File: | llvm/lib/Target/X86/X86PartialReduction.cpp |
Warning: | line 238, column 26 Division by zero |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | //===-- X86PartialReduction.cpp -------------------------------------------===// | ||||||||||||
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 pass looks for add instructions used by a horizontal reduction to see | ||||||||||||
10 | // if we might be able to use pmaddwd or psadbw. Some cases of this require | ||||||||||||
11 | // cross basic block knowledge and can't be done in SelectionDAG. | ||||||||||||
12 | // | ||||||||||||
13 | //===----------------------------------------------------------------------===// | ||||||||||||
14 | |||||||||||||
15 | #include "X86.h" | ||||||||||||
16 | #include "llvm/Analysis/ValueTracking.h" | ||||||||||||
17 | #include "llvm/CodeGen/TargetPassConfig.h" | ||||||||||||
18 | #include "llvm/IR/Constants.h" | ||||||||||||
19 | #include "llvm/IR/Instructions.h" | ||||||||||||
20 | #include "llvm/IR/IntrinsicsX86.h" | ||||||||||||
21 | #include "llvm/IR/IRBuilder.h" | ||||||||||||
22 | #include "llvm/IR/Operator.h" | ||||||||||||
23 | #include "llvm/Pass.h" | ||||||||||||
24 | #include "X86TargetMachine.h" | ||||||||||||
25 | |||||||||||||
26 | using namespace llvm; | ||||||||||||
27 | |||||||||||||
28 | #define DEBUG_TYPE"x86-partial-reduction" "x86-partial-reduction" | ||||||||||||
29 | |||||||||||||
30 | namespace { | ||||||||||||
31 | |||||||||||||
32 | class X86PartialReduction : public FunctionPass { | ||||||||||||
33 | const DataLayout *DL; | ||||||||||||
34 | const X86Subtarget *ST; | ||||||||||||
35 | |||||||||||||
36 | public: | ||||||||||||
37 | static char ID; // Pass identification, replacement for typeid. | ||||||||||||
38 | |||||||||||||
39 | X86PartialReduction() : FunctionPass(ID) { } | ||||||||||||
40 | |||||||||||||
41 | bool runOnFunction(Function &Fn) override; | ||||||||||||
42 | |||||||||||||
43 | void getAnalysisUsage(AnalysisUsage &AU) const override { | ||||||||||||
44 | AU.setPreservesCFG(); | ||||||||||||
45 | } | ||||||||||||
46 | |||||||||||||
47 | StringRef getPassName() const override { | ||||||||||||
48 | return "X86 Partial Reduction"; | ||||||||||||
49 | } | ||||||||||||
50 | |||||||||||||
51 | private: | ||||||||||||
52 | bool tryMAddReplacement(Instruction *Op); | ||||||||||||
53 | bool trySADReplacement(Instruction *Op); | ||||||||||||
54 | }; | ||||||||||||
55 | } | ||||||||||||
56 | |||||||||||||
57 | FunctionPass *llvm::createX86PartialReductionPass() { | ||||||||||||
58 | return new X86PartialReduction(); | ||||||||||||
59 | } | ||||||||||||
60 | |||||||||||||
61 | char X86PartialReduction::ID = 0; | ||||||||||||
62 | |||||||||||||
63 | INITIALIZE_PASS(X86PartialReduction, DEBUG_TYPE,static void *initializeX86PartialReductionPassOnce(PassRegistry &Registry) { PassInfo *PI = new PassInfo( "X86 Partial Reduction" , "x86-partial-reduction", &X86PartialReduction::ID, PassInfo ::NormalCtor_t(callDefaultCtor<X86PartialReduction>), false , false); Registry.registerPass(*PI, true); return PI; } static llvm::once_flag InitializeX86PartialReductionPassFlag; void llvm ::initializeX86PartialReductionPass(PassRegistry &Registry ) { llvm::call_once(InitializeX86PartialReductionPassFlag, initializeX86PartialReductionPassOnce , std::ref(Registry)); } | ||||||||||||
64 | "X86 Partial Reduction", false, false)static void *initializeX86PartialReductionPassOnce(PassRegistry &Registry) { PassInfo *PI = new PassInfo( "X86 Partial Reduction" , "x86-partial-reduction", &X86PartialReduction::ID, PassInfo ::NormalCtor_t(callDefaultCtor<X86PartialReduction>), false , false); Registry.registerPass(*PI, true); return PI; } static llvm::once_flag InitializeX86PartialReductionPassFlag; void llvm ::initializeX86PartialReductionPass(PassRegistry &Registry ) { llvm::call_once(InitializeX86PartialReductionPassFlag, initializeX86PartialReductionPassOnce , std::ref(Registry)); } | ||||||||||||
65 | |||||||||||||
66 | bool X86PartialReduction::tryMAddReplacement(Instruction *Op) { | ||||||||||||
67 | if (!ST->hasSSE2()) | ||||||||||||
68 | return false; | ||||||||||||
69 | |||||||||||||
70 | // Need at least 8 elements. | ||||||||||||
71 | if (cast<FixedVectorType>(Op->getType())->getNumElements() < 8) | ||||||||||||
72 | return false; | ||||||||||||
73 | |||||||||||||
74 | // Element type should be i32. | ||||||||||||
75 | if (!cast<VectorType>(Op->getType())->getElementType()->isIntegerTy(32)) | ||||||||||||
76 | return false; | ||||||||||||
77 | |||||||||||||
78 | auto *Mul = dyn_cast<BinaryOperator>(Op); | ||||||||||||
79 | if (!Mul || Mul->getOpcode() != Instruction::Mul) | ||||||||||||
80 | return false; | ||||||||||||
81 | |||||||||||||
82 | Value *LHS = Mul->getOperand(0); | ||||||||||||
83 | Value *RHS = Mul->getOperand(1); | ||||||||||||
84 | |||||||||||||
85 | // LHS and RHS should be only used once or if they are the same then only | ||||||||||||
86 | // used twice. Only check this when SSE4.1 is enabled and we have zext/sext | ||||||||||||
87 | // instructions, otherwise we use punpck to emulate zero extend in stages. The | ||||||||||||
88 | // trunc/ we need to do likely won't introduce new instructions in that case. | ||||||||||||
89 | if (ST->hasSSE41()) { | ||||||||||||
90 | if (LHS == RHS) { | ||||||||||||
91 | if (!isa<Constant>(LHS) && !LHS->hasNUses(2)) | ||||||||||||
92 | return false; | ||||||||||||
93 | } else { | ||||||||||||
94 | if (!isa<Constant>(LHS) && !LHS->hasOneUse()) | ||||||||||||
95 | return false; | ||||||||||||
96 | if (!isa<Constant>(RHS) && !RHS->hasOneUse()) | ||||||||||||
97 | return false; | ||||||||||||
98 | } | ||||||||||||
99 | } | ||||||||||||
100 | |||||||||||||
101 | auto CanShrinkOp = [&](Value *Op) { | ||||||||||||
102 | auto IsFreeTruncation = [&](Value *Op) { | ||||||||||||
103 | if (auto *Cast = dyn_cast<CastInst>(Op)) { | ||||||||||||
104 | if (Cast->getParent() == Mul->getParent() && | ||||||||||||
105 | (Cast->getOpcode() == Instruction::SExt || | ||||||||||||
106 | Cast->getOpcode() == Instruction::ZExt) && | ||||||||||||
107 | Cast->getOperand(0)->getType()->getScalarSizeInBits() <= 16) | ||||||||||||
108 | return true; | ||||||||||||
109 | } | ||||||||||||
110 | |||||||||||||
111 | return isa<Constant>(Op); | ||||||||||||
112 | }; | ||||||||||||
113 | |||||||||||||
114 | // If the operation can be freely truncated and has enough sign bits we | ||||||||||||
115 | // can shrink. | ||||||||||||
116 | if (IsFreeTruncation(Op) && | ||||||||||||
117 | ComputeNumSignBits(Op, *DL, 0, nullptr, Mul) > 16) | ||||||||||||
118 | return true; | ||||||||||||
119 | |||||||||||||
120 | // SelectionDAG has limited support for truncating through an add or sub if | ||||||||||||
121 | // the inputs are freely truncatable. | ||||||||||||
122 | if (auto *BO = dyn_cast<BinaryOperator>(Op)) { | ||||||||||||
123 | if (BO->getParent() == Mul->getParent() && | ||||||||||||
124 | IsFreeTruncation(BO->getOperand(0)) && | ||||||||||||
125 | IsFreeTruncation(BO->getOperand(1)) && | ||||||||||||
126 | ComputeNumSignBits(Op, *DL, 0, nullptr, Mul) > 16) | ||||||||||||
127 | return true; | ||||||||||||
128 | } | ||||||||||||
129 | |||||||||||||
130 | return false; | ||||||||||||
131 | }; | ||||||||||||
132 | |||||||||||||
133 | // Both Ops need to be shrinkable. | ||||||||||||
134 | if (!CanShrinkOp(LHS) && !CanShrinkOp(RHS)) | ||||||||||||
135 | return false; | ||||||||||||
136 | |||||||||||||
137 | IRBuilder<> Builder(Mul); | ||||||||||||
138 | |||||||||||||
139 | auto *MulTy = cast<FixedVectorType>(Op->getType()); | ||||||||||||
140 | unsigned NumElts = MulTy->getNumElements(); | ||||||||||||
141 | |||||||||||||
142 | // Extract even elements and odd elements and add them together. This will | ||||||||||||
143 | // be pattern matched by SelectionDAG to pmaddwd. This instruction will be | ||||||||||||
144 | // half the original width. | ||||||||||||
145 | SmallVector<int, 16> EvenMask(NumElts / 2); | ||||||||||||
146 | SmallVector<int, 16> OddMask(NumElts / 2); | ||||||||||||
147 | for (int i = 0, e = NumElts / 2; i != e; ++i) { | ||||||||||||
148 | EvenMask[i] = i * 2; | ||||||||||||
149 | OddMask[i] = i * 2 + 1; | ||||||||||||
150 | } | ||||||||||||
151 | // Creating a new mul so the replaceAllUsesWith below doesn't replace the | ||||||||||||
152 | // uses in the shuffles we're creating. | ||||||||||||
153 | Value *NewMul = Builder.CreateMul(Mul->getOperand(0), Mul->getOperand(1)); | ||||||||||||
154 | Value *EvenElts = Builder.CreateShuffleVector(NewMul, NewMul, EvenMask); | ||||||||||||
155 | Value *OddElts = Builder.CreateShuffleVector(NewMul, NewMul, OddMask); | ||||||||||||
156 | Value *MAdd = Builder.CreateAdd(EvenElts, OddElts); | ||||||||||||
157 | |||||||||||||
158 | // Concatenate zeroes to extend back to the original type. | ||||||||||||
159 | SmallVector<int, 32> ConcatMask(NumElts); | ||||||||||||
160 | std::iota(ConcatMask.begin(), ConcatMask.end(), 0); | ||||||||||||
161 | Value *Zero = Constant::getNullValue(MAdd->getType()); | ||||||||||||
162 | Value *Concat = Builder.CreateShuffleVector(MAdd, Zero, ConcatMask); | ||||||||||||
163 | |||||||||||||
164 | Mul->replaceAllUsesWith(Concat); | ||||||||||||
165 | Mul->eraseFromParent(); | ||||||||||||
166 | |||||||||||||
167 | return true; | ||||||||||||
168 | } | ||||||||||||
169 | |||||||||||||
170 | bool X86PartialReduction::trySADReplacement(Instruction *Op) { | ||||||||||||
171 | if (!ST->hasSSE2()) | ||||||||||||
172 | return false; | ||||||||||||
173 | |||||||||||||
174 | // TODO: There's nothing special about i32, any integer type above i16 should | ||||||||||||
175 | // work just as well. | ||||||||||||
176 | if (!cast<VectorType>(Op->getType())->getElementType()->isIntegerTy(32)) | ||||||||||||
177 | return false; | ||||||||||||
178 | |||||||||||||
179 | // Operand should be a select. | ||||||||||||
180 | auto *SI = dyn_cast<SelectInst>(Op); | ||||||||||||
181 | if (!SI
| ||||||||||||
182 | return false; | ||||||||||||
183 | |||||||||||||
184 | // Select needs to implement absolute value. | ||||||||||||
185 | Value *LHS, *RHS; | ||||||||||||
186 | auto SPR = matchSelectPattern(SI, LHS, RHS); | ||||||||||||
187 | if (SPR.Flavor != SPF_ABS) | ||||||||||||
188 | return false; | ||||||||||||
189 | |||||||||||||
190 | // Need a subtract of two values. | ||||||||||||
191 | auto *Sub = dyn_cast<BinaryOperator>(LHS); | ||||||||||||
192 | if (!Sub
| ||||||||||||
193 | return false; | ||||||||||||
194 | |||||||||||||
195 | // Look for zero extend from i8. | ||||||||||||
196 | auto getZeroExtendedVal = [](Value *Op) -> Value * { | ||||||||||||
197 | if (auto *ZExt = dyn_cast<ZExtInst>(Op)) | ||||||||||||
198 | if (cast<VectorType>(ZExt->getOperand(0)->getType()) | ||||||||||||
199 | ->getElementType() | ||||||||||||
200 | ->isIntegerTy(8)) | ||||||||||||
201 | return ZExt->getOperand(0); | ||||||||||||
202 | |||||||||||||
203 | return nullptr; | ||||||||||||
204 | }; | ||||||||||||
205 | |||||||||||||
206 | // Both operands of the subtract should be extends from vXi8. | ||||||||||||
207 | Value *Op0 = getZeroExtendedVal(Sub->getOperand(0)); | ||||||||||||
208 | Value *Op1 = getZeroExtendedVal(Sub->getOperand(1)); | ||||||||||||
209 | if (!Op0
| ||||||||||||
210 | return false; | ||||||||||||
211 | |||||||||||||
212 | IRBuilder<> Builder(SI); | ||||||||||||
213 | |||||||||||||
214 | auto *OpTy = cast<FixedVectorType>(Op->getType()); | ||||||||||||
215 | unsigned NumElts = OpTy->getNumElements(); | ||||||||||||
216 | |||||||||||||
217 | unsigned IntrinsicNumElts; | ||||||||||||
218 | Intrinsic::ID IID; | ||||||||||||
219 | if (ST->hasBWI() && NumElts >= 64) { | ||||||||||||
220 | IID = Intrinsic::x86_avx512_psad_bw_512; | ||||||||||||
221 | IntrinsicNumElts = 64; | ||||||||||||
222 | } else if (ST->hasAVX2() && NumElts >= 32) { | ||||||||||||
223 | IID = Intrinsic::x86_avx2_psad_bw; | ||||||||||||
224 | IntrinsicNumElts = 32; | ||||||||||||
225 | } else { | ||||||||||||
226 | IID = Intrinsic::x86_sse2_psad_bw; | ||||||||||||
227 | IntrinsicNumElts = 16; | ||||||||||||
228 | } | ||||||||||||
229 | |||||||||||||
230 | Function *PSADBWFn = Intrinsic::getDeclaration(SI->getModule(), IID); | ||||||||||||
231 | |||||||||||||
232 | if (NumElts < 16) { | ||||||||||||
233 | // Pad input with zeroes. | ||||||||||||
234 | SmallVector<int, 32> ConcatMask(16); | ||||||||||||
235 | for (unsigned i = 0; i != NumElts; ++i) | ||||||||||||
236 | ConcatMask[i] = i; | ||||||||||||
237 | for (unsigned i = NumElts; i != 16; ++i) | ||||||||||||
238 | ConcatMask[i] = (i % NumElts) + NumElts; | ||||||||||||
| |||||||||||||
239 | |||||||||||||
240 | Value *Zero = Constant::getNullValue(Op0->getType()); | ||||||||||||
241 | Op0 = Builder.CreateShuffleVector(Op0, Zero, ConcatMask); | ||||||||||||
242 | Op1 = Builder.CreateShuffleVector(Op1, Zero, ConcatMask); | ||||||||||||
243 | NumElts = 16; | ||||||||||||
244 | } | ||||||||||||
245 | |||||||||||||
246 | // Intrinsics produce vXi64 and need to be casted to vXi32. | ||||||||||||
247 | auto *I32Ty = | ||||||||||||
248 | FixedVectorType::get(Builder.getInt32Ty(), IntrinsicNumElts / 4); | ||||||||||||
249 | |||||||||||||
250 | assert(NumElts % IntrinsicNumElts == 0 && "Unexpected number of elements!")(static_cast <bool> (NumElts % IntrinsicNumElts == 0 && "Unexpected number of elements!") ? void (0) : __assert_fail ("NumElts % IntrinsicNumElts == 0 && \"Unexpected number of elements!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/lib/Target/X86/X86PartialReduction.cpp" , 250, __extension__ __PRETTY_FUNCTION__)); | ||||||||||||
251 | unsigned NumSplits = NumElts / IntrinsicNumElts; | ||||||||||||
252 | |||||||||||||
253 | // First collect the pieces we need. | ||||||||||||
254 | SmallVector<Value *, 4> Ops(NumSplits); | ||||||||||||
255 | for (unsigned i = 0; i != NumSplits; ++i) { | ||||||||||||
256 | SmallVector<int, 64> ExtractMask(IntrinsicNumElts); | ||||||||||||
257 | std::iota(ExtractMask.begin(), ExtractMask.end(), i * IntrinsicNumElts); | ||||||||||||
258 | Value *ExtractOp0 = Builder.CreateShuffleVector(Op0, Op0, ExtractMask); | ||||||||||||
259 | Value *ExtractOp1 = Builder.CreateShuffleVector(Op1, Op0, ExtractMask); | ||||||||||||
260 | Ops[i] = Builder.CreateCall(PSADBWFn, {ExtractOp0, ExtractOp1}); | ||||||||||||
261 | Ops[i] = Builder.CreateBitCast(Ops[i], I32Ty); | ||||||||||||
262 | } | ||||||||||||
263 | |||||||||||||
264 | assert(isPowerOf2_32(NumSplits) && "Expected power of 2 splits")(static_cast <bool> (isPowerOf2_32(NumSplits) && "Expected power of 2 splits") ? void (0) : __assert_fail ("isPowerOf2_32(NumSplits) && \"Expected power of 2 splits\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/lib/Target/X86/X86PartialReduction.cpp" , 264, __extension__ __PRETTY_FUNCTION__)); | ||||||||||||
265 | unsigned Stages = Log2_32(NumSplits); | ||||||||||||
266 | for (unsigned s = Stages; s > 0; --s) { | ||||||||||||
267 | unsigned NumConcatElts = | ||||||||||||
268 | cast<FixedVectorType>(Ops[0]->getType())->getNumElements() * 2; | ||||||||||||
269 | for (unsigned i = 0; i != 1U << (s - 1); ++i) { | ||||||||||||
270 | SmallVector<int, 64> ConcatMask(NumConcatElts); | ||||||||||||
271 | std::iota(ConcatMask.begin(), ConcatMask.end(), 0); | ||||||||||||
272 | Ops[i] = Builder.CreateShuffleVector(Ops[i*2], Ops[i*2+1], ConcatMask); | ||||||||||||
273 | } | ||||||||||||
274 | } | ||||||||||||
275 | |||||||||||||
276 | // At this point the final value should be in Ops[0]. Now we need to adjust | ||||||||||||
277 | // it to the final original type. | ||||||||||||
278 | NumElts = cast<FixedVectorType>(OpTy)->getNumElements(); | ||||||||||||
279 | if (NumElts == 2) { | ||||||||||||
280 | // Extract down to 2 elements. | ||||||||||||
281 | Ops[0] = Builder.CreateShuffleVector(Ops[0], Ops[0], ArrayRef<int>{0, 1}); | ||||||||||||
282 | } else if (NumElts >= 8) { | ||||||||||||
283 | SmallVector<int, 32> ConcatMask(NumElts); | ||||||||||||
284 | unsigned SubElts = | ||||||||||||
285 | cast<FixedVectorType>(Ops[0]->getType())->getNumElements(); | ||||||||||||
286 | for (unsigned i = 0; i != SubElts; ++i) | ||||||||||||
287 | ConcatMask[i] = i; | ||||||||||||
288 | for (unsigned i = SubElts; i != NumElts; ++i) | ||||||||||||
289 | ConcatMask[i] = (i % SubElts) + SubElts; | ||||||||||||
290 | |||||||||||||
291 | Value *Zero = Constant::getNullValue(Ops[0]->getType()); | ||||||||||||
292 | Ops[0] = Builder.CreateShuffleVector(Ops[0], Zero, ConcatMask); | ||||||||||||
293 | } | ||||||||||||
294 | |||||||||||||
295 | SI->replaceAllUsesWith(Ops[0]); | ||||||||||||
296 | SI->eraseFromParent(); | ||||||||||||
297 | |||||||||||||
298 | return true; | ||||||||||||
299 | } | ||||||||||||
300 | |||||||||||||
301 | // Walk backwards from the ExtractElementInst and determine if it is the end of | ||||||||||||
302 | // a horizontal reduction. Return the input to the reduction if we find one. | ||||||||||||
303 | static Value *matchAddReduction(const ExtractElementInst &EE) { | ||||||||||||
304 | // Make sure we're extracting index 0. | ||||||||||||
305 | auto *Index = dyn_cast<ConstantInt>(EE.getIndexOperand()); | ||||||||||||
306 | if (!Index || !Index->isNullValue()) | ||||||||||||
307 | return nullptr; | ||||||||||||
308 | |||||||||||||
309 | const auto *BO = dyn_cast<BinaryOperator>(EE.getVectorOperand()); | ||||||||||||
310 | if (!BO || BO->getOpcode() != Instruction::Add || !BO->hasOneUse()) | ||||||||||||
311 | return nullptr; | ||||||||||||
312 | |||||||||||||
313 | unsigned NumElems = cast<FixedVectorType>(BO->getType())->getNumElements(); | ||||||||||||
314 | // Ensure the reduction size is a power of 2. | ||||||||||||
315 | if (!isPowerOf2_32(NumElems)) | ||||||||||||
316 | return nullptr; | ||||||||||||
317 | |||||||||||||
318 | const Value *Op = BO; | ||||||||||||
319 | unsigned Stages = Log2_32(NumElems); | ||||||||||||
320 | for (unsigned i = 0; i != Stages; ++i) { | ||||||||||||
321 | const auto *BO = dyn_cast<BinaryOperator>(Op); | ||||||||||||
322 | if (!BO || BO->getOpcode() != Instruction::Add) | ||||||||||||
323 | return nullptr; | ||||||||||||
324 | |||||||||||||
325 | // If this isn't the first add, then it should only have 2 users, the | ||||||||||||
326 | // shuffle and another add which we checked in the previous iteration. | ||||||||||||
327 | if (i != 0 && !BO->hasNUses(2)) | ||||||||||||
328 | return nullptr; | ||||||||||||
329 | |||||||||||||
330 | Value *LHS = BO->getOperand(0); | ||||||||||||
331 | Value *RHS = BO->getOperand(1); | ||||||||||||
332 | |||||||||||||
333 | auto *Shuffle = dyn_cast<ShuffleVectorInst>(LHS); | ||||||||||||
334 | if (Shuffle) { | ||||||||||||
335 | Op = RHS; | ||||||||||||
336 | } else { | ||||||||||||
337 | Shuffle = dyn_cast<ShuffleVectorInst>(RHS); | ||||||||||||
338 | Op = LHS; | ||||||||||||
339 | } | ||||||||||||
340 | |||||||||||||
341 | // The first operand of the shuffle should be the same as the other operand | ||||||||||||
342 | // of the bin op. | ||||||||||||
343 | if (!Shuffle || Shuffle->getOperand(0) != Op) | ||||||||||||
344 | return nullptr; | ||||||||||||
345 | |||||||||||||
346 | // Verify the shuffle has the expected (at this stage of the pyramid) mask. | ||||||||||||
347 | unsigned MaskEnd = 1 << i; | ||||||||||||
348 | for (unsigned Index = 0; Index < MaskEnd; ++Index) | ||||||||||||
349 | if (Shuffle->getMaskValue(Index) != (int)(MaskEnd + Index)) | ||||||||||||
350 | return nullptr; | ||||||||||||
351 | } | ||||||||||||
352 | |||||||||||||
353 | return const_cast<Value *>(Op); | ||||||||||||
354 | } | ||||||||||||
355 | |||||||||||||
356 | // See if this BO is reachable from this Phi by walking forward through single | ||||||||||||
357 | // use BinaryOperators with the same opcode. If we get back then we know we've | ||||||||||||
358 | // found a loop and it is safe to step through this Add to find more leaves. | ||||||||||||
359 | static bool isReachableFromPHI(PHINode *Phi, BinaryOperator *BO) { | ||||||||||||
360 | // The PHI itself should only have one use. | ||||||||||||
361 | if (!Phi->hasOneUse()) | ||||||||||||
362 | return false; | ||||||||||||
363 | |||||||||||||
364 | Instruction *U = cast<Instruction>(*Phi->user_begin()); | ||||||||||||
365 | if (U == BO) | ||||||||||||
366 | return true; | ||||||||||||
367 | |||||||||||||
368 | while (U->hasOneUse() && U->getOpcode() == BO->getOpcode()) | ||||||||||||
369 | U = cast<Instruction>(*U->user_begin()); | ||||||||||||
370 | |||||||||||||
371 | return U == BO; | ||||||||||||
372 | } | ||||||||||||
373 | |||||||||||||
374 | // Collect all the leaves of the tree of adds that feeds into the horizontal | ||||||||||||
375 | // reduction. Root is the Value that is used by the horizontal reduction. | ||||||||||||
376 | // We look through single use phis, single use adds, or adds that are used by | ||||||||||||
377 | // a phi that forms a loop with the add. | ||||||||||||
378 | static void collectLeaves(Value *Root, SmallVectorImpl<Instruction *> &Leaves) { | ||||||||||||
379 | SmallPtrSet<Value *, 8> Visited; | ||||||||||||
380 | SmallVector<Value *, 8> Worklist; | ||||||||||||
381 | Worklist.push_back(Root); | ||||||||||||
382 | |||||||||||||
383 | while (!Worklist.empty()) { | ||||||||||||
384 | Value *V = Worklist.pop_back_val(); | ||||||||||||
385 | if (!Visited.insert(V).second) | ||||||||||||
386 | continue; | ||||||||||||
387 | |||||||||||||
388 | if (auto *PN = dyn_cast<PHINode>(V)) { | ||||||||||||
389 | // PHI node should have single use unless it is the root node, then it | ||||||||||||
390 | // has 2 uses. | ||||||||||||
391 | if (!PN->hasNUses(PN == Root ? 2 : 1)) | ||||||||||||
392 | break; | ||||||||||||
393 | |||||||||||||
394 | // Push incoming values to the worklist. | ||||||||||||
395 | append_range(Worklist, PN->incoming_values()); | ||||||||||||
396 | |||||||||||||
397 | continue; | ||||||||||||
398 | } | ||||||||||||
399 | |||||||||||||
400 | if (auto *BO = dyn_cast<BinaryOperator>(V)) { | ||||||||||||
401 | if (BO->getOpcode() == Instruction::Add) { | ||||||||||||
402 | // Simple case. Single use, just push its operands to the worklist. | ||||||||||||
403 | if (BO->hasNUses(BO == Root ? 2 : 1)) { | ||||||||||||
404 | append_range(Worklist, BO->operands()); | ||||||||||||
405 | continue; | ||||||||||||
406 | } | ||||||||||||
407 | |||||||||||||
408 | // If there is additional use, make sure it is an unvisited phi that | ||||||||||||
409 | // gets us back to this node. | ||||||||||||
410 | if (BO->hasNUses(BO == Root ? 3 : 2)) { | ||||||||||||
411 | PHINode *PN = nullptr; | ||||||||||||
412 | for (auto *U : Root->users()) | ||||||||||||
413 | if (auto *P = dyn_cast<PHINode>(U)) | ||||||||||||
414 | if (!Visited.count(P)) | ||||||||||||
415 | PN = P; | ||||||||||||
416 | |||||||||||||
417 | // If we didn't find a 2-input PHI then this isn't a case we can | ||||||||||||
418 | // handle. | ||||||||||||
419 | if (!PN || PN->getNumIncomingValues() != 2) | ||||||||||||
420 | continue; | ||||||||||||
421 | |||||||||||||
422 | // Walk forward from this phi to see if it reaches back to this add. | ||||||||||||
423 | if (!isReachableFromPHI(PN, BO)) | ||||||||||||
424 | continue; | ||||||||||||
425 | |||||||||||||
426 | // The phi forms a loop with this Add, push its operands. | ||||||||||||
427 | append_range(Worklist, BO->operands()); | ||||||||||||
428 | } | ||||||||||||
429 | } | ||||||||||||
430 | } | ||||||||||||
431 | |||||||||||||
432 | // Not an add or phi, make it a leaf. | ||||||||||||
433 | if (auto *I = dyn_cast<Instruction>(V)) { | ||||||||||||
434 | if (!V->hasNUses(I == Root ? 2 : 1)) | ||||||||||||
435 | continue; | ||||||||||||
436 | |||||||||||||
437 | // Add this as a leaf. | ||||||||||||
438 | Leaves.push_back(I); | ||||||||||||
439 | } | ||||||||||||
440 | } | ||||||||||||
441 | } | ||||||||||||
442 | |||||||||||||
443 | bool X86PartialReduction::runOnFunction(Function &F) { | ||||||||||||
444 | if (skipFunction(F)) | ||||||||||||
| |||||||||||||
445 | return false; | ||||||||||||
446 | |||||||||||||
447 | auto *TPC = getAnalysisIfAvailable<TargetPassConfig>(); | ||||||||||||
448 | if (!TPC) | ||||||||||||
449 | return false; | ||||||||||||
450 | |||||||||||||
451 | auto &TM = TPC->getTM<X86TargetMachine>(); | ||||||||||||
452 | ST = TM.getSubtargetImpl(F); | ||||||||||||
453 | |||||||||||||
454 | DL = &F.getParent()->getDataLayout(); | ||||||||||||
455 | |||||||||||||
456 | bool MadeChange = false; | ||||||||||||
