6974 | if (cast<llvm::ConstantInt>(RetentionPolicy)->isZero()) { | |||
6975 | // Temporal fetch, needs to convert cache level to locality. | |||
6976 | Locality = llvm::ConstantInt::get(Int32Ty, | |||
6977 | -cast<llvm::ConstantInt>(CacheLevel)->getValue() + 3); | |||
6978 | } else { | |||
6979 | // Streaming fetch. | |||
6980 | Locality = llvm::ConstantInt::get(Int32Ty, 0); | |||
6981 | } | |||
6982 | ||||
6983 | // FIXME: We need AArch64 specific LLVM intrinsic if we want to specify | |||
6984 | // PLDL3STRM or PLDL2STRM. | |||
6985 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); | |||
6986 | return Builder.CreateCall(F, {Address, RW, Locality, IsData}); | |||
6987 | } | |||
6988 | ||||
6989 | if (BuiltinID == AArch64::BI__builtin_arm_rbit) { | |||
6990 | assert((getContext().getTypeSize(E->getType()) == 32) &&(((getContext().getTypeSize(E->getType()) == 32) && "rbit of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6991, __PRETTY_FUNCTION__)) | |||
6991 | "rbit of unusual size!")(((getContext().getTypeSize(E->getType()) == 32) && "rbit of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6991, __PRETTY_FUNCTION__)); | |||
6992 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
6993 | return Builder.CreateCall( | |||
6994 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); | |||
6995 | } | |||
6996 | if (BuiltinID == AArch64::BI__builtin_arm_rbit64) { | |||
6997 | assert((getContext().getTypeSize(E->getType()) == 64) &&(((getContext().getTypeSize(E->getType()) == 64) && "rbit of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6998, __PRETTY_FUNCTION__)) | |||
6998 | "rbit of unusual size!")(((getContext().getTypeSize(E->getType()) == 64) && "rbit of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6998, __PRETTY_FUNCTION__)); | |||
6999 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
7000 | return Builder.CreateCall( | |||
7001 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); | |||
7002 | } | |||
7003 | ||||
7004 | if (BuiltinID == AArch64::BI__builtin_arm_jcvt) { | |||
7005 | assert((getContext().getTypeSize(E->getType()) == 32) &&(((getContext().getTypeSize(E->getType()) == 32) && "__jcvt of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"__jcvt of unusual size!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7006, __PRETTY_FUNCTION__)) | |||
7006 | "__jcvt of unusual size!")(((getContext().getTypeSize(E->getType()) == 32) && "__jcvt of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"__jcvt of unusual size!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7006, __PRETTY_FUNCTION__)); | |||
7007 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
7008 | return Builder.CreateCall( | |||
7009 | CGM.getIntrinsic(Intrinsic::aarch64_fjcvtzs), Arg); | |||
7010 | } | |||
7011 | ||||
7012 | if (BuiltinID == AArch64::BI__clear_cache) { | |||
7013 | assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments")((E->getNumArgs() == 2 && "__clear_cache takes 2 arguments" ) ? static_cast<void> (0) : __assert_fail ("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7013, __PRETTY_FUNCTION__)); | |||
7014 | const FunctionDecl *FD = E->getDirectCallee(); | |||
7015 | Value *Ops[2]; | |||
7016 | for (unsigned i = 0; i < 2; i++) | |||
7017 | Ops[i] = EmitScalarExpr(E->getArg(i)); | |||
7018 | llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType()); | |||
7019 | llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty); | |||
7020 | StringRef Name = FD->getName(); | |||
7021 | return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops); | |||
7022 | } | |||
7023 | ||||
7024 | if ((BuiltinID == AArch64::BI__builtin_arm_ldrex || | |||
7025 | BuiltinID == AArch64::BI__builtin_arm_ldaex) && | |||
7026 | getContext().getTypeSize(E->getType()) == 128) { | |||
7027 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex | |||
7028 | ? Intrinsic::aarch64_ldaxp | |||
7029 | : Intrinsic::aarch64_ldxp); | |||
7030 | ||||
7031 | Value *LdPtr = EmitScalarExpr(E->getArg(0)); | |||
7032 | Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy), | |||
7033 | "ldxp"); | |||
7034 | ||||
7035 | Value *Val0 = Builder.CreateExtractValue(Val, 1); | |||
7036 | Value *Val1 = Builder.CreateExtractValue(Val, 0); | |||
7037 | llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128); | |||
7038 | Val0 = Builder.CreateZExt(Val0, Int128Ty); | |||
7039 | Val1 = Builder.CreateZExt(Val1, Int128Ty); | |||
7040 | ||||
7041 | Value *ShiftCst = llvm::ConstantInt::get(Int128Ty, 64); | |||
7042 | Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */); | |||
7043 | Val = Builder.CreateOr(Val, Val1); | |||
7044 | return Builder.CreateBitCast(Val, ConvertType(E->getType())); | |||
7045 | } else if (BuiltinID == AArch64::BI__builtin_arm_ldrex || | |||
7046 | BuiltinID == AArch64::BI__builtin_arm_ldaex) { | |||
7047 | Value *LoadAddr = EmitScalarExpr(E->getArg(0)); | |||
7048 | ||||
7049 | QualType Ty = E->getType(); | |||
7050 | llvm::Type *RealResTy = ConvertType(Ty); | |||
7051 | llvm::Type *PtrTy = llvm::IntegerType::get( | |||
7052 | getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo(); | |||
7053 | LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy); | |||
7054 | ||||
7055 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex | |||
7056 | ? Intrinsic::aarch64_ldaxr | |||
7057 | : Intrinsic::aarch64_ldxr, | |||
7058 | PtrTy); | |||
7059 | Value *Val = Builder.CreateCall(F, LoadAddr, "ldxr"); | |||
7060 | ||||
7061 | if (RealResTy->isPointerTy()) | |||
7062 | return Builder.CreateIntToPtr(Val, RealResTy); | |||
7063 | ||||
7064 | llvm::Type *IntResTy = llvm::IntegerType::get( | |||
7065 | getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy)); | |||
7066 | Val = Builder.CreateTruncOrBitCast(Val, IntResTy); | |||
7067 | return Builder.CreateBitCast(Val, RealResTy); | |||
7068 | } | |||
7069 | ||||
7070 | if ((BuiltinID == AArch64::BI__builtin_arm_strex || | |||
7071 | BuiltinID == AArch64::BI__builtin_arm_stlex) && | |||
7072 | getContext().getTypeSize(E->getArg(0)->getType()) == 128) { | |||
7073 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex | |||
7074 | ? Intrinsic::aarch64_stlxp | |||
7075 | : Intrinsic::aarch64_stxp); | |||
7076 | llvm::Type *STy = llvm::StructType::get(Int64Ty, Int64Ty); | |||
7077 | ||||
7078 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); | |||
7079 | EmitAnyExprToMem(E->getArg(0), Tmp, Qualifiers(), /*init*/ true); | |||
7080 | ||||
7081 | Tmp = Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(STy)); | |||
7082 | llvm::Value *Val = Builder.CreateLoad(Tmp); | |||
7083 | ||||
7084 | Value *Arg0 = Builder.CreateExtractValue(Val, 0); | |||
7085 | Value *Arg1 = Builder.CreateExtractValue(Val, 1); | |||
7086 | Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), | |||
7087 | Int8PtrTy); | |||
7088 | return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "stxp"); | |||
7089 | } | |||
7090 | ||||
7091 | if (BuiltinID == AArch64::BI__builtin_arm_strex || | |||
7092 | BuiltinID == AArch64::BI__builtin_arm_stlex) { | |||
7093 | Value *StoreVal = EmitScalarExpr(E->getArg(0)); | |||
7094 | Value *StoreAddr = EmitScalarExpr(E->getArg(1)); | |||
7095 | ||||
7096 | QualType Ty = E->getArg(0)->getType(); | |||
7097 | llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(), | |||
7098 | getContext().getTypeSize(Ty)); | |||
7099 | StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo()); | |||
7100 | ||||
7101 | if (StoreVal->getType()->isPointerTy()) | |||
7102 | StoreVal = Builder.CreatePtrToInt(StoreVal, Int64Ty); | |||
7103 | else { | |||
7104 | llvm::Type *IntTy = llvm::IntegerType::get( | |||
7105 | getLLVMContext(), | |||
7106 | CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType())); | |||
7107 | StoreVal = Builder.CreateBitCast(StoreVal, IntTy); | |||
7108 | StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int64Ty); | |||
7109 | } | |||
7110 | ||||
7111 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex | |||
7112 | ? Intrinsic::aarch64_stlxr | |||
7113 | : Intrinsic::aarch64_stxr, | |||
7114 | StoreAddr->getType()); | |||
7115 | return Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr"); | |||
7116 | } | |||
7117 | ||||
7118 | if (BuiltinID == AArch64::BI__getReg) { | |||
7119 | Expr::EvalResult Result; | |||
7120 | if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext())) | |||
7121 | llvm_unreachable("Sema will ensure that the parameter is constant")::llvm::llvm_unreachable_internal("Sema will ensure that the parameter is constant" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7121); | |||
7122 | ||||
7123 | llvm::APSInt Value = Result.Val.getInt(); | |||
7124 | LLVMContext &Context = CGM.getLLVMContext(); | |||
7125 | std::string Reg = Value == 31 ? "sp" : "x" + Value.toString(10); | |||
7126 | ||||
7127 | llvm::Metadata *Ops[] = {llvm::MDString::get(Context, Reg)}; | |||
7128 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); | |||
7129 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); | |||
7130 | ||||
7131 | llvm::Function *F = | |||
7132 | CGM.getIntrinsic(llvm::Intrinsic::read_register, {Int64Ty}); | |||
7133 | return Builder.CreateCall(F, Metadata); | |||
7134 | } | |||
7135 | ||||
7136 | if (BuiltinID == AArch64::BI__builtin_arm_clrex) { | |||
7137 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_clrex); | |||
7138 | return Builder.CreateCall(F); | |||
7139 | } | |||
7140 | ||||
7141 | if (BuiltinID == AArch64::BI_ReadWriteBarrier) | |||
7142 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, | |||
7143 | llvm::SyncScope::SingleThread); | |||
7144 | ||||
7145 | // CRC32 | |||
7146 | Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic; | |||
7147 | switch (BuiltinID) { | |||
7148 | case AArch64::BI__builtin_arm_crc32b: | |||
7149 | CRCIntrinsicID = Intrinsic::aarch64_crc32b; break; | |||
7150 | case AArch64::BI__builtin_arm_crc32cb: | |||
7151 | CRCIntrinsicID = Intrinsic::aarch64_crc32cb; break; | |||
7152 | case AArch64::BI__builtin_arm_crc32h: | |||
7153 | CRCIntrinsicID = Intrinsic::aarch64_crc32h; break; | |||
7154 | case AArch64::BI__builtin_arm_crc32ch: | |||
7155 | CRCIntrinsicID = Intrinsic::aarch64_crc32ch; break; | |||
7156 | case AArch64::BI__builtin_arm_crc32w: | |||
7157 | CRCIntrinsicID = Intrinsic::aarch64_crc32w; break; | |||
7158 | case AArch64::BI__builtin_arm_crc32cw: | |||
7159 | CRCIntrinsicID = Intrinsic::aarch64_crc32cw; break; | |||
7160 | case AArch64::BI__builtin_arm_crc32d: | |||
7161 | CRCIntrinsicID = Intrinsic::aarch64_crc32x; break; | |||
7162 | case AArch64::BI__builtin_arm_crc32cd: | |||
7163 | CRCIntrinsicID = Intrinsic::aarch64_crc32cx; break; | |||
7164 | } | |||
7165 | ||||
7166 | if (CRCIntrinsicID != Intrinsic::not_intrinsic) { | |||
7167 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); | |||
7168 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
7169 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); | |||
7170 | ||||
7171 | llvm::Type *DataTy = F->getFunctionType()->getParamType(1); | |||
7172 | Arg1 = Builder.CreateZExtOrBitCast(Arg1, DataTy); | |||
7173 | ||||
7174 | return Builder.CreateCall(F, {Arg0, Arg1}); | |||
7175 | } | |||
7176 | ||||
7177 | // Memory Tagging Extensions (MTE) Intrinsics | |||
7178 | Intrinsic::ID MTEIntrinsicID = Intrinsic::not_intrinsic; | |||
7179 | switch (BuiltinID) { | |||
7180 | case AArch64::BI__builtin_arm_irg: | |||
7181 | MTEIntrinsicID = Intrinsic::aarch64_irg; break; | |||
7182 | case AArch64::BI__builtin_arm_addg: | |||
7183 | MTEIntrinsicID = Intrinsic::aarch64_addg; break; | |||
7184 | case AArch64::BI__builtin_arm_gmi: | |||
7185 | MTEIntrinsicID = Intrinsic::aarch64_gmi; break; | |||
7186 | case AArch64::BI__builtin_arm_ldg: | |||
7187 | MTEIntrinsicID = Intrinsic::aarch64_ldg; break; | |||
7188 | case AArch64::BI__builtin_arm_stg: | |||
7189 | MTEIntrinsicID = Intrinsic::aarch64_stg; break; | |||
7190 | case AArch64::BI__builtin_arm_subp: | |||
7191 | MTEIntrinsicID = Intrinsic::aarch64_subp; break; | |||
7192 | } | |||
7193 | ||||
7194 | if (MTEIntrinsicID != Intrinsic::not_intrinsic) { | |||
7195 | llvm::Type *T = ConvertType(E->getType()); | |||
7196 | ||||
7197 | if (MTEIntrinsicID == Intrinsic::aarch64_irg) { | |||
7198 | Value *Pointer = EmitScalarExpr(E->getArg(0)); | |||
7199 | Value *Mask = EmitScalarExpr(E->getArg(1)); | |||
7200 | ||||
7201 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); | |||
7202 | Mask = Builder.CreateZExt(Mask, Int64Ty); | |||
7203 | Value *RV = Builder.CreateCall( | |||
7204 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, Mask}); | |||
7205 | return Builder.CreatePointerCast(RV, T); | |||
7206 | } | |||
7207 | if (MTEIntrinsicID == Intrinsic::aarch64_addg) { | |||
7208 | Value *Pointer = EmitScalarExpr(E->getArg(0)); | |||
7209 | Value *TagOffset = EmitScalarExpr(E->getArg(1)); | |||
7210 | ||||
7211 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); | |||
7212 | TagOffset = Builder.CreateZExt(TagOffset, Int64Ty); | |||
7213 | Value *RV = Builder.CreateCall( | |||
7214 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, TagOffset}); | |||
7215 | return Builder.CreatePointerCast(RV, T); | |||
7216 | } | |||
7217 | if (MTEIntrinsicID == Intrinsic::aarch64_gmi) { | |||
7218 | Value *Pointer = EmitScalarExpr(E->getArg(0)); | |||
7219 | Value *ExcludedMask = EmitScalarExpr(E->getArg(1)); | |||
7220 | ||||
7221 | ExcludedMask = Builder.CreateZExt(ExcludedMask, Int64Ty); | |||
7222 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); | |||
7223 | return Builder.CreateCall( | |||
7224 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, ExcludedMask}); | |||
7225 | } | |||
7226 | // Although it is possible to supply a different return | |||
7227 | // address (first arg) to this intrinsic, for now we set | |||
7228 | // return address same as input address. | |||
7229 | if (MTEIntrinsicID == Intrinsic::aarch64_ldg) { | |||
7230 | Value *TagAddress = EmitScalarExpr(E->getArg(0)); | |||
7231 | TagAddress = Builder.CreatePointerCast(TagAddress, Int8PtrTy); | |||
7232 | Value *RV = Builder.CreateCall( | |||
7233 | CGM.getIntrinsic(MTEIntrinsicID), {TagAddress, TagAddress}); | |||
7234 | return Builder.CreatePointerCast(RV, T); | |||
7235 | } | |||
7236 | // Although it is possible to supply a different tag (to set) | |||
7237 | // to this intrinsic (as first arg), for now we supply | |||
7238 | // the tag that is in input address arg (common use case). | |||
7239 | if (MTEIntrinsicID == Intrinsic::aarch64_stg) { | |||
7240 | Value *TagAddress = EmitScalarExpr(E->getArg(0)); | |||
7241 | TagAddress = Builder.CreatePointerCast(TagAddress, Int8PtrTy); | |||
7242 | return Builder.CreateCall( | |||
7243 | CGM.getIntrinsic(MTEIntrinsicID), {TagAddress, TagAddress}); | |||
7244 | } | |||
7245 | if (MTEIntrinsicID == Intrinsic::aarch64_subp) { | |||
7246 | Value *PointerA = EmitScalarExpr(E->getArg(0)); | |||
7247 | Value *PointerB = EmitScalarExpr(E->getArg(1)); | |||
7248 | PointerA = Builder.CreatePointerCast(PointerA, Int8PtrTy); | |||
7249 | PointerB = Builder.CreatePointerCast(PointerB, Int8PtrTy); | |||
7250 | return Builder.CreateCall( | |||
7251 | CGM.getIntrinsic(MTEIntrinsicID), {PointerA, PointerB}); | |||
7252 | } | |||
7253 | } | |||
7254 | ||||
7255 | if (BuiltinID == AArch64::BI__builtin_arm_rsr || | |||
7256 | BuiltinID == AArch64::BI__builtin_arm_rsr64 || | |||
7257 | BuiltinID == AArch64::BI__builtin_arm_rsrp || | |||
7258 | BuiltinID == AArch64::BI__builtin_arm_wsr || | |||
7259 | BuiltinID == AArch64::BI__builtin_arm_wsr64 || | |||
7260 | BuiltinID == AArch64::BI__builtin_arm_wsrp) { | |||
7261 | ||||
7262 | bool IsRead = BuiltinID == AArch64::BI__builtin_arm_rsr || | |||
7263 | BuiltinID == AArch64::BI__builtin_arm_rsr64 || | |||
7264 | BuiltinID == AArch64::BI__builtin_arm_rsrp; | |||
7265 | ||||
7266 | bool IsPointerBuiltin = BuiltinID == AArch64::BI__builtin_arm_rsrp || | |||
7267 | BuiltinID == AArch64::BI__builtin_arm_wsrp; | |||
7268 | ||||
7269 | bool Is64Bit = BuiltinID != AArch64::BI__builtin_arm_rsr && | |||
7270 | BuiltinID != AArch64::BI__builtin_arm_wsr; | |||
7271 | ||||
7272 | llvm::Type *ValueType; | |||
7273 | llvm::Type *RegisterType = Int64Ty; | |||
7274 | if (IsPointerBuiltin) { | |||
7275 | ValueType = VoidPtrTy; | |||
7276 | } else if (Is64Bit) { | |||
7277 | ValueType = Int64Ty; | |||
7278 | } else { | |||
7279 | ValueType = Int32Ty; | |||
7280 | } | |||
7281 | ||||
7282 | return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, IsRead); | |||
7283 | } | |||
7284 | ||||
7285 | if (BuiltinID == AArch64::BI_ReadStatusReg || | |||
7286 | BuiltinID == AArch64::BI_WriteStatusReg) { | |||
7287 | LLVMContext &Context = CGM.getLLVMContext(); | |||
7288 | ||||
7289 | unsigned SysReg = | |||
7290 | E->getArg(0)->EvaluateKnownConstInt(getContext()).getZExtValue(); | |||
7291 | ||||
7292 | std::string SysRegStr; | |||
7293 | llvm::raw_string_ostream(SysRegStr) << | |||
7294 | ((1 << 1) | ((SysReg >> 14) & 1)) << ":" << | |||
7295 | ((SysReg >> 11) & 7) << ":" << | |||
7296 | ((SysReg >> 7) & 15) << ":" << | |||
7297 | ((SysReg >> 3) & 15) << ":" << | |||
7298 | ( SysReg & 7); | |||
7299 | ||||
7300 | llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysRegStr) }; | |||
7301 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); | |||
7302 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); | |||
7303 | ||||
7304 | llvm::Type *RegisterType = Int64Ty; | |||
7305 | llvm::Type *Types[] = { RegisterType }; | |||
7306 | ||||
7307 | if (BuiltinID == AArch64::BI_ReadStatusReg) { | |||
7308 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types); | |||
7309 | ||||
7310 | return Builder.CreateCall(F, Metadata); | |||
7311 | } | |||
7312 | ||||
7313 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types); | |||
7314 | llvm::Value *ArgValue = EmitScalarExpr(E->getArg(1)); | |||
7315 | ||||
7316 | return Builder.CreateCall(F, { Metadata, ArgValue }); | |||
7317 | } | |||
7318 | ||||
7319 | if (BuiltinID == AArch64::BI_AddressOfReturnAddress) { | |||
7320 | llvm::Function *F = | |||
7321 | CGM.getIntrinsic(Intrinsic::addressofreturnaddress, AllocaInt8PtrTy); | |||
7322 | return Builder.CreateCall(F); | |||
7323 | } | |||
7324 | ||||
7325 | if (BuiltinID == AArch64::BI__builtin_sponentry) { | |||
7326 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::sponentry, AllocaInt8PtrTy); | |||
7327 | return Builder.CreateCall(F); | |||
7328 | } | |||
7329 | ||||
7330 | // Find out if any arguments are required to be integer constant | |||
7331 | // expressions. | |||
7332 | unsigned ICEArguments = 0; | |||
7333 | ASTContext::GetBuiltinTypeError Error; | |||
7334 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
7335 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7335, __PRETTY_FUNCTION__)); | |||
7336 | ||||
7337 | llvm::SmallVector<Value*, 4> Ops; | |||
7338 | for (unsigned i = 0, e = E->getNumArgs() - 1; i != e; i++) { | |||
7339 | if ((ICEArguments & (1 << i)) == 0) { | |||
7340 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
7341 | } else { | |||
7342 | // If this is required to be a constant, constant fold it so that we know | |||
7343 | // that the generated intrinsic gets a ConstantInt. | |||
7344 | llvm::APSInt Result; | |||
7345 | bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext()); | |||
7346 | assert(IsConst && "Constant arg isn't actually constant?")((IsConst && "Constant arg isn't actually constant?") ? static_cast<void> (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7346, __PRETTY_FUNCTION__)); | |||
7347 | (void)IsConst; | |||
7348 | Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result)); | |||
7349 | } | |||
7350 | } | |||
7351 | ||||
7352 | auto SISDMap = makeArrayRef(AArch64SISDIntrinsicMap); | |||
7353 | const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap( | |||
7354 | SISDMap, BuiltinID, AArch64SISDIntrinsicsProvenSorted); | |||
7355 | ||||
7356 | if (Builtin) { | |||
7357 | Ops.push_back(EmitScalarExpr(E->getArg(E->getNumArgs() - 1))); | |||
7358 | Value *Result = EmitCommonNeonSISDBuiltinExpr(*this, *Builtin, Ops, E); | |||
7359 | assert(Result && "SISD intrinsic should have been handled")((Result && "SISD intrinsic should have been handled" ) ? static_cast<void> (0) : __assert_fail ("Result && \"SISD intrinsic should have been handled\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7359, __PRETTY_FUNCTION__)); | |||
7360 | return Result; | |||
7361 | } | |||
7362 | ||||
7363 | llvm::APSInt Result; | |||
7364 | const Expr *Arg = E->getArg(E->getNumArgs()-1); | |||
7365 | NeonTypeFlags Type(0); | |||
7366 | if (Arg->isIntegerConstantExpr(Result, getContext())) | |||
7367 | // Determine the type of this overloaded NEON intrinsic. | |||
7368 | Type = NeonTypeFlags(Result.getZExtValue()); | |||
7369 | ||||
7370 | bool usgn = Type.isUnsigned(); | |||
7371 | bool quad = Type.isQuad(); | |||
7372 | ||||
7373 | // Handle non-overloaded intrinsics first. | |||
7374 | switch (BuiltinID) { | |||
7375 | default: break; | |||
7376 | case NEON::BI__builtin_neon_vabsh_f16: | |||
7377 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7378 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, HalfTy), Ops, "vabs"); | |||
7379 | case NEON::BI__builtin_neon_vldrq_p128: { | |||
7380 | llvm::Type *Int128Ty = llvm::Type::getIntNTy(getLLVMContext(), 128); | |||
7381 | llvm::Type *Int128PTy = llvm::PointerType::get(Int128Ty, 0); | |||
7382 | Value *Ptr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int128PTy); | |||
7383 | return Builder.CreateAlignedLoad(Int128Ty, Ptr, | |||
7384 | CharUnits::fromQuantity(16)); | |||
7385 | } | |||
7386 | case NEON::BI__builtin_neon_vstrq_p128: { | |||
7387 | llvm::Type *Int128PTy = llvm::Type::getIntNPtrTy(getLLVMContext(), 128); | |||
7388 | Value *Ptr = Builder.CreateBitCast(Ops[0], Int128PTy); | |||
7389 | return Builder.CreateDefaultAlignedStore(EmitScalarExpr(E->getArg(1)), Ptr); | |||
7390 | } | |||
7391 | case NEON::BI__builtin_neon_vcvts_u32_f32: | |||
7392 | case NEON::BI__builtin_neon_vcvtd_u64_f64: | |||
7393 | usgn = true; | |||
7394 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
7395 | case NEON::BI__builtin_neon_vcvts_s32_f32: | |||
7396 | case NEON::BI__builtin_neon_vcvtd_s64_f64: { | |||
7397 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7398 | bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64; | |||
7399 | llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty; | |||
7400 | llvm::Type *FTy = Is64 ? DoubleTy : FloatTy; | |||
7401 | Ops[0] = Builder.CreateBitCast(Ops[0], FTy); | |||
7402 | if (usgn) | |||
7403 | return Builder.CreateFPToUI(Ops[0], InTy); | |||
7404 | return Builder.CreateFPToSI(Ops[0], InTy); | |||
7405 | } | |||
7406 | case NEON::BI__builtin_neon_vcvts_f32_u32: | |||
7407 | case NEON::BI__builtin_neon_vcvtd_f64_u64: | |||
7408 | usgn = true; | |||
7409 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
7410 | case NEON::BI__builtin_neon_vcvts_f32_s32: | |||
7411 | case NEON::BI__builtin_neon_vcvtd_f64_s64: { | |||
7412 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7413 | bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64; | |||
7414 | llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty; | |||
7415 | llvm::Type *FTy = Is64 ? DoubleTy : FloatTy; | |||
7416 | Ops[0] = Builder.CreateBitCast(Ops[0], InTy); | |||
7417 | if (usgn) | |||
7418 | return Builder.CreateUIToFP(Ops[0], FTy); | |||
7419 | return Builder.CreateSIToFP(Ops[0], FTy); | |||
7420 | } | |||
7421 | case NEON::BI__builtin_neon_vcvth_f16_u16: | |||
7422 | case NEON::BI__builtin_neon_vcvth_f16_u32: | |||
7423 | case NEON::BI__builtin_neon_vcvth_f16_u64: | |||
7424 | usgn = true; | |||
7425 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
7426 | case NEON::BI__builtin_neon_vcvth_f16_s16: | |||
7427 | case NEON::BI__builtin_neon_vcvth_f16_s32: | |||
7428 | case NEON::BI__builtin_neon_vcvth_f16_s64: { | |||
7429 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7430 | llvm::Type *FTy = HalfTy; | |||
7431 | llvm::Type *InTy; | |||
7432 | if (Ops[0]->getType()->getPrimitiveSizeInBits() == 64) | |||
7433 | InTy = Int64Ty; | |||
7434 | else if (Ops[0]->getType()->getPrimitiveSizeInBits() == 32) | |||
7435 | InTy = Int32Ty; | |||
7436 | else | |||
7437 | InTy = Int16Ty; | |||
7438 | Ops[0] = Builder.CreateBitCast(Ops[0], InTy); | |||
7439 | if (usgn) | |||
7440 | return Builder.CreateUIToFP(Ops[0], FTy); | |||
7441 | return Builder.CreateSIToFP(Ops[0], FTy); | |||
7442 | } | |||
7443 | case NEON::BI__builtin_neon_vcvth_u16_f16: | |||
7444 | usgn = true; | |||
7445 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
7446 | case NEON::BI__builtin_neon_vcvth_s16_f16: { | |||
7447 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7448 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); | |||
7449 | if (usgn) | |||
7450 | return Builder.CreateFPToUI(Ops[0], Int16Ty); | |||
7451 | return Builder.CreateFPToSI(Ops[0], Int16Ty); | |||
7452 | } | |||
7453 | case NEON::BI__builtin_neon_vcvth_u32_f16: | |||
7454 | usgn = true; | |||
7455 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
7456 | case NEON::BI__builtin_neon_vcvth_s32_f16: { | |||
7457 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7458 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); | |||
7459 | if (usgn) | |||
7460 | return Builder.CreateFPToUI(Ops[0], Int32Ty); | |||
7461 | return Builder.CreateFPToSI(Ops[0], Int32Ty); | |||
7462 | } | |||
7463 | case NEON::BI__builtin_neon_vcvth_u64_f16: | |||
7464 | usgn = true; | |||
7465 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
7466 | case NEON::BI__builtin_neon_vcvth_s64_f16: { | |||
7467 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7468 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); | |||
7469 | if (usgn) | |||
7470 | return Builder.CreateFPToUI(Ops[0], Int64Ty); | |||
7471 | return Builder.CreateFPToSI(Ops[0], Int64Ty); | |||
7472 | } | |||
7473 | case NEON::BI__builtin_neon_vcvtah_u16_f16: | |||
7474 | case NEON::BI__builtin_neon_vcvtmh_u16_f16: | |||
7475 | case NEON::BI__builtin_neon_vcvtnh_u16_f16: | |||
7476 | case NEON::BI__builtin_neon_vcvtph_u16_f16: | |||
7477 | case NEON::BI__builtin_neon_vcvtah_s16_f16: | |||
7478 | case NEON::BI__builtin_neon_vcvtmh_s16_f16: | |||
7479 | case NEON::BI__builtin_neon_vcvtnh_s16_f16: | |||
7480 | case NEON::BI__builtin_neon_vcvtph_s16_f16: { | |||
7481 | unsigned Int; | |||
7482 | llvm::Type* InTy = Int32Ty; | |||
7483 | llvm::Type* FTy = HalfTy; | |||
7484 | llvm::Type *Tys[2] = {InTy, FTy}; | |||
7485 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7486 | switch (BuiltinID) { | |||
7487 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7487); | |||
7488 | case NEON::BI__builtin_neon_vcvtah_u16_f16: | |||
7489 | Int = Intrinsic::aarch64_neon_fcvtau; break; | |||
7490 | case NEON::BI__builtin_neon_vcvtmh_u16_f16: | |||
7491 | Int = Intrinsic::aarch64_neon_fcvtmu; break; | |||
7492 | case NEON::BI__builtin_neon_vcvtnh_u16_f16: | |||
7493 | Int = Intrinsic::aarch64_neon_fcvtnu; break; | |||
7494 | case NEON::BI__builtin_neon_vcvtph_u16_f16: | |||
7495 | Int = Intrinsic::aarch64_neon_fcvtpu; break; | |||
7496 | case NEON::BI__builtin_neon_vcvtah_s16_f16: | |||
7497 | Int = Intrinsic::aarch64_neon_fcvtas; break; | |||
7498 | case NEON::BI__builtin_neon_vcvtmh_s16_f16: | |||
7499 | Int = Intrinsic::aarch64_neon_fcvtms; break; | |||
7500 | case NEON::BI__builtin_neon_vcvtnh_s16_f16: | |||
7501 | Int = Intrinsic::aarch64_neon_fcvtns; break; | |||
7502 | case NEON::BI__builtin_neon_vcvtph_s16_f16: | |||
7503 | Int = Intrinsic::aarch64_neon_fcvtps; break; | |||
7504 | } | |||
7505 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvt"); | |||
7506 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
7507 | } | |||
7508 | case NEON::BI__builtin_neon_vcaleh_f16: | |||
7509 | case NEON::BI__builtin_neon_vcalth_f16: | |||
7510 | case NEON::BI__builtin_neon_vcageh_f16: | |||
7511 | case NEON::BI__builtin_neon_vcagth_f16: { | |||
7512 | unsigned Int; | |||
7513 | llvm::Type* InTy = Int32Ty; | |||
7514 | llvm::Type* FTy = HalfTy; | |||
7515 | llvm::Type *Tys[2] = {InTy, FTy}; | |||
7516 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7517 | switch (BuiltinID) { | |||
7518 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7518); | |||
7519 | case NEON::BI__builtin_neon_vcageh_f16: | |||
7520 | Int = Intrinsic::aarch64_neon_facge; break; | |||
7521 | case NEON::BI__builtin_neon_vcagth_f16: | |||
7522 | Int = Intrinsic::aarch64_neon_facgt; break; | |||
7523 | case NEON::BI__builtin_neon_vcaleh_f16: | |||
7524 | Int = Intrinsic::aarch64_neon_facge; std::swap(Ops[0], Ops[1]); break; | |||
7525 | case NEON::BI__builtin_neon_vcalth_f16: | |||
7526 | Int = Intrinsic::aarch64_neon_facgt; std::swap(Ops[0], Ops[1]); break; | |||
7527 | } | |||
7528 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "facg"); | |||
7529 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
7530 | } | |||
7531 | case NEON::BI__builtin_neon_vcvth_n_s16_f16: | |||
7532 | case NEON::BI__builtin_neon_vcvth_n_u16_f16: { | |||
7533 | unsigned Int; | |||
7534 | llvm::Type* InTy = Int32Ty; | |||
7535 | llvm::Type* FTy = HalfTy; | |||
7536 | llvm::Type *Tys[2] = {InTy, FTy}; | |||
7537 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7538 | switch (BuiltinID) { | |||
7539 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7539); | |||
7540 | case NEON::BI__builtin_neon_vcvth_n_s16_f16: | |||
7541 | Int = Intrinsic::aarch64_neon_vcvtfp2fxs; break; | |||
7542 | case NEON::BI__builtin_neon_vcvth_n_u16_f16: | |||
7543 | Int = Intrinsic::aarch64_neon_vcvtfp2fxu; break; | |||
7544 | } | |||
7545 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n"); | |||
7546 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
7547 | } | |||
7548 | case NEON::BI__builtin_neon_vcvth_n_f16_s16: | |||
7549 | case NEON::BI__builtin_neon_vcvth_n_f16_u16: { | |||
7550 | unsigned Int; | |||
7551 | llvm::Type* FTy = HalfTy; | |||
7552 | llvm::Type* InTy = Int32Ty; | |||
7553 | llvm::Type *Tys[2] = {FTy, InTy}; | |||
7554 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7555 | switch (BuiltinID) { | |||
7556 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7556); | |||
7557 | case NEON::BI__builtin_neon_vcvth_n_f16_s16: | |||
7558 | Int = Intrinsic::aarch64_neon_vcvtfxs2fp; | |||
7559 | Ops[0] = Builder.CreateSExt(Ops[0], InTy, "sext"); | |||
7560 | break; | |||
7561 | case NEON::BI__builtin_neon_vcvth_n_f16_u16: | |||
7562 | Int = Intrinsic::aarch64_neon_vcvtfxu2fp; | |||
7563 | Ops[0] = Builder.CreateZExt(Ops[0], InTy); | |||
7564 | break; | |||
7565 | } | |||
7566 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n"); | |||
7567 | } | |||
7568 | case NEON::BI__builtin_neon_vpaddd_s64: { | |||
7569 | llvm::Type *Ty = llvm::VectorType::get(Int64Ty, 2); | |||
7570 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
7571 | // The vector is v2f64, so make sure it's bitcast to that. | |||
7572 | Vec = Builder.CreateBitCast(Vec, Ty, "v2i64"); | |||
7573 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); | |||
7574 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); | |||
7575 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); | |||
7576 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); | |||
7577 | // Pairwise addition of a v2f64 into a scalar f64. | |||
7578 | return Builder.CreateAdd(Op0, Op1, "vpaddd"); | |||
7579 | } | |||
7580 | case NEON::BI__builtin_neon_vpaddd_f64: { | |||
7581 | llvm::Type *Ty = | |||
7582 | llvm::VectorType::get(DoubleTy, 2); | |||
7583 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
7584 | // The vector is v2f64, so make sure it's bitcast to that. | |||
7585 | Vec = Builder.CreateBitCast(Vec, Ty, "v2f64"); | |||
7586 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); | |||
7587 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); | |||
7588 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); | |||
7589 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); | |||
7590 | // Pairwise addition of a v2f64 into a scalar f64. | |||
7591 | return Builder.CreateFAdd(Op0, Op1, "vpaddd"); | |||
7592 | } | |||
7593 | case NEON::BI__builtin_neon_vpadds_f32: { | |||
7594 | llvm::Type *Ty = | |||
7595 | llvm::VectorType::get(FloatTy, 2); | |||
7596 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
7597 | // The vector is v2f32, so make sure it's bitcast to that. | |||
7598 | Vec = Builder.CreateBitCast(Vec, Ty, "v2f32"); | |||
7599 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); | |||
7600 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); | |||
7601 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); | |||
7602 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); | |||
7603 | // Pairwise addition of a v2f32 into a scalar f32. | |||
7604 | return Builder.CreateFAdd(Op0, Op1, "vpaddd"); | |||
7605 | } | |||
7606 | case NEON::BI__builtin_neon_vceqzd_s64: | |||
7607 | case NEON::BI__builtin_neon_vceqzd_f64: | |||
7608 | case NEON::BI__builtin_neon_vceqzs_f32: | |||
7609 | case NEON::BI__builtin_neon_vceqzh_f16: | |||
7610 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7611 | return EmitAArch64CompareBuiltinExpr( | |||
7612 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
7613 | ICmpInst::FCMP_OEQ, ICmpInst::ICMP_EQ, "vceqz"); | |||
7614 | case NEON::BI__builtin_neon_vcgezd_s64: | |||
7615 | case NEON::BI__builtin_neon_vcgezd_f64: | |||
7616 | case NEON::BI__builtin_neon_vcgezs_f32: | |||
7617 | case NEON::BI__builtin_neon_vcgezh_f16: | |||
7618 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7619 | return EmitAArch64CompareBuiltinExpr( | |||
7620 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
7621 | ICmpInst::FCMP_OGE, ICmpInst::ICMP_SGE, "vcgez"); | |||
7622 | case NEON::BI__builtin_neon_vclezd_s64: | |||
7623 | case NEON::BI__builtin_neon_vclezd_f64: | |||
7624 | case NEON::BI__builtin_neon_vclezs_f32: | |||
7625 | case NEON::BI__builtin_neon_vclezh_f16: | |||
7626 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7627 | return EmitAArch64CompareBuiltinExpr( | |||
7628 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
7629 | ICmpInst::FCMP_OLE, ICmpInst::ICMP_SLE, "vclez"); | |||
7630 | case NEON::BI__builtin_neon_vcgtzd_s64: | |||
7631 | case NEON::BI__builtin_neon_vcgtzd_f64: | |||
7632 | case NEON::BI__builtin_neon_vcgtzs_f32: | |||
7633 | case NEON::BI__builtin_neon_vcgtzh_f16: | |||
7634 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7635 | return EmitAArch64CompareBuiltinExpr( | |||
7636 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
7637 | ICmpInst::FCMP_OGT, ICmpInst::ICMP_SGT, "vcgtz"); | |||
7638 | case NEON::BI__builtin_neon_vcltzd_s64: | |||
7639 | case NEON::BI__builtin_neon_vcltzd_f64: | |||
7640 | case NEON::BI__builtin_neon_vcltzs_f32: | |||
7641 | case NEON::BI__builtin_neon_vcltzh_f16: | |||
7642 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7643 | return EmitAArch64CompareBuiltinExpr( | |||
7644 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
7645 | ICmpInst::FCMP_OLT, ICmpInst::ICMP_SLT, "vcltz"); | |||
7646 | ||||
7647 | case NEON::BI__builtin_neon_vceqzd_u64: { | |||
7648 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
7649 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); | |||
7650 | Ops[0] = | |||
7651 | Builder.CreateICmpEQ(Ops[0], llvm::Constant::getNullValue(Int64Ty)); | |||
7652 | return Builder.CreateSExt(Ops[0], Int64Ty, "vceqzd"); | |||
7653 | } | |||
7654 | case NEON::BI__builtin_neon_vceqd_f64: | |||
7655 | case NEON::BI__builtin_neon_vcled_f64: | |||
7656 | case NEON::BI__builtin_neon_vcltd_f64: | |||
7657 | case NEON::BI__builtin_neon_vcged_f64: | |||
7658 | case NEON::BI__builtin_neon_vcgtd_f64: { | |||
7659 | llvm::CmpInst::Predicate P; | |||
7660 | switch (BuiltinID) { | |||
7661 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7661); | |||
7662 | case NEON::BI__builtin_neon_vceqd_f64: P = llvm::FCmpInst::FCMP_OEQ; break; | |||
7663 | case NEON::BI__builtin_neon_vcled_f64: P = llvm::FCmpInst::FCMP_OLE; break; | |||
7664 | case NEON::BI__builtin_neon_vcltd_f64: P = llvm::FCmpInst::FCMP_OLT; break; | |||
7665 | case NEON::BI__builtin_neon_vcged_f64: P = llvm::FCmpInst::FCMP_OGE; break; | |||
7666 | case NEON::BI__builtin_neon_vcgtd_f64: P = llvm::FCmpInst::FCMP_OGT; break; | |||
7667 | } | |||
7668 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7669 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); | |||
7670 | Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy); | |||
7671 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); | |||
7672 | return Builder.CreateSExt(Ops[0], Int64Ty, "vcmpd"); | |||
7673 | } | |||
7674 | case NEON::BI__builtin_neon_vceqs_f32: | |||
7675 | case NEON::BI__builtin_neon_vcles_f32: | |||
7676 | case NEON::BI__builtin_neon_vclts_f32: | |||
7677 | case NEON::BI__builtin_neon_vcges_f32: | |||
7678 | case NEON::BI__builtin_neon_vcgts_f32: { | |||
7679 | llvm::CmpInst::Predicate P; | |||
7680 | switch (BuiltinID) { | |||
7681 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7681); | |||
7682 | case NEON::BI__builtin_neon_vceqs_f32: P = llvm::FCmpInst::FCMP_OEQ; break; | |||
7683 | case NEON::BI__builtin_neon_vcles_f32: P = llvm::FCmpInst::FCMP_OLE; break; | |||
7684 | case NEON::BI__builtin_neon_vclts_f32: P = llvm::FCmpInst::FCMP_OLT; break; | |||
7685 | case NEON::BI__builtin_neon_vcges_f32: P = llvm::FCmpInst::FCMP_OGE; break; | |||
7686 | case NEON::BI__builtin_neon_vcgts_f32: P = llvm::FCmpInst::FCMP_OGT; break; | |||
7687 | } | |||
7688 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7689 | Ops[0] = Builder.CreateBitCast(Ops[0], FloatTy); | |||
7690 | Ops[1] = Builder.CreateBitCast(Ops[1], FloatTy); | |||
7691 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); | |||
7692 | return Builder.CreateSExt(Ops[0], Int32Ty, "vcmpd"); | |||
7693 | } | |||
7694 | case NEON::BI__builtin_neon_vceqh_f16: | |||
7695 | case NEON::BI__builtin_neon_vcleh_f16: | |||
7696 | case NEON::BI__builtin_neon_vclth_f16: | |||
7697 | case NEON::BI__builtin_neon_vcgeh_f16: | |||
7698 | case NEON::BI__builtin_neon_vcgth_f16: { | |||
7699 | llvm::CmpInst::Predicate P; | |||
7700 | switch (BuiltinID) { | |||
7701 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7701); | |||
7702 | case NEON::BI__builtin_neon_vceqh_f16: P = llvm::FCmpInst::FCMP_OEQ; break; | |||
7703 | case NEON::BI__builtin_neon_vcleh_f16: P = llvm::FCmpInst::FCMP_OLE; break; | |||
7704 | case NEON::BI__builtin_neon_vclth_f16: P = llvm::FCmpInst::FCMP_OLT; break; | |||
7705 | case NEON::BI__builtin_neon_vcgeh_f16: P = llvm::FCmpInst::FCMP_OGE; break; | |||
7706 | case NEON::BI__builtin_neon_vcgth_f16: P = llvm::FCmpInst::FCMP_OGT; break; | |||
7707 | } | |||
7708 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7709 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); | |||
7710 | Ops[1] = Builder.CreateBitCast(Ops[1], HalfTy); | |||
7711 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); | |||
7712 | return Builder.CreateSExt(Ops[0], Int16Ty, "vcmpd"); | |||
7713 | } | |||
7714 | case NEON::BI__builtin_neon_vceqd_s64: | |||
7715 | case NEON::BI__builtin_neon_vceqd_u64: | |||
7716 | case NEON::BI__builtin_neon_vcgtd_s64: | |||
7717 | case NEON::BI__builtin_neon_vcgtd_u64: | |||
7718 | case NEON::BI__builtin_neon_vcltd_s64: | |||
7719 | case NEON::BI__builtin_neon_vcltd_u64: | |||
7720 | case NEON::BI__builtin_neon_vcged_u64: | |||
7721 | case NEON::BI__builtin_neon_vcged_s64: | |||
7722 | case NEON::BI__builtin_neon_vcled_u64: | |||
7723 | case NEON::BI__builtin_neon_vcled_s64: { | |||
7724 | llvm::CmpInst::Predicate P; | |||
7725 | switch (BuiltinID) { | |||
7726 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7726); | |||
7727 | case NEON::BI__builtin_neon_vceqd_s64: | |||
7728 | case NEON::BI__builtin_neon_vceqd_u64:P = llvm::ICmpInst::ICMP_EQ;break; | |||
7729 | case NEON::BI__builtin_neon_vcgtd_s64:P = llvm::ICmpInst::ICMP_SGT;break; | |||
7730 | case NEON::BI__builtin_neon_vcgtd_u64:P = llvm::ICmpInst::ICMP_UGT;break; | |||
7731 | case NEON::BI__builtin_neon_vcltd_s64:P = llvm::ICmpInst::ICMP_SLT;break; | |||
7732 | case NEON::BI__builtin_neon_vcltd_u64:P = llvm::ICmpInst::ICMP_ULT;break; | |||
7733 | case NEON::BI__builtin_neon_vcged_u64:P = llvm::ICmpInst::ICMP_UGE;break; | |||
7734 | case NEON::BI__builtin_neon_vcged_s64:P = llvm::ICmpInst::ICMP_SGE;break; | |||
7735 | case NEON::BI__builtin_neon_vcled_u64:P = llvm::ICmpInst::ICMP_ULE;break; | |||
7736 | case NEON::BI__builtin_neon_vcled_s64:P = llvm::ICmpInst::ICMP_SLE;break; | |||
7737 | } | |||
7738 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7739 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); | |||
7740 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); | |||
7741 | Ops[0] = Builder.CreateICmp(P, Ops[0], Ops[1]); | |||
7742 | return Builder.CreateSExt(Ops[0], Int64Ty, "vceqd"); | |||
7743 | } | |||
7744 | case NEON::BI__builtin_neon_vtstd_s64: | |||
7745 | case NEON::BI__builtin_neon_vtstd_u64: { | |||
7746 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7747 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); | |||
7748 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); | |||
7749 | Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]); | |||
7750 | Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0], | |||
7751 | llvm::Constant::getNullValue(Int64Ty)); | |||
7752 | return Builder.CreateSExt(Ops[0], Int64Ty, "vtstd"); | |||
7753 | } | |||
7754 | case NEON::BI__builtin_neon_vset_lane_i8: | |||
7755 | case NEON::BI__builtin_neon_vset_lane_i16: | |||
7756 | case NEON::BI__builtin_neon_vset_lane_i32: | |||
7757 | case NEON::BI__builtin_neon_vset_lane_i64: | |||
7758 | case NEON::BI__builtin_neon_vset_lane_f32: | |||
7759 | case NEON::BI__builtin_neon_vsetq_lane_i8: | |||
7760 | case NEON::BI__builtin_neon_vsetq_lane_i16: | |||
7761 | case NEON::BI__builtin_neon_vsetq_lane_i32: | |||
7762 | case NEON::BI__builtin_neon_vsetq_lane_i64: | |||
7763 | case NEON::BI__builtin_neon_vsetq_lane_f32: | |||
7764 | Ops.push_back(EmitScalarExpr(E->getArg(2))); | |||
7765 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); | |||
7766 | case NEON::BI__builtin_neon_vset_lane_f64: | |||
7767 | // The vector type needs a cast for the v1f64 variant. | |||
7768 | Ops[1] = Builder.CreateBitCast(Ops[1], | |||
7769 | llvm::VectorType::get(DoubleTy, 1)); | |||
7770 | Ops.push_back(EmitScalarExpr(E->getArg(2))); | |||
7771 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); | |||
7772 | case NEON::BI__builtin_neon_vsetq_lane_f64: | |||
7773 | // The vector type needs a cast for the v2f64 variant. | |||
7774 | Ops[1] = Builder.CreateBitCast(Ops[1], | |||
7775 | llvm::VectorType::get(DoubleTy, 2)); | |||
7776 | Ops.push_back(EmitScalarExpr(E->getArg(2))); | |||
7777 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); | |||
7778 | ||||
7779 | case NEON::BI__builtin_neon_vget_lane_i8: | |||
7780 | case NEON::BI__builtin_neon_vdupb_lane_i8: | |||
7781 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int8Ty, 8)); | |||
7782 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7783 | "vget_lane"); | |||
7784 | case NEON::BI__builtin_neon_vgetq_lane_i8: | |||
7785 | case NEON::BI__builtin_neon_vdupb_laneq_i8: | |||
7786 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int8Ty, 16)); | |||
7787 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7788 | "vgetq_lane"); | |||
7789 | case NEON::BI__builtin_neon_vget_lane_i16: | |||
7790 | case NEON::BI__builtin_neon_vduph_lane_i16: | |||
7791 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int16Ty, 4)); | |||
7792 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7793 | "vget_lane"); | |||
7794 | case NEON::BI__builtin_neon_vgetq_lane_i16: | |||
7795 | case NEON::BI__builtin_neon_vduph_laneq_i16: | |||
7796 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int16Ty, 8)); | |||
7797 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7798 | "vgetq_lane"); | |||
7799 | case NEON::BI__builtin_neon_vget_lane_i32: | |||
7800 | case NEON::BI__builtin_neon_vdups_lane_i32: | |||
7801 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 2)); | |||
7802 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7803 | "vget_lane"); | |||
7804 | case NEON::BI__builtin_neon_vdups_lane_f32: | |||
7805 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
7806 | llvm::VectorType::get(FloatTy, 2)); | |||
7807 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7808 | "vdups_lane"); | |||
7809 | case NEON::BI__builtin_neon_vgetq_lane_i32: | |||
7810 | case NEON::BI__builtin_neon_vdups_laneq_i32: | |||
7811 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4)); | |||
7812 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7813 | "vgetq_lane"); | |||
7814 | case NEON::BI__builtin_neon_vget_lane_i64: | |||
7815 | case NEON::BI__builtin_neon_vdupd_lane_i64: | |||
7816 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 1)); | |||
7817 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7818 | "vget_lane"); | |||
7819 | case NEON::BI__builtin_neon_vdupd_lane_f64: | |||
7820 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
7821 | llvm::VectorType::get(DoubleTy, 1)); | |||
7822 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7823 | "vdupd_lane"); | |||
7824 | case NEON::BI__builtin_neon_vgetq_lane_i64: | |||
7825 | case NEON::BI__builtin_neon_vdupd_laneq_i64: | |||
7826 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2)); | |||
7827 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7828 | "vgetq_lane"); | |||
7829 | case NEON::BI__builtin_neon_vget_lane_f32: | |||
7830 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
7831 | llvm::VectorType::get(FloatTy, 2)); | |||
7832 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7833 | "vget_lane"); | |||
7834 | case NEON::BI__builtin_neon_vget_lane_f64: | |||
7835 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
7836 | llvm::VectorType::get(DoubleTy, 1)); | |||
7837 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7838 | "vget_lane"); | |||
7839 | case NEON::BI__builtin_neon_vgetq_lane_f32: | |||
7840 | case NEON::BI__builtin_neon_vdups_laneq_f32: | |||
7841 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
7842 | llvm::VectorType::get(FloatTy, 4)); | |||
7843 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7844 | "vgetq_lane"); | |||
7845 | case NEON::BI__builtin_neon_vgetq_lane_f64: | |||
7846 | case NEON::BI__builtin_neon_vdupd_laneq_f64: | |||
7847 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
7848 | llvm::VectorType::get(DoubleTy, 2)); | |||
7849 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
7850 | "vgetq_lane"); | |||
7851 | case NEON::BI__builtin_neon_vaddh_f16: | |||
7852 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7853 | return Builder.CreateFAdd(Ops[0], Ops[1], "vaddh"); | |||
7854 | case NEON::BI__builtin_neon_vsubh_f16: | |||
7855 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7856 | return Builder.CreateFSub(Ops[0], Ops[1], "vsubh"); | |||
7857 | case NEON::BI__builtin_neon_vmulh_f16: | |||
7858 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7859 | return Builder.CreateFMul(Ops[0], Ops[1], "vmulh"); | |||
7860 | case NEON::BI__builtin_neon_vdivh_f16: | |||
7861 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7862 | return Builder.CreateFDiv(Ops[0], Ops[1], "vdivh"); | |||
7863 | case NEON::BI__builtin_neon_vfmah_f16: { | |||
7864 | Function *F = CGM.getIntrinsic(Intrinsic::fma, HalfTy); | |||
7865 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. | |||
7866 | return Builder.CreateCall(F, | |||
7867 | {EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), Ops[0]}); | |||
7868 | } | |||
7869 | case NEON::BI__builtin_neon_vfmsh_f16: { | |||
7870 | Function *F = CGM.getIntrinsic(Intrinsic::fma, HalfTy); | |||
7871 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(HalfTy); | |||
7872 | Value* Sub = Builder.CreateFSub(Zero, EmitScalarExpr(E->getArg(1)), "vsubh"); | |||
7873 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. | |||
7874 | return Builder.CreateCall(F, {Sub, EmitScalarExpr(E->getArg(2)), Ops[0]}); | |||
7875 | } | |||
7876 | case NEON::BI__builtin_neon_vaddd_s64: | |||
7877 | case NEON::BI__builtin_neon_vaddd_u64: | |||
7878 | return Builder.CreateAdd(Ops[0], EmitScalarExpr(E->getArg(1)), "vaddd"); | |||
7879 | case NEON::BI__builtin_neon_vsubd_s64: | |||
7880 | case NEON::BI__builtin_neon_vsubd_u64: | |||
7881 | return Builder.CreateSub(Ops[0], EmitScalarExpr(E->getArg(1)), "vsubd"); | |||
7882 | case NEON::BI__builtin_neon_vqdmlalh_s16: | |||
7883 | case NEON::BI__builtin_neon_vqdmlslh_s16: { | |||
7884 | SmallVector<Value *, 2> ProductOps; | |||
7885 | ProductOps.push_back(vectorWrapScalar16(Ops[1])); | |||
7886 | ProductOps.push_back(vectorWrapScalar16(EmitScalarExpr(E->getArg(2)))); | |||
7887 | llvm::Type *VTy = llvm::VectorType::get(Int32Ty, 4); | |||
7888 | Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy), | |||
7889 | ProductOps, "vqdmlXl"); | |||
