1 | //===- Instructions.cpp - Implement the LLVM instructions -----------------===// |
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 implements all of the non-inline methods for the LLVM instruction |
10 | // classes. |
11 | // |
12 | //===----------------------------------------------------------------------===// |
13 | |
14 | #include "llvm/IR/Instructions.h" |
15 | #include "LLVMContextImpl.h" |
16 | #include "llvm/ADT/SmallBitVector.h" |
17 | #include "llvm/ADT/SmallVector.h" |
18 | #include "llvm/ADT/Twine.h" |
19 | #include "llvm/IR/Attributes.h" |
20 | #include "llvm/IR/BasicBlock.h" |
21 | #include "llvm/IR/Constant.h" |
22 | #include "llvm/IR/ConstantRange.h" |
23 | #include "llvm/IR/Constants.h" |
24 | #include "llvm/IR/DataLayout.h" |
25 | #include "llvm/IR/DerivedTypes.h" |
26 | #include "llvm/IR/Function.h" |
27 | #include "llvm/IR/InstrTypes.h" |
28 | #include "llvm/IR/Instruction.h" |
29 | #include "llvm/IR/Intrinsics.h" |
30 | #include "llvm/IR/LLVMContext.h" |
31 | #include "llvm/IR/MDBuilder.h" |
32 | #include "llvm/IR/Metadata.h" |
33 | #include "llvm/IR/Module.h" |
34 | #include "llvm/IR/Operator.h" |
35 | #include "llvm/IR/PatternMatch.h" |
36 | #include "llvm/IR/ProfDataUtils.h" |
37 | #include "llvm/IR/Type.h" |
38 | #include "llvm/IR/Value.h" |
39 | #include "llvm/Support/AtomicOrdering.h" |
40 | #include "llvm/Support/Casting.h" |
41 | #include "llvm/Support/CheckedArithmetic.h" |
42 | #include "llvm/Support/Compiler.h" |
43 | #include "llvm/Support/ErrorHandling.h" |
44 | #include "llvm/Support/KnownBits.h" |
45 | #include "llvm/Support/MathExtras.h" |
46 | #include "llvm/Support/ModRef.h" |
47 | #include "llvm/Support/TypeSize.h" |
48 | #include <algorithm> |
49 | #include <cassert> |
50 | #include <cstdint> |
51 | #include <optional> |
52 | #include <vector> |
53 | |
54 | using namespace llvm; |
55 | |
56 | static cl::opt<bool> DisableI2pP2iOpt( |
57 | "disable-i2p-p2i-opt" , cl::init(Val: false), |
58 | cl::desc("Disables inttoptr/ptrtoint roundtrip optimization" )); |
59 | |
60 | //===----------------------------------------------------------------------===// |
61 | // AllocaInst Class |
62 | //===----------------------------------------------------------------------===// |
63 | |
64 | std::optional<TypeSize> |
65 | AllocaInst::getAllocationSize(const DataLayout &DL) const { |
66 | TypeSize Size = DL.getTypeAllocSize(Ty: getAllocatedType()); |
67 | if (isArrayAllocation()) { |
68 | auto *C = dyn_cast<ConstantInt>(Val: getArraySize()); |
69 | if (!C) |
70 | return std::nullopt; |
71 | assert(!Size.isScalable() && "Array elements cannot have a scalable size" ); |
72 | auto CheckedProd = |
73 | checkedMulUnsigned(LHS: Size.getKnownMinValue(), RHS: C->getZExtValue()); |
74 | if (!CheckedProd) |
75 | return std::nullopt; |
76 | return TypeSize::getFixed(ExactSize: *CheckedProd); |
77 | } |
78 | return Size; |
79 | } |
80 | |
81 | std::optional<TypeSize> |
82 | AllocaInst::getAllocationSizeInBits(const DataLayout &DL) const { |
83 | std::optional<TypeSize> Size = getAllocationSize(DL); |
84 | if (!Size) |
85 | return std::nullopt; |
86 | auto CheckedProd = checkedMulUnsigned(LHS: Size->getKnownMinValue(), |
87 | RHS: static_cast<TypeSize::ScalarTy>(8)); |
88 | if (!CheckedProd) |
89 | return std::nullopt; |
90 | return TypeSize::get(Quantity: *CheckedProd, Scalable: Size->isScalable()); |
91 | } |
92 | |
93 | //===----------------------------------------------------------------------===// |
94 | // SelectInst Class |
95 | //===----------------------------------------------------------------------===// |
96 | |
97 | /// areInvalidOperands - Return a string if the specified operands are invalid |
98 | /// for a select operation, otherwise return null. |
99 | const char *SelectInst::areInvalidOperands(Value *Op0, Value *Op1, Value *Op2) { |
100 | if (Op1->getType() != Op2->getType()) |
101 | return "both values to select must have same type" ; |
102 | |
103 | if (Op1->getType()->isTokenTy()) |
104 | return "select values cannot have token type" ; |
105 | |
106 | if (VectorType *VT = dyn_cast<VectorType>(Val: Op0->getType())) { |
107 | // Vector select. |
108 | if (VT->getElementType() != Type::getInt1Ty(C&: Op0->getContext())) |
109 | return "vector select condition element type must be i1" ; |
110 | VectorType *ET = dyn_cast<VectorType>(Val: Op1->getType()); |
111 | if (!ET) |
112 | return "selected values for vector select must be vectors" ; |
113 | if (ET->getElementCount() != VT->getElementCount()) |
114 | return "vector select requires selected vectors to have " |
115 | "the same vector length as select condition" ; |
116 | } else if (Op0->getType() != Type::getInt1Ty(C&: Op0->getContext())) { |
117 | return "select condition must be i1 or <n x i1>" ; |
118 | } |
119 | return nullptr; |
120 | } |
121 | |
122 | //===----------------------------------------------------------------------===// |
123 | // PHINode Class |
124 | //===----------------------------------------------------------------------===// |
125 | |
126 | PHINode::PHINode(const PHINode &PN) |
127 | : Instruction(PN.getType(), Instruction::PHI, AllocMarker), |
128 | ReservedSpace(PN.getNumOperands()) { |
129 | NumUserOperands = PN.getNumOperands(); |
130 | allocHungoffUses(N: PN.getNumOperands()); |
131 | std::copy(first: PN.op_begin(), last: PN.op_end(), result: op_begin()); |
132 | copyIncomingBlocks(BBRange: make_range(x: PN.block_begin(), y: PN.block_end())); |
133 | SubclassOptionalData = PN.SubclassOptionalData; |
134 | } |
135 | |