457 | for (auto &BB : F) { | ||||||||||||
458 | for (auto &I : BB) { | ||||||||||||
459 | auto *EE = dyn_cast<ExtractElementInst>(&I); | ||||||||||||
460 | if (!EE
| ||||||||||||
461 | continue; | ||||||||||||
462 | |||||||||||||
463 | // First find a reduction tree. | ||||||||||||
464 | // FIXME: Do we need to handle other opcodes than Add? | ||||||||||||
465 | Value *Root = matchAddReduction(*EE); | ||||||||||||
466 | if (!Root
| ||||||||||||
467 | continue; | ||||||||||||
468 | |||||||||||||
469 | SmallVector<Instruction *, 8> Leaves; | ||||||||||||
470 | collectLeaves(Root, Leaves); | ||||||||||||
471 | |||||||||||||
472 | for (Instruction *I : Leaves) { | ||||||||||||
473 | if (tryMAddReplacement(I)) { | ||||||||||||
474 | MadeChange = true; | ||||||||||||
475 | continue; | ||||||||||||
476 | } | ||||||||||||
477 | |||||||||||||
478 | // Don't do SAD matching on the root node. SelectionDAG already | ||||||||||||
479 | // has support for that and currently generates better code. | ||||||||||||
480 | if (I != Root && trySADReplacement(I)) | ||||||||||||
481 | MadeChange = true; | ||||||||||||
482 | } | ||||||||||||
483 | } | ||||||||||||
484 | } | ||||||||||||
485 | |||||||||||||
486 | return MadeChange; | ||||||||||||
487 | } |
1 | //===-- X86Subtarget.h - Define Subtarget for the X86 ----------*- C++ -*--===// | ||||||
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 file declares the X86 specific subclass of TargetSubtargetInfo. | ||||||
10 | // | ||||||
11 | //===----------------------------------------------------------------------===// | ||||||
12 | |||||||
13 | #ifndef LLVM_LIB_TARGET_X86_X86SUBTARGET_H | ||||||
14 | #define LLVM_LIB_TARGET_X86_X86SUBTARGET_H | ||||||
15 | |||||||
16 | #include "X86FrameLowering.h" | ||||||
17 | #include "X86ISelLowering.h" | ||||||
18 | #include "X86InstrInfo.h" | ||||||
19 | #include "X86SelectionDAGInfo.h" | ||||||
20 | #include "llvm/ADT/Triple.h" | ||||||
21 | #include "llvm/CodeGen/TargetSubtargetInfo.h" | ||||||
22 | #include "llvm/IR/CallingConv.h" | ||||||
23 | #include <climits> | ||||||
24 | #include <memory> | ||||||
25 | |||||||
26 | #define GET_SUBTARGETINFO_HEADER | ||||||
27 | #include "X86GenSubtargetInfo.inc" | ||||||
28 | |||||||
29 | namespace llvm { | ||||||
30 | |||||||
31 | class CallLowering; | ||||||
32 | class GlobalValue; | ||||||
33 | class InstructionSelector; | ||||||
34 | class LegalizerInfo; | ||||||
35 | class RegisterBankInfo; | ||||||
36 | class StringRef; | ||||||
37 | class TargetMachine; | ||||||
38 | |||||||
39 | /// The X86 backend supports a number of different styles of PIC. | ||||||
40 | /// | ||||||
41 | namespace PICStyles { | ||||||
42 | |||||||
43 | enum class Style { | ||||||
44 | StubPIC, // Used on i386-darwin in pic mode. | ||||||
45 | GOT, // Used on 32 bit elf on when in pic mode. | ||||||
46 | RIPRel, // Used on X86-64 when in pic mode. | ||||||
47 | None // Set when not in pic mode. | ||||||
48 | }; | ||||||
49 | |||||||
50 | } // end namespace PICStyles | ||||||
51 | |||||||
52 | class X86Subtarget final : public X86GenSubtargetInfo { | ||||||
53 | // NOTE: Do not add anything new to this list. Coarse, CPU name based flags | ||||||
54 | // are not a good idea. We should be migrating away from these. | ||||||
55 | enum X86ProcFamilyEnum { | ||||||
56 | Others, | ||||||
57 | IntelAtom, | ||||||
58 | IntelSLM | ||||||
59 | }; | ||||||
60 | |||||||
61 | enum X86SSEEnum { | ||||||
62 | NoSSE, SSE1, SSE2, SSE3, SSSE3, SSE41, SSE42, AVX, AVX2, AVX512F | ||||||
63 | }; | ||||||
64 | |||||||
65 | enum X863DNowEnum { | ||||||
66 | NoThreeDNow, MMX, ThreeDNow, ThreeDNowA | ||||||
67 | }; | ||||||
68 | |||||||
69 | /// X86 processor family: Intel Atom, and others | ||||||
70 | X86ProcFamilyEnum X86ProcFamily = Others; | ||||||
71 | |||||||
72 | /// Which PIC style to use | ||||||
73 | PICStyles::Style PICStyle; | ||||||
74 | |||||||
75 | const TargetMachine &TM; | ||||||
76 | |||||||
77 | /// SSE1, SSE2, SSE3, SSSE3, SSE41, SSE42, or none supported. | ||||||
78 | X86SSEEnum X86SSELevel = NoSSE; | ||||||
79 | |||||||
80 | /// MMX, 3DNow, 3DNow Athlon, or none supported. | ||||||
81 | X863DNowEnum X863DNowLevel = NoThreeDNow; | ||||||
82 | |||||||
83 | /// True if the processor supports X87 instructions. | ||||||
84 | bool HasX87 = false; | ||||||
85 | |||||||
86 | /// True if the processor supports CMPXCHG8B. | ||||||
87 | bool HasCmpxchg8b = false; | ||||||
88 | |||||||
89 | /// True if this processor has NOPL instruction | ||||||
90 | /// (generally pentium pro+). | ||||||
91 | bool HasNOPL = false; | ||||||
92 | |||||||
93 | /// True if this processor has conditional move instructions | ||||||
94 | /// (generally pentium pro+). | ||||||
95 | bool HasCMov = false; | ||||||
96 | |||||||
97 | /// True if the processor supports X86-64 instructions. | ||||||
98 | bool HasX86_64 = false; | ||||||
99 | |||||||
100 | /// True if the processor supports POPCNT. | ||||||
101 | bool HasPOPCNT = false; | ||||||
102 | |||||||
103 | /// True if the processor supports SSE4A instructions. | ||||||
104 | bool HasSSE4A = false; | ||||||
105 | |||||||
106 | /// Target has AES instructions | ||||||
107 | bool HasAES = false; | ||||||
108 | bool HasVAES = false; | ||||||
109 | |||||||
110 | /// Target has FXSAVE/FXRESTOR instructions | ||||||
111 | bool HasFXSR = false; | ||||||
112 | |||||||
113 | /// Target has XSAVE instructions | ||||||
114 | bool HasXSAVE = false; | ||||||
115 | |||||||
116 | /// Target has XSAVEOPT instructions | ||||||
117 | bool HasXSAVEOPT = false; | ||||||
118 | |||||||
119 | /// Target has XSAVEC instructions | ||||||
120 | bool HasXSAVEC = false; | ||||||
121 | |||||||
122 | /// Target has XSAVES instructions | ||||||
123 | bool HasXSAVES = false; | ||||||
124 | |||||||
125 | /// Target has carry-less multiplication | ||||||
126 | bool HasPCLMUL = false; | ||||||
127 | bool HasVPCLMULQDQ = false; | ||||||
128 | |||||||
129 | /// Target has Galois Field Arithmetic instructions | ||||||
130 | bool HasGFNI = false; | ||||||
131 | |||||||
132 | /// Target has 3-operand fused multiply-add | ||||||
133 | bool HasFMA = false; | ||||||
134 | |||||||
135 | /// Target has 4-operand fused multiply-add | ||||||
136 | bool HasFMA4 = false; | ||||||
137 | |||||||
138 | /// Target has XOP instructions | ||||||
139 | bool HasXOP = false; | ||||||
140 | |||||||
141 | /// Target has TBM instructions. | ||||||
142 | bool HasTBM = false; | ||||||
143 | |||||||
144 | /// Target has LWP instructions | ||||||
145 | bool HasLWP = false; | ||||||
146 | |||||||
147 | /// True if the processor has the MOVBE instruction. | ||||||
148 | bool HasMOVBE = false; | ||||||
149 | |||||||
150 | /// True if the processor has the RDRAND instruction. | ||||||
151 | bool HasRDRAND = false; | ||||||
152 | |||||||
153 | /// Processor has 16-bit floating point conversion instructions. | ||||||
154 | bool HasF16C = false; | ||||||
155 | |||||||
156 | /// Processor has FS/GS base insturctions. | ||||||
157 | bool HasFSGSBase = false; | ||||||
158 | |||||||
159 | /// Processor has LZCNT instruction. | ||||||
160 | bool HasLZCNT = false; | ||||||
161 | |||||||
162 | /// Processor has BMI1 instructions. | ||||||
163 | bool HasBMI = false; | ||||||
164 | |||||||
165 | /// Processor has BMI2 instructions. | ||||||
166 | bool HasBMI2 = false; | ||||||
167 | |||||||
168 | /// Processor has VBMI instructions. | ||||||
169 | bool HasVBMI = false; | ||||||
170 | |||||||
171 | /// Processor has VBMI2 instructions. | ||||||
172 | bool HasVBMI2 = false; | ||||||
173 | |||||||
174 | /// Processor has Integer Fused Multiply Add | ||||||
175 | bool HasIFMA = false; | ||||||
176 | |||||||
177 | /// Processor has RTM instructions. | ||||||
178 | bool HasRTM = false; | ||||||
179 | |||||||
180 | /// Processor has ADX instructions. | ||||||
181 | bool HasADX = false; | ||||||
182 | |||||||
183 | /// Processor has SHA instructions. | ||||||
184 | bool HasSHA = false; | ||||||
185 | |||||||
186 | /// Processor has PRFCHW instructions. | ||||||
187 | bool HasPRFCHW = false; | ||||||
188 | |||||||
189 | /// Processor has RDSEED instructions. | ||||||
190 | bool HasRDSEED = false; | ||||||
191 | |||||||
192 | /// Processor has LAHF/SAHF instructions in 64-bit mode. | ||||||
193 | bool HasLAHFSAHF64 = false; | ||||||
194 | |||||||
195 | /// Processor has MONITORX/MWAITX instructions. | ||||||
196 | bool HasMWAITX = false; | ||||||
197 | |||||||
198 | /// Processor has Cache Line Zero instruction | ||||||
199 | bool HasCLZERO = false; | ||||||
200 | |||||||
201 | /// Processor has Cache Line Demote instruction | ||||||
202 | bool HasCLDEMOTE = false; | ||||||
203 | |||||||
204 | /// Processor has MOVDIRI instruction (direct store integer). | ||||||
205 | bool HasMOVDIRI = false; | ||||||
206 | |||||||
207 | /// Processor has MOVDIR64B instruction (direct store 64 bytes). | ||||||
208 | bool HasMOVDIR64B = false; | ||||||
209 | |||||||
210 | /// Processor has ptwrite instruction. | ||||||
211 | bool HasPTWRITE = false; | ||||||
212 | |||||||
213 | /// Processor has Prefetch with intent to Write instruction | ||||||
214 | bool HasPREFETCHWT1 = false; | ||||||
215 | |||||||
216 | /// True if SHLD instructions are slow. | ||||||
217 | bool IsSHLDSlow = false; | ||||||
218 | |||||||
219 | /// True if the PMULLD instruction is slow compared to PMULLW/PMULHW and | ||||||
220 | // PMULUDQ. | ||||||
221 | bool IsPMULLDSlow = false; | ||||||
222 | |||||||
223 | /// True if the PMADDWD instruction is slow compared to PMULLD. | ||||||
224 | bool IsPMADDWDSlow = false; | ||||||
225 | |||||||
226 | /// True if unaligned memory accesses of 16-bytes are slow. | ||||||
227 | bool IsUAMem16Slow = false; | ||||||
228 | |||||||
229 | /// True if unaligned memory accesses of 32-bytes are slow. | ||||||
230 | bool IsUAMem32Slow = false; | ||||||
231 | |||||||
232 | /// True if SSE operations can have unaligned memory operands. | ||||||
233 | /// This may require setting a configuration bit in the processor. | ||||||
234 | bool HasSSEUnalignedMem = false; | ||||||
235 | |||||||
236 | /// True if this processor has the CMPXCHG16B instruction; | ||||||
237 | /// this is true for most x86-64 chips, but not the first AMD chips. | ||||||
238 | bool HasCmpxchg16b = false; | ||||||
239 | |||||||
240 | /// True if the LEA instruction should be used for adjusting | ||||||
241 | /// the stack pointer. This is an optimization for Intel Atom processors. | ||||||
242 | bool UseLeaForSP = false; | ||||||
243 | |||||||
244 | /// True if POPCNT instruction has a false dependency on the destination register. | ||||||
245 | bool HasPOPCNTFalseDeps = false; | ||||||
246 | |||||||
247 | /// True if LZCNT/TZCNT instructions have a false dependency on the destination register. | ||||||
248 | bool HasLZCNTFalseDeps = false; | ||||||
249 | |||||||
250 | /// True if its preferable to combine to a single cross-lane shuffle | ||||||
251 | /// using a variable mask over multiple fixed shuffles. | ||||||
252 | bool HasFastVariableCrossLaneShuffle = false; | ||||||
253 | |||||||
254 | /// True if its preferable to combine to a single per-lane shuffle | ||||||
255 | /// using a variable mask over multiple fixed shuffles. | ||||||
256 | bool HasFastVariablePerLaneShuffle = false; | ||||||
257 | |||||||
258 | /// True if vzeroupper instructions should be inserted after code that uses | ||||||
259 | /// ymm or zmm registers. | ||||||
260 | bool InsertVZEROUPPER = false; | ||||||
261 | |||||||
262 | /// True if there is no performance penalty for writing NOPs with up to | ||||||
263 | /// 7 bytes. | ||||||
264 | bool HasFast7ByteNOP = false; | ||||||
265 | |||||||
266 | /// True if there is no performance penalty for writing NOPs with up to | ||||||
267 | /// 11 bytes. | ||||||
268 | bool HasFast11ByteNOP = false; | ||||||
269 | |||||||
270 | /// True if there is no performance penalty for writing NOPs with up to | ||||||
271 | /// 15 bytes. | ||||||
272 | bool HasFast15ByteNOP = false; | ||||||
273 | |||||||
274 | /// True if gather is reasonably fast. This is true for Skylake client and | ||||||
275 | /// all AVX-512 CPUs. | ||||||
276 | bool HasFastGather = false; | ||||||
277 | |||||||
278 | /// True if hardware SQRTSS instruction is at least as fast (latency) as | ||||||
279 | /// RSQRTSS followed by a Newton-Raphson iteration. | ||||||
280 | bool HasFastScalarFSQRT = false; | ||||||
281 | |||||||
282 | /// True if hardware SQRTPS/VSQRTPS instructions are at least as fast | ||||||
283 | /// (throughput) as RSQRTPS/VRSQRTPS followed by a Newton-Raphson iteration. | ||||||
284 | bool HasFastVectorFSQRT = false; | ||||||
285 | |||||||
286 | /// True if 8-bit divisions are significantly faster than | ||||||
287 | /// 32-bit divisions and should be used when possible. | ||||||
288 | bool HasSlowDivide32 = false; | ||||||
289 | |||||||
290 | /// True if 32-bit divides are significantly faster than | ||||||
291 | /// 64-bit divisions and should be used when possible. | ||||||
292 | bool HasSlowDivide64 = false; | ||||||
293 | |||||||
294 | /// True if LZCNT instruction is fast. | ||||||
295 | bool HasFastLZCNT = false; | ||||||
296 | |||||||
297 | /// True if SHLD based rotate is fast. | ||||||
298 | bool HasFastSHLDRotate = false; | ||||||
299 | |||||||
300 | /// True if the processor supports macrofusion. | ||||||
301 | bool HasMacroFusion = false; | ||||||
302 | |||||||
303 | /// True if the processor supports branch fusion. | ||||||
304 | bool HasBranchFusion = false; | ||||||
305 | |||||||
306 | /// True if the processor has enhanced REP MOVSB/STOSB. | ||||||
307 | bool HasERMSB = false; | ||||||
308 | |||||||
309 | /// True if the processor has fast short REP MOV. | ||||||
310 | bool HasFSRM = false; | ||||||
311 | |||||||
312 | /// True if the short functions should be padded to prevent | ||||||
313 | /// a stall when returning too early. | ||||||
314 | bool PadShortFunctions = false; | ||||||
315 | |||||||
316 | /// True if two memory operand instructions should use a temporary register | ||||||
317 | /// instead. | ||||||
318 | bool SlowTwoMemOps = false; | ||||||
319 | |||||||
320 | /// True if the LEA instruction inputs have to be ready at address generation | ||||||
321 | /// (AG) time. | ||||||
322 | bool LEAUsesAG = false; | ||||||
323 | |||||||
324 | /// True if the LEA instruction with certain arguments is slow | ||||||
325 | bool SlowLEA = false; | ||||||
326 | |||||||
327 | /// True if the LEA instruction has all three source operands: base, index, | ||||||
328 | /// and offset or if the LEA instruction uses base and index registers where | ||||||
329 | /// the base is EBP, RBP,or R13 | ||||||
330 | bool Slow3OpsLEA = false; | ||||||
331 | |||||||
332 | /// True if INC and DEC instructions are slow when writing to flags | ||||||
333 | bool SlowIncDec = false; | ||||||
334 | |||||||
335 | /// Processor has AVX-512 PreFetch Instructions | ||||||
336 | bool HasPFI = false; | ||||||
337 | |||||||
338 | /// Processor has AVX-512 Exponential and Reciprocal Instructions | ||||||
339 | bool HasERI = false; | ||||||
340 | |||||||
341 | /// Processor has AVX-512 Conflict Detection Instructions | ||||||
342 | bool HasCDI = false; | ||||||
343 | |||||||
344 | /// Processor has AVX-512 population count Instructions | ||||||
345 | bool HasVPOPCNTDQ = false; | ||||||
346 | |||||||
347 | /// Processor has AVX-512 Doubleword and Quadword instructions | ||||||
348 | bool HasDQI = false; | ||||||
349 | |||||||
350 | /// Processor has AVX-512 Byte and Word instructions | ||||||
351 | bool HasBWI = false; | ||||||
352 | |||||||
353 | /// Processor has AVX-512 Vector Length eXtenstions | ||||||
354 | bool HasVLX = false; | ||||||
355 | |||||||
356 | /// Processor has PKU extenstions | ||||||
357 | bool HasPKU = false; | ||||||
358 | |||||||
359 | /// Processor has AVX-512 Vector Neural Network Instructions | ||||||
360 | bool HasVNNI = false; | ||||||
361 | |||||||
362 | /// Processor has AVX Vector Neural Network Instructions | ||||||
363 | bool HasAVXVNNI = false; | ||||||
364 | |||||||
365 | /// Processor has AVX-512 bfloat16 floating-point extensions | ||||||
366 | bool HasBF16 = false; | ||||||
367 | |||||||
368 | /// Processor supports ENQCMD instructions | ||||||
369 | bool HasENQCMD = false; | ||||||
370 | |||||||
371 | /// Processor has AVX-512 Bit Algorithms instructions | ||||||
372 | bool HasBITALG = false; | ||||||
373 | |||||||
374 | /// Processor has AVX-512 vp2intersect instructions | ||||||
375 | bool HasVP2INTERSECT = false; | ||||||
376 | |||||||
377 | /// Processor supports CET SHSTK - Control-Flow Enforcement Technology | ||||||
378 | /// using Shadow Stack | ||||||
379 | bool HasSHSTK = false; | ||||||
380 | |||||||
381 | /// Processor supports Invalidate Process-Context Identifier | ||||||
382 | bool HasINVPCID = false; | ||||||
383 | |||||||
384 | /// Processor has Software Guard Extensions | ||||||
385 | bool HasSGX = false; | ||||||
386 | |||||||
387 | /// Processor supports Flush Cache Line instruction | ||||||
388 | bool HasCLFLUSHOPT = false; | ||||||
389 | |||||||
390 | /// Processor supports Cache Line Write Back instruction | ||||||
391 | bool HasCLWB = false; | ||||||
392 | |||||||
393 | /// Processor supports Write Back No Invalidate instruction | ||||||
394 | bool HasWBNOINVD = false; | ||||||
395 | |||||||
396 | /// Processor support RDPID instruction | ||||||
397 | bool HasRDPID = false; | ||||||
398 | |||||||
399 | /// Processor supports WaitPKG instructions | ||||||
400 | bool HasWAITPKG = false; | ||||||
401 | |||||||
402 | /// Processor supports PCONFIG instruction | ||||||
403 | bool HasPCONFIG = false; | ||||||
404 | |||||||
405 | /// Processor support key locker instructions | ||||||
406 | bool HasKL = false; | ||||||
407 | |||||||
408 | /// Processor support key locker wide instructions | ||||||
409 | bool HasWIDEKL = false; | ||||||
410 | |||||||
411 | /// Processor supports HRESET instruction | ||||||
412 | bool HasHRESET = false; | ||||||
413 | |||||||
414 | /// Processor supports SERIALIZE instruction | ||||||
415 | bool HasSERIALIZE = false; | ||||||
416 | |||||||
417 | /// Processor supports TSXLDTRK instruction | ||||||
418 | bool HasTSXLDTRK = false; | ||||||
419 | |||||||
420 | /// Processor has AMX support | ||||||
421 | bool HasAMXTILE = false; | ||||||
422 | bool HasAMXBF16 = false; | ||||||
423 | bool HasAMXINT8 = false; | ||||||
424 | |||||||
425 | /// Processor supports User Level Interrupt instructions | ||||||
426 | bool HasUINTR = false; | ||||||
427 | |||||||
428 | /// Processor has a single uop BEXTR implementation. | ||||||
429 | bool HasFastBEXTR = false; | ||||||
430 | |||||||
431 | /// Try harder to combine to horizontal vector ops if they are fast. | ||||||
432 | bool HasFastHorizontalOps = false; | ||||||
433 | |||||||
434 | /// Prefer a left/right scalar logical shifts pair over a shift+and pair. | ||||||
435 | bool HasFastScalarShiftMasks = false; | ||||||
436 | |||||||
437 | /// Prefer a left/right vector logical shifts pair over a shift+and pair. | ||||||
438 | bool HasFastVectorShiftMasks = false; | ||||||
439 | |||||||
440 | /// Prefer a movbe over a single-use load + bswap / single-use bswap + store. | ||||||
441 | bool HasFastMOVBE = false; | ||||||
442 | |||||||
443 | /// Use a retpoline thunk rather than indirect calls to block speculative | ||||||
444 | /// execution. | ||||||
445 | bool UseRetpolineIndirectCalls = false; | ||||||
446 | |||||||
447 | /// Use a retpoline thunk or remove any indirect branch to block speculative | ||||||
448 | /// execution. | ||||||
449 | bool UseRetpolineIndirectBranches = false; | ||||||
450 | |||||||
451 | /// Deprecated flag, query `UseRetpolineIndirectCalls` and | ||||||
452 | /// `UseRetpolineIndirectBranches` instead. | ||||||
453 | bool DeprecatedUseRetpoline = false; | ||||||
454 | |||||||
455 | /// When using a retpoline thunk, call an externally provided thunk rather | ||||||
456 | /// than emitting one inside the compiler. | ||||||
457 | bool UseRetpolineExternalThunk = false; | ||||||
458 | |||||||
459 | /// Prevent generation of indirect call/branch instructions from memory, | ||||||
460 | /// and force all indirect call/branch instructions from a register to be | ||||||
461 | /// preceded by an LFENCE. Also decompose RET instructions into a | ||||||
462 | /// POP+LFENCE+JMP sequence. | ||||||
463 | bool UseLVIControlFlowIntegrity = false; | ||||||
464 | |||||||
465 | /// Enable Speculative Execution Side Effect Suppression | ||||||
466 | bool UseSpeculativeExecutionSideEffectSuppression = false; | ||||||
467 | |||||||
468 | /// Insert LFENCE instructions to prevent data speculatively injected into | ||||||
469 | /// loads from being used maliciously. | ||||||
470 | bool UseLVILoadHardening = false; | ||||||
471 | |||||||
472 | /// Use software floating point for code generation. | ||||||
473 | bool UseSoftFloat = false; | ||||||
474 | |||||||
475 | /// Use alias analysis during code generation. | ||||||
476 | bool UseAA = false; | ||||||
477 | |||||||
478 | /// The minimum alignment known to hold of the stack frame on | ||||||
479 | /// entry to the function and which must be maintained by every function. | ||||||
480 | Align stackAlignment = Align(4); | ||||||
481 | |||||||
482 | Align TileConfigAlignment = Align(4); | ||||||
483 | |||||||
484 | /// Max. memset / memcpy size that is turned into rep/movs, rep/stos ops. | ||||||
485 | /// | ||||||
486 | // FIXME: this is a known good value for Yonah. How about others? | ||||||
487 | unsigned MaxInlineSizeThreshold = 128; | ||||||
488 | |||||||
489 | /// Indicates target prefers 128 bit instructions. | ||||||
490 | bool Prefer128Bit = false; | ||||||
491 | |||||||
492 | /// Indicates target prefers 256 bit instructions. | ||||||
493 | bool Prefer256Bit = false; | ||||||
494 | |||||||
495 | /// Indicates target prefers AVX512 mask registers. | ||||||
496 | bool PreferMaskRegisters = false; | ||||||
497 | |||||||
498 | /// Use Goldmont specific floating point div/sqrt costs. | ||||||
499 | bool UseGLMDivSqrtCosts = false; | ||||||
500 | |||||||
501 | /// What processor and OS we're targeting. | ||||||
502 | Triple TargetTriple; | ||||||
503 | |||||||
504 | /// GlobalISel related APIs. | ||||||
505 | std::unique_ptr<CallLowering> CallLoweringInfo; | ||||||
506 | std::unique_ptr<LegalizerInfo> Legalizer; | ||||||
507 | std::unique_ptr<RegisterBankInfo> RegBankInfo; | ||||||
508 | std::unique_ptr<InstructionSelector> InstSelector; | ||||||
509 | |||||||
510 | private: | ||||||
511 | /// Override the stack alignment. | ||||||
512 | MaybeAlign StackAlignOverride; | ||||||
513 | |||||||
514 | /// Preferred vector width from function attribute. | ||||||
515 | unsigned PreferVectorWidthOverride; | ||||||
516 | |||||||
517 | /// Resolved preferred vector width from function attribute and subtarget | ||||||
518 | /// features. | ||||||
519 | unsigned PreferVectorWidth = UINT32_MAX(4294967295U); | ||||||
520 | |||||||
521 | /// Required vector width from function attribute. | ||||||
522 | unsigned RequiredVectorWidth; | ||||||
523 | |||||||
524 | /// True if compiling for 64-bit, false for 16-bit or 32-bit. | ||||||
525 | bool In64BitMode = false; | ||||||
526 | |||||||
527 | /// True if compiling for 32-bit, false for 16-bit or 64-bit. | ||||||
528 | bool In32BitMode = false; | ||||||
529 | |||||||
530 | /// True if compiling for 16-bit, false for 32-bit or 64-bit. | ||||||
531 | bool In16BitMode = false; | ||||||
532 | |||||||
533 | X86SelectionDAGInfo TSInfo; | ||||||
534 | // Ordering here is important. X86InstrInfo initializes X86RegisterInfo which | ||||||
535 | // X86TargetLowering needs. | ||||||
536 | X86InstrInfo InstrInfo; | ||||||
537 | X86TargetLowering TLInfo; | ||||||
538 | X86FrameLowering FrameLowering; | ||||||
539 | |||||||
540 | public: | ||||||
541 | /// This constructor initializes the data members to match that | ||||||
542 | /// of the specified triple. | ||||||
543 | /// | ||||||
544 | X86Subtarget(const Triple &TT, StringRef CPU, StringRef TuneCPU, StringRef FS, | ||||||
545 | const X86TargetMachine &TM, MaybeAlign StackAlignOverride, | ||||||
546 | unsigned PreferVectorWidthOverride, | ||||||
547 | unsigned RequiredVectorWidth); | ||||||
548 | |||||||
549 | const X86TargetLowering *getTargetLowering() const override { | ||||||
550 | return &TLInfo; | ||||||
551 | } | ||||||
552 | |||||||
553 | const X86InstrInfo *getInstrInfo() const override { return &InstrInfo; } | ||||||