7890 | Constant *CI = ConstantInt::get(SizeTy, 0); | |||
7891 | Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0"); | |||
7892 | ||||
7893 | unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlalh_s16 | |||
7894 | ? Intrinsic::aarch64_neon_sqadd | |||
7895 | : Intrinsic::aarch64_neon_sqsub; | |||
7896 | return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int32Ty), Ops, "vqdmlXl"); | |||
7897 | } | |||
7898 | case NEON::BI__builtin_neon_vqshlud_n_s64: { | |||
7899 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7900 | Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty); | |||
7901 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqshlu, Int64Ty), | |||
7902 | Ops, "vqshlu_n"); | |||
7903 | } | |||
7904 | case NEON::BI__builtin_neon_vqshld_n_u64: | |||
7905 | case NEON::BI__builtin_neon_vqshld_n_s64: { | |||
7906 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vqshld_n_u64 | |||
7907 | ? Intrinsic::aarch64_neon_uqshl | |||
7908 | : Intrinsic::aarch64_neon_sqshl; | |||
7909 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7910 | Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty); | |||
7911 | return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vqshl_n"); | |||
7912 | } | |||
7913 | case NEON::BI__builtin_neon_vrshrd_n_u64: | |||
7914 | case NEON::BI__builtin_neon_vrshrd_n_s64: { | |||
7915 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrshrd_n_u64 | |||
7916 | ? Intrinsic::aarch64_neon_urshl | |||
7917 | : Intrinsic::aarch64_neon_srshl; | |||
7918 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
7919 | int SV = cast<ConstantInt>(Ops[1])->getSExtValue(); | |||
7920 | Ops[1] = ConstantInt::get(Int64Ty, -SV); | |||
7921 | return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vrshr_n"); | |||
7922 | } | |||
7923 | case NEON::BI__builtin_neon_vrsrad_n_u64: | |||
7924 | case NEON::BI__builtin_neon_vrsrad_n_s64: { | |||
7925 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrsrad_n_u64 | |||
7926 | ? Intrinsic::aarch64_neon_urshl | |||
7927 | : Intrinsic::aarch64_neon_srshl; | |||
7928 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); | |||
7929 | Ops.push_back(Builder.CreateNeg(EmitScalarExpr(E->getArg(2)))); | |||
7930 | Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Int64Ty), | |||
7931 | {Ops[1], Builder.CreateSExt(Ops[2], Int64Ty)}); | |||
7932 | return Builder.CreateAdd(Ops[0], Builder.CreateBitCast(Ops[1], Int64Ty)); | |||
7933 | } | |||
7934 | case NEON::BI__builtin_neon_vshld_n_s64: | |||
7935 | case NEON::BI__builtin_neon_vshld_n_u64: { | |||
7936 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
7937 | return Builder.CreateShl( | |||
7938 | Ops[0], ConstantInt::get(Int64Ty, Amt->getZExtValue()), "shld_n"); | |||
7939 | } | |||
7940 | case NEON::BI__builtin_neon_vshrd_n_s64: { | |||
7941 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
7942 | return Builder.CreateAShr( | |||
7943 | Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63), | |||
7944 | Amt->getZExtValue())), | |||
7945 | "shrd_n"); | |||
7946 | } | |||
7947 | case NEON::BI__builtin_neon_vshrd_n_u64: { | |||
7948 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
7949 | uint64_t ShiftAmt = Amt->getZExtValue(); | |||
7950 | // Right-shifting an unsigned value by its size yields 0. | |||
7951 | if (ShiftAmt == 64) | |||
7952 | return ConstantInt::get(Int64Ty, 0); | |||
7953 | return Builder.CreateLShr(Ops[0], ConstantInt::get(Int64Ty, ShiftAmt), | |||
7954 | "shrd_n"); | |||
7955 | } | |||
7956 | case NEON::BI__builtin_neon_vsrad_n_s64: { | |||
7957 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2))); | |||
7958 | Ops[1] = Builder.CreateAShr( | |||
7959 | Ops[1], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63), | |||
7960 | Amt->getZExtValue())), | |||
7961 | "shrd_n"); | |||
7962 | return Builder.CreateAdd(Ops[0], Ops[1]); | |||
7963 | } | |||
7964 | case NEON::BI__builtin_neon_vsrad_n_u64: { | |||
7965 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2))); | |||
7966 | uint64_t ShiftAmt = Amt->getZExtValue(); | |||
7967 | // Right-shifting an unsigned value by its size yields 0. | |||
7968 | // As Op + 0 = Op, return Ops[0] directly. | |||
7969 | if (ShiftAmt == 64) | |||
7970 | return Ops[0]; | |||
7971 | Ops[1] = Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, ShiftAmt), | |||
7972 | "shrd_n"); | |||
7973 | return Builder.CreateAdd(Ops[0], Ops[1]); | |||
7974 | } | |||
7975 | case NEON::BI__builtin_neon_vqdmlalh_lane_s16: | |||
7976 | case NEON::BI__builtin_neon_vqdmlalh_laneq_s16: | |||
7977 | case NEON::BI__builtin_neon_vqdmlslh_lane_s16: | |||
7978 | case NEON::BI__builtin_neon_vqdmlslh_laneq_s16: { | |||
7979 | Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)), | |||
7980 | "lane"); | |||
7981 | SmallVector<Value *, 2> ProductOps; | |||
7982 | ProductOps.push_back(vectorWrapScalar16(Ops[1])); | |||
7983 | ProductOps.push_back(vectorWrapScalar16(Ops[2])); | |||
7984 | llvm::Type *VTy = llvm::VectorType::get(Int32Ty, 4); | |||
7985 | Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy), | |||
7986 | ProductOps, "vqdmlXl"); | |||
7987 | Constant *CI = ConstantInt::get(SizeTy, 0); | |||
7988 | Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0"); | |||
7989 | Ops.pop_back(); | |||
7990 | ||||
7991 | unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlalh_lane_s16 || | |||
7992 | BuiltinID == NEON::BI__builtin_neon_vqdmlalh_laneq_s16) | |||
7993 | ? Intrinsic::aarch64_neon_sqadd | |||
7994 | : Intrinsic::aarch64_neon_sqsub; | |||
7995 | return EmitNeonCall(CGM.getIntrinsic(AccInt, Int32Ty), Ops, "vqdmlXl"); | |||
7996 | } | |||
7997 | case NEON::BI__builtin_neon_vqdmlals_s32: | |||
7998 | case NEON::BI__builtin_neon_vqdmlsls_s32: { | |||
7999 | SmallVector<Value *, 2> ProductOps; | |||
8000 | ProductOps.push_back(Ops[1]); | |||
8001 | ProductOps.push_back(EmitScalarExpr(E->getArg(2))); | |||
8002 | Ops[1] = | |||
8003 | EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar), | |||
8004 | ProductOps, "vqdmlXl"); | |||
8005 | ||||
8006 | unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlals_s32 | |||
8007 | ? Intrinsic::aarch64_neon_sqadd | |||
8008 | : Intrinsic::aarch64_neon_sqsub; | |||
8009 | return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int64Ty), Ops, "vqdmlXl"); | |||
8010 | } | |||
8011 | case NEON::BI__builtin_neon_vqdmlals_lane_s32: | |||
8012 | case NEON::BI__builtin_neon_vqdmlals_laneq_s32: | |||
8013 | case NEON::BI__builtin_neon_vqdmlsls_lane_s32: | |||
8014 | case NEON::BI__builtin_neon_vqdmlsls_laneq_s32: { | |||
8015 | Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)), | |||
8016 | "lane"); | |||
8017 | SmallVector<Value *, 2> ProductOps; | |||
8018 | ProductOps.push_back(Ops[1]); | |||
8019 | ProductOps.push_back(Ops[2]); | |||
8020 | Ops[1] = | |||
8021 | EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar), | |||
8022 | ProductOps, "vqdmlXl"); | |||
8023 | Ops.pop_back(); | |||
8024 | ||||
8025 | unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlals_lane_s32 || | |||
8026 | BuiltinID == NEON::BI__builtin_neon_vqdmlals_laneq_s32) | |||
8027 | ? Intrinsic::aarch64_neon_sqadd | |||
8028 | : Intrinsic::aarch64_neon_sqsub; | |||
8029 | return EmitNeonCall(CGM.getIntrinsic(AccInt, Int64Ty), Ops, "vqdmlXl"); | |||
8030 | } | |||
8031 | case NEON::BI__builtin_neon_vduph_lane_f16: { | |||
8032 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
8033 | "vget_lane"); | |||
8034 | } | |||
8035 | case NEON::BI__builtin_neon_vduph_laneq_f16: { | |||
8036 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
8037 | "vgetq_lane"); | |||
8038 | } | |||
8039 | case AArch64::BI_BitScanForward: | |||
8040 | case AArch64::BI_BitScanForward64: | |||
8041 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E); | |||
8042 | case AArch64::BI_BitScanReverse: | |||
8043 | case AArch64::BI_BitScanReverse64: | |||
8044 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E); | |||
8045 | case AArch64::BI_InterlockedAnd64: | |||
8046 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E); | |||
8047 | case AArch64::BI_InterlockedExchange64: | |||
8048 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E); | |||
8049 | case AArch64::BI_InterlockedExchangeAdd64: | |||
8050 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E); | |||
8051 | case AArch64::BI_InterlockedExchangeSub64: | |||
8052 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E); | |||
8053 | case AArch64::BI_InterlockedOr64: | |||
8054 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E); | |||
8055 | case AArch64::BI_InterlockedXor64: | |||
8056 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E); | |||
8057 | case AArch64::BI_InterlockedDecrement64: | |||
8058 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E); | |||
8059 | case AArch64::BI_InterlockedIncrement64: | |||
8060 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E); | |||
8061 | case AArch64::BI_InterlockedExchangeAdd8_acq: | |||
8062 | case AArch64::BI_InterlockedExchangeAdd16_acq: | |||
8063 | case AArch64::BI_InterlockedExchangeAdd_acq: | |||
8064 | case AArch64::BI_InterlockedExchangeAdd64_acq: | |||
8065 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_acq, E); | |||
8066 | case AArch64::BI_InterlockedExchangeAdd8_rel: | |||
8067 | case AArch64::BI_InterlockedExchangeAdd16_rel: | |||
8068 | case AArch64::BI_InterlockedExchangeAdd_rel: | |||
8069 | case AArch64::BI_InterlockedExchangeAdd64_rel: | |||
8070 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_rel, E); | |||
8071 | case AArch64::BI_InterlockedExchangeAdd8_nf: | |||
8072 | case AArch64::BI_InterlockedExchangeAdd16_nf: | |||
8073 | case AArch64::BI_InterlockedExchangeAdd_nf: | |||
8074 | case AArch64::BI_InterlockedExchangeAdd64_nf: | |||
8075 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_nf, E); | |||
8076 | case AArch64::BI_InterlockedExchange8_acq: | |||
8077 | case AArch64::BI_InterlockedExchange16_acq: | |||
8078 | case AArch64::BI_InterlockedExchange_acq: | |||
8079 | case AArch64::BI_InterlockedExchange64_acq: | |||
8080 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_acq, E); | |||
8081 | case AArch64::BI_InterlockedExchange8_rel: | |||
8082 | case AArch64::BI_InterlockedExchange16_rel: | |||
8083 | case AArch64::BI_InterlockedExchange_rel: | |||
8084 | case AArch64::BI_InterlockedExchange64_rel: | |||
8085 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_rel, E); | |||
8086 | case AArch64::BI_InterlockedExchange8_nf: | |||
8087 | case AArch64::BI_InterlockedExchange16_nf: | |||
8088 | case AArch64::BI_InterlockedExchange_nf: | |||
8089 | case AArch64::BI_InterlockedExchange64_nf: | |||
8090 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_nf, E); | |||
8091 | case AArch64::BI_InterlockedCompareExchange8_acq: | |||
8092 | case AArch64::BI_InterlockedCompareExchange16_acq: | |||
8093 | case AArch64::BI_InterlockedCompareExchange_acq: | |||
8094 | case AArch64::BI_InterlockedCompareExchange64_acq: | |||
8095 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_acq, E); | |||
8096 | case AArch64::BI_InterlockedCompareExchange8_rel: | |||
8097 | case AArch64::BI_InterlockedCompareExchange16_rel: | |||
8098 | case AArch64::BI_InterlockedCompareExchange_rel: | |||
8099 | case AArch64::BI_InterlockedCompareExchange64_rel: | |||
8100 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_rel, E); | |||
8101 | case AArch64::BI_InterlockedCompareExchange8_nf: | |||
8102 | case AArch64::BI_InterlockedCompareExchange16_nf: | |||
8103 | case AArch64::BI_InterlockedCompareExchange_nf: | |||
8104 | case AArch64::BI_InterlockedCompareExchange64_nf: | |||
8105 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_nf, E); | |||
8106 | case AArch64::BI_InterlockedOr8_acq: | |||
8107 | case AArch64::BI_InterlockedOr16_acq: | |||
8108 | case AArch64::BI_InterlockedOr_acq: | |||
8109 | case AArch64::BI_InterlockedOr64_acq: | |||
8110 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_acq, E); | |||
8111 | case AArch64::BI_InterlockedOr8_rel: | |||
8112 | case AArch64::BI_InterlockedOr16_rel: | |||
8113 | case AArch64::BI_InterlockedOr_rel: | |||
8114 | case AArch64::BI_InterlockedOr64_rel: | |||
8115 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_rel, E); | |||
8116 | case AArch64::BI_InterlockedOr8_nf: | |||
8117 | case AArch64::BI_InterlockedOr16_nf: | |||
8118 | case AArch64::BI_InterlockedOr_nf: | |||
8119 | case AArch64::BI_InterlockedOr64_nf: | |||
8120 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_nf, E); | |||
8121 | case AArch64::BI_InterlockedXor8_acq: | |||
8122 | case AArch64::BI_InterlockedXor16_acq: | |||
8123 | case AArch64::BI_InterlockedXor_acq: | |||
8124 | case AArch64::BI_InterlockedXor64_acq: | |||
8125 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_acq, E); | |||
8126 | case AArch64::BI_InterlockedXor8_rel: | |||
8127 | case AArch64::BI_InterlockedXor16_rel: | |||
8128 | case AArch64::BI_InterlockedXor_rel: | |||
8129 | case AArch64::BI_InterlockedXor64_rel: | |||
8130 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_rel, E); | |||
8131 | case AArch64::BI_InterlockedXor8_nf: | |||
8132 | case AArch64::BI_InterlockedXor16_nf: | |||
8133 | case AArch64::BI_InterlockedXor_nf: | |||
8134 | case AArch64::BI_InterlockedXor64_nf: | |||
8135 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_nf, E); | |||
8136 | case AArch64::BI_InterlockedAnd8_acq: | |||
8137 | case AArch64::BI_InterlockedAnd16_acq: | |||
8138 | case AArch64::BI_InterlockedAnd_acq: | |||
8139 | case AArch64::BI_InterlockedAnd64_acq: | |||
8140 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_acq, E); | |||
8141 | case AArch64::BI_InterlockedAnd8_rel: | |||
8142 | case AArch64::BI_InterlockedAnd16_rel: | |||
8143 | case AArch64::BI_InterlockedAnd_rel: | |||
8144 | case AArch64::BI_InterlockedAnd64_rel: | |||
8145 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_rel, E); | |||
8146 | case AArch64::BI_InterlockedAnd8_nf: | |||
8147 | case AArch64::BI_InterlockedAnd16_nf: | |||
8148 | case AArch64::BI_InterlockedAnd_nf: | |||
8149 | case AArch64::BI_InterlockedAnd64_nf: | |||
8150 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_nf, E); | |||
8151 | case AArch64::BI_InterlockedIncrement16_acq: | |||
8152 | case AArch64::BI_InterlockedIncrement_acq: | |||
8153 | case AArch64::BI_InterlockedIncrement64_acq: | |||
8154 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_acq, E); | |||
8155 | case AArch64::BI_InterlockedIncrement16_rel: | |||
8156 | case AArch64::BI_InterlockedIncrement_rel: | |||
8157 | case AArch64::BI_InterlockedIncrement64_rel: | |||
8158 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_rel, E); | |||
8159 | case AArch64::BI_InterlockedIncrement16_nf: | |||
8160 | case AArch64::BI_InterlockedIncrement_nf: | |||
8161 | case AArch64::BI_InterlockedIncrement64_nf: | |||
8162 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_nf, E); | |||
8163 | case AArch64::BI_InterlockedDecrement16_acq: | |||
8164 | case AArch64::BI_InterlockedDecrement_acq: | |||
8165 | case AArch64::BI_InterlockedDecrement64_acq: | |||
8166 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_acq, E); | |||
8167 | case AArch64::BI_InterlockedDecrement16_rel: | |||
8168 | case AArch64::BI_InterlockedDecrement_rel: | |||
8169 | case AArch64::BI_InterlockedDecrement64_rel: | |||
8170 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_rel, E); | |||
8171 | case AArch64::BI_InterlockedDecrement16_nf: | |||
8172 | case AArch64::BI_InterlockedDecrement_nf: | |||
8173 | case AArch64::BI_InterlockedDecrement64_nf: | |||
8174 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_nf, E); | |||
8175 | ||||
8176 | case AArch64::BI_InterlockedAdd: { | |||
8177 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); | |||
8178 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
8179 | AtomicRMWInst *RMWI = Builder.CreateAtomicRMW( | |||
8180 | AtomicRMWInst::Add, Arg0, Arg1, | |||
8181 | llvm::AtomicOrdering::SequentiallyConsistent); | |||
8182 | return Builder.CreateAdd(RMWI, Arg1); | |||
8183 | } | |||
8184 | } | |||
8185 | ||||
8186 | llvm::VectorType *VTy = GetNeonType(this, Type); | |||
8187 | llvm::Type *Ty = VTy; | |||
8188 | if (!Ty) | |||
8189 | return nullptr; | |||
8190 | ||||
8191 | // Not all intrinsics handled by the common case work for AArch64 yet, so only | |||
8192 | // defer to common code if it's been added to our special map. | |||
8193 | Builtin = findNeonIntrinsicInMap(AArch64SIMDIntrinsicMap, BuiltinID, | |||
8194 | AArch64SIMDIntrinsicsProvenSorted); | |||
8195 | ||||
8196 | if (Builtin) | |||
8197 | return EmitCommonNeonBuiltinExpr( | |||
8198 | Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic, | |||
8199 | Builtin->NameHint, Builtin->TypeModifier, E, Ops, | |||
8200 | /*never use addresses*/ Address::invalid(), Address::invalid(), Arch); | |||
8201 | ||||
8202 | if (Value *V = EmitAArch64TblBuiltinExpr(*this, BuiltinID, E, Ops, Arch)) | |||
8203 | return V; | |||
8204 | ||||
8205 | unsigned Int; | |||
8206 | switch (BuiltinID) { | |||
8207 | default: return nullptr; | |||
8208 | case NEON::BI__builtin_neon_vbsl_v: | |||
8209 | case NEON::BI__builtin_neon_vbslq_v: { | |||
8210 | llvm::Type *BitTy = llvm::VectorType::getInteger(VTy); | |||
8211 | Ops[0] = Builder.CreateBitCast(Ops[0], BitTy, "vbsl"); | |||
8212 | Ops[1] = Builder.CreateBitCast(Ops[1], BitTy, "vbsl"); | |||
8213 | Ops[2] = Builder.CreateBitCast(Ops[2], BitTy, "vbsl"); | |||
8214 | ||||
8215 | Ops[1] = Builder.CreateAnd(Ops[0], Ops[1], "vbsl"); | |||
8216 | Ops[2] = Builder.CreateAnd(Builder.CreateNot(Ops[0]), Ops[2], "vbsl"); | |||
8217 | Ops[0] = Builder.CreateOr(Ops[1], Ops[2], "vbsl"); | |||
8218 | return Builder.CreateBitCast(Ops[0], Ty); | |||
8219 | } | |||
8220 | case NEON::BI__builtin_neon_vfma_lane_v: | |||
8221 | case NEON::BI__builtin_neon_vfmaq_lane_v: { // Only used for FP types | |||
8222 | // The ARM builtins (and instructions) have the addend as the first | |||
8223 | // operand, but the 'fma' intrinsics have it last. Swap it around here. | |||
8224 | Value *Addend = Ops[0]; | |||
8225 | Value *Multiplicand = Ops[1]; | |||
8226 | Value *LaneSource = Ops[2]; | |||
8227 | Ops[0] = Multiplicand; | |||
8228 | Ops[1] = LaneSource; | |||
8229 | Ops[2] = Addend; | |||
8230 | ||||
8231 | // Now adjust things to handle the lane access. | |||
8232 | llvm::Type *SourceTy = BuiltinID == NEON::BI__builtin_neon_vfmaq_lane_v ? | |||
8233 | llvm::VectorType::get(VTy->getElementType(), VTy->getNumElements() / 2) : | |||
8234 | VTy; | |||
8235 | llvm::Constant *cst = cast<Constant>(Ops[3]); | |||
8236 | Value *SV = llvm::ConstantVector::getSplat(VTy->getNumElements(), cst); | |||
8237 | Ops[1] = Builder.CreateBitCast(Ops[1], SourceTy); | |||
8238 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV, "lane"); | |||
8239 | ||||
8240 | Ops.pop_back(); | |||
8241 | Int = Intrinsic::fma; | |||
8242 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "fmla"); | |||
8243 | } | |||
8244 | case NEON::BI__builtin_neon_vfma_laneq_v: { | |||
8245 | llvm::VectorType *VTy = cast<llvm::VectorType>(Ty); | |||
8246 | // v1f64 fma should be mapped to Neon scalar f64 fma | |||
8247 | if (VTy && VTy->getElementType() == DoubleTy) { | |||
8248 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); | |||
8249 | Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy); | |||
8250 | llvm::Type *VTy = GetNeonType(this, | |||
8251 | NeonTypeFlags(NeonTypeFlags::Float64, false, true)); | |||
8252 | Ops[2] = Builder.CreateBitCast(Ops[2], VTy); | |||
8253 | Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract"); | |||
8254 | Function *F = CGM.getIntrinsic(Intrinsic::fma, DoubleTy); | |||
8255 | Value *Result = Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]}); | |||
8256 | return Builder.CreateBitCast(Result, Ty); | |||
8257 | } | |||
8258 | Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty); | |||
8259 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
8260 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
8261 | ||||
8262 | llvm::Type *STy = llvm::VectorType::get(VTy->getElementType(), | |||
8263 | VTy->getNumElements() * 2); | |||
8264 | Ops[2] = Builder.CreateBitCast(Ops[2], STy); | |||
8265 | Value* SV = llvm::ConstantVector::getSplat(VTy->getNumElements(), | |||
8266 | cast<ConstantInt>(Ops[3])); | |||
8267 | Ops[2] = Builder.CreateShuffleVector(Ops[2], Ops[2], SV, "lane"); | |||
8268 | ||||
8269 | return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]}); | |||
8270 | } | |||
8271 | case NEON::BI__builtin_neon_vfmaq_laneq_v: { | |||
8272 | Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty); | |||
8273 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
8274 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
8275 | ||||
8276 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
8277 | Ops[2] = EmitNeonSplat(Ops[2], cast<ConstantInt>(Ops[3])); | |||
8278 | return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]}); | |||
8279 | } | |||
8280 | case NEON::BI__builtin_neon_vfmah_lane_f16: | |||
8281 | case NEON::BI__builtin_neon_vfmas_lane_f32: | |||
8282 | case NEON::BI__builtin_neon_vfmah_laneq_f16: | |||
8283 | case NEON::BI__builtin_neon_vfmas_laneq_f32: | |||
8284 | case NEON::BI__builtin_neon_vfmad_lane_f64: | |||
8285 | case NEON::BI__builtin_neon_vfmad_laneq_f64: { | |||
8286 | Ops.push_back(EmitScalarExpr(E->getArg(3))); | |||
8287 | llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext())); | |||
8288 | Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty); | |||
8289 | Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract"); | |||
8290 | return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]}); | |||
8291 | } | |||
8292 | case NEON::BI__builtin_neon_vmull_v: | |||
8293 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
8294 | Int = usgn ? Intrinsic::aarch64_neon_umull : Intrinsic::aarch64_neon_smull; | |||
8295 | if (Type.isPoly()) Int = Intrinsic::aarch64_neon_pmull; | |||
8296 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull"); | |||
8297 | case NEON::BI__builtin_neon_vmax_v: | |||
8298 | case NEON::BI__builtin_neon_vmaxq_v: | |||
8299 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
8300 | Int = usgn ? Intrinsic::aarch64_neon_umax : Intrinsic::aarch64_neon_smax; | |||
8301 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmax; | |||
8302 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmax"); | |||
8303 | case NEON::BI__builtin_neon_vmaxh_f16: { | |||
8304 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
8305 | Int = Intrinsic::aarch64_neon_fmax; | |||
8306 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmax"); | |||
8307 | } | |||
8308 | case NEON::BI__builtin_neon_vmin_v: | |||
8309 | case NEON::BI__builtin_neon_vminq_v: | |||
8310 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
8311 | Int = usgn ? Intrinsic::aarch64_neon_umin : Intrinsic::aarch64_neon_smin; | |||
8312 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmin; | |||
8313 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmin"); | |||
8314 | case NEON::BI__builtin_neon_vminh_f16: { | |||
8315 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
8316 | Int = Intrinsic::aarch64_neon_fmin; | |||
8317 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmin"); | |||
8318 | } | |||
8319 | case NEON::BI__builtin_neon_vabd_v: | |||
8320 | case NEON::BI__builtin_neon_vabdq_v: | |||
8321 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
8322 | Int = usgn ? Intrinsic::aarch64_neon_uabd : Intrinsic::aarch64_neon_sabd; | |||
8323 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fabd; | |||
8324 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vabd"); | |||
8325 | case NEON::BI__builtin_neon_vpadal_v: | |||
8326 | case NEON::BI__builtin_neon_vpadalq_v: { | |||
8327 | unsigned ArgElts = VTy->getNumElements(); | |||
8328 | llvm::IntegerType *EltTy = cast<IntegerType>(VTy->getElementType()); | |||
8329 | unsigned BitWidth = EltTy->getBitWidth(); | |||
8330 | llvm::Type *ArgTy = llvm::VectorType::get( | |||
8331 | llvm::IntegerType::get(getLLVMContext(), BitWidth/2), 2*ArgElts); | |||
8332 | llvm::Type* Tys[2] = { VTy, ArgTy }; | |||
8333 | Int = usgn ? Intrinsic::aarch64_neon_uaddlp : Intrinsic::aarch64_neon_saddlp; | |||
8334 | SmallVector<llvm::Value*, 1> TmpOps; | |||
8335 | TmpOps.push_back(Ops[1]); | |||
8336 | Function *F = CGM.getIntrinsic(Int, Tys); | |||
8337 | llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vpadal"); | |||
8338 | llvm::Value *addend = Builder.CreateBitCast(Ops[0], tmp->getType()); | |||
8339 | return Builder.CreateAdd(tmp, addend); | |||
8340 | } | |||
8341 | case NEON::BI__builtin_neon_vpmin_v: | |||
8342 | case NEON::BI__builtin_neon_vpminq_v: | |||
8343 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
8344 | Int = usgn ? Intrinsic::aarch64_neon_uminp : Intrinsic::aarch64_neon_sminp; | |||
8345 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fminp; | |||
8346 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmin"); | |||
8347 | case NEON::BI__builtin_neon_vpmax_v: | |||
8348 | case NEON::BI__builtin_neon_vpmaxq_v: | |||
8349 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
8350 | Int = usgn ? Intrinsic::aarch64_neon_umaxp : Intrinsic::aarch64_neon_smaxp; | |||
8351 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmaxp; | |||
8352 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmax"); | |||
8353 | case NEON::BI__builtin_neon_vminnm_v: | |||
8354 | case NEON::BI__builtin_neon_vminnmq_v: | |||
8355 | Int = Intrinsic::aarch64_neon_fminnm; | |||
8356 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vminnm"); | |||
8357 | case NEON::BI__builtin_neon_vminnmh_f16: | |||
8358 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
8359 | Int = Intrinsic::aarch64_neon_fminnm; | |||
8360 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vminnm"); | |||
8361 | case NEON::BI__builtin_neon_vmaxnm_v: | |||
8362 | case NEON::BI__builtin_neon_vmaxnmq_v: | |||
8363 | Int = Intrinsic::aarch64_neon_fmaxnm; | |||
8364 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmaxnm"); | |||
8365 | case NEON::BI__builtin_neon_vmaxnmh_f16: | |||
8366 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
8367 | Int = Intrinsic::aarch64_neon_fmaxnm; | |||
8368 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmaxnm"); | |||
8369 | case NEON::BI__builtin_neon_vrecpss_f32: { | |||
8370 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
8371 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, FloatTy), | |||
8372 | Ops, "vrecps"); | |||
8373 | } | |||
8374 | case NEON::BI__builtin_neon_vrecpsd_f64: | |||
8375 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
8376 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, DoubleTy), | |||
8377 | Ops, "vrecps"); | |||
8378 | case NEON::BI__builtin_neon_vrecpsh_f16: | |||
8379 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
8380 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, HalfTy), | |||
8381 | Ops, "vrecps"); | |||
8382 | case NEON::BI__builtin_neon_vqshrun_n_v: | |||
8383 | Int = Intrinsic::aarch64_neon_sqshrun; | |||
8384 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrun_n"); | |||
8385 | case NEON::BI__builtin_neon_vqrshrun_n_v: | |||
8386 | Int = Intrinsic::aarch64_neon_sqrshrun; | |||
8387 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrun_n"); | |||
8388 | case NEON::BI__builtin_neon_vqshrn_n_v: | |||
8389 | Int = usgn ? Intrinsic::aarch64_neon_uqshrn : Intrinsic::aarch64_neon_sqshrn; | |||
8390 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n"); | |||
8391 | case NEON::BI__builtin_neon_vrshrn_n_v: | |||
8392 | Int = Intrinsic::aarch64_neon_rshrn; | |||
8393 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshrn_n"); | |||
8394 | case NEON::BI__builtin_neon_vqrshrn_n_v: | |||
8395 | Int = usgn ? Intrinsic::aarch64_neon_uqrshrn : Intrinsic::aarch64_neon_sqrshrn; | |||
8396 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n"); | |||
8397 | case NEON::BI__builtin_neon_vrndah_f16: { | |||
8398 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8399 | Int = Intrinsic::round; | |||
8400 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrnda"); | |||
8401 | } | |||
8402 | case NEON::BI__builtin_neon_vrnda_v: | |||
8403 | case NEON::BI__builtin_neon_vrndaq_v: { | |||
8404 | Int = Intrinsic::round; | |||
8405 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnda"); | |||
8406 | } | |||
8407 | case NEON::BI__builtin_neon_vrndih_f16: { | |||
8408 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8409 | Int = Intrinsic::nearbyint; | |||
8410 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndi"); | |||
8411 | } | |||
8412 | case NEON::BI__builtin_neon_vrndmh_f16: { | |||
8413 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8414 | Int = Intrinsic::floor; | |||
8415 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndm"); | |||
8416 | } | |||
8417 | case NEON::BI__builtin_neon_vrndm_v: | |||
8418 | case NEON::BI__builtin_neon_vrndmq_v: { | |||
8419 | Int = Intrinsic::floor; | |||
8420 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndm"); | |||
8421 | } | |||
8422 | case NEON::BI__builtin_neon_vrndnh_f16: { | |||
8423 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8424 | Int = Intrinsic::aarch64_neon_frintn; | |||
8425 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndn"); | |||
8426 | } | |||
8427 | case NEON::BI__builtin_neon_vrndn_v: | |||
8428 | case NEON::BI__builtin_neon_vrndnq_v: { | |||
8429 | Int = Intrinsic::aarch64_neon_frintn; | |||
8430 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndn"); | |||
8431 | } | |||
8432 | case NEON::BI__builtin_neon_vrndns_f32: { | |||
8433 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8434 | Int = Intrinsic::aarch64_neon_frintn; | |||
8435 | return EmitNeonCall(CGM.getIntrinsic(Int, FloatTy), Ops, "vrndn"); | |||
8436 | } | |||
8437 | case NEON::BI__builtin_neon_vrndph_f16: { | |||
8438 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8439 | Int = Intrinsic::ceil; | |||
8440 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndp"); | |||
8441 | } | |||
8442 | case NEON::BI__builtin_neon_vrndp_v: | |||
8443 | case NEON::BI__builtin_neon_vrndpq_v: { | |||
8444 | Int = Intrinsic::ceil; | |||
8445 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndp"); | |||
8446 | } | |||
8447 | case NEON::BI__builtin_neon_vrndxh_f16: { | |||
8448 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8449 | Int = Intrinsic::rint; | |||
8450 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndx"); | |||
8451 | } | |||
8452 | case NEON::BI__builtin_neon_vrndx_v: | |||
8453 | case NEON::BI__builtin_neon_vrndxq_v: { | |||
8454 | Int = Intrinsic::rint; | |||
8455 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndx"); | |||
8456 | } | |||
8457 | case NEON::BI__builtin_neon_vrndh_f16: { | |||
8458 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8459 | Int = Intrinsic::trunc; | |||
8460 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndz"); | |||
8461 | } | |||
8462 | case NEON::BI__builtin_neon_vrnd_v: | |||
8463 | case NEON::BI__builtin_neon_vrndq_v: { | |||
8464 | Int = Intrinsic::trunc; | |||
8465 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndz"); | |||
8466 | } | |||
8467 | case NEON::BI__builtin_neon_vcvt_f64_v: | |||
8468 | case NEON::BI__builtin_neon_vcvtq_f64_v: | |||
8469 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
8470 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, quad)); | |||
8471 | return usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") | |||
8472 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); | |||
8473 | case NEON::BI__builtin_neon_vcvt_f64_f32: { | |||
8474 | assert(Type.getEltType() == NeonTypeFlags::Float64 && quad &&((Type.getEltType() == NeonTypeFlags::Float64 && quad && "unexpected vcvt_f64_f32 builtin") ? static_cast< void> (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float64 && quad && \"unexpected vcvt_f64_f32 builtin\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 8475, __PRETTY_FUNCTION__)) | |||
8475 | "unexpected vcvt_f64_f32 builtin")((Type.getEltType() == NeonTypeFlags::Float64 && quad && "unexpected vcvt_f64_f32 builtin") ? static_cast< void> (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float64 && quad && \"unexpected vcvt_f64_f32 builtin\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 8475, __PRETTY_FUNCTION__)); | |||
8476 | NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float32, false, false); | |||
8477 | Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag)); | |||
8478 | ||||
8479 | return Builder.CreateFPExt(Ops[0], Ty, "vcvt"); | |||
8480 | } | |||
8481 | case NEON::BI__builtin_neon_vcvt_f32_f64: { | |||
8482 | assert(Type.getEltType() == NeonTypeFlags::Float32 &&((Type.getEltType() == NeonTypeFlags::Float32 && "unexpected vcvt_f32_f64 builtin" ) ? static_cast<void> (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float32 && \"unexpected vcvt_f32_f64 builtin\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 8483, __PRETTY_FUNCTION__)) | |||
8483 | "unexpected vcvt_f32_f64 builtin")((Type.getEltType() == NeonTypeFlags::Float32 && "unexpected vcvt_f32_f64 builtin" ) ? static_cast<void> (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float32 && \"unexpected vcvt_f32_f64 builtin\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 8483, __PRETTY_FUNCTION__)); | |||
8484 | NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float64, false, true); | |||
8485 | Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag)); | |||
8486 | ||||
8487 | return Builder.CreateFPTrunc(Ops[0], Ty, "vcvt"); | |||
8488 | } | |||
8489 | case NEON::BI__builtin_neon_vcvt_s32_v: | |||
8490 | case NEON::BI__builtin_neon_vcvt_u32_v: | |||
8491 | case NEON::BI__builtin_neon_vcvt_s64_v: | |||
8492 | case NEON::BI__builtin_neon_vcvt_u64_v: | |||
8493 | case NEON::BI__builtin_neon_vcvt_s16_v: | |||
8494 | case NEON::BI__builtin_neon_vcvt_u16_v: | |||
8495 | case NEON::BI__builtin_neon_vcvtq_s32_v: | |||
8496 | case NEON::BI__builtin_neon_vcvtq_u32_v: | |||
8497 | case NEON::BI__builtin_neon_vcvtq_s64_v: | |||
8498 | case NEON::BI__builtin_neon_vcvtq_u64_v: | |||
8499 | case NEON::BI__builtin_neon_vcvtq_s16_v: | |||
8500 | case NEON::BI__builtin_neon_vcvtq_u16_v: { | |||
8501 | Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type)); | |||
8502 | if (usgn) | |||
8503 | return Builder.CreateFPToUI(Ops[0], Ty); | |||
8504 | return Builder.CreateFPToSI(Ops[0], Ty); | |||
8505 | } | |||
8506 | case NEON::BI__builtin_neon_vcvta_s16_v: | |||
8507 | case NEON::BI__builtin_neon_vcvta_u16_v: | |||
8508 | case NEON::BI__builtin_neon_vcvta_s32_v: | |||
8509 | case NEON::BI__builtin_neon_vcvtaq_s16_v: | |||
8510 | case NEON::BI__builtin_neon_vcvtaq_s32_v: | |||
8511 | case NEON::BI__builtin_neon_vcvta_u32_v: | |||
8512 | case NEON::BI__builtin_neon_vcvtaq_u16_v: | |||
8513 | case NEON::BI__builtin_neon_vcvtaq_u32_v: | |||
8514 | case NEON::BI__builtin_neon_vcvta_s64_v: | |||
8515 | case NEON::BI__builtin_neon_vcvtaq_s64_v: | |||
8516 | case NEON::BI__builtin_neon_vcvta_u64_v: | |||
8517 | case NEON::BI__builtin_neon_vcvtaq_u64_v: { | |||
8518 | Int = usgn ? Intrinsic::aarch64_neon_fcvtau : Intrinsic::aarch64_neon_fcvtas; | |||
8519 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
8520 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvta"); | |||
8521 | } | |||
8522 | case NEON::BI__builtin_neon_vcvtm_s16_v: | |||
8523 | case NEON::BI__builtin_neon_vcvtm_s32_v: | |||
8524 | case NEON::BI__builtin_neon_vcvtmq_s16_v: | |||
8525 | case NEON::BI__builtin_neon_vcvtmq_s32_v: | |||
8526 | case NEON::BI__builtin_neon_vcvtm_u16_v: | |||
8527 | case NEON::BI__builtin_neon_vcvtm_u32_v: | |||
8528 | case NEON::BI__builtin_neon_vcvtmq_u16_v: | |||
8529 | case NEON::BI__builtin_neon_vcvtmq_u32_v: | |||
8530 | case NEON::BI__builtin_neon_vcvtm_s64_v: | |||
8531 | case NEON::BI__builtin_neon_vcvtmq_s64_v: | |||
8532 | case NEON::BI__builtin_neon_vcvtm_u64_v: | |||
8533 | case NEON::BI__builtin_neon_vcvtmq_u64_v: { | |||
8534 | Int = usgn ? Intrinsic::aarch64_neon_fcvtmu : Intrinsic::aarch64_neon_fcvtms; | |||
8535 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
8536 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtm"); | |||
8537 | } | |||
8538 | case NEON::BI__builtin_neon_vcvtn_s16_v: | |||
8539 | case NEON::BI__builtin_neon_vcvtn_s32_v: | |||
8540 | case NEON::BI__builtin_neon_vcvtnq_s16_v: | |||
8541 | case NEON::BI__builtin_neon_vcvtnq_s32_v: | |||
8542 | case NEON::BI__builtin_neon_vcvtn_u16_v: | |||
8543 | case NEON::BI__builtin_neon_vcvtn_u32_v: | |||
8544 | case NEON::BI__builtin_neon_vcvtnq_u16_v: | |||
8545 | case NEON::BI__builtin_neon_vcvtnq_u32_v: | |||
8546 | case NEON::BI__builtin_neon_vcvtn_s64_v: | |||
8547 | case NEON::BI__builtin_neon_vcvtnq_s64_v: | |||
8548 | case NEON::BI__builtin_neon_vcvtn_u64_v: | |||
8549 | case NEON::BI__builtin_neon_vcvtnq_u64_v: { | |||
8550 | Int = usgn ? Intrinsic::aarch64_neon_fcvtnu : Intrinsic::aarch64_neon_fcvtns; | |||
8551 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
8552 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtn"); | |||
8553 | } | |||
8554 | case NEON::BI__builtin_neon_vcvtp_s16_v: | |||
8555 | case NEON::BI__builtin_neon_vcvtp_s32_v: | |||
8556 | case NEON::BI__builtin_neon_vcvtpq_s16_v: | |||
8557 | case NEON::BI__builtin_neon_vcvtpq_s32_v: | |||
8558 | case NEON::BI__builtin_neon_vcvtp_u16_v: | |||
8559 | case NEON::BI__builtin_neon_vcvtp_u32_v: | |||
8560 | case NEON::BI__builtin_neon_vcvtpq_u16_v: | |||
8561 | case NEON::BI__builtin_neon_vcvtpq_u32_v: | |||
8562 | case NEON::BI__builtin_neon_vcvtp_s64_v: | |||
8563 | case NEON::BI__builtin_neon_vcvtpq_s64_v: | |||
8564 | case NEON::BI__builtin_neon_vcvtp_u64_v: | |||
8565 | case NEON::BI__builtin_neon_vcvtpq_u64_v: { | |||
8566 | Int = usgn ? Intrinsic::aarch64_neon_fcvtpu : Intrinsic::aarch64_neon_fcvtps; | |||
8567 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
8568 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtp"); | |||
8569 | } | |||
8570 | case NEON::BI__builtin_neon_vmulx_v: | |||
8571 | case NEON::BI__builtin_neon_vmulxq_v: { | |||
8572 | Int = Intrinsic::aarch64_neon_fmulx; | |||
8573 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmulx"); | |||
8574 | } | |||
8575 | case NEON::BI__builtin_neon_vmulxh_lane_f16: | |||
8576 | case NEON::BI__builtin_neon_vmulxh_laneq_f16: { | |||
8577 | // vmulx_lane should be mapped to Neon scalar mulx after | |||
8578 | // extracting the scalar element | |||
8579 | Ops.push_back(EmitScalarExpr(E->getArg(2))); | |||
8580 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract"); | |||
8581 | Ops.pop_back(); | |||
8582 | Int = Intrinsic::aarch64_neon_fmulx; | |||
8583 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmulx"); | |||
8584 | } | |||
8585 | case NEON::BI__builtin_neon_vmul_lane_v: | |||
8586 | case NEON::BI__builtin_neon_vmul_laneq_v: { | |||
8587 | // v1f64 vmul_lane should be mapped to Neon scalar mul lane | |||
8588 | bool Quad = false; | |||
8589 | if (BuiltinID == NEON::BI__builtin_neon_vmul_laneq_v) | |||
8590 | Quad = true; | |||
8591 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); | |||
8592 | llvm::Type *VTy = GetNeonType(this, | |||
8593 | NeonTypeFlags(NeonTypeFlags::Float64, false, Quad)); | |||
8594 | Ops[1] = Builder.CreateBitCast(Ops[1], VTy); | |||
8595 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract"); | |||
8596 | Value *Result = Builder.CreateFMul(Ops[0], Ops[1]); | |||
8597 | return Builder.CreateBitCast(Result, Ty); | |||
8598 | } | |||
8599 | case NEON::BI__builtin_neon_vnegd_s64: | |||
8600 | return Builder.CreateNeg(EmitScalarExpr(E->getArg(0)), "vnegd"); | |||
8601 | case NEON::BI__builtin_neon_vnegh_f16: | |||
8602 | return Builder.CreateFNeg(EmitScalarExpr(E->getArg(0)), "vnegh"); | |||
8603 | case NEON::BI__builtin_neon_vpmaxnm_v: | |||
8604 | case NEON::BI__builtin_neon_vpmaxnmq_v: { | |||
8605 | Int = Intrinsic::aarch64_neon_fmaxnmp; | |||
8606 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmaxnm"); | |||
8607 | } | |||
8608 | case NEON::BI__builtin_neon_vpminnm_v: | |||
8609 | case NEON::BI__builtin_neon_vpminnmq_v: { | |||
8610 | Int = Intrinsic::aarch64_neon_fminnmp; | |||
8611 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpminnm"); | |||
8612 | } | |||
8613 | case NEON::BI__builtin_neon_vsqrth_f16: { | |||
8614 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8615 | Int = Intrinsic::sqrt; | |||
8616 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vsqrt"); | |||
8617 | } | |||
8618 | case NEON::BI__builtin_neon_vsqrt_v: | |||
8619 | case NEON::BI__builtin_neon_vsqrtq_v: { | |||
8620 | Int = Intrinsic::sqrt; | |||
8621 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
8622 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqrt"); | |||
8623 | } | |||
8624 | case NEON::BI__builtin_neon_vrbit_v: | |||
8625 | case NEON::BI__builtin_neon_vrbitq_v: { | |||
8626 | Int = Intrinsic::aarch64_neon_rbit; | |||
8627 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrbit"); | |||
8628 | } | |||
8629 | case NEON::BI__builtin_neon_vaddv_u8: | |||
8630 | // FIXME: These are handled by the AArch64 scalar code. | |||
8631 | usgn = true; | |||
8632 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
8633 | case NEON::BI__builtin_neon_vaddv_s8: { | |||
8634 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; | |||
8635 | Ty = Int32Ty; | |||
8636 | VTy = llvm::VectorType::get(Int8Ty, 8); | |||
8637 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8638 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8639 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); | |||
8640 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8641 | } | |||
8642 | case NEON::BI__builtin_neon_vaddv_u16: | |||
8643 | usgn = true; | |||
8644 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
8645 | case NEON::BI__builtin_neon_vaddv_s16: { | |||
8646 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; | |||
8647 | Ty = Int32Ty; | |||
8648 | VTy = llvm::VectorType::get(Int16Ty, 4); | |||
8649 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8650 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8651 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); | |||
8652 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8653 | } | |||
8654 | case NEON::BI__builtin_neon_vaddvq_u8: | |||
8655 | usgn = true; | |||
8656 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
8657 | case NEON::BI__builtin_neon_vaddvq_s8: { | |||
8658 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; | |||
8659 | Ty = Int32Ty; | |||
8660 | VTy = llvm::VectorType::get(Int8Ty, 16); | |||
8661 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8662 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8663 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); | |||
8664 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8665 | } | |||
8666 | case NEON::BI__builtin_neon_vaddvq_u16: | |||
8667 | usgn = true; | |||
8668 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
8669 | case NEON::BI__builtin_neon_vaddvq_s16: { | |||
8670 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; | |||
8671 | Ty = Int32Ty; | |||
8672 | VTy = llvm::VectorType::get(Int16Ty, 8); | |||
8673 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8674 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8675 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); | |||
8676 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8677 | } | |||
8678 | case NEON::BI__builtin_neon_vmaxv_u8: { | |||
8679 | Int = Intrinsic::aarch64_neon_umaxv; | |||
8680 | Ty = Int32Ty; | |||
8681 | VTy = llvm::VectorType::get(Int8Ty, 8); | |||
8682 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8683 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8684 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8685 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8686 | } | |||
8687 | case NEON::BI__builtin_neon_vmaxv_u16: { | |||
8688 | Int = Intrinsic::aarch64_neon_umaxv; | |||
8689 | Ty = Int32Ty; | |||
8690 | VTy = llvm::VectorType::get(Int16Ty, 4); | |||
8691 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8692 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8693 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8694 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8695 | } | |||
8696 | case NEON::BI__builtin_neon_vmaxvq_u8: { | |||
8697 | Int = Intrinsic::aarch64_neon_umaxv; | |||
8698 | Ty = Int32Ty; | |||
8699 | VTy = llvm::VectorType::get(Int8Ty, 16); | |||
8700 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8701 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8702 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8703 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8704 | } | |||
8705 | case NEON::BI__builtin_neon_vmaxvq_u16: { | |||
8706 | Int = Intrinsic::aarch64_neon_umaxv; | |||
8707 | Ty = Int32Ty; | |||
8708 | VTy = llvm::VectorType::get(Int16Ty, 8); | |||
8709 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8710 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8711 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8712 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8713 | } | |||
8714 | case NEON::BI__builtin_neon_vmaxv_s8: { | |||
8715 | Int = Intrinsic::aarch64_neon_smaxv; | |||
8716 | Ty = Int32Ty; | |||
8717 | VTy = llvm::VectorType::get(Int8Ty, 8); | |||
8718 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8719 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8720 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8721 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8722 | } | |||
8723 | case NEON::BI__builtin_neon_vmaxv_s16: { | |||
8724 | Int = Intrinsic::aarch64_neon_smaxv; | |||
8725 | Ty = Int32Ty; | |||
8726 | VTy = llvm::VectorType::get(Int16Ty, 4); | |||
8727 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8728 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8729 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8730 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8731 | } | |||
8732 | case NEON::BI__builtin_neon_vmaxvq_s8: { | |||
8733 | Int = Intrinsic::aarch64_neon_smaxv; | |||
8734 | Ty = Int32Ty; | |||
8735 | VTy = llvm::VectorType::get(Int8Ty, 16); | |||
8736 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8737 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8738 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8739 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8740 | } | |||
8741 | case NEON::BI__builtin_neon_vmaxvq_s16: { | |||
8742 | Int = Intrinsic::aarch64_neon_smaxv; | |||
8743 | Ty = Int32Ty; | |||
8744 | VTy = llvm::VectorType::get(Int16Ty, 8); | |||
8745 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8746 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8747 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8748 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8749 | } | |||
8750 | case NEON::BI__builtin_neon_vmaxv_f16: { | |||
8751 | Int = Intrinsic::aarch64_neon_fmaxv; | |||
8752 | Ty = HalfTy; | |||
8753 | VTy = llvm::VectorType::get(HalfTy, 4); | |||
8754 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8755 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8756 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8757 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
8758 | } | |||
8759 | case NEON::BI__builtin_neon_vmaxvq_f16: { | |||
8760 | Int = Intrinsic::aarch64_neon_fmaxv; | |||
8761 | Ty = HalfTy; | |||
8762 | VTy = llvm::VectorType::get(HalfTy, 8); | |||
8763 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8764 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8765 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
8766 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
8767 | } | |||
8768 | case NEON::BI__builtin_neon_vminv_u8: { | |||
8769 | Int = Intrinsic::aarch64_neon_uminv; | |||
8770 | Ty = Int32Ty; | |||
8771 | VTy = llvm::VectorType::get(Int8Ty, 8); | |||
8772 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8773 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8774 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8775 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8776 | } | |||
8777 | case NEON::BI__builtin_neon_vminv_u16: { | |||
8778 | Int = Intrinsic::aarch64_neon_uminv; | |||
8779 | Ty = Int32Ty; | |||
8780 | VTy = llvm::VectorType::get(Int16Ty, 4); | |||
8781 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8782 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8783 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8784 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8785 | } | |||
8786 | case NEON::BI__builtin_neon_vminvq_u8: { | |||
8787 | Int = Intrinsic::aarch64_neon_uminv; | |||
8788 | Ty = Int32Ty; | |||
8789 | VTy = llvm::VectorType::get(Int8Ty, 16); | |||