136 | // removeIncomingValue - Remove an incoming value. This is useful if a |
137 | // predecessor basic block is deleted. |
138 | Value *PHINode::removeIncomingValue(unsigned Idx, bool DeletePHIIfEmpty) { |
139 | Value *Removed = getIncomingValue(i: Idx); |
140 | |
141 | // Move everything after this operand down. |
142 | // |
143 | // FIXME: we could just swap with the end of the list, then erase. However, |
144 | // clients might not expect this to happen. The code as it is thrashes the |
145 | // use/def lists, which is kinda lame. |
146 | std::copy(first: op_begin() + Idx + 1, last: op_end(), result: op_begin() + Idx); |
147 | copyIncomingBlocks(BBRange: drop_begin(RangeOrContainer: blocks(), N: Idx + 1), ToIdx: Idx); |
148 | |
149 | // Nuke the last value. |
150 | Op<-1>().set(nullptr); |
151 | setNumHungOffUseOperands(getNumOperands() - 1); |
152 | |
153 | // If the PHI node is dead, because it has zero entries, nuke it now. |
154 | if (getNumOperands() == 0 && DeletePHIIfEmpty) { |
155 | // If anyone is using this PHI, make them use a dummy value instead... |
156 | replaceAllUsesWith(V: PoisonValue::get(T: getType())); |
157 | eraseFromParent(); |
158 | } |
159 | return Removed; |
160 | } |
161 | |
162 | void PHINode::removeIncomingValueIf(function_ref<bool(unsigned)> Predicate, |
163 | bool DeletePHIIfEmpty) { |
164 | SmallDenseSet<unsigned> RemoveIndices; |
165 | for (unsigned Idx = 0; Idx < getNumIncomingValues(); ++Idx) |
166 | if (Predicate(Idx)) |
167 | RemoveIndices.insert(V: Idx); |
168 | |
169 | if (RemoveIndices.empty()) |
170 | return; |
171 | |
172 | // Remove operands. |
173 | auto NewOpEnd = remove_if(Range: operands(), P: [&](Use &U) { |
174 | return RemoveIndices.contains(V: U.getOperandNo()); |
175 | }); |
176 | for (Use &U : make_range(x: NewOpEnd, y: op_end())) |
177 | U.set(nullptr); |
178 | |
179 | // Remove incoming blocks. |
180 | (void)std::remove_if(first: const_cast<block_iterator>(block_begin()), |
181 | last: const_cast<block_iterator>(block_end()), pred: [&](BasicBlock *&BB) { |
182 | return RemoveIndices.contains(V: &BB - block_begin()); |
183 | }); |
184 | |
185 | setNumHungOffUseOperands(getNumOperands() - RemoveIndices.size()); |
186 | |
187 | // If the PHI node is dead, because it has zero entries, nuke it now. |
188 | if (getNumOperands() == 0 && DeletePHIIfEmpty) { |
189 | // If anyone is using this PHI, make them use a dummy value instead... |
190 | replaceAllUsesWith(V: PoisonValue::get(T: getType())); |
191 | eraseFromParent(); |
192 | } |
193 | } |
194 | |
195 | /// growOperands - grow operands - This grows the operand list in response |
196 | /// to a push_back style of operation. This grows the number of ops by 1.5 |
197 | /// times. |
198 | /// |
199 | void PHINode::growOperands() { |
200 | unsigned e = getNumOperands(); |
201 | unsigned NumOps = e + e / 2; |
202 | if (NumOps < 2) NumOps = 2; // 2 op PHI nodes are VERY common. |
203 | |
204 | ReservedSpace = NumOps; |
205 | growHungoffUses(N: ReservedSpace, /* IsPhi */ true); |
206 | } |
207 | |
208 | /// hasConstantValue - If the specified PHI node always merges together the same |
209 | /// value, return the value, otherwise return null. |
210 | Value *PHINode::hasConstantValue() const { |
211 | // Exploit the fact that phi nodes always have at least one entry. |
212 | Value *ConstantValue = getIncomingValue(i: 0); |
213 | for (unsigned i = 1, e = getNumIncomingValues(); i != e; ++i) |
214 | if (getIncomingValue(i) != ConstantValue && getIncomingValue(i) != this) { |
215 | if (ConstantValue != this) |
216 | return nullptr; // Incoming values not all the same. |
217 | // The case where the first value is this PHI. |
218 | ConstantValue = getIncomingValue(i); |
219 | } |
220 | if (ConstantValue == this) |
221 | return PoisonValue::get(T: getType()); |
222 | return ConstantValue; |
223 | } |
224 | |
225 | /// hasConstantOrUndefValue - Whether the specified PHI node always merges |
226 | /// together the same value, assuming that undefs result in the same value as |
227 | /// non-undefs. |
228 | /// Unlike \ref hasConstantValue, this does not return a value because the |
229 | /// unique non-undef incoming value need not dominate the PHI node. |
230 | bool PHINode::hasConstantOrUndefValue() const { |
231 | Value *ConstantValue = nullptr; |
232 | for (unsigned i = 0, e = getNumIncomingValues(); i != e; ++i) { |
233 | Value *Incoming = getIncomingValue(i); |
234 | if (Incoming != this && !isa<UndefValue>(Val: Incoming)) { |
235 | if (ConstantValue && ConstantValue != Incoming) |
236 | return false; |
237 | ConstantValue = Incoming; |
238 | } |
239 | } |
240 | return true; |
241 | } |
242 | |
243 | //===----------------------------------------------------------------------===// |
244 | // LandingPadInst Implementation |
245 | //===----------------------------------------------------------------------===// |
246 | |
247 | LandingPadInst::LandingPadInst(Type *RetTy, unsigned NumReservedValues, |
248 | const Twine &NameStr, |
249 | InsertPosition InsertBefore) |
250 | : Instruction(RetTy, Instruction::LandingPad, AllocMarker, InsertBefore) { |
251 | init(NumReservedValues, NameStr); |
252 | } |
253 | |
254 | LandingPadInst::LandingPadInst(const LandingPadInst &LP) |
255 | : Instruction(LP.getType(), Instruction::LandingPad, AllocMarker), |
256 | ReservedSpace(LP.getNumOperands()) { |
257 | NumUserOperands = LP.getNumOperands(); |
258 | allocHungoffUses(N: LP.getNumOperands()); |
259 | Use *OL = getOperandList(); |
260 | const Use *InOL = LP.getOperandList(); |
261 | for (unsigned I = 0, E = ReservedSpace; I != E; ++I) |