554 | |||||||
555 | const X86FrameLowering *getFrameLowering() const override { | ||||||
556 | return &FrameLowering; | ||||||
557 | } | ||||||
558 | |||||||
559 | const X86SelectionDAGInfo *getSelectionDAGInfo() const override { | ||||||
560 | return &TSInfo; | ||||||
561 | } | ||||||
562 | |||||||
563 | const X86RegisterInfo *getRegisterInfo() const override { | ||||||
564 | return &getInstrInfo()->getRegisterInfo(); | ||||||
565 | } | ||||||
566 | |||||||
567 | unsigned getTileConfigSize() const { return 64; } | ||||||
568 | Align getTileConfigAlignment() const { return TileConfigAlignment; } | ||||||
569 | |||||||
570 | /// Returns the minimum alignment known to hold of the | ||||||
571 | /// stack frame on entry to the function and which must be maintained by every | ||||||
572 | /// function for this subtarget. | ||||||
573 | Align getStackAlignment() const { return stackAlignment; } | ||||||
574 | |||||||
575 | /// Returns the maximum memset / memcpy size | ||||||
576 | /// that still makes it profitable to inline the call. | ||||||
577 | unsigned getMaxInlineSizeThreshold() const { return MaxInlineSizeThreshold; } | ||||||
578 | |||||||
579 | /// ParseSubtargetFeatures - Parses features string setting specified | ||||||
580 | /// subtarget options. Definition of function is auto generated by tblgen. | ||||||
581 | void ParseSubtargetFeatures(StringRef CPU, StringRef TuneCPU, StringRef FS); | ||||||
582 | |||||||
583 | /// Methods used by Global ISel | ||||||
584 | const CallLowering *getCallLowering() const override; | ||||||
585 | InstructionSelector *getInstructionSelector() const override; | ||||||
586 | const LegalizerInfo *getLegalizerInfo() const override; | ||||||
587 | const RegisterBankInfo *getRegBankInfo() const override; | ||||||
588 | |||||||
589 | private: | ||||||
590 | /// Initialize the full set of dependencies so we can use an initializer | ||||||
591 | /// list for X86Subtarget. | ||||||
592 | X86Subtarget &initializeSubtargetDependencies(StringRef CPU, | ||||||
593 | StringRef TuneCPU, | ||||||
594 | StringRef FS); | ||||||
595 | void initSubtargetFeatures(StringRef CPU, StringRef TuneCPU, StringRef FS); | ||||||
596 | |||||||
597 | public: | ||||||
598 | /// Is this x86_64? (disregarding specific ABI / programming model) | ||||||
599 | bool is64Bit() const { | ||||||
600 | return In64BitMode; | ||||||
601 | } | ||||||
602 | |||||||
603 | bool is32Bit() const { | ||||||
604 | return In32BitMode; | ||||||
605 | } | ||||||
606 | |||||||
607 | bool is16Bit() const { | ||||||
608 | return In16BitMode; | ||||||
609 | } | ||||||
610 | |||||||
611 | /// Is this x86_64 with the ILP32 programming model (x32 ABI)? | ||||||
612 | bool isTarget64BitILP32() const { | ||||||
613 | return In64BitMode && (TargetTriple.isX32() || TargetTriple.isOSNaCl()); | ||||||
614 | } | ||||||
615 | |||||||
616 | /// Is this x86_64 with the LP64 programming model (standard AMD64, no x32)? | ||||||
617 | bool isTarget64BitLP64() const { | ||||||
618 | return In64BitMode && (!TargetTriple.isX32() && !TargetTriple.isOSNaCl()); | ||||||
619 | } | ||||||
620 | |||||||
621 | PICStyles::Style getPICStyle() const { return PICStyle; } | ||||||
622 | void setPICStyle(PICStyles::Style Style) { PICStyle = Style; } | ||||||
623 | |||||||
624 | bool hasX87() const { return HasX87; } | ||||||
625 | bool hasCmpxchg8b() const { return HasCmpxchg8b; } | ||||||
626 | bool hasNOPL() const { return HasNOPL; } | ||||||
627 | // SSE codegen depends on cmovs, and all SSE1+ processors support them. | ||||||
628 | // All 64-bit processors support cmov. | ||||||
629 | bool hasCMov() const { return HasCMov || X86SSELevel >= SSE1 || is64Bit(); } | ||||||
630 | bool hasSSE1() const { return X86SSELevel >= SSE1; } | ||||||
631 | bool hasSSE2() const { return X86SSELevel
| ||||||
632 | bool hasSSE3() const { return X86SSELevel >= SSE3; } | ||||||
633 | bool hasSSSE3() const { return X86SSELevel >= SSSE3; } | ||||||
634 | bool hasSSE41() const { return X86SSELevel >= SSE41; } | ||||||
635 | bool hasSSE42() const { return X86SSELevel >= SSE42; } | ||||||
636 | bool hasAVX() const { return X86SSELevel >= AVX; } | ||||||
637 | bool hasAVX2() const { return X86SSELevel >= AVX2; } | ||||||
638 | bool hasAVX512() const { return X86SSELevel >= AVX512F; } | ||||||
639 | bool hasInt256() const { return hasAVX2(); } | ||||||
640 | bool hasSSE4A() const { return HasSSE4A; } | ||||||
641 | bool hasMMX() const { return X863DNowLevel >= MMX; } | ||||||
642 | bool has3DNow() const { return X863DNowLevel >= ThreeDNow; } | ||||||
643 | bool has3DNowA() const { return X863DNowLevel >= ThreeDNowA; } | ||||||
644 | bool hasPOPCNT() const { return HasPOPCNT; } | ||||||
645 | bool hasAES() const { return HasAES; } | ||||||
646 | bool hasVAES() const { return HasVAES; } | ||||||
647 | bool hasFXSR() const { return HasFXSR; } | ||||||
648 | bool hasXSAVE() const { return HasXSAVE; } | ||||||
649 | bool hasXSAVEOPT() const { return HasXSAVEOPT; } | ||||||
650 | bool hasXSAVEC() const { return HasXSAVEC; } | ||||||
651 | bool hasXSAVES() const { return HasXSAVES; } | ||||||
652 | bool hasPCLMUL() const { return HasPCLMUL; } | ||||||
653 | bool hasVPCLMULQDQ() const { return HasVPCLMULQDQ; } | ||||||
654 | bool hasGFNI() const { return HasGFNI; } | ||||||
655 | // Prefer FMA4 to FMA - its better for commutation/memory folding and | ||||||
656 | // has equal or better performance on all supported targets. | ||||||
657 | bool hasFMA() const { return HasFMA; } | ||||||
658 | bool hasFMA4() const { return HasFMA4; } | ||||||
659 | bool hasAnyFMA() const { return hasFMA() || hasFMA4(); } | ||||||
660 | bool hasXOP() const { return HasXOP; } | ||||||
661 | bool hasTBM() const { return HasTBM; } | ||||||
662 | bool hasLWP() const { return HasLWP; } | ||||||
663 | bool hasMOVBE() const { return HasMOVBE; } | ||||||
664 | bool hasRDRAND() const { return HasRDRAND; } | ||||||
665 | bool hasF16C() const { return HasF16C; } | ||||||
666 | bool hasFSGSBase() const { return HasFSGSBase; } | ||||||
667 | bool hasLZCNT() const { return HasLZCNT; } | ||||||
668 | bool hasBMI() const { return HasBMI; } | ||||||
669 | bool hasBMI2() const { return HasBMI2; } | ||||||
670 | bool hasVBMI() const { return HasVBMI; } | ||||||
671 | bool hasVBMI2() const { return HasVBMI2; } | ||||||
672 | bool hasIFMA() const { return HasIFMA; } | ||||||
673 | bool hasRTM() const { return HasRTM; } | ||||||
674 | bool hasADX() const { return HasADX; } | ||||||
675 | bool hasSHA() const { return HasSHA; } | ||||||
676 | bool hasPRFCHW() const { return HasPRFCHW; } | ||||||
677 | bool hasPREFETCHWT1() const { return HasPREFETCHWT1; } | ||||||
678 | bool hasPrefetchW() const { | ||||||
679 | // The PREFETCHW instruction was added with 3DNow but later CPUs gave it | ||||||
680 | // its own CPUID bit as part of deprecating 3DNow. Intel eventually added | ||||||
681 | // it and KNL has another that prefetches to L2 cache. We assume the | ||||||
682 | // L1 version exists if the L2 version does. | ||||||
683 | return has3DNow() || hasPRFCHW() || hasPREFETCHWT1(); | ||||||
684 | } | ||||||
685 | bool hasSSEPrefetch() const { | ||||||
686 | // We implicitly enable these when we have a write prefix supporting cache | ||||||
687 | // level OR if we have prfchw, but don't already have a read prefetch from | ||||||
688 | // 3dnow. | ||||||
689 | return hasSSE1() || (hasPRFCHW() && !has3DNow()) || hasPREFETCHWT1(); | ||||||
690 | } | ||||||
691 | bool hasRDSEED() const { return HasRDSEED; } | ||||||
692 | bool hasLAHFSAHF() const { return HasLAHFSAHF64 || !is64Bit(); } | ||||||
693 | bool hasMWAITX() const { return HasMWAITX; } | ||||||
694 | bool hasCLZERO() const { return HasCLZERO; } | ||||||
695 | bool hasCLDEMOTE() const { return HasCLDEMOTE; } | ||||||
696 | bool hasMOVDIRI() const { return HasMOVDIRI; } | ||||||
697 | bool hasMOVDIR64B() const { return HasMOVDIR64B; } | ||||||
698 | bool hasPTWRITE() const { return HasPTWRITE; } | ||||||
699 | bool isSHLDSlow() const { return IsSHLDSlow; } | ||||||
700 | bool isPMULLDSlow() const { return IsPMULLDSlow; } | ||||||
701 | bool isPMADDWDSlow() const { return IsPMADDWDSlow; } | ||||||
702 | bool isUnalignedMem16Slow() const { return IsUAMem16Slow; } | ||||||
703 | bool isUnalignedMem32Slow() const { return IsUAMem32Slow; } | ||||||
704 | bool hasSSEUnalignedMem() const { return HasSSEUnalignedMem; } | ||||||
705 | bool hasCmpxchg16b() const { return HasCmpxchg16b && is64Bit(); } | ||||||
706 | bool useLeaForSP() const { return UseLeaForSP; } | ||||||
707 | bool hasPOPCNTFalseDeps() const { return HasPOPCNTFalseDeps; } | ||||||
708 | bool hasLZCNTFalseDeps() const { return HasLZCNTFalseDeps; } | ||||||
709 | bool hasFastVariableCrossLaneShuffle() const { | ||||||
710 | return HasFastVariableCrossLaneShuffle; | ||||||
711 | } | ||||||
712 | bool hasFastVariablePerLaneShuffle() const { | ||||||
713 | return HasFastVariablePerLaneShuffle; | ||||||
714 | } | ||||||
715 | bool insertVZEROUPPER() const { return InsertVZEROUPPER; } | ||||||
716 | bool hasFastGather() const { return HasFastGather; } | ||||||
717 | bool hasFastScalarFSQRT() const { return HasFastScalarFSQRT; } | ||||||
718 | bool hasFastVectorFSQRT() const { return HasFastVectorFSQRT; } | ||||||
719 | bool hasFastLZCNT() const { return HasFastLZCNT; } | ||||||
720 | bool hasFastSHLDRotate() const { return HasFastSHLDRotate; } | ||||||
721 | bool hasFastBEXTR() const { return HasFastBEXTR; } | ||||||
722 | bool hasFastHorizontalOps() const { return HasFastHorizontalOps; } | ||||||
723 | bool hasFastScalarShiftMasks() const { return HasFastScalarShiftMasks; } | ||||||
724 | bool hasFastVectorShiftMasks() const { return HasFastVectorShiftMasks; } | ||||||
725 | bool hasFastMOVBE() const { return HasFastMOVBE; } | ||||||
726 | bool hasMacroFusion() const { return HasMacroFusion; } | ||||||
727 | bool hasBranchFusion() const { return HasBranchFusion; } | ||||||
728 | bool hasERMSB() const { return HasERMSB; } | ||||||
729 | bool hasFSRM() const { return HasFSRM; } | ||||||
730 | bool hasSlowDivide32() const { return HasSlowDivide32; } | ||||||
731 | bool hasSlowDivide64() const { return HasSlowDivide64; } | ||||||
732 | bool padShortFunctions() const { return PadShortFunctions; } | ||||||
733 | bool slowTwoMemOps() const { return SlowTwoMemOps; } | ||||||
734 | bool LEAusesAG() const { return LEAUsesAG; } | ||||||
735 | bool slowLEA() const { return SlowLEA; } | ||||||
736 | bool slow3OpsLEA() const { return Slow3OpsLEA; } | ||||||
737 | bool slowIncDec() const { return SlowIncDec; } | ||||||
738 | bool hasCDI() const { return HasCDI; } | ||||||
739 | bool hasVPOPCNTDQ() const { return HasVPOPCNTDQ; } | ||||||
740 | bool hasPFI() const { return HasPFI; } | ||||||
741 | bool hasERI() const { return HasERI; } | ||||||
742 | bool hasDQI() const { return HasDQI; } | ||||||
743 | bool hasBWI() const { return HasBWI; } | ||||||
744 | bool hasVLX() const { return HasVLX; } | ||||||
745 | bool hasPKU() const { return HasPKU; } | ||||||
746 | bool hasVNNI() const { return HasVNNI; } | ||||||
747 | bool hasBF16() const { return HasBF16; } | ||||||
748 | bool hasVP2INTERSECT() const { return HasVP2INTERSECT; } | ||||||
749 | bool hasBITALG() const { return HasBITALG; } | ||||||
750 | bool hasSHSTK() const { return HasSHSTK; } | ||||||
751 | bool hasCLFLUSHOPT() const { return HasCLFLUSHOPT; } | ||||||
752 | bool hasCLWB() const { return HasCLWB; } | ||||||
753 | bool hasWBNOINVD() const { return HasWBNOINVD; } | ||||||
754 | bool hasRDPID() const { return HasRDPID; } | ||||||
755 | bool hasWAITPKG() const { return HasWAITPKG; } | ||||||
756 | bool hasPCONFIG() const { return HasPCONFIG; } | ||||||
757 | bool hasSGX() const { return HasSGX; } | ||||||
758 | bool hasINVPCID() const { return HasINVPCID; } | ||||||
759 | bool hasENQCMD() const { return HasENQCMD; } | ||||||
760 | bool hasKL() const { return HasKL; } | ||||||
761 | bool hasWIDEKL() const { return HasWIDEKL; } | ||||||
762 | bool hasHRESET() const { return HasHRESET; } | ||||||
763 | bool hasSERIALIZE() const { return HasSERIALIZE; } | ||||||
764 | bool hasTSXLDTRK() const { return HasTSXLDTRK; } | ||||||
765 | bool hasUINTR() const { return HasUINTR; } | ||||||
766 | bool useRetpolineIndirectCalls() const { return UseRetpolineIndirectCalls; } | ||||||
767 | bool useRetpolineIndirectBranches() const { | ||||||
768 | return UseRetpolineIndirectBranches; | ||||||
769 | } | ||||||
770 | bool hasAVXVNNI() const { return HasAVXVNNI; } | ||||||
771 | bool hasAMXTILE() const { return HasAMXTILE; } | ||||||
772 | bool hasAMXBF16() const { return HasAMXBF16; } | ||||||
773 | bool hasAMXINT8() const { return HasAMXINT8; } | ||||||
774 | bool useRetpolineExternalThunk() const { return UseRetpolineExternalThunk; } | ||||||
775 | |||||||
776 | // These are generic getters that OR together all of the thunk types | ||||||
777 | // supported by the subtarget. Therefore useIndirectThunk*() will return true | ||||||
778 | // if any respective thunk feature is enabled. | ||||||
779 | bool useIndirectThunkCalls() const { | ||||||
780 | return useRetpolineIndirectCalls() || useLVIControlFlowIntegrity(); | ||||||
781 | } | ||||||
782 | bool useIndirectThunkBranches() const { | ||||||
783 | return useRetpolineIndirectBranches() || useLVIControlFlowIntegrity(); | ||||||
784 | } | ||||||
785 | |||||||
786 | bool preferMaskRegisters() const { return PreferMaskRegisters; } | ||||||
787 | bool useGLMDivSqrtCosts() const { return UseGLMDivSqrtCosts; } | ||||||
788 | bool useLVIControlFlowIntegrity() const { return UseLVIControlFlowIntegrity; } | ||||||
789 | bool useLVILoadHardening() const { return UseLVILoadHardening; } | ||||||
790 | bool useSpeculativeExecutionSideEffectSuppression() const { | ||||||
791 | return UseSpeculativeExecutionSideEffectSuppression; | ||||||
792 | } | ||||||
793 | |||||||
794 | unsigned getPreferVectorWidth() const { return PreferVectorWidth; } | ||||||
795 | unsigned getRequiredVectorWidth() const { return RequiredVectorWidth; } | ||||||
796 | |||||||
797 | // Helper functions to determine when we should allow widening to 512-bit | ||||||
798 | // during codegen. | ||||||
799 | // TODO: Currently we're always allowing widening on CPUs without VLX, | ||||||
800 | // because for many cases we don't have a better option. | ||||||
801 | bool canExtendTo512DQ() const { | ||||||
802 | return hasAVX512() && (!hasVLX() || getPreferVectorWidth() >= 512); | ||||||
803 | } | ||||||
804 | bool canExtendTo512BW() const { | ||||||
805 | return hasBWI() && canExtendTo512DQ(); | ||||||
806 | } | ||||||
807 | |||||||
808 | // If there are no 512-bit vectors and we prefer not to use 512-bit registers, | ||||||
809 | // disable them in the legalizer. | ||||||
810 | bool useAVX512Regs() const { | ||||||
811 | return hasAVX512() && (canExtendTo512DQ() || RequiredVectorWidth > 256); | ||||||
812 | } | ||||||
813 | |||||||
814 | bool useBWIRegs() const { | ||||||
815 | return hasBWI() && useAVX512Regs(); | ||||||
816 | } | ||||||
817 | |||||||
818 | bool isXRaySupported() const override { return is64Bit(); } | ||||||
819 | |||||||
820 | /// TODO: to be removed later and replaced with suitable properties | ||||||
821 | bool isAtom() const { return X86ProcFamily == IntelAtom; } | ||||||
822 | bool isSLM() const { return X86ProcFamily == IntelSLM; } | ||||||
823 | bool useSoftFloat() const { return UseSoftFloat; } | ||||||
824 | bool useAA() const override { return UseAA; } | ||||||
825 | |||||||
826 | /// Use mfence if we have SSE2 or we're on x86-64 (even if we asked for | ||||||
827 | /// no-sse2). There isn't any reason to disable it if the target processor | ||||||
828 | /// supports it. | ||||||
829 | bool hasMFence() const { return hasSSE2() || is64Bit(); } | ||||||
830 | |||||||
831 | const Triple &getTargetTriple() const { return TargetTriple; } | ||||||
832 | |||||||
833 | bool isTargetDarwin() const { return TargetTriple.isOSDarwin(); } | ||||||
834 | bool isTargetFreeBSD() const { return TargetTriple.isOSFreeBSD(); } | ||||||
835 | bool isTargetDragonFly() const { return TargetTriple.isOSDragonFly(); } | ||||||
836 | bool isTargetSolaris() const { return TargetTriple.isOSSolaris(); } | ||||||
837 | bool isTargetPS4() const { return TargetTriple.isPS4CPU(); } | ||||||
838 | |||||||
839 | bool isTargetELF() const { return TargetTriple.isOSBinFormatELF(); } | ||||||
840 | bool isTargetCOFF() const { return TargetTriple.isOSBinFormatCOFF(); } | ||||||
841 | bool isTargetMachO() const { return TargetTriple.isOSBinFormatMachO(); } | ||||||
842 | |||||||
843 | bool isTargetLinux() const { return TargetTriple.isOSLinux(); } | ||||||
844 | bool isTargetKFreeBSD() const { return TargetTriple.isOSKFreeBSD(); } | ||||||
845 | bool isTargetGlibc() const { return TargetTriple.isOSGlibc(); } | ||||||
846 | bool isTargetAndroid() const { return TargetTriple.isAndroid(); } | ||||||
847 | bool isTargetNaCl() const { return TargetTriple.isOSNaCl(); } | ||||||
848 | bool isTargetNaCl32() const { return isTargetNaCl() && !is64Bit(); } | ||||||
849 | bool isTargetNaCl64() const { return isTargetNaCl() && is64Bit(); } | ||||||
850 | bool isTargetMCU() const { return TargetTriple.isOSIAMCU(); } | ||||||
851 | bool isTargetFuchsia() const { return TargetTriple.isOSFuchsia(); } | ||||||
852 | |||||||
853 | bool isTargetWindowsMSVC() const { | ||||||
854 | return TargetTriple.isWindowsMSVCEnvironment(); | ||||||
855 | } | ||||||
856 | |||||||
857 | bool isTargetWindowsCoreCLR() const { | ||||||
858 | return TargetTriple.isWindowsCoreCLREnvironment(); | ||||||
859 | } | ||||||
860 | |||||||
861 | bool isTargetWindowsCygwin() const { | ||||||
862 | return TargetTriple.isWindowsCygwinEnvironment(); | ||||||
863 | } | ||||||
864 | |||||||
865 | bool isTargetWindowsGNU() const { | ||||||
866 | return TargetTriple.isWindowsGNUEnvironment(); | ||||||
867 | } | ||||||
868 | |||||||
869 | bool isTargetWindowsItanium() const { | ||||||
870 | return TargetTriple.isWindowsItaniumEnvironment(); | ||||||
871 | } | ||||||
872 | |||||||
873 | bool isTargetCygMing() const { return TargetTriple.isOSCygMing(); } | ||||||
874 | |||||||
875 | bool isOSWindows() const { return TargetTriple.isOSWindows(); } | ||||||
876 | |||||||
877 | bool isTargetWin64() const { return In64BitMode && isOSWindows(); } | ||||||
878 | |||||||
879 | bool isTargetWin32() const { return !In64BitMode && isOSWindows(); } | ||||||
880 | |||||||
881 | bool isPICStyleGOT() const { return PICStyle == PICStyles::Style::GOT; } | ||||||
882 | bool isPICStyleRIPRel() const { return PICStyle == PICStyles::Style::RIPRel; } | ||||||
883 | |||||||
884 | bool isPICStyleStubPIC() const { | ||||||
885 | return PICStyle == PICStyles::Style::StubPIC; | ||||||
886 | } | ||||||
887 | |||||||
888 | bool isPositionIndependent() const; | ||||||
889 | |||||||
890 | bool isCallingConvWin64(CallingConv::ID CC) const { | ||||||
891 | switch (CC) { | ||||||
892 | // On Win64, all these conventions just use the default convention. | ||||||
893 | case CallingConv::C: | ||||||
894 | case CallingConv::Fast: | ||||||
895 | case CallingConv::Tail: | ||||||
896 | case CallingConv::Swift: | ||||||
897 | case CallingConv::SwiftTail: | ||||||
898 | case CallingConv::X86_FastCall: | ||||||
899 | case CallingConv::X86_StdCall: | ||||||
900 | case CallingConv::X86_ThisCall: | ||||||
901 | case CallingConv::X86_VectorCall: | ||||||
902 | case CallingConv::Intel_OCL_BI: | ||||||
903 | return isTargetWin64(); | ||||||
904 | // This convention allows using the Win64 convention on other targets. | ||||||
905 | case CallingConv::Win64: | ||||||
906 | return true; | ||||||
907 | // This convention allows using the SysV convention on Windows targets. | ||||||
908 | case CallingConv::X86_64_SysV: | ||||||
909 | return false; | ||||||
910 | // Otherwise, who knows what this is. | ||||||
911 | default: | ||||||
912 | return false; | ||||||
913 | } | ||||||
914 | } | ||||||
915 | |||||||
916 | /// Classify a global variable reference for the current subtarget according | ||||||
917 | /// to how we should reference it in a non-pcrel context. | ||||||
918 | unsigned char classifyLocalReference(const GlobalValue *GV) const; | ||||||
919 | |||||||
920 | unsigned char classifyGlobalReference(const GlobalValue *GV, | ||||||
921 | const Module &M) const; | ||||||
922 | unsigned char classifyGlobalReference(const GlobalValue *GV) const; | ||||||
923 | |||||||
924 | /// Classify a global function reference for the current subtarget. | ||||||
925 | unsigned char classifyGlobalFunctionReference(const GlobalValue *GV, | ||||||
926 | const Module &M) const; | ||||||
927 | unsigned char classifyGlobalFunctionReference(const GlobalValue *GV) const; | ||||||
928 | |||||||
929 | /// Classify a blockaddress reference for the current subtarget according to | ||||||
930 | /// how we should reference it in a non-pcrel context. | ||||||
931 | unsigned char classifyBlockAddressReference() const; | ||||||
932 | |||||||
933 | /// Return true if the subtarget allows calls to immediate address. | ||||||
934 | bool isLegalToCallImmediateAddr() const; | ||||||
935 | |||||||
936 | /// If we are using indirect thunks, we need to expand indirectbr to avoid it | ||||||
937 | /// lowering to an actual indirect jump. | ||||||
938 | bool enableIndirectBrExpand() const override { | ||||||
939 | return useIndirectThunkBranches(); | ||||||
940 | } | ||||||
941 | |||||||
942 | /// Enable the MachineScheduler pass for all X86 subtargets. | ||||||
943 | bool enableMachineScheduler() const override { return true; } | ||||||
944 | |||||||
945 | bool enableEarlyIfConversion() const override; | ||||||
946 | |||||||
947 | void getPostRAMutations(std::vector<std::unique_ptr<ScheduleDAGMutation>> | ||||||
948 | &Mutations) const override; | ||||||
949 | |||||||
950 | AntiDepBreakMode getAntiDepBreakMode() const override { | ||||||
951 | return TargetSubtargetInfo::ANTIDEP_CRITICAL; | ||||||
952 | } | ||||||
953 | |||||||
954 | bool enableAdvancedRASplitCost() const override { return false; } | ||||||
955 | }; | ||||||
956 | |||||||
957 | } // end namespace llvm | ||||||
958 | |||||||
959 | #endif // LLVM_LIB_TARGET_X86_X86SUBTARGET_H |
1 | //===- llvm/InstrTypes.h - Important Instruction subclasses -----*- C++ -*-===// |
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 file defines various meta classes of instructions that exist in the VM |
10 | // representation. Specific concrete subclasses of these may be found in the |
11 | // i*.h files... |
12 | // |
13 | //===----------------------------------------------------------------------===// |
14 | |
15 | #ifndef LLVM_IR_INSTRTYPES_H |
16 | #define LLVM_IR_INSTRTYPES_H |
17 | |
18 | #include "llvm/ADT/ArrayRef.h" |
19 | #include "llvm/ADT/None.h" |
20 | #include "llvm/ADT/Optional.h" |
21 | #include "llvm/ADT/STLExtras.h" |
22 | #include "llvm/ADT/StringMap.h" |
23 | #include "llvm/ADT/StringRef.h" |
24 | #include "llvm/ADT/Twine.h" |
25 | #include "llvm/ADT/iterator_range.h" |
26 | #include "llvm/IR/Attributes.h" |
27 | #include "llvm/IR/CallingConv.h" |
28 | #include "llvm/IR/Constants.h" |
29 | #include "llvm/IR/DerivedTypes.h" |
30 | #include "llvm/IR/Function.h" |
31 | #include "llvm/IR/Instruction.h" |
32 | #include "llvm/IR/LLVMContext.h" |
33 | #include "llvm/IR/OperandTraits.h" |
34 | #include "llvm/IR/Type.h" |
35 | #include "llvm/IR/User.h" |
36 | #include "llvm/IR/Value.h" |
37 | #include "llvm/Support/Casting.h" |
38 | #include "llvm/Support/ErrorHandling.h" |
39 | #include <algorithm> |
40 | #include <cassert> |
41 | #include <cstddef> |
42 | #include <cstdint> |
43 | #include <iterator> |
44 | #include <string> |
45 | #include <vector> |
46 | |
47 | namespace llvm { |
48 | |
49 | namespace Intrinsic { |
50 | typedef unsigned ID; |
51 | } |
52 | |
53 | //===----------------------------------------------------------------------===// |
54 | // UnaryInstruction Class |
55 | //===----------------------------------------------------------------------===// |
56 | |
57 | class UnaryInstruction : public Instruction { |
58 | protected: |
59 | UnaryInstruction(Type *Ty, unsigned iType, Value *V, |
60 | Instruction *IB = nullptr) |
61 | : Instruction(Ty, iType, &Op<0>(), 1, IB) { |
62 | Op<0>() = V; |
63 | } |
64 | UnaryInstruction(Type *Ty, unsigned iType, Value *V, BasicBlock *IAE) |
65 | : Instruction(Ty, iType, &Op<0>(), 1, IAE) { |
66 | Op<0>() = V; |
67 | } |