8790 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8791 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8792 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8793 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8794 | } | |||
8795 | case NEON::BI__builtin_neon_vminvq_u16: { | |||
8796 | Int = Intrinsic::aarch64_neon_uminv; | |||
8797 | Ty = Int32Ty; | |||
8798 | VTy = llvm::VectorType::get(Int16Ty, 8); | |||
8799 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8800 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8801 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8802 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8803 | } | |||
8804 | case NEON::BI__builtin_neon_vminv_s8: { | |||
8805 | Int = Intrinsic::aarch64_neon_sminv; | |||
8806 | Ty = Int32Ty; | |||
8807 | VTy = llvm::VectorType::get(Int8Ty, 8); | |||
8808 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8809 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8810 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8811 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8812 | } | |||
8813 | case NEON::BI__builtin_neon_vminv_s16: { | |||
8814 | Int = Intrinsic::aarch64_neon_sminv; | |||
8815 | Ty = Int32Ty; | |||
8816 | VTy = llvm::VectorType::get(Int16Ty, 4); | |||
8817 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8818 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8819 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8820 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8821 | } | |||
8822 | case NEON::BI__builtin_neon_vminvq_s8: { | |||
8823 | Int = Intrinsic::aarch64_neon_sminv; | |||
8824 | Ty = Int32Ty; | |||
8825 | VTy = llvm::VectorType::get(Int8Ty, 16); | |||
8826 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8827 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8828 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8829 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
8830 | } | |||
8831 | case NEON::BI__builtin_neon_vminvq_s16: { | |||
8832 | Int = Intrinsic::aarch64_neon_sminv; | |||
8833 | Ty = Int32Ty; | |||
8834 | VTy = llvm::VectorType::get(Int16Ty, 8); | |||
8835 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8836 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8837 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8838 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8839 | } | |||
8840 | case NEON::BI__builtin_neon_vminv_f16: { | |||
8841 | Int = Intrinsic::aarch64_neon_fminv; | |||
8842 | Ty = HalfTy; | |||
8843 | VTy = llvm::VectorType::get(HalfTy, 4); | |||
8844 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8845 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8846 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8847 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
8848 | } | |||
8849 | case NEON::BI__builtin_neon_vminvq_f16: { | |||
8850 | Int = Intrinsic::aarch64_neon_fminv; | |||
8851 | Ty = HalfTy; | |||
8852 | VTy = llvm::VectorType::get(HalfTy, 8); | |||
8853 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8854 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8855 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
8856 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
8857 | } | |||
8858 | case NEON::BI__builtin_neon_vmaxnmv_f16: { | |||
8859 | Int = Intrinsic::aarch64_neon_fmaxnmv; | |||
8860 | Ty = HalfTy; | |||
8861 | VTy = llvm::VectorType::get(HalfTy, 4); | |||
8862 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8863 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8864 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv"); | |||
8865 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
8866 | } | |||
8867 | case NEON::BI__builtin_neon_vmaxnmvq_f16: { | |||
8868 | Int = Intrinsic::aarch64_neon_fmaxnmv; | |||
8869 | Ty = HalfTy; | |||
8870 | VTy = llvm::VectorType::get(HalfTy, 8); | |||
8871 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8872 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8873 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv"); | |||
8874 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
8875 | } | |||
8876 | case NEON::BI__builtin_neon_vminnmv_f16: { | |||
8877 | Int = Intrinsic::aarch64_neon_fminnmv; | |||
8878 | Ty = HalfTy; | |||
8879 | VTy = llvm::VectorType::get(HalfTy, 4); | |||
8880 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8881 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8882 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv"); | |||
8883 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
8884 | } | |||
8885 | case NEON::BI__builtin_neon_vminnmvq_f16: { | |||
8886 | Int = Intrinsic::aarch64_neon_fminnmv; | |||
8887 | Ty = HalfTy; | |||
8888 | VTy = llvm::VectorType::get(HalfTy, 8); | |||
8889 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8890 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8891 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv"); | |||
8892 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
8893 | } | |||
8894 | case NEON::BI__builtin_neon_vmul_n_f64: { | |||
8895 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); | |||
8896 | Value *RHS = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), DoubleTy); | |||
8897 | return Builder.CreateFMul(Ops[0], RHS); | |||
8898 | } | |||
8899 | case NEON::BI__builtin_neon_vaddlv_u8: { | |||
8900 | Int = Intrinsic::aarch64_neon_uaddlv; | |||
8901 | Ty = Int32Ty; | |||
8902 | VTy = llvm::VectorType::get(Int8Ty, 8); | |||
8903 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8904 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8905 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
8906 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8907 | } | |||
8908 | case NEON::BI__builtin_neon_vaddlv_u16: { | |||
8909 | Int = Intrinsic::aarch64_neon_uaddlv; | |||
8910 | Ty = Int32Ty; | |||
8911 | VTy = llvm::VectorType::get(Int16Ty, 4); | |||
8912 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8913 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8914 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
8915 | } | |||
8916 | case NEON::BI__builtin_neon_vaddlvq_u8: { | |||
8917 | Int = Intrinsic::aarch64_neon_uaddlv; | |||
8918 | Ty = Int32Ty; | |||
8919 | VTy = llvm::VectorType::get(Int8Ty, 16); | |||
8920 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8921 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8922 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
8923 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8924 | } | |||
8925 | case NEON::BI__builtin_neon_vaddlvq_u16: { | |||
8926 | Int = Intrinsic::aarch64_neon_uaddlv; | |||
8927 | Ty = Int32Ty; | |||
8928 | VTy = llvm::VectorType::get(Int16Ty, 8); | |||
8929 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8930 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8931 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
8932 | } | |||
8933 | case NEON::BI__builtin_neon_vaddlv_s8: { | |||
8934 | Int = Intrinsic::aarch64_neon_saddlv; | |||
8935 | Ty = Int32Ty; | |||
8936 | VTy = llvm::VectorType::get(Int8Ty, 8); | |||
8937 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8938 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8939 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
8940 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8941 | } | |||
8942 | case NEON::BI__builtin_neon_vaddlv_s16: { | |||
8943 | Int = Intrinsic::aarch64_neon_saddlv; | |||
8944 | Ty = Int32Ty; | |||
8945 | VTy = llvm::VectorType::get(Int16Ty, 4); | |||
8946 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8947 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8948 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
8949 | } | |||
8950 | case NEON::BI__builtin_neon_vaddlvq_s8: { | |||
8951 | Int = Intrinsic::aarch64_neon_saddlv; | |||
8952 | Ty = Int32Ty; | |||
8953 | VTy = llvm::VectorType::get(Int8Ty, 16); | |||
8954 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8955 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8956 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
8957 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
8958 | } | |||
8959 | case NEON::BI__builtin_neon_vaddlvq_s16: { | |||
8960 | Int = Intrinsic::aarch64_neon_saddlv; | |||
8961 | Ty = Int32Ty; | |||
8962 | VTy = llvm::VectorType::get(Int16Ty, 8); | |||
8963 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
8964 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
8965 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
8966 | } | |||
8967 | case NEON::BI__builtin_neon_vsri_n_v: | |||
8968 | case NEON::BI__builtin_neon_vsriq_n_v: { | |||
8969 | Int = Intrinsic::aarch64_neon_vsri; | |||
8970 | llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty); | |||
8971 | return EmitNeonCall(Intrin, Ops, "vsri_n"); | |||
8972 | } | |||
8973 | case NEON::BI__builtin_neon_vsli_n_v: | |||
8974 | case NEON::BI__builtin_neon_vsliq_n_v: { | |||
8975 | Int = Intrinsic::aarch64_neon_vsli; | |||
8976 | llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty); | |||
8977 | return EmitNeonCall(Intrin, Ops, "vsli_n"); | |||
8978 | } | |||
8979 | case NEON::BI__builtin_neon_vsra_n_v: | |||
8980 | case NEON::BI__builtin_neon_vsraq_n_v: | |||
8981 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
8982 | Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n"); | |||
8983 | return Builder.CreateAdd(Ops[0], Ops[1]); | |||
8984 | case NEON::BI__builtin_neon_vrsra_n_v: | |||
8985 | case NEON::BI__builtin_neon_vrsraq_n_v: { | |||
8986 | Int = usgn ? Intrinsic::aarch64_neon_urshl : Intrinsic::aarch64_neon_srshl; | |||
8987 | SmallVector<llvm::Value*,2> TmpOps; | |||
8988 | TmpOps.push_back(Ops[1]); | |||
8989 | TmpOps.push_back(Ops[2]); | |||
8990 | Function* F = CGM.getIntrinsic(Int, Ty); | |||
8991 | llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vrshr_n", 1, true); | |||
8992 | Ops[0] = Builder.CreateBitCast(Ops[0], VTy); | |||
8993 | return Builder.CreateAdd(Ops[0], tmp); | |||
8994 | } | |||
8995 | case NEON::BI__builtin_neon_vld1_v: | |||
8996 | case NEON::BI__builtin_neon_vld1q_v: { | |||
8997 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy)); | |||
8998 | auto Alignment = CharUnits::fromQuantity( | |||
8999 | BuiltinID == NEON::BI__builtin_neon_vld1_v ? 8 : 16); | |||
9000 | return Builder.CreateAlignedLoad(VTy, Ops[0], Alignment); | |||
9001 | } | |||
9002 | case NEON::BI__builtin_neon_vst1_v: | |||
9003 | case NEON::BI__builtin_neon_vst1q_v: | |||
9004 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy)); | |||
9005 | Ops[1] = Builder.CreateBitCast(Ops[1], VTy); | |||
9006 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9007 | case NEON::BI__builtin_neon_vld1_lane_v: | |||
9008 | case NEON::BI__builtin_neon_vld1q_lane_v: { | |||
9009 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
9010 | Ty = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
9011 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
9012 | auto Alignment = CharUnits::fromQuantity( | |||
9013 | BuiltinID == NEON::BI__builtin_neon_vld1_lane_v ? 8 : 16); | |||
9014 | Ops[0] = | |||
9015 | Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment); | |||
9016 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vld1_lane"); | |||
9017 | } | |||
9018 | case NEON::BI__builtin_neon_vld1_dup_v: | |||
9019 | case NEON::BI__builtin_neon_vld1q_dup_v: { | |||
9020 | Value *V = UndefValue::get(Ty); | |||
9021 | Ty = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
9022 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
9023 | auto Alignment = CharUnits::fromQuantity( | |||
9024 | BuiltinID == NEON::BI__builtin_neon_vld1_dup_v ? 8 : 16); | |||
9025 | Ops[0] = | |||
9026 | Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment); | |||
9027 | llvm::Constant *CI = ConstantInt::get(Int32Ty, 0); | |||
9028 | Ops[0] = Builder.CreateInsertElement(V, Ops[0], CI); | |||
9029 | return EmitNeonSplat(Ops[0], CI); | |||
9030 | } | |||
9031 | case NEON::BI__builtin_neon_vst1_lane_v: | |||
9032 | case NEON::BI__builtin_neon_vst1q_lane_v: | |||
9033 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
9034 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]); | |||
9035 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
9036 | return Builder.CreateDefaultAlignedStore(Ops[1], | |||
9037 | Builder.CreateBitCast(Ops[0], Ty)); | |||
9038 | case NEON::BI__builtin_neon_vld2_v: | |||
9039 | case NEON::BI__builtin_neon_vld2q_v: { | |||
9040 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); | |||
9041 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
9042 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
9043 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2, Tys); | |||
9044 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld2"); | |||
9045 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
9046 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
9047 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9048 | } | |||
9049 | case NEON::BI__builtin_neon_vld3_v: | |||
9050 | case NEON::BI__builtin_neon_vld3q_v: { | |||
9051 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); | |||
9052 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
9053 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
9054 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3, Tys); | |||
9055 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld3"); | |||
9056 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
9057 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
9058 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9059 | } | |||
9060 | case NEON::BI__builtin_neon_vld4_v: | |||
9061 | case NEON::BI__builtin_neon_vld4q_v: { | |||
9062 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); | |||
9063 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
9064 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
9065 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4, Tys); | |||
9066 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld4"); | |||
9067 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
9068 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
9069 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9070 | } | |||
9071 | case NEON::BI__builtin_neon_vld2_dup_v: | |||
9072 | case NEON::BI__builtin_neon_vld2q_dup_v: { | |||
9073 | llvm::Type *PTy = | |||
9074 | llvm::PointerType::getUnqual(VTy->getElementType()); | |||
9075 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
9076 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
9077 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2r, Tys); | |||
9078 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld2"); | |||
9079 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
9080 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
9081 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9082 | } | |||
9083 | case NEON::BI__builtin_neon_vld3_dup_v: | |||
9084 | case NEON::BI__builtin_neon_vld3q_dup_v: { | |||
9085 | llvm::Type *PTy = | |||
9086 | llvm::PointerType::getUnqual(VTy->getElementType()); | |||
9087 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
9088 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
9089 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3r, Tys); | |||
9090 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld3"); | |||
9091 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
9092 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
9093 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9094 | } | |||
9095 | case NEON::BI__builtin_neon_vld4_dup_v: | |||
9096 | case NEON::BI__builtin_neon_vld4q_dup_v: { | |||
9097 | llvm::Type *PTy = | |||
9098 | llvm::PointerType::getUnqual(VTy->getElementType()); | |||
9099 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
9100 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
9101 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4r, Tys); | |||
9102 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld4"); | |||
9103 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
9104 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
9105 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9106 | } | |||
9107 | case NEON::BI__builtin_neon_vld2_lane_v: | |||
9108 | case NEON::BI__builtin_neon_vld2q_lane_v: { | |||
9109 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; | |||
9110 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2lane, Tys); | |||
9111 | Ops.push_back(Ops[1]); | |||
9112 | Ops.erase(Ops.begin()+1); | |||
9113 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
9114 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
9115 | Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty); | |||
9116 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld2_lane"); | |||
9117 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
9118 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
9119 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9120 | } | |||
9121 | case NEON::BI__builtin_neon_vld3_lane_v: | |||
9122 | case NEON::BI__builtin_neon_vld3q_lane_v: { | |||
9123 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; | |||
9124 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3lane, Tys); | |||
9125 | Ops.push_back(Ops[1]); | |||
9126 | Ops.erase(Ops.begin()+1); | |||
9127 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
9128 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
9129 | Ops[3] = Builder.CreateBitCast(Ops[3], Ty); | |||
9130 | Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty); | |||
9131 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld3_lane"); | |||
9132 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
9133 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
9134 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9135 | } | |||
9136 | case NEON::BI__builtin_neon_vld4_lane_v: | |||
9137 | case NEON::BI__builtin_neon_vld4q_lane_v: { | |||
9138 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; | |||
9139 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4lane, Tys); | |||
9140 | Ops.push_back(Ops[1]); | |||
9141 | Ops.erase(Ops.begin()+1); | |||
9142 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
9143 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
9144 | Ops[3] = Builder.CreateBitCast(Ops[3], Ty); | |||
9145 | Ops[4] = Builder.CreateBitCast(Ops[4], Ty); | |||
9146 | Ops[5] = Builder.CreateZExt(Ops[5], Int64Ty); | |||
9147 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld4_lane"); | |||
9148 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
9149 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
9150 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
9151 | } | |||
9152 | case NEON::BI__builtin_neon_vst2_v: | |||
9153 | case NEON::BI__builtin_neon_vst2q_v: { | |||
9154 | Ops.push_back(Ops[0]); | |||
9155 | Ops.erase(Ops.begin()); | |||
9156 | llvm::Type *Tys[2] = { VTy, Ops[2]->getType() }; | |||
9157 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2, Tys), | |||
9158 | Ops, ""); | |||
9159 | } | |||
9160 | case NEON::BI__builtin_neon_vst2_lane_v: | |||
9161 | case NEON::BI__builtin_neon_vst2q_lane_v: { | |||
9162 | Ops.push_back(Ops[0]); | |||
9163 | Ops.erase(Ops.begin()); | |||
9164 | Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty); | |||
9165 | llvm::Type *Tys[2] = { VTy, Ops[3]->getType() }; | |||
9166 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2lane, Tys), | |||
9167 | Ops, ""); | |||
9168 | } | |||
9169 | case NEON::BI__builtin_neon_vst3_v: | |||
9170 | case NEON::BI__builtin_neon_vst3q_v: { | |||
9171 | Ops.push_back(Ops[0]); | |||
9172 | Ops.erase(Ops.begin()); | |||
9173 | llvm::Type *Tys[2] = { VTy, Ops[3]->getType() }; | |||
9174 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3, Tys), | |||
9175 | Ops, ""); | |||
9176 | } | |||
9177 | case NEON::BI__builtin_neon_vst3_lane_v: | |||
9178 | case NEON::BI__builtin_neon_vst3q_lane_v: { | |||
9179 | Ops.push_back(Ops[0]); | |||
9180 | Ops.erase(Ops.begin()); | |||
9181 | Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty); | |||
9182 | llvm::Type *Tys[2] = { VTy, Ops[4]->getType() }; | |||
9183 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3lane, Tys), | |||
9184 | Ops, ""); | |||
9185 | } | |||
9186 | case NEON::BI__builtin_neon_vst4_v: | |||
9187 | case NEON::BI__builtin_neon_vst4q_v: { | |||
9188 | Ops.push_back(Ops[0]); | |||
9189 | Ops.erase(Ops.begin()); | |||
9190 | llvm::Type *Tys[2] = { VTy, Ops[4]->getType() }; | |||
9191 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4, Tys), | |||
9192 | Ops, ""); | |||
9193 | } | |||
9194 | case NEON::BI__builtin_neon_vst4_lane_v: | |||
9195 | case NEON::BI__builtin_neon_vst4q_lane_v: { | |||
9196 | Ops.push_back(Ops[0]); | |||
9197 | Ops.erase(Ops.begin()); | |||
9198 | Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty); | |||
9199 | llvm::Type *Tys[2] = { VTy, Ops[5]->getType() }; | |||
9200 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4lane, Tys), | |||
9201 | Ops, ""); | |||
9202 | } | |||
9203 | case NEON::BI__builtin_neon_vtrn_v: | |||
9204 | case NEON::BI__builtin_neon_vtrnq_v: { | |||
9205 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
9206 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
9207 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
9208 | Value *SV = nullptr; | |||
9209 | ||||
9210 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
9211 | SmallVector<uint32_t, 16> Indices; | |||
9212 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { | |||
9213 | Indices.push_back(i+vi); | |||
9214 | Indices.push_back(i+e+vi); | |||
9215 | } | |||
9216 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
9217 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn"); | |||
9218 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
9219 | } | |||
9220 | return SV; | |||
9221 | } | |||
9222 | case NEON::BI__builtin_neon_vuzp_v: | |||
9223 | case NEON::BI__builtin_neon_vuzpq_v: { | |||
9224 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
9225 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
9226 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
9227 | Value *SV = nullptr; | |||
9228 | ||||
9229 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
9230 | SmallVector<uint32_t, 16> Indices; | |||
9231 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) | |||
9232 | Indices.push_back(2*i+vi); | |||
9233 | ||||
9234 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
9235 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp"); | |||
9236 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
9237 | } | |||
9238 | return SV; | |||
9239 | } | |||
9240 | case NEON::BI__builtin_neon_vzip_v: | |||
9241 | case NEON::BI__builtin_neon_vzipq_v: { | |||
9242 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
9243 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
9244 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
9245 | Value *SV = nullptr; | |||
9246 | ||||
9247 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
9248 | SmallVector<uint32_t, 16> Indices; | |||
9249 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { | |||
9250 | Indices.push_back((i + vi*e) >> 1); | |||
9251 | Indices.push_back(((i + vi*e) >> 1)+e); | |||
9252 | } | |||
9253 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
9254 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip"); | |||
9255 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
9256 | } | |||
9257 | return SV; | |||
9258 | } | |||
9259 | case NEON::BI__builtin_neon_vqtbl1q_v: { | |||
9260 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl1, Ty), | |||
9261 | Ops, "vtbl1"); | |||
9262 | } | |||
9263 | case NEON::BI__builtin_neon_vqtbl2q_v: { | |||
9264 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl2, Ty), | |||
9265 | Ops, "vtbl2"); | |||
9266 | } | |||
9267 | case NEON::BI__builtin_neon_vqtbl3q_v: { | |||
9268 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl3, Ty), | |||
9269 | Ops, "vtbl3"); | |||
9270 | } | |||
9271 | case NEON::BI__builtin_neon_vqtbl4q_v: { | |||
9272 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl4, Ty), | |||
9273 | Ops, "vtbl4"); | |||
9274 | } | |||
9275 | case NEON::BI__builtin_neon_vqtbx1q_v: { | |||
9276 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx1, Ty), | |||
9277 | Ops, "vtbx1"); | |||
9278 | } | |||
9279 | case NEON::BI__builtin_neon_vqtbx2q_v: { | |||
9280 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx2, Ty), | |||
9281 | Ops, "vtbx2"); | |||
9282 | } | |||
9283 | case NEON::BI__builtin_neon_vqtbx3q_v: { | |||
9284 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx3, Ty), | |||
9285 | Ops, "vtbx3"); | |||
9286 | } | |||
9287 | case NEON::BI__builtin_neon_vqtbx4q_v: { | |||
9288 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx4, Ty), | |||
9289 | Ops, "vtbx4"); | |||
9290 | } | |||
9291 | case NEON::BI__builtin_neon_vsqadd_v: | |||
9292 | case NEON::BI__builtin_neon_vsqaddq_v: { | |||
9293 | Int = Intrinsic::aarch64_neon_usqadd; | |||
9294 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqadd"); | |||
9295 | } | |||
9296 | case NEON::BI__builtin_neon_vuqadd_v: | |||
9297 | case NEON::BI__builtin_neon_vuqaddq_v: { | |||
9298 | Int = Intrinsic::aarch64_neon_suqadd; | |||
9299 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vuqadd"); | |||
9300 | } | |||
9301 | } | |||
9302 | } | |||
9303 | ||||
9304 | Value *CodeGenFunction::EmitBPFBuiltinExpr(unsigned BuiltinID, | |||
9305 | const CallExpr *E) { | |||
9306 | assert(BuiltinID == BPF::BI__builtin_preserve_field_info &&((BuiltinID == BPF::BI__builtin_preserve_field_info && "unexpected ARM builtin") ? static_cast<void> (0) : __assert_fail ("BuiltinID == BPF::BI__builtin_preserve_field_info && \"unexpected ARM builtin\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9307, __PRETTY_FUNCTION__)) | |||
9307 | "unexpected ARM builtin")((BuiltinID == BPF::BI__builtin_preserve_field_info && "unexpected ARM builtin") ? static_cast<void> (0) : __assert_fail ("BuiltinID == BPF::BI__builtin_preserve_field_info && \"unexpected ARM builtin\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9307, __PRETTY_FUNCTION__)); | |||
9308 | ||||
9309 | const Expr *Arg = E->getArg(0); | |||
9310 | bool IsBitField = Arg->IgnoreParens()->getObjectKind() == OK_BitField; | |||
9311 | ||||
9312 | if (!getDebugInfo()) { | |||
9313 | CGM.Error(E->getExprLoc(), "using builtin_preserve_field_info() without -g"); | |||
9314 | return IsBitField ? EmitLValue(Arg).getBitFieldPointer() | |||
9315 | : EmitLValue(Arg).getPointer(); | |||
9316 | } | |||
9317 | ||||
9318 | // Enable underlying preserve_*_access_index() generation. | |||
9319 | bool OldIsInPreservedAIRegion = IsInPreservedAIRegion; | |||
9320 | IsInPreservedAIRegion = true; | |||
9321 | Value *FieldAddr = IsBitField ? EmitLValue(Arg).getBitFieldPointer() | |||
9322 | : EmitLValue(Arg).getPointer(); | |||
9323 | IsInPreservedAIRegion = OldIsInPreservedAIRegion; | |||
9324 | ||||
9325 | ConstantInt *C = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
9326 | Value *InfoKind = ConstantInt::get(Int64Ty, C->getSExtValue()); | |||
9327 | ||||
9328 | // Built the IR for the preserve_field_info intrinsic. | |||
9329 | llvm::Function *FnGetFieldInfo = llvm::Intrinsic::getDeclaration( | |||
9330 | &CGM.getModule(), llvm::Intrinsic::bpf_preserve_field_info, | |||
9331 | {FieldAddr->getType()}); | |||
9332 | return Builder.CreateCall(FnGetFieldInfo, {FieldAddr, InfoKind}); | |||
9333 | } | |||
9334 | ||||
9335 | llvm::Value *CodeGenFunction:: | |||
9336 | BuildVector(ArrayRef<llvm::Value*> Ops) { | |||
9337 | assert((Ops.size() & (Ops.size() - 1)) == 0 &&(((Ops.size() & (Ops.size() - 1)) == 0 && "Not a power-of-two sized vector!" ) ? static_cast<void> (0) : __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9338, __PRETTY_FUNCTION__)) | |||
9338 | "Not a power-of-two sized vector!")(((Ops.size() & (Ops.size() - 1)) == 0 && "Not a power-of-two sized vector!" ) ? static_cast<void> (0) : __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9338, __PRETTY_FUNCTION__)); | |||
9339 | bool AllConstants = true; | |||
9340 | for (unsigned i = 0, e = Ops.size(); i != e && AllConstants; ++i) | |||
9341 | AllConstants &= isa<Constant>(Ops[i]); | |||
9342 | ||||
9343 | // If this is a constant vector, create a ConstantVector. | |||
9344 | if (AllConstants) { | |||
9345 | SmallVector<llvm::Constant*, 16> CstOps; | |||
9346 | for (unsigned i = 0, e = Ops.size(); i != e; ++i) | |||
9347 | CstOps.push_back(cast<Constant>(Ops[i])); | |||
9348 | return llvm::ConstantVector::get(CstOps); | |||
9349 | } | |||
9350 | ||||
9351 | // Otherwise, insertelement the values to build the vector. | |||
9352 | Value *Result = | |||
9353 | llvm::UndefValue::get(llvm::VectorType::get(Ops[0]->getType(), Ops.size())); | |||
9354 | ||||
9355 | for (unsigned i = 0, e = Ops.size(); i != e; ++i) | |||
9356 | Result = Builder.CreateInsertElement(Result, Ops[i], Builder.getInt32(i)); | |||
9357 | ||||
9358 | return Result; | |||
9359 | } | |||
9360 | ||||
9361 | // Convert the mask from an integer type to a vector of i1. | |||
9362 | static Value *getMaskVecValue(CodeGenFunction &CGF, Value *Mask, | |||
9363 | unsigned NumElts) { | |||
9364 | ||||
9365 | llvm::VectorType *MaskTy = llvm::VectorType::get(CGF.Builder.getInt1Ty(), | |||
9366 | cast<IntegerType>(Mask->getType())->getBitWidth()); | |||
9367 | Value *MaskVec = CGF.Builder.CreateBitCast(Mask, MaskTy); | |||
9368 | ||||
9369 | // If we have less than 8 elements, then the starting mask was an i8 and | |||
9370 | // we need to extract down to the right number of elements. | |||
9371 | if (NumElts < 8) { | |||
9372 | uint32_t Indices[4]; | |||
9373 | for (unsigned i = 0; i != NumElts; ++i) | |||
9374 | Indices[i] = i; | |||
9375 | MaskVec = CGF.Builder.CreateShuffleVector(MaskVec, MaskVec, | |||
9376 | makeArrayRef(Indices, NumElts), | |||
9377 | "extract"); | |||
9378 | } | |||
9379 | return MaskVec; | |||
9380 | } | |||
9381 | ||||
9382 | static Value *EmitX86MaskedStore(CodeGenFunction &CGF, | |||
9383 | ArrayRef<Value *> Ops, | |||
9384 | unsigned Align) { | |||
9385 | // Cast the pointer to right type. | |||
9386 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], | |||
9387 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
9388 | ||||
9389 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], | |||
9390 | Ops[1]->getType()->getVectorNumElements()); | |||
9391 | ||||
9392 | return CGF.Builder.CreateMaskedStore(Ops[1], Ptr, Align, MaskVec); | |||
9393 | } | |||
9394 | ||||
9395 | static Value *EmitX86MaskedLoad(CodeGenFunction &CGF, | |||
9396 | ArrayRef<Value *> Ops, unsigned Align) { | |||
9397 | // Cast the pointer to right type. | |||
9398 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], | |||
9399 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
9400 | ||||
9401 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], | |||
9402 | Ops[1]->getType()->getVectorNumElements()); | |||
9403 | ||||
9404 | return CGF.Builder.CreateMaskedLoad(Ptr, Align, MaskVec, Ops[1]); | |||
9405 | } | |||
9406 | ||||
9407 | static Value *EmitX86ExpandLoad(CodeGenFunction &CGF, | |||
9408 | ArrayRef<Value *> Ops) { | |||
9409 | llvm::Type *ResultTy = Ops[1]->getType(); | |||
9410 | llvm::Type *PtrTy = ResultTy->getVectorElementType(); | |||
9411 | ||||
9412 | // Cast the pointer to element type. | |||
9413 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], | |||
9414 | llvm::PointerType::getUnqual(PtrTy)); | |||
9415 | ||||
9416 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], | |||
9417 | ResultTy->getVectorNumElements()); | |||
9418 | ||||
9419 | llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_expandload, | |||
9420 | ResultTy); | |||
9421 | return CGF.Builder.CreateCall(F, { Ptr, MaskVec, Ops[1] }); | |||
9422 | } | |||
9423 | ||||
9424 | static Value *EmitX86CompressExpand(CodeGenFunction &CGF, | |||
9425 | ArrayRef<Value *> Ops, | |||
9426 | bool IsCompress) { | |||
9427 | llvm::Type *ResultTy = Ops[1]->getType(); | |||
9428 | ||||
9429 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], | |||
9430 | ResultTy->getVectorNumElements()); | |||
9431 | ||||
9432 | Intrinsic::ID IID = IsCompress ? Intrinsic::x86_avx512_mask_compress | |||
9433 | : Intrinsic::x86_avx512_mask_expand; | |||
9434 | llvm::Function *F = CGF.CGM.getIntrinsic(IID, ResultTy); | |||
9435 | return CGF.Builder.CreateCall(F, { Ops[0], Ops[1], MaskVec }); | |||
9436 | } | |||
9437 | ||||
9438 | static Value *EmitX86CompressStore(CodeGenFunction &CGF, | |||
9439 | ArrayRef<Value *> Ops) { | |||
9440 | llvm::Type *ResultTy = Ops[1]->getType(); | |||
9441 | llvm::Type *PtrTy = ResultTy->getVectorElementType(); | |||
9442 | ||||
9443 | // Cast the pointer to element type. | |||
9444 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], | |||
9445 | llvm::PointerType::getUnqual(PtrTy)); | |||
9446 | ||||
9447 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], | |||
9448 | ResultTy->getVectorNumElements()); | |||
9449 | ||||
9450 | llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_compressstore, | |||
9451 | ResultTy); | |||
9452 | return CGF.Builder.CreateCall(F, { Ops[1], Ptr, MaskVec }); | |||
9453 | } | |||
9454 | ||||
9455 | static Value *EmitX86MaskLogic(CodeGenFunction &CGF, Instruction::BinaryOps Opc, | |||
9456 | ArrayRef<Value *> Ops, | |||
9457 | bool InvertLHS = false) { | |||
9458 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
9459 | Value *LHS = getMaskVecValue(CGF, Ops[0], NumElts); | |||
9460 | Value *RHS = getMaskVecValue(CGF, Ops[1], NumElts); | |||
9461 | ||||
9462 | if (InvertLHS) | |||
9463 | LHS = CGF.Builder.CreateNot(LHS); | |||
9464 | ||||
9465 | return CGF.Builder.CreateBitCast(CGF.Builder.CreateBinOp(Opc, LHS, RHS), | |||
9466 | Ops[0]->getType()); | |||
9467 | } | |||
9468 | ||||
9469 | static Value *EmitX86FunnelShift(CodeGenFunction &CGF, Value *Op0, Value *Op1, | |||
9470 | Value *Amt, bool IsRight) { | |||
9471 | llvm::Type *Ty = Op0->getType(); | |||
9472 | ||||
9473 | // Amount may be scalar immediate, in which case create a splat vector. | |||
9474 | // Funnel shifts amounts are treated as modulo and types are all power-of-2 so | |||
9475 | // we only care about the lowest log2 bits anyway. | |||
9476 | if (Amt->getType() != Ty) { | |||
9477 | unsigned NumElts = Ty->getVectorNumElements(); | |||
9478 | Amt = CGF.Builder.CreateIntCast(Amt, Ty->getScalarType(), false); | |||
9479 | Amt = CGF.Builder.CreateVectorSplat(NumElts, Amt); | |||
9480 | } | |||
9481 | ||||
9482 | unsigned IID = IsRight ? Intrinsic::fshr : Intrinsic::fshl; | |||
9483 | Function *F = CGF.CGM.getIntrinsic(IID, Ty); | |||
9484 | return CGF.Builder.CreateCall(F, {Op0, Op1, Amt}); | |||
9485 | } | |||
9486 | ||||
9487 | static Value *EmitX86vpcom(CodeGenFunction &CGF, ArrayRef<Value *> Ops, | |||
9488 | bool IsSigned) { | |||
9489 | Value *Op0 = Ops[0]; | |||
9490 | Value *Op1 = Ops[1]; | |||
9491 | llvm::Type *Ty = Op0->getType(); | |||
9492 | uint64_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; | |||
9493 | ||||
9494 | CmpInst::Predicate Pred; | |||
9495 | switch (Imm) { | |||
9496 | case 0x0: | |||
9497 | Pred = IsSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; | |||
9498 | break; | |||
9499 | case 0x1: | |||
9500 | Pred = IsSigned ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; | |||
9501 | break; | |||
9502 | case 0x2: | |||
9503 | Pred = IsSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; | |||
9504 | break; | |||
9505 | case 0x3: | |||
9506 | Pred = IsSigned ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; | |||
9507 | break; | |||
9508 | case 0x4: | |||
9509 | Pred = ICmpInst::ICMP_EQ; | |||
9510 | break; | |||
9511 | case 0x5: | |||
9512 | Pred = ICmpInst::ICMP_NE; | |||
9513 | break; | |||
9514 | case 0x6: | |||
9515 | return llvm::Constant::getNullValue(Ty); // FALSE | |||
9516 | case 0x7: | |||
9517 | return llvm::Constant::getAllOnesValue(Ty); // TRUE | |||
9518 | default: | |||
9519 | llvm_unreachable("Unexpected XOP vpcom/vpcomu predicate")::llvm::llvm_unreachable_internal("Unexpected XOP vpcom/vpcomu predicate" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9519); | |||
9520 | } | |||
9521 | ||||
9522 | Value *Cmp = CGF.Builder.CreateICmp(Pred, Op0, Op1); | |||
9523 | Value *Res = CGF.Builder.CreateSExt(Cmp, Ty); | |||
9524 | return Res; | |||
9525 | } | |||
9526 | ||||
9527 | static Value *EmitX86Select(CodeGenFunction &CGF, | |||
9528 | Value *Mask, Value *Op0, Value *Op1) { | |||
9529 | ||||
9530 | // If the mask is all ones just return first argument. | |||
9531 | if (const auto *C = dyn_cast<Constant>(Mask)) | |||
9532 | if (C->isAllOnesValue()) | |||
9533 | return Op0; | |||
9534 | ||||
9535 | Mask = getMaskVecValue(CGF, Mask, Op0->getType()->getVectorNumElements()); | |||
9536 | ||||
9537 | return CGF.Builder.CreateSelect(Mask, Op0, Op1); | |||
9538 | } | |||
9539 | ||||
9540 | static Value *EmitX86ScalarSelect(CodeGenFunction &CGF, | |||
9541 | Value *Mask, Value *Op0, Value *Op1) { | |||
9542 | // If the mask is all ones just return first argument. | |||
9543 | if (const auto *C = dyn_cast<Constant>(Mask)) | |||
9544 | if (C->isAllOnesValue()) | |||
9545 | return Op0; | |||
9546 | ||||
9547 | llvm::VectorType *MaskTy = | |||
9548 | llvm::VectorType::get(CGF.Builder.getInt1Ty(), | |||
9549 | Mask->getType()->getIntegerBitWidth()); | |||
9550 | Mask = CGF.Builder.CreateBitCast(Mask, MaskTy); | |||
9551 | Mask = CGF.Builder.CreateExtractElement(Mask, (uint64_t)0); | |||
9552 | return CGF.Builder.CreateSelect(Mask, Op0, Op1); | |||
9553 | } | |||
9554 | ||||
9555 | static Value *EmitX86MaskedCompareResult(CodeGenFunction &CGF, Value *Cmp, | |||
9556 | unsigned NumElts, Value *MaskIn) { | |||
9557 | if (MaskIn) { | |||
9558 | const auto *C = dyn_cast<Constant>(MaskIn); | |||
9559 | if (!C || !C->isAllOnesValue()) | |||
9560 | Cmp = CGF.Builder.CreateAnd(Cmp, getMaskVecValue(CGF, MaskIn, NumElts)); | |||
9561 | } | |||
9562 | ||||
9563 | if (NumElts < 8) { | |||
9564 | uint32_t Indices[8]; | |||
9565 | for (unsigned i = 0; i != NumElts; ++i) | |||
9566 | Indices[i] = i; | |||
9567 | for (unsigned i = NumElts; i != 8; ++i) | |||
9568 | Indices[i] = i % NumElts + NumElts; | |||
9569 | Cmp = CGF.Builder.CreateShuffleVector( | |||
9570 | Cmp, llvm::Constant::getNullValue(Cmp->getType()), Indices); | |||
9571 | } | |||
9572 | ||||
9573 | return CGF.Builder.CreateBitCast(Cmp, | |||
9574 | IntegerType::get(CGF.getLLVMContext(), | |||
9575 | std::max(NumElts, 8U))); | |||
9576 | } | |||
9577 | ||||
9578 | static Value *EmitX86MaskedCompare(CodeGenFunction &CGF, unsigned CC, | |||
9579 | bool Signed, ArrayRef<Value *> Ops) { | |||
9580 | assert((Ops.size() == 2 || Ops.size() == 4) &&(((Ops.size() == 2 || Ops.size() == 4) && "Unexpected number of arguments" ) ? static_cast<void> (0) : __assert_fail ("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9581, __PRETTY_FUNCTION__)) | |||
9581 | "Unexpected number of arguments")(((Ops.size() == 2 || Ops.size() == 4) && "Unexpected number of arguments" ) ? static_cast<void> (0) : __assert_fail ("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9581, __PRETTY_FUNCTION__)); | |||
9582 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
9583 | Value *Cmp; | |||
9584 | ||||
9585 | if (CC == 3) { | |||
9586 | Cmp = Constant::getNullValue( | |||
9587 | llvm::VectorType::get(CGF.Builder.getInt1Ty(), NumElts)); | |||
9588 | } else if (CC == 7) { | |||
9589 | Cmp = Constant::getAllOnesValue( | |||
9590 | llvm::VectorType::get(CGF.Builder.getInt1Ty(), NumElts)); | |||
9591 | } else { | |||
9592 | ICmpInst::Predicate Pred; | |||
9593 | switch (CC) { | |||
9594 | default: llvm_unreachable("Unknown condition code")::llvm::llvm_unreachable_internal("Unknown condition code", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9594); | |||
9595 | case 0: Pred = ICmpInst::ICMP_EQ; break; | |||
9596 | case 1: Pred = Signed ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; break; | |||
9597 | case 2: Pred = Signed ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; break; | |||
9598 | case 4: Pred = ICmpInst::ICMP_NE; break; | |||
9599 | case 5: Pred = Signed ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; break; | |||
9600 | case 6: Pred = Signed ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; break; | |||
9601 | } | |||
9602 | Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]); | |||
9603 | } | |||
9604 | ||||
9605 | Value *MaskIn = nullptr; | |||
9606 | if (Ops.size() == 4) | |||
9607 | MaskIn = Ops[3]; | |||
9608 | ||||
9609 | return EmitX86MaskedCompareResult(CGF, Cmp, NumElts, MaskIn); | |||
9610 | } | |||
9611 | ||||
9612 | static Value *EmitX86ConvertToMask(CodeGenFunction &CGF, Value *In) { | |||
9613 | Value *Zero = Constant::getNullValue(In->getType()); | |||
9614 | return EmitX86MaskedCompare(CGF, 1, true, { In, Zero }); | |||
9615 | } | |||
9616 | ||||
9617 | static Value *EmitX86ConvertIntToFp(CodeGenFunction &CGF, | |||
9618 | ArrayRef<Value *> Ops, bool IsSigned) { | |||
9619 | unsigned Rnd = cast<llvm::ConstantInt>(Ops[3])->getZExtValue(); | |||
9620 | llvm::Type *Ty = Ops[1]->getType(); | |||
9621 | ||||
9622 | Value *Res; | |||
9623 | if (Rnd != 4) { | |||
9624 | Intrinsic::ID IID = IsSigned ? Intrinsic::x86_avx512_sitofp_round | |||
9625 | : Intrinsic::x86_avx512_uitofp_round; | |||
9626 | Function *F = CGF.CGM.getIntrinsic(IID, { Ty, Ops[0]->getType() }); | |||
9627 | Res = CGF.Builder.CreateCall(F, { Ops[0], Ops[3] }); | |||
9628 | } else { | |||
9629 | Res = IsSigned ? CGF.Builder.CreateSIToFP(Ops[0], Ty) | |||
9630 | : CGF.Builder.CreateUIToFP(Ops[0], Ty); | |||
9631 | } | |||
9632 | ||||
9633 | return EmitX86Select(CGF, Ops[2], Res, Ops[1]); | |||
9634 | } | |||
9635 | ||||
9636 | static Value *EmitX86Abs(CodeGenFunction &CGF, ArrayRef<Value *> Ops) { | |||
9637 | ||||
9638 | llvm::Type *Ty = Ops[0]->getType(); | |||
9639 | Value *Zero = llvm::Constant::getNullValue(Ty); | |||
9640 | Value *Sub = CGF.Builder.CreateSub(Zero, Ops[0]); | |||
9641 | Value *Cmp = CGF.Builder.CreateICmp(ICmpInst::ICMP_SGT, Ops[0], Zero); | |||
9642 | Value *Res = CGF.Builder.CreateSelect(Cmp, Ops[0], Sub); | |||
9643 | return Res; | |||
9644 | } | |||
9645 | ||||
9646 | static Value *EmitX86MinMax(CodeGenFunction &CGF, ICmpInst::Predicate Pred, | |||
9647 | ArrayRef<Value *> Ops) { | |||
9648 | Value *Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]); | |||
9649 | Value *Res = CGF.Builder.CreateSelect(Cmp, Ops[0], Ops[1]); | |||
9650 | ||||
9651 | assert(Ops.size() == 2)((Ops.size() == 2) ? static_cast<void> (0) : __assert_fail ("Ops.size() == 2", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9651, __PRETTY_FUNCTION__)); | |||
9652 | return Res; | |||
9653 | } | |||
9654 | ||||
9655 | // Lowers X86 FMA intrinsics to IR. | |||
9656 | static Value *EmitX86FMAExpr(CodeGenFunction &CGF, ArrayRef<Value *> Ops, | |||
9657 | unsigned BuiltinID, bool IsAddSub) { | |||
9658 | ||||
9659 | bool Subtract = false; | |||
9660 | Intrinsic::ID IID = Intrinsic::not_intrinsic; | |||
9661 | switch (BuiltinID) { | |||
9662 | default: break; | |||
9663 | case clang::X86::BI__builtin_ia32_vfmsubps512_mask3: | |||
9664 | Subtract = true; | |||
9665 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
9666 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask: | |||
9667 | case clang::X86::BI__builtin_ia32_vfmaddps512_maskz: | |||
9668 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask3: | |||
9669 | IID = llvm::Intrinsic::x86_avx512_vfmadd_ps_512; break; | |||
9670 | case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3: | |||
9671 | Subtract = true; | |||
9672 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
9673 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask: | |||
9674 | case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz: | |||
9675 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3: | |||
9676 | IID = llvm::Intrinsic::x86_avx512_vfmadd_pd_512; break; | |||
9677 | case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3: | |||
9678 | Subtract = true; | |||
9679 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
9680 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask: | |||
9681 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz: | |||
9682 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3: | |||
9683 | IID = llvm::Intrinsic::x86_avx512_vfmaddsub_ps_512; | |||
9684 | break; | |||
9685 | case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3: | |||
9686 | Subtract = true; | |||
9687 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
9688 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask: | |||
9689 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz: | |||
9690 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3: | |||
9691 | IID = llvm::Intrinsic::x86_avx512_vfmaddsub_pd_512; | |||
9692 | break; | |||
9693 | } | |||
9694 | ||||
9695 | Value *A = Ops[0]; | |||
9696 | Value *B = Ops[1]; | |||
9697 | Value *C = Ops[2]; | |||
9698 | ||||
9699 | if (Subtract) | |||
9700 | C = CGF.Builder.CreateFNeg(C); | |||
9701 | ||||
9702 | Value *Res; | |||
9703 | ||||
9704 | // Only handle in case of _MM_FROUND_CUR_DIRECTION/4 (no rounding). | |||
9705 | if (IID != Intrinsic::not_intrinsic && | |||
9706 | cast<llvm::ConstantInt>(Ops.back())->getZExtValue() != (uint64_t)4) { | |||
9707 | Function *Intr = CGF.CGM.getIntrinsic(IID); | |||
9708 | Res = CGF.Builder.CreateCall(Intr, {A, B, C, Ops.back() }); | |||
9709 | } else { | |||
9710 | llvm::Type *Ty = A->getType(); | |||
9711 | Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ty); | |||
9712 | Res = CGF.Builder.CreateCall(FMA, {A, B, C} ); | |||
9713 | ||||
9714 | if (IsAddSub) { | |||
9715 | // Negate even elts in C using a mask. | |||
9716 | unsigned NumElts = Ty->getVectorNumElements(); | |||
9717 | SmallVector<uint32_t, 16> Indices(NumElts); | |||
9718 | for (unsigned i = 0; i != NumElts; ++i) | |||
9719 | Indices[i] = i + (i % 2) * NumElts; | |||
9720 | ||||
9721 | Value *NegC = CGF.Builder.CreateFNeg(C); | |||
9722 | Value *FMSub = CGF.Builder.CreateCall(FMA, {A, B, NegC} ); | |||
9723 | Res = CGF.Builder.CreateShuffleVector(FMSub, Res, Indices); | |||
9724 | } | |||
9725 | } | |||
9726 | ||||
9727 | // Handle any required masking. | |||
9728 | Value *MaskFalseVal = nullptr; | |||
9729 | switch (BuiltinID) { | |||
9730 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask: | |||
9731 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask: | |||
9732 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask: | |||
9733 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask: | |||
9734 | MaskFalseVal = Ops[0]; | |||
9735 | break; | |||
9736 | case clang::X86::BI__builtin_ia32_vfmaddps512_maskz: | |||
9737 | case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz: | |||
9738 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz: | |||
9739 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz: | |||
9740 | MaskFalseVal = Constant::getNullValue(Ops[0]->getType()); | |||
9741 | break; | |||
9742 | case clang::X86::BI__builtin_ia32_vfmsubps512_mask3: | |||
9743 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask3: | |||
9744 | case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3: | |||
9745 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3: | |||
9746 | case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3: | |||
9747 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3: | |||
9748 | case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3: | |||
9749 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3: | |||
9750 | MaskFalseVal = Ops[2]; | |||
9751 | break; | |||
9752 | } | |||
9753 | ||||
9754 | if (MaskFalseVal) | |||
9755 | return EmitX86Select(CGF, Ops[3], Res, MaskFalseVal); | |||
9756 | ||||
9757 | return Res; | |||
9758 | } | |||
9759 | ||||
9760 | static Value * | |||
9761 | EmitScalarFMAExpr(CodeGenFunction &CGF, MutableArrayRef<Value *> Ops, | |||
9762 | Value *Upper, bool ZeroMask = false, unsigned PTIdx = 0, | |||
9763 | bool NegAcc = false) { | |||
9764 | unsigned Rnd = 4; | |||
9765 | if (Ops.size() > 4) | |||
9766 | Rnd = cast<llvm::ConstantInt>(Ops[4])->getZExtValue(); | |||
9767 | ||||
9768 | if (NegAcc) | |||
9769 | Ops[2] = CGF.Builder.CreateFNeg(Ops[2]); | |||
9770 | ||||
9771 | Ops[0] = CGF.Builder.CreateExtractElement(Ops[0], (uint64_t)0); | |||
9772 | Ops[1] = CGF.Builder.CreateExtractElement(Ops[1], (uint64_t)0); | |||
9773 | Ops[2] = CGF.Builder.CreateExtractElement(Ops[2], (uint64_t)0); | |||
9774 | Value *Res; | |||
9775 | if (Rnd != 4) { | |||
9776 | Intrinsic::ID IID = Ops[0]->getType()->getPrimitiveSizeInBits() == 32 ? | |||
9777 | Intrinsic::x86_avx512_vfmadd_f32 : | |||
9778 | Intrinsic::x86_avx512_vfmadd_f64; | |||
9779 | Res = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID), | |||
9780 | {Ops[0], Ops[1], Ops[2], Ops[4]}); | |||
9781 | } else { | |||
9782 | Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ops[0]->getType()); | |||