262 | OL[I] = InOL[I]; |
263 | |
264 | setCleanup(LP.isCleanup()); |
265 | } |
266 | |
267 | LandingPadInst *LandingPadInst::Create(Type *RetTy, unsigned NumReservedClauses, |
268 | const Twine &NameStr, |
269 | InsertPosition InsertBefore) { |
270 | return new LandingPadInst(RetTy, NumReservedClauses, NameStr, InsertBefore); |
271 | } |
272 | |
273 | void LandingPadInst::init(unsigned NumReservedValues, const Twine &NameStr) { |
274 | ReservedSpace = NumReservedValues; |
275 | setNumHungOffUseOperands(0); |
276 | allocHungoffUses(N: ReservedSpace); |
277 | setName(NameStr); |
278 | setCleanup(false); |
279 | } |
280 | |
281 | /// growOperands - grow operands - This grows the operand list in response to a |
282 | /// push_back style of operation. This grows the number of ops by 2 times. |
283 | void LandingPadInst::growOperands(unsigned Size) { |
284 | unsigned e = getNumOperands(); |
285 | if (ReservedSpace >= e + Size) return; |
286 | ReservedSpace = (std::max(a: e, b: 1U) + Size / 2) * 2; |
287 | growHungoffUses(N: ReservedSpace); |
288 | } |
289 | |
290 | void LandingPadInst::addClause(Constant *Val) { |
291 | unsigned OpNo = getNumOperands(); |
292 | growOperands(Size: 1); |
293 | assert(OpNo < ReservedSpace && "Growing didn't work!" ); |
294 | setNumHungOffUseOperands(getNumOperands() + 1); |
295 | getOperandList()[OpNo] = Val; |
296 | } |
297 | |
298 | //===----------------------------------------------------------------------===// |
299 | // CallBase Implementation |
300 | //===----------------------------------------------------------------------===// |
301 | |
302 | CallBase *CallBase::Create(CallBase *CB, ArrayRef<OperandBundleDef> Bundles, |
303 | InsertPosition InsertPt) { |
304 | switch (CB->getOpcode()) { |
305 | case Instruction::Call: |
306 | return CallInst::Create(CI: cast<CallInst>(Val: CB), Bundles, InsertPt); |
307 | case Instruction::Invoke: |
308 | return InvokeInst::Create(II: cast<InvokeInst>(Val: CB), Bundles, InsertPt); |
309 | case Instruction::CallBr: |
310 | return CallBrInst::Create(CBI: cast<CallBrInst>(Val: CB), Bundles, InsertBefore: InsertPt); |
311 | default: |
312 | llvm_unreachable("Unknown CallBase sub-class!" ); |
313 | } |
314 | } |
315 | |
316 | CallBase *CallBase::Create(CallBase *CI, OperandBundleDef OpB, |
317 | InsertPosition InsertPt) { |
318 | SmallVector<OperandBundleDef, 2> OpDefs; |
319 | for (unsigned i = 0, e = CI->getNumOperandBundles(); i < e; ++i) { |
320 | auto ChildOB = CI->getOperandBundleAt(Index: i); |
321 | if (ChildOB.getTagName() != OpB.getTag()) |
322 | OpDefs.emplace_back(Args&: ChildOB); |
323 | } |
324 | OpDefs.emplace_back(Args&: OpB); |
325 | return CallBase::Create(CB: CI, Bundles: OpDefs, InsertPt); |
326 | } |
327 | |
328 | Function *CallBase::getCaller() { return getParent()->getParent(); } |
329 | |
330 | unsigned CallBase::getNumSubclassExtraOperandsDynamic() const { |
331 | assert(getOpcode() == Instruction::CallBr && "Unexpected opcode!" ); |
332 | return cast<CallBrInst>(Val: this)->getNumIndirectDests() + 1; |
333 | } |
334 | |
335 | bool CallBase::isIndirectCall() const { |
336 | const Value *V = getCalledOperand(); |
337 | if (isa<Function>(Val: V) || isa<Constant>(Val: V)) |
338 | return false; |
339 | return !isInlineAsm(); |
340 | } |
341 | |
342 | /// Tests if this call site must be tail call optimized. Only a CallInst can |
343 | /// be tail call optimized. |
344 | bool CallBase::isMustTailCall() const { |
345 | if (auto *CI = dyn_cast<CallInst>(Val: this)) |
346 | return CI->isMustTailCall(); |
347 | return false; |
348 | } |
349 | |
350 | /// Tests if this call site is marked as a tail call. |
351 | bool CallBase::isTailCall() const { |
352 | if (auto *CI = dyn_cast<CallInst>(Val: this)) |
353 | return CI->isTailCall(); |
354 | return false; |
355 | } |
356 | |
357 | Intrinsic::ID CallBase::getIntrinsicID() const { |
358 | if (auto *F = getCalledFunction()) |
359 | return F->getIntrinsicID(); |
360 | return Intrinsic::not_intrinsic; |
361 | } |
362 | |
363 | FPClassTest CallBase::getRetNoFPClass() const { |
364 | FPClassTest Mask = Attrs.getRetNoFPClass(); |
365 | |
366 | if (const Function *F = getCalledFunction()) |
367 | Mask |= F->getAttributes().getRetNoFPClass(); |
368 | return Mask; |
369 | } |
370 | |
371 | FPClassTest CallBase::getParamNoFPClass(unsigned i) const { |
372 | FPClassTest Mask = Attrs.getParamNoFPClass(ArgNo: i); |
373 | |
374 | if (const Function *F = getCalledFunction()) |
375 | Mask |= F->getAttributes().getParamNoFPClass(ArgNo: i); |
376 | return Mask; |
377 | } |
378 | |
379 | std::optional<ConstantRange> CallBase::getRange() const { |
380 | Attribute CallAttr = Attrs.getRetAttr(Kind: Attribute::Range); |
381 | Attribute FnAttr; |
382 | if (const Function *F = getCalledFunction()) |
383 | FnAttr = F->getRetAttribute(Kind: Attribute::Range); |
384 | |
385 | if (CallAttr.isValid() && FnAttr.isValid()) |
386 | return CallAttr.getRange().intersectWith(CR: FnAttr.getRange()); |
387 | if (CallAttr.isValid()) |
388 | return CallAttr.getRange(); |
389 | if (FnAttr.isValid()) |
390 | return FnAttr.getRange(); |
391 | return std::nullopt; |
392 | } |
393 | |
394 | bool CallBase::isReturnNonNull() const { |
395 | if (hasRetAttr(Kind: Attribute::NonNull)) |
396 | return true; |
397 | |
398 | if (getRetDereferenceableBytes() > 0 && |
399 | !NullPointerIsDefined(F: getCaller(), AS: getType()->getPointerAddressSpace())) |
400 | return true; |
401 | |
402 | return false; |
403 | } |
404 | |
405 | Value *CallBase::getArgOperandWithAttribute(Attribute::AttrKind Kind) const { |