68 | |
69 | public: |
70 | // allocate space for exactly one operand |
71 | void *operator new(size_t S) { return User::operator new(S, 1); } |
72 | void operator delete(void *Ptr) { User::operator delete(Ptr); } |
73 | |
74 | /// Transparently provide more efficient getOperand methods. |
75 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(Value)public: inline Value *getOperand(unsigned) const; inline void setOperand(unsigned, Value*); inline op_iterator op_begin(); inline const_op_iterator op_begin() const; inline op_iterator op_end(); inline const_op_iterator op_end() const; protected : template <int> inline Use &Op(); template <int > inline const Use &Op() const; public: inline unsigned getNumOperands() const; |
76 | |
77 | // Methods for support type inquiry through isa, cast, and dyn_cast: |
78 | static bool classof(const Instruction *I) { |
79 | return I->isUnaryOp() || |
80 | I->getOpcode() == Instruction::Alloca || |
81 | I->getOpcode() == Instruction::Load || |
82 | I->getOpcode() == Instruction::VAArg || |
83 | I->getOpcode() == Instruction::ExtractValue || |
84 | (I->getOpcode() >= CastOpsBegin && I->getOpcode() < CastOpsEnd); |
85 | } |
86 | static bool classof(const Value *V) { |
87 | return isa<Instruction>(V) && classof(cast<Instruction>(V)); |
88 | } |
89 | }; |
90 | |
91 | template <> |
92 | struct OperandTraits<UnaryInstruction> : |
93 | public FixedNumOperandTraits<UnaryInstruction, 1> { |
94 | }; |
95 | |
96 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(UnaryInstruction, Value)UnaryInstruction::op_iterator UnaryInstruction::op_begin() { return OperandTraits<UnaryInstruction>::op_begin(this); } UnaryInstruction ::const_op_iterator UnaryInstruction::op_begin() const { return OperandTraits<UnaryInstruction>::op_begin(const_cast< UnaryInstruction*>(this)); } UnaryInstruction::op_iterator UnaryInstruction::op_end() { return OperandTraits<UnaryInstruction >::op_end(this); } UnaryInstruction::const_op_iterator UnaryInstruction ::op_end() const { return OperandTraits<UnaryInstruction> ::op_end(const_cast<UnaryInstruction*>(this)); } Value * UnaryInstruction::getOperand(unsigned i_nocapture) const { (static_cast <bool> (i_nocapture < OperandTraits<UnaryInstruction >::operands(this) && "getOperand() out of range!") ? void (0) : __assert_fail ("i_nocapture < OperandTraits<UnaryInstruction>::operands(this) && \"getOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 96, __extension__ __PRETTY_FUNCTION__)); return cast_or_null <Value>( OperandTraits<UnaryInstruction>::op_begin (const_cast<UnaryInstruction*>(this))[i_nocapture].get( )); } void UnaryInstruction::setOperand(unsigned i_nocapture, Value *Val_nocapture) { (static_cast <bool> (i_nocapture < OperandTraits<UnaryInstruction>::operands(this) && "setOperand() out of range!") ? void (0) : __assert_fail ("i_nocapture < OperandTraits<UnaryInstruction>::operands(this) && \"setOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 96, __extension__ __PRETTY_FUNCTION__)); OperandTraits<UnaryInstruction >::op_begin(this)[i_nocapture] = Val_nocapture; } unsigned UnaryInstruction::getNumOperands() const { return OperandTraits <UnaryInstruction>::operands(this); } template <int Idx_nocapture > Use &UnaryInstruction::Op() { return this->OpFrom <Idx_nocapture>(this); } template <int Idx_nocapture > const Use &UnaryInstruction::Op() const { return this ->OpFrom<Idx_nocapture>(this); } |
97 | |
98 | //===----------------------------------------------------------------------===// |
99 | // UnaryOperator Class |
100 | //===----------------------------------------------------------------------===// |
101 | |
102 | class UnaryOperator : public UnaryInstruction { |
103 | void AssertOK(); |
104 | |
105 | protected: |
106 | UnaryOperator(UnaryOps iType, Value *S, Type *Ty, |
107 | const Twine &Name, Instruction *InsertBefore); |
108 | UnaryOperator(UnaryOps iType, Value *S, Type *Ty, |
109 | const Twine &Name, BasicBlock *InsertAtEnd); |
110 | |
111 | // Note: Instruction needs to be a friend here to call cloneImpl. |
112 | friend class Instruction; |
113 | |
114 | UnaryOperator *cloneImpl() const; |
115 | |
116 | public: |
117 | |
118 | /// Construct a unary instruction, given the opcode and an operand. |
119 | /// Optionally (if InstBefore is specified) insert the instruction |
120 | /// into a BasicBlock right before the specified instruction. The specified |
121 | /// Instruction is allowed to be a dereferenced end iterator. |
122 | /// |
123 | static UnaryOperator *Create(UnaryOps Op, Value *S, |
124 | const Twine &Name = Twine(), |
125 | Instruction *InsertBefore = nullptr); |
126 | |
127 | /// Construct a unary instruction, given the opcode and an operand. |
128 | /// Also automatically insert this instruction to the end of the |
129 | /// BasicBlock specified. |
130 | /// |
131 | static UnaryOperator *Create(UnaryOps Op, Value *S, |
132 | const Twine &Name, |
133 | BasicBlock *InsertAtEnd); |
134 | |
135 | /// These methods just forward to Create, and are useful when you |
136 | /// statically know what type of instruction you're going to create. These |
137 | /// helpers just save some typing. |
138 | #define HANDLE_UNARY_INST(N, OPC, CLASS) \ |
139 | static UnaryOperator *Create##OPC(Value *V, const Twine &Name = "") {\ |
140 | return Create(Instruction::OPC, V, Name);\ |
141 | } |
142 | #include "llvm/IR/Instruction.def" |
143 | #define HANDLE_UNARY_INST(N, OPC, CLASS) \ |
144 | static UnaryOperator *Create##OPC(Value *V, const Twine &Name, \ |
145 | BasicBlock *BB) {\ |
146 | return Create(Instruction::OPC, V, Name, BB);\ |
147 | } |
148 | #include "llvm/IR/Instruction.def" |
149 | #define HANDLE_UNARY_INST(N, OPC, CLASS) \ |
150 | static UnaryOperator *Create##OPC(Value *V, const Twine &Name, \ |
151 | Instruction *I) {\ |
152 | return Create(Instruction::OPC, V, Name, I);\ |
153 | } |
154 | #include "llvm/IR/Instruction.def" |
155 | |
156 | static UnaryOperator * |
157 | CreateWithCopiedFlags(UnaryOps Opc, Value *V, Instruction *CopyO, |
158 | const Twine &Name = "", |
159 | Instruction *InsertBefore = nullptr) { |
160 | UnaryOperator *UO = Create(Opc, V, Name, InsertBefore); |
161 | UO->copyIRFlags(CopyO); |
162 | return UO; |
163 | } |
164 | |
165 | static UnaryOperator *CreateFNegFMF(Value *Op, Instruction *FMFSource, |
166 | const Twine &Name = "", |
167 | Instruction *InsertBefore = nullptr) { |
168 | return CreateWithCopiedFlags(Instruction::FNeg, Op, FMFSource, Name, |
169 | InsertBefore); |
170 | } |
171 | |
172 | UnaryOps getOpcode() const { |
173 | return static_cast<UnaryOps>(Instruction::getOpcode()); |
174 | } |
175 | |
176 | // Methods for support type inquiry through isa, cast, and dyn_cast: |
177 | static bool classof(const Instruction *I) { |
178 | return I->isUnaryOp(); |
179 | } |
180 | static bool classof(const Value *V) { |
181 | return isa<Instruction>(V) && classof(cast<Instruction>(V)); |
182 | } |
183 | }; |
184 | |
185 | //===----------------------------------------------------------------------===// |
186 | // BinaryOperator Class |
187 | //===----------------------------------------------------------------------===// |
188 | |
189 | class BinaryOperator : public Instruction { |
190 | void AssertOK(); |
191 | |
192 | protected: |
193 | BinaryOperator(BinaryOps iType, Value *S1, Value *S2, Type *Ty, |
194 | const Twine &Name, Instruction *InsertBefore); |
195 | BinaryOperator(BinaryOps iType, Value *S1, Value *S2, Type *Ty, |
196 | const Twine &Name, BasicBlock *InsertAtEnd); |
197 | |
198 | // Note: Instruction needs to be a friend here to call cloneImpl. |
199 | friend class Instruction; |
200 | |
201 | BinaryOperator *cloneImpl() const; |
202 | |
203 | public: |
204 | // allocate space for exactly two operands |
205 | void *operator new(size_t S) { return User::operator new(S, 2); } |
206 | void operator delete(void *Ptr) { User::operator delete(Ptr); } |
207 | |
208 | /// Transparently provide more efficient getOperand methods. |
209 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(Value)public: inline Value *getOperand(unsigned) const; inline void setOperand(unsigned, Value*); inline op_iterator op_begin(); inline const_op_iterator op_begin() const; inline op_iterator op_end(); inline const_op_iterator op_end() const; protected : template <int> inline Use &Op(); template <int > inline const Use &Op() const; public: inline unsigned getNumOperands() const; |
210 | |
211 | /// Construct a binary instruction, given the opcode and the two |
212 | /// operands. Optionally (if InstBefore is specified) insert the instruction |
213 | /// into a BasicBlock right before the specified instruction. The specified |
214 | /// Instruction is allowed to be a dereferenced end iterator. |
215 | /// |
216 | static BinaryOperator *Create(BinaryOps Op, Value *S1, Value *S2, |
217 | const Twine &Name = Twine(), |
218 | Instruction *InsertBefore = nullptr); |
219 | |
220 | /// Construct a binary instruction, given the opcode and the two |
221 | /// operands. Also automatically insert this instruction to the end of the |
222 | /// BasicBlock specified. |
223 | /// |
224 | static BinaryOperator *Create(BinaryOps Op, Value *S1, Value *S2, |
225 | const Twine &Name, BasicBlock *InsertAtEnd); |
226 | |
227 | /// These methods just forward to Create, and are useful when you |
228 | /// statically know what type of instruction you're going to create. These |
229 | /// helpers just save some typing. |
230 | #define HANDLE_BINARY_INST(N, OPC, CLASS) \ |
231 | static BinaryOperator *Create##OPC(Value *V1, Value *V2, \ |
232 | const Twine &Name = "") {\ |
233 | return Create(Instruction::OPC, V1, V2, Name);\ |
234 | } |
235 | #include "llvm/IR/Instruction.def" |
236 | #define HANDLE_BINARY_INST(N, OPC, CLASS) \ |
237 | static BinaryOperator *Create##OPC(Value *V1, Value *V2, \ |
238 | const Twine &Name, BasicBlock *BB) {\ |
239 | return Create(Instruction::OPC, V1, V2, Name, BB);\ |
240 | } |
241 | #include "llvm/IR/Instruction.def" |
242 | #define HANDLE_BINARY_INST(N, OPC, CLASS) \ |
243 | static BinaryOperator *Create##OPC(Value *V1, Value *V2, \ |
244 | const Twine &Name, Instruction *I) {\ |
245 | return Create(Instruction::OPC, V1, V2, Name, I);\ |
246 | } |
247 | #include "llvm/IR/Instruction.def" |
248 | |
249 | static BinaryOperator * |
250 | CreateWithCopiedFlags(BinaryOps Opc, Value *V1, Value *V2, Instruction *CopyO, |
251 | const Twine &Name = "", |
252 | Instruction *InsertBefore = nullptr) { |
253 | BinaryOperator *BO = Create(Opc, V1, V2, Name, InsertBefore); |
254 | BO->copyIRFlags(CopyO); |
255 | return BO; |
256 | } |
257 | |
258 | static BinaryOperator *CreateFAddFMF(Value *V1, Value *V2, |
259 | Instruction *FMFSource, |
260 | const Twine &Name = "") { |
261 | return CreateWithCopiedFlags(Instruction::FAdd, V1, V2, FMFSource, Name); |
262 | } |
263 | static BinaryOperator *CreateFSubFMF(Value *V1, Value *V2, |
264 | Instruction *FMFSource, |
265 | const Twine &Name = "") { |
266 | return CreateWithCopiedFlags(Instruction::FSub, V1, V2, FMFSource, Name); |
267 | } |
268 | static BinaryOperator *CreateFMulFMF(Value *V1, Value *V2, |
269 | Instruction *FMFSource, |
270 | const Twine &Name = "") { |
271 | return CreateWithCopiedFlags(Instruction::FMul, V1, V2, FMFSource, Name); |
272 | } |
273 | static BinaryOperator *CreateFDivFMF(Value *V1, Value *V2, |
274 | Instruction *FMFSource, |
275 | const Twine &Name = "") { |
276 | return CreateWithCopiedFlags(Instruction::FDiv, V1, V2, FMFSource, Name); |
277 | } |
278 | static BinaryOperator *CreateFRemFMF(Value *V1, Value *V2, |
279 | Instruction *FMFSource, |
280 | const Twine &Name = "") { |
281 | return CreateWithCopiedFlags(Instruction::FRem, V1, V2, FMFSource, Name); |
282 | } |
283 | |
284 | static BinaryOperator *CreateNSW(BinaryOps Opc, Value *V1, Value *V2, |
285 | const Twine &Name = "") { |
286 | BinaryOperator *BO = Create(Opc, V1, V2, Name); |
287 | BO->setHasNoSignedWrap(true); |
288 | return BO; |
289 | } |
290 | static BinaryOperator *CreateNSW(BinaryOps Opc, Value *V1, Value *V2, |
291 | const Twine &Name, BasicBlock *BB) { |
292 | BinaryOperator *BO = Create(Opc, V1, V2, Name, BB); |
293 | BO->setHasNoSignedWrap(true); |
294 | return BO; |
295 | } |
296 | static BinaryOperator *CreateNSW(BinaryOps Opc, Value *V1, Value *V2, |
297 | const Twine &Name, Instruction *I) { |
298 | BinaryOperator *BO = Create(Opc, V1, V2, Name, I); |
299 | BO->setHasNoSignedWrap(true); |
300 | return BO; |
301 | } |
302 | |
303 | static BinaryOperator *CreateNUW(BinaryOps Opc, Value *V1, Value *V2, |
304 | const Twine &Name = "") { |
305 | BinaryOperator *BO = Create(Opc, V1, V2, Name); |
306 | BO->setHasNoUnsignedWrap(true); |
307 | return BO; |
308 | } |
309 | static BinaryOperator *CreateNUW(BinaryOps Opc, Value *V1, Value *V2, |
310 | const Twine &Name, BasicBlock *BB) { |
311 | BinaryOperator *BO = Create(Opc, V1, V2, Name, BB); |
312 | BO->setHasNoUnsignedWrap(true); |
313 | return BO; |
314 | } |
315 | static BinaryOperator *CreateNUW(BinaryOps Opc, Value *V1, Value *V2, |
316 | const Twine &Name, Instruction *I) { |
317 | BinaryOperator *BO = Create(Opc, V1, V2, Name, I); |
318 | BO->setHasNoUnsignedWrap(true); |
319 | return BO; |
320 | } |
321 | |
322 | static BinaryOperator *CreateExact(BinaryOps Opc, Value *V1, Value *V2, |
323 | const Twine &Name = "") { |
324 | BinaryOperator *BO = Create(Opc, V1, V2, Name); |
325 | BO->setIsExact(true); |
326 | return BO; |
327 | } |
328 | static BinaryOperator *CreateExact(BinaryOps Opc, Value *V1, Value *V2, |
329 | const Twine &Name, BasicBlock *BB) { |
330 | BinaryOperator *BO = Create(Opc, V1, V2, Name, BB); |
331 | BO->setIsExact(true); |
332 | return BO; |
333 | } |
334 | static BinaryOperator *CreateExact(BinaryOps Opc, Value *V1, Value *V2, |
335 | const Twine &Name, Instruction *I) { |
336 | BinaryOperator *BO = Create(Opc, V1, V2, Name, I); |
337 | BO->setIsExact(true); |
338 | return BO; |
339 | } |
340 | |
341 | #define DEFINE_HELPERS(OPC, NUWNSWEXACT) \ |
342 | static BinaryOperator *Create##NUWNSWEXACT##OPC(Value *V1, Value *V2, \ |
343 | const Twine &Name = "") { \ |
344 | return Create##NUWNSWEXACT(Instruction::OPC, V1, V2, Name); \ |
345 | } \ |
346 | static BinaryOperator *Create##NUWNSWEXACT##OPC( \ |
347 | Value *V1, Value *V2, const Twine &Name, BasicBlock *BB) { \ |
348 | return Create##NUWNSWEXACT(Instruction::OPC, V1, V2, Name, BB); \ |
349 | } \ |
350 | static BinaryOperator *Create##NUWNSWEXACT##OPC( \ |
351 | Value *V1, Value *V2, const Twine &Name, Instruction *I) { \ |
352 | return Create##NUWNSWEXACT(Instruction::OPC, V1, V2, Name, I); \ |
353 | } |
354 | |
355 | DEFINE_HELPERS(Add, NSW) // CreateNSWAdd |
356 | DEFINE_HELPERS(Add, NUW) // CreateNUWAdd |
357 | DEFINE_HELPERS(Sub, NSW) // CreateNSWSub |
358 | DEFINE_HELPERS(Sub, NUW) // CreateNUWSub |
359 | DEFINE_HELPERS(Mul, NSW) // CreateNSWMul |
360 | DEFINE_HELPERS(Mul, NUW) // CreateNUWMul |
361 | DEFINE_HELPERS(Shl, NSW) // CreateNSWShl |
362 | DEFINE_HELPERS(Shl, NUW) // CreateNUWShl |
363 | |
364 | DEFINE_HELPERS(SDiv, Exact) // CreateExactSDiv |
365 | DEFINE_HELPERS(UDiv, Exact) // CreateExactUDiv |
366 | DEFINE_HELPERS(AShr, Exact) // CreateExactAShr |
367 | DEFINE_HELPERS(LShr, Exact) // CreateExactLShr |
368 | |
369 | #undef DEFINE_HELPERS |
370 | |
371 | /// Helper functions to construct and inspect unary operations (NEG and NOT) |
372 | /// via binary operators SUB and XOR: |
373 | /// |
374 | /// Create the NEG and NOT instructions out of SUB and XOR instructions. |
375 | /// |
376 | static BinaryOperator *CreateNeg(Value *Op, const Twine &Name = "", |
377 | Instruction *InsertBefore = nullptr); |
378 | static BinaryOperator *CreateNeg(Value *Op, const Twine &Name, |
379 | BasicBlock *InsertAtEnd); |
380 | static BinaryOperator *CreateNSWNeg(Value *Op, const Twine &Name = "", |
381 | Instruction *InsertBefore = nullptr); |
382 | static BinaryOperator *CreateNSWNeg(Value *Op, const Twine &Name, |
383 | BasicBlock *InsertAtEnd); |
384 | static BinaryOperator *CreateNUWNeg(Value *Op, const Twine &Name = "", |
385 | Instruction *InsertBefore = nullptr); |
386 | static BinaryOperator *CreateNUWNeg(Value *Op, const Twine &Name, |
387 | BasicBlock *InsertAtEnd); |
388 | static BinaryOperator *CreateNot(Value *Op, const Twine &Name = "", |
389 | Instruction *InsertBefore = nullptr); |
390 | static BinaryOperator *CreateNot(Value *Op, const Twine &Name, |
391 | BasicBlock *InsertAtEnd); |
392 | |
393 | BinaryOps getOpcode() const { |
394 | return static_cast<BinaryOps>(Instruction::getOpcode()); |
395 | } |
396 | |
397 | /// Exchange the two operands to this instruction. |
398 | /// This instruction is safe to use on any binary instruction and |
399 | /// does not modify the semantics of the instruction. If the instruction |
400 | /// cannot be reversed (ie, it's a Div), then return true. |
401 | /// |
402 | bool swapOperands(); |
403 | |
404 | // Methods for support type inquiry through isa, cast, and dyn_cast: |
405 | static bool classof(const Instruction *I) { |
406 | return I->isBinaryOp(); |
407 | } |
408 | static bool classof(const Value *V) { |
409 | return isa<Instruction>(V) && classof(cast<Instruction>(V)); |
410 | } |
411 | }; |
412 | |
413 | template <> |
414 | struct OperandTraits<BinaryOperator> : |
415 | public FixedNumOperandTraits<BinaryOperator, 2> { |
416 | }; |
417 | |
418 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(BinaryOperator, Value)BinaryOperator::op_iterator BinaryOperator::op_begin() { return OperandTraits<BinaryOperator>::op_begin(this); } BinaryOperator ::const_op_iterator BinaryOperator::op_begin() const { return OperandTraits<BinaryOperator>::op_begin(const_cast< BinaryOperator*>(this)); } BinaryOperator::op_iterator BinaryOperator ::op_end() { return OperandTraits<BinaryOperator>::op_end (this); } BinaryOperator::const_op_iterator BinaryOperator::op_end () const { return OperandTraits<BinaryOperator>::op_end (const_cast<BinaryOperator*>(this)); } Value *BinaryOperator ::getOperand(unsigned i_nocapture) const { (static_cast <bool > (i_nocapture < OperandTraits<BinaryOperator>::operands (this) && "getOperand() out of range!") ? void (0) : __assert_fail ("i_nocapture < OperandTraits<BinaryOperator>::operands(this) && \"getOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 418, __extension__ __PRETTY_FUNCTION__)); return cast_or_null <Value>( OperandTraits<BinaryOperator>::op_begin( const_cast<BinaryOperator*>(this))[i_nocapture].get()); } void BinaryOperator::setOperand(unsigned i_nocapture, Value *Val_nocapture) { (static_cast <bool> (i_nocapture < OperandTraits<BinaryOperator>::operands(this) && "setOperand() out of range!") ? void (0) : __assert_fail ("i_nocapture < OperandTraits<BinaryOperator>::operands(this) && \"setOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 418, __extension__ __PRETTY_FUNCTION__)); OperandTraits< BinaryOperator>::op_begin(this)[i_nocapture] = Val_nocapture ; } unsigned BinaryOperator::getNumOperands() const { return OperandTraits <BinaryOperator>::operands(this); } template <int Idx_nocapture > Use &BinaryOperator::Op() { return this->OpFrom< Idx_nocapture>(this); } template <int Idx_nocapture> const Use &BinaryOperator::Op() const { return this-> OpFrom<Idx_nocapture>(this); } |
419 | |
420 | //===----------------------------------------------------------------------===// |
421 | // CastInst Class |
422 | //===----------------------------------------------------------------------===// |
423 | |
424 | /// This is the base class for all instructions that perform data |
425 | /// casts. It is simply provided so that instruction category testing |
426 | /// can be performed with code like: |
427 | /// |
428 | /// if (isa<CastInst>(Instr)) { ... } |
429 | /// Base class of casting instructions. |
430 | class CastInst : public UnaryInstruction { |
431 | protected: |
432 | /// Constructor with insert-before-instruction semantics for subclasses |
433 | CastInst(Type *Ty, unsigned iType, Value *S, |
434 | const Twine &NameStr = "", Instruction *InsertBefore = nullptr) |
435 | : UnaryInstruction(Ty, iType, S, InsertBefore) { |
436 | setName(NameStr); |
437 | } |
438 | /// Constructor with insert-at-end-of-block semantics for subclasses |
439 | CastInst(Type *Ty, unsigned iType, Value *S, |
440 | const Twine &NameStr, BasicBlock *InsertAtEnd) |
441 | : UnaryInstruction(Ty, iType, S, InsertAtEnd) { |
442 | setName(NameStr); |
443 | } |
444 | |
445 | public: |
446 | /// Provides a way to construct any of the CastInst subclasses using an |
447 | /// opcode instead of the subclass's constructor. The opcode must be in the |
448 | /// CastOps category (Instruction::isCast(opcode) returns true). This |
449 | /// constructor has insert-before-instruction semantics to automatically |
450 | /// insert the new CastInst before InsertBefore (if it is non-null). |
451 | /// Construct any of the CastInst subclasses |
452 | static CastInst *Create( |
453 | Instruction::CastOps, ///< The opcode of the cast instruction |
454 | Value *S, ///< The value to be casted (operand 0) |
455 | Type *Ty, ///< The type to which cast should be made |
456 | const Twine &Name = "", ///< Name for the instruction |
457 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
458 | ); |
459 | /// Provides a way to construct any of the CastInst subclasses using an |
460 | /// opcode instead of the subclass's constructor. The opcode must be in the |
461 | /// CastOps category. This constructor has insert-at-end-of-block semantics |
462 | /// to automatically insert the new CastInst at the end of InsertAtEnd (if |
463 | /// its non-null). |
464 | /// Construct any of the CastInst subclasses |
465 | static CastInst *Create( |
466 | Instruction::CastOps, ///< The opcode for the cast instruction |
467 | Value *S, ///< The value to be casted (operand 0) |
468 | Type *Ty, ///< The type to which operand is casted |
469 | const Twine &Name, ///< The name for the instruction |
470 | BasicBlock *InsertAtEnd ///< The block to insert the instruction into |
471 | ); |
472 | |
473 | /// Create a ZExt or BitCast cast instruction |
474 | static CastInst *CreateZExtOrBitCast( |
475 | Value *S, ///< The value to be casted (operand 0) |
476 | Type *Ty, ///< The type to which cast should be made |
477 | const Twine &Name = "", ///< Name for the instruction |
478 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
479 | ); |
480 | |
481 | /// Create a ZExt or BitCast cast instruction |
482 | static CastInst *CreateZExtOrBitCast( |
483 | Value *S, ///< The value to be casted (operand 0) |
484 | Type *Ty, ///< The type to which operand is casted |
485 | const Twine &Name, ///< The name for the instruction |
486 | BasicBlock *InsertAtEnd ///< The block to insert the instruction into |
487 | ); |
488 | |
489 | /// Create a SExt or BitCast cast instruction |
490 | static CastInst *CreateSExtOrBitCast( |
491 | Value *S, ///< The value to be casted (operand 0) |
492 | Type *Ty, ///< The type to which cast should be made |
493 | const Twine &Name = "", ///< Name for the instruction |
494 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
495 | ); |
496 | |
497 | /// Create a SExt or BitCast cast instruction |
498 | static CastInst *CreateSExtOrBitCast( |
499 | Value *S, ///< The value to be casted (operand 0) |
500 | Type *Ty, ///< The type to which operand is casted |
501 | const Twine &Name, ///< The name for the instruction |
502 | BasicBlock *InsertAtEnd ///< The block to insert the instruction into |
503 | ); |
504 | |
505 | /// Create a BitCast AddrSpaceCast, or a PtrToInt cast instruction. |
506 | static CastInst *CreatePointerCast( |
507 | Value *S, ///< The pointer value to be casted (operand 0) |
508 | Type *Ty, ///< The type to which operand is casted |
509 | const Twine &Name, ///< The name for the instruction |
510 | BasicBlock *InsertAtEnd ///< The block to insert the instruction into |
511 | ); |
512 | |
513 | /// Create a BitCast, AddrSpaceCast or a PtrToInt cast instruction. |
514 | static CastInst *CreatePointerCast( |
515 | Value *S, ///< The pointer value to be casted (operand 0) |
516 | Type *Ty, ///< The type to which cast should be made |
517 | const Twine &Name = "", ///< Name for the instruction |
518 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
519 | ); |
520 | |
521 | /// Create a BitCast or an AddrSpaceCast cast instruction. |
522 | static CastInst *CreatePointerBitCastOrAddrSpaceCast( |