9783 | Res = CGF.Builder.CreateCall(FMA, Ops.slice(0, 3)); | |||
9784 | } | |||
9785 | // If we have more than 3 arguments, we need to do masking. | |||
9786 | if (Ops.size() > 3) { | |||
9787 | Value *PassThru = ZeroMask ? Constant::getNullValue(Res->getType()) | |||
9788 | : Ops[PTIdx]; | |||
9789 | ||||
9790 | // If we negated the accumulator and the its the PassThru value we need to | |||
9791 | // bypass the negate. Conveniently Upper should be the same thing in this | |||
9792 | // case. | |||
9793 | if (NegAcc && PTIdx == 2) | |||
9794 | PassThru = CGF.Builder.CreateExtractElement(Upper, (uint64_t)0); | |||
9795 | ||||
9796 | Res = EmitX86ScalarSelect(CGF, Ops[3], Res, PassThru); | |||
9797 | } | |||
9798 | return CGF.Builder.CreateInsertElement(Upper, Res, (uint64_t)0); | |||
9799 | } | |||
9800 | ||||
9801 | static Value *EmitX86Muldq(CodeGenFunction &CGF, bool IsSigned, | |||
9802 | ArrayRef<Value *> Ops) { | |||
9803 | llvm::Type *Ty = Ops[0]->getType(); | |||
9804 | // Arguments have a vXi32 type so cast to vXi64. | |||
9805 | Ty = llvm::VectorType::get(CGF.Int64Ty, | |||
9806 | Ty->getPrimitiveSizeInBits() / 64); | |||
9807 | Value *LHS = CGF.Builder.CreateBitCast(Ops[0], Ty); | |||
9808 | Value *RHS = CGF.Builder.CreateBitCast(Ops[1], Ty); | |||
9809 | ||||
9810 | if (IsSigned) { | |||
9811 | // Shift left then arithmetic shift right. | |||
9812 | Constant *ShiftAmt = ConstantInt::get(Ty, 32); | |||
9813 | LHS = CGF.Builder.CreateShl(LHS, ShiftAmt); | |||
9814 | LHS = CGF.Builder.CreateAShr(LHS, ShiftAmt); | |||
9815 | RHS = CGF.Builder.CreateShl(RHS, ShiftAmt); | |||
9816 | RHS = CGF.Builder.CreateAShr(RHS, ShiftAmt); | |||
9817 | } else { | |||
9818 | // Clear the upper bits. | |||
9819 | Constant *Mask = ConstantInt::get(Ty, 0xffffffff); | |||
9820 | LHS = CGF.Builder.CreateAnd(LHS, Mask); | |||
9821 | RHS = CGF.Builder.CreateAnd(RHS, Mask); | |||
9822 | } | |||
9823 | ||||
9824 | return CGF.Builder.CreateMul(LHS, RHS); | |||
9825 | } | |||
9826 | ||||
9827 | // Emit a masked pternlog intrinsic. This only exists because the header has to | |||
9828 | // use a macro and we aren't able to pass the input argument to a pternlog | |||
9829 | // builtin and a select builtin without evaluating it twice. | |||
9830 | static Value *EmitX86Ternlog(CodeGenFunction &CGF, bool ZeroMask, | |||
9831 | ArrayRef<Value *> Ops) { | |||
9832 | llvm::Type *Ty = Ops[0]->getType(); | |||
9833 | ||||
9834 | unsigned VecWidth = Ty->getPrimitiveSizeInBits(); | |||
9835 | unsigned EltWidth = Ty->getScalarSizeInBits(); | |||
9836 | Intrinsic::ID IID; | |||
9837 | if (VecWidth == 128 && EltWidth == 32) | |||
9838 | IID = Intrinsic::x86_avx512_pternlog_d_128; | |||
9839 | else if (VecWidth == 256 && EltWidth == 32) | |||
9840 | IID = Intrinsic::x86_avx512_pternlog_d_256; | |||
9841 | else if (VecWidth == 512 && EltWidth == 32) | |||
9842 | IID = Intrinsic::x86_avx512_pternlog_d_512; | |||
9843 | else if (VecWidth == 128 && EltWidth == 64) | |||
9844 | IID = Intrinsic::x86_avx512_pternlog_q_128; | |||
9845 | else if (VecWidth == 256 && EltWidth == 64) | |||
9846 | IID = Intrinsic::x86_avx512_pternlog_q_256; | |||
9847 | else if (VecWidth == 512 && EltWidth == 64) | |||
9848 | IID = Intrinsic::x86_avx512_pternlog_q_512; | |||
9849 | else | |||
9850 | llvm_unreachable("Unexpected intrinsic")::llvm::llvm_unreachable_internal("Unexpected intrinsic", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9850); | |||
9851 | ||||
9852 | Value *Ternlog = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID), | |||
9853 | Ops.drop_back()); | |||
9854 | Value *PassThru = ZeroMask ? ConstantAggregateZero::get(Ty) : Ops[0]; | |||
9855 | return EmitX86Select(CGF, Ops[4], Ternlog, PassThru); | |||
9856 | } | |||
9857 | ||||
9858 | static Value *EmitX86SExtMask(CodeGenFunction &CGF, Value *Op, | |||
9859 | llvm::Type *DstTy) { | |||
9860 | unsigned NumberOfElements = DstTy->getVectorNumElements(); | |||
9861 | Value *Mask = getMaskVecValue(CGF, Op, NumberOfElements); | |||
9862 | return CGF.Builder.CreateSExt(Mask, DstTy, "vpmovm2"); | |||
9863 | } | |||
9864 | ||||
9865 | // Emit addition or subtraction with signed/unsigned saturation. | |||
9866 | static Value *EmitX86AddSubSatExpr(CodeGenFunction &CGF, | |||
9867 | ArrayRef<Value *> Ops, bool IsSigned, | |||
9868 | bool IsAddition) { | |||
9869 | Intrinsic::ID IID = | |||
9870 | IsSigned ? (IsAddition ? Intrinsic::sadd_sat : Intrinsic::ssub_sat) | |||
9871 | : (IsAddition ? Intrinsic::uadd_sat : Intrinsic::usub_sat); | |||
9872 | llvm::Function *F = CGF.CGM.getIntrinsic(IID, Ops[0]->getType()); | |||
9873 | return CGF.Builder.CreateCall(F, {Ops[0], Ops[1]}); | |||
9874 | } | |||
9875 | ||||
9876 | Value *CodeGenFunction::EmitX86CpuIs(const CallExpr *E) { | |||
9877 | const Expr *CPUExpr = E->getArg(0)->IgnoreParenCasts(); | |||
9878 | StringRef CPUStr = cast<clang::StringLiteral>(CPUExpr)->getString(); | |||
9879 | return EmitX86CpuIs(CPUStr); | |||
9880 | } | |||
9881 | ||||
9882 | // Convert a BF16 to a float. | |||
9883 | static Value *EmitX86CvtBF16ToFloatExpr(CodeGenFunction &CGF, | |||
9884 | const CallExpr *E, | |||
9885 | ArrayRef<Value *> Ops) { | |||
9886 | llvm::Type *Int32Ty = CGF.Builder.getInt32Ty(); | |||
9887 | Value *ZeroExt = CGF.Builder.CreateZExt(Ops[0], Int32Ty); | |||
9888 | Value *Shl = CGF.Builder.CreateShl(ZeroExt, 16); | |||
9889 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); | |||
9890 | Value *BitCast = CGF.Builder.CreateBitCast(Shl, ResultType); | |||
9891 | return BitCast; | |||
9892 | } | |||
9893 | ||||
9894 | Value *CodeGenFunction::EmitX86CpuIs(StringRef CPUStr) { | |||
9895 | ||||
9896 | llvm::Type *Int32Ty = Builder.getInt32Ty(); | |||
9897 | ||||
9898 | // Matching the struct layout from the compiler-rt/libgcc structure that is | |||
9899 | // filled in: | |||
9900 | // unsigned int __cpu_vendor; | |||
9901 | // unsigned int __cpu_type; | |||
9902 | // unsigned int __cpu_subtype; | |||
9903 | // unsigned int __cpu_features[1]; | |||
9904 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty, | |||
9905 | llvm::ArrayType::get(Int32Ty, 1)); | |||
9906 | ||||
9907 | // Grab the global __cpu_model. | |||
9908 | llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model"); | |||
9909 | cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true); | |||
9910 | ||||
9911 | // Calculate the index needed to access the correct field based on the | |||
9912 | // range. Also adjust the expected value. | |||
9913 | unsigned Index; | |||
9914 | unsigned Value; | |||
9915 | std::tie(Index, Value) = StringSwitch<std::pair<unsigned, unsigned>>(CPUStr) | |||
9916 | #define X86_VENDOR(ENUM, STRING) \ | |||
9917 | .Case(STRING, {0u, static_cast<unsigned>(llvm::X86::ENUM)}) | |||
9918 | #define X86_CPU_TYPE_COMPAT_WITH_ALIAS(ARCHNAME, ENUM, STR, ALIAS) \ | |||
9919 | .Cases(STR, ALIAS, {1u, static_cast<unsigned>(llvm::X86::ENUM)}) | |||
9920 | #define X86_CPU_TYPE_COMPAT(ARCHNAME, ENUM, STR) \ | |||
9921 | .Case(STR, {1u, static_cast<unsigned>(llvm::X86::ENUM)}) | |||
9922 | #define X86_CPU_SUBTYPE_COMPAT(ARCHNAME, ENUM, STR) \ | |||
9923 | .Case(STR, {2u, static_cast<unsigned>(llvm::X86::ENUM)}) | |||
9924 | #include "llvm/Support/X86TargetParser.def" | |||
9925 | .Default({0, 0}); | |||
9926 | assert(Value != 0 && "Invalid CPUStr passed to CpuIs")((Value != 0 && "Invalid CPUStr passed to CpuIs") ? static_cast <void> (0) : __assert_fail ("Value != 0 && \"Invalid CPUStr passed to CpuIs\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9926, __PRETTY_FUNCTION__)); | |||
9927 | ||||
9928 | // Grab the appropriate field from __cpu_model. | |||
9929 | llvm::Value *Idxs[] = {ConstantInt::get(Int32Ty, 0), | |||
9930 | ConstantInt::get(Int32Ty, Index)}; | |||
9931 | llvm::Value *CpuValue = Builder.CreateGEP(STy, CpuModel, Idxs); | |||
9932 | CpuValue = Builder.CreateAlignedLoad(CpuValue, CharUnits::fromQuantity(4)); | |||
9933 | ||||
9934 | // Check the value of the field against the requested value. | |||
9935 | return Builder.CreateICmpEQ(CpuValue, | |||
9936 | llvm::ConstantInt::get(Int32Ty, Value)); | |||
9937 | } | |||
9938 | ||||
9939 | Value *CodeGenFunction::EmitX86CpuSupports(const CallExpr *E) { | |||
9940 | const Expr *FeatureExpr = E->getArg(0)->IgnoreParenCasts(); | |||
9941 | StringRef FeatureStr = cast<StringLiteral>(FeatureExpr)->getString(); | |||
9942 | return EmitX86CpuSupports(FeatureStr); | |||
9943 | } | |||
9944 | ||||
9945 | uint64_t | |||
9946 | CodeGenFunction::GetX86CpuSupportsMask(ArrayRef<StringRef> FeatureStrs) { | |||
9947 | // Processor features and mapping to processor feature value. | |||
9948 | uint64_t FeaturesMask = 0; | |||
9949 | for (const StringRef &FeatureStr : FeatureStrs) { | |||
9950 | unsigned Feature = | |||
9951 | StringSwitch<unsigned>(FeatureStr) | |||
9952 | #define X86_FEATURE_COMPAT(VAL, ENUM, STR) .Case(STR, VAL) | |||
9953 | #include "llvm/Support/X86TargetParser.def" | |||
9954 | ; | |||
9955 | FeaturesMask |= (1ULL << Feature); | |||
9956 | } | |||
9957 | return FeaturesMask; | |||
9958 | } | |||
9959 | ||||
9960 | Value *CodeGenFunction::EmitX86CpuSupports(ArrayRef<StringRef> FeatureStrs) { | |||
9961 | return EmitX86CpuSupports(GetX86CpuSupportsMask(FeatureStrs)); | |||
9962 | } | |||
9963 | ||||
9964 | llvm::Value *CodeGenFunction::EmitX86CpuSupports(uint64_t FeaturesMask) { | |||
9965 | uint32_t Features1 = Lo_32(FeaturesMask); | |||
9966 | uint32_t Features2 = Hi_32(FeaturesMask); | |||
9967 | ||||
9968 | Value *Result = Builder.getTrue(); | |||
9969 | ||||
9970 | if (Features1 != 0) { | |||
9971 | // Matching the struct layout from the compiler-rt/libgcc structure that is | |||
9972 | // filled in: | |||
9973 | // unsigned int __cpu_vendor; | |||
9974 | // unsigned int __cpu_type; | |||
9975 | // unsigned int __cpu_subtype; | |||
9976 | // unsigned int __cpu_features[1]; | |||
9977 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty, | |||
9978 | llvm::ArrayType::get(Int32Ty, 1)); | |||
9979 | ||||
9980 | // Grab the global __cpu_model. | |||
9981 | llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model"); | |||
9982 | cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true); | |||
9983 | ||||
9984 | // Grab the first (0th) element from the field __cpu_features off of the | |||
9985 | // global in the struct STy. | |||
9986 | Value *Idxs[] = {Builder.getInt32(0), Builder.getInt32(3), | |||
9987 | Builder.getInt32(0)}; | |||
9988 | Value *CpuFeatures = Builder.CreateGEP(STy, CpuModel, Idxs); | |||
9989 | Value *Features = | |||
9990 | Builder.CreateAlignedLoad(CpuFeatures, CharUnits::fromQuantity(4)); | |||
9991 | ||||
9992 | // Check the value of the bit corresponding to the feature requested. | |||
9993 | Value *Mask = Builder.getInt32(Features1); | |||
9994 | Value *Bitset = Builder.CreateAnd(Features, Mask); | |||
9995 | Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask); | |||
9996 | Result = Builder.CreateAnd(Result, Cmp); | |||
9997 | } | |||
9998 | ||||
9999 | if (Features2 != 0) { | |||
10000 | llvm::Constant *CpuFeatures2 = CGM.CreateRuntimeVariable(Int32Ty, | |||
10001 | "__cpu_features2"); | |||
10002 | cast<llvm::GlobalValue>(CpuFeatures2)->setDSOLocal(true); | |||
10003 | ||||
10004 | Value *Features = | |||
10005 | Builder.CreateAlignedLoad(CpuFeatures2, CharUnits::fromQuantity(4)); | |||
10006 | ||||
10007 | // Check the value of the bit corresponding to the feature requested. | |||
10008 | Value *Mask = Builder.getInt32(Features2); | |||
10009 | Value *Bitset = Builder.CreateAnd(Features, Mask); | |||
10010 | Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask); | |||
10011 | Result = Builder.CreateAnd(Result, Cmp); | |||
10012 | } | |||
10013 | ||||
10014 | return Result; | |||
10015 | } | |||
10016 | ||||
10017 | Value *CodeGenFunction::EmitX86CpuInit() { | |||
10018 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, | |||
10019 | /*Variadic*/ false); | |||
10020 | llvm::FunctionCallee Func = | |||
10021 | CGM.CreateRuntimeFunction(FTy, "__cpu_indicator_init"); | |||
10022 | cast<llvm::GlobalValue>(Func.getCallee())->setDSOLocal(true); | |||
10023 | cast<llvm::GlobalValue>(Func.getCallee()) | |||
10024 | ->setDLLStorageClass(llvm::GlobalValue::DefaultStorageClass); | |||
10025 | return Builder.CreateCall(Func); | |||
10026 | } | |||
10027 | ||||
10028 | Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID, | |||
10029 | const CallExpr *E) { | |||
10030 | if (BuiltinID == X86::BI__builtin_cpu_is) | |||
10031 | return EmitX86CpuIs(E); | |||
10032 | if (BuiltinID == X86::BI__builtin_cpu_supports) | |||
10033 | return EmitX86CpuSupports(E); | |||
10034 | if (BuiltinID == X86::BI__builtin_cpu_init) | |||
10035 | return EmitX86CpuInit(); | |||
10036 | ||||
10037 | SmallVector<Value*, 4> Ops; | |||
10038 | ||||
10039 | // Find out if any arguments are required to be integer constant expressions. | |||
10040 | unsigned ICEArguments = 0; | |||
10041 | ASTContext::GetBuiltinTypeError Error; | |||
10042 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
10043 | 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-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10043, __PRETTY_FUNCTION__)); | |||
10044 | ||||
10045 | for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { | |||
10046 | // If this is a normal argument, just emit it as a scalar. | |||
10047 | if ((ICEArguments & (1 << i)) == 0) { | |||
10048 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
10049 | continue; | |||
10050 | } | |||
10051 | ||||
10052 | // If this is required to be a constant, constant fold it so that we know | |||
10053 | // that the generated intrinsic gets a ConstantInt. | |||
10054 | llvm::APSInt Result; | |||
10055 | bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext()); | |||
10056 | assert(IsConst && "Constant arg isn't actually constant?")((IsConst && "Constant arg isn't actually constant?") ? static_cast<void> (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10056, __PRETTY_FUNCTION__)); (void)IsConst; | |||
10057 | Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result)); | |||
10058 | } | |||
10059 | ||||
10060 | // These exist so that the builtin that takes an immediate can be bounds | |||
10061 | // checked by clang to avoid passing bad immediates to the backend. Since | |||
10062 | // AVX has a larger immediate than SSE we would need separate builtins to | |||
10063 | // do the different bounds checking. Rather than create a clang specific | |||
10064 | // SSE only builtin, this implements eight separate builtins to match gcc | |||
10065 | // implementation. | |||
10066 | auto getCmpIntrinsicCall = [this, &Ops](Intrinsic::ID ID, unsigned Imm) { | |||
10067 | Ops.push_back(llvm::ConstantInt::get(Int8Ty, Imm)); | |||
10068 | llvm::Function *F = CGM.getIntrinsic(ID); | |||
10069 | return Builder.CreateCall(F, Ops); | |||
10070 | }; | |||
10071 | ||||
10072 | // For the vector forms of FP comparisons, translate the builtins directly to | |||
10073 | // IR. | |||
10074 | // TODO: The builtins could be removed if the SSE header files used vector | |||
10075 | // extension comparisons directly (vector ordered/unordered may need | |||
10076 | // additional support via __builtin_isnan()). | |||
10077 | auto getVectorFCmpIR = [this, &Ops](CmpInst::Predicate Pred) { | |||
10078 | Value *Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]); | |||
10079 | llvm::VectorType *FPVecTy = cast<llvm::VectorType>(Ops[0]->getType()); | |||
10080 | llvm::VectorType *IntVecTy = llvm::VectorType::getInteger(FPVecTy); | |||
10081 | Value *Sext = Builder.CreateSExt(Cmp, IntVecTy); | |||
10082 | return Builder.CreateBitCast(Sext, FPVecTy); | |||
10083 | }; | |||
10084 | ||||
10085 | switch (BuiltinID) { | |||
10086 | default: return nullptr; | |||
10087 | case X86::BI_mm_prefetch: { | |||
10088 | Value *Address = Ops[0]; | |||
10089 | ConstantInt *C = cast<ConstantInt>(Ops[1]); | |||
10090 | Value *RW = ConstantInt::get(Int32Ty, (C->getZExtValue() >> 2) & 0x1); | |||
10091 | Value *Locality = ConstantInt::get(Int32Ty, C->getZExtValue() & 0x3); | |||
10092 | Value *Data = ConstantInt::get(Int32Ty, 1); | |||
10093 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); | |||
10094 | return Builder.CreateCall(F, {Address, RW, Locality, Data}); | |||
10095 | } | |||
10096 | case X86::BI_mm_clflush: { | |||
10097 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_clflush), | |||
10098 | Ops[0]); | |||
10099 | } | |||
10100 | case X86::BI_mm_lfence: { | |||
10101 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_lfence)); | |||
10102 | } | |||
10103 | case X86::BI_mm_mfence: { | |||
10104 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_mfence)); | |||
10105 | } | |||
10106 | case X86::BI_mm_sfence: { | |||
10107 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_sfence)); | |||
10108 | } | |||
10109 | case X86::BI_mm_pause: { | |||
10110 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_pause)); | |||
10111 | } | |||
10112 | case X86::BI__rdtsc: { | |||
10113 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtsc)); | |||
10114 | } | |||
10115 | case X86::BI__builtin_ia32_rdtscp: { | |||
10116 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtscp)); | |||
10117 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1), | |||
10118 | Ops[0]); | |||
10119 | return Builder.CreateExtractValue(Call, 0); | |||
10120 | } | |||
10121 | case X86::BI__builtin_ia32_lzcnt_u16: | |||
10122 | case X86::BI__builtin_ia32_lzcnt_u32: | |||
10123 | case X86::BI__builtin_ia32_lzcnt_u64: { | |||
10124 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType()); | |||
10125 | return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)}); | |||
10126 | } | |||
10127 | case X86::BI__builtin_ia32_tzcnt_u16: | |||
10128 | case X86::BI__builtin_ia32_tzcnt_u32: | |||
10129 | case X86::BI__builtin_ia32_tzcnt_u64: { | |||
10130 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, Ops[0]->getType()); | |||
10131 | return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)}); | |||
10132 | } | |||
10133 | case X86::BI__builtin_ia32_undef128: | |||
10134 | case X86::BI__builtin_ia32_undef256: | |||
10135 | case X86::BI__builtin_ia32_undef512: | |||
10136 | // The x86 definition of "undef" is not the same as the LLVM definition | |||
10137 | // (PR32176). We leave optimizing away an unnecessary zero constant to the | |||
10138 | // IR optimizer and backend. | |||
10139 | // TODO: If we had a "freeze" IR instruction to generate a fixed undef | |||
10140 | // value, we should use that here instead of a zero. | |||
10141 | return llvm::Constant::getNullValue(ConvertType(E->getType())); | |||
10142 | case X86::BI__builtin_ia32_vec_init_v8qi: | |||
10143 | case X86::BI__builtin_ia32_vec_init_v4hi: | |||
10144 | case X86::BI__builtin_ia32_vec_init_v2si: | |||
10145 | return Builder.CreateBitCast(BuildVector(Ops), | |||
10146 | llvm::Type::getX86_MMXTy(getLLVMContext())); | |||
10147 | case X86::BI__builtin_ia32_vec_ext_v2si: | |||
10148 | case X86::BI__builtin_ia32_vec_ext_v16qi: | |||
10149 | case X86::BI__builtin_ia32_vec_ext_v8hi: | |||
10150 | case X86::BI__builtin_ia32_vec_ext_v4si: | |||
10151 | case X86::BI__builtin_ia32_vec_ext_v4sf: | |||
10152 | case X86::BI__builtin_ia32_vec_ext_v2di: | |||
10153 | case X86::BI__builtin_ia32_vec_ext_v32qi: | |||
10154 | case X86::BI__builtin_ia32_vec_ext_v16hi: | |||
10155 | case X86::BI__builtin_ia32_vec_ext_v8si: | |||
10156 | case X86::BI__builtin_ia32_vec_ext_v4di: { | |||
10157 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
10158 | uint64_t Index = cast<ConstantInt>(Ops[1])->getZExtValue(); | |||
10159 | Index &= NumElts - 1; | |||
10160 | // These builtins exist so we can ensure the index is an ICE and in range. | |||
10161 | // Otherwise we could just do this in the header file. | |||
10162 | return Builder.CreateExtractElement(Ops[0], Index); | |||
10163 | } | |||
10164 | case X86::BI__builtin_ia32_vec_set_v16qi: | |||
10165 | case X86::BI__builtin_ia32_vec_set_v8hi: | |||
10166 | case X86::BI__builtin_ia32_vec_set_v4si: | |||
10167 | case X86::BI__builtin_ia32_vec_set_v2di: | |||
10168 | case X86::BI__builtin_ia32_vec_set_v32qi: | |||
10169 | case X86::BI__builtin_ia32_vec_set_v16hi: | |||
10170 | case X86::BI__builtin_ia32_vec_set_v8si: | |||
10171 | case X86::BI__builtin_ia32_vec_set_v4di: { | |||
10172 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
10173 | unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue(); | |||
10174 | Index &= NumElts - 1; | |||
10175 | // These builtins exist so we can ensure the index is an ICE and in range. | |||
10176 | // Otherwise we could just do this in the header file. | |||
10177 | return Builder.CreateInsertElement(Ops[0], Ops[1], Index); | |||
10178 | } | |||
10179 | case X86::BI_mm_setcsr: | |||
10180 | case X86::BI__builtin_ia32_ldmxcsr: { | |||
10181 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); | |||
10182 | Builder.CreateStore(Ops[0], Tmp); | |||
10183 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_ldmxcsr), | |||
10184 | Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy)); | |||
10185 | } | |||
10186 | case X86::BI_mm_getcsr: | |||
10187 | case X86::BI__builtin_ia32_stmxcsr: { | |||
10188 | Address Tmp = CreateMemTemp(E->getType()); | |||
10189 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_stmxcsr), | |||
10190 | Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy)); | |||
10191 | return Builder.CreateLoad(Tmp, "stmxcsr"); | |||
10192 | } | |||
10193 | case X86::BI__builtin_ia32_xsave: | |||
10194 | case X86::BI__builtin_ia32_xsave64: | |||
10195 | case X86::BI__builtin_ia32_xrstor: | |||
10196 | case X86::BI__builtin_ia32_xrstor64: | |||
10197 | case X86::BI__builtin_ia32_xsaveopt: | |||
10198 | case X86::BI__builtin_ia32_xsaveopt64: | |||
10199 | case X86::BI__builtin_ia32_xrstors: | |||
10200 | case X86::BI__builtin_ia32_xrstors64: | |||
10201 | case X86::BI__builtin_ia32_xsavec: | |||
10202 | case X86::BI__builtin_ia32_xsavec64: | |||
10203 | case X86::BI__builtin_ia32_xsaves: | |||
10204 | case X86::BI__builtin_ia32_xsaves64: | |||
10205 | case X86::BI__builtin_ia32_xsetbv: | |||
10206 | case X86::BI_xsetbv: { | |||
10207 | Intrinsic::ID ID; | |||
10208 | #define INTRINSIC_X86_XSAVE_ID(NAME) \ | |||
10209 | case X86::BI__builtin_ia32_##NAME: \ | |||
10210 | ID = Intrinsic::x86_##NAME; \ | |||
10211 | break | |||
10212 | switch (BuiltinID) { | |||
10213 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10213); | |||
10214 | INTRINSIC_X86_XSAVE_ID(xsave); | |||
10215 | INTRINSIC_X86_XSAVE_ID(xsave64); | |||
10216 | INTRINSIC_X86_XSAVE_ID(xrstor); | |||
10217 | INTRINSIC_X86_XSAVE_ID(xrstor64); | |||
10218 | INTRINSIC_X86_XSAVE_ID(xsaveopt); | |||
10219 | INTRINSIC_X86_XSAVE_ID(xsaveopt64); | |||
10220 | INTRINSIC_X86_XSAVE_ID(xrstors); | |||
10221 | INTRINSIC_X86_XSAVE_ID(xrstors64); | |||
10222 | INTRINSIC_X86_XSAVE_ID(xsavec); | |||
10223 | INTRINSIC_X86_XSAVE_ID(xsavec64); | |||
10224 | INTRINSIC_X86_XSAVE_ID(xsaves); | |||
10225 | INTRINSIC_X86_XSAVE_ID(xsaves64); | |||
10226 | INTRINSIC_X86_XSAVE_ID(xsetbv); | |||
10227 | case X86::BI_xsetbv: | |||
10228 | ID = Intrinsic::x86_xsetbv; | |||
10229 | break; | |||
10230 | } | |||
10231 | #undef INTRINSIC_X86_XSAVE_ID | |||
10232 | Value *Mhi = Builder.CreateTrunc( | |||
10233 | Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, 32)), Int32Ty); | |||
10234 | Value *Mlo = Builder.CreateTrunc(Ops[1], Int32Ty); | |||
10235 | Ops[1] = Mhi; | |||
10236 | Ops.push_back(Mlo); | |||
10237 | return Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
10238 | } | |||
10239 | case X86::BI__builtin_ia32_xgetbv: | |||
10240 | case X86::BI_xgetbv: | |||
10241 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_xgetbv), Ops); | |||
10242 | case X86::BI__builtin_ia32_storedqudi128_mask: | |||
10243 | case X86::BI__builtin_ia32_storedqusi128_mask: | |||
10244 | case X86::BI__builtin_ia32_storedquhi128_mask: | |||
10245 | case X86::BI__builtin_ia32_storedquqi128_mask: | |||
10246 | case X86::BI__builtin_ia32_storeupd128_mask: | |||
10247 | case X86::BI__builtin_ia32_storeups128_mask: | |||
10248 | case X86::BI__builtin_ia32_storedqudi256_mask: | |||
10249 | case X86::BI__builtin_ia32_storedqusi256_mask: | |||
10250 | case X86::BI__builtin_ia32_storedquhi256_mask: | |||
10251 | case X86::BI__builtin_ia32_storedquqi256_mask: | |||
10252 | case X86::BI__builtin_ia32_storeupd256_mask: | |||
10253 | case X86::BI__builtin_ia32_storeups256_mask: | |||
10254 | case X86::BI__builtin_ia32_storedqudi512_mask: | |||
10255 | case X86::BI__builtin_ia32_storedqusi512_mask: | |||
10256 | case X86::BI__builtin_ia32_storedquhi512_mask: | |||
10257 | case X86::BI__builtin_ia32_storedquqi512_mask: | |||
10258 | case X86::BI__builtin_ia32_storeupd512_mask: | |||
10259 | case X86::BI__builtin_ia32_storeups512_mask: | |||
10260 | return EmitX86MaskedStore(*this, Ops, 1); | |||
10261 | ||||
10262 | case X86::BI__builtin_ia32_storess128_mask: | |||
10263 | case X86::BI__builtin_ia32_storesd128_mask: { | |||
10264 | return EmitX86MaskedStore(*this, Ops, 1); | |||
10265 | } | |||
10266 | case X86::BI__builtin_ia32_vpopcntb_128: | |||
10267 | case X86::BI__builtin_ia32_vpopcntd_128: | |||
10268 | case X86::BI__builtin_ia32_vpopcntq_128: | |||
10269 | case X86::BI__builtin_ia32_vpopcntw_128: | |||
10270 | case X86::BI__builtin_ia32_vpopcntb_256: | |||
10271 | case X86::BI__builtin_ia32_vpopcntd_256: | |||
10272 | case X86::BI__builtin_ia32_vpopcntq_256: | |||
10273 | case X86::BI__builtin_ia32_vpopcntw_256: | |||
10274 | case X86::BI__builtin_ia32_vpopcntb_512: | |||
10275 | case X86::BI__builtin_ia32_vpopcntd_512: | |||
10276 | case X86::BI__builtin_ia32_vpopcntq_512: | |||
10277 | case X86::BI__builtin_ia32_vpopcntw_512: { | |||
10278 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
10279 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); | |||
10280 | return Builder.CreateCall(F, Ops); | |||
10281 | } | |||
10282 | case X86::BI__builtin_ia32_cvtmask2b128: | |||
10283 | case X86::BI__builtin_ia32_cvtmask2b256: | |||
10284 | case X86::BI__builtin_ia32_cvtmask2b512: | |||
10285 | case X86::BI__builtin_ia32_cvtmask2w128: | |||
10286 | case X86::BI__builtin_ia32_cvtmask2w256: | |||
10287 | case X86::BI__builtin_ia32_cvtmask2w512: | |||
10288 | case X86::BI__builtin_ia32_cvtmask2d128: | |||
10289 | case X86::BI__builtin_ia32_cvtmask2d256: | |||
10290 | case X86::BI__builtin_ia32_cvtmask2d512: | |||
10291 | case X86::BI__builtin_ia32_cvtmask2q128: | |||
10292 | case X86::BI__builtin_ia32_cvtmask2q256: | |||
10293 | case X86::BI__builtin_ia32_cvtmask2q512: | |||
10294 | return EmitX86SExtMask(*this, Ops[0], ConvertType(E->getType())); | |||
10295 | ||||
10296 | case X86::BI__builtin_ia32_cvtb2mask128: | |||
10297 | case X86::BI__builtin_ia32_cvtb2mask256: | |||
10298 | case X86::BI__builtin_ia32_cvtb2mask512: | |||
10299 | case X86::BI__builtin_ia32_cvtw2mask128: | |||
10300 | case X86::BI__builtin_ia32_cvtw2mask256: | |||
10301 | case X86::BI__builtin_ia32_cvtw2mask512: | |||
10302 | case X86::BI__builtin_ia32_cvtd2mask128: | |||
10303 | case X86::BI__builtin_ia32_cvtd2mask256: | |||
10304 | case X86::BI__builtin_ia32_cvtd2mask512: | |||
10305 | case X86::BI__builtin_ia32_cvtq2mask128: | |||
10306 | case X86::BI__builtin_ia32_cvtq2mask256: | |||
10307 | case X86::BI__builtin_ia32_cvtq2mask512: | |||
10308 | return EmitX86ConvertToMask(*this, Ops[0]); | |||
10309 | ||||
10310 | case X86::BI__builtin_ia32_cvtdq2ps512_mask: | |||
10311 | case X86::BI__builtin_ia32_cvtqq2ps512_mask: | |||
10312 | case X86::BI__builtin_ia32_cvtqq2pd512_mask: | |||
10313 | return EmitX86ConvertIntToFp(*this, Ops, /*IsSigned*/true); | |||
10314 | case X86::BI__builtin_ia32_cvtudq2ps512_mask: | |||
10315 | case X86::BI__builtin_ia32_cvtuqq2ps512_mask: | |||
10316 | case X86::BI__builtin_ia32_cvtuqq2pd512_mask: | |||
10317 | return EmitX86ConvertIntToFp(*this, Ops, /*IsSigned*/false); | |||
10318 | ||||
10319 | case X86::BI__builtin_ia32_vfmaddss3: | |||
10320 | case X86::BI__builtin_ia32_vfmaddsd3: | |||
10321 | case X86::BI__builtin_ia32_vfmaddss3_mask: | |||
10322 | case X86::BI__builtin_ia32_vfmaddsd3_mask: | |||
10323 | return EmitScalarFMAExpr(*this, Ops, Ops[0]); | |||
10324 | case X86::BI__builtin_ia32_vfmaddss: | |||
10325 | case X86::BI__builtin_ia32_vfmaddsd: | |||
10326 | return EmitScalarFMAExpr(*this, Ops, | |||
10327 | Constant::getNullValue(Ops[0]->getType())); | |||
10328 | case X86::BI__builtin_ia32_vfmaddss3_maskz: | |||
10329 | case X86::BI__builtin_ia32_vfmaddsd3_maskz: | |||
10330 | return EmitScalarFMAExpr(*this, Ops, Ops[0], /*ZeroMask*/true); | |||
10331 | case X86::BI__builtin_ia32_vfmaddss3_mask3: | |||
10332 | case X86::BI__builtin_ia32_vfmaddsd3_mask3: | |||
10333 | return EmitScalarFMAExpr(*this, Ops, Ops[2], /*ZeroMask*/false, 2); | |||
10334 | case X86::BI__builtin_ia32_vfmsubss3_mask3: | |||
10335 | case X86::BI__builtin_ia32_vfmsubsd3_mask3: | |||
10336 | return EmitScalarFMAExpr(*this, Ops, Ops[2], /*ZeroMask*/false, 2, | |||
10337 | /*NegAcc*/true); | |||
10338 | case X86::BI__builtin_ia32_vfmaddps: | |||
10339 | case X86::BI__builtin_ia32_vfmaddpd: | |||
10340 | case X86::BI__builtin_ia32_vfmaddps256: | |||
10341 | case X86::BI__builtin_ia32_vfmaddpd256: | |||
10342 | case X86::BI__builtin_ia32_vfmaddps512_mask: | |||
10343 | case X86::BI__builtin_ia32_vfmaddps512_maskz: | |||
10344 | case X86::BI__builtin_ia32_vfmaddps512_mask3: | |||
10345 | case X86::BI__builtin_ia32_vfmsubps512_mask3: | |||
10346 | case X86::BI__builtin_ia32_vfmaddpd512_mask: | |||
10347 | case X86::BI__builtin_ia32_vfmaddpd512_maskz: | |||
10348 | case X86::BI__builtin_ia32_vfmaddpd512_mask3: | |||
10349 | case X86::BI__builtin_ia32_vfmsubpd512_mask3: | |||
10350 | return EmitX86FMAExpr(*this, Ops, BuiltinID, /*IsAddSub*/false); | |||
10351 | case X86::BI__builtin_ia32_vfmaddsubps: | |||
10352 | case X86::BI__builtin_ia32_vfmaddsubpd: | |||
10353 | case X86::BI__builtin_ia32_vfmaddsubps256: | |||
10354 | case X86::BI__builtin_ia32_vfmaddsubpd256: | |||
10355 | case X86::BI__builtin_ia32_vfmaddsubps512_mask: | |||
10356 | case X86::BI__builtin_ia32_vfmaddsubps512_maskz: | |||
10357 | case X86::BI__builtin_ia32_vfmaddsubps512_mask3: | |||
10358 | case X86::BI__builtin_ia32_vfmsubaddps512_mask3: | |||
10359 | case X86::BI__builtin_ia32_vfmaddsubpd512_mask: | |||
10360 | case X86::BI__builtin_ia32_vfmaddsubpd512_maskz: | |||
10361 | case X86::BI__builtin_ia32_vfmaddsubpd512_mask3: | |||
10362 | case X86::BI__builtin_ia32_vfmsubaddpd512_mask3: | |||
10363 | return EmitX86FMAExpr(*this, Ops, BuiltinID, /*IsAddSub*/true); | |||
10364 | ||||
10365 | case X86::BI__builtin_ia32_movdqa32store128_mask: | |||
10366 | case X86::BI__builtin_ia32_movdqa64store128_mask: | |||
10367 | case X86::BI__builtin_ia32_storeaps128_mask: | |||
10368 | case X86::BI__builtin_ia32_storeapd128_mask: | |||
10369 | case X86::BI__builtin_ia32_movdqa32store256_mask: | |||
10370 | case X86::BI__builtin_ia32_movdqa64store256_mask: | |||
10371 | case X86::BI__builtin_ia32_storeaps256_mask: | |||
10372 | case X86::BI__builtin_ia32_storeapd256_mask: | |||
10373 | case X86::BI__builtin_ia32_movdqa32store512_mask: | |||
10374 | case X86::BI__builtin_ia32_movdqa64store512_mask: | |||
10375 | case X86::BI__builtin_ia32_storeaps512_mask: | |||
10376 | case X86::BI__builtin_ia32_storeapd512_mask: { | |||
10377 | unsigned Align = | |||
10378 | getContext().getTypeAlignInChars(E->getArg(1)->getType()).getQuantity(); | |||
10379 | return EmitX86MaskedStore(*this, Ops, Align); | |||
10380 | } | |||
10381 | case X86::BI__builtin_ia32_loadups128_mask: | |||
10382 | case X86::BI__builtin_ia32_loadups256_mask: | |||
10383 | case X86::BI__builtin_ia32_loadups512_mask: | |||
10384 | case X86::BI__builtin_ia32_loadupd128_mask: | |||
10385 | case X86::BI__builtin_ia32_loadupd256_mask: | |||
10386 | case X86::BI__builtin_ia32_loadupd512_mask: | |||
10387 | case X86::BI__builtin_ia32_loaddquqi128_mask: | |||
10388 | case X86::BI__builtin_ia32_loaddquqi256_mask: | |||
10389 | case X86::BI__builtin_ia32_loaddquqi512_mask: | |||
10390 | case X86::BI__builtin_ia32_loaddquhi128_mask: | |||
10391 | case X86::BI__builtin_ia32_loaddquhi256_mask: | |||
10392 | case X86::BI__builtin_ia32_loaddquhi512_mask: | |||
10393 | case X86::BI__builtin_ia32_loaddqusi128_mask: | |||
10394 | case X86::BI__builtin_ia32_loaddqusi256_mask: | |||
10395 | case X86::BI__builtin_ia32_loaddqusi512_mask: | |||
10396 | case X86::BI__builtin_ia32_loaddqudi128_mask: | |||
10397 | case X86::BI__builtin_ia32_loaddqudi256_mask: | |||
10398 | case X86::BI__builtin_ia32_loaddqudi512_mask: | |||
10399 | return EmitX86MaskedLoad(*this, Ops, 1); | |||
10400 | ||||
10401 | case X86::BI__builtin_ia32_loadss128_mask: | |||
10402 | case X86::BI__builtin_ia32_loadsd128_mask: | |||
10403 | return EmitX86MaskedLoad(*this, Ops, 1); | |||
10404 | ||||
10405 | case X86::BI__builtin_ia32_loadaps128_mask: | |||
10406 | case X86::BI__builtin_ia32_loadaps256_mask: | |||
10407 | case X86::BI__builtin_ia32_loadaps512_mask: | |||
10408 | case X86::BI__builtin_ia32_loadapd128_mask: | |||
10409 | case X86::BI__builtin_ia32_loadapd256_mask: | |||
10410 | case X86::BI__builtin_ia32_loadapd512_mask: | |||
10411 | case X86::BI__builtin_ia32_movdqa32load128_mask: | |||
10412 | case X86::BI__builtin_ia32_movdqa32load256_mask: | |||
10413 | case X86::BI__builtin_ia32_movdqa32load512_mask: | |||
10414 | case X86::BI__builtin_ia32_movdqa64load128_mask: | |||
10415 | case X86::BI__builtin_ia32_movdqa64load256_mask: | |||
10416 | case X86::BI__builtin_ia32_movdqa64load512_mask: { | |||
10417 | unsigned Align = | |||
10418 | getContext().getTypeAlignInChars(E->getArg(1)->getType()).getQuantity(); | |||
10419 | return EmitX86MaskedLoad(*this, Ops, Align); | |||
10420 | } | |||
10421 | ||||
10422 | case X86::BI__builtin_ia32_expandloaddf128_mask: | |||
10423 | case X86::BI__builtin_ia32_expandloaddf256_mask: | |||
10424 | case X86::BI__builtin_ia32_expandloaddf512_mask: | |||
10425 | case X86::BI__builtin_ia32_expandloadsf128_mask: | |||
10426 | case X86::BI__builtin_ia32_expandloadsf256_mask: | |||
10427 | case X86::BI__builtin_ia32_expandloadsf512_mask: | |||
10428 | case X86::BI__builtin_ia32_expandloaddi128_mask: | |||
10429 | case X86::BI__builtin_ia32_expandloaddi256_mask: | |||
10430 | case X86::BI__builtin_ia32_expandloaddi512_mask: | |||
10431 | case X86::BI__builtin_ia32_expandloadsi128_mask: | |||
10432 | case X86::BI__builtin_ia32_expandloadsi256_mask: | |||
10433 | case X86::BI__builtin_ia32_expandloadsi512_mask: | |||
10434 | case X86::BI__builtin_ia32_expandloadhi128_mask: | |||
10435 | case X86::BI__builtin_ia32_expandloadhi256_mask: | |||
10436 | case X86::BI__builtin_ia32_expandloadhi512_mask: | |||
10437 | case X86::BI__builtin_ia32_expandloadqi128_mask: | |||
10438 | case X86::BI__builtin_ia32_expandloadqi256_mask: | |||
10439 | case X86::BI__builtin_ia32_expandloadqi512_mask: | |||
10440 | return EmitX86ExpandLoad(*this, Ops); | |||
10441 | ||||
10442 | case X86::BI__builtin_ia32_compressstoredf128_mask: | |||
10443 | case X86::BI__builtin_ia32_compressstoredf256_mask: | |||
10444 | case X86::BI__builtin_ia32_compressstoredf512_mask: | |||
10445 | case X86::BI__builtin_ia32_compressstoresf128_mask: | |||
10446 | case X86::BI__builtin_ia32_compressstoresf256_mask: | |||
10447 | case X86::BI__builtin_ia32_compressstoresf512_mask: | |||
10448 | case X86::BI__builtin_ia32_compressstoredi128_mask: | |||
10449 | case X86::BI__builtin_ia32_compressstoredi256_mask: | |||
10450 | case X86::BI__builtin_ia32_compressstoredi512_mask: | |||
10451 | case X86::BI__builtin_ia32_compressstoresi128_mask: | |||
10452 | case X86::BI__builtin_ia32_compressstoresi256_mask: | |||
10453 | case X86::BI__builtin_ia32_compressstoresi512_mask: | |||
10454 | case X86::BI__builtin_ia32_compressstorehi128_mask: | |||
10455 | case X86::BI__builtin_ia32_compressstorehi256_mask: | |||
10456 | case X86::BI__builtin_ia32_compressstorehi512_mask: | |||
10457 | case X86::BI__builtin_ia32_compressstoreqi128_mask: | |||
10458 | case X86::BI__builtin_ia32_compressstoreqi256_mask: | |||
10459 | case X86::BI__builtin_ia32_compressstoreqi512_mask: | |||
10460 | return EmitX86CompressStore(*this, Ops); | |||
10461 | ||||
10462 | case X86::BI__builtin_ia32_expanddf128_mask: | |||
10463 | case X86::BI__builtin_ia32_expanddf256_mask: | |||
10464 | case X86::BI__builtin_ia32_expanddf512_mask: | |||
10465 | case X86::BI__builtin_ia32_expandsf128_mask: | |||
10466 | case X86::BI__builtin_ia32_expandsf256_mask: | |||
10467 | case X86::BI__builtin_ia32_expandsf512_mask: | |||
10468 | case X86::BI__builtin_ia32_expanddi128_mask: | |||
10469 | case X86::BI__builtin_ia32_expanddi256_mask: | |||
10470 | case X86::BI__builtin_ia32_expanddi512_mask: | |||
10471 | case X86::BI__builtin_ia32_expandsi128_mask: | |||
10472 | case X86::BI__builtin_ia32_expandsi256_mask: | |||
10473 | case X86::BI__builtin_ia32_expandsi512_mask: | |||
10474 | case X86::BI__builtin_ia32_expandhi128_mask: | |||
10475 | case X86::BI__builtin_ia32_expandhi256_mask: | |||
10476 | case X86::BI__builtin_ia32_expandhi512_mask: | |||
10477 | case X86::BI__builtin_ia32_expandqi128_mask: | |||
10478 | case X86::BI__builtin_ia32_expandqi256_mask: | |||
10479 | case X86::BI__builtin_ia32_expandqi512_mask: | |||
10480 | return EmitX86CompressExpand(*this, Ops, /*IsCompress*/false); | |||
10481 | ||||
10482 | case X86::BI__builtin_ia32_compressdf128_mask: | |||
10483 | case X86::BI__builtin_ia32_compressdf256_mask: | |||
10484 | case X86::BI__builtin_ia32_compressdf512_mask: | |||
10485 | case X86::BI__builtin_ia32_compresssf128_mask: | |||
10486 | case X86::BI__builtin_ia32_compresssf256_mask: | |||
10487 | case X86::BI__builtin_ia32_compresssf512_mask: | |||
10488 | case X86::BI__builtin_ia32_compressdi128_mask: | |||
10489 | case X86::BI__builtin_ia32_compressdi256_mask: | |||
10490 | case X86::BI__builtin_ia32_compressdi512_mask: | |||
10491 | case X86::BI__builtin_ia32_compresssi128_mask: | |||
10492 | case X86::BI__builtin_ia32_compresssi256_mask: | |||
10493 | case X86::BI__builtin_ia32_compresssi512_mask: | |||
10494 | case X86::BI__builtin_ia32_compresshi128_mask: | |||
10495 | case X86::BI__builtin_ia32_compresshi256_mask: | |||
10496 | case X86::BI__builtin_ia32_compresshi512_mask: | |||
10497 | case X86::BI__builtin_ia32_compressqi128_mask: | |||
10498 | case X86::BI__builtin_ia32_compressqi256_mask: | |||
10499 | case X86::BI__builtin_ia32_compressqi512_mask: | |||
10500 | return EmitX86CompressExpand(*this, Ops, /*IsCompress*/true); | |||
10501 | ||||
10502 | case X86::BI__builtin_ia32_gather3div2df: | |||
10503 | case X86::BI__builtin_ia32_gather3div2di: | |||
10504 | case X86::BI__builtin_ia32_gather3div4df: | |||
10505 | case X86::BI__builtin_ia32_gather3div4di: | |||
10506 | case X86::BI__builtin_ia32_gather3div4sf: | |||
10507 | case X86::BI__builtin_ia32_gather3div4si: | |||
10508 | case X86::BI__builtin_ia32_gather3div8sf: | |||
10509 | case X86::BI__builtin_ia32_gather3div8si: | |||
10510 | case X86::BI__builtin_ia32_gather3siv2df: | |||
10511 | case X86::BI__builtin_ia32_gather3siv2di: | |||
10512 | case X86::BI__builtin_ia32_gather3siv4df: | |||
10513 | case X86::BI__builtin_ia32_gather3siv4di: | |||
10514 | case X86::BI__builtin_ia32_gather3siv4sf: | |||
10515 | case X86::BI__builtin_ia32_gather3siv4si: | |||
10516 | case X86::BI__builtin_ia32_gather3siv8sf: | |||
10517 | case X86::BI__builtin_ia32_gather3siv8si: | |||
10518 | case X86::BI__builtin_ia32_gathersiv8df: | |||
10519 | case X86::BI__builtin_ia32_gathersiv16sf: | |||
10520 | case X86::BI__builtin_ia32_gatherdiv8df: | |||
10521 | case X86::BI__builtin_ia32_gatherdiv16sf: | |||
10522 | case X86::BI__builtin_ia32_gathersiv8di: | |||
10523 | case X86::BI__builtin_ia32_gathersiv16si: | |||
10524 | case X86::BI__builtin_ia32_gatherdiv8di: | |||
10525 | case X86::BI__builtin_ia32_gatherdiv16si: { | |||
10526 | Intrinsic::ID IID; | |||
10527 | switch (BuiltinID) { | |||
10528 | default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10528); | |||
10529 | case X86::BI__builtin_ia32_gather3div2df: | |||
10530 | IID = Intrinsic::x86_avx512_mask_gather3div2_df; | |||
10531 | break; | |||
10532 | case X86::BI__builtin_ia32_gather3div2di: | |||
10533 | IID = Intrinsic::x86_avx512_mask_gather3div2_di; | |||
10534 | break; | |||
10535 | case X86::BI__builtin_ia32_gather3div4df: | |||
10536 | IID = Intrinsic::x86_avx512_mask_gather3div4_df; | |||
10537 | break; | |||
10538 | case X86::BI__builtin_ia32_gather3div4di: | |||
10539 | IID = Intrinsic::x86_avx512_mask_gather3div4_di; | |||
10540 | break; | |||
10541 | case X86::BI__builtin_ia32_gather3div4sf: | |||
10542 | IID = Intrinsic::x86_avx512_mask_gather3div4_sf; | |||
10543 | break; | |||
10544 | case X86::BI__builtin_ia32_gather3div4si: | |||
10545 | IID = Intrinsic::x86_avx512_mask_gather3div4_si; | |||
10546 | break; | |||
10547 | case X86::BI__builtin_ia32_gather3div8sf: | |||
10548 | IID = Intrinsic::x86_avx512_mask_gather3div8_sf; | |||
10549 | break; | |||
10550 | case X86::BI__builtin_ia32_gather3div8si: | |||
10551 | IID = Intrinsic::x86_avx512_mask_gather3div8_si; | |||
10552 | break; | |||
10553 | case X86::BI__builtin_ia32_gather3siv2df: | |||
10554 | IID = Intrinsic::x86_avx512_mask_gather3siv2_df; | |||
10555 | break; | |||
10556 | case X86::BI__builtin_ia32_gather3siv2di: | |||
10557 | IID = Intrinsic::x86_avx512_mask_gather3siv2_di; | |||
10558 | break; | |||
10559 | case X86::BI__builtin_ia32_gather3siv4df: | |||
10560 | IID = Intrinsic::x86_avx512_mask_gather3siv4_df; | |||
10561 | break; | |||
10562 | case X86::BI__builtin_ia32_gather3siv4di: | |||
10563 | IID = Intrinsic::x86_avx512_mask_gather3siv4_di; | |||
10564 | break; | |||
10565 | case X86::BI__builtin_ia32_gather3siv4sf: | |||
10566 | IID = Intrinsic::x86_avx512_mask_gather3siv4_sf; | |||
10567 | break; | |||
10568 | case X86::BI__builtin_ia32_gather3siv4si: | |||
10569 | IID = Intrinsic::x86_avx512_mask_gather3siv4_si; | |||
10570 | break; | |||
10571 | case X86::BI__builtin_ia32_gather3siv8sf: | |||
10572 | IID = Intrinsic::x86_avx512_mask_gather3siv8_sf; | |||
10573 | break; | |||
10574 | case X86::BI__builtin_ia32_gather3siv8si: | |||
10575 | IID = Intrinsic::x86_avx512_mask_gather3siv8_si; | |||
10576 | break; | |||
10577 | case X86::BI__builtin_ia32_gathersiv8df: | |||
10578 | IID = Intrinsic::x86_avx512_mask_gather_dpd_512; | |||
10579 | break; | |||
10580 | case X86::BI__builtin_ia32_gathersiv16sf: | |||
10581 | IID = Intrinsic::x86_avx512_mask_gather_dps_512; | |||
10582 | break; | |||
10583 | case X86::BI__builtin_ia32_gatherdiv8df: | |||
10584 | IID = Intrinsic::x86_avx512_mask_gather_qpd_512; | |||
10585 | break; | |||
10586 | case X86::BI__builtin_ia32_gatherdiv16sf: | |||
10587 | IID = Intrinsic::x86_avx512_mask_gather_qps_512; | |||
10588 | break; | |||
10589 | case X86::BI__builtin_ia32_gathersiv8di: | |||
10590 | IID = Intrinsic::x86_avx512_mask_gather_dpq_512; | |||
10591 | break; | |||
10592 | case X86::BI__builtin_ia32_gathersiv16si: | |||
10593 | IID = Intrinsic::x86_avx512_mask_gather_dpi_512; | |||
10594 | break; | |||
10595 | case X86::BI__builtin_ia32_gatherdiv8di: | |||
10596 | IID = Intrinsic::x86_avx512_mask_gather_qpq_512; | |||
10597 | break; | |||
10598 | case X86::BI__builtin_ia32_gatherdiv16si: | |||
10599 | IID = Intrinsic::x86_avx512_mask_gather_qpi_512; | |||
10600 | break; | |||
10601 | } | |||
10602 | ||||
10603 | unsigned MinElts = std::min(Ops[0]->getType()->getVectorNumElements(), | |||
10604 | Ops[2]->getType()->getVectorNumElements()); | |||
10605 | Ops[3] = getMaskVecValue(*this, Ops[3], MinElts); | |||
10606 | Function *Intr = CGM.getIntrinsic(IID); | |||
10607 | return Builder.CreateCall(Intr, Ops); | |||
10608 | } | |||
10609 | ||||
10610 | case X86::BI__builtin_ia32_scattersiv8df: | |||
10611 | case X86::BI__builtin_ia32_scattersiv16sf: | |||
10612 | case X86::BI__builtin_ia32_scatterdiv8df: | |||
10613 | case X86::BI__builtin_ia32_scatterdiv16sf: | |||
10614 | case X86::BI__builtin_ia32_scattersiv8di: | |||
10615 | case X86::BI__builtin_ia32_scattersiv16si: | |||
10616 | case X86::BI__builtin_ia32_scatterdiv8di: | |||
10617 | case X86::BI__builtin_ia32_scatterdiv16si: | |||
10618 | case X86::BI__builtin_ia32_scatterdiv2df: | |||
10619 | case X86::BI__builtin_ia32_scatterdiv2di: | |||
10620 | case X86::BI__builtin_ia32_scatterdiv4df: | |||
10621 | case X86::BI__builtin_ia32_scatterdiv4di: | |||
10622 | case X86::BI__builtin_ia32_scatterdiv4sf: | |||
10623 | case X86::BI__builtin_ia32_scatterdiv4si: | |||
10624 | case X86::BI__builtin_ia32_scatterdiv8sf: | |||
10625 | case X86::BI__builtin_ia32_scatterdiv8si: | |||
10626 | case X86::BI__builtin_ia32_scattersiv2df: | |||
10627 | case X86::BI__builtin_ia32_scattersiv2di: | |||
10628 | case X86::BI__builtin_ia32_scattersiv4df: | |||
10629 | case X86::BI__builtin_ia32_scattersiv4di: | |||
10630 | case X86::BI__builtin_ia32_scattersiv4sf: | |||
10631 | case X86::BI__builtin_ia32_scattersiv4si: | |||
10632 | case X86::BI__builtin_ia32_scattersiv8sf: | |||
10633 | case X86::BI__builtin_ia32_scattersiv8si: { | |||
10634 | Intrinsic::ID IID; | |||
10635 | switch (BuiltinID) { | |||
10636 | default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10636); | |||
10637 | case X86::BI__builtin_ia32_scattersiv8df: | |||
10638 | IID = Intrinsic::x86_avx512_mask_scatter_dpd_512; | |||
10639 | break; | |||
10640 | case X86::BI__builtin_ia32_scattersiv16sf: | |||
10641 | IID = Intrinsic::x86_avx512_mask_scatter_dps_512; | |||
10642 | break; | |||
10643 | case X86::BI__builtin_ia32_scatterdiv8df: | |||
10644 | IID = Intrinsic::x86_avx512_mask_scatter_qpd_512; | |||
10645 | break; | |||
10646 | case X86::BI__builtin_ia32_scatterdiv16sf: | |||
10647 | IID = Intrinsic::x86_avx512_mask_scatter_qps_512; | |||
10648 | break; | |||
10649 | case X86::BI__builtin_ia32_scattersiv8di: | |||
10650 | IID = Intrinsic::x86_avx512_mask_scatter_dpq_512; | |||
10651 | break; | |||
10652 | case X86::BI__builtin_ia32_scattersiv16si: | |||
10653 | IID = Intrinsic::x86_avx512_mask_scatter_dpi_512; | |||
10654 | break; | |||
10655 | case X86::BI__builtin_ia32_scatterdiv8di: | |||
10656 | IID = Intrinsic::x86_avx512_mask_scatter_qpq_512; | |||
10657 | break; | |||
10658 | case X86::BI__builtin_ia32_scatterdiv16si: | |||
10659 | IID = Intrinsic::x86_avx512_mask_scatter_qpi_512; | |||
10660 | break; | |||
10661 | case X86::BI__builtin_ia32_scatterdiv2df: | |||
10662 | IID = Intrinsic::x86_avx512_mask_scatterdiv2_df; | |||
10663 | break; | |||
10664 | case X86::BI__builtin_ia32_scatterdiv2di: | |||
10665 | IID = Intrinsic::x86_avx512_mask_scatterdiv2_di; | |||
10666 | break; | |||
10667 | case X86::BI__builtin_ia32_scatterdiv4df: | |||
10668 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_df; | |||
10669 | break; | |||
10670 | case X86::BI__builtin_ia32_scatterdiv4di: | |||
10671 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_di; | |||
10672 | break; | |||
10673 | case X86::BI__builtin_ia32_scatterdiv4sf: | |||
10674 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_sf; | |||
10675 | break; | |||
10676 | case X86::BI__builtin_ia32_scatterdiv4si: | |||
10677 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_si; | |||
10678 | break; | |||
10679 | case X86::BI__builtin_ia32_scatterdiv8sf: | |||
10680 | IID = Intrinsic::x86_avx512_mask_scatterdiv8_sf; | |||
10681 | break; | |||
10682 | case X86::BI__builtin_ia32_scatterdiv8si: | |||
10683 | IID = Intrinsic::x86_avx512_mask_scatterdiv8_si; | |||
10684 | break; | |||
10685 | case X86::BI__builtin_ia32_scattersiv2df: | |||
10686 | IID = Intrinsic::x86_avx512_mask_scattersiv2_df; | |||
10687 | break; | |||
10688 | case X86::BI__builtin_ia32_scattersiv2di: | |||
10689 | IID = Intrinsic::x86_avx512_mask_scattersiv2_di; | |||
10690 | break; | |||
10691 | case X86::BI__builtin_ia32_scattersiv4df: | |||
10692 | IID = Intrinsic::x86_avx512_mask_scattersiv4_df; | |||
10693 | break; | |||
10694 | case X86::BI__builtin_ia32_scattersiv4di: | |||
10695 | IID = Intrinsic::x86_avx512_mask_scattersiv4_di; | |||
10696 | break; | |||
10697 | case X86::BI__builtin_ia32_scattersiv4sf: | |||
10698 | IID = Intrinsic::x86_avx512_mask_scattersiv4_sf; | |||
10699 | break; | |||
10700 | case X86::BI__builtin_ia32_scattersiv4si: | |||
10701 | IID = Intrinsic::x86_avx512_mask_scattersiv4_si; | |||
10702 | break; | |||
10703 | case X86::BI__builtin_ia32_scattersiv8sf: | |||
10704 | IID = Intrinsic::x86_avx512_mask_scattersiv8_sf; | |||
10705 | break; | |||
10706 | case X86::BI__builtin_ia32_scattersiv8si: | |||
10707 | IID = Intrinsic::x86_avx512_mask_scattersiv8_si; | |||
10708 | break; | |||
10709 | } | |||
10710 | ||||
10711 | unsigned MinElts = std::min(Ops[2]->getType()->getVectorNumElements(), | |||
10712 | Ops[3]->getType()->getVectorNumElements()); | |||
10713 | Ops[1] = getMaskVecValue(*this, Ops[1], MinElts); | |||
10714 | Function *Intr = CGM.getIntrinsic(IID); | |||