406 | unsigned Index; |
407 | |
408 | if (Attrs.hasAttrSomewhere(Kind, Index: &Index)) |
409 | return getArgOperand(i: Index - AttributeList::FirstArgIndex); |
410 | if (const Function *F = getCalledFunction()) |
411 | if (F->getAttributes().hasAttrSomewhere(Kind, Index: &Index)) |
412 | return getArgOperand(i: Index - AttributeList::FirstArgIndex); |
413 | |
414 | return nullptr; |
415 | } |
416 | |
417 | /// Determine whether the argument or parameter has the given attribute. |
418 | bool CallBase::paramHasAttr(unsigned ArgNo, Attribute::AttrKind Kind) const { |
419 | assert(ArgNo < arg_size() && "Param index out of bounds!" ); |
420 | |
421 | if (Attrs.hasParamAttr(ArgNo, Kind)) |
422 | return true; |
423 | |
424 | const Function *F = getCalledFunction(); |
425 | if (!F) |
426 | return false; |
427 | |
428 | if (!F->getAttributes().hasParamAttr(ArgNo, Kind)) |
429 | return false; |
430 | |
431 | // Take into account mod/ref by operand bundles. |
432 | switch (Kind) { |
433 | case Attribute::ReadNone: |
434 | return !hasReadingOperandBundles() && !hasClobberingOperandBundles(); |
435 | case Attribute::ReadOnly: |
436 | return !hasClobberingOperandBundles(); |
437 | case Attribute::WriteOnly: |
438 | return !hasReadingOperandBundles(); |
439 | default: |
440 | return true; |
441 | } |
442 | } |
443 | |
444 | bool CallBase::paramHasNonNullAttr(unsigned ArgNo, |
445 | bool AllowUndefOrPoison) const { |
446 | assert(getArgOperand(ArgNo)->getType()->isPointerTy() && |
447 | "Argument must be a pointer" ); |
448 | if (paramHasAttr(ArgNo, Kind: Attribute::NonNull) && |
449 | (AllowUndefOrPoison || paramHasAttr(ArgNo, Kind: Attribute::NoUndef))) |
450 | return true; |
451 | |
452 | if (paramHasAttr(ArgNo, Kind: Attribute::Dereferenceable) && |
453 | !NullPointerIsDefined( |
454 | F: getCaller(), |
455 | AS: getArgOperand(i: ArgNo)->getType()->getPointerAddressSpace())) |
456 | return true; |
457 | |
458 | return false; |
459 | } |
460 | |
461 | bool CallBase::hasFnAttrOnCalledFunction(Attribute::AttrKind Kind) const { |
462 | if (auto *F = dyn_cast<Function>(Val: getCalledOperand())) |
463 | return F->getAttributes().hasFnAttr(Kind); |
464 | |
465 | return false; |
466 | } |
467 | |
468 | bool CallBase::hasFnAttrOnCalledFunction(StringRef Kind) const { |
469 | if (auto *F = dyn_cast<Function>(Val: getCalledOperand())) |
470 | return F->getAttributes().hasFnAttr(Kind); |
471 | |
472 | return false; |
473 | } |
474 | |
475 | template <typename AK> |
476 | Attribute CallBase::getFnAttrOnCalledFunction(AK Kind) const { |
477 | if constexpr (std::is_same_v<AK, Attribute::AttrKind>) { |
478 | // getMemoryEffects() correctly combines memory effects from the call-site, |
479 | // operand bundles and function. |
480 | assert(Kind != Attribute::Memory && "Use getMemoryEffects() instead" ); |
481 | } |
482 | |
483 | if (auto *F = dyn_cast<Function>(Val: getCalledOperand())) |
484 | return F->getAttributes().getFnAttr(Kind); |
485 | |
486 | return Attribute(); |
487 | } |
488 | |
489 | template LLVM_ABI Attribute |
490 | CallBase::getFnAttrOnCalledFunction(Attribute::AttrKind Kind) const; |
491 | template LLVM_ABI Attribute |
492 | CallBase::getFnAttrOnCalledFunction(StringRef Kind) const; |
493 | |
494 | template <typename AK> |
495 | Attribute CallBase::getParamAttrOnCalledFunction(unsigned ArgNo, |
496 | AK Kind) const { |
497 | Value *V = getCalledOperand(); |
498 | |
499 | if (auto *F = dyn_cast<Function>(Val: V)) |
500 | return F->getAttributes().getParamAttr(ArgNo, Kind); |
501 | |
502 | return Attribute(); |
503 | } |
504 | template LLVM_ABI Attribute CallBase::getParamAttrOnCalledFunction( |
505 | unsigned ArgNo, Attribute::AttrKind Kind) const; |
506 | template LLVM_ABI Attribute |
507 | CallBase::getParamAttrOnCalledFunction(unsigned ArgNo, StringRef Kind) const; |
508 | |
509 | void CallBase::getOperandBundlesAsDefs( |
510 | SmallVectorImpl<OperandBundleDef> &Defs) const { |
511 | for (unsigned i = 0, e = getNumOperandBundles(); i != e; ++i) |
512 | Defs.emplace_back(Args: getOperandBundleAt(Index: i)); |
513 | } |
514 | |
515 | CallBase::op_iterator |
516 | CallBase::populateBundleOperandInfos(ArrayRef<OperandBundleDef> Bundles, |
517 | const unsigned BeginIndex) { |
518 | auto It = op_begin() + BeginIndex; |
519 | for (auto &B : Bundles) |
520 | It = std::copy(first: B.input_begin(), last: B.input_end(), result: It); |
521 | |
522 | auto *ContextImpl = getContext().pImpl; |
523 | auto BI = Bundles.begin(); |
524 | unsigned CurrentIndex = BeginIndex; |
525 | |
526 | for (auto &BOI : bundle_op_infos()) { |
527 | assert(BI != Bundles.end() && "Incorrect allocation?" ); |
528 | |
529 | BOI.Tag = ContextImpl->getOrInsertBundleTag(Tag: BI->getTag()); |
530 | BOI.Begin = CurrentIndex; |
531 | BOI.End = CurrentIndex + BI->input_size(); |
532 | CurrentIndex = BOI.End; |
533 | BI++; |
534 | } |
535 | |
536 | assert(BI == Bundles.end() && "Incorrect allocation?" ); |
537 | |
538 | return It; |
539 | } |
540 | |
541 | CallBase::BundleOpInfo &CallBase::getBundleOpInfoForOperand(unsigned OpIdx) { |
542 | /// When there isn't many bundles, we do a simple linear search. |
543 | /// Else fallback to a binary-search that use the fact that bundles usually |
544 | /// have similar number of argument to get faster convergence. |
545 | if (bundle_op_info_end() - bundle_op_info_begin() < 8) { |
546 | for (auto &BOI : bundle_op_infos()) |
547 | if (BOI.Begin <= OpIdx && OpIdx < BOI.End) |
548 | return BOI; |
549 | |
550 | llvm_unreachable("Did not find operand bundle for operand!" ); |