523 | Value *S, ///< The pointer value to be casted (operand 0) |
524 | Type *Ty, ///< The type to which operand is casted |
525 | const Twine &Name, ///< The name for the instruction |
526 | BasicBlock *InsertAtEnd ///< The block to insert the instruction into |
527 | ); |
528 | |
529 | /// Create a BitCast or an AddrSpaceCast cast instruction. |
530 | static CastInst *CreatePointerBitCastOrAddrSpaceCast( |
531 | Value *S, ///< The pointer value to be casted (operand 0) |
532 | Type *Ty, ///< The type to which cast should be made |
533 | const Twine &Name = "", ///< Name for the instruction |
534 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
535 | ); |
536 | |
537 | /// Create a BitCast, a PtrToInt, or an IntToPTr cast instruction. |
538 | /// |
539 | /// If the value is a pointer type and the destination an integer type, |
540 | /// creates a PtrToInt cast. If the value is an integer type and the |
541 | /// destination a pointer type, creates an IntToPtr cast. Otherwise, creates |
542 | /// a bitcast. |
543 | static CastInst *CreateBitOrPointerCast( |
544 | Value *S, ///< The pointer value to be casted (operand 0) |
545 | Type *Ty, ///< The type to which cast should be made |
546 | const Twine &Name = "", ///< Name for the instruction |
547 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
548 | ); |
549 | |
550 | /// Create a ZExt, BitCast, or Trunc for int -> int casts. |
551 | static CastInst *CreateIntegerCast( |
552 | Value *S, ///< The pointer value to be casted (operand 0) |
553 | Type *Ty, ///< The type to which cast should be made |
554 | bool isSigned, ///< Whether to regard S as signed or not |
555 | const Twine &Name = "", ///< Name for the instruction |
556 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
557 | ); |
558 | |
559 | /// Create a ZExt, BitCast, or Trunc for int -> int casts. |
560 | static CastInst *CreateIntegerCast( |
561 | Value *S, ///< The integer value to be casted (operand 0) |
562 | Type *Ty, ///< The integer type to which operand is casted |
563 | bool isSigned, ///< Whether to regard S as signed or not |
564 | const Twine &Name, ///< The name for the instruction |
565 | BasicBlock *InsertAtEnd ///< The block to insert the instruction into |
566 | ); |
567 | |
568 | /// Create an FPExt, BitCast, or FPTrunc for fp -> fp casts |
569 | static CastInst *CreateFPCast( |
570 | Value *S, ///< The floating point value to be casted |
571 | Type *Ty, ///< The floating point type to cast to |
572 | const Twine &Name = "", ///< Name for the instruction |
573 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
574 | ); |
575 | |
576 | /// Create an FPExt, BitCast, or FPTrunc for fp -> fp casts |
577 | static CastInst *CreateFPCast( |
578 | Value *S, ///< The floating point value to be casted |
579 | Type *Ty, ///< The floating point type to cast to |
580 | const Twine &Name, ///< The name for the instruction |
581 | BasicBlock *InsertAtEnd ///< The block to insert the instruction into |
582 | ); |
583 | |
584 | /// Create a Trunc or BitCast cast instruction |
585 | static CastInst *CreateTruncOrBitCast( |
586 | Value *S, ///< The value to be casted (operand 0) |
587 | Type *Ty, ///< The type to which cast should be made |
588 | const Twine &Name = "", ///< Name for the instruction |
589 | Instruction *InsertBefore = nullptr ///< Place to insert the instruction |
590 | ); |
591 | |
592 | /// Create a Trunc or BitCast cast instruction |
593 | static CastInst *CreateTruncOrBitCast( |
594 | Value *S, ///< The value to be casted (operand 0) |
595 | Type *Ty, ///< The type to which operand is casted |
596 | const Twine &Name, ///< The name for the instruction |
597 | BasicBlock *InsertAtEnd ///< The block to insert the instruction into |
598 | ); |
599 | |
600 | /// Check whether a bitcast between these types is valid |
601 | static bool isBitCastable( |
602 | Type *SrcTy, ///< The Type from which the value should be cast. |
603 | Type *DestTy ///< The Type to which the value should be cast. |
604 | ); |
605 | |
606 | /// Check whether a bitcast, inttoptr, or ptrtoint cast between these |
607 | /// types is valid and a no-op. |
608 | /// |
609 | /// This ensures that any pointer<->integer cast has enough bits in the |
610 | /// integer and any other cast is a bitcast. |
611 | static bool isBitOrNoopPointerCastable( |
612 | Type *SrcTy, ///< The Type from which the value should be cast. |
613 | Type *DestTy, ///< The Type to which the value should be cast. |
614 | const DataLayout &DL); |
615 | |
616 | /// Returns the opcode necessary to cast Val into Ty using usual casting |
617 | /// rules. |
618 | /// Infer the opcode for cast operand and type |
619 | static Instruction::CastOps getCastOpcode( |
620 | const Value *Val, ///< The value to cast |
621 | bool SrcIsSigned, ///< Whether to treat the source as signed |
622 | Type *Ty, ///< The Type to which the value should be casted |
623 | bool DstIsSigned ///< Whether to treate the dest. as signed |
624 | ); |
625 | |
626 | /// There are several places where we need to know if a cast instruction |
627 | /// only deals with integer source and destination types. To simplify that |
628 | /// logic, this method is provided. |
629 | /// @returns true iff the cast has only integral typed operand and dest type. |
630 | /// Determine if this is an integer-only cast. |
631 | bool isIntegerCast() const; |
632 | |
633 | /// A lossless cast is one that does not alter the basic value. It implies |
634 | /// a no-op cast but is more stringent, preventing things like int->float, |
635 | /// long->double, or int->ptr. |
636 | /// @returns true iff the cast is lossless. |
637 | /// Determine if this is a lossless cast. |
638 | bool isLosslessCast() const; |
639 | |
640 | /// A no-op cast is one that can be effected without changing any bits. |
641 | /// It implies that the source and destination types are the same size. The |
642 | /// DataLayout argument is to determine the pointer size when examining casts |
643 | /// involving Integer and Pointer types. They are no-op casts if the integer |
644 | /// is the same size as the pointer. However, pointer size varies with |
645 | /// platform. Note that a precondition of this method is that the cast is |
646 | /// legal - i.e. the instruction formed with these operands would verify. |
647 | static bool isNoopCast( |
648 | Instruction::CastOps Opcode, ///< Opcode of cast |
649 | Type *SrcTy, ///< SrcTy of cast |
650 | Type *DstTy, ///< DstTy of cast |
651 | const DataLayout &DL ///< DataLayout to get the Int Ptr type from. |
652 | ); |
653 | |
654 | /// Determine if this cast is a no-op cast. |
655 | /// |
656 | /// \param DL is the DataLayout to determine pointer size. |
657 | bool isNoopCast(const DataLayout &DL) const; |
658 | |
659 | /// Determine how a pair of casts can be eliminated, if they can be at all. |
660 | /// This is a helper function for both CastInst and ConstantExpr. |
661 | /// @returns 0 if the CastInst pair can't be eliminated, otherwise |
662 | /// returns Instruction::CastOps value for a cast that can replace |
663 | /// the pair, casting SrcTy to DstTy. |
664 | /// Determine if a cast pair is eliminable |
665 | static unsigned isEliminableCastPair( |
666 | Instruction::CastOps firstOpcode, ///< Opcode of first cast |
667 | Instruction::CastOps secondOpcode, ///< Opcode of second cast |
668 | Type *SrcTy, ///< SrcTy of 1st cast |
669 | Type *MidTy, ///< DstTy of 1st cast & SrcTy of 2nd cast |
670 | Type *DstTy, ///< DstTy of 2nd cast |
671 | Type *SrcIntPtrTy, ///< Integer type corresponding to Ptr SrcTy, or null |
672 | Type *MidIntPtrTy, ///< Integer type corresponding to Ptr MidTy, or null |
673 | Type *DstIntPtrTy ///< Integer type corresponding to Ptr DstTy, or null |
674 | ); |
675 | |
676 | /// Return the opcode of this CastInst |
677 | Instruction::CastOps getOpcode() const { |
678 | return Instruction::CastOps(Instruction::getOpcode()); |
679 | } |
680 | |
681 | /// Return the source type, as a convenience |
682 | Type* getSrcTy() const { return getOperand(0)->getType(); } |
683 | /// Return the destination type, as a convenience |
684 | Type* getDestTy() const { return getType(); } |
685 | |
686 | /// This method can be used to determine if a cast from SrcTy to DstTy using |
687 | /// Opcode op is valid or not. |
688 | /// @returns true iff the proposed cast is valid. |
689 | /// Determine if a cast is valid without creating one. |
690 | static bool castIsValid(Instruction::CastOps op, Type *SrcTy, Type *DstTy); |
691 | static bool castIsValid(Instruction::CastOps op, Value *S, Type *DstTy) { |
692 | return castIsValid(op, S->getType(), DstTy); |
693 | } |
694 | |
695 | /// Methods for support type inquiry through isa, cast, and dyn_cast: |
696 | static bool classof(const Instruction *I) { |
697 | return I->isCast(); |
698 | } |
699 | static bool classof(const Value *V) { |
700 | return isa<Instruction>(V) && classof(cast<Instruction>(V)); |
701 | } |
702 | }; |
703 | |
704 | //===----------------------------------------------------------------------===// |
705 | // CmpInst Class |
706 | //===----------------------------------------------------------------------===// |
707 | |
708 | /// This class is the base class for the comparison instructions. |
709 | /// Abstract base class of comparison instructions. |
710 | class CmpInst : public Instruction { |
711 | public: |
712 | /// This enumeration lists the possible predicates for CmpInst subclasses. |
713 | /// Values in the range 0-31 are reserved for FCmpInst, while values in the |
714 | /// range 32-64 are reserved for ICmpInst. This is necessary to ensure the |
715 | /// predicate values are not overlapping between the classes. |
716 | /// |
717 | /// Some passes (e.g. InstCombine) depend on the bit-wise characteristics of |
718 | /// FCMP_* values. Changing the bit patterns requires a potential change to |
719 | /// those passes. |
720 | enum Predicate : unsigned { |
721 | // Opcode U L G E Intuitive operation |
722 | FCMP_FALSE = 0, ///< 0 0 0 0 Always false (always folded) |
723 | FCMP_OEQ = 1, ///< 0 0 0 1 True if ordered and equal |
724 | FCMP_OGT = 2, ///< 0 0 1 0 True if ordered and greater than |
725 | FCMP_OGE = 3, ///< 0 0 1 1 True if ordered and greater than or equal |
726 | FCMP_OLT = 4, ///< 0 1 0 0 True if ordered and less than |
727 | FCMP_OLE = 5, ///< 0 1 0 1 True if ordered and less than or equal |
728 | FCMP_ONE = 6, ///< 0 1 1 0 True if ordered and operands are unequal |
729 | FCMP_ORD = 7, ///< 0 1 1 1 True if ordered (no nans) |
730 | FCMP_UNO = 8, ///< 1 0 0 0 True if unordered: isnan(X) | isnan(Y) |
731 | FCMP_UEQ = 9, ///< 1 0 0 1 True if unordered or equal |
732 | FCMP_UGT = 10, ///< 1 0 1 0 True if unordered or greater than |
733 | FCMP_UGE = 11, ///< 1 0 1 1 True if unordered, greater than, or equal |
734 | FCMP_ULT = 12, ///< 1 1 0 0 True if unordered or less than |
735 | FCMP_ULE = 13, ///< 1 1 0 1 True if unordered, less than, or equal |
736 | FCMP_UNE = 14, ///< 1 1 1 0 True if unordered or not equal |
737 | FCMP_TRUE = 15, ///< 1 1 1 1 Always true (always folded) |
738 | FIRST_FCMP_PREDICATE = FCMP_FALSE, |
739 | LAST_FCMP_PREDICATE = FCMP_TRUE, |
740 | BAD_FCMP_PREDICATE = FCMP_TRUE + 1, |
741 | ICMP_EQ = 32, ///< equal |
742 | ICMP_NE = 33, ///< not equal |
743 | ICMP_UGT = 34, ///< unsigned greater than |
744 | ICMP_UGE = 35, ///< unsigned greater or equal |
745 | ICMP_ULT = 36, ///< unsigned less than |
746 | ICMP_ULE = 37, ///< unsigned less or equal |
747 | ICMP_SGT = 38, ///< signed greater than |
748 | ICMP_SGE = 39, ///< signed greater or equal |
749 | ICMP_SLT = 40, ///< signed less than |
750 | ICMP_SLE = 41, ///< signed less or equal |
751 | FIRST_ICMP_PREDICATE = ICMP_EQ, |
752 | LAST_ICMP_PREDICATE = ICMP_SLE, |
753 | BAD_ICMP_PREDICATE = ICMP_SLE + 1 |
754 | }; |
755 | using PredicateField = |
756 | Bitfield::Element<Predicate, 0, 6, LAST_ICMP_PREDICATE>; |
757 | |
758 | protected: |
759 | CmpInst(Type *ty, Instruction::OtherOps op, Predicate pred, |
760 | Value *LHS, Value *RHS, const Twine &Name = "", |
761 | Instruction *InsertBefore = nullptr, |
762 | Instruction *FlagsSource = nullptr); |
763 | |
764 | CmpInst(Type *ty, Instruction::OtherOps op, Predicate pred, |
765 | Value *LHS, Value *RHS, const Twine &Name, |
766 | BasicBlock *InsertAtEnd); |
767 | |
768 | public: |
769 | // allocate space for exactly two operands |
770 | void *operator new(size_t S) { return User::operator new(S, 2); } |
771 | void operator delete(void *Ptr) { User::operator delete(Ptr); } |
772 | |
773 | /// Construct a compare instruction, given the opcode, the predicate and |
774 | /// the two operands. Optionally (if InstBefore is specified) insert the |
775 | /// instruction into a BasicBlock right before the specified instruction. |
776 | /// The specified Instruction is allowed to be a dereferenced end iterator. |
777 | /// Create a CmpInst |
778 | static CmpInst *Create(OtherOps Op, |
779 | Predicate predicate, Value *S1, |
780 | Value *S2, const Twine &Name = "", |
781 | Instruction *InsertBefore = nullptr); |
782 | |
783 | /// Construct a compare instruction, given the opcode, the predicate and the |
784 | /// two operands. Also automatically insert this instruction to the end of |
785 | /// the BasicBlock specified. |
786 | /// Create a CmpInst |
787 | static CmpInst *Create(OtherOps Op, Predicate predicate, Value *S1, |
788 | Value *S2, const Twine &Name, BasicBlock *InsertAtEnd); |
789 | |
790 | /// Get the opcode casted to the right type |
791 | OtherOps getOpcode() const { |
792 | return static_cast<OtherOps>(Instruction::getOpcode()); |
793 | } |
794 | |
795 | /// Return the predicate for this instruction. |
796 | Predicate getPredicate() const { return getSubclassData<PredicateField>(); } |
797 | |
798 | /// Set the predicate for this instruction to the specified value. |
799 | void setPredicate(Predicate P) { setSubclassData<PredicateField>(P); } |
800 | |
801 | static bool isFPPredicate(Predicate P) { |
802 | static_assert(FIRST_FCMP_PREDICATE == 0, |
803 | "FIRST_FCMP_PREDICATE is required to be 0"); |
804 | return P <= LAST_FCMP_PREDICATE; |
805 | } |
806 | |
807 | static bool isIntPredicate(Predicate P) { |
808 | return P >= FIRST_ICMP_PREDICATE && P <= LAST_ICMP_PREDICATE; |
809 | } |
810 | |
811 | static StringRef getPredicateName(Predicate P); |
812 | |
813 | bool isFPPredicate() const { return isFPPredicate(getPredicate()); } |
814 | bool isIntPredicate() const { return isIntPredicate(getPredicate()); } |
815 | |
816 | /// For example, EQ -> NE, UGT -> ULE, SLT -> SGE, |
817 | /// OEQ -> UNE, UGT -> OLE, OLT -> UGE, etc. |
818 | /// @returns the inverse predicate for the instruction's current predicate. |
819 | /// Return the inverse of the instruction's predicate. |
820 | Predicate getInversePredicate() const { |
821 | return getInversePredicate(getPredicate()); |
822 | } |
823 | |
824 | /// For example, EQ -> NE, UGT -> ULE, SLT -> SGE, |
825 | /// OEQ -> UNE, UGT -> OLE, OLT -> UGE, etc. |
826 | /// @returns the inverse predicate for predicate provided in \p pred. |
827 | /// Return the inverse of a given predicate |
828 | static Predicate getInversePredicate(Predicate pred); |
829 | |
830 | /// For example, EQ->EQ, SLE->SGE, ULT->UGT, |
831 | /// OEQ->OEQ, ULE->UGE, OLT->OGT, etc. |
832 | /// @returns the predicate that would be the result of exchanging the two |
833 | /// operands of the CmpInst instruction without changing the result |
834 | /// produced. |
835 | /// Return the predicate as if the operands were swapped |
836 | Predicate getSwappedPredicate() const { |
837 | return getSwappedPredicate(getPredicate()); |
838 | } |
839 | |
840 | /// This is a static version that you can use without an instruction |
841 | /// available. |
842 | /// Return the predicate as if the operands were swapped. |
843 | static Predicate getSwappedPredicate(Predicate pred); |
844 | |
845 | /// This is a static version that you can use without an instruction |
846 | /// available. |
847 | /// @returns true if the comparison predicate is strict, false otherwise. |
848 | static bool isStrictPredicate(Predicate predicate); |
849 | |
850 | /// @returns true if the comparison predicate is strict, false otherwise. |
851 | /// Determine if this instruction is using an strict comparison predicate. |
852 | bool isStrictPredicate() const { return isStrictPredicate(getPredicate()); } |
853 | |
854 | /// This is a static version that you can use without an instruction |
855 | /// available. |
856 | /// @returns true if the comparison predicate is non-strict, false otherwise. |
857 | static bool isNonStrictPredicate(Predicate predicate); |
858 | |
859 | /// @returns true if the comparison predicate is non-strict, false otherwise. |
860 | /// Determine if this instruction is using an non-strict comparison predicate. |
861 | bool isNonStrictPredicate() const { |
862 | return isNonStrictPredicate(getPredicate()); |
863 | } |
864 | |
865 | /// For example, SGE -> SGT, SLE -> SLT, ULE -> ULT, UGE -> UGT. |
866 | /// Returns the strict version of non-strict comparisons. |
867 | Predicate getStrictPredicate() const { |
868 | return getStrictPredicate(getPredicate()); |
869 | } |
870 | |
871 | /// This is a static version that you can use without an instruction |
872 | /// available. |
873 | /// @returns the strict version of comparison provided in \p pred. |
874 | /// If \p pred is not a strict comparison predicate, returns \p pred. |
875 | /// Returns the strict version of non-strict comparisons. |
876 | static Predicate getStrictPredicate(Predicate pred); |
877 | |
878 | /// For example, SGT -> SGE, SLT -> SLE, ULT -> ULE, UGT -> UGE. |
879 | /// Returns the non-strict version of strict comparisons. |
880 | Predicate getNonStrictPredicate() const { |
881 | return getNonStrictPredicate(getPredicate()); |
882 | } |
883 | |
884 | /// This is a static version that you can use without an instruction |
885 | /// available. |
886 | /// @returns the non-strict version of comparison provided in \p pred. |
887 | /// If \p pred is not a strict comparison predicate, returns \p pred. |
888 | /// Returns the non-strict version of strict comparisons. |
889 | static Predicate getNonStrictPredicate(Predicate pred); |
890 | |
891 | /// This is a static version that you can use without an instruction |
892 | /// available. |
893 | /// Return the flipped strictness of predicate |
894 | static Predicate getFlippedStrictnessPredicate(Predicate pred); |
895 | |
896 | /// For predicate of kind "is X or equal to 0" returns the predicate "is X". |
897 | /// For predicate of kind "is X" returns the predicate "is X or equal to 0". |
898 | /// does not support other kind of predicates. |
899 | /// @returns the predicate that does not contains is equal to zero if |
900 | /// it had and vice versa. |
901 | /// Return the flipped strictness of predicate |
902 | Predicate getFlippedStrictnessPredicate() const { |
903 | return getFlippedStrictnessPredicate(getPredicate()); |
904 | } |
905 | |
906 | /// Provide more efficient getOperand methods. |
907 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(Value)public: inline Value *getOperand(unsigned) const; inline void setOperand(unsigned, Value*); inline op_iterator op_begin(); inline const_op_iterator op_begin() const; inline op_iterator op_end(); inline const_op_iterator op_end() const; protected : template <int> inline Use &Op(); template <int > inline const Use &Op() const; public: inline unsigned getNumOperands() const; |
908 | |
909 | /// This is just a convenience that dispatches to the subclasses. |
910 | /// Swap the operands and adjust predicate accordingly to retain |
911 | /// the same comparison. |
912 | void swapOperands(); |
913 | |
914 | /// This is just a convenience that dispatches to the subclasses. |
915 | /// Determine if this CmpInst is commutative. |
916 | bool isCommutative() const; |
917 | |
918 | /// Determine if this is an equals/not equals predicate. |
919 | /// This is a static version that you can use without an instruction |
920 | /// available. |
921 | static bool isEquality(Predicate pred); |
922 | |
923 | /// Determine if this is an equals/not equals predicate. |
924 | bool isEquality() const { return isEquality(getPredicate()); } |
925 | |
926 | /// Return true if the predicate is relational (not EQ or NE). |
927 | static bool isRelational(Predicate P) { return !isEquality(P); } |
928 | |
929 | /// Return true if the predicate is relational (not EQ or NE). |
930 | bool isRelational() const { return !isEquality(); } |
931 | |
932 | /// @returns true if the comparison is signed, false otherwise. |
933 | /// Determine if this instruction is using a signed comparison. |
934 | bool isSigned() const { |
935 | return isSigned(getPredicate()); |
936 | } |
937 | |
938 | /// @returns true if the comparison is unsigned, false otherwise. |
939 | /// Determine if this instruction is using an unsigned comparison. |
940 | bool isUnsigned() const { |
941 | return isUnsigned(getPredicate()); |
942 | } |
943 | |
944 | /// For example, ULT->SLT, ULE->SLE, UGT->SGT, UGE->SGE, SLT->Failed assert |
945 | /// @returns the signed version of the unsigned predicate pred. |
946 | /// return the signed version of a predicate |
947 | static Predicate getSignedPredicate(Predicate pred); |
948 | |
949 | /// For example, ULT->SLT, ULE->SLE, UGT->SGT, UGE->SGE, SLT->Failed assert |
950 | /// @returns the signed version of the predicate for this instruction (which |
951 | /// has to be an unsigned predicate). |
952 | /// return the signed version of a predicate |
953 | Predicate getSignedPredicate() { |
954 | return getSignedPredicate(getPredicate()); |
955 | } |
956 | |
957 | /// For example, SLT->ULT, SLE->ULE, SGT->UGT, SGE->UGE, ULT->Failed assert |
958 | /// @returns the unsigned version of the signed predicate pred. |
959 | static Predicate getUnsignedPredicate(Predicate pred); |
960 | |
961 | /// For example, SLT->ULT, SLE->ULE, SGT->UGT, SGE->UGE, ULT->Failed assert |
962 | /// @returns the unsigned version of the predicate for this instruction (which |
963 | /// has to be an signed predicate). |
964 | /// return the unsigned version of a predicate |
965 | Predicate getUnsignedPredicate() { |
966 | return getUnsignedPredicate(getPredicate()); |
967 | } |
968 | |
969 | /// For example, SLT->ULT, ULT->SLT, SLE->ULE, ULE->SLE, EQ->Failed assert |
970 | /// @returns the unsigned version of the signed predicate pred or |
971 | /// the signed version of the signed predicate pred. |
972 | static Predicate getFlippedSignednessPredicate(Predicate pred); |
973 | |
974 | /// For example, SLT->ULT, ULT->SLT, SLE->ULE, ULE->SLE, EQ->Failed assert |
975 | /// @returns the unsigned version of the signed predicate pred or |
976 | /// the signed version of the signed predicate pred. |
977 | Predicate getFlippedSignednessPredicate() { |
978 | return getFlippedSignednessPredicate(getPredicate()); |
979 | } |
980 | |
981 | /// This is just a convenience. |
982 | /// Determine if this is true when both operands are the same. |
983 | bool isTrueWhenEqual() const { |
984 | return isTrueWhenEqual(getPredicate()); |
985 | } |
986 | |
987 | /// This is just a convenience. |
988 | /// Determine if this is false when both operands are the same. |
989 | bool isFalseWhenEqual() const { |
990 | return isFalseWhenEqual(getPredicate()); |
991 | } |
992 | |
993 | /// @returns true if the predicate is unsigned, false otherwise. |
994 | /// Determine if the predicate is an unsigned operation. |
995 | static bool isUnsigned(Predicate predicate); |
996 | |
997 | /// @returns true if the predicate is signed, false otherwise. |
998 | /// Determine if the predicate is an signed operation. |
999 | static bool isSigned(Predicate predicate); |
1000 | |
1001 | /// Determine if the predicate is an ordered operation. |
1002 | static bool isOrdered(Predicate predicate); |
1003 | |
1004 | /// Determine if the predicate is an unordered operation. |
1005 | static bool isUnordered(Predicate predicate); |
1006 | |
1007 | /// Determine if the predicate is true when comparing a value with itself. |
1008 | static bool isTrueWhenEqual(Predicate predicate); |
1009 | |
1010 | /// Determine if the predicate is false when comparing a value with itself. |
1011 | static bool isFalseWhenEqual(Predicate predicate); |
1012 | |
1013 | /// Determine if Pred1 implies Pred2 is true when two compares have matching |