10715 | return Builder.CreateCall(Intr, Ops); | |||
10716 | } | |||
10717 | ||||
10718 | case X86::BI__builtin_ia32_vextractf128_pd256: | |||
10719 | case X86::BI__builtin_ia32_vextractf128_ps256: | |||
10720 | case X86::BI__builtin_ia32_vextractf128_si256: | |||
10721 | case X86::BI__builtin_ia32_extract128i256: | |||
10722 | case X86::BI__builtin_ia32_extractf64x4_mask: | |||
10723 | case X86::BI__builtin_ia32_extractf32x4_mask: | |||
10724 | case X86::BI__builtin_ia32_extracti64x4_mask: | |||
10725 | case X86::BI__builtin_ia32_extracti32x4_mask: | |||
10726 | case X86::BI__builtin_ia32_extractf32x8_mask: | |||
10727 | case X86::BI__builtin_ia32_extracti32x8_mask: | |||
10728 | case X86::BI__builtin_ia32_extractf32x4_256_mask: | |||
10729 | case X86::BI__builtin_ia32_extracti32x4_256_mask: | |||
10730 | case X86::BI__builtin_ia32_extractf64x2_256_mask: | |||
10731 | case X86::BI__builtin_ia32_extracti64x2_256_mask: | |||
10732 | case X86::BI__builtin_ia32_extractf64x2_512_mask: | |||
10733 | case X86::BI__builtin_ia32_extracti64x2_512_mask: { | |||
10734 | llvm::Type *DstTy = ConvertType(E->getType()); | |||
10735 | unsigned NumElts = DstTy->getVectorNumElements(); | |||
10736 | unsigned SrcNumElts = Ops[0]->getType()->getVectorNumElements(); | |||
10737 | unsigned SubVectors = SrcNumElts / NumElts; | |||
10738 | unsigned Index = cast<ConstantInt>(Ops[1])->getZExtValue(); | |||
10739 | assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors")((llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors" ) ? static_cast<void> (0) : __assert_fail ("llvm::isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10739, __PRETTY_FUNCTION__)); | |||
10740 | Index &= SubVectors - 1; // Remove any extra bits. | |||
10741 | Index *= NumElts; | |||
10742 | ||||
10743 | uint32_t Indices[16]; | |||
10744 | for (unsigned i = 0; i != NumElts; ++i) | |||
10745 | Indices[i] = i + Index; | |||
10746 | ||||
10747 | Value *Res = Builder.CreateShuffleVector(Ops[0], | |||
10748 | UndefValue::get(Ops[0]->getType()), | |||
10749 | makeArrayRef(Indices, NumElts), | |||
10750 | "extract"); | |||
10751 | ||||
10752 | if (Ops.size() == 4) | |||
10753 | Res = EmitX86Select(*this, Ops[3], Res, Ops[2]); | |||
10754 | ||||
10755 | return Res; | |||
10756 | } | |||
10757 | case X86::BI__builtin_ia32_vinsertf128_pd256: | |||
10758 | case X86::BI__builtin_ia32_vinsertf128_ps256: | |||
10759 | case X86::BI__builtin_ia32_vinsertf128_si256: | |||
10760 | case X86::BI__builtin_ia32_insert128i256: | |||
10761 | case X86::BI__builtin_ia32_insertf64x4: | |||
10762 | case X86::BI__builtin_ia32_insertf32x4: | |||
10763 | case X86::BI__builtin_ia32_inserti64x4: | |||
10764 | case X86::BI__builtin_ia32_inserti32x4: | |||
10765 | case X86::BI__builtin_ia32_insertf32x8: | |||
10766 | case X86::BI__builtin_ia32_inserti32x8: | |||
10767 | case X86::BI__builtin_ia32_insertf32x4_256: | |||
10768 | case X86::BI__builtin_ia32_inserti32x4_256: | |||
10769 | case X86::BI__builtin_ia32_insertf64x2_256: | |||
10770 | case X86::BI__builtin_ia32_inserti64x2_256: | |||
10771 | case X86::BI__builtin_ia32_insertf64x2_512: | |||
10772 | case X86::BI__builtin_ia32_inserti64x2_512: { | |||
10773 | unsigned DstNumElts = Ops[0]->getType()->getVectorNumElements(); | |||
10774 | unsigned SrcNumElts = Ops[1]->getType()->getVectorNumElements(); | |||
10775 | unsigned SubVectors = DstNumElts / SrcNumElts; | |||
10776 | unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue(); | |||
10777 | assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors")((llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors" ) ? static_cast<void> (0) : __assert_fail ("llvm::isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10777, __PRETTY_FUNCTION__)); | |||
10778 | Index &= SubVectors - 1; // Remove any extra bits. | |||
10779 | Index *= SrcNumElts; | |||
10780 | ||||
10781 | uint32_t Indices[16]; | |||
10782 | for (unsigned i = 0; i != DstNumElts; ++i) | |||
10783 | Indices[i] = (i >= SrcNumElts) ? SrcNumElts + (i % SrcNumElts) : i; | |||
10784 | ||||
10785 | Value *Op1 = Builder.CreateShuffleVector(Ops[1], | |||
10786 | UndefValue::get(Ops[1]->getType()), | |||
10787 | makeArrayRef(Indices, DstNumElts), | |||
10788 | "widen"); | |||
10789 | ||||
10790 | for (unsigned i = 0; i != DstNumElts; ++i) { | |||
10791 | if (i >= Index && i < (Index + SrcNumElts)) | |||
10792 | Indices[i] = (i - Index) + DstNumElts; | |||
10793 | else | |||
10794 | Indices[i] = i; | |||
10795 | } | |||
10796 | ||||
10797 | return Builder.CreateShuffleVector(Ops[0], Op1, | |||
10798 | makeArrayRef(Indices, DstNumElts), | |||
10799 | "insert"); | |||
10800 | } | |||
10801 | case X86::BI__builtin_ia32_pmovqd512_mask: | |||
10802 | case X86::BI__builtin_ia32_pmovwb512_mask: { | |||
10803 | Value *Res = Builder.CreateTrunc(Ops[0], Ops[1]->getType()); | |||
10804 | return EmitX86Select(*this, Ops[2], Res, Ops[1]); | |||
10805 | } | |||
10806 | case X86::BI__builtin_ia32_pmovdb512_mask: | |||
10807 | case X86::BI__builtin_ia32_pmovdw512_mask: | |||
10808 | case X86::BI__builtin_ia32_pmovqw512_mask: { | |||
10809 | if (const auto *C = dyn_cast<Constant>(Ops[2])) | |||
10810 | if (C->isAllOnesValue()) | |||
10811 | return Builder.CreateTrunc(Ops[0], Ops[1]->getType()); | |||
10812 | ||||
10813 | Intrinsic::ID IID; | |||
10814 | switch (BuiltinID) { | |||
10815 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10815); | |||
10816 | case X86::BI__builtin_ia32_pmovdb512_mask: | |||
10817 | IID = Intrinsic::x86_avx512_mask_pmov_db_512; | |||
10818 | break; | |||
10819 | case X86::BI__builtin_ia32_pmovdw512_mask: | |||
10820 | IID = Intrinsic::x86_avx512_mask_pmov_dw_512; | |||
10821 | break; | |||
10822 | case X86::BI__builtin_ia32_pmovqw512_mask: | |||
10823 | IID = Intrinsic::x86_avx512_mask_pmov_qw_512; | |||
10824 | break; | |||
10825 | } | |||
10826 | ||||
10827 | Function *Intr = CGM.getIntrinsic(IID); | |||
10828 | return Builder.CreateCall(Intr, Ops); | |||
10829 | } | |||
10830 | case X86::BI__builtin_ia32_pblendw128: | |||
10831 | case X86::BI__builtin_ia32_blendpd: | |||
10832 | case X86::BI__builtin_ia32_blendps: | |||
10833 | case X86::BI__builtin_ia32_blendpd256: | |||
10834 | case X86::BI__builtin_ia32_blendps256: | |||
10835 | case X86::BI__builtin_ia32_pblendw256: | |||
10836 | case X86::BI__builtin_ia32_pblendd128: | |||
10837 | case X86::BI__builtin_ia32_pblendd256: { | |||
10838 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
10839 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); | |||
10840 | ||||
10841 | uint32_t Indices[16]; | |||
10842 | // If there are more than 8 elements, the immediate is used twice so make | |||
10843 | // sure we handle that. | |||
10844 | for (unsigned i = 0; i != NumElts; ++i) | |||
10845 | Indices[i] = ((Imm >> (i % 8)) & 0x1) ? NumElts + i : i; | |||
10846 | ||||
10847 | return Builder.CreateShuffleVector(Ops[0], Ops[1], | |||
10848 | makeArrayRef(Indices, NumElts), | |||
10849 | "blend"); | |||
10850 | } | |||
10851 | case X86::BI__builtin_ia32_pshuflw: | |||
10852 | case X86::BI__builtin_ia32_pshuflw256: | |||
10853 | case X86::BI__builtin_ia32_pshuflw512: { | |||
10854 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
10855 | llvm::Type *Ty = Ops[0]->getType(); | |||
10856 | unsigned NumElts = Ty->getVectorNumElements(); | |||
10857 | ||||
10858 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. | |||
10859 | Imm = (Imm & 0xff) * 0x01010101; | |||
10860 | ||||
10861 | uint32_t Indices[32]; | |||
10862 | for (unsigned l = 0; l != NumElts; l += 8) { | |||
10863 | for (unsigned i = 0; i != 4; ++i) { | |||
10864 | Indices[l + i] = l + (Imm & 3); | |||
10865 | Imm >>= 2; | |||
10866 | } | |||
10867 | for (unsigned i = 4; i != 8; ++i) | |||
10868 | Indices[l + i] = l + i; | |||
10869 | } | |||
10870 | ||||
10871 | return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty), | |||
10872 | makeArrayRef(Indices, NumElts), | |||
10873 | "pshuflw"); | |||
10874 | } | |||
10875 | case X86::BI__builtin_ia32_pshufhw: | |||
10876 | case X86::BI__builtin_ia32_pshufhw256: | |||
10877 | case X86::BI__builtin_ia32_pshufhw512: { | |||
10878 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
10879 | llvm::Type *Ty = Ops[0]->getType(); | |||
10880 | unsigned NumElts = Ty->getVectorNumElements(); | |||
10881 | ||||
10882 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. | |||
10883 | Imm = (Imm & 0xff) * 0x01010101; | |||
10884 | ||||
10885 | uint32_t Indices[32]; | |||
10886 | for (unsigned l = 0; l != NumElts; l += 8) { | |||
10887 | for (unsigned i = 0; i != 4; ++i) | |||
10888 | Indices[l + i] = l + i; | |||
10889 | for (unsigned i = 4; i != 8; ++i) { | |||
10890 | Indices[l + i] = l + 4 + (Imm & 3); | |||
10891 | Imm >>= 2; | |||
10892 | } | |||
10893 | } | |||
10894 | ||||
10895 | return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty), | |||
10896 | makeArrayRef(Indices, NumElts), | |||
10897 | "pshufhw"); | |||
10898 | } | |||
10899 | case X86::BI__builtin_ia32_pshufd: | |||
10900 | case X86::BI__builtin_ia32_pshufd256: | |||
10901 | case X86::BI__builtin_ia32_pshufd512: | |||
10902 | case X86::BI__builtin_ia32_vpermilpd: | |||
10903 | case X86::BI__builtin_ia32_vpermilps: | |||
10904 | case X86::BI__builtin_ia32_vpermilpd256: | |||
10905 | case X86::BI__builtin_ia32_vpermilps256: | |||
10906 | case X86::BI__builtin_ia32_vpermilpd512: | |||
10907 | case X86::BI__builtin_ia32_vpermilps512: { | |||
10908 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
10909 | llvm::Type *Ty = Ops[0]->getType(); | |||
10910 | unsigned NumElts = Ty->getVectorNumElements(); | |||
10911 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128; | |||
10912 | unsigned NumLaneElts = NumElts / NumLanes; | |||
10913 | ||||
10914 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. | |||
10915 | Imm = (Imm & 0xff) * 0x01010101; | |||
10916 | ||||
10917 | uint32_t Indices[16]; | |||
10918 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { | |||
10919 | for (unsigned i = 0; i != NumLaneElts; ++i) { | |||
10920 | Indices[i + l] = (Imm % NumLaneElts) + l; | |||
10921 | Imm /= NumLaneElts; | |||
10922 | } | |||
10923 | } | |||
10924 | ||||
10925 | return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty), | |||
10926 | makeArrayRef(Indices, NumElts), | |||
10927 | "permil"); | |||
10928 | } | |||
10929 | case X86::BI__builtin_ia32_shufpd: | |||
10930 | case X86::BI__builtin_ia32_shufpd256: | |||
10931 | case X86::BI__builtin_ia32_shufpd512: | |||
10932 | case X86::BI__builtin_ia32_shufps: | |||
10933 | case X86::BI__builtin_ia32_shufps256: | |||
10934 | case X86::BI__builtin_ia32_shufps512: { | |||
10935 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); | |||
10936 | llvm::Type *Ty = Ops[0]->getType(); | |||
10937 | unsigned NumElts = Ty->getVectorNumElements(); | |||
10938 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128; | |||
10939 | unsigned NumLaneElts = NumElts / NumLanes; | |||
10940 | ||||
10941 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. | |||
10942 | Imm = (Imm & 0xff) * 0x01010101; | |||
10943 | ||||
10944 | uint32_t Indices[16]; | |||
10945 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { | |||
10946 | for (unsigned i = 0; i != NumLaneElts; ++i) { | |||
10947 | unsigned Index = Imm % NumLaneElts; | |||
10948 | Imm /= NumLaneElts; | |||
10949 | if (i >= (NumLaneElts / 2)) | |||
10950 | Index += NumElts; | |||
10951 | Indices[l + i] = l + Index; | |||
10952 | } | |||
10953 | } | |||
10954 | ||||
10955 | return Builder.CreateShuffleVector(Ops[0], Ops[1], | |||
10956 | makeArrayRef(Indices, NumElts), | |||
10957 | "shufp"); | |||
10958 | } | |||
10959 | case X86::BI__builtin_ia32_permdi256: | |||
10960 | case X86::BI__builtin_ia32_permdf256: | |||
10961 | case X86::BI__builtin_ia32_permdi512: | |||
10962 | case X86::BI__builtin_ia32_permdf512: { | |||
10963 | unsigned Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
10964 | llvm::Type *Ty = Ops[0]->getType(); | |||
10965 | unsigned NumElts = Ty->getVectorNumElements(); | |||
10966 | ||||
10967 | // These intrinsics operate on 256-bit lanes of four 64-bit elements. | |||
10968 | uint32_t Indices[8]; | |||
10969 | for (unsigned l = 0; l != NumElts; l += 4) | |||
10970 | for (unsigned i = 0; i != 4; ++i) | |||
10971 | Indices[l + i] = l + ((Imm >> (2 * i)) & 0x3); | |||
10972 | ||||
10973 | return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty), | |||
10974 | makeArrayRef(Indices, NumElts), | |||
10975 | "perm"); | |||
10976 | } | |||
10977 | case X86::BI__builtin_ia32_palignr128: | |||
10978 | case X86::BI__builtin_ia32_palignr256: | |||
10979 | case X86::BI__builtin_ia32_palignr512: { | |||
10980 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff; | |||
10981 | ||||
10982 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
10983 | assert(NumElts % 16 == 0)((NumElts % 16 == 0) ? static_cast<void> (0) : __assert_fail ("NumElts % 16 == 0", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10983, __PRETTY_FUNCTION__)); | |||
10984 | ||||
10985 | // If palignr is shifting the pair of vectors more than the size of two | |||
10986 | // lanes, emit zero. | |||
10987 | if (ShiftVal >= 32) | |||
10988 | return llvm::Constant::getNullValue(ConvertType(E->getType())); | |||
10989 | ||||
10990 | // If palignr is shifting the pair of input vectors more than one lane, | |||
10991 | // but less than two lanes, convert to shifting in zeroes. | |||
10992 | if (ShiftVal > 16) { | |||
10993 | ShiftVal -= 16; | |||
10994 | Ops[1] = Ops[0]; | |||
10995 | Ops[0] = llvm::Constant::getNullValue(Ops[0]->getType()); | |||
10996 | } | |||
10997 | ||||
10998 | uint32_t Indices[64]; | |||
10999 | // 256-bit palignr operates on 128-bit lanes so we need to handle that | |||
11000 | for (unsigned l = 0; l != NumElts; l += 16) { | |||
11001 | for (unsigned i = 0; i != 16; ++i) { | |||
11002 | unsigned Idx = ShiftVal + i; | |||
11003 | if (Idx >= 16) | |||
11004 | Idx += NumElts - 16; // End of lane, switch operand. | |||
11005 | Indices[l + i] = Idx + l; | |||
11006 | } | |||
11007 | } | |||
11008 | ||||
11009 | return Builder.CreateShuffleVector(Ops[1], Ops[0], | |||
11010 | makeArrayRef(Indices, NumElts), | |||
11011 | "palignr"); | |||
11012 | } | |||
11013 | case X86::BI__builtin_ia32_alignd128: | |||
11014 | case X86::BI__builtin_ia32_alignd256: | |||
11015 | case X86::BI__builtin_ia32_alignd512: | |||
11016 | case X86::BI__builtin_ia32_alignq128: | |||
11017 | case X86::BI__builtin_ia32_alignq256: | |||
11018 | case X86::BI__builtin_ia32_alignq512: { | |||
11019 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
11020 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff; | |||
11021 | ||||
11022 | // Mask the shift amount to width of two vectors. | |||
11023 | ShiftVal &= (2 * NumElts) - 1; | |||
11024 | ||||
11025 | uint32_t Indices[16]; | |||
11026 | for (unsigned i = 0; i != NumElts; ++i) | |||
11027 | Indices[i] = i + ShiftVal; | |||
11028 | ||||
11029 | return Builder.CreateShuffleVector(Ops[1], Ops[0], | |||
11030 | makeArrayRef(Indices, NumElts), | |||
11031 | "valign"); | |||
11032 | } | |||
11033 | case X86::BI__builtin_ia32_shuf_f32x4_256: | |||
11034 | case X86::BI__builtin_ia32_shuf_f64x2_256: | |||
11035 | case X86::BI__builtin_ia32_shuf_i32x4_256: | |||
11036 | case X86::BI__builtin_ia32_shuf_i64x2_256: | |||
11037 | case X86::BI__builtin_ia32_shuf_f32x4: | |||
11038 | case X86::BI__builtin_ia32_shuf_f64x2: | |||
11039 | case X86::BI__builtin_ia32_shuf_i32x4: | |||
11040 | case X86::BI__builtin_ia32_shuf_i64x2: { | |||
11041 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); | |||
11042 | llvm::Type *Ty = Ops[0]->getType(); | |||
11043 | unsigned NumElts = Ty->getVectorNumElements(); | |||
11044 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() == 512 ? 4 : 2; | |||
11045 | unsigned NumLaneElts = NumElts / NumLanes; | |||
11046 | ||||
11047 | uint32_t Indices[16]; | |||
11048 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { | |||
11049 | unsigned Index = (Imm % NumLanes) * NumLaneElts; | |||
11050 | Imm /= NumLanes; // Discard the bits we just used. | |||
11051 | if (l >= (NumElts / 2)) | |||
11052 | Index += NumElts; // Switch to other source. | |||
11053 | for (unsigned i = 0; i != NumLaneElts; ++i) { | |||
11054 | Indices[l + i] = Index + i; | |||
11055 | } | |||
11056 | } | |||
11057 | ||||
11058 | return Builder.CreateShuffleVector(Ops[0], Ops[1], | |||
11059 | makeArrayRef(Indices, NumElts), | |||
11060 | "shuf"); | |||
11061 | } | |||
11062 | ||||
11063 | case X86::BI__builtin_ia32_vperm2f128_pd256: | |||
11064 | case X86::BI__builtin_ia32_vperm2f128_ps256: | |||
11065 | case X86::BI__builtin_ia32_vperm2f128_si256: | |||
11066 | case X86::BI__builtin_ia32_permti256: { | |||
11067 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); | |||
11068 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
11069 | ||||
11070 | // This takes a very simple approach since there are two lanes and a | |||
11071 | // shuffle can have 2 inputs. So we reserve the first input for the first | |||
11072 | // lane and the second input for the second lane. This may result in | |||
11073 | // duplicate sources, but this can be dealt with in the backend. | |||
11074 | ||||
11075 | Value *OutOps[2]; | |||
11076 | uint32_t Indices[8]; | |||
11077 | for (unsigned l = 0; l != 2; ++l) { | |||
11078 | // Determine the source for this lane. | |||
11079 | if (Imm & (1 << ((l * 4) + 3))) | |||
11080 | OutOps[l] = llvm::ConstantAggregateZero::get(Ops[0]->getType()); | |||
11081 | else if (Imm & (1 << ((l * 4) + 1))) | |||
11082 | OutOps[l] = Ops[1]; | |||
11083 | else | |||
11084 | OutOps[l] = Ops[0]; | |||
11085 | ||||
11086 | for (unsigned i = 0; i != NumElts/2; ++i) { | |||
11087 | // Start with ith element of the source for this lane. | |||
11088 | unsigned Idx = (l * NumElts) + i; | |||
11089 | // If bit 0 of the immediate half is set, switch to the high half of | |||
11090 | // the source. | |||
11091 | if (Imm & (1 << (l * 4))) | |||
11092 | Idx += NumElts/2; | |||
11093 | Indices[(l * (NumElts/2)) + i] = Idx; | |||
11094 | } | |||
11095 | } | |||
11096 | ||||
11097 | return Builder.CreateShuffleVector(OutOps[0], OutOps[1], | |||
11098 | makeArrayRef(Indices, NumElts), | |||
11099 | "vperm"); | |||
11100 | } | |||
11101 | ||||
11102 | case X86::BI__builtin_ia32_pslldqi128_byteshift: | |||
11103 | case X86::BI__builtin_ia32_pslldqi256_byteshift: | |||
11104 | case X86::BI__builtin_ia32_pslldqi512_byteshift: { | |||
11105 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; | |||
11106 | llvm::Type *ResultType = Ops[0]->getType(); | |||
11107 | // Builtin type is vXi64 so multiply by 8 to get bytes. | |||
11108 | unsigned NumElts = ResultType->getVectorNumElements() * 8; | |||
11109 | ||||
11110 | // If pslldq is shifting the vector more than 15 bytes, emit zero. | |||
11111 | if (ShiftVal >= 16) | |||
11112 | return llvm::Constant::getNullValue(ResultType); | |||
11113 | ||||
11114 | uint32_t Indices[64]; | |||
11115 | // 256/512-bit pslldq operates on 128-bit lanes so we need to handle that | |||
11116 | for (unsigned l = 0; l != NumElts; l += 16) { | |||
11117 | for (unsigned i = 0; i != 16; ++i) { | |||
11118 | unsigned Idx = NumElts + i - ShiftVal; | |||
11119 | if (Idx < NumElts) Idx -= NumElts - 16; // end of lane, switch operand. | |||
11120 | Indices[l + i] = Idx + l; | |||
11121 | } | |||
11122 | } | |||
11123 | ||||
11124 | llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, NumElts); | |||
11125 | Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast"); | |||
11126 | Value *Zero = llvm::Constant::getNullValue(VecTy); | |||
11127 | Value *SV = Builder.CreateShuffleVector(Zero, Cast, | |||
11128 | makeArrayRef(Indices, NumElts), | |||
11129 | "pslldq"); | |||
11130 | return Builder.CreateBitCast(SV, Ops[0]->getType(), "cast"); | |||
11131 | } | |||
11132 | case X86::BI__builtin_ia32_psrldqi128_byteshift: | |||
11133 | case X86::BI__builtin_ia32_psrldqi256_byteshift: | |||
11134 | case X86::BI__builtin_ia32_psrldqi512_byteshift: { | |||
11135 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; | |||
11136 | llvm::Type *ResultType = Ops[0]->getType(); | |||
11137 | // Builtin type is vXi64 so multiply by 8 to get bytes. | |||
11138 | unsigned NumElts = ResultType->getVectorNumElements() * 8; | |||
11139 | ||||
11140 | // If psrldq is shifting the vector more than 15 bytes, emit zero. | |||
11141 | if (ShiftVal >= 16) | |||
11142 | return llvm::Constant::getNullValue(ResultType); | |||
11143 | ||||
11144 | uint32_t Indices[64]; | |||
11145 | // 256/512-bit psrldq operates on 128-bit lanes so we need to handle that | |||
11146 | for (unsigned l = 0; l != NumElts; l += 16) { | |||
11147 | for (unsigned i = 0; i != 16; ++i) { | |||
11148 | unsigned Idx = i + ShiftVal; | |||
11149 | if (Idx >= 16) Idx += NumElts - 16; // end of lane, switch operand. | |||
11150 | Indices[l + i] = Idx + l; | |||
11151 | } | |||
11152 | } | |||
11153 | ||||
11154 | llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, NumElts); | |||
11155 | Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast"); | |||
11156 | Value *Zero = llvm::Constant::getNullValue(VecTy); | |||
11157 | Value *SV = Builder.CreateShuffleVector(Cast, Zero, | |||
11158 | makeArrayRef(Indices, NumElts), | |||
11159 | "psrldq"); | |||
11160 | return Builder.CreateBitCast(SV, ResultType, "cast"); | |||
11161 | } | |||
11162 | case X86::BI__builtin_ia32_kshiftliqi: | |||
11163 | case X86::BI__builtin_ia32_kshiftlihi: | |||
11164 | case X86::BI__builtin_ia32_kshiftlisi: | |||
11165 | case X86::BI__builtin_ia32_kshiftlidi: { | |||
11166 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; | |||
11167 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
11168 | ||||
11169 | if (ShiftVal >= NumElts) | |||
11170 | return llvm::Constant::getNullValue(Ops[0]->getType()); | |||
11171 | ||||
11172 | Value *In = getMaskVecValue(*this, Ops[0], NumElts); | |||
11173 | ||||
11174 | uint32_t Indices[64]; | |||
11175 | for (unsigned i = 0; i != NumElts; ++i) | |||
11176 | Indices[i] = NumElts + i - ShiftVal; | |||
11177 | ||||
11178 | Value *Zero = llvm::Constant::getNullValue(In->getType()); | |||
11179 | Value *SV = Builder.CreateShuffleVector(Zero, In, | |||
11180 | makeArrayRef(Indices, NumElts), | |||
11181 | "kshiftl"); | |||
11182 | return Builder.CreateBitCast(SV, Ops[0]->getType()); | |||
11183 | } | |||
11184 | case X86::BI__builtin_ia32_kshiftriqi: | |||
11185 | case X86::BI__builtin_ia32_kshiftrihi: | |||
11186 | case X86::BI__builtin_ia32_kshiftrisi: | |||
11187 | case X86::BI__builtin_ia32_kshiftridi: { | |||
11188 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; | |||
11189 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
11190 | ||||
11191 | if (ShiftVal >= NumElts) | |||
11192 | return llvm::Constant::getNullValue(Ops[0]->getType()); | |||
11193 | ||||
11194 | Value *In = getMaskVecValue(*this, Ops[0], NumElts); | |||
11195 | ||||
11196 | uint32_t Indices[64]; | |||
11197 | for (unsigned i = 0; i != NumElts; ++i) | |||
11198 | Indices[i] = i + ShiftVal; | |||
11199 | ||||
11200 | Value *Zero = llvm::Constant::getNullValue(In->getType()); | |||
11201 | Value *SV = Builder.CreateShuffleVector(In, Zero, | |||
11202 | makeArrayRef(Indices, NumElts), | |||
11203 | "kshiftr"); | |||
11204 | return Builder.CreateBitCast(SV, Ops[0]->getType()); | |||
11205 | } | |||
11206 | case X86::BI__builtin_ia32_movnti: | |||
11207 | case X86::BI__builtin_ia32_movnti64: | |||
11208 | case X86::BI__builtin_ia32_movntsd: | |||
11209 | case X86::BI__builtin_ia32_movntss: { | |||
11210 | llvm::MDNode *Node = llvm::MDNode::get( | |||
11211 | getLLVMContext(), llvm::ConstantAsMetadata::get(Builder.getInt32(1))); | |||
11212 | ||||
11213 | Value *Ptr = Ops[0]; | |||
11214 | Value *Src = Ops[1]; | |||
11215 | ||||
11216 | // Extract the 0'th element of the source vector. | |||
11217 | if (BuiltinID == X86::BI__builtin_ia32_movntsd || | |||
11218 | BuiltinID == X86::BI__builtin_ia32_movntss) | |||
11219 | Src = Builder.CreateExtractElement(Src, (uint64_t)0, "extract"); | |||
11220 | ||||
11221 | // Convert the type of the pointer to a pointer to the stored type. | |||
11222 | Value *BC = Builder.CreateBitCast( | |||
11223 | Ptr, llvm::PointerType::getUnqual(Src->getType()), "cast"); | |||
11224 | ||||
11225 | // Unaligned nontemporal store of the scalar value. | |||
11226 | StoreInst *SI = Builder.CreateDefaultAlignedStore(Src, BC); | |||
11227 | SI->setMetadata(CGM.getModule().getMDKindID("nontemporal"), Node); | |||
11228 | SI->setAlignment(llvm::Align::None()); | |||
11229 | return SI; | |||
11230 | } | |||
11231 | // Rotate is a special case of funnel shift - 1st 2 args are the same. | |||
11232 | case X86::BI__builtin_ia32_vprotb: | |||
11233 | case X86::BI__builtin_ia32_vprotw: | |||
11234 | case X86::BI__builtin_ia32_vprotd: | |||
11235 | case X86::BI__builtin_ia32_vprotq: | |||
11236 | case X86::BI__builtin_ia32_vprotbi: | |||
11237 | case X86::BI__builtin_ia32_vprotwi: | |||
11238 | case X86::BI__builtin_ia32_vprotdi: | |||
11239 | case X86::BI__builtin_ia32_vprotqi: | |||
11240 | case X86::BI__builtin_ia32_prold128: | |||
11241 | case X86::BI__builtin_ia32_prold256: | |||
11242 | case X86::BI__builtin_ia32_prold512: | |||
11243 | case X86::BI__builtin_ia32_prolq128: | |||
11244 | case X86::BI__builtin_ia32_prolq256: | |||
11245 | case X86::BI__builtin_ia32_prolq512: | |||
11246 | case X86::BI__builtin_ia32_prolvd128: | |||
11247 | case X86::BI__builtin_ia32_prolvd256: | |||
11248 | case X86::BI__builtin_ia32_prolvd512: | |||
11249 | case X86::BI__builtin_ia32_prolvq128: | |||
11250 | case X86::BI__builtin_ia32_prolvq256: | |||
11251 | case X86::BI__builtin_ia32_prolvq512: | |||
11252 | return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], false); | |||
11253 | case X86::BI__builtin_ia32_prord128: | |||
11254 | case X86::BI__builtin_ia32_prord256: | |||
11255 | case X86::BI__builtin_ia32_prord512: | |||
11256 | case X86::BI__builtin_ia32_prorq128: | |||
11257 | case X86::BI__builtin_ia32_prorq256: | |||
11258 | case X86::BI__builtin_ia32_prorq512: | |||
11259 | case X86::BI__builtin_ia32_prorvd128: | |||
11260 | case X86::BI__builtin_ia32_prorvd256: | |||
11261 | case X86::BI__builtin_ia32_prorvd512: | |||
11262 | case X86::BI__builtin_ia32_prorvq128: | |||
11263 | case X86::BI__builtin_ia32_prorvq256: | |||
11264 | case X86::BI__builtin_ia32_prorvq512: | |||
11265 | return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], true); | |||
11266 | case X86::BI__builtin_ia32_selectb_128: | |||
11267 | case X86::BI__builtin_ia32_selectb_256: | |||
11268 | case X86::BI__builtin_ia32_selectb_512: | |||
11269 | case X86::BI__builtin_ia32_selectw_128: | |||
11270 | case X86::BI__builtin_ia32_selectw_256: | |||
11271 | case X86::BI__builtin_ia32_selectw_512: | |||
11272 | case X86::BI__builtin_ia32_selectd_128: | |||
11273 | case X86::BI__builtin_ia32_selectd_256: | |||
11274 | case X86::BI__builtin_ia32_selectd_512: | |||
11275 | case X86::BI__builtin_ia32_selectq_128: | |||
11276 | case X86::BI__builtin_ia32_selectq_256: | |||
11277 | case X86::BI__builtin_ia32_selectq_512: | |||
11278 | case X86::BI__builtin_ia32_selectps_128: | |||
11279 | case X86::BI__builtin_ia32_selectps_256: | |||
11280 | case X86::BI__builtin_ia32_selectps_512: | |||
11281 | case X86::BI__builtin_ia32_selectpd_128: | |||
11282 | case X86::BI__builtin_ia32_selectpd_256: | |||
11283 | case X86::BI__builtin_ia32_selectpd_512: | |||
11284 | return EmitX86Select(*this, Ops[0], Ops[1], Ops[2]); | |||
11285 | case X86::BI__builtin_ia32_selectss_128: | |||
11286 | case X86::BI__builtin_ia32_selectsd_128: { | |||
11287 | Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0); | |||
11288 | Value *B = Builder.CreateExtractElement(Ops[2], (uint64_t)0); | |||
11289 | A = EmitX86ScalarSelect(*this, Ops[0], A, B); | |||
11290 | return Builder.CreateInsertElement(Ops[1], A, (uint64_t)0); | |||
11291 | } | |||
11292 | case X86::BI__builtin_ia32_cmpb128_mask: | |||
11293 | case X86::BI__builtin_ia32_cmpb256_mask: | |||
11294 | case X86::BI__builtin_ia32_cmpb512_mask: | |||
11295 | case X86::BI__builtin_ia32_cmpw128_mask: | |||
11296 | case X86::BI__builtin_ia32_cmpw256_mask: | |||
11297 | case X86::BI__builtin_ia32_cmpw512_mask: | |||
11298 | case X86::BI__builtin_ia32_cmpd128_mask: | |||
11299 | case X86::BI__builtin_ia32_cmpd256_mask: | |||
11300 | case X86::BI__builtin_ia32_cmpd512_mask: | |||
11301 | case X86::BI__builtin_ia32_cmpq128_mask: | |||
11302 | case X86::BI__builtin_ia32_cmpq256_mask: | |||
11303 | case X86::BI__builtin_ia32_cmpq512_mask: { | |||
11304 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; | |||
11305 | return EmitX86MaskedCompare(*this, CC, true, Ops); | |||
11306 | } | |||
11307 | case X86::BI__builtin_ia32_ucmpb128_mask: | |||
11308 | case X86::BI__builtin_ia32_ucmpb256_mask: | |||
11309 | case X86::BI__builtin_ia32_ucmpb512_mask: | |||
11310 | case X86::BI__builtin_ia32_ucmpw128_mask: | |||
11311 | case X86::BI__builtin_ia32_ucmpw256_mask: | |||
11312 | case X86::BI__builtin_ia32_ucmpw512_mask: | |||
11313 | case X86::BI__builtin_ia32_ucmpd128_mask: | |||
11314 | case X86::BI__builtin_ia32_ucmpd256_mask: | |||
11315 | case X86::BI__builtin_ia32_ucmpd512_mask: | |||
11316 | case X86::BI__builtin_ia32_ucmpq128_mask: | |||
11317 | case X86::BI__builtin_ia32_ucmpq256_mask: | |||
11318 | case X86::BI__builtin_ia32_ucmpq512_mask: { | |||
11319 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; | |||
11320 | return EmitX86MaskedCompare(*this, CC, false, Ops); | |||
11321 | } | |||
11322 | case X86::BI__builtin_ia32_vpcomb: | |||
11323 | case X86::BI__builtin_ia32_vpcomw: | |||
11324 | case X86::BI__builtin_ia32_vpcomd: | |||
11325 | case X86::BI__builtin_ia32_vpcomq: | |||
11326 | return EmitX86vpcom(*this, Ops, true); | |||
11327 | case X86::BI__builtin_ia32_vpcomub: | |||
11328 | case X86::BI__builtin_ia32_vpcomuw: | |||
11329 | case X86::BI__builtin_ia32_vpcomud: | |||
11330 | case X86::BI__builtin_ia32_vpcomuq: | |||
11331 | return EmitX86vpcom(*this, Ops, false); | |||
11332 | ||||
11333 | case X86::BI__builtin_ia32_kortestcqi: | |||
11334 | case X86::BI__builtin_ia32_kortestchi: | |||
11335 | case X86::BI__builtin_ia32_kortestcsi: | |||
11336 | case X86::BI__builtin_ia32_kortestcdi: { | |||
11337 | Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops); | |||
11338 | Value *C = llvm::Constant::getAllOnesValue(Ops[0]->getType()); | |||
11339 | Value *Cmp = Builder.CreateICmpEQ(Or, C); | |||
11340 | return Builder.CreateZExt(Cmp, ConvertType(E->getType())); | |||
11341 | } | |||
11342 | case X86::BI__builtin_ia32_kortestzqi: | |||
11343 | case X86::BI__builtin_ia32_kortestzhi: | |||
11344 | case X86::BI__builtin_ia32_kortestzsi: | |||
11345 | case X86::BI__builtin_ia32_kortestzdi: { | |||
11346 | Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops); | |||
11347 | Value *C = llvm::Constant::getNullValue(Ops[0]->getType()); | |||
11348 | Value *Cmp = Builder.CreateICmpEQ(Or, C); | |||
11349 | return Builder.CreateZExt(Cmp, ConvertType(E->getType())); | |||
11350 | } | |||
11351 | ||||
11352 | case X86::BI__builtin_ia32_ktestcqi: | |||
11353 | case X86::BI__builtin_ia32_ktestzqi: | |||
11354 | case X86::BI__builtin_ia32_ktestchi: | |||
11355 | case X86::BI__builtin_ia32_ktestzhi: | |||
11356 | case X86::BI__builtin_ia32_ktestcsi: | |||
11357 | case X86::BI__builtin_ia32_ktestzsi: | |||
11358 | case X86::BI__builtin_ia32_ktestcdi: | |||
11359 | case X86::BI__builtin_ia32_ktestzdi: { | |||
11360 | Intrinsic::ID IID; | |||
11361 | switch (BuiltinID) { | |||
11362 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11362); | |||
11363 | case X86::BI__builtin_ia32_ktestcqi: | |||
11364 | IID = Intrinsic::x86_avx512_ktestc_b; | |||
11365 | break; | |||
11366 | case X86::BI__builtin_ia32_ktestzqi: | |||
11367 | IID = Intrinsic::x86_avx512_ktestz_b; | |||
11368 | break; | |||
11369 | case X86::BI__builtin_ia32_ktestchi: | |||
11370 | IID = Intrinsic::x86_avx512_ktestc_w; | |||
11371 | break; | |||
11372 | case X86::BI__builtin_ia32_ktestzhi: | |||
11373 | IID = Intrinsic::x86_avx512_ktestz_w; | |||
11374 | break; | |||
11375 | case X86::BI__builtin_ia32_ktestcsi: | |||
11376 | IID = Intrinsic::x86_avx512_ktestc_d; | |||
11377 | break; | |||
11378 | case X86::BI__builtin_ia32_ktestzsi: | |||
11379 | IID = Intrinsic::x86_avx512_ktestz_d; | |||
11380 | break; | |||
11381 | case X86::BI__builtin_ia32_ktestcdi: | |||
11382 | IID = Intrinsic::x86_avx512_ktestc_q; | |||
11383 | break; | |||
11384 | case X86::BI__builtin_ia32_ktestzdi: | |||
11385 | IID = Intrinsic::x86_avx512_ktestz_q; | |||
11386 | break; | |||
11387 | } | |||
11388 | ||||
11389 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
11390 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); | |||
11391 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); | |||
11392 | Function *Intr = CGM.getIntrinsic(IID); | |||
11393 | return Builder.CreateCall(Intr, {LHS, RHS}); | |||
11394 | } | |||
11395 | ||||
11396 | case X86::BI__builtin_ia32_kaddqi: | |||
11397 | case X86::BI__builtin_ia32_kaddhi: | |||
11398 | case X86::BI__builtin_ia32_kaddsi: | |||
11399 | case X86::BI__builtin_ia32_kadddi: { | |||
11400 | Intrinsic::ID IID; | |||
11401 | switch (BuiltinID) { | |||
11402 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11402); | |||
11403 | case X86::BI__builtin_ia32_kaddqi: | |||
11404 | IID = Intrinsic::x86_avx512_kadd_b; | |||
11405 | break; | |||
11406 | case X86::BI__builtin_ia32_kaddhi: | |||
11407 | IID = Intrinsic::x86_avx512_kadd_w; | |||
11408 | break; | |||
11409 | case X86::BI__builtin_ia32_kaddsi: | |||
11410 | IID = Intrinsic::x86_avx512_kadd_d; | |||
11411 | break; | |||
11412 | case X86::BI__builtin_ia32_kadddi: | |||
11413 | IID = Intrinsic::x86_avx512_kadd_q; | |||
11414 | break; | |||
11415 | } | |||
11416 | ||||
11417 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
11418 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); | |||
11419 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); | |||
11420 | Function *Intr = CGM.getIntrinsic(IID); | |||
11421 | Value *Res = Builder.CreateCall(Intr, {LHS, RHS}); | |||
11422 | return Builder.CreateBitCast(Res, Ops[0]->getType()); | |||
11423 | } | |||
11424 | case X86::BI__builtin_ia32_kandqi: | |||
11425 | case X86::BI__builtin_ia32_kandhi: | |||
11426 | case X86::BI__builtin_ia32_kandsi: | |||
11427 | case X86::BI__builtin_ia32_kanddi: | |||
11428 | return EmitX86MaskLogic(*this, Instruction::And, Ops); | |||
11429 | case X86::BI__builtin_ia32_kandnqi: | |||
11430 | case X86::BI__builtin_ia32_kandnhi: | |||
11431 | case X86::BI__builtin_ia32_kandnsi: | |||
11432 | case X86::BI__builtin_ia32_kandndi: | |||
11433 | return EmitX86MaskLogic(*this, Instruction::And, Ops, true); | |||
11434 | case X86::BI__builtin_ia32_korqi: | |||
11435 | case X86::BI__builtin_ia32_korhi: | |||
11436 | case X86::BI__builtin_ia32_korsi: | |||
11437 | case X86::BI__builtin_ia32_kordi: | |||
11438 | return EmitX86MaskLogic(*this, Instruction::Or, Ops); | |||
11439 | case X86::BI__builtin_ia32_kxnorqi: | |||
11440 | case X86::BI__builtin_ia32_kxnorhi: | |||
11441 | case X86::BI__builtin_ia32_kxnorsi: | |||
11442 | case X86::BI__builtin_ia32_kxnordi: | |||
11443 | return EmitX86MaskLogic(*this, Instruction::Xor, Ops, true); | |||
11444 | case X86::BI__builtin_ia32_kxorqi: | |||
11445 | case X86::BI__builtin_ia32_kxorhi: | |||
11446 | case X86::BI__builtin_ia32_kxorsi: | |||
11447 | case X86::BI__builtin_ia32_kxordi: | |||
11448 | return EmitX86MaskLogic(*this, Instruction::Xor, Ops); | |||
11449 | case X86::BI__builtin_ia32_knotqi: | |||
11450 | case X86::BI__builtin_ia32_knothi: | |||
11451 | case X86::BI__builtin_ia32_knotsi: | |||
11452 | case X86::BI__builtin_ia32_knotdi: { | |||
11453 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
11454 | Value *Res = getMaskVecValue(*this, Ops[0], NumElts); | |||
11455 | return Builder.CreateBitCast(Builder.CreateNot(Res), | |||
11456 | Ops[0]->getType()); | |||
11457 | } | |||
11458 | case X86::BI__builtin_ia32_kmovb: | |||
11459 | case X86::BI__builtin_ia32_kmovw: | |||
11460 | case X86::BI__builtin_ia32_kmovd: | |||
11461 | case X86::BI__builtin_ia32_kmovq: { | |||
11462 | // Bitcast to vXi1 type and then back to integer. This gets the mask | |||
11463 | // register type into the IR, but might be optimized out depending on | |||
11464 | // what's around it. | |||
11465 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
11466 | Value *Res = getMaskVecValue(*this, Ops[0], NumElts); | |||
11467 | return Builder.CreateBitCast(Res, Ops[0]->getType()); | |||
11468 | } | |||
11469 | ||||
11470 | case X86::BI__builtin_ia32_kunpckdi: | |||
11471 | case X86::BI__builtin_ia32_kunpcksi: | |||
11472 | case X86::BI__builtin_ia32_kunpckhi: { | |||
11473 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
11474 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); | |||
11475 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); | |||
11476 | uint32_t Indices[64]; | |||
11477 | for (unsigned i = 0; i != NumElts; ++i) | |||
11478 | Indices[i] = i; | |||
11479 | ||||
11480 | // First extract half of each vector. This gives better codegen than | |||
11481 | // doing it in a single shuffle. | |||
11482 | LHS = Builder.CreateShuffleVector(LHS, LHS, | |||
11483 | makeArrayRef(Indices, NumElts / 2)); | |||
11484 | RHS = Builder.CreateShuffleVector(RHS, RHS, | |||
11485 | makeArrayRef(Indices, NumElts / 2)); | |||
11486 | // Concat the vectors. | |||
11487 | // NOTE: Operands are swapped to match the intrinsic definition. | |||
11488 | Value *Res = Builder.CreateShuffleVector(RHS, LHS, | |||
11489 | makeArrayRef(Indices, NumElts)); | |||
11490 | return Builder.CreateBitCast(Res, Ops[0]->getType()); | |||
11491 | } | |||
11492 | ||||
11493 | case X86::BI__builtin_ia32_vplzcntd_128: | |||
11494 | case X86::BI__builtin_ia32_vplzcntd_256: | |||
11495 | case X86::BI__builtin_ia32_vplzcntd_512: | |||
11496 | case X86::BI__builtin_ia32_vplzcntq_128: | |||
11497 | case X86::BI__builtin_ia32_vplzcntq_256: | |||
11498 | case X86::BI__builtin_ia32_vplzcntq_512: { | |||
11499 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType()); | |||
11500 | return Builder.CreateCall(F, {Ops[0],Builder.getInt1(false)}); | |||
11501 | } | |||
11502 | case X86::BI__builtin_ia32_sqrtss: | |||
11503 | case X86::BI__builtin_ia32_sqrtsd: { | |||
11504 | Value *A = Builder.CreateExtractElement(Ops[0], (uint64_t)0); | |||
11505 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType()); | |||
11506 | A = Builder.CreateCall(F, {A}); | |||
11507 | return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0); | |||
11508 | } | |||
11509 | case X86::BI__builtin_ia32_sqrtsd_round_mask: | |||
11510 | case X86::BI__builtin_ia32_sqrtss_round_mask: { | |||
11511 | unsigned CC = cast<llvm::ConstantInt>(Ops[4])->getZExtValue(); | |||
11512 | // Support only if the rounding mode is 4 (AKA CUR_DIRECTION), | |||
11513 | // otherwise keep the intrinsic. | |||
11514 | if (CC != 4) { | |||
11515 | Intrinsic::ID IID = BuiltinID == X86::BI__builtin_ia32_sqrtsd_round_mask ? | |||
11516 | Intrinsic::x86_avx512_mask_sqrt_sd : | |||
11517 | Intrinsic::x86_avx512_mask_sqrt_ss; | |||
11518 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
11519 | } | |||
11520 | Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0); | |||
11521 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType()); | |||
11522 | A = Builder.CreateCall(F, A); | |||
11523 | Value *Src = Builder.CreateExtractElement(Ops[2], (uint64_t)0); | |||
11524 | A = EmitX86ScalarSelect(*this, Ops[3], A, Src); | |||
11525 | return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0); | |||
11526 | } | |||
11527 | case X86::BI__builtin_ia32_sqrtpd256: | |||
11528 | case X86::BI__builtin_ia32_sqrtpd: | |||
11529 | case X86::BI__builtin_ia32_sqrtps256: | |||
11530 | case X86::BI__builtin_ia32_sqrtps: | |||
11531 | case X86::BI__builtin_ia32_sqrtps512: | |||
11532 | case X86::BI__builtin_ia32_sqrtpd512: { | |||
11533 | if (Ops.size() == 2) { | |||
11534 | unsigned CC = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
11535 | // Support only if the rounding mode is 4 (AKA CUR_DIRECTION), | |||
11536 | // otherwise keep the intrinsic. | |||
11537 | if (CC != 4) { | |||
11538 | Intrinsic::ID IID = BuiltinID == X86::BI__builtin_ia32_sqrtps512 ? | |||
11539 | Intrinsic::x86_avx512_sqrt_ps_512 : | |||
11540 | Intrinsic::x86_avx512_sqrt_pd_512; | |||
11541 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
11542 | } | |||
11543 | } | |||
11544 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, Ops[0]->getType()); | |||
11545 | return Builder.CreateCall(F, Ops[0]); | |||
11546 | } | |||
11547 | case X86::BI__builtin_ia32_pabsb128: | |||
11548 | case X86::BI__builtin_ia32_pabsw128: | |||
11549 | case X86::BI__builtin_ia32_pabsd128: | |||
11550 | case X86::BI__builtin_ia32_pabsb256: | |||
11551 | case X86::BI__builtin_ia32_pabsw256: | |||
11552 | case X86::BI__builtin_ia32_pabsd256: | |||
11553 | case X86::BI__builtin_ia32_pabsq128: | |||
11554 | case X86::BI__builtin_ia32_pabsq256: | |||
11555 | case X86::BI__builtin_ia32_pabsb512: | |||
11556 | case X86::BI__builtin_ia32_pabsw512: | |||
11557 | case X86::BI__builtin_ia32_pabsd512: | |||
11558 | case X86::BI__builtin_ia32_pabsq512: | |||
11559 | return EmitX86Abs(*this, Ops); | |||
11560 | ||||
11561 | case X86::BI__builtin_ia32_pmaxsb128: | |||
11562 | case X86::BI__builtin_ia32_pmaxsw128: | |||
11563 | case X86::BI__builtin_ia32_pmaxsd128: | |||
11564 | case X86::BI__builtin_ia32_pmaxsq128: | |||
11565 | case X86::BI__builtin_ia32_pmaxsb256: | |||
11566 | case X86::BI__builtin_ia32_pmaxsw256: | |||
11567 | case X86::BI__builtin_ia32_pmaxsd256: | |||
11568 | case X86::BI__builtin_ia32_pmaxsq256: | |||
11569 | case X86::BI__builtin_ia32_pmaxsb512: | |||
11570 | case X86::BI__builtin_ia32_pmaxsw512: | |||
11571 | case X86::BI__builtin_ia32_pmaxsd512: | |||
11572 | case X86::BI__builtin_ia32_pmaxsq512: | |||
11573 | return EmitX86MinMax(*this, ICmpInst::ICMP_SGT, Ops); | |||
11574 | case X86::BI__builtin_ia32_pmaxub128: | |||
11575 | case X86::BI__builtin_ia32_pmaxuw128: | |||
11576 | case X86::BI__builtin_ia32_pmaxud128: | |||
11577 | case X86::BI__builtin_ia32_pmaxuq128: | |||
11578 | case X86::BI__builtin_ia32_pmaxub256: | |||
11579 | case X86::BI__builtin_ia32_pmaxuw256: | |||
11580 | case X86::BI__builtin_ia32_pmaxud256: | |||
11581 | case X86::BI__builtin_ia32_pmaxuq256: | |||
11582 | case X86::BI__builtin_ia32_pmaxub512: | |||
11583 | case X86::BI__builtin_ia32_pmaxuw512: | |||
11584 | case X86::BI__builtin_ia32_pmaxud512: | |||
11585 | case X86::BI__builtin_ia32_pmaxuq512: | |||
11586 | return EmitX86MinMax(*this, ICmpInst::ICMP_UGT, Ops); | |||
11587 | case X86::BI__builtin_ia32_pminsb128: | |||
11588 | case X86::BI__builtin_ia32_pminsw128: | |||
11589 | case X86::BI__builtin_ia32_pminsd128: | |||
11590 | case X86::BI__builtin_ia32_pminsq128: | |||
11591 | case X86::BI__builtin_ia32_pminsb256: | |||
11592 | case X86::BI__builtin_ia32_pminsw256: | |||
11593 | case X86::BI__builtin_ia32_pminsd256: | |||
11594 | case X86::BI__builtin_ia32_pminsq256: | |||
11595 | case X86::BI__builtin_ia32_pminsb512: | |||
11596 | case X86::BI__builtin_ia32_pminsw512: | |||
11597 | case X86::BI__builtin_ia32_pminsd512: | |||
11598 | case X86::BI__builtin_ia32_pminsq512: | |||
11599 | return EmitX86MinMax(*this, ICmpInst::ICMP_SLT, Ops); | |||
11600 | case X86::BI__builtin_ia32_pminub128: | |||
11601 | case X86::BI__builtin_ia32_pminuw128: | |||
11602 | case X86::BI__builtin_ia32_pminud128: | |||
11603 | case X86::BI__builtin_ia32_pminuq128: | |||
11604 | case X86::BI__builtin_ia32_pminub256: | |||
11605 | case X86::BI__builtin_ia32_pminuw256: | |||
11606 | case X86::BI__builtin_ia32_pminud256: | |||
11607 | case X86::BI__builtin_ia32_pminuq256: | |||
11608 | case X86::BI__builtin_ia32_pminub512: | |||
11609 | case X86::BI__builtin_ia32_pminuw512: | |||
11610 | case X86::BI__builtin_ia32_pminud512: | |||
11611 | case X86::BI__builtin_ia32_pminuq512: | |||
11612 | return EmitX86MinMax(*this, ICmpInst::ICMP_ULT, Ops); | |||
11613 | ||||
11614 | case X86::BI__builtin_ia32_pmuludq128: | |||
11615 | case X86::BI__builtin_ia32_pmuludq256: | |||
11616 | case X86::BI__builtin_ia32_pmuludq512: | |||
11617 | return EmitX86Muldq(*this, /*IsSigned*/false, Ops); | |||
11618 | ||||
11619 | case X86::BI__builtin_ia32_pmuldq128: | |||
11620 | case X86::BI__builtin_ia32_pmuldq256: | |||
11621 | case X86::BI__builtin_ia32_pmuldq512: | |||
11622 | return EmitX86Muldq(*this, /*IsSigned*/true, Ops); | |||
11623 | ||||
11624 | case X86::BI__builtin_ia32_pternlogd512_mask: | |||
11625 | case X86::BI__builtin_ia32_pternlogq512_mask: | |||
11626 | case X86::BI__builtin_ia32_pternlogd128_mask: | |||
11627 | case X86::BI__builtin_ia32_pternlogd256_mask: | |||
11628 | case X86::BI__builtin_ia32_pternlogq128_mask: | |||
11629 | case X86::BI__builtin_ia32_pternlogq256_mask: | |||
11630 | return EmitX86Ternlog(*this, /*ZeroMask*/false, Ops); | |||
11631 | ||||
11632 | case X86::BI__builtin_ia32_pternlogd512_maskz: | |||
11633 | case X86::BI__builtin_ia32_pternlogq512_maskz: | |||
11634 | case X86::BI__builtin_ia32_pternlogd128_maskz: | |||
11635 | case X86::BI__builtin_ia32_pternlogd256_maskz: | |||
11636 | case X86::BI__builtin_ia32_pternlogq128_maskz: | |||
11637 | case X86::BI__builtin_ia32_pternlogq256_maskz: | |||
11638 | return EmitX86Ternlog(*this, /*ZeroMask*/true, Ops); | |||
11639 | ||||
11640 | case X86::BI__builtin_ia32_vpshldd128: | |||
11641 | case X86::BI__builtin_ia32_vpshldd256: | |||
11642 | case X86::BI__builtin_ia32_vpshldd512: | |||
11643 | case X86::BI__builtin_ia32_vpshldq128: | |||
11644 | case X86::BI__builtin_ia32_vpshldq256: | |||
11645 | case X86::BI__builtin_ia32_vpshldq512: | |||
11646 | case X86::BI__builtin_ia32_vpshldw128: | |||
11647 | case X86::BI__builtin_ia32_vpshldw256: | |||
11648 | case X86::BI__builtin_ia32_vpshldw512: | |||
11649 | return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false); | |||
11650 | ||||
11651 | case X86::BI__builtin_ia32_vpshrdd128: | |||
11652 | case X86::BI__builtin_ia32_vpshrdd256: | |||
11653 | case X86::BI__builtin_ia32_vpshrdd512: | |||
11654 | case X86::BI__builtin_ia32_vpshrdq128: | |||
11655 | case X86::BI__builtin_ia32_vpshrdq256: | |||
11656 | case X86::BI__builtin_ia32_vpshrdq512: | |||
11657 | case X86::BI__builtin_ia32_vpshrdw128: | |||
11658 | case X86::BI__builtin_ia32_vpshrdw256: | |||
11659 | case X86::BI__builtin_ia32_vpshrdw512: | |||
11660 | // Ops 0 and 1 are swapped. | |||
11661 | return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true); | |||
11662 | ||||
11663 | case X86::BI__builtin_ia32_vpshldvd128: | |||
11664 | case X86::BI__builtin_ia32_vpshldvd256: | |||
11665 | case X86::BI__builtin_ia32_vpshldvd512: | |||
11666 | case X86::BI__builtin_ia32_vpshldvq128: | |||
11667 | case X86::BI__builtin_ia32_vpshldvq256: | |||
11668 | case X86::BI__builtin_ia32_vpshldvq512: | |||
11669 | case X86::BI__builtin_ia32_vpshldvw128: | |||
11670 | case X86::BI__builtin_ia32_vpshldvw256: | |||
11671 | case X86::BI__builtin_ia32_vpshldvw512: | |||
11672 | return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false); | |||
11673 | ||||
11674 | case X86::BI__builtin_ia32_vpshrdvd128: | |||
11675 | case X86::BI__builtin_ia32_vpshrdvd256: | |||
11676 | case X86::BI__builtin_ia32_vpshrdvd512: | |||
11677 | case X86::BI__builtin_ia32_vpshrdvq128: | |||
11678 | case X86::BI__builtin_ia32_vpshrdvq256: | |||
11679 | case X86::BI__builtin_ia32_vpshrdvq512: | |||
11680 | case X86::BI__builtin_ia32_vpshrdvw128: | |||
11681 | case X86::BI__builtin_ia32_vpshrdvw256: | |||
11682 | case X86::BI__builtin_ia32_vpshrdvw512: | |||
11683 | // Ops 0 and 1 are swapped. | |||
11684 | return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true); | |||
11685 | ||||
11686 | // 3DNow! | |||
11687 | case X86::BI__builtin_ia32_pswapdsf: | |||
11688 | case X86::BI__builtin_ia32_pswapdsi: { | |||
11689 | llvm::Type *MMXTy = llvm::Type::getX86_MMXTy(getLLVMContext()); | |||
11690 | Ops[0] = Builder.CreateBitCast(Ops[0], MMXTy, "cast"); | |||
11691 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_3dnowa_pswapd); | |||
11692 | return Builder.CreateCall(F, Ops, "pswapd"); | |||
11693 | } | |||
11694 | case X86::BI__builtin_ia32_rdrand16_step: | |||
11695 | case X86::BI__builtin_ia32_rdrand32_step: | |||
11696 | case X86::BI__builtin_ia32_rdrand64_step: | |||
11697 | case X86::BI__builtin_ia32_rdseed16_step: | |||
11698 | case X86::BI__builtin_ia32_rdseed32_step: | |||
11699 | case X86::BI__builtin_ia32_rdseed64_step: { | |||
11700 | Intrinsic::ID ID; | |||
11701 | switch (BuiltinID) { | |||
11702 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11702); | |||
11703 | case X86::BI__builtin_ia32_rdrand16_step: | |||
11704 | ID = Intrinsic::x86_rdrand_16; | |||
11705 | break; | |||
11706 | case X86::BI__builtin_ia32_rdrand32_step: | |||