551 | } |
552 | |
553 | assert(OpIdx >= arg_size() && "the Idx is not in the operand bundles" ); |
554 | assert(bundle_op_info_end() - bundle_op_info_begin() > 0 && |
555 | OpIdx < std::prev(bundle_op_info_end())->End && |
556 | "The Idx isn't in the operand bundle" ); |
557 | |
558 | /// We need a decimal number below and to prevent using floating point numbers |
559 | /// we use an intergal value multiplied by this constant. |
560 | constexpr unsigned NumberScaling = 1024; |
561 | |
562 | bundle_op_iterator Begin = bundle_op_info_begin(); |
563 | bundle_op_iterator End = bundle_op_info_end(); |
564 | bundle_op_iterator Current = Begin; |
565 | |
566 | while (Begin != End) { |
567 | unsigned ScaledOperandPerBundle = |
568 | NumberScaling * (std::prev(x: End)->End - Begin->Begin) / (End - Begin); |
569 | Current = Begin + (((OpIdx - Begin->Begin) * NumberScaling) / |
570 | ScaledOperandPerBundle); |
571 | if (Current >= End) |
572 | Current = std::prev(x: End); |
573 | assert(Current < End && Current >= Begin && |
574 | "the operand bundle doesn't cover every value in the range" ); |
575 | if (OpIdx >= Current->Begin && OpIdx < Current->End) |
576 | break; |
577 | if (OpIdx >= Current->End) |
578 | Begin = Current + 1; |
579 | else |
580 | End = Current; |
581 | } |
582 | |
583 | assert(OpIdx >= Current->Begin && OpIdx < Current->End && |
584 | "the operand bundle doesn't cover every value in the range" ); |
585 | return *Current; |
586 | } |
587 | |
588 | CallBase *CallBase::addOperandBundle(CallBase *CB, uint32_t ID, |
589 | OperandBundleDef OB, |
590 | InsertPosition InsertPt) { |
591 | if (CB->getOperandBundle(ID)) |
592 | return CB; |
593 | |
594 | SmallVector<OperandBundleDef, 1> Bundles; |
595 | CB->getOperandBundlesAsDefs(Defs&: Bundles); |
596 | Bundles.push_back(Elt: OB); |
597 | return Create(CB, Bundles, InsertPt); |
598 | } |
599 | |
600 | CallBase *CallBase::removeOperandBundle(CallBase *CB, uint32_t ID, |
601 | InsertPosition InsertPt) { |
602 | SmallVector<OperandBundleDef, 1> Bundles; |
603 | bool CreateNew = false; |
604 | |
605 | for (unsigned I = 0, E = CB->getNumOperandBundles(); I != E; ++I) { |
606 | auto Bundle = CB->getOperandBundleAt(Index: I); |
607 | if (Bundle.getTagID() == ID) { |
608 | CreateNew = true; |
609 | continue; |
610 | } |
611 | Bundles.emplace_back(Args&: Bundle); |
612 | } |
613 | |
614 | return CreateNew ? Create(CB, Bundles, InsertPt) : CB; |
615 | } |
616 | |
617 | bool CallBase::hasReadingOperandBundles() const { |
618 | // Implementation note: this is a conservative implementation of operand |
619 | // bundle semantics, where *any* non-assume operand bundle (other than |
620 | // ptrauth) forces a callsite to be at least readonly. |
621 | return hasOperandBundlesOtherThan(IDs: {LLVMContext::OB_ptrauth, |
622 | LLVMContext::OB_kcfi, |
623 | LLVMContext::OB_convergencectrl}) && |
624 | getIntrinsicID() != Intrinsic::assume; |
625 | } |
626 | |
627 | bool CallBase::hasClobberingOperandBundles() const { |
628 | return hasOperandBundlesOtherThan( |
629 | IDs: {LLVMContext::OB_deopt, LLVMContext::OB_funclet, |
630 | LLVMContext::OB_ptrauth, LLVMContext::OB_kcfi, |
631 | LLVMContext::OB_convergencectrl}) && |
632 | getIntrinsicID() != Intrinsic::assume; |
633 | } |
634 | |
635 | MemoryEffects CallBase::getMemoryEffects() const { |
636 | MemoryEffects ME = getAttributes().getMemoryEffects(); |
637 | if (auto *Fn = dyn_cast<Function>(Val: getCalledOperand())) { |
638 | MemoryEffects FnME = Fn->getMemoryEffects(); |
639 | if (hasOperandBundles()) { |
640 | // TODO: Add a method to get memory effects for operand bundles instead. |
641 | if (hasReadingOperandBundles()) |
642 | FnME |= MemoryEffects::readOnly(); |
643 | if (hasClobberingOperandBundles()) |
644 | FnME |= MemoryEffects::writeOnly(); |
645 | } |
646 | if (isVolatile()) { |
647 | // Volatile operations also access inaccessible memory. |
648 | FnME |= MemoryEffects::inaccessibleMemOnly(); |
649 | } |
650 | ME &= FnME; |
651 | } |
652 | return ME; |
653 | } |
654 | void CallBase::setMemoryEffects(MemoryEffects ME) { |
655 | addFnAttr(Attr: Attribute::getWithMemoryEffects(Context&: getContext(), ME)); |
656 | } |
657 | |
658 | /// Determine if the function does not access memory. |
659 | bool CallBase::doesNotAccessMemory() const { |
660 | return getMemoryEffects().doesNotAccessMemory(); |
661 | } |
662 | void CallBase::setDoesNotAccessMemory() { |
663 | setMemoryEffects(MemoryEffects::none()); |
664 | } |
665 | |
666 | /// Determine if the function does not access or only reads memory. |
667 | bool CallBase::onlyReadsMemory() const { |
668 | return getMemoryEffects().onlyReadsMemory(); |
669 | } |
670 | void CallBase::setOnlyReadsMemory() { |
671 | setMemoryEffects(getMemoryEffects() & MemoryEffects::readOnly()); |
672 | } |
673 | |
674 | /// Determine if the function does not access or only writes memory. |
675 | bool CallBase::onlyWritesMemory() const { |
676 | return getMemoryEffects().onlyWritesMemory(); |
677 | } |
678 | void CallBase::setOnlyWritesMemory() { |
679 | setMemoryEffects(getMemoryEffects() & MemoryEffects::writeOnly()); |
680 | } |
681 | |
682 | /// Determine if the call can access memmory only using pointers based |
683 | /// on its arguments. |
684 | bool CallBase::onlyAccessesArgMemory() const { |
685 | return getMemoryEffects().onlyAccessesArgPointees(); |
686 | } |
687 | void CallBase::setOnlyAccessesArgMemory() { |
688 | setMemoryEffects(getMemoryEffects() & MemoryEffects::argMemOnly()); |
689 | } |
690 | |
691 | /// Determine if the function may only access memory that is |