1014 | /// operands. |
1015 | static bool isImpliedTrueByMatchingCmp(Predicate Pred1, Predicate Pred2); |
1016 | |
1017 | /// Determine if Pred1 implies Pred2 is false when two compares have matching |
1018 | /// operands. |
1019 | static bool isImpliedFalseByMatchingCmp(Predicate Pred1, Predicate Pred2); |
1020 | |
1021 | /// Methods for support type inquiry through isa, cast, and dyn_cast: |
1022 | static bool classof(const Instruction *I) { |
1023 | return I->getOpcode() == Instruction::ICmp || |
1024 | I->getOpcode() == Instruction::FCmp; |
1025 | } |
1026 | static bool classof(const Value *V) { |
1027 | return isa<Instruction>(V) && classof(cast<Instruction>(V)); |
1028 | } |
1029 | |
1030 | /// Create a result type for fcmp/icmp |
1031 | static Type* makeCmpResultType(Type* opnd_type) { |
1032 | if (VectorType* vt = dyn_cast<VectorType>(opnd_type)) { |
1033 | return VectorType::get(Type::getInt1Ty(opnd_type->getContext()), |
1034 | vt->getElementCount()); |
1035 | } |
1036 | return Type::getInt1Ty(opnd_type->getContext()); |
1037 | } |
1038 | |
1039 | private: |
1040 | // Shadow Value::setValueSubclassData with a private forwarding method so that |
1041 | // subclasses cannot accidentally use it. |
1042 | void setValueSubclassData(unsigned short D) { |
1043 | Value::setValueSubclassData(D); |
1044 | } |
1045 | }; |
1046 | |
1047 | // FIXME: these are redundant if CmpInst < BinaryOperator |
1048 | template <> |
1049 | struct OperandTraits<CmpInst> : public FixedNumOperandTraits<CmpInst, 2> { |
1050 | }; |
1051 | |
1052 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(CmpInst, Value)CmpInst::op_iterator CmpInst::op_begin() { return OperandTraits <CmpInst>::op_begin(this); } CmpInst::const_op_iterator CmpInst::op_begin() const { return OperandTraits<CmpInst> ::op_begin(const_cast<CmpInst*>(this)); } CmpInst::op_iterator CmpInst::op_end() { return OperandTraits<CmpInst>::op_end (this); } CmpInst::const_op_iterator CmpInst::op_end() const { return OperandTraits<CmpInst>::op_end(const_cast<CmpInst *>(this)); } Value *CmpInst::getOperand(unsigned i_nocapture ) const { (static_cast <bool> (i_nocapture < OperandTraits <CmpInst>::operands(this) && "getOperand() out of range!" ) ? void (0) : __assert_fail ("i_nocapture < OperandTraits<CmpInst>::operands(this) && \"getOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1052, __extension__ __PRETTY_FUNCTION__)); return cast_or_null <Value>( OperandTraits<CmpInst>::op_begin(const_cast <CmpInst*>(this))[i_nocapture].get()); } void CmpInst:: setOperand(unsigned i_nocapture, Value *Val_nocapture) { (static_cast <bool> (i_nocapture < OperandTraits<CmpInst>:: operands(this) && "setOperand() out of range!") ? void (0) : __assert_fail ("i_nocapture < OperandTraits<CmpInst>::operands(this) && \"setOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1052, __extension__ __PRETTY_FUNCTION__)); OperandTraits< CmpInst>::op_begin(this)[i_nocapture] = Val_nocapture; } unsigned CmpInst::getNumOperands() const { return OperandTraits<CmpInst >::operands(this); } template <int Idx_nocapture> Use &CmpInst::Op() { return this->OpFrom<Idx_nocapture >(this); } template <int Idx_nocapture> const Use & CmpInst::Op() const { return this->OpFrom<Idx_nocapture >(this); } |
1053 | |
1054 | /// A lightweight accessor for an operand bundle meant to be passed |
1055 | /// around by value. |
1056 | struct OperandBundleUse { |
1057 | ArrayRef<Use> Inputs; |
1058 | |
1059 | OperandBundleUse() = default; |
1060 | explicit OperandBundleUse(StringMapEntry<uint32_t> *Tag, ArrayRef<Use> Inputs) |
1061 | : Inputs(Inputs), Tag(Tag) {} |
1062 | |
1063 | /// Return true if the operand at index \p Idx in this operand bundle |
1064 | /// has the attribute A. |
1065 | bool operandHasAttr(unsigned Idx, Attribute::AttrKind A) const { |
1066 | if (isDeoptOperandBundle()) |
1067 | if (A == Attribute::ReadOnly || A == Attribute::NoCapture) |
1068 | return Inputs[Idx]->getType()->isPointerTy(); |
1069 | |
1070 | // Conservative answer: no operands have any attributes. |
1071 | return false; |
1072 | } |
1073 | |
1074 | /// Return the tag of this operand bundle as a string. |
1075 | StringRef getTagName() const { |
1076 | return Tag->getKey(); |
1077 | } |
1078 | |
1079 | /// Return the tag of this operand bundle as an integer. |
1080 | /// |
1081 | /// Operand bundle tags are interned by LLVMContextImpl::getOrInsertBundleTag, |
1082 | /// and this function returns the unique integer getOrInsertBundleTag |
1083 | /// associated the tag of this operand bundle to. |
1084 | uint32_t getTagID() const { |
1085 | return Tag->getValue(); |
1086 | } |
1087 | |
1088 | /// Return true if this is a "deopt" operand bundle. |
1089 | bool isDeoptOperandBundle() const { |
1090 | return getTagID() == LLVMContext::OB_deopt; |
1091 | } |
1092 | |
1093 | /// Return true if this is a "funclet" operand bundle. |
1094 | bool isFuncletOperandBundle() const { |
1095 | return getTagID() == LLVMContext::OB_funclet; |
1096 | } |
1097 | |
1098 | /// Return true if this is a "cfguardtarget" operand bundle. |
1099 | bool isCFGuardTargetOperandBundle() const { |
1100 | return getTagID() == LLVMContext::OB_cfguardtarget; |
1101 | } |
1102 | |
1103 | private: |
1104 | /// Pointer to an entry in LLVMContextImpl::getOrInsertBundleTag. |
1105 | StringMapEntry<uint32_t> *Tag; |
1106 | }; |
1107 | |
1108 | /// A container for an operand bundle being viewed as a set of values |
1109 | /// rather than a set of uses. |
1110 | /// |
1111 | /// Unlike OperandBundleUse, OperandBundleDefT owns the memory it carries, and |
1112 | /// so it is possible to create and pass around "self-contained" instances of |
1113 | /// OperandBundleDef and ConstOperandBundleDef. |
1114 | template <typename InputTy> class OperandBundleDefT { |
1115 | std::string Tag; |
1116 | std::vector<InputTy> Inputs; |
1117 | |
1118 | public: |
1119 | explicit OperandBundleDefT(std::string Tag, std::vector<InputTy> Inputs) |
1120 | : Tag(std::move(Tag)), Inputs(std::move(Inputs)) {} |
1121 | explicit OperandBundleDefT(std::string Tag, ArrayRef<InputTy> Inputs) |
1122 | : Tag(std::move(Tag)), Inputs(Inputs) {} |
1123 | |
1124 | explicit OperandBundleDefT(const OperandBundleUse &OBU) { |
1125 | Tag = std::string(OBU.getTagName()); |
1126 | llvm::append_range(Inputs, OBU.Inputs); |
1127 | } |
1128 | |
1129 | ArrayRef<InputTy> inputs() const { return Inputs; } |
1130 | |
1131 | using input_iterator = typename std::vector<InputTy>::const_iterator; |
1132 | |
1133 | size_t input_size() const { return Inputs.size(); } |
1134 | input_iterator input_begin() const { return Inputs.begin(); } |
1135 | input_iterator input_end() const { return Inputs.end(); } |
1136 | |
1137 | StringRef getTag() const { return Tag; } |
1138 | }; |
1139 | |
1140 | using OperandBundleDef = OperandBundleDefT<Value *>; |
1141 | using ConstOperandBundleDef = OperandBundleDefT<const Value *>; |
1142 | |
1143 | //===----------------------------------------------------------------------===// |
1144 | // CallBase Class |
1145 | //===----------------------------------------------------------------------===// |
1146 | |
1147 | /// Base class for all callable instructions (InvokeInst and CallInst) |
1148 | /// Holds everything related to calling a function. |
1149 | /// |
1150 | /// All call-like instructions are required to use a common operand layout: |
1151 | /// - Zero or more arguments to the call, |
1152 | /// - Zero or more operand bundles with zero or more operand inputs each |
1153 | /// bundle, |
1154 | /// - Zero or more subclass controlled operands |
1155 | /// - The called function. |
1156 | /// |
1157 | /// This allows this base class to easily access the called function and the |
1158 | /// start of the arguments without knowing how many other operands a particular |
1159 | /// subclass requires. Note that accessing the end of the argument list isn't |
1160 | /// as cheap as most other operations on the base class. |
1161 | class CallBase : public Instruction { |
1162 | protected: |
1163 | // The first two bits are reserved by CallInst for fast retrieval, |
1164 | using CallInstReservedField = Bitfield::Element<unsigned, 0, 2>; |
1165 | using CallingConvField = |
1166 | Bitfield::Element<CallingConv::ID, CallInstReservedField::NextBit, 10, |
1167 | CallingConv::MaxID>; |
1168 | static_assert( |
1169 | Bitfield::areContiguous<CallInstReservedField, CallingConvField>(), |
1170 | "Bitfields must be contiguous"); |
1171 | |
1172 | /// The last operand is the called operand. |
1173 | static constexpr int CalledOperandOpEndIdx = -1; |
1174 | |
1175 | AttributeList Attrs; ///< parameter attributes for callable |
1176 | FunctionType *FTy; |
1177 | |
1178 | template <class... ArgsTy> |
1179 | CallBase(AttributeList const &A, FunctionType *FT, ArgsTy &&... Args) |
1180 | : Instruction(std::forward<ArgsTy>(Args)...), Attrs(A), FTy(FT) {} |
1181 | |
1182 | using Instruction::Instruction; |
1183 | |
1184 | bool hasDescriptor() const { return Value::HasDescriptor; } |
1185 | |
1186 | unsigned getNumSubclassExtraOperands() const { |
1187 | switch (getOpcode()) { |
1188 | case Instruction::Call: |
1189 | return 0; |
1190 | case Instruction::Invoke: |
1191 | return 2; |
1192 | case Instruction::CallBr: |
1193 | return getNumSubclassExtraOperandsDynamic(); |
1194 | } |
1195 | llvm_unreachable("Invalid opcode!")::llvm::llvm_unreachable_internal("Invalid opcode!", "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1195); |
1196 | } |
1197 | |
1198 | /// Get the number of extra operands for instructions that don't have a fixed |
1199 | /// number of extra operands. |
1200 | unsigned getNumSubclassExtraOperandsDynamic() const; |
1201 | |
1202 | public: |
1203 | using Instruction::getContext; |
1204 | |
1205 | /// Create a clone of \p CB with a different set of operand bundles and |
1206 | /// insert it before \p InsertPt. |
1207 | /// |
1208 | /// The returned call instruction is identical \p CB in every way except that |
1209 | /// the operand bundles for the new instruction are set to the operand bundles |
1210 | /// in \p Bundles. |
1211 | static CallBase *Create(CallBase *CB, ArrayRef<OperandBundleDef> Bundles, |
1212 | Instruction *InsertPt = nullptr); |
1213 | |
1214 | /// Create a clone of \p CB with the operand bundle with the tag matching |
1215 | /// \p Bundle's tag replaced with Bundle, and insert it before \p InsertPt. |
1216 | /// |
1217 | /// The returned call instruction is identical \p CI in every way except that |
1218 | /// the specified operand bundle has been replaced. |
1219 | static CallBase *Create(CallBase *CB, |
1220 | OperandBundleDef Bundle, |
1221 | Instruction *InsertPt = nullptr); |
1222 | |
1223 | /// Create a clone of \p CB with operand bundle \p OB added. |
1224 | static CallBase *addOperandBundle(CallBase *CB, uint32_t ID, |
1225 | OperandBundleDef OB, |
1226 | Instruction *InsertPt = nullptr); |
1227 | |
1228 | /// Create a clone of \p CB with operand bundle \p ID removed. |
1229 | static CallBase *removeOperandBundle(CallBase *CB, uint32_t ID, |
1230 | Instruction *InsertPt = nullptr); |
1231 | |
1232 | static bool classof(const Instruction *I) { |
1233 | return I->getOpcode() == Instruction::Call || |
1234 | I->getOpcode() == Instruction::Invoke || |
1235 | I->getOpcode() == Instruction::CallBr; |
1236 | } |
1237 | static bool classof(const Value *V) { |
1238 | return isa<Instruction>(V) && classof(cast<Instruction>(V)); |
1239 | } |
1240 | |
1241 | FunctionType *getFunctionType() const { return FTy; } |
1242 | |
1243 | void mutateFunctionType(FunctionType *FTy) { |
1244 | Value::mutateType(FTy->getReturnType()); |
1245 | this->FTy = FTy; |
1246 | } |
1247 | |
1248 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(Value)public: inline Value *getOperand(unsigned) const; inline void setOperand(unsigned, Value*); inline op_iterator op_begin(); inline const_op_iterator op_begin() const; inline op_iterator op_end(); inline const_op_iterator op_end() const; protected : template <int> inline Use &Op(); template <int > inline const Use &Op() const; public: inline unsigned getNumOperands() const; |
1249 | |
1250 | /// data_operands_begin/data_operands_end - Return iterators iterating over |
1251 | /// the call / invoke argument list and bundle operands. For invokes, this is |
1252 | /// the set of instruction operands except the invoke target and the two |
1253 | /// successor blocks; and for calls this is the set of instruction operands |
1254 | /// except the call target. |
1255 | User::op_iterator data_operands_begin() { return op_begin(); } |
1256 | User::const_op_iterator data_operands_begin() const { |
1257 | return const_cast<CallBase *>(this)->data_operands_begin(); |
1258 | } |
1259 | User::op_iterator data_operands_end() { |
1260 | // Walk from the end of the operands over the called operand and any |
1261 | // subclass operands. |
1262 | return op_end() - getNumSubclassExtraOperands() - 1; |
1263 | } |
1264 | User::const_op_iterator data_operands_end() const { |
1265 | return const_cast<CallBase *>(this)->data_operands_end(); |
1266 | } |
1267 | iterator_range<User::op_iterator> data_ops() { |
1268 | return make_range(data_operands_begin(), data_operands_end()); |
1269 | } |
1270 | iterator_range<User::const_op_iterator> data_ops() const { |
1271 | return make_range(data_operands_begin(), data_operands_end()); |
1272 | } |
1273 | bool data_operands_empty() const { |
1274 | return data_operands_end() == data_operands_begin(); |
1275 | } |
1276 | unsigned data_operands_size() const { |
1277 | return std::distance(data_operands_begin(), data_operands_end()); |
1278 | } |
1279 | |
1280 | bool isDataOperand(const Use *U) const { |
1281 | assert(this == U->getUser() &&(static_cast <bool> (this == U->getUser() && "Only valid to query with a use of this instruction!") ? void (0) : __assert_fail ("this == U->getUser() && \"Only valid to query with a use of this instruction!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1282, __extension__ __PRETTY_FUNCTION__)) |
1282 | "Only valid to query with a use of this instruction!")(static_cast <bool> (this == U->getUser() && "Only valid to query with a use of this instruction!") ? void (0) : __assert_fail ("this == U->getUser() && \"Only valid to query with a use of this instruction!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1282, __extension__ __PRETTY_FUNCTION__)); |
1283 | return data_operands_begin() <= U && U < data_operands_end(); |
1284 | } |
1285 | bool isDataOperand(Value::const_user_iterator UI) const { |
1286 | return isDataOperand(&UI.getUse()); |
1287 | } |
1288 | |
1289 | /// Given a value use iterator, return the data operand corresponding to it. |
1290 | /// Iterator must actually correspond to a data operand. |
1291 | unsigned getDataOperandNo(Value::const_user_iterator UI) const { |
1292 | return getDataOperandNo(&UI.getUse()); |
1293 | } |
1294 | |
1295 | /// Given a use for a data operand, get the data operand number that |
1296 | /// corresponds to it. |
1297 | unsigned getDataOperandNo(const Use *U) const { |
1298 | assert(isDataOperand(U) && "Data operand # out of range!")(static_cast <bool> (isDataOperand(U) && "Data operand # out of range!" ) ? void (0) : __assert_fail ("isDataOperand(U) && \"Data operand # out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1298, __extension__ __PRETTY_FUNCTION__)); |
1299 | return U - data_operands_begin(); |
1300 | } |
1301 | |
1302 | /// Return the iterator pointing to the beginning of the argument list. |
1303 | User::op_iterator arg_begin() { return op_begin(); } |
1304 | User::const_op_iterator arg_begin() const { |
1305 | return const_cast<CallBase *>(this)->arg_begin(); |
1306 | } |
1307 | |
1308 | /// Return the iterator pointing to the end of the argument list. |
1309 | User::op_iterator arg_end() { |
1310 | // From the end of the data operands, walk backwards past the bundle |
1311 | // operands. |
1312 | return data_operands_end() - getNumTotalBundleOperands(); |
1313 | } |
1314 | User::const_op_iterator arg_end() const { |
1315 | return const_cast<CallBase *>(this)->arg_end(); |
1316 | } |
1317 | |
1318 | /// Iteration adapter for range-for loops. |
1319 | iterator_range<User::op_iterator> args() { |
1320 | return make_range(arg_begin(), arg_end()); |
1321 | } |
1322 | iterator_range<User::const_op_iterator> args() const { |
1323 | return make_range(arg_begin(), arg_end()); |
1324 | } |
1325 | bool arg_empty() const { return arg_end() == arg_begin(); } |
1326 | unsigned arg_size() const { return arg_end() - arg_begin(); } |
1327 | |
1328 | // Legacy API names that duplicate the above and will be removed once users |
1329 | // are migrated. |
1330 | iterator_range<User::op_iterator> arg_operands() { |
1331 | return make_range(arg_begin(), arg_end()); |
1332 | } |
1333 | iterator_range<User::const_op_iterator> arg_operands() const { |
1334 | return make_range(arg_begin(), arg_end()); |
1335 | } |
1336 | unsigned getNumArgOperands() const { return arg_size(); } |
1337 | |
1338 | Value *getArgOperand(unsigned i) const { |
1339 | assert(i < getNumArgOperands() && "Out of bounds!")(static_cast <bool> (i < getNumArgOperands() && "Out of bounds!") ? void (0) : __assert_fail ("i < getNumArgOperands() && \"Out of bounds!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1339, __extension__ __PRETTY_FUNCTION__)); |
1340 | return getOperand(i); |
1341 | } |
1342 | |
1343 | void setArgOperand(unsigned i, Value *v) { |
1344 | assert(i < getNumArgOperands() && "Out of bounds!")(static_cast <bool> (i < getNumArgOperands() && "Out of bounds!") ? void (0) : __assert_fail ("i < getNumArgOperands() && \"Out of bounds!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1344, __extension__ __PRETTY_FUNCTION__)); |
1345 | setOperand(i, v); |
1346 | } |
1347 | |
1348 | /// Wrappers for getting the \c Use of a call argument. |
1349 | const Use &getArgOperandUse(unsigned i) const { |
1350 | assert(i < getNumArgOperands() && "Out of bounds!")(static_cast <bool> (i < getNumArgOperands() && "Out of bounds!") ? void (0) : __assert_fail ("i < getNumArgOperands() && \"Out of bounds!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1350, __extension__ __PRETTY_FUNCTION__)); |
1351 | return User::getOperandUse(i); |
1352 | } |
1353 | Use &getArgOperandUse(unsigned i) { |
1354 | assert(i < getNumArgOperands() && "Out of bounds!")(static_cast <bool> (i < getNumArgOperands() && "Out of bounds!") ? void (0) : __assert_fail ("i < getNumArgOperands() && \"Out of bounds!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1354, __extension__ __PRETTY_FUNCTION__)); |
1355 | return User::getOperandUse(i); |
1356 | } |
1357 | |
1358 | bool isArgOperand(const Use *U) const { |
1359 | assert(this == U->getUser() &&(static_cast <bool> (this == U->getUser() && "Only valid to query with a use of this instruction!") ? void (0) : __assert_fail ("this == U->getUser() && \"Only valid to query with a use of this instruction!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1360, __extension__ __PRETTY_FUNCTION__)) |
1360 | "Only valid to query with a use of this instruction!")(static_cast <bool> (this == U->getUser() && "Only valid to query with a use of this instruction!") ? void (0) : __assert_fail ("this == U->getUser() && \"Only valid to query with a use of this instruction!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1360, __extension__ __PRETTY_FUNCTION__)); |
1361 | return arg_begin() <= U && U < arg_end(); |
1362 | } |
1363 | bool isArgOperand(Value::const_user_iterator UI) const { |
1364 | return isArgOperand(&UI.getUse()); |
1365 | } |
1366 | |
1367 | /// Given a use for a arg operand, get the arg operand number that |
1368 | /// corresponds to it. |
1369 | unsigned getArgOperandNo(const Use *U) const { |
1370 | assert(isArgOperand(U) && "Arg operand # out of range!")(static_cast <bool> (isArgOperand(U) && "Arg operand # out of range!" ) ? void (0) : __assert_fail ("isArgOperand(U) && \"Arg operand # out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1370, __extension__ __PRETTY_FUNCTION__)); |
1371 | return U - arg_begin(); |
1372 | } |
1373 | |
1374 | /// Given a value use iterator, return the arg operand number corresponding to |
1375 | /// it. Iterator must actually correspond to a data operand. |
1376 | unsigned getArgOperandNo(Value::const_user_iterator UI) const { |
1377 | return getArgOperandNo(&UI.getUse()); |
1378 | } |
1379 | |
1380 | /// Returns true if this CallSite passes the given Value* as an argument to |
1381 | /// the called function. |
1382 | bool hasArgument(const Value *V) const { |
1383 | return llvm::is_contained(args(), V); |
1384 | } |
1385 | |
1386 | Value *getCalledOperand() const { return Op<CalledOperandOpEndIdx>(); } |
1387 | |
1388 | const Use &getCalledOperandUse() const { return Op<CalledOperandOpEndIdx>(); } |
1389 | Use &getCalledOperandUse() { return Op<CalledOperandOpEndIdx>(); } |
1390 | |
1391 | /// Returns the function called, or null if this is an |
1392 | /// indirect function invocation. |
1393 | Function *getCalledFunction() const { |
1394 | return dyn_cast_or_null<Function>(getCalledOperand()); |
1395 | } |
1396 | |
1397 | /// Return true if the callsite is an indirect call. |
1398 | bool isIndirectCall() const; |
1399 | |
1400 | /// Determine whether the passed iterator points to the callee operand's Use. |
1401 | bool isCallee(Value::const_user_iterator UI) const { |
1402 | return isCallee(&UI.getUse()); |
1403 | } |
1404 | |
1405 | /// Determine whether this Use is the callee operand's Use. |
1406 | bool isCallee(const Use *U) const { return &getCalledOperandUse() == U; } |
1407 | |
1408 | /// Helper to get the caller (the parent function). |
1409 | Function *getCaller(); |
1410 | const Function *getCaller() const { |
1411 | return const_cast<CallBase *>(this)->getCaller(); |
1412 | } |
1413 | |
1414 | /// Tests if this call site must be tail call optimized. Only a CallInst can |
1415 | /// be tail call optimized. |
1416 | bool isMustTailCall() const; |
1417 | |
1418 | /// Tests if this call site is marked as a tail call. |
1419 | bool isTailCall() const; |
1420 | |
1421 | /// Returns the intrinsic ID of the intrinsic called or |
1422 | /// Intrinsic::not_intrinsic if the called function is not an intrinsic, or if |
1423 | /// this is an indirect call. |
1424 | Intrinsic::ID getIntrinsicID() const; |
1425 | |
1426 | void setCalledOperand(Value *V) { Op<CalledOperandOpEndIdx>() = V; } |
1427 | |
1428 | /// Sets the function called, including updating the function type. |
1429 | void setCalledFunction(Function *Fn) { |
1430 | setCalledFunction(Fn->getFunctionType(), Fn); |
1431 | } |
1432 | |
1433 | /// Sets the function called, including updating the function type. |
1434 | void setCalledFunction(FunctionCallee Fn) { |
1435 | setCalledFunction(Fn.getFunctionType(), Fn.getCallee()); |
1436 | } |
1437 | |
1438 | /// Sets the function called, including updating to the specified function |
1439 | /// type. |
1440 | void setCalledFunction(FunctionType *FTy, Value *Fn) { |
1441 | this->FTy = FTy; |
1442 | assert(cast<PointerType>(Fn->getType())->isOpaqueOrPointeeTypeMatches(FTy))(static_cast <bool> (cast<PointerType>(Fn->getType ())->isOpaqueOrPointeeTypeMatches(FTy)) ? void (0) : __assert_fail ("cast<PointerType>(Fn->getType())->isOpaqueOrPointeeTypeMatches(FTy)" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1442, __extension__ __PRETTY_FUNCTION__)); |
1443 | // This function doesn't mutate the return type, only the function |
1444 | // type. Seems broken, but I'm just gonna stick an assert in for now. |
1445 | assert(getType() == FTy->getReturnType())(static_cast <bool> (getType() == FTy->getReturnType ()) ? void (0) : __assert_fail ("getType() == FTy->getReturnType()" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1445, __extension__ __PRETTY_FUNCTION__)); |
1446 | setCalledOperand(Fn); |
1447 | } |
1448 | |
1449 | CallingConv::ID getCallingConv() const { |
1450 | return getSubclassData<CallingConvField>(); |
1451 | } |
1452 | |
1453 | void setCallingConv(CallingConv::ID CC) { |
1454 | setSubclassData<CallingConvField>(CC); |
1455 | } |
1456 | |
1457 | /// Check if this call is an inline asm statement. |
1458 | bool isInlineAsm() const { return isa<InlineAsm>(getCalledOperand()); } |
1459 | |
1460 | /// \name Attribute API |
1461 | /// |
1462 | /// These methods access and modify attributes on this call (including |