11707 | ID = Intrinsic::x86_rdrand_32; | |||
11708 | break; | |||
11709 | case X86::BI__builtin_ia32_rdrand64_step: | |||
11710 | ID = Intrinsic::x86_rdrand_64; | |||
11711 | break; | |||
11712 | case X86::BI__builtin_ia32_rdseed16_step: | |||
11713 | ID = Intrinsic::x86_rdseed_16; | |||
11714 | break; | |||
11715 | case X86::BI__builtin_ia32_rdseed32_step: | |||
11716 | ID = Intrinsic::x86_rdseed_32; | |||
11717 | break; | |||
11718 | case X86::BI__builtin_ia32_rdseed64_step: | |||
11719 | ID = Intrinsic::x86_rdseed_64; | |||
11720 | break; | |||
11721 | } | |||
11722 | ||||
11723 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID)); | |||
11724 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 0), | |||
11725 | Ops[0]); | |||
11726 | return Builder.CreateExtractValue(Call, 1); | |||
11727 | } | |||
11728 | case X86::BI__builtin_ia32_addcarryx_u32: | |||
11729 | case X86::BI__builtin_ia32_addcarryx_u64: | |||
11730 | case X86::BI__builtin_ia32_subborrow_u32: | |||
11731 | case X86::BI__builtin_ia32_subborrow_u64: { | |||
11732 | Intrinsic::ID IID; | |||
11733 | switch (BuiltinID) { | |||
11734 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11734); | |||
11735 | case X86::BI__builtin_ia32_addcarryx_u32: | |||
11736 | IID = Intrinsic::x86_addcarry_32; | |||
11737 | break; | |||
11738 | case X86::BI__builtin_ia32_addcarryx_u64: | |||
11739 | IID = Intrinsic::x86_addcarry_64; | |||
11740 | break; | |||
11741 | case X86::BI__builtin_ia32_subborrow_u32: | |||
11742 | IID = Intrinsic::x86_subborrow_32; | |||
11743 | break; | |||
11744 | case X86::BI__builtin_ia32_subborrow_u64: | |||
11745 | IID = Intrinsic::x86_subborrow_64; | |||
11746 | break; | |||
11747 | } | |||
11748 | ||||
11749 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), | |||
11750 | { Ops[0], Ops[1], Ops[2] }); | |||
11751 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1), | |||
11752 | Ops[3]); | |||
11753 | return Builder.CreateExtractValue(Call, 0); | |||
11754 | } | |||
11755 | ||||
11756 | case X86::BI__builtin_ia32_fpclassps128_mask: | |||
11757 | case X86::BI__builtin_ia32_fpclassps256_mask: | |||
11758 | case X86::BI__builtin_ia32_fpclassps512_mask: | |||
11759 | case X86::BI__builtin_ia32_fpclasspd128_mask: | |||
11760 | case X86::BI__builtin_ia32_fpclasspd256_mask: | |||
11761 | case X86::BI__builtin_ia32_fpclasspd512_mask: { | |||
11762 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
11763 | Value *MaskIn = Ops[2]; | |||
11764 | Ops.erase(&Ops[2]); | |||
11765 | ||||
11766 | Intrinsic::ID ID; | |||
11767 | switch (BuiltinID) { | |||
11768 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11768); | |||
11769 | case X86::BI__builtin_ia32_fpclassps128_mask: | |||
11770 | ID = Intrinsic::x86_avx512_fpclass_ps_128; | |||
11771 | break; | |||
11772 | case X86::BI__builtin_ia32_fpclassps256_mask: | |||
11773 | ID = Intrinsic::x86_avx512_fpclass_ps_256; | |||
11774 | break; | |||
11775 | case X86::BI__builtin_ia32_fpclassps512_mask: | |||
11776 | ID = Intrinsic::x86_avx512_fpclass_ps_512; | |||
11777 | break; | |||
11778 | case X86::BI__builtin_ia32_fpclasspd128_mask: | |||
11779 | ID = Intrinsic::x86_avx512_fpclass_pd_128; | |||
11780 | break; | |||
11781 | case X86::BI__builtin_ia32_fpclasspd256_mask: | |||
11782 | ID = Intrinsic::x86_avx512_fpclass_pd_256; | |||
11783 | break; | |||
11784 | case X86::BI__builtin_ia32_fpclasspd512_mask: | |||
11785 | ID = Intrinsic::x86_avx512_fpclass_pd_512; | |||
11786 | break; | |||
11787 | } | |||
11788 | ||||
11789 | Value *Fpclass = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
11790 | return EmitX86MaskedCompareResult(*this, Fpclass, NumElts, MaskIn); | |||
11791 | } | |||
11792 | ||||
11793 | case X86::BI__builtin_ia32_vp2intersect_q_512: | |||
11794 | case X86::BI__builtin_ia32_vp2intersect_q_256: | |||
11795 | case X86::BI__builtin_ia32_vp2intersect_q_128: | |||
11796 | case X86::BI__builtin_ia32_vp2intersect_d_512: | |||
11797 | case X86::BI__builtin_ia32_vp2intersect_d_256: | |||
11798 | case X86::BI__builtin_ia32_vp2intersect_d_128: { | |||
11799 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
11800 | Intrinsic::ID ID; | |||
11801 | ||||
11802 | switch (BuiltinID) { | |||
11803 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11803); | |||
11804 | case X86::BI__builtin_ia32_vp2intersect_q_512: | |||
11805 | ID = Intrinsic::x86_avx512_vp2intersect_q_512; | |||
11806 | break; | |||
11807 | case X86::BI__builtin_ia32_vp2intersect_q_256: | |||
11808 | ID = Intrinsic::x86_avx512_vp2intersect_q_256; | |||
11809 | break; | |||
11810 | case X86::BI__builtin_ia32_vp2intersect_q_128: | |||
11811 | ID = Intrinsic::x86_avx512_vp2intersect_q_128; | |||
11812 | break; | |||
11813 | case X86::BI__builtin_ia32_vp2intersect_d_512: | |||
11814 | ID = Intrinsic::x86_avx512_vp2intersect_d_512; | |||
11815 | break; | |||
11816 | case X86::BI__builtin_ia32_vp2intersect_d_256: | |||
11817 | ID = Intrinsic::x86_avx512_vp2intersect_d_256; | |||
11818 | break; | |||
11819 | case X86::BI__builtin_ia32_vp2intersect_d_128: | |||
11820 | ID = Intrinsic::x86_avx512_vp2intersect_d_128; | |||
11821 | break; | |||
11822 | } | |||
11823 | ||||
11824 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID), {Ops[0], Ops[1]}); | |||
11825 | Value *Result = Builder.CreateExtractValue(Call, 0); | |||
11826 | Result = EmitX86MaskedCompareResult(*this, Result, NumElts, nullptr); | |||
11827 | Builder.CreateDefaultAlignedStore(Result, Ops[2]); | |||
11828 | ||||
11829 | Result = Builder.CreateExtractValue(Call, 1); | |||
11830 | Result = EmitX86MaskedCompareResult(*this, Result, NumElts, nullptr); | |||
11831 | return Builder.CreateDefaultAlignedStore(Result, Ops[3]); | |||
11832 | } | |||
11833 | ||||
11834 | case X86::BI__builtin_ia32_vpmultishiftqb128: | |||
11835 | case X86::BI__builtin_ia32_vpmultishiftqb256: | |||
11836 | case X86::BI__builtin_ia32_vpmultishiftqb512: { | |||
11837 | Intrinsic::ID ID; | |||
11838 | switch (BuiltinID) { | |||
11839 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11839); | |||
11840 | case X86::BI__builtin_ia32_vpmultishiftqb128: | |||
11841 | ID = Intrinsic::x86_avx512_pmultishift_qb_128; | |||
11842 | break; | |||
11843 | case X86::BI__builtin_ia32_vpmultishiftqb256: | |||
11844 | ID = Intrinsic::x86_avx512_pmultishift_qb_256; | |||
11845 | break; | |||
11846 | case X86::BI__builtin_ia32_vpmultishiftqb512: | |||
11847 | ID = Intrinsic::x86_avx512_pmultishift_qb_512; | |||
11848 | break; | |||
11849 | } | |||
11850 | ||||
11851 | return Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
11852 | } | |||
11853 | ||||
11854 | case X86::BI__builtin_ia32_vpshufbitqmb128_mask: | |||
11855 | case X86::BI__builtin_ia32_vpshufbitqmb256_mask: | |||
11856 | case X86::BI__builtin_ia32_vpshufbitqmb512_mask: { | |||
11857 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
11858 | Value *MaskIn = Ops[2]; | |||
11859 | Ops.erase(&Ops[2]); | |||
11860 | ||||
11861 | Intrinsic::ID ID; | |||
11862 | switch (BuiltinID) { | |||
11863 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11863); | |||
11864 | case X86::BI__builtin_ia32_vpshufbitqmb128_mask: | |||
11865 | ID = Intrinsic::x86_avx512_vpshufbitqmb_128; | |||
11866 | break; | |||
11867 | case X86::BI__builtin_ia32_vpshufbitqmb256_mask: | |||
11868 | ID = Intrinsic::x86_avx512_vpshufbitqmb_256; | |||
11869 | break; | |||
11870 | case X86::BI__builtin_ia32_vpshufbitqmb512_mask: | |||
11871 | ID = Intrinsic::x86_avx512_vpshufbitqmb_512; | |||
11872 | break; | |||
11873 | } | |||
11874 | ||||
11875 | Value *Shufbit = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
11876 | return EmitX86MaskedCompareResult(*this, Shufbit, NumElts, MaskIn); | |||
11877 | } | |||
11878 | ||||
11879 | // packed comparison intrinsics | |||
11880 | case X86::BI__builtin_ia32_cmpeqps: | |||
11881 | case X86::BI__builtin_ia32_cmpeqpd: | |||
11882 | return getVectorFCmpIR(CmpInst::FCMP_OEQ); | |||
11883 | case X86::BI__builtin_ia32_cmpltps: | |||
11884 | case X86::BI__builtin_ia32_cmpltpd: | |||
11885 | return getVectorFCmpIR(CmpInst::FCMP_OLT); | |||
11886 | case X86::BI__builtin_ia32_cmpleps: | |||
11887 | case X86::BI__builtin_ia32_cmplepd: | |||
11888 | return getVectorFCmpIR(CmpInst::FCMP_OLE); | |||
11889 | case X86::BI__builtin_ia32_cmpunordps: | |||
11890 | case X86::BI__builtin_ia32_cmpunordpd: | |||
11891 | return getVectorFCmpIR(CmpInst::FCMP_UNO); | |||
11892 | case X86::BI__builtin_ia32_cmpneqps: | |||
11893 | case X86::BI__builtin_ia32_cmpneqpd: | |||
11894 | return getVectorFCmpIR(CmpInst::FCMP_UNE); | |||
11895 | case X86::BI__builtin_ia32_cmpnltps: | |||
11896 | case X86::BI__builtin_ia32_cmpnltpd: | |||
11897 | return getVectorFCmpIR(CmpInst::FCMP_UGE); | |||
11898 | case X86::BI__builtin_ia32_cmpnleps: | |||
11899 | case X86::BI__builtin_ia32_cmpnlepd: | |||
11900 | return getVectorFCmpIR(CmpInst::FCMP_UGT); | |||
11901 | case X86::BI__builtin_ia32_cmpordps: | |||
11902 | case X86::BI__builtin_ia32_cmpordpd: | |||
11903 | return getVectorFCmpIR(CmpInst::FCMP_ORD); | |||
11904 | case X86::BI__builtin_ia32_cmpps: | |||
11905 | case X86::BI__builtin_ia32_cmpps256: | |||
11906 | case X86::BI__builtin_ia32_cmppd: | |||
11907 | case X86::BI__builtin_ia32_cmppd256: | |||
11908 | case X86::BI__builtin_ia32_cmpps128_mask: | |||
11909 | case X86::BI__builtin_ia32_cmpps256_mask: | |||
11910 | case X86::BI__builtin_ia32_cmpps512_mask: | |||
11911 | case X86::BI__builtin_ia32_cmppd128_mask: | |||
11912 | case X86::BI__builtin_ia32_cmppd256_mask: | |||
11913 | case X86::BI__builtin_ia32_cmppd512_mask: { | |||
11914 | // Lowering vector comparisons to fcmp instructions, while | |||
11915 | // ignoring signalling behaviour requested | |||
11916 | // ignoring rounding mode requested | |||
11917 | // This is is only possible as long as FENV_ACCESS is not implemented. | |||
11918 | // See also: https://reviews.llvm.org/D45616 | |||
11919 | ||||
11920 | // The third argument is the comparison condition, and integer in the | |||
11921 | // range [0, 31] | |||
11922 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x1f; | |||
11923 | ||||
11924 | // Lowering to IR fcmp instruction. | |||
11925 | // Ignoring requested signaling behaviour, | |||
11926 | // e.g. both _CMP_GT_OS & _CMP_GT_OQ are translated to FCMP_OGT. | |||
11927 | FCmpInst::Predicate Pred; | |||
11928 | switch (CC) { | |||
11929 | case 0x00: Pred = FCmpInst::FCMP_OEQ; break; | |||
11930 | case 0x01: Pred = FCmpInst::FCMP_OLT; break; | |||
11931 | case 0x02: Pred = FCmpInst::FCMP_OLE; break; | |||
11932 | case 0x03: Pred = FCmpInst::FCMP_UNO; break; | |||
11933 | case 0x04: Pred = FCmpInst::FCMP_UNE; break; | |||
11934 | case 0x05: Pred = FCmpInst::FCMP_UGE; break; | |||
11935 | case 0x06: Pred = FCmpInst::FCMP_UGT; break; | |||
11936 | case 0x07: Pred = FCmpInst::FCMP_ORD; break; | |||
11937 | case 0x08: Pred = FCmpInst::FCMP_UEQ; break; | |||
11938 | case 0x09: Pred = FCmpInst::FCMP_ULT; break; | |||
11939 | case 0x0a: Pred = FCmpInst::FCMP_ULE; break; | |||
11940 | case 0x0b: Pred = FCmpInst::FCMP_FALSE; break; | |||
11941 | case 0x0c: Pred = FCmpInst::FCMP_ONE; break; | |||
11942 | case 0x0d: Pred = FCmpInst::FCMP_OGE; break; | |||
11943 | case 0x0e: Pred = FCmpInst::FCMP_OGT; break; | |||
11944 | case 0x0f: Pred = FCmpInst::FCMP_TRUE; break; | |||
11945 | case 0x10: Pred = FCmpInst::FCMP_OEQ; break; | |||
11946 | case 0x11: Pred = FCmpInst::FCMP_OLT; break; | |||
11947 | case 0x12: Pred = FCmpInst::FCMP_OLE; break; | |||
11948 | case 0x13: Pred = FCmpInst::FCMP_UNO; break; | |||
11949 | case 0x14: Pred = FCmpInst::FCMP_UNE; break; | |||
11950 | case 0x15: Pred = FCmpInst::FCMP_UGE; break; | |||
11951 | case 0x16: Pred = FCmpInst::FCMP_UGT; break; | |||
11952 | case 0x17: Pred = FCmpInst::FCMP_ORD; break; | |||
11953 | case 0x18: Pred = FCmpInst::FCMP_UEQ; break; | |||
11954 | case 0x19: Pred = FCmpInst::FCMP_ULT; break; | |||
11955 | case 0x1a: Pred = FCmpInst::FCMP_ULE; break; | |||
11956 | case 0x1b: Pred = FCmpInst::FCMP_FALSE; break; | |||
11957 | case 0x1c: Pred = FCmpInst::FCMP_ONE; break; | |||
11958 | case 0x1d: Pred = FCmpInst::FCMP_OGE; break; | |||
11959 | case 0x1e: Pred = FCmpInst::FCMP_OGT; break; | |||
11960 | case 0x1f: Pred = FCmpInst::FCMP_TRUE; break; | |||
11961 | default: llvm_unreachable("Unhandled CC")::llvm::llvm_unreachable_internal("Unhandled CC", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11961); | |||
11962 | } | |||
11963 | ||||
11964 | // Builtins without the _mask suffix return a vector of integers | |||
11965 | // of the same width as the input vectors | |||
11966 | switch (BuiltinID) { | |||
11967 | case X86::BI__builtin_ia32_cmpps512_mask: | |||
11968 | case X86::BI__builtin_ia32_cmppd512_mask: | |||
11969 | case X86::BI__builtin_ia32_cmpps128_mask: | |||
11970 | case X86::BI__builtin_ia32_cmpps256_mask: | |||
11971 | case X86::BI__builtin_ia32_cmppd128_mask: | |||
11972 | case X86::BI__builtin_ia32_cmppd256_mask: { | |||
11973 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); | |||
11974 | Value *Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]); | |||
11975 | return EmitX86MaskedCompareResult(*this, Cmp, NumElts, Ops[3]); | |||
11976 | } | |||
11977 | default: | |||
11978 | return getVectorFCmpIR(Pred); | |||
11979 | } | |||
11980 | } | |||
11981 | ||||
11982 | // SSE scalar comparison intrinsics | |||
11983 | case X86::BI__builtin_ia32_cmpeqss: | |||
11984 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 0); | |||
11985 | case X86::BI__builtin_ia32_cmpltss: | |||
11986 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 1); | |||
11987 | case X86::BI__builtin_ia32_cmpless: | |||
11988 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 2); | |||
11989 | case X86::BI__builtin_ia32_cmpunordss: | |||
11990 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 3); | |||
11991 | case X86::BI__builtin_ia32_cmpneqss: | |||
11992 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 4); | |||
11993 | case X86::BI__builtin_ia32_cmpnltss: | |||
11994 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 5); | |||
11995 | case X86::BI__builtin_ia32_cmpnless: | |||
11996 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 6); | |||
11997 | case X86::BI__builtin_ia32_cmpordss: | |||
11998 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 7); | |||
11999 | case X86::BI__builtin_ia32_cmpeqsd: | |||
12000 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 0); | |||
12001 | case X86::BI__builtin_ia32_cmpltsd: | |||
12002 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 1); | |||
12003 | case X86::BI__builtin_ia32_cmplesd: | |||
12004 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 2); | |||
12005 | case X86::BI__builtin_ia32_cmpunordsd: | |||
12006 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 3); | |||
12007 | case X86::BI__builtin_ia32_cmpneqsd: | |||
12008 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 4); | |||
12009 | case X86::BI__builtin_ia32_cmpnltsd: | |||
12010 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 5); | |||
12011 | case X86::BI__builtin_ia32_cmpnlesd: | |||
12012 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 6); | |||
12013 | case X86::BI__builtin_ia32_cmpordsd: | |||
12014 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 7); | |||
12015 | ||||
12016 | // AVX512 bf16 intrinsics | |||
12017 | case X86::BI__builtin_ia32_cvtneps2bf16_128_mask: { | |||
12018 | Ops[2] = getMaskVecValue(*this, Ops[2], | |||
12019 | Ops[0]->getType()->getVectorNumElements()); | |||
12020 | Intrinsic::ID IID = Intrinsic::x86_avx512bf16_mask_cvtneps2bf16_128; | |||
12021 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
12022 | } | |||
12023 | case X86::BI__builtin_ia32_cvtsbf162ss_32: | |||
12024 | return EmitX86CvtBF16ToFloatExpr(*this, E, Ops); | |||
12025 | ||||
12026 | case X86::BI__builtin_ia32_cvtneps2bf16_256_mask: | |||
12027 | case X86::BI__builtin_ia32_cvtneps2bf16_512_mask: { | |||
12028 | Intrinsic::ID IID; | |||
12029 | switch (BuiltinID) { | |||
12030 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12030); | |||
12031 | case X86::BI__builtin_ia32_cvtneps2bf16_256_mask: | |||
12032 | IID = Intrinsic::x86_avx512bf16_cvtneps2bf16_256; | |||
12033 | break; | |||
12034 | case X86::BI__builtin_ia32_cvtneps2bf16_512_mask: | |||
12035 | IID = Intrinsic::x86_avx512bf16_cvtneps2bf16_512; | |||
12036 | break; | |||
12037 | } | |||
12038 | Value *Res = Builder.CreateCall(CGM.getIntrinsic(IID), Ops[0]); | |||
12039 | return EmitX86Select(*this, Ops[2], Res, Ops[1]); | |||
12040 | } | |||
12041 | ||||
12042 | case X86::BI__emul: | |||
12043 | case X86::BI__emulu: { | |||
12044 | llvm::Type *Int64Ty = llvm::IntegerType::get(getLLVMContext(), 64); | |||
12045 | bool isSigned = (BuiltinID == X86::BI__emul); | |||
12046 | Value *LHS = Builder.CreateIntCast(Ops[0], Int64Ty, isSigned); | |||
12047 | Value *RHS = Builder.CreateIntCast(Ops[1], Int64Ty, isSigned); | |||
12048 | return Builder.CreateMul(LHS, RHS, "", !isSigned, isSigned); | |||
12049 | } | |||
12050 | case X86::BI__mulh: | |||
12051 | case X86::BI__umulh: | |||
12052 | case X86::BI_mul128: | |||
12053 | case X86::BI_umul128: { | |||
12054 | llvm::Type *ResType = ConvertType(E->getType()); | |||
12055 | llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128); | |||
12056 | ||||
12057 | bool IsSigned = (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI_mul128); | |||
12058 | Value *LHS = Builder.CreateIntCast(Ops[0], Int128Ty, IsSigned); | |||
12059 | Value *RHS = Builder.CreateIntCast(Ops[1], Int128Ty, IsSigned); | |||
12060 | ||||
12061 | Value *MulResult, *HigherBits; | |||
12062 | if (IsSigned) { | |||
12063 | MulResult = Builder.CreateNSWMul(LHS, RHS); | |||
12064 | HigherBits = Builder.CreateAShr(MulResult, 64); | |||
12065 | } else { | |||
12066 | MulResult = Builder.CreateNUWMul(LHS, RHS); | |||
12067 | HigherBits = Builder.CreateLShr(MulResult, 64); | |||
12068 | } | |||
12069 | HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned); | |||
12070 | ||||
12071 | if (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI__umulh) | |||
12072 | return HigherBits; | |||
12073 | ||||
12074 | Address HighBitsAddress = EmitPointerWithAlignment(E->getArg(2)); | |||
12075 | Builder.CreateStore(HigherBits, HighBitsAddress); | |||
12076 | return Builder.CreateIntCast(MulResult, ResType, IsSigned); | |||
12077 | } | |||
12078 | ||||
12079 | case X86::BI__faststorefence: { | |||
12080 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, | |||
12081 | llvm::SyncScope::System); | |||
12082 | } | |||
12083 | case X86::BI__shiftleft128: | |||
12084 | case X86::BI__shiftright128: { | |||
12085 | // FIXME: Once fshl/fshr no longer add an unneeded and and cmov, do this: | |||
12086 | // llvm::Function *F = CGM.getIntrinsic( | |||
12087 | // BuiltinID == X86::BI__shiftleft128 ? Intrinsic::fshl : Intrinsic::fshr, | |||
12088 | // Int64Ty); | |||
12089 | // Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty); | |||
12090 | // return Builder.CreateCall(F, Ops); | |||
12091 | llvm::Type *Int128Ty = Builder.getInt128Ty(); | |||
12092 | Value *HighPart128 = | |||
12093 | Builder.CreateShl(Builder.CreateZExt(Ops[1], Int128Ty), 64); | |||
12094 | Value *LowPart128 = Builder.CreateZExt(Ops[0], Int128Ty); | |||
12095 | Value *Val = Builder.CreateOr(HighPart128, LowPart128); | |||
12096 | Value *Amt = Builder.CreateAnd(Builder.CreateZExt(Ops[2], Int128Ty), | |||
12097 | llvm::ConstantInt::get(Int128Ty, 0x3f)); | |||
12098 | Value *Res; | |||
12099 | if (BuiltinID == X86::BI__shiftleft128) | |||
12100 | Res = Builder.CreateLShr(Builder.CreateShl(Val, Amt), 64); | |||
12101 | else | |||
12102 | Res = Builder.CreateLShr(Val, Amt); | |||
12103 | return Builder.CreateTrunc(Res, Int64Ty); | |||
12104 | } | |||
12105 | case X86::BI_ReadWriteBarrier: | |||
12106 | case X86::BI_ReadBarrier: | |||
12107 | case X86::BI_WriteBarrier: { | |||
12108 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, | |||
12109 | llvm::SyncScope::SingleThread); | |||
12110 | } | |||
12111 | case X86::BI_BitScanForward: | |||
12112 | case X86::BI_BitScanForward64: | |||
12113 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E); | |||
12114 | case X86::BI_BitScanReverse: | |||
12115 | case X86::BI_BitScanReverse64: | |||
12116 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E); | |||
12117 | ||||
12118 | case X86::BI_InterlockedAnd64: | |||
12119 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E); | |||
12120 | case X86::BI_InterlockedExchange64: | |||
12121 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E); | |||
12122 | case X86::BI_InterlockedExchangeAdd64: | |||
12123 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E); | |||
12124 | case X86::BI_InterlockedExchangeSub64: | |||
12125 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E); | |||
12126 | case X86::BI_InterlockedOr64: | |||
12127 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E); | |||
12128 | case X86::BI_InterlockedXor64: | |||
12129 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E); | |||
12130 | case X86::BI_InterlockedDecrement64: | |||
12131 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E); | |||
12132 | case X86::BI_InterlockedIncrement64: | |||
12133 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E); | |||
12134 | case X86::BI_InterlockedCompareExchange128: { | |||
12135 | // InterlockedCompareExchange128 doesn't directly refer to 128bit ints, | |||
12136 | // instead it takes pointers to 64bit ints for Destination and | |||
12137 | // ComparandResult, and exchange is taken as two 64bit ints (high & low). | |||
12138 | // The previous value is written to ComparandResult, and success is | |||
12139 | // returned. | |||
12140 | ||||
12141 | llvm::Type *Int128Ty = Builder.getInt128Ty(); | |||
12142 | llvm::Type *Int128PtrTy = Int128Ty->getPointerTo(); | |||
12143 | ||||
12144 | Value *Destination = | |||
12145 | Builder.CreateBitCast(Ops[0], Int128PtrTy); | |||
12146 | Value *ExchangeHigh128 = Builder.CreateZExt(Ops[1], Int128Ty); | |||
12147 | Value *ExchangeLow128 = Builder.CreateZExt(Ops[2], Int128Ty); | |||
12148 | Address ComparandResult(Builder.CreateBitCast(Ops[3], Int128PtrTy), | |||
12149 | getContext().toCharUnitsFromBits(128)); | |||
12150 | ||||
12151 | Value *Exchange = Builder.CreateOr( | |||
12152 | Builder.CreateShl(ExchangeHigh128, 64, "", false, false), | |||
12153 | ExchangeLow128); | |||
12154 | ||||
12155 | Value *Comparand = Builder.CreateLoad(ComparandResult); | |||
12156 | ||||
12157 | AtomicCmpXchgInst *CXI = | |||
12158 | Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, | |||
12159 | AtomicOrdering::SequentiallyConsistent, | |||
12160 | AtomicOrdering::SequentiallyConsistent); | |||
12161 | CXI->setVolatile(true); | |||
12162 | ||||
12163 | // Write the result back to the inout pointer. | |||
12164 | Builder.CreateStore(Builder.CreateExtractValue(CXI, 0), ComparandResult); | |||
12165 | ||||
12166 | // Get the success boolean and zero extend it to i8. | |||
12167 | Value *Success = Builder.CreateExtractValue(CXI, 1); | |||
12168 | return Builder.CreateZExt(Success, ConvertType(E->getType())); | |||
12169 | } | |||
12170 | ||||
12171 | case X86::BI_AddressOfReturnAddress: { | |||
12172 | Function *F = | |||
12173 | CGM.getIntrinsic(Intrinsic::addressofreturnaddress, AllocaInt8PtrTy); | |||
12174 | return Builder.CreateCall(F); | |||
12175 | } | |||
12176 | case X86::BI__stosb: { | |||
12177 | // We treat __stosb as a volatile memset - it may not generate "rep stosb" | |||
12178 | // instruction, but it will create a memset that won't be optimized away. | |||
12179 | return Builder.CreateMemSet(Ops[0], Ops[1], Ops[2], 1, true); | |||
12180 | } | |||
12181 | case X86::BI__ud2: | |||
12182 | // llvm.trap makes a ud2a instruction on x86. | |||
12183 | return EmitTrapCall(Intrinsic::trap); | |||
12184 | case X86::BI__int2c: { | |||
12185 | // This syscall signals a driver assertion failure in x86 NT kernels. | |||
12186 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false); | |||
12187 | llvm::InlineAsm *IA = | |||
12188 | llvm::InlineAsm::get(FTy, "int $$0x2c", "", /*hasSideEffects=*/true); | |||
12189 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( | |||
12190 | getLLVMContext(), llvm::AttributeList::FunctionIndex, | |||
12191 | llvm::Attribute::NoReturn); | |||
12192 | llvm::CallInst *CI = Builder.CreateCall(IA); | |||
12193 | CI->setAttributes(NoReturnAttr); | |||
12194 | return CI; | |||
12195 | } | |||
12196 | case X86::BI__readfsbyte: | |||
12197 | case X86::BI__readfsword: | |||
12198 | case X86::BI__readfsdword: | |||
12199 | case X86::BI__readfsqword: { | |||
12200 | llvm::Type *IntTy = ConvertType(E->getType()); | |||
12201 | Value *Ptr = | |||
12202 | Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 257)); | |||
12203 | LoadInst *Load = Builder.CreateAlignedLoad( | |||
12204 | IntTy, Ptr, getContext().getTypeAlignInChars(E->getType())); | |||
12205 | Load->setVolatile(true); | |||
12206 | return Load; | |||
12207 | } | |||
12208 | case X86::BI__readgsbyte: | |||
12209 | case X86::BI__readgsword: | |||
12210 | case X86::BI__readgsdword: | |||
12211 | case X86::BI__readgsqword: { | |||
12212 | llvm::Type *IntTy = ConvertType(E->getType()); | |||
12213 | Value *Ptr = | |||
12214 | Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 256)); | |||
12215 | LoadInst *Load = Builder.CreateAlignedLoad( | |||
12216 | IntTy, Ptr, getContext().getTypeAlignInChars(E->getType())); | |||
12217 | Load->setVolatile(true); | |||
12218 | return Load; | |||
12219 | } | |||
12220 | case X86::BI__builtin_ia32_paddsb512: | |||
12221 | case X86::BI__builtin_ia32_paddsw512: | |||
12222 | case X86::BI__builtin_ia32_paddsb256: | |||
12223 | case X86::BI__builtin_ia32_paddsw256: | |||
12224 | case X86::BI__builtin_ia32_paddsb128: | |||
12225 | case X86::BI__builtin_ia32_paddsw128: | |||
12226 | return EmitX86AddSubSatExpr(*this, Ops, true, true); | |||
12227 | case X86::BI__builtin_ia32_paddusb512: | |||
12228 | case X86::BI__builtin_ia32_paddusw512: | |||
12229 | case X86::BI__builtin_ia32_paddusb256: | |||
12230 | case X86::BI__builtin_ia32_paddusw256: | |||
12231 | case X86::BI__builtin_ia32_paddusb128: | |||
12232 | case X86::BI__builtin_ia32_paddusw128: | |||
12233 | return EmitX86AddSubSatExpr(*this, Ops, false, true); | |||
12234 | case X86::BI__builtin_ia32_psubsb512: | |||
12235 | case X86::BI__builtin_ia32_psubsw512: | |||
12236 | case X86::BI__builtin_ia32_psubsb256: | |||
12237 | case X86::BI__builtin_ia32_psubsw256: | |||
12238 | case X86::BI__builtin_ia32_psubsb128: | |||
12239 | case X86::BI__builtin_ia32_psubsw128: | |||
12240 | return EmitX86AddSubSatExpr(*this, Ops, true, false); | |||
12241 | case X86::BI__builtin_ia32_psubusb512: | |||
12242 | case X86::BI__builtin_ia32_psubusw512: | |||
12243 | case X86::BI__builtin_ia32_psubusb256: | |||
12244 | case X86::BI__builtin_ia32_psubusw256: | |||
12245 | case X86::BI__builtin_ia32_psubusb128: | |||
12246 | case X86::BI__builtin_ia32_psubusw128: | |||
12247 | return EmitX86AddSubSatExpr(*this, Ops, false, false); | |||
12248 | } | |||
12249 | } | |||
12250 | ||||
12251 | Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID, | |||
12252 | const CallExpr *E) { | |||
12253 | SmallVector<Value*, 4> Ops; | |||
12254 | ||||
12255 | for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) | |||
12256 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
12257 | ||||
12258 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
12259 | ||||
12260 | switch (BuiltinID) { | |||
12261 | default: return nullptr; | |||
12262 | ||||
12263 | // __builtin_ppc_get_timebase is GCC 4.8+'s PowerPC-specific name for what we | |||
12264 | // call __builtin_readcyclecounter. | |||
12265 | case PPC::BI__builtin_ppc_get_timebase: | |||
12266 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::readcyclecounter)); | |||
12267 | ||||
12268 | // vec_ld, vec_xl_be, vec_lvsl, vec_lvsr | |||
12269 | case PPC::BI__builtin_altivec_lvx: | |||
12270 | case PPC::BI__builtin_altivec_lvxl: | |||
12271 | case PPC::BI__builtin_altivec_lvebx: | |||
12272 | case PPC::BI__builtin_altivec_lvehx: | |||
12273 | case PPC::BI__builtin_altivec_lvewx: | |||
12274 | case PPC::BI__builtin_altivec_lvsl: | |||
12275 | case PPC::BI__builtin_altivec_lvsr: | |||
12276 | case PPC::BI__builtin_vsx_lxvd2x: | |||
12277 | case PPC::BI__builtin_vsx_lxvw4x: | |||
12278 | case PPC::BI__builtin_vsx_lxvd2x_be: | |||
12279 | case PPC::BI__builtin_vsx_lxvw4x_be: | |||
12280 | case PPC::BI__builtin_vsx_lxvl: | |||
12281 | case PPC::BI__builtin_vsx_lxvll: | |||
12282 | { | |||
12283 | if(BuiltinID == PPC::BI__builtin_vsx_lxvl || | |||
12284 | BuiltinID == PPC::BI__builtin_vsx_lxvll){ | |||
12285 | Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy); | |||
12286 | }else { | |||
12287 | Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy); | |||
12288 | Ops[0] = Builder.CreateGEP(Ops[1], Ops[0]); | |||
12289 | Ops.pop_back(); | |||
12290 | } | |||
12291 | ||||
12292 | switch (BuiltinID) { | |||
12293 | default: llvm_unreachable("Unsupported ld/lvsl/lvsr intrinsic!")::llvm::llvm_unreachable_internal("Unsupported ld/lvsl/lvsr intrinsic!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12293); | |||
12294 | case PPC::BI__builtin_altivec_lvx: | |||
12295 | ID = Intrinsic::ppc_altivec_lvx; | |||
12296 | break; | |||
12297 | case PPC::BI__builtin_altivec_lvxl: | |||
12298 | ID = Intrinsic::ppc_altivec_lvxl; | |||
12299 | break; | |||
12300 | case PPC::BI__builtin_altivec_lvebx: | |||
12301 | ID = Intrinsic::ppc_altivec_lvebx; | |||
12302 | break; | |||
12303 | case PPC::BI__builtin_altivec_lvehx: | |||
12304 | ID = Intrinsic::ppc_altivec_lvehx; | |||
12305 | break; | |||
12306 | case PPC::BI__builtin_altivec_lvewx: | |||
12307 | ID = Intrinsic::ppc_altivec_lvewx; | |||
12308 | break; | |||
12309 | case PPC::BI__builtin_altivec_lvsl: | |||
12310 | ID = Intrinsic::ppc_altivec_lvsl; | |||
12311 | break; | |||
12312 | case PPC::BI__builtin_altivec_lvsr: | |||
12313 | ID = Intrinsic::ppc_altivec_lvsr; | |||
12314 | break; | |||
12315 | case PPC::BI__builtin_vsx_lxvd2x: | |||
12316 | ID = Intrinsic::ppc_vsx_lxvd2x; | |||
12317 | break; | |||
12318 | case PPC::BI__builtin_vsx_lxvw4x: | |||
12319 | ID = Intrinsic::ppc_vsx_lxvw4x; | |||
12320 | break; | |||
12321 | case PPC::BI__builtin_vsx_lxvd2x_be: | |||
12322 | ID = Intrinsic::ppc_vsx_lxvd2x_be; | |||
12323 | break; | |||
12324 | case PPC::BI__builtin_vsx_lxvw4x_be: | |||
12325 | ID = Intrinsic::ppc_vsx_lxvw4x_be; | |||
12326 | break; | |||
12327 | case PPC::BI__builtin_vsx_lxvl: | |||
12328 | ID = Intrinsic::ppc_vsx_lxvl; | |||
12329 | break; | |||
12330 | case PPC::BI__builtin_vsx_lxvll: | |||
12331 | ID = Intrinsic::ppc_vsx_lxvll; | |||
12332 | break; | |||
12333 | } | |||
12334 | llvm::Function *F = CGM.getIntrinsic(ID); | |||
12335 | return Builder.CreateCall(F, Ops, ""); | |||
12336 | } | |||
12337 | ||||
12338 | // vec_st, vec_xst_be | |||
12339 | case PPC::BI__builtin_altivec_stvx: | |||
12340 | case PPC::BI__builtin_altivec_stvxl: | |||
12341 | case PPC::BI__builtin_altivec_stvebx: | |||
12342 | case PPC::BI__builtin_altivec_stvehx: | |||
12343 | case PPC::BI__builtin_altivec_stvewx: | |||
12344 | case PPC::BI__builtin_vsx_stxvd2x: | |||
12345 | case PPC::BI__builtin_vsx_stxvw4x: | |||
12346 | case PPC::BI__builtin_vsx_stxvd2x_be: | |||
12347 | case PPC::BI__builtin_vsx_stxvw4x_be: | |||
12348 | case PPC::BI__builtin_vsx_stxvl: | |||
12349 | case PPC::BI__builtin_vsx_stxvll: | |||
12350 | { | |||
12351 | if(BuiltinID == PPC::BI__builtin_vsx_stxvl || | |||
12352 | BuiltinID == PPC::BI__builtin_vsx_stxvll ){ | |||
12353 | Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy); | |||
12354 | }else { | |||
12355 | Ops[2] = Builder.CreateBitCast(Ops[2], Int8PtrTy); | |||
12356 | Ops[1] = Builder.CreateGEP(Ops[2], Ops[1]); | |||
12357 | Ops.pop_back(); | |||
12358 | } | |||
12359 | ||||
12360 | switch (BuiltinID) { | |||
12361 | default: llvm_unreachable("Unsupported st intrinsic!")::llvm::llvm_unreachable_internal("Unsupported st intrinsic!" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12361); | |||
12362 | case PPC::BI__builtin_altivec_stvx: | |||
12363 | ID = Intrinsic::ppc_altivec_stvx; | |||
12364 | break; | |||
12365 | case PPC::BI__builtin_altivec_stvxl: | |||
12366 | ID = Intrinsic::ppc_altivec_stvxl; | |||
12367 | break; | |||
12368 | case PPC::BI__builtin_altivec_stvebx: | |||
12369 | ID = Intrinsic::ppc_altivec_stvebx; | |||
12370 | break; | |||
12371 | case PPC::BI__builtin_altivec_stvehx: | |||
12372 | ID = Intrinsic::ppc_altivec_stvehx; | |||
12373 | break; | |||
12374 | case PPC::BI__builtin_altivec_stvewx: | |||
12375 | ID = Intrinsic::ppc_altivec_stvewx; | |||
12376 | break; | |||
12377 | case PPC::BI__builtin_vsx_stxvd2x: | |||
12378 | ID = Intrinsic::ppc_vsx_stxvd2x; | |||
12379 | break; | |||
12380 | case PPC::BI__builtin_vsx_stxvw4x: | |||
12381 | ID = Intrinsic::ppc_vsx_stxvw4x; | |||
12382 | break; | |||
12383 | case PPC::BI__builtin_vsx_stxvd2x_be: | |||
12384 | ID = Intrinsic::ppc_vsx_stxvd2x_be; | |||
12385 | break; | |||
12386 | case PPC::BI__builtin_vsx_stxvw4x_be: | |||
12387 | ID = Intrinsic::ppc_vsx_stxvw4x_be; | |||
12388 | break; | |||
12389 | case PPC::BI__builtin_vsx_stxvl: | |||
12390 | ID = Intrinsic::ppc_vsx_stxvl; | |||
12391 | break; | |||
12392 | case PPC::BI__builtin_vsx_stxvll: | |||
12393 | ID = Intrinsic::ppc_vsx_stxvll; | |||
12394 | break; | |||
12395 | } | |||
12396 | llvm::Function *F = CGM.getIntrinsic(ID); | |||
12397 | return Builder.CreateCall(F, Ops, ""); | |||
12398 | } | |||
12399 | // Square root | |||
12400 | case PPC::BI__builtin_vsx_xvsqrtsp: | |||
12401 | case PPC::BI__builtin_vsx_xvsqrtdp: { | |||
12402 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12403 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12404 | ID = Intrinsic::sqrt; | |||
12405 | llvm::Function *F = CGM.getIntrinsic(ID, ResultType); | |||
12406 | return Builder.CreateCall(F, X); | |||
12407 | } | |||
12408 | // Count leading zeros | |||
12409 | case PPC::BI__builtin_altivec_vclzb: | |||
12410 | case PPC::BI__builtin_altivec_vclzh: | |||
12411 | case PPC::BI__builtin_altivec_vclzw: | |||
12412 | case PPC::BI__builtin_altivec_vclzd: { | |||
12413 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12414 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12415 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); | |||
12416 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType); | |||
12417 | return Builder.CreateCall(F, {X, Undef}); | |||
12418 | } | |||
12419 | case PPC::BI__builtin_altivec_vctzb: | |||
12420 | case PPC::BI__builtin_altivec_vctzh: | |||
12421 | case PPC::BI__builtin_altivec_vctzw: | |||
12422 | case PPC::BI__builtin_altivec_vctzd: { | |||
12423 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12424 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12425 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); | |||
12426 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType); | |||
12427 | return Builder.CreateCall(F, {X, Undef}); | |||
12428 | } | |||
12429 | case PPC::BI__builtin_altivec_vpopcntb: | |||
12430 | case PPC::BI__builtin_altivec_vpopcnth: | |||
12431 | case PPC::BI__builtin_altivec_vpopcntw: | |||
12432 | case PPC::BI__builtin_altivec_vpopcntd: { | |||
12433 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12434 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12435 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); | |||
12436 | return Builder.CreateCall(F, X); | |||
12437 | } | |||
12438 | // Copy sign | |||
12439 | case PPC::BI__builtin_vsx_xvcpsgnsp: | |||
12440 | case PPC::BI__builtin_vsx_xvcpsgndp: { | |||
12441 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12442 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12443 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
12444 | ID = Intrinsic::copysign; | |||
12445 | llvm::Function *F = CGM.getIntrinsic(ID, ResultType); | |||
12446 | return Builder.CreateCall(F, {X, Y}); | |||
12447 | } | |||
12448 | // Rounding/truncation | |||
12449 | case PPC::BI__builtin_vsx_xvrspip: | |||
12450 | case PPC::BI__builtin_vsx_xvrdpip: | |||
12451 | case PPC::BI__builtin_vsx_xvrdpim: | |||
12452 | case PPC::BI__builtin_vsx_xvrspim: | |||
12453 | case PPC::BI__builtin_vsx_xvrdpi: | |||
12454 | case PPC::BI__builtin_vsx_xvrspi: | |||
12455 | case PPC::BI__builtin_vsx_xvrdpic: | |||
12456 | case PPC::BI__builtin_vsx_xvrspic: | |||
12457 | case PPC::BI__builtin_vsx_xvrdpiz: | |||
12458 | case PPC::BI__builtin_vsx_xvrspiz: { | |||
12459 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12460 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12461 | if (BuiltinID == PPC::BI__builtin_vsx_xvrdpim || | |||
12462 | BuiltinID == PPC::BI__builtin_vsx_xvrspim) | |||
12463 | ID = Intrinsic::floor; | |||
12464 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpi || | |||
12465 | BuiltinID == PPC::BI__builtin_vsx_xvrspi) | |||
12466 | ID = Intrinsic::round; | |||
12467 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpic || | |||
12468 | BuiltinID == PPC::BI__builtin_vsx_xvrspic) | |||
12469 | ID = Intrinsic::nearbyint; | |||
12470 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpip || | |||
12471 | BuiltinID == PPC::BI__builtin_vsx_xvrspip) | |||
12472 | ID = Intrinsic::ceil; | |||
12473 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpiz || | |||
12474 | BuiltinID == PPC::BI__builtin_vsx_xvrspiz) | |||
12475 | ID = Intrinsic::trunc; | |||
12476 | llvm::Function *F = CGM.getIntrinsic(ID, ResultType); | |||
12477 | return Builder.CreateCall(F, X); | |||
12478 | } | |||
12479 | ||||
12480 | // Absolute value | |||
12481 | case PPC::BI__builtin_vsx_xvabsdp: | |||
12482 | case PPC::BI__builtin_vsx_xvabssp: { | |||
12483 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12484 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12485 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); | |||
12486 | return Builder.CreateCall(F, X); | |||
12487 | } | |||
12488 | ||||
12489 | // FMA variations | |||
12490 | case PPC::BI__builtin_vsx_xvmaddadp: | |||
12491 | case PPC::BI__builtin_vsx_xvmaddasp: | |||
12492 | case PPC::BI__builtin_vsx_xvnmaddadp: | |||
12493 | case PPC::BI__builtin_vsx_xvnmaddasp: | |||
12494 | case PPC::BI__builtin_vsx_xvmsubadp: | |||
12495 | case PPC::BI__builtin_vsx_xvmsubasp: | |||
12496 | case PPC::BI__builtin_vsx_xvnmsubadp: | |||
12497 | case PPC::BI__builtin_vsx_xvnmsubasp: { | |||
12498 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12499 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12500 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
12501 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
12502 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); | |||
12503 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
12504 | switch (BuiltinID) { | |||
12505 | case PPC::BI__builtin_vsx_xvmaddadp: | |||
12506 | case PPC::BI__builtin_vsx_xvmaddasp: | |||
12507 | return Builder.CreateCall(F, {X, Y, Z}); | |||
12508 | case PPC::BI__builtin_vsx_xvnmaddadp: | |||
12509 | case PPC::BI__builtin_vsx_xvnmaddasp: | |||
12510 | return Builder.CreateFSub(Zero, | |||
12511 | Builder.CreateCall(F, {X, Y, Z}), "sub"); | |||
12512 | case PPC::BI__builtin_vsx_xvmsubadp: | |||
12513 | case PPC::BI__builtin_vsx_xvmsubasp: | |||
12514 | return Builder.CreateCall(F, | |||
12515 | {X, Y, Builder.CreateFSub(Zero, Z, "sub")}); | |||
12516 | case PPC::BI__builtin_vsx_xvnmsubadp: | |||
12517 | case PPC::BI__builtin_vsx_xvnmsubasp: | |||
12518 | Value *FsubRes = | |||
12519 | Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")}); | |||
12520 | return Builder.CreateFSub(Zero, FsubRes, "sub"); | |||
12521 | } | |||
12522 | llvm_unreachable("Unknown FMA operation")::llvm::llvm_unreachable_internal("Unknown FMA operation", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12522); | |||
12523 | return nullptr; // Suppress no-return warning | |||
12524 | } | |||
12525 | ||||
12526 | case PPC::BI__builtin_vsx_insertword: { | |||
12527 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxinsertw); | |||
12528 | ||||
12529 | // Third argument is a compile time constant int. It must be clamped to | |||
12530 | // to the range [0, 12]. | |||
12531 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); | |||
12532 | assert(ArgCI &&((ArgCI && "Third arg to xxinsertw intrinsic must be constant integer" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12533, __PRETTY_FUNCTION__)) | |||
12533 | "Third arg to xxinsertw intrinsic must be constant integer")((ArgCI && "Third arg to xxinsertw intrinsic must be constant integer" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12533, __PRETTY_FUNCTION__)); | |||
12534 | const int64_t MaxIndex = 12; | |||
12535 | int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex); | |||
12536 | ||||
12537 | // The builtin semantics don't exactly match the xxinsertw instructions | |||
12538 | // semantics (which ppc_vsx_xxinsertw follows). The builtin extracts the | |||
12539 | // word from the first argument, and inserts it in the second argument. The | |||
12540 | // instruction extracts the word from its second input register and inserts | |||
12541 | // it into its first input register, so swap the first and second arguments. | |||
12542 | std::swap(Ops[0], Ops[1]); | |||
12543 | ||||
12544 | // Need to cast the second argument from a vector of unsigned int to a | |||
12545 | // vector of long long. | |||
12546 | Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int64Ty, 2)); | |||
12547 | ||||
12548 | if (getTarget().isLittleEndian()) { | |||
12549 | // Create a shuffle mask of (1, 0) | |||
12550 | Constant *ShuffleElts[2] = { ConstantInt::get(Int32Ty, 1), | |||
12551 | ConstantInt::get(Int32Ty, 0) | |||
12552 | }; | |||
12553 | Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts); | |||
12554 | ||||
12555 | // Reverse the double words in the vector we will extract from. | |||
12556 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2)); | |||
12557 | Ops[0] = Builder.CreateShuffleVector(Ops[0], Ops[0], ShuffleMask); | |||
12558 | ||||
12559 | // Reverse the index. | |||
12560 | Index = MaxIndex - Index; | |||
12561 | } | |||
12562 | ||||
12563 | // Intrinsic expects the first arg to be a vector of int. | |||
12564 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4)); | |||
12565 | Ops[2] = ConstantInt::getSigned(Int32Ty, Index); | |||
12566 | return Builder.CreateCall(F, Ops); | |||
12567 | } | |||
12568 | ||||
12569 | case PPC::BI__builtin_vsx_extractuword: { | |||
12570 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxextractuw); | |||
12571 | ||||
12572 | // Intrinsic expects the first argument to be a vector of doublewords. | |||
12573 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2)); | |||
12574 | ||||
12575 | // The second argument is a compile time constant int that needs to | |||
12576 | // be clamped to the range [0, 12]. | |||
12577 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[1]); | |||
12578 | assert(ArgCI &&((ArgCI && "Second Arg to xxextractuw intrinsic must be a constant integer!" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12579, __PRETTY_FUNCTION__)) | |||
12579 | "Second Arg to xxextractuw intrinsic must be a constant integer!")((ArgCI && "Second Arg to xxextractuw intrinsic must be a constant integer!" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12579, __PRETTY_FUNCTION__)); | |||
12580 | const int64_t MaxIndex = 12; | |||
12581 | int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex); | |||
12582 | ||||
12583 | if (getTarget().isLittleEndian()) { | |||
12584 | // Reverse the index. | |||
12585 | Index = MaxIndex - Index; | |||
12586 | Ops[1] = ConstantInt::getSigned(Int32Ty, Index); | |||
12587 | ||||
12588 | // Emit the call, then reverse the double words of the results vector. | |||
12589 | Value *Call = Builder.CreateCall(F, Ops); | |||
12590 | ||||
12591 | // Create a shuffle mask of (1, 0) | |||
12592 | Constant *ShuffleElts[2] = { ConstantInt::get(Int32Ty, 1), | |||
12593 | ConstantInt::get(Int32Ty, 0) | |||
12594 | }; | |||
12595 | Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts); | |||
12596 | ||||
12597 | Value *ShuffleCall = Builder.CreateShuffleVector(Call, Call, ShuffleMask); | |||
12598 | return ShuffleCall; | |||
12599 | } else { | |||
12600 | Ops[1] = ConstantInt::getSigned(Int32Ty, Index); | |||
12601 | return Builder.CreateCall(F, Ops); | |||
12602 | } | |||
12603 | } | |||
12604 | ||||
12605 | case PPC::BI__builtin_vsx_xxpermdi: { | |||
12606 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); | |||
12607 | assert(ArgCI && "Third arg must be constant integer!")((ArgCI && "Third arg must be constant integer!") ? static_cast <void> (0) : __assert_fail ("ArgCI && \"Third arg must be constant integer!\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12607, __PRETTY_FUNCTION__)); | |||
12608 | ||||
12609 | unsigned Index = ArgCI->getZExtValue(); | |||
12610 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2)); | |||
12611 | Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int64Ty, 2)); | |||
12612 | ||||
12613 | // Account for endianness by treating this as just a shuffle. So we use the | |||
12614 | // same indices for both LE and BE in order to produce expected results in | |||
12615 | // both cases. | |||
12616 | unsigned ElemIdx0 = (Index & 2) >> 1; | |||
12617 | unsigned ElemIdx1 = 2 + (Index & 1); | |||
12618 | ||||
12619 | Constant *ShuffleElts[2] = {ConstantInt::get(Int32Ty, ElemIdx0), | |||
12620 | ConstantInt::get(Int32Ty, ElemIdx1)}; | |||
12621 | Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts); | |||
12622 | ||||
12623 | Value *ShuffleCall = | |||
12624 | Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleMask); | |||
12625 | QualType BIRetType = E->getType(); | |||
12626 | auto RetTy = ConvertType(BIRetType); | |||
12627 | return Builder.CreateBitCast(ShuffleCall, RetTy); | |||
12628 | } | |||
12629 | ||||
12630 | case PPC::BI__builtin_vsx_xxsldwi: { | |||
12631 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); | |||
12632 | assert(ArgCI && "Third argument must be a compile time constant")((ArgCI && "Third argument must be a compile time constant" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Third argument must be a compile time constant\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12632, __PRETTY_FUNCTION__)); | |||
12633 | unsigned Index = ArgCI->getZExtValue() & 0x3; | |||
12634 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4)); | |||
12635 | Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int32Ty, 4)); | |||
12636 | ||||
12637 | // Create a shuffle mask | |||
12638 | unsigned ElemIdx0; | |||
12639 | unsigned ElemIdx1; | |||
12640 | unsigned ElemIdx2; | |||
12641 | unsigned ElemIdx3; | |||
12642 | if (getTarget().isLittleEndian()) { | |||
12643 | // Little endian element N comes from element 8+N-Index of the | |||
12644 | // concatenated wide vector (of course, using modulo arithmetic on | |||
12645 | // the total number of elements). | |||
12646 | ElemIdx0 = (8 - Index) % 8; | |||
12647 | ElemIdx1 = (9 - Index) % 8; | |||
12648 | ElemIdx2 = (10 - Index) % 8; | |||
12649 | ElemIdx3 = (11 - Index) % 8; | |||
12650 | } else { | |||
12651 | // Big endian ElemIdx<N> = Index + N | |||
12652 | ElemIdx0 = Index; | |||
12653 | ElemIdx1 = Index + 1; | |||
12654 | ElemIdx2 = Index + 2; | |||
12655 | ElemIdx3 = Index + 3; | |||
12656 | } | |||
12657 | ||||
12658 | Constant *ShuffleElts[4] = {ConstantInt::get(Int32Ty, ElemIdx0), | |||
12659 | ConstantInt::get(Int32Ty, ElemIdx1), | |||
12660 | ConstantInt::get(Int32Ty, ElemIdx2), | |||
12661 | ConstantInt::get(Int32Ty, ElemIdx3)}; | |||
12662 | ||||
12663 | Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts); | |||
12664 | Value *ShuffleCall = | |||
12665 | Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleMask); | |||
12666 | QualType BIRetType = E->getType(); | |||
12667 | auto RetTy = ConvertType(BIRetType); | |||
12668 | return Builder.CreateBitCast(ShuffleCall, RetTy); | |||
12669 | } | |||
12670 | ||||
12671 | case PPC::BI__builtin_pack_vector_int128: { | |||
12672 | bool isLittleEndian = getTarget().isLittleEndian(); | |||
12673 | Value *UndefValue = | |||
12674 | llvm::UndefValue::get(llvm::VectorType::get(Ops[0]->getType(), 2)); | |||
12675 | Value *Res = Builder.CreateInsertElement( | |||
12676 | UndefValue, Ops[0], (uint64_t)(isLittleEndian ? 1 : 0)); | |||
12677 | Res = Builder.CreateInsertElement(Res, Ops[1], | |||
12678 | (uint64_t)(isLittleEndian ? 0 : 1)); | |||
12679 | return Builder.CreateBitCast(Res, ConvertType(E->getType())); | |||
12680 | } | |||
12681 | ||||
12682 | case PPC::BI__builtin_unpack_vector_int128: { | |||
12683 | ConstantInt *Index = cast<ConstantInt>(Ops[1]); | |||
12684 | Value *Unpacked = Builder.CreateBitCast( | |||
12685 | Ops[0], llvm::VectorType::get(ConvertType(E->getType()), 2)); | |||