692 | /// inaccessible from the IR. |
693 | bool CallBase::onlyAccessesInaccessibleMemory() const { |
694 | return getMemoryEffects().onlyAccessesInaccessibleMem(); |
695 | } |
696 | void CallBase::setOnlyAccessesInaccessibleMemory() { |
697 | setMemoryEffects(getMemoryEffects() & MemoryEffects::inaccessibleMemOnly()); |
698 | } |
699 | |
700 | /// Determine if the function may only access memory that is |
701 | /// either inaccessible from the IR or pointed to by its arguments. |
702 | bool CallBase::onlyAccessesInaccessibleMemOrArgMem() const { |
703 | return getMemoryEffects().onlyAccessesInaccessibleOrArgMem(); |
704 | } |
705 | void CallBase::setOnlyAccessesInaccessibleMemOrArgMem() { |
706 | setMemoryEffects(getMemoryEffects() & |
707 | MemoryEffects::inaccessibleOrArgMemOnly()); |
708 | } |
709 | |
710 | CaptureInfo CallBase::getCaptureInfo(unsigned OpNo) const { |
711 | if (OpNo < arg_size()) { |
712 | // If the argument is passed byval, the callee does not have access to the |
713 | // original pointer and thus cannot capture it. |
714 | if (isByValArgument(ArgNo: OpNo)) |
715 | return CaptureInfo::none(); |
716 | |
717 | CaptureInfo CI = getParamAttributes(ArgNo: OpNo).getCaptureInfo(); |
718 | if (auto *Fn = dyn_cast<Function>(Val: getCalledOperand())) |
719 | CI &= Fn->getAttributes().getParamAttrs(ArgNo: OpNo).getCaptureInfo(); |
720 | return CI; |
721 | } |
722 | |
723 | // deopt operand bundles are captures(none) |
724 | auto &BOI = getBundleOpInfoForOperand(OpIdx: OpNo); |
725 | auto OBU = operandBundleFromBundleOpInfo(BOI); |
726 | return OBU.isDeoptOperandBundle() ? CaptureInfo::none() : CaptureInfo::all(); |
727 | } |
728 | |
729 | bool CallBase::hasArgumentWithAdditionalReturnCaptureComponents() const { |
730 | for (unsigned I = 0, E = arg_size(); I < E; ++I) { |
731 | if (!getArgOperand(i: I)->getType()->isPointerTy()) |
732 | continue; |
733 | |
734 | CaptureInfo CI = getParamAttributes(ArgNo: I).getCaptureInfo(); |
735 | if (auto *Fn = dyn_cast<Function>(Val: getCalledOperand())) |
736 | CI &= Fn->getAttributes().getParamAttrs(ArgNo: I).getCaptureInfo(); |
737 | if (capturesAnything(CC: CI.getRetComponents() & ~CI.getOtherComponents())) |
738 | return true; |
739 | } |
740 | return false; |
741 | } |
742 | |
743 | //===----------------------------------------------------------------------===// |
744 | // CallInst Implementation |
745 | //===----------------------------------------------------------------------===// |
746 | |
747 | void CallInst::init(FunctionType *FTy, Value *Func, ArrayRef<Value *> Args, |
748 | ArrayRef<OperandBundleDef> Bundles, const Twine &NameStr) { |
749 | this->FTy = FTy; |
750 | assert(getNumOperands() == Args.size() + CountBundleInputs(Bundles) + 1 && |
751 | "NumOperands not set up?" ); |
752 | |
753 | #ifndef NDEBUG |
754 | assert((Args.size() == FTy->getNumParams() || |
755 | (FTy->isVarArg() && Args.size() > FTy->getNumParams())) && |
756 | "Calling a function with bad signature!" ); |
757 | |
758 | for (unsigned i = 0; i != Args.size(); ++i) |
759 | assert((i >= FTy->getNumParams() || |
760 | FTy->getParamType(i) == Args[i]->getType()) && |
761 | "Calling a function with a bad signature!" ); |
762 | #endif |
763 | |
764 | // Set operands in order of their index to match use-list-order |
765 | // prediction. |
766 | llvm::copy(Range&: Args, Out: op_begin()); |
767 | setCalledOperand(Func); |
768 | |
769 | auto It = populateBundleOperandInfos(Bundles, BeginIndex: Args.size()); |
770 | (void)It; |
771 | assert(It + 1 == op_end() && "Should add up!" ); |
772 | |
773 | setName(NameStr); |
774 | } |
775 | |
776 | void CallInst::init(FunctionType *FTy, Value *Func, const Twine &NameStr) { |
777 | this->FTy = FTy; |
778 | assert(getNumOperands() == 1 && "NumOperands not set up?" ); |
779 | setCalledOperand(Func); |
780 | |
781 | assert(FTy->getNumParams() == 0 && "Calling a function with bad signature" ); |
782 | |
783 | setName(NameStr); |
784 | } |
785 | |
786 | CallInst::CallInst(FunctionType *Ty, Value *Func, const Twine &Name, |
787 | AllocInfo AllocInfo, InsertPosition InsertBefore) |
788 | : CallBase(Ty->getReturnType(), Instruction::Call, AllocInfo, |
789 | InsertBefore) { |
790 | init(FTy: Ty, Func, NameStr: Name); |
791 | } |
792 | |
793 | CallInst::CallInst(const CallInst &CI, AllocInfo AllocInfo) |
794 | : CallBase(CI.Attrs, CI.FTy, CI.getType(), Instruction::Call, AllocInfo) { |
795 | assert(getNumOperands() == CI.getNumOperands() && |
796 | "Wrong number of operands allocated" ); |
797 | setTailCallKind(CI.getTailCallKind()); |
798 | setCallingConv(CI.getCallingConv()); |
799 | |
800 | std::copy(first: CI.op_begin(), last: CI.op_end(), result: op_begin()); |
801 | std::copy(first: CI.bundle_op_info_begin(), last: CI.bundle_op_info_end(), |
802 | result: bundle_op_info_begin()); |
803 | SubclassOptionalData = CI.SubclassOptionalData; |
804 | } |
805 | |
806 | CallInst *CallInst::Create(CallInst *CI, ArrayRef<OperandBundleDef> OpB, |
807 | InsertPosition InsertPt) { |
808 | std::vector<Value *> Args(CI->arg_begin(), CI->arg_end()); |
809 | |
810 | auto *NewCI = CallInst::Create(Ty: CI->getFunctionType(), Func: CI->getCalledOperand(), |
811 | Args, Bundles: OpB, NameStr: CI->getName(), InsertBefore: InsertPt); |
812 | NewCI->setTailCallKind(CI->getTailCallKind()); |
813 | NewCI->setCallingConv(CI->getCallingConv()); |
814 | NewCI->SubclassOptionalData = CI->SubclassOptionalData; |
815 | NewCI->setAttributes(CI->getAttributes()); |
816 | NewCI->setDebugLoc(CI->getDebugLoc()); |
817 | return NewCI; |