1463 | /// looking through to the attributes on the called function when necessary). |
1464 | ///@{ |
1465 | |
1466 | /// Return the parameter attributes for this call. |
1467 | /// |
1468 | AttributeList getAttributes() const { return Attrs; } |
1469 | |
1470 | /// Set the parameter attributes for this call. |
1471 | /// |
1472 | void setAttributes(AttributeList A) { Attrs = A; } |
1473 | |
1474 | /// Determine whether this call has the given attribute. If it does not |
1475 | /// then determine if the called function has the attribute, but only if |
1476 | /// the attribute is allowed for the call. |
1477 | bool hasFnAttr(Attribute::AttrKind Kind) const { |
1478 | assert(Kind != Attribute::NoBuiltin &&(static_cast <bool> (Kind != Attribute::NoBuiltin && "Use CallBase::isNoBuiltin() to check for Attribute::NoBuiltin" ) ? void (0) : __assert_fail ("Kind != Attribute::NoBuiltin && \"Use CallBase::isNoBuiltin() to check for Attribute::NoBuiltin\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1479, __extension__ __PRETTY_FUNCTION__)) |
1479 | "Use CallBase::isNoBuiltin() to check for Attribute::NoBuiltin")(static_cast <bool> (Kind != Attribute::NoBuiltin && "Use CallBase::isNoBuiltin() to check for Attribute::NoBuiltin" ) ? void (0) : __assert_fail ("Kind != Attribute::NoBuiltin && \"Use CallBase::isNoBuiltin() to check for Attribute::NoBuiltin\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1479, __extension__ __PRETTY_FUNCTION__)); |
1480 | return hasFnAttrImpl(Kind); |
1481 | } |
1482 | |
1483 | /// Determine whether this call has the given attribute. If it does not |
1484 | /// then determine if the called function has the attribute, but only if |
1485 | /// the attribute is allowed for the call. |
1486 | bool hasFnAttr(StringRef Kind) const { return hasFnAttrImpl(Kind); } |
1487 | |
1488 | /// adds the attribute to the list of attributes. |
1489 | void addAttribute(unsigned i, Attribute::AttrKind Kind) { |
1490 | AttributeList PAL = getAttributes(); |
1491 | PAL = PAL.addAttribute(getContext(), i, Kind); |
1492 | setAttributes(PAL); |
1493 | } |
1494 | |
1495 | /// adds the attribute to the list of attributes. |
1496 | void addAttribute(unsigned i, Attribute Attr) { |
1497 | AttributeList PAL = getAttributes(); |
1498 | PAL = PAL.addAttribute(getContext(), i, Attr); |
1499 | setAttributes(PAL); |
1500 | } |
1501 | |
1502 | /// Adds the attribute to the indicated argument |
1503 | void addParamAttr(unsigned ArgNo, Attribute::AttrKind Kind) { |
1504 | assert(ArgNo < getNumArgOperands() && "Out of bounds")(static_cast <bool> (ArgNo < getNumArgOperands() && "Out of bounds") ? void (0) : __assert_fail ("ArgNo < getNumArgOperands() && \"Out of bounds\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1504, __extension__ __PRETTY_FUNCTION__)); |
1505 | AttributeList PAL = getAttributes(); |
1506 | PAL = PAL.addParamAttribute(getContext(), ArgNo, Kind); |
1507 | setAttributes(PAL); |
1508 | } |
1509 | |
1510 | /// Adds the attribute to the indicated argument |
1511 | void addParamAttr(unsigned ArgNo, Attribute Attr) { |
1512 | assert(ArgNo < getNumArgOperands() && "Out of bounds")(static_cast <bool> (ArgNo < getNumArgOperands() && "Out of bounds") ? void (0) : __assert_fail ("ArgNo < getNumArgOperands() && \"Out of bounds\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1512, __extension__ __PRETTY_FUNCTION__)); |
1513 | AttributeList PAL = getAttributes(); |
1514 | PAL = PAL.addParamAttribute(getContext(), ArgNo, Attr); |
1515 | setAttributes(PAL); |
1516 | } |
1517 | |
1518 | /// removes the attribute from the list of attributes. |
1519 | void removeAttribute(unsigned i, Attribute::AttrKind Kind) { |
1520 | AttributeList PAL = getAttributes(); |
1521 | PAL = PAL.removeAttribute(getContext(), i, Kind); |
1522 | setAttributes(PAL); |
1523 | } |
1524 | |
1525 | /// removes the attribute from the list of attributes. |
1526 | void removeAttribute(unsigned i, StringRef Kind) { |
1527 | AttributeList PAL = getAttributes(); |
1528 | PAL = PAL.removeAttribute(getContext(), i, Kind); |
1529 | setAttributes(PAL); |
1530 | } |
1531 | |
1532 | void removeAttributes(unsigned i, const AttrBuilder &Attrs) { |
1533 | AttributeList PAL = getAttributes(); |
1534 | PAL = PAL.removeAttributes(getContext(), i, Attrs); |
1535 | setAttributes(PAL); |
1536 | } |
1537 | |
1538 | /// Removes the attribute from the given argument |
1539 | void removeParamAttr(unsigned ArgNo, Attribute::AttrKind Kind) { |
1540 | assert(ArgNo < getNumArgOperands() && "Out of bounds")(static_cast <bool> (ArgNo < getNumArgOperands() && "Out of bounds") ? void (0) : __assert_fail ("ArgNo < getNumArgOperands() && \"Out of bounds\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1540, __extension__ __PRETTY_FUNCTION__)); |
1541 | AttributeList PAL = getAttributes(); |
1542 | PAL = PAL.removeParamAttribute(getContext(), ArgNo, Kind); |
1543 | setAttributes(PAL); |
1544 | } |
1545 | |
1546 | /// Removes the attribute from the given argument |
1547 | void removeParamAttr(unsigned ArgNo, StringRef Kind) { |
1548 | assert(ArgNo < getNumArgOperands() && "Out of bounds")(static_cast <bool> (ArgNo < getNumArgOperands() && "Out of bounds") ? void (0) : __assert_fail ("ArgNo < getNumArgOperands() && \"Out of bounds\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1548, __extension__ __PRETTY_FUNCTION__)); |
1549 | AttributeList PAL = getAttributes(); |
1550 | PAL = PAL.removeParamAttribute(getContext(), ArgNo, Kind); |
1551 | setAttributes(PAL); |
1552 | } |
1553 | |
1554 | /// Removes the attributes from the given argument |
1555 | void removeParamAttrs(unsigned ArgNo, const AttrBuilder &Attrs) { |
1556 | AttributeList PAL = getAttributes(); |
1557 | PAL = PAL.removeParamAttributes(getContext(), ArgNo, Attrs); |
1558 | setAttributes(PAL); |
1559 | } |
1560 | |
1561 | /// adds the dereferenceable attribute to the list of attributes. |
1562 | void addDereferenceableAttr(unsigned i, uint64_t Bytes) { |
1563 | AttributeList PAL = getAttributes(); |
1564 | PAL = PAL.addDereferenceableAttr(getContext(), i, Bytes); |
1565 | setAttributes(PAL); |
1566 | } |
1567 | |
1568 | /// adds the dereferenceable_or_null attribute to the list of |
1569 | /// attributes. |
1570 | void addDereferenceableOrNullAttr(unsigned i, uint64_t Bytes) { |
1571 | AttributeList PAL = getAttributes(); |
1572 | PAL = PAL.addDereferenceableOrNullAttr(getContext(), i, Bytes); |
1573 | setAttributes(PAL); |
1574 | } |
1575 | |
1576 | /// Determine whether the return value has the given attribute. |
1577 | bool hasRetAttr(Attribute::AttrKind Kind) const { |
1578 | return hasRetAttrImpl(Kind); |
1579 | } |
1580 | /// Determine whether the return value has the given attribute. |
1581 | bool hasRetAttr(StringRef Kind) const { return hasRetAttrImpl(Kind); } |
1582 | |
1583 | /// Determine whether the argument or parameter has the given attribute. |
1584 | bool paramHasAttr(unsigned ArgNo, Attribute::AttrKind Kind) const; |
1585 | |
1586 | /// Get the attribute of a given kind at a position. |
1587 | Attribute getAttribute(unsigned i, Attribute::AttrKind Kind) const { |
1588 | return getAttributes().getAttribute(i, Kind); |
1589 | } |
1590 | |
1591 | /// Get the attribute of a given kind at a position. |
1592 | Attribute getAttribute(unsigned i, StringRef Kind) const { |
1593 | return getAttributes().getAttribute(i, Kind); |
1594 | } |
1595 | |
1596 | /// Get the attribute of a given kind from a given arg |
1597 | Attribute getParamAttr(unsigned ArgNo, Attribute::AttrKind Kind) const { |
1598 | assert(ArgNo < getNumArgOperands() && "Out of bounds")(static_cast <bool> (ArgNo < getNumArgOperands() && "Out of bounds") ? void (0) : __assert_fail ("ArgNo < getNumArgOperands() && \"Out of bounds\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1598, __extension__ __PRETTY_FUNCTION__)); |
1599 | return getAttributes().getParamAttr(ArgNo, Kind); |
1600 | } |
1601 | |
1602 | /// Get the attribute of a given kind from a given arg |
1603 | Attribute getParamAttr(unsigned ArgNo, StringRef Kind) const { |
1604 | assert(ArgNo < getNumArgOperands() && "Out of bounds")(static_cast <bool> (ArgNo < getNumArgOperands() && "Out of bounds") ? void (0) : __assert_fail ("ArgNo < getNumArgOperands() && \"Out of bounds\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1604, __extension__ __PRETTY_FUNCTION__)); |
1605 | return getAttributes().getParamAttr(ArgNo, Kind); |
1606 | } |
1607 | |
1608 | /// Return true if the data operand at index \p i has the attribute \p |
1609 | /// A. |
1610 | /// |
1611 | /// Data operands include call arguments and values used in operand bundles, |
1612 | /// but does not include the callee operand. This routine dispatches to the |
1613 | /// underlying AttributeList or the OperandBundleUser as appropriate. |
1614 | /// |
1615 | /// The index \p i is interpreted as |
1616 | /// |
1617 | /// \p i == Attribute::ReturnIndex -> the return value |
1618 | /// \p i in [1, arg_size + 1) -> argument number (\p i - 1) |
1619 | /// \p i in [arg_size + 1, data_operand_size + 1) -> bundle operand at index |
1620 | /// (\p i - 1) in the operand list. |
1621 | bool dataOperandHasImpliedAttr(unsigned i, Attribute::AttrKind Kind) const { |
1622 | // Note that we have to add one because `i` isn't zero-indexed. |
1623 | assert(i < (getNumArgOperands() + getNumTotalBundleOperands() + 1) &&(static_cast <bool> (i < (getNumArgOperands() + getNumTotalBundleOperands () + 1) && "Data operand index out of bounds!") ? void (0) : __assert_fail ("i < (getNumArgOperands() + getNumTotalBundleOperands() + 1) && \"Data operand index out of bounds!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1624, __extension__ __PRETTY_FUNCTION__)) |
1624 | "Data operand index out of bounds!")(static_cast <bool> (i < (getNumArgOperands() + getNumTotalBundleOperands () + 1) && "Data operand index out of bounds!") ? void (0) : __assert_fail ("i < (getNumArgOperands() + getNumTotalBundleOperands() + 1) && \"Data operand index out of bounds!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1624, __extension__ __PRETTY_FUNCTION__)); |
1625 | |
1626 | // The attribute A can either be directly specified, if the operand in |
1627 | // question is a call argument; or be indirectly implied by the kind of its |
1628 | // containing operand bundle, if the operand is a bundle operand. |
1629 | |
1630 | if (i == AttributeList::ReturnIndex) |
1631 | return hasRetAttr(Kind); |
1632 | |
1633 | // FIXME: Avoid these i - 1 calculations and update the API to use |
1634 | // zero-based indices. |
1635 | if (i < (getNumArgOperands() + 1)) |
1636 | return paramHasAttr(i - 1, Kind); |
1637 | |
1638 | assert(hasOperandBundles() && i >= (getBundleOperandsStartIndex() + 1) &&(static_cast <bool> (hasOperandBundles() && i >= (getBundleOperandsStartIndex() + 1) && "Must be either a call argument or an operand bundle!" ) ? void (0) : __assert_fail ("hasOperandBundles() && i >= (getBundleOperandsStartIndex() + 1) && \"Must be either a call argument or an operand bundle!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1639, __extension__ __PRETTY_FUNCTION__)) |
1639 | "Must be either a call argument or an operand bundle!")(static_cast <bool> (hasOperandBundles() && i >= (getBundleOperandsStartIndex() + 1) && "Must be either a call argument or an operand bundle!" ) ? void (0) : __assert_fail ("hasOperandBundles() && i >= (getBundleOperandsStartIndex() + 1) && \"Must be either a call argument or an operand bundle!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1639, __extension__ __PRETTY_FUNCTION__)); |
1640 | return bundleOperandHasAttr(i - 1, Kind); |
1641 | } |
1642 | |
1643 | /// Determine whether this data operand is not captured. |
1644 | // FIXME: Once this API is no longer duplicated in `CallSite`, rename this to |
1645 | // better indicate that this may return a conservative answer. |
1646 | bool doesNotCapture(unsigned OpNo) const { |
1647 | return dataOperandHasImpliedAttr(OpNo + 1, Attribute::NoCapture); |
1648 | } |
1649 | |
1650 | /// Determine whether this argument is passed by value. |
1651 | bool isByValArgument(unsigned ArgNo) const { |
1652 | return paramHasAttr(ArgNo, Attribute::ByVal); |
1653 | } |
1654 | |
1655 | /// Determine whether this argument is passed in an alloca. |
1656 | bool isInAllocaArgument(unsigned ArgNo) const { |
1657 | return paramHasAttr(ArgNo, Attribute::InAlloca); |
1658 | } |
1659 | |
1660 | /// Determine whether this argument is passed by value, in an alloca, or is |
1661 | /// preallocated. |
1662 | bool isPassPointeeByValueArgument(unsigned ArgNo) const { |
1663 | return paramHasAttr(ArgNo, Attribute::ByVal) || |
1664 | paramHasAttr(ArgNo, Attribute::InAlloca) || |
1665 | paramHasAttr(ArgNo, Attribute::Preallocated); |
1666 | } |
1667 | |
1668 | /// Determine whether passing undef to this argument is undefined behavior. |
1669 | /// If passing undef to this argument is UB, passing poison is UB as well |
1670 | /// because poison is more undefined than undef. |
1671 | bool isPassingUndefUB(unsigned ArgNo) const { |
1672 | return paramHasAttr(ArgNo, Attribute::NoUndef) || |
1673 | // dereferenceable implies noundef. |
1674 | paramHasAttr(ArgNo, Attribute::Dereferenceable) || |
1675 | // dereferenceable implies noundef, and null is a well-defined value. |
1676 | paramHasAttr(ArgNo, Attribute::DereferenceableOrNull); |
1677 | } |
1678 | |
1679 | /// Determine if there are is an inalloca argument. Only the last argument can |
1680 | /// have the inalloca attribute. |
1681 | bool hasInAllocaArgument() const { |
1682 | return !arg_empty() && paramHasAttr(arg_size() - 1, Attribute::InAlloca); |
1683 | } |
1684 | |
1685 | // FIXME: Once this API is no longer duplicated in `CallSite`, rename this to |
1686 | // better indicate that this may return a conservative answer. |
1687 | bool doesNotAccessMemory(unsigned OpNo) const { |
1688 | return dataOperandHasImpliedAttr(OpNo + 1, Attribute::ReadNone); |
1689 | } |
1690 | |
1691 | // FIXME: Once this API is no longer duplicated in `CallSite`, rename this to |
1692 | // better indicate that this may return a conservative answer. |
1693 | bool onlyReadsMemory(unsigned OpNo) const { |
1694 | return dataOperandHasImpliedAttr(OpNo + 1, Attribute::ReadOnly) || |
1695 | dataOperandHasImpliedAttr(OpNo + 1, Attribute::ReadNone); |
1696 | } |
1697 | |
1698 | // FIXME: Once this API is no longer duplicated in `CallSite`, rename this to |
1699 | // better indicate that this may return a conservative answer. |
1700 | bool doesNotReadMemory(unsigned OpNo) const { |
1701 | return dataOperandHasImpliedAttr(OpNo + 1, Attribute::WriteOnly) || |
1702 | dataOperandHasImpliedAttr(OpNo + 1, Attribute::ReadNone); |
1703 | } |
1704 | |
1705 | /// Extract the alignment of the return value. |
1706 | MaybeAlign getRetAlign() const { return Attrs.getRetAlignment(); } |
1707 | |
1708 | /// Extract the alignment for a call or parameter (0=unknown). |
1709 | MaybeAlign getParamAlign(unsigned ArgNo) const { |
1710 | return Attrs.getParamAlignment(ArgNo); |
1711 | } |
1712 | |
1713 | MaybeAlign getParamStackAlign(unsigned ArgNo) const { |
1714 | return Attrs.getParamStackAlignment(ArgNo); |
1715 | } |
1716 | |
1717 | /// Extract the byval type for a call or parameter. |
1718 | Type *getParamByValType(unsigned ArgNo) const { |
1719 | if (auto *Ty = Attrs.getParamByValType(ArgNo)) |
1720 | return Ty; |
1721 | if (const Function *F = getCalledFunction()) |
1722 | return F->getAttributes().getParamByValType(ArgNo); |
1723 | return nullptr; |
1724 | } |
1725 | |
1726 | /// Extract the preallocated type for a call or parameter. |
1727 | Type *getParamPreallocatedType(unsigned ArgNo) const { |
1728 | if (auto *Ty = Attrs.getParamPreallocatedType(ArgNo)) |
1729 | return Ty; |
1730 | if (const Function *F = getCalledFunction()) |
1731 | return F->getAttributes().getParamPreallocatedType(ArgNo); |
1732 | return nullptr; |
1733 | } |
1734 | |
1735 | /// Extract the preallocated type for a call or parameter. |
1736 | Type *getParamInAllocaType(unsigned ArgNo) const { |
1737 | if (auto *Ty = Attrs.getParamInAllocaType(ArgNo)) |
1738 | return Ty; |
1739 | if (const Function *F = getCalledFunction()) |
1740 | return F->getAttributes().getParamInAllocaType(ArgNo); |
1741 | return nullptr; |
1742 | } |
1743 | |
1744 | /// Extract the number of dereferenceable bytes for a call or |
1745 | /// parameter (0=unknown). |
1746 | uint64_t getDereferenceableBytes(unsigned i) const { |
1747 | return Attrs.getDereferenceableBytes(i); |
1748 | } |
1749 | |
1750 | /// Extract the number of dereferenceable_or_null bytes for a call or |
1751 | /// parameter (0=unknown). |
1752 | uint64_t getDereferenceableOrNullBytes(unsigned i) const { |
1753 | return Attrs.getDereferenceableOrNullBytes(i); |
1754 | } |
1755 | |
1756 | /// Return true if the return value is known to be not null. |
1757 | /// This may be because it has the nonnull attribute, or because at least |
1758 | /// one byte is dereferenceable and the pointer is in addrspace(0). |
1759 | bool isReturnNonNull() const; |
1760 | |
1761 | /// Determine if the return value is marked with NoAlias attribute. |
1762 | bool returnDoesNotAlias() const { |
1763 | return Attrs.hasAttribute(AttributeList::ReturnIndex, Attribute::NoAlias); |
1764 | } |
1765 | |
1766 | /// If one of the arguments has the 'returned' attribute, returns its |
1767 | /// operand value. Otherwise, return nullptr. |
1768 | Value *getReturnedArgOperand() const; |
1769 | |
1770 | /// Return true if the call should not be treated as a call to a |
1771 | /// builtin. |
1772 | bool isNoBuiltin() const { |
1773 | return hasFnAttrImpl(Attribute::NoBuiltin) && |
1774 | !hasFnAttrImpl(Attribute::Builtin); |
1775 | } |
1776 | |
1777 | /// Determine if the call requires strict floating point semantics. |
1778 | bool isStrictFP() const { return hasFnAttr(Attribute::StrictFP); } |
1779 | |
1780 | /// Return true if the call should not be inlined. |
1781 | bool isNoInline() const { return hasFnAttr(Attribute::NoInline); } |
1782 | void setIsNoInline() { |
1783 | addAttribute(AttributeList::FunctionIndex, Attribute::NoInline); |
1784 | } |
1785 | /// Determine if the call does not access memory. |
1786 | bool doesNotAccessMemory() const { return hasFnAttr(Attribute::ReadNone); } |
1787 | void setDoesNotAccessMemory() { |
1788 | addAttribute(AttributeList::FunctionIndex, Attribute::ReadNone); |
1789 | } |
1790 | |
1791 | /// Determine if the call does not access or only reads memory. |
1792 | bool onlyReadsMemory() const { |
1793 | return doesNotAccessMemory() || hasFnAttr(Attribute::ReadOnly); |
1794 | } |
1795 | |
1796 | void setOnlyReadsMemory() { |
1797 | addAttribute(AttributeList::FunctionIndex, Attribute::ReadOnly); |
1798 | } |
1799 | |
1800 | /// Determine if the call does not access or only writes memory. |
1801 | bool doesNotReadMemory() const { |
1802 | return doesNotAccessMemory() || hasFnAttr(Attribute::WriteOnly); |
1803 | } |
1804 | void setDoesNotReadMemory() { |
1805 | addAttribute(AttributeList::FunctionIndex, Attribute::WriteOnly); |
1806 | } |
1807 | |
1808 | /// Determine if the call can access memmory only using pointers based |
1809 | /// on its arguments. |
1810 | bool onlyAccessesArgMemory() const { |
1811 | return hasFnAttr(Attribute::ArgMemOnly); |
1812 | } |
1813 | void setOnlyAccessesArgMemory() { |
1814 | addAttribute(AttributeList::FunctionIndex, Attribute::ArgMemOnly); |
1815 | } |
1816 | |
1817 | /// Determine if the function may only access memory that is |
1818 | /// inaccessible from the IR. |
1819 | bool onlyAccessesInaccessibleMemory() const { |
1820 | return hasFnAttr(Attribute::InaccessibleMemOnly); |
1821 | } |
1822 | void setOnlyAccessesInaccessibleMemory() { |
1823 | addAttribute(AttributeList::FunctionIndex, Attribute::InaccessibleMemOnly); |
1824 | } |
1825 | |
1826 | /// Determine if the function may only access memory that is |
1827 | /// either inaccessible from the IR or pointed to by its arguments. |
1828 | bool onlyAccessesInaccessibleMemOrArgMem() const { |
1829 | return hasFnAttr(Attribute::InaccessibleMemOrArgMemOnly); |
1830 | } |
1831 | void setOnlyAccessesInaccessibleMemOrArgMem() { |
1832 | addAttribute(AttributeList::FunctionIndex, |
1833 | Attribute::InaccessibleMemOrArgMemOnly); |
1834 | } |
1835 | /// Determine if the call cannot return. |
1836 | bool doesNotReturn() const { return hasFnAttr(Attribute::NoReturn); } |
1837 | void setDoesNotReturn() { |
1838 | addAttribute(AttributeList::FunctionIndex, Attribute::NoReturn); |
1839 | } |
1840 | |
1841 | /// Determine if the call should not perform indirect branch tracking. |
1842 | bool doesNoCfCheck() const { return hasFnAttr(Attribute::NoCfCheck); } |
1843 | |
1844 | /// Determine if the call cannot unwind. |
1845 | bool doesNotThrow() const { return hasFnAttr(Attribute::NoUnwind); } |
1846 | void setDoesNotThrow() { |
1847 | addAttribute(AttributeList::FunctionIndex, Attribute::NoUnwind); |
1848 | } |
1849 | |
1850 | /// Determine if the invoke cannot be duplicated. |
1851 | bool cannotDuplicate() const { return hasFnAttr(Attribute::NoDuplicate); } |
1852 | void setCannotDuplicate() { |
1853 | addAttribute(AttributeList::FunctionIndex, Attribute::NoDuplicate); |
1854 | } |
1855 | |
1856 | /// Determine if the call cannot be tail merged. |
1857 | bool cannotMerge() const { return hasFnAttr(Attribute::NoMerge); } |
1858 | void setCannotMerge() { |
1859 | addAttribute(AttributeList::FunctionIndex, Attribute::NoMerge); |
1860 | } |
1861 | |
1862 | /// Determine if the invoke is convergent |
1863 | bool isConvergent() const { return hasFnAttr(Attribute::Convergent); } |
1864 | void setConvergent() { |
1865 | addAttribute(AttributeList::FunctionIndex, Attribute::Convergent); |
1866 | } |
1867 | void setNotConvergent() { |
1868 | removeAttribute(AttributeList::FunctionIndex, Attribute::Convergent); |
1869 | } |
1870 | |
1871 | /// Determine if the call returns a structure through first |
1872 | /// pointer argument. |
1873 | bool hasStructRetAttr() const { |
1874 | if (getNumArgOperands() == 0) |
1875 | return false; |
1876 | |
1877 | // Be friendly and also check the callee. |
1878 | return paramHasAttr(0, Attribute::StructRet); |
1879 | } |
1880 | |
1881 | /// Determine if any call argument is an aggregate passed by value. |
1882 | bool hasByValArgument() const { |
1883 | return Attrs.hasAttrSomewhere(Attribute::ByVal); |
1884 | } |
1885 | |
1886 | ///@{ |
1887 | // End of attribute API. |
1888 | |
1889 | /// \name Operand Bundle API |
1890 | /// |
1891 | /// This group of methods provides the API to access and manipulate operand |
1892 | /// bundles on this call. |
1893 | /// @{ |
1894 | |
1895 | /// Return the number of operand bundles associated with this User. |
1896 | unsigned getNumOperandBundles() const { |
1897 | return std::distance(bundle_op_info_begin(), bundle_op_info_end()); |
1898 | } |
1899 | |
1900 | /// Return true if this User has any operand bundles. |
1901 | bool hasOperandBundles() const { return getNumOperandBundles() != 0; } |
1902 | |
1903 | /// Return the index of the first bundle operand in the Use array. |
1904 | unsigned getBundleOperandsStartIndex() const { |
1905 | assert(hasOperandBundles() && "Don't call otherwise!")(static_cast <bool> (hasOperandBundles() && "Don't call otherwise!" ) ? void (0) : __assert_fail ("hasOperandBundles() && \"Don't call otherwise!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1905, __extension__ __PRETTY_FUNCTION__)); |
1906 | return bundle_op_info_begin()->Begin; |
1907 | } |
1908 | |
1909 | /// Return the index of the last bundle operand in the Use array. |
1910 | unsigned getBundleOperandsEndIndex() const { |
1911 | assert(hasOperandBundles() && "Don't call otherwise!")(static_cast <bool> (hasOperandBundles() && "Don't call otherwise!" ) ? void (0) : __assert_fail ("hasOperandBundles() && \"Don't call otherwise!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1911, __extension__ __PRETTY_FUNCTION__)); |
1912 | return bundle_op_info_end()[-1].End; |
1913 | } |
1914 | |
1915 | /// Return true if the operand at index \p Idx is a bundle operand. |
1916 | bool isBundleOperand(unsigned Idx) const { |
1917 | return hasOperandBundles() && Idx >= getBundleOperandsStartIndex() && |
1918 | Idx < getBundleOperandsEndIndex(); |
1919 | } |
1920 | |
1921 | /// Returns true if the use is a bundle operand. |
1922 | bool isBundleOperand(const Use *U) const { |