12686 | ||||
12687 | if (getTarget().isLittleEndian()) | |||
12688 | Index = ConstantInt::get(Index->getType(), 1 - Index->getZExtValue()); | |||
12689 | ||||
12690 | return Builder.CreateExtractElement(Unpacked, Index); | |||
12691 | } | |||
12692 | } | |||
12693 | } | |||
12694 | ||||
12695 | Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID, | |||
12696 | const CallExpr *E) { | |||
12697 | switch (BuiltinID) { | |||
12698 | case AMDGPU::BI__builtin_amdgcn_div_scale: | |||
12699 | case AMDGPU::BI__builtin_amdgcn_div_scalef: { | |||
12700 | // Translate from the intrinsics's struct return to the builtin's out | |||
12701 | // argument. | |||
12702 | ||||
12703 | Address FlagOutPtr = EmitPointerWithAlignment(E->getArg(3)); | |||
12704 | ||||
12705 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); | |||
12706 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); | |||
12707 | llvm::Value *Z = EmitScalarExpr(E->getArg(2)); | |||
12708 | ||||
12709 | llvm::Function *Callee = CGM.getIntrinsic(Intrinsic::amdgcn_div_scale, | |||
12710 | X->getType()); | |||
12711 | ||||
12712 | llvm::Value *Tmp = Builder.CreateCall(Callee, {X, Y, Z}); | |||
12713 | ||||
12714 | llvm::Value *Result = Builder.CreateExtractValue(Tmp, 0); | |||
12715 | llvm::Value *Flag = Builder.CreateExtractValue(Tmp, 1); | |||
12716 | ||||
12717 | llvm::Type *RealFlagType | |||
12718 | = FlagOutPtr.getPointer()->getType()->getPointerElementType(); | |||
12719 | ||||
12720 | llvm::Value *FlagExt = Builder.CreateZExt(Flag, RealFlagType); | |||
12721 | Builder.CreateStore(FlagExt, FlagOutPtr); | |||
12722 | return Result; | |||
12723 | } | |||
12724 | case AMDGPU::BI__builtin_amdgcn_div_fmas: | |||
12725 | case AMDGPU::BI__builtin_amdgcn_div_fmasf: { | |||
12726 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
12727 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
12728 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); | |||
12729 | llvm::Value *Src3 = EmitScalarExpr(E->getArg(3)); | |||
12730 | ||||
12731 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_div_fmas, | |||
12732 | Src0->getType()); | |||
12733 | llvm::Value *Src3ToBool = Builder.CreateIsNotNull(Src3); | |||
12734 | return Builder.CreateCall(F, {Src0, Src1, Src2, Src3ToBool}); | |||
12735 | } | |||
12736 | ||||
12737 | case AMDGPU::BI__builtin_amdgcn_ds_swizzle: | |||
12738 | return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_ds_swizzle); | |||
12739 | case AMDGPU::BI__builtin_amdgcn_mov_dpp8: | |||
12740 | return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_mov_dpp8); | |||
12741 | case AMDGPU::BI__builtin_amdgcn_mov_dpp: | |||
12742 | case AMDGPU::BI__builtin_amdgcn_update_dpp: { | |||
12743 | llvm::SmallVector<llvm::Value *, 6> Args; | |||
12744 | for (unsigned I = 0; I != E->getNumArgs(); ++I) | |||
12745 | Args.push_back(EmitScalarExpr(E->getArg(I))); | |||
12746 | assert(Args.size() == 5 || Args.size() == 6)((Args.size() == 5 || Args.size() == 6) ? static_cast<void > (0) : __assert_fail ("Args.size() == 5 || Args.size() == 6" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12746, __PRETTY_FUNCTION__)); | |||
12747 | if (Args.size() == 5) | |||
12748 | Args.insert(Args.begin(), llvm::UndefValue::get(Args[0]->getType())); | |||
12749 | Function *F = | |||
12750 | CGM.getIntrinsic(Intrinsic::amdgcn_update_dpp, Args[0]->getType()); | |||
12751 | return Builder.CreateCall(F, Args); | |||
12752 | } | |||
12753 | case AMDGPU::BI__builtin_amdgcn_div_fixup: | |||
12754 | case AMDGPU::BI__builtin_amdgcn_div_fixupf: | |||
12755 | case AMDGPU::BI__builtin_amdgcn_div_fixuph: | |||
12756 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_div_fixup); | |||
12757 | case AMDGPU::BI__builtin_amdgcn_trig_preop: | |||
12758 | case AMDGPU::BI__builtin_amdgcn_trig_preopf: | |||
12759 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_trig_preop); | |||
12760 | case AMDGPU::BI__builtin_amdgcn_rcp: | |||
12761 | case AMDGPU::BI__builtin_amdgcn_rcpf: | |||
12762 | case AMDGPU::BI__builtin_amdgcn_rcph: | |||
12763 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rcp); | |||
12764 | case AMDGPU::BI__builtin_amdgcn_rsq: | |||
12765 | case AMDGPU::BI__builtin_amdgcn_rsqf: | |||
12766 | case AMDGPU::BI__builtin_amdgcn_rsqh: | |||
12767 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq); | |||
12768 | case AMDGPU::BI__builtin_amdgcn_rsq_clamp: | |||
12769 | case AMDGPU::BI__builtin_amdgcn_rsq_clampf: | |||
12770 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq_clamp); | |||
12771 | case AMDGPU::BI__builtin_amdgcn_sinf: | |||
12772 | case AMDGPU::BI__builtin_amdgcn_sinh: | |||
12773 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_sin); | |||
12774 | case AMDGPU::BI__builtin_amdgcn_cosf: | |||
12775 | case AMDGPU::BI__builtin_amdgcn_cosh: | |||
12776 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_cos); | |||
12777 | case AMDGPU::BI__builtin_amdgcn_log_clampf: | |||
12778 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_log_clamp); | |||
12779 | case AMDGPU::BI__builtin_amdgcn_ldexp: | |||
12780 | case AMDGPU::BI__builtin_amdgcn_ldexpf: | |||
12781 | case AMDGPU::BI__builtin_amdgcn_ldexph: | |||
12782 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_ldexp); | |||
12783 | case AMDGPU::BI__builtin_amdgcn_frexp_mant: | |||
12784 | case AMDGPU::BI__builtin_amdgcn_frexp_mantf: | |||
12785 | case AMDGPU::BI__builtin_amdgcn_frexp_manth: | |||
12786 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_frexp_mant); | |||
12787 | case AMDGPU::BI__builtin_amdgcn_frexp_exp: | |||
12788 | case AMDGPU::BI__builtin_amdgcn_frexp_expf: { | |||
12789 | Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
12790 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp, | |||
12791 | { Builder.getInt32Ty(), Src0->getType() }); | |||
12792 | return Builder.CreateCall(F, Src0); | |||
12793 | } | |||
12794 | case AMDGPU::BI__builtin_amdgcn_frexp_exph: { | |||
12795 | Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
12796 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp, | |||
12797 | { Builder.getInt16Ty(), Src0->getType() }); | |||
12798 | return Builder.CreateCall(F, Src0); | |||
12799 | } | |||
12800 | case AMDGPU::BI__builtin_amdgcn_fract: | |||
12801 | case AMDGPU::BI__builtin_amdgcn_fractf: | |||
12802 | case AMDGPU::BI__builtin_amdgcn_fracth: | |||
12803 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_fract); | |||
12804 | case AMDGPU::BI__builtin_amdgcn_lerp: | |||
12805 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_lerp); | |||
12806 | case AMDGPU::BI__builtin_amdgcn_ubfe: | |||
12807 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_ubfe); | |||
12808 | case AMDGPU::BI__builtin_amdgcn_sbfe: | |||
12809 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_sbfe); | |||
12810 | case AMDGPU::BI__builtin_amdgcn_uicmp: | |||
12811 | case AMDGPU::BI__builtin_amdgcn_uicmpl: | |||
12812 | case AMDGPU::BI__builtin_amdgcn_sicmp: | |||
12813 | case AMDGPU::BI__builtin_amdgcn_sicmpl: { | |||
12814 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
12815 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
12816 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); | |||
12817 | ||||
12818 | // FIXME-GFX10: How should 32 bit mask be handled? | |||
12819 | Value *F = CGM.getIntrinsic(Intrinsic::amdgcn_icmp, | |||
12820 | { Builder.getInt64Ty(), Src0->getType() }); | |||
12821 | return Builder.CreateCall(F, { Src0, Src1, Src2 }); | |||
12822 | } | |||
12823 | case AMDGPU::BI__builtin_amdgcn_fcmp: | |||
12824 | case AMDGPU::BI__builtin_amdgcn_fcmpf: { | |||
12825 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
12826 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
12827 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); | |||
12828 | ||||
12829 | // FIXME-GFX10: How should 32 bit mask be handled? | |||
12830 | Value *F = CGM.getIntrinsic(Intrinsic::amdgcn_fcmp, | |||
12831 | { Builder.getInt64Ty(), Src0->getType() }); | |||
12832 | return Builder.CreateCall(F, { Src0, Src1, Src2 }); | |||
12833 | } | |||
12834 | case AMDGPU::BI__builtin_amdgcn_class: | |||
12835 | case AMDGPU::BI__builtin_amdgcn_classf: | |||
12836 | case AMDGPU::BI__builtin_amdgcn_classh: | |||
12837 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_class); | |||
12838 | case AMDGPU::BI__builtin_amdgcn_fmed3f: | |||
12839 | case AMDGPU::BI__builtin_amdgcn_fmed3h: | |||
12840 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fmed3); | |||
12841 | case AMDGPU::BI__builtin_amdgcn_ds_append: | |||
12842 | case AMDGPU::BI__builtin_amdgcn_ds_consume: { | |||
12843 | Intrinsic::ID Intrin = BuiltinID == AMDGPU::BI__builtin_amdgcn_ds_append ? | |||
12844 | Intrinsic::amdgcn_ds_append : Intrinsic::amdgcn_ds_consume; | |||
12845 | Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
12846 | Function *F = CGM.getIntrinsic(Intrin, { Src0->getType() }); | |||
12847 | return Builder.CreateCall(F, { Src0, Builder.getFalse() }); | |||
12848 | } | |||
12849 | case AMDGPU::BI__builtin_amdgcn_read_exec: { | |||
12850 | CallInst *CI = cast<CallInst>( | |||
12851 | EmitSpecialRegisterBuiltin(*this, E, Int64Ty, Int64Ty, true, "exec")); | |||
12852 | CI->setConvergent(); | |||
12853 | return CI; | |||
12854 | } | |||
12855 | case AMDGPU::BI__builtin_amdgcn_read_exec_lo: | |||
12856 | case AMDGPU::BI__builtin_amdgcn_read_exec_hi: { | |||
12857 | StringRef RegName = BuiltinID == AMDGPU::BI__builtin_amdgcn_read_exec_lo ? | |||
12858 | "exec_lo" : "exec_hi"; | |||
12859 | CallInst *CI = cast<CallInst>( | |||
12860 | EmitSpecialRegisterBuiltin(*this, E, Int32Ty, Int32Ty, true, RegName)); | |||
12861 | CI->setConvergent(); | |||
12862 | return CI; | |||
12863 | } | |||
12864 | // amdgcn workitem | |||
12865 | case AMDGPU::BI__builtin_amdgcn_workitem_id_x: | |||
12866 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_x, 0, 1024); | |||
12867 | case AMDGPU::BI__builtin_amdgcn_workitem_id_y: | |||
12868 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_y, 0, 1024); | |||
12869 | case AMDGPU::BI__builtin_amdgcn_workitem_id_z: | |||
12870 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_z, 0, 1024); | |||
12871 | ||||
12872 | // r600 intrinsics | |||
12873 | case AMDGPU::BI__builtin_r600_recipsqrt_ieee: | |||
12874 | case AMDGPU::BI__builtin_r600_recipsqrt_ieeef: | |||
12875 | return emitUnaryBuiltin(*this, E, Intrinsic::r600_recipsqrt_ieee); | |||
12876 | case AMDGPU::BI__builtin_r600_read_tidig_x: | |||
12877 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_x, 0, 1024); | |||
12878 | case AMDGPU::BI__builtin_r600_read_tidig_y: | |||
12879 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_y, 0, 1024); | |||
12880 | case AMDGPU::BI__builtin_r600_read_tidig_z: | |||
12881 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_z, 0, 1024); | |||
12882 | default: | |||
12883 | return nullptr; | |||
12884 | } | |||
12885 | } | |||
12886 | ||||
12887 | /// Handle a SystemZ function in which the final argument is a pointer | |||
12888 | /// to an int that receives the post-instruction CC value. At the LLVM level | |||
12889 | /// this is represented as a function that returns a {result, cc} pair. | |||
12890 | static Value *EmitSystemZIntrinsicWithCC(CodeGenFunction &CGF, | |||
12891 | unsigned IntrinsicID, | |||
12892 | const CallExpr *E) { | |||
12893 | unsigned NumArgs = E->getNumArgs() - 1; | |||
12894 | SmallVector<Value *, 8> Args(NumArgs); | |||
12895 | for (unsigned I = 0; I < NumArgs; ++I) | |||
12896 | Args[I] = CGF.EmitScalarExpr(E->getArg(I)); | |||
12897 | Address CCPtr = CGF.EmitPointerWithAlignment(E->getArg(NumArgs)); | |||
12898 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID); | |||
12899 | Value *Call = CGF.Builder.CreateCall(F, Args); | |||
12900 | Value *CC = CGF.Builder.CreateExtractValue(Call, 1); | |||
12901 | CGF.Builder.CreateStore(CC, CCPtr); | |||
12902 | return CGF.Builder.CreateExtractValue(Call, 0); | |||
12903 | } | |||
12904 | ||||
12905 | Value *CodeGenFunction::EmitSystemZBuiltinExpr(unsigned BuiltinID, | |||
12906 | const CallExpr *E) { | |||
12907 | switch (BuiltinID) { | |||
12908 | case SystemZ::BI__builtin_tbegin: { | |||
12909 | Value *TDB = EmitScalarExpr(E->getArg(0)); | |||
12910 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c); | |||
12911 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin); | |||
12912 | return Builder.CreateCall(F, {TDB, Control}); | |||
12913 | } | |||
12914 | case SystemZ::BI__builtin_tbegin_nofloat: { | |||
12915 | Value *TDB = EmitScalarExpr(E->getArg(0)); | |||
12916 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c); | |||
12917 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin_nofloat); | |||
12918 | return Builder.CreateCall(F, {TDB, Control}); | |||
12919 | } | |||
12920 | case SystemZ::BI__builtin_tbeginc: { | |||
12921 | Value *TDB = llvm::ConstantPointerNull::get(Int8PtrTy); | |||
12922 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff08); | |||
12923 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbeginc); | |||
12924 | return Builder.CreateCall(F, {TDB, Control}); | |||
12925 | } | |||
12926 | case SystemZ::BI__builtin_tabort: { | |||
12927 | Value *Data = EmitScalarExpr(E->getArg(0)); | |||
12928 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tabort); | |||
12929 | return Builder.CreateCall(F, Builder.CreateSExt(Data, Int64Ty, "tabort")); | |||
12930 | } | |||
12931 | case SystemZ::BI__builtin_non_tx_store: { | |||
12932 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
12933 | Value *Data = EmitScalarExpr(E->getArg(1)); | |||
12934 | Function *F = CGM.getIntrinsic(Intrinsic::s390_ntstg); | |||
12935 | return Builder.CreateCall(F, {Data, Address}); | |||
12936 | } | |||
12937 | ||||
12938 | // Vector builtins. Note that most vector builtins are mapped automatically | |||
12939 | // to target-specific LLVM intrinsics. The ones handled specially here can | |||
12940 | // be represented via standard LLVM IR, which is preferable to enable common | |||
12941 | // LLVM optimizations. | |||
12942 | ||||
12943 | case SystemZ::BI__builtin_s390_vpopctb: | |||
12944 | case SystemZ::BI__builtin_s390_vpopcth: | |||
12945 | case SystemZ::BI__builtin_s390_vpopctf: | |||
12946 | case SystemZ::BI__builtin_s390_vpopctg: { | |||
12947 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12948 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12949 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); | |||
12950 | return Builder.CreateCall(F, X); | |||
12951 | } | |||
12952 | ||||
12953 | case SystemZ::BI__builtin_s390_vclzb: | |||
12954 | case SystemZ::BI__builtin_s390_vclzh: | |||
12955 | case SystemZ::BI__builtin_s390_vclzf: | |||
12956 | case SystemZ::BI__builtin_s390_vclzg: { | |||
12957 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12958 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12959 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); | |||
12960 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType); | |||
12961 | return Builder.CreateCall(F, {X, Undef}); | |||
12962 | } | |||
12963 | ||||
12964 | case SystemZ::BI__builtin_s390_vctzb: | |||
12965 | case SystemZ::BI__builtin_s390_vctzh: | |||
12966 | case SystemZ::BI__builtin_s390_vctzf: | |||
12967 | case SystemZ::BI__builtin_s390_vctzg: { | |||
12968 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12969 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12970 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); | |||
12971 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType); | |||
12972 | return Builder.CreateCall(F, {X, Undef}); | |||
12973 | } | |||
12974 | ||||
12975 | case SystemZ::BI__builtin_s390_vfsqsb: | |||
12976 | case SystemZ::BI__builtin_s390_vfsqdb: { | |||
12977 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12978 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12979 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType); | |||
12980 | return Builder.CreateCall(F, X); | |||
12981 | } | |||
12982 | case SystemZ::BI__builtin_s390_vfmasb: | |||
12983 | case SystemZ::BI__builtin_s390_vfmadb: { | |||
12984 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12985 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12986 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
12987 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
12988 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
12989 | return Builder.CreateCall(F, {X, Y, Z}); | |||
12990 | } | |||
12991 | case SystemZ::BI__builtin_s390_vfmssb: | |||
12992 | case SystemZ::BI__builtin_s390_vfmsdb: { | |||
12993 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
12994 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
12995 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
12996 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
12997 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); | |||
12998 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
12999 | return Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")}); | |||
13000 | } | |||
13001 | case SystemZ::BI__builtin_s390_vfnmasb: | |||
13002 | case SystemZ::BI__builtin_s390_vfnmadb: { | |||
13003 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13004 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
13005 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
13006 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
13007 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); | |||
13008 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
13009 | return Builder.CreateFSub(Zero, Builder.CreateCall(F, {X, Y, Z}), "sub"); | |||
13010 | } | |||
13011 | case SystemZ::BI__builtin_s390_vfnmssb: | |||
13012 | case SystemZ::BI__builtin_s390_vfnmsdb: { | |||
13013 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13014 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
13015 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
13016 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
13017 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); | |||
13018 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
13019 | Value *NegZ = Builder.CreateFSub(Zero, Z, "sub"); | |||
13020 | return Builder.CreateFSub(Zero, Builder.CreateCall(F, {X, Y, NegZ})); | |||
13021 | } | |||
13022 | case SystemZ::BI__builtin_s390_vflpsb: | |||
13023 | case SystemZ::BI__builtin_s390_vflpdb: { | |||
13024 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13025 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
13026 | Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); | |||
13027 | return Builder.CreateCall(F, X); | |||
13028 | } | |||
13029 | case SystemZ::BI__builtin_s390_vflnsb: | |||
13030 | case SystemZ::BI__builtin_s390_vflndb: { | |||
13031 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13032 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
13033 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); | |||
13034 | Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); | |||
13035 | return Builder.CreateFSub(Zero, Builder.CreateCall(F, X), "sub"); | |||
13036 | } | |||
13037 | case SystemZ::BI__builtin_s390_vfisb: | |||
13038 | case SystemZ::BI__builtin_s390_vfidb: { | |||
13039 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13040 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
13041 | // Constant-fold the M4 and M5 mask arguments. | |||
13042 | llvm::APSInt M4, M5; | |||
13043 | bool IsConstM4 = E->getArg(1)->isIntegerConstantExpr(M4, getContext()); | |||
13044 | bool IsConstM5 = E->getArg(2)->isIntegerConstantExpr(M5, getContext()); | |||
13045 | assert(IsConstM4 && IsConstM5 && "Constant arg isn't actually constant?")((IsConstM4 && IsConstM5 && "Constant arg isn't actually constant?" ) ? static_cast<void> (0) : __assert_fail ("IsConstM4 && IsConstM5 && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13045, __PRETTY_FUNCTION__)); | |||
13046 | (void)IsConstM4; (void)IsConstM5; | |||
13047 | // Check whether this instance can be represented via a LLVM standard | |||
13048 | // intrinsic. We only support some combinations of M4 and M5. | |||
13049 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
13050 | switch (M4.getZExtValue()) { | |||
13051 | default: break; | |||
13052 | case 0: // IEEE-inexact exception allowed | |||
13053 | switch (M5.getZExtValue()) { | |||
13054 | default: break; | |||
13055 | case 0: ID = Intrinsic::rint; break; | |||
13056 | } | |||
13057 | break; | |||
13058 | case 4: // IEEE-inexact exception suppressed | |||
13059 | switch (M5.getZExtValue()) { | |||
13060 | default: break; | |||
13061 | case 0: ID = Intrinsic::nearbyint; break; | |||
13062 | case 1: ID = Intrinsic::round; break; | |||
13063 | case 5: ID = Intrinsic::trunc; break; | |||
13064 | case 6: ID = Intrinsic::ceil; break; | |||
13065 | case 7: ID = Intrinsic::floor; break; | |||
13066 | } | |||
13067 | break; | |||
13068 | } | |||
13069 | if (ID != Intrinsic::not_intrinsic) { | |||
13070 | Function *F = CGM.getIntrinsic(ID, ResultType); | |||
13071 | return Builder.CreateCall(F, X); | |||
13072 | } | |||
13073 | switch (BuiltinID) { | |||
13074 | case SystemZ::BI__builtin_s390_vfisb: ID = Intrinsic::s390_vfisb; break; | |||
13075 | case SystemZ::BI__builtin_s390_vfidb: ID = Intrinsic::s390_vfidb; break; | |||
13076 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13076); | |||
13077 | } | |||
13078 | Function *F = CGM.getIntrinsic(ID); | |||
13079 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); | |||
13080 | Value *M5Value = llvm::ConstantInt::get(getLLVMContext(), M5); | |||
13081 | return Builder.CreateCall(F, {X, M4Value, M5Value}); | |||
13082 | } | |||
13083 | case SystemZ::BI__builtin_s390_vfmaxsb: | |||
13084 | case SystemZ::BI__builtin_s390_vfmaxdb: { | |||
13085 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13086 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
13087 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
13088 | // Constant-fold the M4 mask argument. | |||
13089 | llvm::APSInt M4; | |||
13090 | bool IsConstM4 = E->getArg(2)->isIntegerConstantExpr(M4, getContext()); | |||
13091 | assert(IsConstM4 && "Constant arg isn't actually constant?")((IsConstM4 && "Constant arg isn't actually constant?" ) ? static_cast<void> (0) : __assert_fail ("IsConstM4 && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13091, __PRETTY_FUNCTION__)); | |||
13092 | (void)IsConstM4; | |||
13093 | // Check whether this instance can be represented via a LLVM standard | |||
13094 | // intrinsic. We only support some values of M4. | |||
13095 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
13096 | switch (M4.getZExtValue()) { | |||
13097 | default: break; | |||
13098 | case 4: ID = Intrinsic::maxnum; break; | |||
13099 | } | |||
13100 | if (ID != Intrinsic::not_intrinsic) { | |||
13101 | Function *F = CGM.getIntrinsic(ID, ResultType); | |||
13102 | return Builder.CreateCall(F, {X, Y}); | |||
13103 | } | |||
13104 | switch (BuiltinID) { | |||
13105 | case SystemZ::BI__builtin_s390_vfmaxsb: ID = Intrinsic::s390_vfmaxsb; break; | |||
13106 | case SystemZ::BI__builtin_s390_vfmaxdb: ID = Intrinsic::s390_vfmaxdb; break; | |||
13107 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13107); | |||
13108 | } | |||
13109 | Function *F = CGM.getIntrinsic(ID); | |||
13110 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); | |||
13111 | return Builder.CreateCall(F, {X, Y, M4Value}); | |||
13112 | } | |||
13113 | case SystemZ::BI__builtin_s390_vfminsb: | |||
13114 | case SystemZ::BI__builtin_s390_vfmindb: { | |||
13115 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13116 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
13117 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
13118 | // Constant-fold the M4 mask argument. | |||
13119 | llvm::APSInt M4; | |||
13120 | bool IsConstM4 = E->getArg(2)->isIntegerConstantExpr(M4, getContext()); | |||
13121 | assert(IsConstM4 && "Constant arg isn't actually constant?")((IsConstM4 && "Constant arg isn't actually constant?" ) ? static_cast<void> (0) : __assert_fail ("IsConstM4 && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13121, __PRETTY_FUNCTION__)); | |||
13122 | (void)IsConstM4; | |||
13123 | // Check whether this instance can be represented via a LLVM standard | |||
13124 | // intrinsic. We only support some values of M4. | |||
13125 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
13126 | switch (M4.getZExtValue()) { | |||
13127 | default: break; | |||
13128 | case 4: ID = Intrinsic::minnum; break; | |||
13129 | } | |||
13130 | if (ID != Intrinsic::not_intrinsic) { | |||
13131 | Function *F = CGM.getIntrinsic(ID, ResultType); | |||
13132 | return Builder.CreateCall(F, {X, Y}); | |||
13133 | } | |||
13134 | switch (BuiltinID) { | |||
13135 | case SystemZ::BI__builtin_s390_vfminsb: ID = Intrinsic::s390_vfminsb; break; | |||
13136 | case SystemZ::BI__builtin_s390_vfmindb: ID = Intrinsic::s390_vfmindb; break; | |||
13137 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13137); | |||
13138 | } | |||
13139 | Function *F = CGM.getIntrinsic(ID); | |||
13140 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); | |||
13141 | return Builder.CreateCall(F, {X, Y, M4Value}); | |||
13142 | } | |||
13143 | ||||
13144 | case SystemZ::BI__builtin_s390_vlbrh: | |||
13145 | case SystemZ::BI__builtin_s390_vlbrf: | |||
13146 | case SystemZ::BI__builtin_s390_vlbrg: { | |||
13147 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13148 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
13149 | Function *F = CGM.getIntrinsic(Intrinsic::bswap, ResultType); | |||
13150 | return Builder.CreateCall(F, X); | |||
13151 | } | |||
13152 | ||||
13153 | // Vector intrinsics that output the post-instruction CC value. | |||
13154 | ||||
13155 | #define INTRINSIC_WITH_CC(NAME) \ | |||
13156 | case SystemZ::BI__builtin_##NAME: \ | |||
13157 | return EmitSystemZIntrinsicWithCC(*this, Intrinsic::NAME, E) | |||
13158 | ||||
13159 | INTRINSIC_WITH_CC(s390_vpkshs); | |||
13160 | INTRINSIC_WITH_CC(s390_vpksfs); | |||
13161 | INTRINSIC_WITH_CC(s390_vpksgs); | |||
13162 | ||||
13163 | INTRINSIC_WITH_CC(s390_vpklshs); | |||
13164 | INTRINSIC_WITH_CC(s390_vpklsfs); | |||
13165 | INTRINSIC_WITH_CC(s390_vpklsgs); | |||
13166 | ||||
13167 | INTRINSIC_WITH_CC(s390_vceqbs); | |||
13168 | INTRINSIC_WITH_CC(s390_vceqhs); | |||
13169 | INTRINSIC_WITH_CC(s390_vceqfs); | |||
13170 | INTRINSIC_WITH_CC(s390_vceqgs); | |||
13171 | ||||
13172 | INTRINSIC_WITH_CC(s390_vchbs); | |||
13173 | INTRINSIC_WITH_CC(s390_vchhs); | |||
13174 | INTRINSIC_WITH_CC(s390_vchfs); | |||
13175 | INTRINSIC_WITH_CC(s390_vchgs); | |||
13176 | ||||
13177 | INTRINSIC_WITH_CC(s390_vchlbs); | |||
13178 | INTRINSIC_WITH_CC(s390_vchlhs); | |||
13179 | INTRINSIC_WITH_CC(s390_vchlfs); | |||
13180 | INTRINSIC_WITH_CC(s390_vchlgs); | |||
13181 | ||||
13182 | INTRINSIC_WITH_CC(s390_vfaebs); | |||
13183 | INTRINSIC_WITH_CC(s390_vfaehs); | |||
13184 | INTRINSIC_WITH_CC(s390_vfaefs); | |||
13185 | ||||
13186 | INTRINSIC_WITH_CC(s390_vfaezbs); | |||
13187 | INTRINSIC_WITH_CC(s390_vfaezhs); | |||
13188 | INTRINSIC_WITH_CC(s390_vfaezfs); | |||
13189 | ||||
13190 | INTRINSIC_WITH_CC(s390_vfeebs); | |||
13191 | INTRINSIC_WITH_CC(s390_vfeehs); | |||
13192 | INTRINSIC_WITH_CC(s390_vfeefs); | |||
13193 | ||||
13194 | INTRINSIC_WITH_CC(s390_vfeezbs); | |||
13195 | INTRINSIC_WITH_CC(s390_vfeezhs); | |||
13196 | INTRINSIC_WITH_CC(s390_vfeezfs); | |||
13197 | ||||
13198 | INTRINSIC_WITH_CC(s390_vfenebs); | |||
13199 | INTRINSIC_WITH_CC(s390_vfenehs); | |||
13200 | INTRINSIC_WITH_CC(s390_vfenefs); | |||
13201 | ||||
13202 | INTRINSIC_WITH_CC(s390_vfenezbs); | |||
13203 | INTRINSIC_WITH_CC(s390_vfenezhs); | |||
13204 | INTRINSIC_WITH_CC(s390_vfenezfs); | |||
13205 | ||||
13206 | INTRINSIC_WITH_CC(s390_vistrbs); | |||
13207 | INTRINSIC_WITH_CC(s390_vistrhs); | |||
13208 | INTRINSIC_WITH_CC(s390_vistrfs); | |||
13209 | ||||
13210 | INTRINSIC_WITH_CC(s390_vstrcbs); | |||
13211 | INTRINSIC_WITH_CC(s390_vstrchs); | |||
13212 | INTRINSIC_WITH_CC(s390_vstrcfs); | |||
13213 | ||||
13214 | INTRINSIC_WITH_CC(s390_vstrczbs); | |||
13215 | INTRINSIC_WITH_CC(s390_vstrczhs); | |||
13216 | INTRINSIC_WITH_CC(s390_vstrczfs); | |||
13217 | ||||
13218 | INTRINSIC_WITH_CC(s390_vfcesbs); | |||
13219 | INTRINSIC_WITH_CC(s390_vfcedbs); | |||
13220 | INTRINSIC_WITH_CC(s390_vfchsbs); | |||
13221 | INTRINSIC_WITH_CC(s390_vfchdbs); | |||
13222 | INTRINSIC_WITH_CC(s390_vfchesbs); | |||
13223 | INTRINSIC_WITH_CC(s390_vfchedbs); | |||
13224 | ||||
13225 | INTRINSIC_WITH_CC(s390_vftcisb); | |||
13226 | INTRINSIC_WITH_CC(s390_vftcidb); | |||
13227 | ||||
13228 | INTRINSIC_WITH_CC(s390_vstrsb); | |||
13229 | INTRINSIC_WITH_CC(s390_vstrsh); | |||
13230 | INTRINSIC_WITH_CC(s390_vstrsf); | |||
13231 | ||||
13232 | INTRINSIC_WITH_CC(s390_vstrszb); | |||
13233 | INTRINSIC_WITH_CC(s390_vstrszh); | |||
13234 | INTRINSIC_WITH_CC(s390_vstrszf); | |||
13235 | ||||
13236 | #undef INTRINSIC_WITH_CC | |||
13237 | ||||
13238 | default: | |||
13239 | return nullptr; | |||
13240 | } | |||
13241 | } | |||
13242 | ||||
13243 | namespace { | |||
13244 | // Helper classes for mapping MMA builtins to particular LLVM intrinsic variant. | |||
13245 | struct NVPTXMmaLdstInfo { | |||
13246 | unsigned NumResults; // Number of elements to load/store | |||
13247 | // Intrinsic IDs for row/col variants. 0 if particular layout is unsupported. | |||
13248 | unsigned IID_col; | |||
13249 | unsigned IID_row; | |||
13250 | }; | |||
13251 | ||||
13252 | #define MMA_INTR(geom_op_type, layout) \ | |||
13253 | Intrinsic::nvvm_wmma_##geom_op_type##_##layout##_stride | |||
13254 | #define MMA_LDST(n, geom_op_type) \ | |||
13255 | { n, MMA_INTR(geom_op_type, col), MMA_INTR(geom_op_type, row) } | |||
13256 | ||||
13257 | static NVPTXMmaLdstInfo getNVPTXMmaLdstInfo(unsigned BuiltinID) { | |||
13258 | switch (BuiltinID) { | |||
13259 | // FP MMA loads | |||
13260 | case NVPTX::BI__hmma_m16n16k16_ld_a: | |||
13261 | return MMA_LDST(8, m16n16k16_load_a_f16); | |||
13262 | case NVPTX::BI__hmma_m16n16k16_ld_b: | |||
13263 | return MMA_LDST(8, m16n16k16_load_b_f16); | |||
13264 | case NVPTX::BI__hmma_m16n16k16_ld_c_f16: | |||
13265 | return MMA_LDST(4, m16n16k16_load_c_f16); | |||
13266 | case NVPTX::BI__hmma_m16n16k16_ld_c_f32: | |||
13267 | return MMA_LDST(8, m16n16k16_load_c_f32); | |||
13268 | case NVPTX::BI__hmma_m32n8k16_ld_a: | |||
13269 | return MMA_LDST(8, m32n8k16_load_a_f16); | |||
13270 | case NVPTX::BI__hmma_m32n8k16_ld_b: | |||
13271 | return MMA_LDST(8, m32n8k16_load_b_f16); | |||
13272 | case NVPTX::BI__hmma_m32n8k16_ld_c_f16: | |||
13273 | return MMA_LDST(4, m32n8k16_load_c_f16); | |||
13274 | case NVPTX::BI__hmma_m32n8k16_ld_c_f32: | |||
13275 | return MMA_LDST(8, m32n8k16_load_c_f32); | |||
13276 | case NVPTX::BI__hmma_m8n32k16_ld_a: | |||
13277 | return MMA_LDST(8, m8n32k16_load_a_f16); | |||
13278 | case NVPTX::BI__hmma_m8n32k16_ld_b: | |||
13279 | return MMA_LDST(8, m8n32k16_load_b_f16); | |||
13280 | case NVPTX::BI__hmma_m8n32k16_ld_c_f16: | |||
13281 | return MMA_LDST(4, m8n32k16_load_c_f16); | |||
13282 | case NVPTX::BI__hmma_m8n32k16_ld_c_f32: | |||
13283 | return MMA_LDST(8, m8n32k16_load_c_f32); | |||
13284 | ||||
13285 | // Integer MMA loads | |||
13286 | case NVPTX::BI__imma_m16n16k16_ld_a_s8: | |||
13287 | return MMA_LDST(2, m16n16k16_load_a_s8); | |||
13288 | case NVPTX::BI__imma_m16n16k16_ld_a_u8: | |||
13289 | return MMA_LDST(2, m16n16k16_load_a_u8); | |||
13290 | case NVPTX::BI__imma_m16n16k16_ld_b_s8: | |||
13291 | return MMA_LDST(2, m16n16k16_load_b_s8); | |||
13292 | case NVPTX::BI__imma_m16n16k16_ld_b_u8: | |||
13293 | return MMA_LDST(2, m16n16k16_load_b_u8); | |||
13294 | case NVPTX::BI__imma_m16n16k16_ld_c: | |||
13295 | return MMA_LDST(8, m16n16k16_load_c_s32); | |||
13296 | case NVPTX::BI__imma_m32n8k16_ld_a_s8: | |||
13297 | return MMA_LDST(4, m32n8k16_load_a_s8); | |||
13298 | case NVPTX::BI__imma_m32n8k16_ld_a_u8: | |||
13299 | return MMA_LDST(4, m32n8k16_load_a_u8); | |||
13300 | case NVPTX::BI__imma_m32n8k16_ld_b_s8: | |||
13301 | return MMA_LDST(1, m32n8k16_load_b_s8); | |||
13302 | case NVPTX::BI__imma_m32n8k16_ld_b_u8: | |||
13303 | return MMA_LDST(1, m32n8k16_load_b_u8); | |||
13304 | case NVPTX::BI__imma_m32n8k16_ld_c: | |||
13305 | return MMA_LDST(8, m32n8k16_load_c_s32); | |||
13306 | case NVPTX::BI__imma_m8n32k16_ld_a_s8: | |||
13307 | return MMA_LDST(1, m8n32k16_load_a_s8); | |||
13308 | case NVPTX::BI__imma_m8n32k16_ld_a_u8: | |||
13309 | return MMA_LDST(1, m8n32k16_load_a_u8); | |||
13310 | case NVPTX::BI__imma_m8n32k16_ld_b_s8: | |||
13311 | return MMA_LDST(4, m8n32k16_load_b_s8); | |||
13312 | case NVPTX::BI__imma_m8n32k16_ld_b_u8: | |||
13313 | return MMA_LDST(4, m8n32k16_load_b_u8); | |||
13314 | case NVPTX::BI__imma_m8n32k16_ld_c: | |||
13315 | return MMA_LDST(8, m8n32k16_load_c_s32); | |||
13316 | ||||
13317 | // Sub-integer MMA loads. | |||
13318 | // Only row/col layout is supported by A/B fragments. | |||
13319 | case NVPTX::BI__imma_m8n8k32_ld_a_s4: | |||
13320 | return {1, 0, MMA_INTR(m8n8k32_load_a_s4, row)}; | |||
13321 | case NVPTX::BI__imma_m8n8k32_ld_a_u4: | |||
13322 | return {1, 0, MMA_INTR(m8n8k32_load_a_u4, row)}; | |||
13323 | case NVPTX::BI__imma_m8n8k32_ld_b_s4: | |||
13324 | return {1, MMA_INTR(m8n8k32_load_b_s4, col), 0}; | |||
13325 | case NVPTX::BI__imma_m8n8k32_ld_b_u4: | |||
13326 | return {1, MMA_INTR(m8n8k32_load_b_u4, col), 0}; | |||
13327 | case NVPTX::BI__imma_m8n8k32_ld_c: | |||
13328 | return MMA_LDST(2, m8n8k32_load_c_s32); | |||
13329 | case NVPTX::BI__bmma_m8n8k128_ld_a_b1: | |||
13330 | return {1, 0, MMA_INTR(m8n8k128_load_a_b1, row)}; | |||
13331 | case NVPTX::BI__bmma_m8n8k128_ld_b_b1: | |||
13332 | return {1, MMA_INTR(m8n8k128_load_b_b1, col), 0}; | |||
13333 | case NVPTX::BI__bmma_m8n8k128_ld_c: | |||
13334 | return MMA_LDST(2, m8n8k128_load_c_s32); | |||
13335 | ||||
13336 | // NOTE: We need to follow inconsitent naming scheme used by NVCC. Unlike | |||
13337 | // PTX and LLVM IR where stores always use fragment D, NVCC builtins always | |||
13338 | // use fragment C for both loads and stores. | |||
13339 | // FP MMA stores. | |||
13340 | case NVPTX::BI__hmma_m16n16k16_st_c_f16: | |||
13341 | return MMA_LDST(4, m16n16k16_store_d_f16); | |||
13342 | case NVPTX::BI__hmma_m16n16k16_st_c_f32: | |||
13343 | return MMA_LDST(8, m16n16k16_store_d_f32); | |||
13344 | case NVPTX::BI__hmma_m32n8k16_st_c_f16: | |||
13345 | return MMA_LDST(4, m32n8k16_store_d_f16); | |||
13346 | case NVPTX::BI__hmma_m32n8k16_st_c_f32: | |||
13347 | return MMA_LDST(8, m32n8k16_store_d_f32); | |||
13348 | case NVPTX::BI__hmma_m8n32k16_st_c_f16: | |||
13349 | return MMA_LDST(4, m8n32k16_store_d_f16); | |||
13350 | case NVPTX::BI__hmma_m8n32k16_st_c_f32: | |||
13351 | return MMA_LDST(8, m8n32k16_store_d_f32); | |||
13352 | ||||
13353 | // Integer and sub-integer MMA stores. | |||
13354 | // Another naming quirk. Unlike other MMA builtins that use PTX types in the | |||
13355 | // name, integer loads/stores use LLVM's i32. | |||
13356 | case NVPTX::BI__imma_m16n16k16_st_c_i32: | |||
13357 | return MMA_LDST(8, m16n16k16_store_d_s32); | |||
13358 | case NVPTX::BI__imma_m32n8k16_st_c_i32: | |||
13359 | return MMA_LDST(8, m32n8k16_store_d_s32); | |||
13360 | case NVPTX::BI__imma_m8n32k16_st_c_i32: | |||
13361 | return MMA_LDST(8, m8n32k16_store_d_s32); | |||
13362 | case NVPTX::BI__imma_m8n8k32_st_c_i32: | |||
13363 | return MMA_LDST(2, m8n8k32_store_d_s32); | |||
13364 | case NVPTX::BI__bmma_m8n8k128_st_c_i32: | |||
13365 | return MMA_LDST(2, m8n8k128_store_d_s32); | |||
13366 | ||||
13367 | default: | |||
13368 | llvm_unreachable("Unknown MMA builtin")::llvm::llvm_unreachable_internal("Unknown MMA builtin", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13368); | |||
13369 | } | |||
13370 | } | |||
13371 | #undef MMA_LDST | |||
13372 | #undef MMA_INTR | |||
13373 | ||||
13374 | ||||
13375 | struct NVPTXMmaInfo { | |||
13376 | unsigned NumEltsA; | |||
13377 | unsigned NumEltsB; | |||
13378 | unsigned NumEltsC; | |||
13379 | unsigned NumEltsD; | |||
13380 | std::array<unsigned, 8> Variants; | |||
13381 | ||||
13382 | unsigned getMMAIntrinsic(int Layout, bool Satf) { | |||
13383 | unsigned Index = Layout * 2 + Satf; | |||
13384 | if (Index >= Variants.size()) | |||
13385 | return 0; | |||
13386 | return Variants[Index]; | |||
13387 | } | |||
13388 | }; | |||
13389 | ||||
13390 | // Returns an intrinsic that matches Layout and Satf for valid combinations of | |||
13391 | // Layout and Satf, 0 otherwise. | |||
13392 | static NVPTXMmaInfo getNVPTXMmaInfo(unsigned BuiltinID) { | |||
13393 | // clang-format off | |||
13394 | #define MMA_VARIANTS(geom, type) {{ \ | |||
13395 | Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type, \ | |||
13396 | Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type##_satfinite, \ | |||
13397 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \ | |||
13398 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \ | |||
13399 | Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type, \ | |||
13400 | Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type##_satfinite, \ | |||
13401 | Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type, \ | |||
13402 | Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type##_satfinite \ | |||
13403 | }} | |||
13404 | // Sub-integer MMA only supports row.col layout. | |||
13405 | #define MMA_VARIANTS_I4(geom, type) {{ \ | |||
13406 | 0, \ | |||
13407 | 0, \ | |||
13408 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \ | |||
13409 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \ | |||
13410 | 0, \ | |||
13411 | 0, \ | |||
13412 | 0, \ | |||
13413 | 0 \ | |||
13414 | }} | |||
13415 | // b1 MMA does not support .satfinite. | |||
13416 | #define MMA_VARIANTS_B1(geom, type) {{ \ | |||
13417 | 0, \ | |||
13418 | 0, \ | |||
13419 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \ | |||
13420 | 0, \ | |||
13421 | 0, \ | |||
13422 | 0, \ | |||
13423 | 0, \ | |||
13424 | 0 \ | |||
13425 | }} | |||
13426 | // clang-format on | |||
13427 | switch (BuiltinID) { | |||
13428 | // FP MMA | |||
13429 | // Note that 'type' argument of MMA_VARIANT uses D_C notation, while | |||
13430 | // NumEltsN of return value are ordered as A,B,C,D. | |||
13431 | case NVPTX::BI__hmma_m16n16k16_mma_f16f16: | |||
13432 | return {8, 8, 4, 4, MMA_VARIANTS(m16n16k16, f16_f16)}; | |||
13433 | case NVPTX::BI__hmma_m16n16k16_mma_f32f16: | |||
13434 | return {8, 8, 4, 8, MMA_VARIANTS(m16n16k16, f32_f16)}; | |||
13435 | case NVPTX::BI__hmma_m16n16k16_mma_f16f32: | |||
13436 | return {8, 8, 8, 4, MMA_VARIANTS(m16n16k16, f16_f32)}; | |||
13437 | case NVPTX::BI__hmma_m16n16k16_mma_f32f32: | |||
13438 | return {8, 8, 8, 8, MMA_VARIANTS(m16n16k16, f32_f32)}; | |||
13439 | case NVPTX::BI__hmma_m32n8k16_mma_f16f16: | |||
13440 | return {8, 8, 4, 4, MMA_VARIANTS(m32n8k16, f16_f16)}; | |||
13441 | case NVPTX::BI__hmma_m32n8k16_mma_f32f16: | |||
13442 | return {8, 8, 4, 8, MMA_VARIANTS(m32n8k16, f32_f16)}; | |||
13443 | case NVPTX::BI__hmma_m32n8k16_mma_f16f32: | |||
13444 | return {8, 8, 8, 4, MMA_VARIANTS(m32n8k16, f16_f32)}; | |||
13445 | case NVPTX::BI__hmma_m32n8k16_mma_f32f32: | |||
13446 | return {8, 8, 8, 8, MMA_VARIANTS(m32n8k16, f32_f32)}; | |||
13447 | case NVPTX::BI__hmma_m8n32k16_mma_f16f16: | |||
13448 | return {8, 8, 4, 4, MMA_VARIANTS(m8n32k16, f16_f16)}; | |||
13449 | case NVPTX::BI__hmma_m8n32k16_mma_f32f16: | |||
13450 | return {8, 8, 4, 8, MMA_VARIANTS(m8n32k16, f32_f16)}; | |||
13451 | case NVPTX::BI__hmma_m8n32k16_mma_f16f32: | |||
13452 | return {8, 8, 8, 4, MMA_VARIANTS(m8n32k16, f16_f32)}; | |||
13453 | case NVPTX::BI__hmma_m8n32k16_mma_f32f32: | |||
13454 | return {8, 8, 8, 8, MMA_VARIANTS(m8n32k16, f32_f32)}; | |||
13455 | ||||
13456 | // Integer MMA | |||
13457 | case NVPTX::BI__imma_m16n16k16_mma_s8: | |||
13458 | return {2, 2, 8, 8, MMA_VARIANTS(m16n16k16, s8)}; | |||
13459 | case NVPTX::BI__imma_m16n16k16_mma_u8: | |||
13460 | return {2, 2, 8, 8, MMA_VARIANTS(m16n16k16, u8)}; | |||
13461 | case NVPTX::BI__imma_m32n8k16_mma_s8: | |||
13462 | return {4, 1, 8, 8, MMA_VARIANTS(m32n8k16, s8)}; | |||
13463 | case NVPTX::BI__imma_m32n8k16_mma_u8: | |||
13464 | return {4, 1, 8, 8, MMA_VARIANTS(m32n8k16, u8)}; | |||
13465 | case NVPTX::BI__imma_m8n32k16_mma_s8: | |||
13466 | return {1, 4, 8, 8, MMA_VARIANTS(m8n32k16, s8)}; | |||
13467 | case NVPTX::BI__imma_m8n32k16_mma_u8: | |||
13468 | return {1, 4, 8, 8, MMA_VARIANTS(m8n32k16, u8)}; | |||
13469 | ||||
13470 | // Sub-integer MMA | |||
13471 | case NVPTX::BI__imma_m8n8k32_mma_s4: | |||
13472 | return {1, 1, 2, 2, MMA_VARIANTS_I4(m8n8k32, s4)}; | |||
13473 | case NVPTX::BI__imma_m8n8k32_mma_u4: | |||
13474 | return {1, 1, 2, 2, MMA_VARIANTS_I4(m8n8k32, u4)}; | |||
13475 | case NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1: | |||
13476 | return {1, 1, 2, 2, MMA_VARIANTS_B1(m8n8k128, b1)}; | |||
13477 | default: | |||
13478 | llvm_unreachable("Unexpected builtin ID.")::llvm::llvm_unreachable_internal("Unexpected builtin ID.", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13478); | |||
13479 | } | |||
13480 | #undef MMA_VARIANTS | |||
13481 | #undef MMA_VARIANTS_I4 | |||
13482 | #undef MMA_VARIANTS_B1 | |||
13483 | } | |||
13484 | ||||
13485 | } // namespace | |||
13486 | ||||
13487 | Value * | |||
13488 | CodeGenFunction::EmitNVPTXBuiltinExpr(unsigned BuiltinID, const CallExpr *E) { | |||
13489 | auto MakeLdg = [&](unsigned IntrinsicID) { | |||
13490 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
13491 | clang::CharUnits Align = | |||
13492 | getNaturalPointeeTypeAlignment(E->getArg(0)->getType()); | |||
13493 | return Builder.CreateCall( | |||
13494 | CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(), | |||
13495 | Ptr->getType()}), | |||
13496 | {Ptr, ConstantInt::get(Builder.getInt32Ty(), Align.getQuantity())}); | |||
13497 | }; | |||
13498 | auto MakeScopedAtomic = [&](unsigned IntrinsicID) { | |||
13499 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
13500 | return Builder.CreateCall( | |||
13501 | CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(), | |||
13502 | Ptr->getType()}), | |||
13503 | {Ptr, EmitScalarExpr(E->getArg(1))}); | |||
13504 | }; | |||
13505 | switch (BuiltinID) { | |||
13506 | case NVPTX::BI__nvvm_atom_add_gen_i: | |||
13507 | case NVPTX::BI__nvvm_atom_add_gen_l: | |||
13508 | case NVPTX::BI__nvvm_atom_add_gen_ll: | |||
13509 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Add, E); | |||
13510 | ||||
13511 | case NVPTX::BI__nvvm_atom_sub_gen_i: | |||
13512 | case NVPTX::BI__nvvm_atom_sub_gen_l: | |||
13513 | case NVPTX::BI__nvvm_atom_sub_gen_ll: | |||
13514 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Sub, E); | |||
13515 | ||||
13516 | case NVPTX::BI__nvvm_atom_and_gen_i: | |||
13517 | case NVPTX::BI__nvvm_atom_and_gen_l: | |||
13518 | case NVPTX::BI__nvvm_atom_and_gen_ll: | |||
13519 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::And, E); | |||
13520 | ||||
13521 | case NVPTX::BI__nvvm_atom_or_gen_i: | |||
13522 | case NVPTX::BI__nvvm_atom_or_gen_l: | |||
13523 | case NVPTX::BI__nvvm_atom_or_gen_ll: | |||
13524 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Or, E); | |||
13525 | ||||
13526 | case NVPTX::BI__nvvm_atom_xor_gen_i: | |||
13527 | case NVPTX::BI__nvvm_atom_xor_gen_l: | |||
13528 | case NVPTX::BI__nvvm_atom_xor_gen_ll: | |||
13529 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xor, E); | |||
13530 | ||||
13531 | case NVPTX::BI__nvvm_atom_xchg_gen_i: | |||
13532 | case NVPTX::BI__nvvm_atom_xchg_gen_l: | |||
13533 | case NVPTX::BI__nvvm_atom_xchg_gen_ll: | |||
13534 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xchg, E); | |||
13535 | ||||
13536 | case NVPTX::BI__nvvm_atom_max_gen_i: | |||
13537 | case NVPTX::BI__nvvm_atom_max_gen_l: | |||
13538 | case NVPTX::BI__nvvm_atom_max_gen_ll: | |||
13539 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Max, E); | |||
13540 | ||||
13541 | case NVPTX::BI__nvvm_atom_max_gen_ui: | |||
13542 | case NVPTX::BI__nvvm_atom_max_gen_ul: | |||
13543 | case NVPTX::BI__nvvm_atom_max_gen_ull: | |||
13544 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMax, E); | |||
13545 | ||||
13546 | case NVPTX::BI__nvvm_atom_min_gen_i: | |||
13547 | case NVPTX::BI__nvvm_atom_min_gen_l: | |||
13548 | case NVPTX::BI__nvvm_atom_min_gen_ll: | |||
13549 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Min, E); | |||
13550 | ||||
13551 | case NVPTX::BI__nvvm_atom_min_gen_ui: | |||
13552 | case NVPTX::BI__nvvm_atom_min_gen_ul: | |||
13553 | case NVPTX::BI__nvvm_atom_min_gen_ull: | |||
13554 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMin, E); | |||
13555 | ||||
13556 | case NVPTX::BI__nvvm_atom_cas_gen_i: | |||
13557 | case NVPTX::BI__nvvm_atom_cas_gen_l: | |||
13558 | case NVPTX::BI__nvvm_atom_cas_gen_ll: | |||
13559 | // __nvvm_atom_cas_gen_* should return the old value rather than the | |||
13560 | // success flag. | |||
13561 | return MakeAtomicCmpXchgValue(*this, E, /*ReturnBool=*/false); | |||
13562 | ||||
13563 | case NVPTX::BI__nvvm_atom_add_gen_f: | |||
13564 | case NVPTX::BI__nvvm_atom_add_gen_d: { | |||
13565 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
13566 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
13567 | return Builder.CreateAtomicRMW(llvm::AtomicRMWInst::FAdd, Ptr, Val, | |||
13568 | AtomicOrdering::SequentiallyConsistent); | |||
13569 | } | |||
13570 | ||||
13571 | case NVPTX::BI__nvvm_atom_inc_gen_ui: { | |||
13572 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
13573 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
13574 | Function *FnALI32 = | |||
13575 | CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_inc_32, Ptr->getType()); | |||
13576 | return Builder.CreateCall(FnALI32, {Ptr, Val}); | |||
13577 | } | |||
13578 | ||||
13579 | case NVPTX::BI__nvvm_atom_dec_gen_ui: { | |||
13580 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
13581 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
13582 | Function *FnALD32 = | |||
13583 | CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_dec_32, Ptr->getType()); | |||
13584 | return Builder.CreateCall(FnALD32, {Ptr, Val}); | |||
13585 | } | |||
13586 | ||||
13587 | case NVPTX::BI__nvvm_ldg_c: | |||
13588 | case NVPTX::BI__nvvm_ldg_c2: | |||
13589 | case NVPTX::BI__nvvm_ldg_c4: | |||
13590 | case NVPTX::BI__nvvm_ldg_s: | |||
13591 | case NVPTX::BI__nvvm_ldg_s2: | |||
13592 | case NVPTX::BI__nvvm_ldg_s4: | |||
13593 | case NVPTX::BI__nvvm_ldg_i: | |||
13594 | case NVPTX::BI__nvvm_ldg_i2: | |||
13595 | case NVPTX::BI__nvvm_ldg_i4: | |||
13596 | case NVPTX::BI__nvvm_ldg_l: | |||
13597 | case NVPTX::BI__nvvm_ldg_ll: | |||
13598 | case NVPTX::BI__nvvm_ldg_ll2: | |||
13599 | case NVPTX::BI__nvvm_ldg_uc: | |||
13600 | case NVPTX::BI__nvvm_ldg_uc2: | |||
13601 | case NVPTX::BI__nvvm_ldg_uc4: | |||
13602 | case NVPTX::BI__nvvm_ldg_us: | |||
13603 | case NVPTX::BI__nvvm_ldg_us2: | |||
13604 | case NVPTX::BI__nvvm_ldg_us4: | |||
13605 | case NVPTX::BI__nvvm_ldg_ui: | |||
13606 | case NVPTX::BI__nvvm_ldg_ui2: | |||
13607 | case NVPTX::BI__nvvm_ldg_ui4: | |||
13608 | case NVPTX::BI__nvvm_ldg_ul: | |||
13609 | case NVPTX::BI__nvvm_ldg_ull: | |||
13610 | case NVPTX::BI__nvvm_ldg_ull2: | |||
13611 | // PTX Interoperability section 2.2: "For a vector with an even number of | |||
13612 | // elements, its alignment is set to number of elements times the alignment | |||
13613 | // of its member: n*alignof(t)." | |||
13614 | return MakeLdg(Intrinsic::nvvm_ldg_global_i); | |||
13615 | case NVPTX::BI__nvvm_ldg_f: | |||
13616 | case NVPTX::BI__nvvm_ldg_f2: | |||
13617 | case NVPTX::BI__nvvm_ldg_f4: | |||
13618 | case NVPTX::BI__nvvm_ldg_d: | |||
13619 | case NVPTX::BI__nvvm_ldg_d2: | |||
13620 | return MakeLdg(Intrinsic::nvvm_ldg_global_f); | |||
13621 | ||||
13622 | case NVPTX::BI__nvvm_atom_cta_add_gen_i: | |||
13623 | case NVPTX::BI__nvvm_atom_cta_add_gen_l: | |||
13624 | case NVPTX::BI__nvvm_atom_cta_add_gen_ll: | |||
13625 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_cta); | |||
13626 | case NVPTX::BI__nvvm_atom_sys_add_gen_i: | |||
13627 | case NVPTX::BI__nvvm_atom_sys_add_gen_l: | |||
13628 | case NVPTX::BI__nvvm_atom_sys_add_gen_ll: | |||
13629 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_sys); | |||
13630 | case NVPTX::BI__nvvm_atom_cta_add_gen_f: | |||
13631 | case NVPTX::BI__nvvm_atom_cta_add_gen_d: | |||
13632 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_cta); | |||
13633 | case NVPTX::BI__nvvm_atom_sys_add_gen_f: | |||
13634 | case NVPTX::BI__nvvm_atom_sys_add_gen_d: | |||
13635 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_sys); | |||
13636 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_i: | |||
13637 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_l: | |||
13638 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_ll: | |||
13639 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_cta); | |||
13640 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_i: | |||