818 | } |
819 | |
820 | // Update profile weight for call instruction by scaling it using the ratio |
821 | // of S/T. The meaning of "branch_weights" meta data for call instruction is |
822 | // transfered to represent call count. |
823 | void CallInst::updateProfWeight(uint64_t S, uint64_t T) { |
824 | if (T == 0) { |
825 | LLVM_DEBUG(dbgs() << "Attempting to update profile weights will result in " |
826 | "div by 0. Ignoring. Likely the function " |
827 | << getParent()->getParent()->getName() |
828 | << " has 0 entry count, and contains call instructions " |
829 | "with non-zero prof info." ); |
830 | return; |
831 | } |
832 | scaleProfData(I&: *this, S, T); |
833 | } |
834 | |
835 | //===----------------------------------------------------------------------===// |
836 | // InvokeInst Implementation |
837 | //===----------------------------------------------------------------------===// |
838 | |
839 | void InvokeInst::init(FunctionType *FTy, Value *Fn, BasicBlock *IfNormal, |
840 | BasicBlock *IfException, ArrayRef<Value *> Args, |
841 | ArrayRef<OperandBundleDef> Bundles, |
842 | const Twine &NameStr) { |
843 | this->FTy = FTy; |
844 | |
845 | assert(getNumOperands() == |
846 | ComputeNumOperands(Args.size(), CountBundleInputs(Bundles)) && |
847 | "NumOperands not set up?" ); |
848 | |
849 | #ifndef NDEBUG |
850 | assert(((Args.size() == FTy->getNumParams()) || |
851 | (FTy->isVarArg() && Args.size() > FTy->getNumParams())) && |
852 | "Invoking a function with bad signature" ); |
853 | |
854 | for (unsigned i = 0, e = Args.size(); i != e; i++) |
855 | assert((i >= FTy->getNumParams() || |
856 | FTy->getParamType(i) == Args[i]->getType()) && |
857 | "Invoking a function with a bad signature!" ); |
858 | #endif |
859 | |
860 | // Set operands in order of their index to match use-list-order |
861 | // prediction. |
862 | llvm::copy(Range&: Args, Out: op_begin()); |
863 | setNormalDest(IfNormal); |
864 | setUnwindDest(IfException); |
865 | setCalledOperand(Fn); |
866 | |
867 | auto It = populateBundleOperandInfos(Bundles, BeginIndex: Args.size()); |
868 | (void)It; |
869 | assert(It + 3 == op_end() && "Should add up!" ); |
870 | |
871 | setName(NameStr); |
872 | } |
873 | |
874 | InvokeInst::InvokeInst(const InvokeInst &II, AllocInfo AllocInfo) |
875 | : CallBase(II.Attrs, II.FTy, II.getType(), Instruction::Invoke, AllocInfo) { |
876 | assert(getNumOperands() == II.getNumOperands() && |
877 | "Wrong number of operands allocated" ); |
878 | setCallingConv(II.getCallingConv()); |
879 | std::copy(first: II.op_begin(), last: II.op_end(), result: op_begin()); |
880 | std::copy(first: II.bundle_op_info_begin(), last: II.bundle_op_info_end(), |
881 | result: bundle_op_info_begin()); |
882 | SubclassOptionalData = II.SubclassOptionalData; |
883 | } |
884 | |
885 | InvokeInst *InvokeInst::Create(InvokeInst *II, ArrayRef<OperandBundleDef> OpB, |
886 | InsertPosition InsertPt) { |
887 | std::vector<Value *> Args(II->arg_begin(), II->arg_end()); |
888 | |
889 | auto *NewII = InvokeInst::Create( |
890 | Ty: II->getFunctionType(), Func: II->getCalledOperand(), IfNormal: II->getNormalDest(), |
891 | IfException: II->getUnwindDest(), Args, Bundles: OpB, NameStr: II->getName(), InsertBefore: InsertPt); |
892 | NewII->setCallingConv(II->getCallingConv()); |
893 | NewII->SubclassOptionalData = II->SubclassOptionalData; |
894 | NewII->setAttributes(II->getAttributes()); |
895 | NewII->setDebugLoc(II->getDebugLoc()); |
896 | return NewII; |
897 | } |
898 | |
899 | LandingPadInst *InvokeInst::getLandingPadInst() const { |
900 | return cast<LandingPadInst>(Val: getUnwindDest()->getFirstNonPHIIt()); |
901 | } |
902 | |
903 | void InvokeInst::updateProfWeight(uint64_t S, uint64_t T) { |
904 | if (T == 0) { |
905 | LLVM_DEBUG(dbgs() << "Attempting to update profile weights will result in " |
906 | "div by 0. Ignoring. Likely the function " |
907 | << getParent()->getParent()->getName() |
908 | << " has 0 entry count, and contains call instructions " |
909 | "with non-zero prof info." ); |
910 | return; |
911 | } |
912 | scaleProfData(I&: *this, S, T); |
913 | } |
914 | |
915 | //===----------------------------------------------------------------------===// |
916 | // CallBrInst Implementation |
917 | //===----------------------------------------------------------------------===// |
918 | |
919 | void CallBrInst::init(FunctionType *FTy, Value *Fn, BasicBlock *Fallthrough, |
920 | ArrayRef<BasicBlock *> IndirectDests, |
921 | ArrayRef<Value *> Args, |
922 | ArrayRef<OperandBundleDef> Bundles, |
923 | const Twine &NameStr) { |
924 | this->FTy = FTy; |
925 | |
926 | assert(getNumOperands() == ComputeNumOperands(Args.size(), |
927 | IndirectDests.size(), |
928 | CountBundleInputs(Bundles)) && |
929 | "NumOperands not set up?" ); |
930 | |
931 | #ifndef NDEBUG |
932 | assert(((Args.size() == FTy->getNumParams()) || |
933 | (FTy->isVarArg() && Args.size() > FTy->getNumParams())) && |
934 | "Calling a function with bad signature" ); |
935 | |
936 | for (unsigned i = 0, e = Args.size(); i != e; i++) |
937 | assert((i >= FTy->getNumParams() || |
938 | FTy->getParamType(i) == Args[i]->getType()) && |
939 | "Calling a function with a bad signature!" ); |
940 | #endif |
941 | |
942 | // Set operands in order of their index to match use-list-order |
943 | // prediction. |
944 | llvm::copy(Range&: Args, Out: op_begin()); |
945 | NumIndirectDests = IndirectDests.size(); |
946 | setDefaultDest(Fallthrough); |
947 | for (unsigned i = 0; i != NumIndirectDests; ++i) |
948 | setIndirectDest(i, B: IndirectDests[i]); |