1923 | assert(this == U->getUser() &&(static_cast <bool> (this == U->getUser() && "Only valid to query with a use of this instruction!") ? void (0) : __assert_fail ("this == U->getUser() && \"Only valid to query with a use of this instruction!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1924, __extension__ __PRETTY_FUNCTION__)) |
1924 | "Only valid to query with a use of this instruction!")(static_cast <bool> (this == U->getUser() && "Only valid to query with a use of this instruction!") ? void (0) : __assert_fail ("this == U->getUser() && \"Only valid to query with a use of this instruction!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1924, __extension__ __PRETTY_FUNCTION__)); |
1925 | return hasOperandBundles() && isBundleOperand(U - op_begin()); |
1926 | } |
1927 | bool isBundleOperand(Value::const_user_iterator UI) const { |
1928 | return isBundleOperand(&UI.getUse()); |
1929 | } |
1930 | |
1931 | /// Return the total number operands (not operand bundles) used by |
1932 | /// every operand bundle in this OperandBundleUser. |
1933 | unsigned getNumTotalBundleOperands() const { |
1934 | if (!hasOperandBundles()) |
1935 | return 0; |
1936 | |
1937 | unsigned Begin = getBundleOperandsStartIndex(); |
1938 | unsigned End = getBundleOperandsEndIndex(); |
1939 | |
1940 | assert(Begin <= End && "Should be!")(static_cast <bool> (Begin <= End && "Should be!" ) ? void (0) : __assert_fail ("Begin <= End && \"Should be!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1940, __extension__ __PRETTY_FUNCTION__)); |
1941 | return End - Begin; |
1942 | } |
1943 | |
1944 | /// Return the operand bundle at a specific index. |
1945 | OperandBundleUse getOperandBundleAt(unsigned Index) const { |
1946 | assert(Index < getNumOperandBundles() && "Index out of bounds!")(static_cast <bool> (Index < getNumOperandBundles() && "Index out of bounds!") ? void (0) : __assert_fail ("Index < getNumOperandBundles() && \"Index out of bounds!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1946, __extension__ __PRETTY_FUNCTION__)); |
1947 | return operandBundleFromBundleOpInfo(*(bundle_op_info_begin() + Index)); |
1948 | } |
1949 | |
1950 | /// Return the number of operand bundles with the tag Name attached to |
1951 | /// this instruction. |
1952 | unsigned countOperandBundlesOfType(StringRef Name) const { |
1953 | unsigned Count = 0; |
1954 | for (unsigned i = 0, e = getNumOperandBundles(); i != e; ++i) |
1955 | if (getOperandBundleAt(i).getTagName() == Name) |
1956 | Count++; |
1957 | |
1958 | return Count; |
1959 | } |
1960 | |
1961 | /// Return the number of operand bundles with the tag ID attached to |
1962 | /// this instruction. |
1963 | unsigned countOperandBundlesOfType(uint32_t ID) const { |
1964 | unsigned Count = 0; |
1965 | for (unsigned i = 0, e = getNumOperandBundles(); i != e; ++i) |
1966 | if (getOperandBundleAt(i).getTagID() == ID) |
1967 | Count++; |
1968 | |
1969 | return Count; |
1970 | } |
1971 | |
1972 | /// Return an operand bundle by name, if present. |
1973 | /// |
1974 | /// It is an error to call this for operand bundle types that may have |
1975 | /// multiple instances of them on the same instruction. |
1976 | Optional<OperandBundleUse> getOperandBundle(StringRef Name) const { |
1977 | assert(countOperandBundlesOfType(Name) < 2 && "Precondition violated!")(static_cast <bool> (countOperandBundlesOfType(Name) < 2 && "Precondition violated!") ? void (0) : __assert_fail ("countOperandBundlesOfType(Name) < 2 && \"Precondition violated!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1977, __extension__ __PRETTY_FUNCTION__)); |
1978 | |
1979 | for (unsigned i = 0, e = getNumOperandBundles(); i != e; ++i) { |
1980 | OperandBundleUse U = getOperandBundleAt(i); |
1981 | if (U.getTagName() == Name) |
1982 | return U; |
1983 | } |
1984 | |
1985 | return None; |
1986 | } |
1987 | |
1988 | /// Return an operand bundle by tag ID, if present. |
1989 | /// |
1990 | /// It is an error to call this for operand bundle types that may have |
1991 | /// multiple instances of them on the same instruction. |
1992 | Optional<OperandBundleUse> getOperandBundle(uint32_t ID) const { |
1993 | assert(countOperandBundlesOfType(ID) < 2 && "Precondition violated!")(static_cast <bool> (countOperandBundlesOfType(ID) < 2 && "Precondition violated!") ? void (0) : __assert_fail ("countOperandBundlesOfType(ID) < 2 && \"Precondition violated!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 1993, __extension__ __PRETTY_FUNCTION__)); |
1994 | |
1995 | for (unsigned i = 0, e = getNumOperandBundles(); i != e; ++i) { |
1996 | OperandBundleUse U = getOperandBundleAt(i); |
1997 | if (U.getTagID() == ID) |
1998 | return U; |
1999 | } |
2000 | |
2001 | return None; |
2002 | } |
2003 | |
2004 | /// Return the list of operand bundles attached to this instruction as |
2005 | /// a vector of OperandBundleDefs. |
2006 | /// |
2007 | /// This function copies the OperandBundeUse instances associated with this |
2008 | /// OperandBundleUser to a vector of OperandBundleDefs. Note: |
2009 | /// OperandBundeUses and OperandBundleDefs are non-trivially *different* |
2010 | /// representations of operand bundles (see documentation above). |
2011 | void getOperandBundlesAsDefs(SmallVectorImpl<OperandBundleDef> &Defs) const; |
2012 | |
2013 | /// Return the operand bundle for the operand at index OpIdx. |
2014 | /// |
2015 | /// It is an error to call this with an OpIdx that does not correspond to an |
2016 | /// bundle operand. |
2017 | OperandBundleUse getOperandBundleForOperand(unsigned OpIdx) const { |
2018 | return operandBundleFromBundleOpInfo(getBundleOpInfoForOperand(OpIdx)); |
2019 | } |
2020 | |
2021 | /// Return true if this operand bundle user has operand bundles that |
2022 | /// may read from the heap. |
2023 | bool hasReadingOperandBundles() const; |
2024 | |
2025 | /// Return true if this operand bundle user has operand bundles that |
2026 | /// may write to the heap. |
2027 | bool hasClobberingOperandBundles() const { |
2028 | for (auto &BOI : bundle_op_infos()) { |
2029 | if (BOI.Tag->second == LLVMContext::OB_deopt || |
2030 | BOI.Tag->second == LLVMContext::OB_funclet) |
2031 | continue; |
2032 | |
2033 | // This instruction has an operand bundle that is not known to us. |
2034 | // Assume the worst. |
2035 | return true; |
2036 | } |
2037 | |
2038 | return false; |
2039 | } |
2040 | |
2041 | /// Return true if the bundle operand at index \p OpIdx has the |
2042 | /// attribute \p A. |
2043 | bool bundleOperandHasAttr(unsigned OpIdx, Attribute::AttrKind A) const { |
2044 | auto &BOI = getBundleOpInfoForOperand(OpIdx); |
2045 | auto OBU = operandBundleFromBundleOpInfo(BOI); |
2046 | return OBU.operandHasAttr(OpIdx - BOI.Begin, A); |
2047 | } |
2048 | |
2049 | /// Return true if \p Other has the same sequence of operand bundle |
2050 | /// tags with the same number of operands on each one of them as this |
2051 | /// OperandBundleUser. |
2052 | bool hasIdenticalOperandBundleSchema(const CallBase &Other) const { |
2053 | if (getNumOperandBundles() != Other.getNumOperandBundles()) |
2054 | return false; |
2055 | |
2056 | return std::equal(bundle_op_info_begin(), bundle_op_info_end(), |
2057 | Other.bundle_op_info_begin()); |
2058 | } |
2059 | |
2060 | /// Return true if this operand bundle user contains operand bundles |
2061 | /// with tags other than those specified in \p IDs. |
2062 | bool hasOperandBundlesOtherThan(ArrayRef<uint32_t> IDs) const { |
2063 | for (unsigned i = 0, e = getNumOperandBundles(); i != e; ++i) { |
2064 | uint32_t ID = getOperandBundleAt(i).getTagID(); |
2065 | if (!is_contained(IDs, ID)) |
2066 | return true; |
2067 | } |
2068 | return false; |
2069 | } |
2070 | |
2071 | /// Is the function attribute S disallowed by some operand bundle on |
2072 | /// this operand bundle user? |
2073 | bool isFnAttrDisallowedByOpBundle(StringRef S) const { |
2074 | // Operand bundles only possibly disallow readnone, readonly and argmemonly |
2075 | // attributes. All String attributes are fine. |
2076 | return false; |
2077 | } |
2078 | |
2079 | /// Is the function attribute A disallowed by some operand bundle on |
2080 | /// this operand bundle user? |
2081 | bool isFnAttrDisallowedByOpBundle(Attribute::AttrKind A) const { |
2082 | switch (A) { |
2083 | default: |
2084 | return false; |
2085 | |
2086 | case Attribute::InaccessibleMemOrArgMemOnly: |
2087 | return hasReadingOperandBundles(); |
2088 | |
2089 | case Attribute::InaccessibleMemOnly: |
2090 | return hasReadingOperandBundles(); |
2091 | |
2092 | case Attribute::ArgMemOnly: |
2093 | return hasReadingOperandBundles(); |
2094 | |
2095 | case Attribute::ReadNone: |
2096 | return hasReadingOperandBundles(); |
2097 | |
2098 | case Attribute::ReadOnly: |
2099 | return hasClobberingOperandBundles(); |
2100 | } |
2101 | |
2102 | llvm_unreachable("switch has a default case!")::llvm::llvm_unreachable_internal("switch has a default case!" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 2102); |
2103 | } |
2104 | |
2105 | /// Used to keep track of an operand bundle. See the main comment on |
2106 | /// OperandBundleUser above. |
2107 | struct BundleOpInfo { |
2108 | /// The operand bundle tag, interned by |
2109 | /// LLVMContextImpl::getOrInsertBundleTag. |
2110 | StringMapEntry<uint32_t> *Tag; |
2111 | |
2112 | /// The index in the Use& vector where operands for this operand |
2113 | /// bundle starts. |
2114 | uint32_t Begin; |
2115 | |
2116 | /// The index in the Use& vector where operands for this operand |
2117 | /// bundle ends. |
2118 | uint32_t End; |
2119 | |
2120 | bool operator==(const BundleOpInfo &Other) const { |
2121 | return Tag == Other.Tag && Begin == Other.Begin && End == Other.End; |
2122 | } |
2123 | }; |
2124 | |
2125 | /// Simple helper function to map a BundleOpInfo to an |
2126 | /// OperandBundleUse. |
2127 | OperandBundleUse |
2128 | operandBundleFromBundleOpInfo(const BundleOpInfo &BOI) const { |
2129 | auto begin = op_begin(); |
2130 | ArrayRef<Use> Inputs(begin + BOI.Begin, begin + BOI.End); |
2131 | return OperandBundleUse(BOI.Tag, Inputs); |
2132 | } |
2133 | |
2134 | using bundle_op_iterator = BundleOpInfo *; |
2135 | using const_bundle_op_iterator = const BundleOpInfo *; |
2136 | |
2137 | /// Return the start of the list of BundleOpInfo instances associated |
2138 | /// with this OperandBundleUser. |
2139 | /// |
2140 | /// OperandBundleUser uses the descriptor area co-allocated with the host User |
2141 | /// to store some meta information about which operands are "normal" operands, |
2142 | /// and which ones belong to some operand bundle. |
2143 | /// |
2144 | /// The layout of an operand bundle user is |
2145 | /// |
2146 | /// +-----------uint32_t End-------------------------------------+ |
2147 | /// | | |
2148 | /// | +--------uint32_t Begin--------------------+ | |
2149 | /// | | | | |
2150 | /// ^ ^ v v |
2151 | /// |------|------|----|----|----|----|----|---------|----|---------|----|----- |
2152 | /// | BOI0 | BOI1 | .. | DU | U0 | U1 | .. | BOI0_U0 | .. | BOI1_U0 | .. | Un |
2153 | /// |------|------|----|----|----|----|----|---------|----|---------|----|----- |
2154 | /// v v ^ ^ |
2155 | /// | | | | |
2156 | /// | +--------uint32_t Begin------------+ | |
2157 | /// | | |
2158 | /// +-----------uint32_t End-----------------------------+ |
2159 | /// |
2160 | /// |
2161 | /// BOI0, BOI1 ... are descriptions of operand bundles in this User's use |
2162 | /// list. These descriptions are installed and managed by this class, and |
2163 | /// they're all instances of OperandBundleUser<T>::BundleOpInfo. |
2164 | /// |
2165 | /// DU is an additional descriptor installed by User's 'operator new' to keep |
2166 | /// track of the 'BOI0 ... BOIN' co-allocation. OperandBundleUser does not |
2167 | /// access or modify DU in any way, it's an implementation detail private to |
2168 | /// User. |
2169 | /// |
2170 | /// The regular Use& vector for the User starts at U0. The operand bundle |
2171 | /// uses are part of the Use& vector, just like normal uses. In the diagram |
2172 | /// above, the operand bundle uses start at BOI0_U0. Each instance of |
2173 | /// BundleOpInfo has information about a contiguous set of uses constituting |
2174 | /// an operand bundle, and the total set of operand bundle uses themselves |
2175 | /// form a contiguous set of uses (i.e. there are no gaps between uses |
2176 | /// corresponding to individual operand bundles). |
2177 | /// |
2178 | /// This class does not know the location of the set of operand bundle uses |
2179 | /// within the use list -- that is decided by the User using this class via |
2180 | /// the BeginIdx argument in populateBundleOperandInfos. |
2181 | /// |
2182 | /// Currently operand bundle users with hung-off operands are not supported. |
2183 | bundle_op_iterator bundle_op_info_begin() { |
2184 | if (!hasDescriptor()) |
2185 | return nullptr; |
2186 | |
2187 | uint8_t *BytesBegin = getDescriptor().begin(); |
2188 | return reinterpret_cast<bundle_op_iterator>(BytesBegin); |
2189 | } |
2190 | |
2191 | /// Return the start of the list of BundleOpInfo instances associated |
2192 | /// with this OperandBundleUser. |
2193 | const_bundle_op_iterator bundle_op_info_begin() const { |
2194 | auto *NonConstThis = const_cast<CallBase *>(this); |
2195 | return NonConstThis->bundle_op_info_begin(); |
2196 | } |
2197 | |
2198 | /// Return the end of the list of BundleOpInfo instances associated |
2199 | /// with this OperandBundleUser. |
2200 | bundle_op_iterator bundle_op_info_end() { |
2201 | if (!hasDescriptor()) |
2202 | return nullptr; |
2203 | |
2204 | uint8_t *BytesEnd = getDescriptor().end(); |
2205 | return reinterpret_cast<bundle_op_iterator>(BytesEnd); |
2206 | } |
2207 | |
2208 | /// Return the end of the list of BundleOpInfo instances associated |
2209 | /// with this OperandBundleUser. |
2210 | const_bundle_op_iterator bundle_op_info_end() const { |
2211 | auto *NonConstThis = const_cast<CallBase *>(this); |
2212 | return NonConstThis->bundle_op_info_end(); |
2213 | } |
2214 | |
2215 | /// Return the range [\p bundle_op_info_begin, \p bundle_op_info_end). |
2216 | iterator_range<bundle_op_iterator> bundle_op_infos() { |
2217 | return make_range(bundle_op_info_begin(), bundle_op_info_end()); |
2218 | } |
2219 | |
2220 | /// Return the range [\p bundle_op_info_begin, \p bundle_op_info_end). |
2221 | iterator_range<const_bundle_op_iterator> bundle_op_infos() const { |
2222 | return make_range(bundle_op_info_begin(), bundle_op_info_end()); |
2223 | } |
2224 | |
2225 | /// Populate the BundleOpInfo instances and the Use& vector from \p |
2226 | /// Bundles. Return the op_iterator pointing to the Use& one past the last |
2227 | /// last bundle operand use. |
2228 | /// |
2229 | /// Each \p OperandBundleDef instance is tracked by a OperandBundleInfo |
2230 | /// instance allocated in this User's descriptor. |
2231 | op_iterator populateBundleOperandInfos(ArrayRef<OperandBundleDef> Bundles, |
2232 | const unsigned BeginIndex); |
2233 | |
2234 | public: |
2235 | /// Return the BundleOpInfo for the operand at index OpIdx. |
2236 | /// |
2237 | /// It is an error to call this with an OpIdx that does not correspond to an |
2238 | /// bundle operand. |
2239 | BundleOpInfo &getBundleOpInfoForOperand(unsigned OpIdx); |
2240 | const BundleOpInfo &getBundleOpInfoForOperand(unsigned OpIdx) const { |
2241 | return const_cast<CallBase *>(this)->getBundleOpInfoForOperand(OpIdx); |
2242 | } |
2243 | |
2244 | protected: |
2245 | /// Return the total number of values used in \p Bundles. |
2246 | static unsigned CountBundleInputs(ArrayRef<OperandBundleDef> Bundles) { |
2247 | unsigned Total = 0; |
2248 | for (auto &B : Bundles) |
2249 | Total += B.input_size(); |
2250 | return Total; |
2251 | } |
2252 | |
2253 | /// @} |
2254 | // End of operand bundle API. |
2255 | |
2256 | private: |
2257 | bool hasFnAttrOnCalledFunction(Attribute::AttrKind Kind) const; |
2258 | bool hasFnAttrOnCalledFunction(StringRef Kind) const; |
2259 | |
2260 | template <typename AttrKind> bool hasFnAttrImpl(AttrKind Kind) const { |
2261 | if (Attrs.hasFnAttribute(Kind)) |
2262 | return true; |
2263 | |
2264 | // Operand bundles override attributes on the called function, but don't |
2265 | // override attributes directly present on the call instruction. |
2266 | if (isFnAttrDisallowedByOpBundle(Kind)) |
2267 | return false; |
2268 | |
2269 | return hasFnAttrOnCalledFunction(Kind); |
2270 | } |
2271 | |
2272 | /// Determine whether the return value has the given attribute. Supports |
2273 | /// Attribute::AttrKind and StringRef as \p AttrKind types. |
2274 | template <typename AttrKind> bool hasRetAttrImpl(AttrKind Kind) const { |
2275 | if (Attrs.hasAttribute(AttributeList::ReturnIndex, Kind)) |
2276 | return true; |
2277 | |
2278 | // Look at the callee, if available. |
2279 | if (const Function *F = getCalledFunction()) |
2280 | return F->getAttributes().hasAttribute(AttributeList::ReturnIndex, Kind); |
2281 | return false; |
2282 | } |
2283 | }; |
2284 | |
2285 | template <> |
2286 | struct OperandTraits<CallBase> : public VariadicOperandTraits<CallBase, 1> {}; |
2287 | |
2288 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(CallBase, Value)CallBase::op_iterator CallBase::op_begin() { return OperandTraits <CallBase>::op_begin(this); } CallBase::const_op_iterator CallBase::op_begin() const { return OperandTraits<CallBase >::op_begin(const_cast<CallBase*>(this)); } CallBase ::op_iterator CallBase::op_end() { return OperandTraits<CallBase >::op_end(this); } CallBase::const_op_iterator CallBase::op_end () const { return OperandTraits<CallBase>::op_end(const_cast <CallBase*>(this)); } Value *CallBase::getOperand(unsigned i_nocapture) const { (static_cast <bool> (i_nocapture < OperandTraits<CallBase>::operands(this) && "getOperand() out of range!" ) ? void (0) : __assert_fail ("i_nocapture < OperandTraits<CallBase>::operands(this) && \"getOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 2288, __extension__ __PRETTY_FUNCTION__)); return cast_or_null <Value>( OperandTraits<CallBase>::op_begin(const_cast <CallBase*>(this))[i_nocapture].get()); } void CallBase ::setOperand(unsigned i_nocapture, Value *Val_nocapture) { (static_cast <bool> (i_nocapture < OperandTraits<CallBase> ::operands(this) && "setOperand() out of range!") ? void (0) : __assert_fail ("i_nocapture < OperandTraits<CallBase>::operands(this) && \"setOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 2288, __extension__ __PRETTY_FUNCTION__)); OperandTraits< CallBase>::op_begin(this)[i_nocapture] = Val_nocapture; } unsigned CallBase::getNumOperands() const { return OperandTraits<CallBase >::operands(this); } template <int Idx_nocapture> Use &CallBase::Op() { return this->OpFrom<Idx_nocapture >(this); } template <int Idx_nocapture> const Use & CallBase::Op() const { return this->OpFrom<Idx_nocapture >(this); } |
2289 | |
2290 | //===----------------------------------------------------------------------===// |
2291 | // FuncletPadInst Class |
2292 | //===----------------------------------------------------------------------===// |
2293 | class FuncletPadInst : public Instruction { |
2294 | private: |
2295 | FuncletPadInst(const FuncletPadInst &CPI); |
2296 | |
2297 | explicit FuncletPadInst(Instruction::FuncletPadOps Op, Value *ParentPad, |
2298 | ArrayRef<Value *> Args, unsigned Values, |
2299 | const Twine &NameStr, Instruction *InsertBefore); |
2300 | explicit FuncletPadInst(Instruction::FuncletPadOps Op, Value *ParentPad, |
2301 | ArrayRef<Value *> Args, unsigned Values, |
2302 | const Twine &NameStr, BasicBlock *InsertAtEnd); |
2303 | |
2304 | void init(Value *ParentPad, ArrayRef<Value *> Args, const Twine &NameStr); |
2305 | |
2306 | protected: |
2307 | // Note: Instruction needs to be a friend here to call cloneImpl. |
2308 | friend class Instruction; |
2309 | friend class CatchPadInst; |
2310 | friend class CleanupPadInst; |
2311 | |
2312 | FuncletPadInst *cloneImpl() const; |
2313 | |
2314 | public: |
2315 | /// Provide fast operand accessors |
2316 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(Value)public: inline Value *getOperand(unsigned) const; inline void setOperand(unsigned, Value*); inline op_iterator op_begin(); inline const_op_iterator op_begin() const; inline op_iterator op_end(); inline const_op_iterator op_end() const; protected : template <int> inline Use &Op(); template <int > inline const Use &Op() const; public: inline unsigned getNumOperands() const; |
2317 | |
2318 | /// getNumArgOperands - Return the number of funcletpad arguments. |
2319 | /// |
2320 | unsigned getNumArgOperands() const { return getNumOperands() - 1; } |
2321 | |
2322 | /// Convenience accessors |
2323 | |
2324 | /// Return the outer EH-pad this funclet is nested within. |
2325 | /// |
2326 | /// Note: This returns the associated CatchSwitchInst if this FuncletPadInst |
2327 | /// is a CatchPadInst. |
2328 | Value *getParentPad() const { return Op<-1>(); } |
2329 | void setParentPad(Value *ParentPad) { |
2330 | assert(ParentPad)(static_cast <bool> (ParentPad) ? void (0) : __assert_fail ("ParentPad", "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 2330, __extension__ __PRETTY_FUNCTION__)); |
2331 | Op<-1>() = ParentPad; |
2332 | } |
2333 | |
2334 | /// getArgOperand/setArgOperand - Return/set the i-th funcletpad argument. |
2335 | /// |
2336 | Value *getArgOperand(unsigned i) const { return getOperand(i); } |
2337 | void setArgOperand(unsigned i, Value *v) { setOperand(i, v); } |
2338 | |
2339 | /// arg_operands - iteration adapter for range-for loops. |
2340 | op_range arg_operands() { return op_range(op_begin(), op_end() - 1); } |
2341 | |
2342 | /// arg_operands - iteration adapter for range-for loops. |
2343 | const_op_range arg_operands() const { |
2344 | return const_op_range(op_begin(), op_end() - 1); |
2345 | } |
2346 | |
2347 | // Methods for support type inquiry through isa, cast, and dyn_cast: |
2348 | static bool classof(const Instruction *I) { return I->isFuncletPad(); } |
2349 | static bool classof(const Value *V) { |
2350 | return isa<Instruction>(V) && classof(cast<Instruction>(V)); |
2351 | } |
2352 | }; |
2353 | |
2354 | template <> |
2355 | struct OperandTraits<FuncletPadInst> |
2356 | : public VariadicOperandTraits<FuncletPadInst, /*MINARITY=*/1> {}; |
2357 | |
2358 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(FuncletPadInst, Value)FuncletPadInst::op_iterator FuncletPadInst::op_begin() { return OperandTraits<FuncletPadInst>::op_begin(this); } FuncletPadInst ::const_op_iterator FuncletPadInst::op_begin() const { return OperandTraits<FuncletPadInst>::op_begin(const_cast< FuncletPadInst*>(this)); } FuncletPadInst::op_iterator FuncletPadInst ::op_end() { return OperandTraits<FuncletPadInst>::op_end (this); } FuncletPadInst::const_op_iterator FuncletPadInst::op_end () const { return OperandTraits<FuncletPadInst>::op_end (const_cast<FuncletPadInst*>(this)); } Value *FuncletPadInst ::getOperand(unsigned i_nocapture) const { (static_cast <bool > (i_nocapture < OperandTraits<FuncletPadInst>::operands (this) && "getOperand() out of range!") ? void (0) : __assert_fail ("i_nocapture < OperandTraits<FuncletPadInst>::operands(this) && \"getOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 2358, __extension__ __PRETTY_FUNCTION__)); return cast_or_null <Value>( OperandTraits<FuncletPadInst>::op_begin( const_cast<FuncletPadInst*>(this))[i_nocapture].get()); } void FuncletPadInst::setOperand(unsigned i_nocapture, Value *Val_nocapture) { (static_cast <bool> (i_nocapture < OperandTraits<FuncletPadInst>::operands(this) && "setOperand() out of range!") ? void (0) : __assert_fail ("i_nocapture < OperandTraits<FuncletPadInst>::operands(this) && \"setOperand() out of range!\"" , "/build/llvm-toolchain-snapshot-13~++20210726100616+dead50d4427c/llvm/include/llvm/IR/InstrTypes.h" , 2358, __extension__ __PRETTY_FUNCTION__)); OperandTraits< FuncletPadInst>::op_begin(this)[i_nocapture] = Val_nocapture ; } unsigned FuncletPadInst::getNumOperands() const { return OperandTraits <FuncletPadInst>::operands(this); } template <int Idx_nocapture > Use &FuncletPadInst::Op() { return this->OpFrom< Idx_nocapture>(this); } template <int Idx_nocapture> const Use &FuncletPadInst::Op() const { return this-> OpFrom<Idx_nocapture>(this); } |
2359 | |
2360 | } // end namespace llvm |
2361 | |
2362 | #endif // LLVM_IR_INSTRTYPES_H |