13641 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_l: | |||
13642 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_ll: | |||
13643 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_sys); | |||
13644 | case NVPTX::BI__nvvm_atom_cta_max_gen_i: | |||
13645 | case NVPTX::BI__nvvm_atom_cta_max_gen_ui: | |||
13646 | case NVPTX::BI__nvvm_atom_cta_max_gen_l: | |||
13647 | case NVPTX::BI__nvvm_atom_cta_max_gen_ul: | |||
13648 | case NVPTX::BI__nvvm_atom_cta_max_gen_ll: | |||
13649 | case NVPTX::BI__nvvm_atom_cta_max_gen_ull: | |||
13650 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_cta); | |||
13651 | case NVPTX::BI__nvvm_atom_sys_max_gen_i: | |||
13652 | case NVPTX::BI__nvvm_atom_sys_max_gen_ui: | |||
13653 | case NVPTX::BI__nvvm_atom_sys_max_gen_l: | |||
13654 | case NVPTX::BI__nvvm_atom_sys_max_gen_ul: | |||
13655 | case NVPTX::BI__nvvm_atom_sys_max_gen_ll: | |||
13656 | case NVPTX::BI__nvvm_atom_sys_max_gen_ull: | |||
13657 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_sys); | |||
13658 | case NVPTX::BI__nvvm_atom_cta_min_gen_i: | |||
13659 | case NVPTX::BI__nvvm_atom_cta_min_gen_ui: | |||
13660 | case NVPTX::BI__nvvm_atom_cta_min_gen_l: | |||
13661 | case NVPTX::BI__nvvm_atom_cta_min_gen_ul: | |||
13662 | case NVPTX::BI__nvvm_atom_cta_min_gen_ll: | |||
13663 | case NVPTX::BI__nvvm_atom_cta_min_gen_ull: | |||
13664 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_cta); | |||
13665 | case NVPTX::BI__nvvm_atom_sys_min_gen_i: | |||
13666 | case NVPTX::BI__nvvm_atom_sys_min_gen_ui: | |||
13667 | case NVPTX::BI__nvvm_atom_sys_min_gen_l: | |||
13668 | case NVPTX::BI__nvvm_atom_sys_min_gen_ul: | |||
13669 | case NVPTX::BI__nvvm_atom_sys_min_gen_ll: | |||
13670 | case NVPTX::BI__nvvm_atom_sys_min_gen_ull: | |||
13671 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_sys); | |||
13672 | case NVPTX::BI__nvvm_atom_cta_inc_gen_ui: | |||
13673 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_cta); | |||
13674 | case NVPTX::BI__nvvm_atom_cta_dec_gen_ui: | |||
13675 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_cta); | |||
13676 | case NVPTX::BI__nvvm_atom_sys_inc_gen_ui: | |||
13677 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_sys); | |||
13678 | case NVPTX::BI__nvvm_atom_sys_dec_gen_ui: | |||
13679 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_sys); | |||
13680 | case NVPTX::BI__nvvm_atom_cta_and_gen_i: | |||
13681 | case NVPTX::BI__nvvm_atom_cta_and_gen_l: | |||
13682 | case NVPTX::BI__nvvm_atom_cta_and_gen_ll: | |||
13683 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_cta); | |||
13684 | case NVPTX::BI__nvvm_atom_sys_and_gen_i: | |||
13685 | case NVPTX::BI__nvvm_atom_sys_and_gen_l: | |||
13686 | case NVPTX::BI__nvvm_atom_sys_and_gen_ll: | |||
13687 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_sys); | |||
13688 | case NVPTX::BI__nvvm_atom_cta_or_gen_i: | |||
13689 | case NVPTX::BI__nvvm_atom_cta_or_gen_l: | |||
13690 | case NVPTX::BI__nvvm_atom_cta_or_gen_ll: | |||
13691 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_cta); | |||
13692 | case NVPTX::BI__nvvm_atom_sys_or_gen_i: | |||
13693 | case NVPTX::BI__nvvm_atom_sys_or_gen_l: | |||
13694 | case NVPTX::BI__nvvm_atom_sys_or_gen_ll: | |||
13695 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_sys); | |||
13696 | case NVPTX::BI__nvvm_atom_cta_xor_gen_i: | |||
13697 | case NVPTX::BI__nvvm_atom_cta_xor_gen_l: | |||
13698 | case NVPTX::BI__nvvm_atom_cta_xor_gen_ll: | |||
13699 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_cta); | |||
13700 | case NVPTX::BI__nvvm_atom_sys_xor_gen_i: | |||
13701 | case NVPTX::BI__nvvm_atom_sys_xor_gen_l: | |||
13702 | case NVPTX::BI__nvvm_atom_sys_xor_gen_ll: | |||
13703 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_sys); | |||
13704 | case NVPTX::BI__nvvm_atom_cta_cas_gen_i: | |||
13705 | case NVPTX::BI__nvvm_atom_cta_cas_gen_l: | |||
13706 | case NVPTX::BI__nvvm_atom_cta_cas_gen_ll: { | |||
13707 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
13708 | return Builder.CreateCall( | |||
13709 | CGM.getIntrinsic( | |||
13710 | Intrinsic::nvvm_atomic_cas_gen_i_cta, | |||
13711 | {Ptr->getType()->getPointerElementType(), Ptr->getType()}), | |||
13712 | {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))}); | |||
13713 | } | |||
13714 | case NVPTX::BI__nvvm_atom_sys_cas_gen_i: | |||
13715 | case NVPTX::BI__nvvm_atom_sys_cas_gen_l: | |||
13716 | case NVPTX::BI__nvvm_atom_sys_cas_gen_ll: { | |||
13717 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
13718 | return Builder.CreateCall( | |||
13719 | CGM.getIntrinsic( | |||
13720 | Intrinsic::nvvm_atomic_cas_gen_i_sys, | |||
13721 | {Ptr->getType()->getPointerElementType(), Ptr->getType()}), | |||
13722 | {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))}); | |||
13723 | } | |||
13724 | case NVPTX::BI__nvvm_match_all_sync_i32p: | |||
13725 | case NVPTX::BI__nvvm_match_all_sync_i64p: { | |||
13726 | Value *Mask = EmitScalarExpr(E->getArg(0)); | |||
13727 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
13728 | Address PredOutPtr = EmitPointerWithAlignment(E->getArg(2)); | |||
13729 | Value *ResultPair = Builder.CreateCall( | |||
13730 | CGM.getIntrinsic(BuiltinID == NVPTX::BI__nvvm_match_all_sync_i32p | |||
13731 | ? Intrinsic::nvvm_match_all_sync_i32p | |||
13732 | : Intrinsic::nvvm_match_all_sync_i64p), | |||
13733 | {Mask, Val}); | |||
13734 | Value *Pred = Builder.CreateZExt(Builder.CreateExtractValue(ResultPair, 1), | |||
13735 | PredOutPtr.getElementType()); | |||
13736 | Builder.CreateStore(Pred, PredOutPtr); | |||
13737 | return Builder.CreateExtractValue(ResultPair, 0); | |||
13738 | } | |||
13739 | ||||
13740 | // FP MMA loads | |||
13741 | case NVPTX::BI__hmma_m16n16k16_ld_a: | |||
13742 | case NVPTX::BI__hmma_m16n16k16_ld_b: | |||
13743 | case NVPTX::BI__hmma_m16n16k16_ld_c_f16: | |||
13744 | case NVPTX::BI__hmma_m16n16k16_ld_c_f32: | |||
13745 | case NVPTX::BI__hmma_m32n8k16_ld_a: | |||
13746 | case NVPTX::BI__hmma_m32n8k16_ld_b: | |||
13747 | case NVPTX::BI__hmma_m32n8k16_ld_c_f16: | |||
13748 | case NVPTX::BI__hmma_m32n8k16_ld_c_f32: | |||
13749 | case NVPTX::BI__hmma_m8n32k16_ld_a: | |||
13750 | case NVPTX::BI__hmma_m8n32k16_ld_b: | |||
13751 | case NVPTX::BI__hmma_m8n32k16_ld_c_f16: | |||
13752 | case NVPTX::BI__hmma_m8n32k16_ld_c_f32: | |||
13753 | // Integer MMA loads. | |||
13754 | case NVPTX::BI__imma_m16n16k16_ld_a_s8: | |||
13755 | case NVPTX::BI__imma_m16n16k16_ld_a_u8: | |||
13756 | case NVPTX::BI__imma_m16n16k16_ld_b_s8: | |||
13757 | case NVPTX::BI__imma_m16n16k16_ld_b_u8: | |||
13758 | case NVPTX::BI__imma_m16n16k16_ld_c: | |||
13759 | case NVPTX::BI__imma_m32n8k16_ld_a_s8: | |||
13760 | case NVPTX::BI__imma_m32n8k16_ld_a_u8: | |||
13761 | case NVPTX::BI__imma_m32n8k16_ld_b_s8: | |||
13762 | case NVPTX::BI__imma_m32n8k16_ld_b_u8: | |||
13763 | case NVPTX::BI__imma_m32n8k16_ld_c: | |||
13764 | case NVPTX::BI__imma_m8n32k16_ld_a_s8: | |||
13765 | case NVPTX::BI__imma_m8n32k16_ld_a_u8: | |||
13766 | case NVPTX::BI__imma_m8n32k16_ld_b_s8: | |||
13767 | case NVPTX::BI__imma_m8n32k16_ld_b_u8: | |||
13768 | case NVPTX::BI__imma_m8n32k16_ld_c: | |||
13769 | // Sub-integer MMA loads. | |||
13770 | case NVPTX::BI__imma_m8n8k32_ld_a_s4: | |||
13771 | case NVPTX::BI__imma_m8n8k32_ld_a_u4: | |||
13772 | case NVPTX::BI__imma_m8n8k32_ld_b_s4: | |||
13773 | case NVPTX::BI__imma_m8n8k32_ld_b_u4: | |||
13774 | case NVPTX::BI__imma_m8n8k32_ld_c: | |||
13775 | case NVPTX::BI__bmma_m8n8k128_ld_a_b1: | |||
13776 | case NVPTX::BI__bmma_m8n8k128_ld_b_b1: | |||
13777 | case NVPTX::BI__bmma_m8n8k128_ld_c: | |||
13778 | { | |||
13779 | Address Dst = EmitPointerWithAlignment(E->getArg(0)); | |||
13780 | Value *Src = EmitScalarExpr(E->getArg(1)); | |||
13781 | Value *Ldm = EmitScalarExpr(E->getArg(2)); | |||
13782 | llvm::APSInt isColMajorArg; | |||
13783 | if (!E->getArg(3)->isIntegerConstantExpr(isColMajorArg, getContext())) | |||
13784 | return nullptr; | |||
13785 | bool isColMajor = isColMajorArg.getSExtValue(); | |||
13786 | NVPTXMmaLdstInfo II = getNVPTXMmaLdstInfo(BuiltinID); | |||
13787 | unsigned IID = isColMajor ? II.IID_col : II.IID_row; | |||
13788 | if (IID == 0) | |||
13789 | return nullptr; | |||
13790 | ||||
13791 | Value *Result = | |||
13792 | Builder.CreateCall(CGM.getIntrinsic(IID, Src->getType()), {Src, Ldm}); | |||
13793 | ||||
13794 | // Save returned values. | |||
13795 | assert(II.NumResults)((II.NumResults) ? static_cast<void> (0) : __assert_fail ("II.NumResults", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13795, __PRETTY_FUNCTION__)); | |||
13796 | if (II.NumResults == 1) { | |||
13797 | Builder.CreateAlignedStore(Result, Dst.getPointer(), | |||
13798 | CharUnits::fromQuantity(4)); | |||
13799 | } else { | |||
13800 | for (unsigned i = 0; i < II.NumResults; ++i) { | |||
13801 | Builder.CreateAlignedStore( | |||
13802 | Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), | |||
13803 | Dst.getElementType()), | |||
13804 | Builder.CreateGEP(Dst.getPointer(), | |||
13805 | llvm::ConstantInt::get(IntTy, i)), | |||
13806 | CharUnits::fromQuantity(4)); | |||
13807 | } | |||
13808 | } | |||
13809 | return Result; | |||
13810 | } | |||
13811 | ||||
13812 | case NVPTX::BI__hmma_m16n16k16_st_c_f16: | |||
13813 | case NVPTX::BI__hmma_m16n16k16_st_c_f32: | |||
13814 | case NVPTX::BI__hmma_m32n8k16_st_c_f16: | |||
13815 | case NVPTX::BI__hmma_m32n8k16_st_c_f32: | |||
13816 | case NVPTX::BI__hmma_m8n32k16_st_c_f16: | |||
13817 | case NVPTX::BI__hmma_m8n32k16_st_c_f32: | |||
13818 | case NVPTX::BI__imma_m16n16k16_st_c_i32: | |||
13819 | case NVPTX::BI__imma_m32n8k16_st_c_i32: | |||
13820 | case NVPTX::BI__imma_m8n32k16_st_c_i32: | |||
13821 | case NVPTX::BI__imma_m8n8k32_st_c_i32: | |||
13822 | case NVPTX::BI__bmma_m8n8k128_st_c_i32: { | |||
13823 | Value *Dst = EmitScalarExpr(E->getArg(0)); | |||
13824 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
13825 | Value *Ldm = EmitScalarExpr(E->getArg(2)); | |||
13826 | llvm::APSInt isColMajorArg; | |||
13827 | if (!E->getArg(3)->isIntegerConstantExpr(isColMajorArg, getContext())) | |||
13828 | return nullptr; | |||
13829 | bool isColMajor = isColMajorArg.getSExtValue(); | |||
13830 | NVPTXMmaLdstInfo II = getNVPTXMmaLdstInfo(BuiltinID); | |||
13831 | unsigned IID = isColMajor ? II.IID_col : II.IID_row; | |||
13832 | if (IID == 0) | |||
13833 | return nullptr; | |||
13834 | Function *Intrinsic = | |||
13835 | CGM.getIntrinsic(IID, Dst->getType()); | |||
13836 | llvm::Type *ParamType = Intrinsic->getFunctionType()->getParamType(1); | |||
13837 | SmallVector<Value *, 10> Values = {Dst}; | |||
13838 | for (unsigned i = 0; i < II.NumResults; ++i) { | |||
13839 | Value *V = Builder.CreateAlignedLoad( | |||
13840 | Builder.CreateGEP(Src.getPointer(), llvm::ConstantInt::get(IntTy, i)), | |||
13841 | CharUnits::fromQuantity(4)); | |||
13842 | Values.push_back(Builder.CreateBitCast(V, ParamType)); | |||
13843 | } | |||
13844 | Values.push_back(Ldm); | |||
13845 | Value *Result = Builder.CreateCall(Intrinsic, Values); | |||
13846 | return Result; | |||
13847 | } | |||
13848 | ||||
13849 | // BI__hmma_m16n16k16_mma_<Dtype><CType>(d, a, b, c, layout, satf) --> | |||
13850 | // Intrinsic::nvvm_wmma_m16n16k16_mma_sync<layout A,B><DType><CType><Satf> | |||
13851 | case NVPTX::BI__hmma_m16n16k16_mma_f16f16: | |||
13852 | case NVPTX::BI__hmma_m16n16k16_mma_f32f16: | |||
13853 | case NVPTX::BI__hmma_m16n16k16_mma_f32f32: | |||
13854 | case NVPTX::BI__hmma_m16n16k16_mma_f16f32: | |||
13855 | case NVPTX::BI__hmma_m32n8k16_mma_f16f16: | |||
13856 | case NVPTX::BI__hmma_m32n8k16_mma_f32f16: | |||
13857 | case NVPTX::BI__hmma_m32n8k16_mma_f32f32: | |||
13858 | case NVPTX::BI__hmma_m32n8k16_mma_f16f32: | |||
13859 | case NVPTX::BI__hmma_m8n32k16_mma_f16f16: | |||
13860 | case NVPTX::BI__hmma_m8n32k16_mma_f32f16: | |||
13861 | case NVPTX::BI__hmma_m8n32k16_mma_f32f32: | |||
13862 | case NVPTX::BI__hmma_m8n32k16_mma_f16f32: | |||
13863 | case NVPTX::BI__imma_m16n16k16_mma_s8: | |||
13864 | case NVPTX::BI__imma_m16n16k16_mma_u8: | |||
13865 | case NVPTX::BI__imma_m32n8k16_mma_s8: | |||
13866 | case NVPTX::BI__imma_m32n8k16_mma_u8: | |||
13867 | case NVPTX::BI__imma_m8n32k16_mma_s8: | |||
13868 | case NVPTX::BI__imma_m8n32k16_mma_u8: | |||
13869 | case NVPTX::BI__imma_m8n8k32_mma_s4: | |||
13870 | case NVPTX::BI__imma_m8n8k32_mma_u4: | |||
13871 | case NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1: { | |||
13872 | Address Dst = EmitPointerWithAlignment(E->getArg(0)); | |||
13873 | Address SrcA = EmitPointerWithAlignment(E->getArg(1)); | |||
13874 | Address SrcB = EmitPointerWithAlignment(E->getArg(2)); | |||
13875 | Address SrcC = EmitPointerWithAlignment(E->getArg(3)); | |||
13876 | llvm::APSInt LayoutArg; | |||
13877 | if (!E->getArg(4)->isIntegerConstantExpr(LayoutArg, getContext())) | |||
13878 | return nullptr; | |||
13879 | int Layout = LayoutArg.getSExtValue(); | |||
13880 | if (Layout < 0 || Layout > 3) | |||
13881 | return nullptr; | |||
13882 | llvm::APSInt SatfArg; | |||
13883 | if (BuiltinID == NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1) | |||
13884 | SatfArg = 0; // .b1 does not have satf argument. | |||
13885 | else if (!E->getArg(5)->isIntegerConstantExpr(SatfArg, getContext())) | |||
13886 | return nullptr; | |||
13887 | bool Satf = SatfArg.getSExtValue(); | |||
13888 | NVPTXMmaInfo MI = getNVPTXMmaInfo(BuiltinID); | |||
13889 | unsigned IID = MI.getMMAIntrinsic(Layout, Satf); | |||
13890 | if (IID == 0) // Unsupported combination of Layout/Satf. | |||
13891 | return nullptr; | |||
13892 | ||||
13893 | SmallVector<Value *, 24> Values; | |||
13894 | Function *Intrinsic = CGM.getIntrinsic(IID); | |||
13895 | llvm::Type *AType = Intrinsic->getFunctionType()->getParamType(0); | |||
13896 | // Load A | |||
13897 | for (unsigned i = 0; i < MI.NumEltsA; ++i) { | |||
13898 | Value *V = Builder.CreateAlignedLoad( | |||
13899 | Builder.CreateGEP(SrcA.getPointer(), | |||
13900 | llvm::ConstantInt::get(IntTy, i)), | |||
13901 | CharUnits::fromQuantity(4)); | |||
13902 | Values.push_back(Builder.CreateBitCast(V, AType)); | |||
13903 | } | |||
13904 | // Load B | |||
13905 | llvm::Type *BType = Intrinsic->getFunctionType()->getParamType(MI.NumEltsA); | |||
13906 | for (unsigned i = 0; i < MI.NumEltsB; ++i) { | |||
13907 | Value *V = Builder.CreateAlignedLoad( | |||
13908 | Builder.CreateGEP(SrcB.getPointer(), | |||
13909 | llvm::ConstantInt::get(IntTy, i)), | |||
13910 | CharUnits::fromQuantity(4)); | |||
13911 | Values.push_back(Builder.CreateBitCast(V, BType)); | |||
13912 | } | |||
13913 | // Load C | |||
13914 | llvm::Type *CType = | |||
13915 | Intrinsic->getFunctionType()->getParamType(MI.NumEltsA + MI.NumEltsB); | |||
13916 | for (unsigned i = 0; i < MI.NumEltsC; ++i) { | |||
13917 | Value *V = Builder.CreateAlignedLoad( | |||
13918 | Builder.CreateGEP(SrcC.getPointer(), | |||
13919 | llvm::ConstantInt::get(IntTy, i)), | |||
13920 | CharUnits::fromQuantity(4)); | |||
13921 | Values.push_back(Builder.CreateBitCast(V, CType)); | |||
13922 | } | |||
13923 | Value *Result = Builder.CreateCall(Intrinsic, Values); | |||
13924 | llvm::Type *DType = Dst.getElementType(); | |||
13925 | for (unsigned i = 0; i < MI.NumEltsD; ++i) | |||
13926 | Builder.CreateAlignedStore( | |||
13927 | Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), DType), | |||
13928 | Builder.CreateGEP(Dst.getPointer(), llvm::ConstantInt::get(IntTy, i)), | |||
13929 | CharUnits::fromQuantity(4)); | |||
13930 | return Result; | |||
13931 | } | |||
13932 | default: | |||
13933 | return nullptr; | |||
13934 | } | |||
13935 | } | |||
13936 | ||||
13937 | Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID, | |||
13938 | const CallExpr *E) { | |||
13939 | switch (BuiltinID) { | |||
13940 | case WebAssembly::BI__builtin_wasm_memory_size: { | |||
13941 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13942 | Value *I = EmitScalarExpr(E->getArg(0)); | |||
13943 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_size, ResultType); | |||
13944 | return Builder.CreateCall(Callee, I); | |||
13945 | } | |||
13946 | case WebAssembly::BI__builtin_wasm_memory_grow: { | |||
13947 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13948 | Value *Args[] = { | |||
13949 | EmitScalarExpr(E->getArg(0)), | |||
13950 | EmitScalarExpr(E->getArg(1)) | |||
13951 | }; | |||
13952 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_grow, ResultType); | |||
13953 | return Builder.CreateCall(Callee, Args); | |||
13954 | } | |||
13955 | case WebAssembly::BI__builtin_wasm_memory_init: { | |||
13956 | llvm::APSInt SegConst; | |||
13957 | if (!E->getArg(0)->isIntegerConstantExpr(SegConst, getContext())) | |||
13958 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13958); | |||
13959 | llvm::APSInt MemConst; | |||
13960 | if (!E->getArg(1)->isIntegerConstantExpr(MemConst, getContext())) | |||
13961 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13961); | |||
13962 | if (!MemConst.isNullValue()) | |||
13963 | ErrorUnsupported(E, "non-zero memory index"); | |||
13964 | Value *Args[] = {llvm::ConstantInt::get(getLLVMContext(), SegConst), | |||
13965 | llvm::ConstantInt::get(getLLVMContext(), MemConst), | |||
13966 | EmitScalarExpr(E->getArg(2)), EmitScalarExpr(E->getArg(3)), | |||
13967 | EmitScalarExpr(E->getArg(4))}; | |||
13968 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_init); | |||
13969 | return Builder.CreateCall(Callee, Args); | |||
13970 | } | |||
13971 | case WebAssembly::BI__builtin_wasm_data_drop: { | |||
13972 | llvm::APSInt SegConst; | |||
13973 | if (!E->getArg(0)->isIntegerConstantExpr(SegConst, getContext())) | |||
13974 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13974); | |||
13975 | Value *Arg = llvm::ConstantInt::get(getLLVMContext(), SegConst); | |||
13976 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_data_drop); | |||
13977 | return Builder.CreateCall(Callee, {Arg}); | |||
13978 | } | |||
13979 | case WebAssembly::BI__builtin_wasm_tls_size: { | |||
13980 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13981 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_size, ResultType); | |||
13982 | return Builder.CreateCall(Callee); | |||
13983 | } | |||
13984 | case WebAssembly::BI__builtin_wasm_tls_align: { | |||
13985 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
13986 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_align, ResultType); | |||
13987 | return Builder.CreateCall(Callee); | |||
13988 | } | |||
13989 | case WebAssembly::BI__builtin_wasm_tls_base: { | |||
13990 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_base); | |||
13991 | return Builder.CreateCall(Callee); | |||
13992 | } | |||
13993 | case WebAssembly::BI__builtin_wasm_throw: { | |||
13994 | Value *Tag = EmitScalarExpr(E->getArg(0)); | |||
13995 | Value *Obj = EmitScalarExpr(E->getArg(1)); | |||
13996 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_throw); | |||
13997 | return Builder.CreateCall(Callee, {Tag, Obj}); | |||
13998 | } | |||
13999 | case WebAssembly::BI__builtin_wasm_rethrow_in_catch: { | |||
14000 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_rethrow_in_catch); | |||
14001 | return Builder.CreateCall(Callee); | |||
14002 | } | |||
14003 | case WebAssembly::BI__builtin_wasm_atomic_wait_i32: { | |||
14004 | Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
14005 | Value *Expected = EmitScalarExpr(E->getArg(1)); | |||
14006 | Value *Timeout = EmitScalarExpr(E->getArg(2)); | |||
14007 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_wait_i32); | |||
14008 | return Builder.CreateCall(Callee, {Addr, Expected, Timeout}); | |||
14009 | } | |||
14010 | case WebAssembly::BI__builtin_wasm_atomic_wait_i64: { | |||
14011 | Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
14012 | Value *Expected = EmitScalarExpr(E->getArg(1)); | |||
14013 | Value *Timeout = EmitScalarExpr(E->getArg(2)); | |||
14014 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_wait_i64); | |||
14015 | return Builder.CreateCall(Callee, {Addr, Expected, Timeout}); | |||
14016 | } | |||
14017 | case WebAssembly::BI__builtin_wasm_atomic_notify: { | |||
14018 | Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
14019 | Value *Count = EmitScalarExpr(E->getArg(1)); | |||
14020 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_notify); | |||
14021 | return Builder.CreateCall(Callee, {Addr, Count}); | |||
14022 | } | |||
14023 | case WebAssembly::BI__builtin_wasm_trunc_s_i32_f32: | |||
14024 | case WebAssembly::BI__builtin_wasm_trunc_s_i32_f64: | |||
14025 | case WebAssembly::BI__builtin_wasm_trunc_s_i64_f32: | |||
14026 | case WebAssembly::BI__builtin_wasm_trunc_s_i64_f64: { | |||
14027 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
14028 | llvm::Type *ResT = ConvertType(E->getType()); | |||
14029 | Function *Callee = | |||
14030 | CGM.getIntrinsic(Intrinsic::wasm_trunc_signed, {ResT, Src->getType()}); | |||
14031 | return Builder.CreateCall(Callee, {Src}); | |||
14032 | } | |||
14033 | case WebAssembly::BI__builtin_wasm_trunc_u_i32_f32: | |||
14034 | case WebAssembly::BI__builtin_wasm_trunc_u_i32_f64: | |||
14035 | case WebAssembly::BI__builtin_wasm_trunc_u_i64_f32: | |||
14036 | case WebAssembly::BI__builtin_wasm_trunc_u_i64_f64: { | |||
14037 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
14038 | llvm::Type *ResT = ConvertType(E->getType()); | |||
14039 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_unsigned, | |||
14040 | {ResT, Src->getType()}); | |||
14041 | return Builder.CreateCall(Callee, {Src}); | |||
14042 | } | |||
14043 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f32: | |||
14044 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f64: | |||
14045 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f32: | |||
14046 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f64: | |||
14047 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32x4_f32x4: | |||
14048 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64x2_f64x2: { | |||
14049 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
14050 | llvm::Type *ResT = ConvertType(E->getType()); | |||
14051 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_saturate_signed, | |||
14052 | {ResT, Src->getType()}); | |||
14053 | return Builder.CreateCall(Callee, {Src}); | |||
14054 | } | |||
14055 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f32: | |||
14056 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f64: | |||
14057 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f32: | |||
14058 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f64: | |||
14059 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32x4_f32x4: | |||
14060 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64x2_f64x2: { | |||
14061 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
14062 | llvm::Type *ResT = ConvertType(E->getType()); | |||
14063 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_saturate_unsigned, | |||
14064 | {ResT, Src->getType()}); | |||
14065 | return Builder.CreateCall(Callee, {Src}); | |||
14066 | } | |||
14067 | case WebAssembly::BI__builtin_wasm_min_f32: | |||
14068 | case WebAssembly::BI__builtin_wasm_min_f64: | |||
14069 | case WebAssembly::BI__builtin_wasm_min_f32x4: | |||
14070 | case WebAssembly::BI__builtin_wasm_min_f64x2: { | |||
14071 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
14072 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
14073 | Function *Callee = CGM.getIntrinsic(Intrinsic::minimum, | |||
14074 | ConvertType(E->getType())); | |||
14075 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
14076 | } | |||
14077 | case WebAssembly::BI__builtin_wasm_max_f32: | |||
14078 | case WebAssembly::BI__builtin_wasm_max_f64: | |||
14079 | case WebAssembly::BI__builtin_wasm_max_f32x4: | |||
14080 | case WebAssembly::BI__builtin_wasm_max_f64x2: { | |||
14081 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
14082 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
14083 | Function *Callee = CGM.getIntrinsic(Intrinsic::maximum, | |||
14084 | ConvertType(E->getType())); | |||
14085 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
14086 | } | |||
14087 | case WebAssembly::BI__builtin_wasm_swizzle_v8x16: { | |||
14088 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
14089 | Value *Indices = EmitScalarExpr(E->getArg(1)); | |||
14090 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_swizzle); | |||
14091 | return Builder.CreateCall(Callee, {Src, Indices}); | |||
14092 | } | |||
14093 | case WebAssembly::BI__builtin_wasm_extract_lane_s_i8x16: | |||
14094 | case WebAssembly::BI__builtin_wasm_extract_lane_u_i8x16: | |||
14095 | case WebAssembly::BI__builtin_wasm_extract_lane_s_i16x8: | |||
14096 | case WebAssembly::BI__builtin_wasm_extract_lane_u_i16x8: | |||
14097 | case WebAssembly::BI__builtin_wasm_extract_lane_i32x4: | |||
14098 | case WebAssembly::BI__builtin_wasm_extract_lane_i64x2: | |||
14099 | case WebAssembly::BI__builtin_wasm_extract_lane_f32x4: | |||
14100 | case WebAssembly::BI__builtin_wasm_extract_lane_f64x2: { | |||
14101 | llvm::APSInt LaneConst; | |||
14102 | if (!E->getArg(1)->isIntegerConstantExpr(LaneConst, getContext())) | |||
14103 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14103); | |||
14104 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
14105 | Value *Lane = llvm::ConstantInt::get(getLLVMContext(), LaneConst); | |||
14106 | Value *Extract = Builder.CreateExtractElement(Vec, Lane); | |||
14107 | switch (BuiltinID) { | |||
14108 | case WebAssembly::BI__builtin_wasm_extract_lane_s_i8x16: | |||
14109 | case WebAssembly::BI__builtin_wasm_extract_lane_s_i16x8: | |||
14110 | return Builder.CreateSExt(Extract, ConvertType(E->getType())); | |||
14111 | case WebAssembly::BI__builtin_wasm_extract_lane_u_i8x16: | |||
14112 | case WebAssembly::BI__builtin_wasm_extract_lane_u_i16x8: | |||
14113 | return Builder.CreateZExt(Extract, ConvertType(E->getType())); | |||
14114 | case WebAssembly::BI__builtin_wasm_extract_lane_i32x4: | |||
14115 | case WebAssembly::BI__builtin_wasm_extract_lane_i64x2: | |||
14116 | case WebAssembly::BI__builtin_wasm_extract_lane_f32x4: | |||
14117 | case WebAssembly::BI__builtin_wasm_extract_lane_f64x2: | |||
14118 | return Extract; | |||
14119 | default: | |||
14120 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14120); | |||
14121 | } | |||
14122 | } | |||
14123 | case WebAssembly::BI__builtin_wasm_replace_lane_i8x16: | |||
14124 | case WebAssembly::BI__builtin_wasm_replace_lane_i16x8: | |||
14125 | case WebAssembly::BI__builtin_wasm_replace_lane_i32x4: | |||
14126 | case WebAssembly::BI__builtin_wasm_replace_lane_i64x2: | |||
14127 | case WebAssembly::BI__builtin_wasm_replace_lane_f32x4: | |||
14128 | case WebAssembly::BI__builtin_wasm_replace_lane_f64x2: { | |||
14129 | llvm::APSInt LaneConst; | |||
14130 | if (!E->getArg(1)->isIntegerConstantExpr(LaneConst, getContext())) | |||
14131 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14131); | |||
14132 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
14133 | Value *Lane = llvm::ConstantInt::get(getLLVMContext(), LaneConst); | |||
14134 | Value *Val = EmitScalarExpr(E->getArg(2)); | |||
14135 | switch (BuiltinID) { | |||
14136 | case WebAssembly::BI__builtin_wasm_replace_lane_i8x16: | |||
14137 | case WebAssembly::BI__builtin_wasm_replace_lane_i16x8: { | |||
14138 | llvm::Type *ElemType = ConvertType(E->getType())->getVectorElementType(); | |||
14139 | Value *Trunc = Builder.CreateTrunc(Val, ElemType); | |||
14140 | return Builder.CreateInsertElement(Vec, Trunc, Lane); | |||
14141 | } | |||
14142 | case WebAssembly::BI__builtin_wasm_replace_lane_i32x4: | |||
14143 | case WebAssembly::BI__builtin_wasm_replace_lane_i64x2: | |||
14144 | case WebAssembly::BI__builtin_wasm_replace_lane_f32x4: | |||
14145 | case WebAssembly::BI__builtin_wasm_replace_lane_f64x2: | |||
14146 | return Builder.CreateInsertElement(Vec, Val, Lane); | |||
14147 | default: | |||
14148 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14148); | |||
14149 | } | |||
14150 | } | |||
14151 | case WebAssembly::BI__builtin_wasm_add_saturate_s_i8x16: | |||
14152 | case WebAssembly::BI__builtin_wasm_add_saturate_u_i8x16: | |||
14153 | case WebAssembly::BI__builtin_wasm_add_saturate_s_i16x8: | |||
14154 | case WebAssembly::BI__builtin_wasm_add_saturate_u_i16x8: | |||
14155 | case WebAssembly::BI__builtin_wasm_sub_saturate_s_i8x16: | |||
14156 | case WebAssembly::BI__builtin_wasm_sub_saturate_u_i8x16: | |||
14157 | case WebAssembly::BI__builtin_wasm_sub_saturate_s_i16x8: | |||
14158 | case WebAssembly::BI__builtin_wasm_sub_saturate_u_i16x8: { | |||
14159 | unsigned IntNo; | |||
14160 | switch (BuiltinID) { | |||
14161 | case WebAssembly::BI__builtin_wasm_add_saturate_s_i8x16: | |||
14162 | case WebAssembly::BI__builtin_wasm_add_saturate_s_i16x8: | |||
14163 | IntNo = Intrinsic::sadd_sat; | |||
14164 | break; | |||
14165 | case WebAssembly::BI__builtin_wasm_add_saturate_u_i8x16: | |||
14166 | case WebAssembly::BI__builtin_wasm_add_saturate_u_i16x8: | |||
14167 | IntNo = Intrinsic::uadd_sat; | |||
14168 | break; | |||
14169 | case WebAssembly::BI__builtin_wasm_sub_saturate_s_i8x16: | |||
14170 | case WebAssembly::BI__builtin_wasm_sub_saturate_s_i16x8: | |||
14171 | IntNo = Intrinsic::wasm_sub_saturate_signed; | |||
14172 | break; | |||
14173 | case WebAssembly::BI__builtin_wasm_sub_saturate_u_i8x16: | |||
14174 | case WebAssembly::BI__builtin_wasm_sub_saturate_u_i16x8: | |||
14175 | IntNo = Intrinsic::wasm_sub_saturate_unsigned; | |||
14176 | break; | |||
14177 | default: | |||
14178 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14178); | |||
14179 | } | |||
14180 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
14181 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
14182 | Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType())); | |||
14183 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
14184 | } | |||
14185 | case WebAssembly::BI__builtin_wasm_bitselect: { | |||
14186 | Value *V1 = EmitScalarExpr(E->getArg(0)); | |||
14187 | Value *V2 = EmitScalarExpr(E->getArg(1)); | |||
14188 | Value *C = EmitScalarExpr(E->getArg(2)); | |||
14189 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_bitselect, | |||
14190 | ConvertType(E->getType())); | |||
14191 | return Builder.CreateCall(Callee, {V1, V2, C}); | |||
14192 | } | |||
14193 | case WebAssembly::BI__builtin_wasm_any_true_i8x16: | |||
14194 | case WebAssembly::BI__builtin_wasm_any_true_i16x8: | |||
14195 | case WebAssembly::BI__builtin_wasm_any_true_i32x4: | |||
14196 | case WebAssembly::BI__builtin_wasm_any_true_i64x2: | |||
14197 | case WebAssembly::BI__builtin_wasm_all_true_i8x16: | |||
14198 | case WebAssembly::BI__builtin_wasm_all_true_i16x8: | |||
14199 | case WebAssembly::BI__builtin_wasm_all_true_i32x4: | |||
14200 | case WebAssembly::BI__builtin_wasm_all_true_i64x2: { | |||
14201 | unsigned IntNo; | |||
14202 | switch (BuiltinID) { | |||
14203 | case WebAssembly::BI__builtin_wasm_any_true_i8x16: | |||
14204 | case WebAssembly::BI__builtin_wasm_any_true_i16x8: | |||
14205 | case WebAssembly::BI__builtin_wasm_any_true_i32x4: | |||
14206 | case WebAssembly::BI__builtin_wasm_any_true_i64x2: | |||
14207 | IntNo = Intrinsic::wasm_anytrue; | |||
14208 | break; | |||
14209 | case WebAssembly::BI__builtin_wasm_all_true_i8x16: | |||
14210 | case WebAssembly::BI__builtin_wasm_all_true_i16x8: | |||
14211 | case WebAssembly::BI__builtin_wasm_all_true_i32x4: | |||
14212 | case WebAssembly::BI__builtin_wasm_all_true_i64x2: | |||
14213 | IntNo = Intrinsic::wasm_alltrue; | |||
14214 | break; | |||
14215 | default: | |||
14216 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14216); | |||
14217 | } | |||
14218 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
14219 | Function *Callee = CGM.getIntrinsic(IntNo, Vec->getType()); | |||
14220 | return Builder.CreateCall(Callee, {Vec}); | |||
14221 | } | |||
14222 | case WebAssembly::BI__builtin_wasm_abs_f32x4: | |||
14223 | case WebAssembly::BI__builtin_wasm_abs_f64x2: { | |||
14224 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
14225 | Function *Callee = CGM.getIntrinsic(Intrinsic::fabs, Vec->getType()); | |||
14226 | return Builder.CreateCall(Callee, {Vec}); | |||
14227 | } | |||
14228 | case WebAssembly::BI__builtin_wasm_sqrt_f32x4: | |||
14229 | case WebAssembly::BI__builtin_wasm_sqrt_f64x2: { | |||
14230 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
14231 | Function *Callee = CGM.getIntrinsic(Intrinsic::sqrt, Vec->getType()); | |||
14232 | return Builder.CreateCall(Callee, {Vec}); | |||
14233 | } | |||
14234 | case WebAssembly::BI__builtin_wasm_qfma_f32x4: | |||
14235 | case WebAssembly::BI__builtin_wasm_qfms_f32x4: | |||
14236 | case WebAssembly::BI__builtin_wasm_qfma_f64x2: | |||
14237 | case WebAssembly::BI__builtin_wasm_qfms_f64x2: { | |||
14238 | Value *A = EmitScalarExpr(E->getArg(0)); | |||
14239 | Value *B = EmitScalarExpr(E->getArg(1)); | |||
14240 | Value *C = EmitScalarExpr(E->getArg(2)); | |||
14241 | unsigned IntNo; | |||
14242 | switch (BuiltinID) { | |||
14243 | case WebAssembly::BI__builtin_wasm_qfma_f32x4: | |||
14244 | case WebAssembly::BI__builtin_wasm_qfma_f64x2: | |||
14245 | IntNo = Intrinsic::wasm_qfma; | |||
14246 | break; | |||
14247 | case WebAssembly::BI__builtin_wasm_qfms_f32x4: | |||
14248 | case WebAssembly::BI__builtin_wasm_qfms_f64x2: | |||
14249 | IntNo = Intrinsic::wasm_qfms; | |||
14250 | break; | |||
14251 | default: | |||
14252 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14252); | |||
14253 | } | |||
14254 | Function *Callee = CGM.getIntrinsic(IntNo, A->getType()); | |||
14255 | return Builder.CreateCall(Callee, {A, B, C}); | |||
14256 | } | |||
14257 | case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8: | |||
14258 | case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8: | |||
14259 | case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4: | |||
14260 | case WebAssembly::BI__builtin_wasm_narrow_u_i16x8_i32x4: { | |||
14261 | Value *Low = EmitScalarExpr(E->getArg(0)); | |||
14262 | Value *High = EmitScalarExpr(E->getArg(1)); | |||
14263 | unsigned IntNo; | |||
14264 | switch (BuiltinID) { | |||
14265 | case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8: | |||
14266 | case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4: | |||
14267 | IntNo = Intrinsic::wasm_narrow_signed; | |||
14268 | break; | |||
14269 | case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8: | |||
14270 | case WebAssembly::BI__builtin_wasm_narrow_u_i16x8_i32x4: | |||
14271 | IntNo = Intrinsic::wasm_narrow_unsigned; | |||
14272 | break; | |||
14273 | default: | |||
14274 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14274); | |||
14275 | } | |||
14276 | Function *Callee = | |||
14277 | CGM.getIntrinsic(IntNo, {ConvertType(E->getType()), Low->getType()}); | |||
14278 | return Builder.CreateCall(Callee, {Low, High}); | |||
14279 | } | |||
14280 | case WebAssembly::BI__builtin_wasm_widen_low_s_i16x8_i8x16: | |||
14281 | case WebAssembly::BI__builtin_wasm_widen_high_s_i16x8_i8x16: | |||
14282 | case WebAssembly::BI__builtin_wasm_widen_low_u_i16x8_i8x16: | |||
14283 | case WebAssembly::BI__builtin_wasm_widen_high_u_i16x8_i8x16: | |||
14284 | case WebAssembly::BI__builtin_wasm_widen_low_s_i32x4_i16x8: | |||
14285 | case WebAssembly::BI__builtin_wasm_widen_high_s_i32x4_i16x8: | |||
14286 | case WebAssembly::BI__builtin_wasm_widen_low_u_i32x4_i16x8: | |||
14287 | case WebAssembly::BI__builtin_wasm_widen_high_u_i32x4_i16x8: { | |||
14288 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
14289 | unsigned IntNo; | |||
14290 | switch (BuiltinID) { | |||
14291 | case WebAssembly::BI__builtin_wasm_widen_low_s_i16x8_i8x16: | |||
14292 | case WebAssembly::BI__builtin_wasm_widen_low_s_i32x4_i16x8: | |||
14293 | IntNo = Intrinsic::wasm_widen_low_signed; | |||
14294 | break; | |||
14295 | case WebAssembly::BI__builtin_wasm_widen_high_s_i16x8_i8x16: | |||
14296 | case WebAssembly::BI__builtin_wasm_widen_high_s_i32x4_i16x8: | |||
14297 | IntNo = Intrinsic::wasm_widen_high_signed; | |||
14298 | break; | |||
14299 | case WebAssembly::BI__builtin_wasm_widen_low_u_i16x8_i8x16: | |||
14300 | case WebAssembly::BI__builtin_wasm_widen_low_u_i32x4_i16x8: | |||
14301 | IntNo = Intrinsic::wasm_widen_low_unsigned; | |||
14302 | break; | |||
14303 | case WebAssembly::BI__builtin_wasm_widen_high_u_i16x8_i8x16: | |||
14304 | case WebAssembly::BI__builtin_wasm_widen_high_u_i32x4_i16x8: | |||
14305 | IntNo = Intrinsic::wasm_widen_high_unsigned; | |||
14306 | break; | |||
14307 | default: | |||
14308 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-10~svn374877/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14308); | |||
14309 | } | |||
14310 | Function *Callee = | |||
14311 | CGM.getIntrinsic(IntNo, {ConvertType(E->getType()), Vec->getType()}); | |||
14312 | return Builder.CreateCall(Callee, Vec); | |||
14313 | } | |||
14314 | default: | |||
14315 | return nullptr; | |||
14316 | } | |||
14317 | } | |||
14318 | ||||
14319 | Value *CodeGenFunction::EmitHexagonBuiltinExpr(unsigned BuiltinID, | |||
14320 | const CallExpr *E) { | |||
14321 | SmallVector<llvm::Value *, 4> Ops; | |||
14322 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
14323 | ||||
14324 | auto MakeCircLd = [&](unsigned IntID, bool HasImm) { | |||
14325 | // The base pointer is passed by address, so it needs to be loaded. | |||
14326 | Address BP = EmitPointerWithAlignment(E->getArg(0)); | |||
14327 | BP = Address(Builder.CreateBitCast(BP.getPointer(), Int8PtrPtrTy), | |||
14328 | BP.getAlignment()); | |||
14329 | llvm::Value *Base = Builder.CreateLoad(BP); | |||
14330 | // Operands are Base, Increment, Modifier, Start. | |||
14331 | if (HasImm) | |||
14332 | Ops = { Base, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), | |||
14333 | EmitScalarExpr(E->getArg(3)) }; | |||
14334 | else | |||
14335 | Ops = { Base, EmitScalarExpr(E->getArg(1)), | |||
14336 | EmitScalarExpr(E->getArg(2)) }; | |||
14337 | ||||
14338 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops); | |||
14339 | llvm::Value *NewBase = Builder.CreateExtractValue(Result, 1); | |||
14340 | llvm::Value *LV = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), | |||
14341 | NewBase->getType()->getPointerTo()); | |||
14342 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
14343 | // The intrinsic generates two results. The new value for the base pointer | |||
14344 | // needs to be stored. | |||
14345 | Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment()); | |||
14346 | return Builder.CreateExtractValue(Result, 0); | |||
14347 | }; | |||
14348 | ||||
14349 | auto MakeCircSt = [&](unsigned IntID, bool HasImm) { | |||
14350 | // The base pointer is passed by address, so it needs to be loaded. | |||
14351 | Address BP = EmitPointerWithAlignment(E->getArg(0)); | |||
14352 | BP = Address(Builder.CreateBitCast(BP.getPointer(), Int8PtrPtrTy), | |||
14353 | BP.getAlignment()); | |||
14354 | llvm::Value *Base = Builder.CreateLoad(BP); | |||
14355 | // Operands are Base, Increment, Modifier, Value, Start. | |||
14356 | if (HasImm) | |||
14357 | Ops = { Base, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), | |||
14358 | EmitScalarExpr(E->getArg(3)), EmitScalarExpr(E->getArg(4)) }; | |||
14359 | else | |||
14360 | Ops = { Base, EmitScalarExpr(E->getArg(1)), | |||
14361 | EmitScalarExpr(E->getArg(2)), EmitScalarExpr(E->getArg(3)) }; | |||
14362 | ||||
14363 | llvm::Value *NewBase = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops); | |||
14364 | llvm::Value *LV = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), | |||
14365 | NewBase->getType()->getPointerTo()); | |||
14366 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
14367 | // The intrinsic generates one result, which is the new value for the base | |||
14368 | // pointer. It needs to be stored. | |||
14369 | return Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment()); | |||
14370 | }; | |||
14371 | ||||
14372 | // Handle the conversion of bit-reverse load intrinsics to bit code. | |||
14373 | // The intrinsic call after this function only reads from memory and the | |||
14374 | // write to memory is dealt by the store instruction. | |||
14375 | auto MakeBrevLd = [&](unsigned IntID, llvm::Type *DestTy) { | |||
14376 | // The intrinsic generates one result, which is the new value for the base | |||
14377 | // pointer. It needs to be returned. The result of the load instruction is | |||
14378 | // passed to intrinsic by address, so the value needs to be stored. | |||
14379 | llvm::Value *BaseAddress = | |||
14380 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int8PtrTy); | |||
14381 | ||||
14382 | // Expressions like &(*pt++) will be incremented per evaluation. | |||
14383 | // EmitPointerWithAlignment and EmitScalarExpr evaluates the expression | |||
14384 | // per call. | |||
14385 | Address DestAddr = EmitPointerWithAlignment(E->getArg(1)); | |||
14386 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), Int8PtrTy), | |||
14387 | DestAddr.getAlignment()); | |||
14388 | llvm::Value *DestAddress = DestAddr.getPointer(); | |||
14389 | ||||
14390 | // Operands are Base, Dest, Modifier. | |||
14391 | // The intrinsic format in LLVM IR is defined as | |||
14392 | // { ValueType, i8* } (i8*, i32). | |||
14393 | Ops = {BaseAddress, EmitScalarExpr(E->getArg(2))}; | |||
14394 | ||||
14395 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops); | |||
14396 | // The value needs to be stored as the variable is passed by reference. | |||
14397 | llvm::Value *DestVal = Builder.CreateExtractValue(Result, 0); | |||
14398 | ||||
14399 | // The store needs to be truncated to fit the destination type. | |||
14400 | // While i32 and i64 are natively supported on Hexagon, i8 and i16 needs | |||
14401 | // to be handled with stores of respective destination type. | |||
14402 | DestVal = Builder.CreateTrunc(DestVal, DestTy); | |||
14403 | ||||
14404 | llvm::Value *DestForStore = | |||
14405 | Builder.CreateBitCast(DestAddress, DestVal->getType()->getPointerTo()); | |||
14406 | Builder.CreateAlignedStore(DestVal, DestForStore, DestAddr.getAlignment()); | |||
14407 | // The updated value of the base pointer is returned. | |||
14408 | return Builder.CreateExtractValue(Result, 1); | |||
14409 | }; | |||
14410 | ||||
14411 | switch (BuiltinID) { | |||
14412 | case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry: | |||
14413 | case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry_128B: { | |||
14414 | Address Dest = EmitPointerWithAlignment(E->getArg(2)); | |||
14415 | unsigned Size; | |||
14416 | if (BuiltinID == Hexagon::BI__builtin_HEXAGON_V6_vaddcarry) { | |||
14417 | Size = 512; | |||
14418 | ID = Intrinsic::hexagon_V6_vaddcarry; | |||
14419 | } else { | |||
14420 | Size = 1024; | |||
14421 | ID = Intrinsic::hexagon_V6_vaddcarry_128B; | |||
14422 | } | |||
14423 | Dest = Builder.CreateBitCast(Dest, | |||
14424 | llvm::VectorType::get(Builder.getInt1Ty(), Size)->getPointerTo(0)); | |||
14425 | LoadInst *QLd = Builder.CreateLoad(Dest); | |||
14426 | Ops = { EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), QLd }; | |||
14427 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
14428 | llvm::Value *Vprd = Builder.CreateExtractValue(Result, 1); | |||
14429 | llvm::Value *Base = Builder.CreateBitCast(EmitScalarExpr(E->getArg(2)), | |||
14430 | Vprd->getType()->getPointerTo(0)); | |||
14431 | Builder.CreateAlignedStore(Vprd, Base, Dest.getAlignment()); | |||
14432 | return Builder.CreateExtractValue(Result, 0); | |||
14433 | } | |||
14434 | case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry: | |||
14435 | case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry_128B: { | |||
14436 | Address Dest = EmitPointerWithAlignment(E->getArg(2)); | |||
14437 | unsigned Size; | |||
14438 | if (BuiltinID == Hexagon::BI__builtin_HEXAGON_V6_vsubcarry) { | |||
14439 | Size = 512; | |||
14440 | ID = Intrinsic::hexagon_V6_vsubcarry; | |||
14441 | } else { | |||
14442 | Size = 1024; | |||
14443 | ID = Intrinsic::hexagon_V6_vsubcarry_128B; | |||
14444 | } | |||
14445 | Dest = Builder.CreateBitCast(Dest, | |||
14446 | llvm::VectorType::get(Builder.getInt1Ty(), Size)->getPointerTo(0)); | |||
14447 | LoadInst *QLd = Builder.CreateLoad(Dest); | |||
14448 | Ops = { EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), QLd }; | |||
14449 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
14450 | llvm::Value *Vprd = Builder.CreateExtractValue(Result, 1); | |||
14451 | llvm::Value *Base = Builder.CreateBitCast(EmitScalarExpr(E->getArg(2)), | |||
14452 | Vprd->getType()->getPointerTo(0)); | |||
14453 | Builder.CreateAlignedStore(Vprd, Base, Dest.getAlignment()); | |||
14454 | return Builder.CreateExtractValue(Result, 0); | |||
14455 | } | |||
14456 | case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pci: | |||
14457 | return MakeCircLd(Intrinsic::hexagon_L2_loadrub_pci, /*HasImm*/true); | |||
14458 | case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pci: | |||
14459 | return MakeCircLd(Intrinsic::hexagon_L2_loadrb_pci, /*HasImm*/true); | |||
14460 | case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pci: | |||
14461 | return MakeCircLd(Intrinsic::hexagon_L2_loadruh_pci, /*HasImm*/true); | |||
14462 | case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pci: | |||
14463 | return MakeCircLd(Intrinsic::hexagon_L2_loadrh_pci, /*HasImm*/true); | |||
14464 | case Hexagon::BI__builtin_HEXAGON_L2_loadri_pci: | |||
14465 | return MakeCircLd(Intrinsic::hexagon_L2_loadri_pci, /*HasImm*/true); | |||
14466 | case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pci: | |||
14467 | return MakeCircLd(Intrinsic::hexagon_L2_loadrd_pci, /*HasImm*/true); | |||
14468 | case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pcr: | |||
14469 | return MakeCircLd(Intrinsic::hexagon_L2_loadrub_pcr, /*HasImm*/false); | |||
14470 | case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pcr: | |||
14471 | return MakeCircLd(Intrinsic::hexagon_L2_loadrb_pcr, /*HasImm*/false); | |||
14472 | case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pcr: | |||
14473 | return MakeCircLd(Intrinsic::hexagon_L2_loadruh_pcr, /*HasImm*/false); | |||
14474 | case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pcr: | |||
14475 | return MakeCircLd(Intrinsic::hexagon_L2_loadrh_pcr, /*HasImm*/false); | |||
14476 | case Hexagon::BI__builtin_HEXAGON_L2_loadri_pcr: | |||
14477 | return MakeCircLd(Intrinsic::hexagon_L2_loadri_pcr, /*HasImm*/false); | |||
14478 | case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pcr: | |||
14479 | return MakeCircLd(Intrinsic::hexagon_L2_loadrd_pcr, /*HasImm*/false); | |||
14480 | case Hexagon::BI__builtin_HEXAGON_S2_storerb_pci: | |||
14481 | return MakeCircSt(Intrinsic::hexagon_S2_storerb_pci, /*HasImm*/true); | |||
14482 | case Hexagon::BI__builtin_HEXAGON_S2_storerh_pci: | |||
14483 | return MakeCircSt(Intrinsic::hexagon_S2_storerh_pci, /*HasImm*/true); | |||
14484 | case Hexagon::BI__builtin_HEXAGON_S2_storerf_pci: | |||
14485 | return MakeCircSt(Intrinsic::hexagon_S2_storerf_pci, /*HasImm*/true); | |||
14486 | case Hexagon::BI__builtin_HEXAGON_S2_storeri_pci: | |||
14487 | return MakeCircSt(Intrinsic::hexagon_S2_storeri_pci, /*HasImm*/true); | |||
14488 | case Hexagon::BI__builtin_HEXAGON_S2_storerd_pci: | |||
14489 | return MakeCircSt(Intrinsic::hexagon_S2_storerd_pci, /*HasImm*/true); | |||
14490 | case Hexagon::BI__builtin_HEXAGON_S2_storerb_pcr: | |||
14491 | return MakeCircSt(Intrinsic::hexagon_S2_storerb_pcr, /*HasImm*/false); | |||
14492 | case Hexagon::BI__builtin_HEXAGON_S2_storerh_pcr: | |||
14493 | return MakeCircSt(Intrinsic::hexagon_S2_storerh_pcr, /*HasImm*/false); | |||
14494 | case Hexagon::BI__builtin_HEXAGON_S2_storerf_pcr: | |||
14495 | return MakeCircSt(Intrinsic::hexagon_S2_storerf_pcr, /*HasImm*/false); | |||
14496 | case Hexagon::BI__builtin_HEXAGON_S2_storeri_pcr: | |||
14497 | return MakeCircSt(Intrinsic::hexagon_S2_storeri_pcr, /*HasImm*/false); | |||
14498 | case Hexagon::BI__builtin_HEXAGON_S2_storerd_pcr: | |||
14499 | return MakeCircSt(Intrinsic::hexagon_S2_storerd_pcr, /*HasImm*/false); | |||
14500 | case Hexagon::BI__builtin_brev_ldub: | |||
14501 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrub_pbr, Int8Ty); | |||
14502 | case Hexagon::BI__builtin_brev_ldb: | |||
14503 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrb_pbr, Int8Ty); | |||
14504 | case Hexagon::BI__builtin_brev_lduh: | |||
14505 | return MakeBrevLd(Intrinsic::hexagon_L2_loadruh_pbr, Int16Ty); | |||
14506 | case Hexagon::BI__builtin_brev_ldh: | |||
14507 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrh_pbr, Int16Ty); | |||
14508 | case Hexagon::BI__builtin_brev_ldw: | |||
14509 | return MakeBrevLd(Intrinsic::hexagon_L2_loadri_pbr, Int32Ty); | |||
14510 | case Hexagon::BI__builtin_brev_ldd: | |||
14511 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrd_pbr, Int64Ty); | |||
14512 | default: | |||
14513 | break; | |||
14514 | } // switch | |||
14515 | ||||
14516 | return nullptr; | |||
14517 | } |