949 | setCalledOperand(Fn); |
950 | |
951 | auto It = populateBundleOperandInfos(Bundles, BeginIndex: Args.size()); |
952 | (void)It; |
953 | assert(It + 2 + IndirectDests.size() == op_end() && "Should add up!" ); |
954 | |
955 | setName(NameStr); |
956 | } |
957 | |
958 | CallBrInst::CallBrInst(const CallBrInst &CBI, AllocInfo AllocInfo) |
959 | : CallBase(CBI.Attrs, CBI.FTy, CBI.getType(), Instruction::CallBr, |
960 | AllocInfo) { |
961 | assert(getNumOperands() == CBI.getNumOperands() && |
962 | "Wrong number of operands allocated" ); |
963 | setCallingConv(CBI.getCallingConv()); |
964 | std::copy(first: CBI.op_begin(), last: CBI.op_end(), result: op_begin()); |
965 | std::copy(first: CBI.bundle_op_info_begin(), last: CBI.bundle_op_info_end(), |
966 | result: bundle_op_info_begin()); |
967 | SubclassOptionalData = CBI.SubclassOptionalData; |
968 | NumIndirectDests = CBI.NumIndirectDests; |
969 | } |
970 | |
971 | CallBrInst *CallBrInst::Create(CallBrInst *CBI, ArrayRef<OperandBundleDef> OpB, |
972 | InsertPosition InsertPt) { |
973 | std::vector<Value *> Args(CBI->arg_begin(), CBI->arg_end()); |
974 | |
975 | auto *NewCBI = CallBrInst::Create( |
976 | Ty: CBI->getFunctionType(), Func: CBI->getCalledOperand(), DefaultDest: CBI->getDefaultDest(), |
977 | IndirectDests: CBI->getIndirectDests(), Args, Bundles: OpB, NameStr: CBI->getName(), InsertBefore: InsertPt); |
978 | NewCBI->setCallingConv(CBI->getCallingConv()); |
979 | NewCBI->SubclassOptionalData = CBI->SubclassOptionalData; |
980 | NewCBI->setAttributes(CBI->getAttributes()); |
981 | NewCBI->setDebugLoc(CBI->getDebugLoc()); |
982 | NewCBI->NumIndirectDests = CBI->NumIndirectDests; |
983 | return NewCBI; |
984 | } |
985 | |
986 | //===----------------------------------------------------------------------===// |
987 | // ReturnInst Implementation |
988 | //===----------------------------------------------------------------------===// |
989 | |
990 | ReturnInst::ReturnInst(const ReturnInst &RI, AllocInfo AllocInfo) |
991 | : Instruction(Type::getVoidTy(C&: RI.getContext()), Instruction::Ret, |
992 | AllocInfo) { |
993 | assert(getNumOperands() == RI.getNumOperands() && |
994 | "Wrong number of operands allocated" ); |
995 | if (RI.getNumOperands()) |
996 | Op<0>() = RI.Op<0>(); |
997 | SubclassOptionalData = RI.SubclassOptionalData; |
998 | } |
999 | |
1000 | ReturnInst::ReturnInst(LLVMContext &C, Value *retVal, AllocInfo AllocInfo, |
1001 | InsertPosition InsertBefore) |
1002 | : Instruction(Type::getVoidTy(C), Instruction::Ret, AllocInfo, |
1003 | InsertBefore) { |
1004 | if (retVal) |
1005 | Op<0>() = retVal; |
1006 | } |
1007 | |
1008 | //===----------------------------------------------------------------------===// |
1009 | // ResumeInst Implementation |
1010 | //===----------------------------------------------------------------------===// |
1011 | |
1012 | ResumeInst::ResumeInst(const ResumeInst &RI) |
1013 | : Instruction(Type::getVoidTy(C&: RI.getContext()), Instruction::Resume, |
1014 | AllocMarker) { |
1015 | Op<0>() = RI.Op<0>(); |
1016 | } |
1017 | |
1018 | ResumeInst::ResumeInst(Value *Exn, InsertPosition InsertBefore) |
1019 | : Instruction(Type::getVoidTy(C&: Exn->getContext()), Instruction::Resume, |
1020 | AllocMarker, InsertBefore) { |
1021 | Op<0>() = Exn; |
1022 | } |
1023 | |
1024 | //===----------------------------------------------------------------------===// |
1025 | // CleanupReturnInst Implementation |
1026 | //===----------------------------------------------------------------------===// |
1027 | |
1028 | CleanupReturnInst::CleanupReturnInst(const CleanupReturnInst &CRI, |
1029 | AllocInfo AllocInfo) |
1030 | : Instruction(CRI.getType(), Instruction::CleanupRet, AllocInfo) { |
1031 | assert(getNumOperands() == CRI.getNumOperands() && |
1032 | "Wrong number of operands allocated" ); |
1033 | setSubclassData<Instruction::OpaqueField>( |
1034 | CRI.getSubclassData<Instruction::OpaqueField>()); |
1035 | Op<0>() = CRI.Op<0>(); |
1036 | if (CRI.hasUnwindDest()) |
1037 | Op<1>() = CRI.Op<1>(); |
1038 | } |
1039 | |
1040 | void CleanupReturnInst::init(Value *CleanupPad, BasicBlock *UnwindBB) { |
1041 | if (UnwindBB) |
1042 | setSubclassData<UnwindDestField>(true); |
1043 | |
1044 | Op<0>() = CleanupPad; |
1045 | if (UnwindBB) |
1046 | Op<1>() = UnwindBB; |
1047 | } |
1048 | |
1049 | CleanupReturnInst::CleanupReturnInst(Value *CleanupPad, BasicBlock *UnwindBB, |
1050 | AllocInfo AllocInfo, |
1051 | InsertPosition InsertBefore) |
1052 | : Instruction(Type::getVoidTy(C&: CleanupPad->getContext()), |
1053 | Instruction::CleanupRet, AllocInfo, InsertBefore) { |
1054 | init(CleanupPad, UnwindBB); |
1055 | } |
1056 | |
1057 | //===----------------------------------------------------------------------===// |
1058 | // CatchReturnInst Implementation |
1059 | //===----------------------------------------------------------------------===// |
1060 | void CatchReturnInst::init(Value *CatchPad, BasicBlock *BB) { |
1061 | Op<0>() = CatchPad; |
1062 | Op<1>() = BB; |
1063 | } |
1064 | |
1065 | CatchReturnInst::CatchReturnInst(const CatchReturnInst &CRI) |
1066 | : Instruction(Type::getVoidTy(C&: CRI.getContext()), Instruction::CatchRet, |
1067 | AllocMarker) { |
1068 | Op<0>() = CRI.Op<0>(); |
1069 | Op<1>() = CRI.Op<1>(); |
1070 | } |
1071 | |
1072 | CatchReturnInst::CatchReturnInst(Value *CatchPad, BasicBlock *BB, |
1073 | InsertPosition InsertBefore) |
1074 | : Instruction(Type::getVoidTy(C&: BB->getContext()), Instruction::CatchRet, |
1075 | AllocMarker, InsertBefore) { |
1076 | init(CatchPad, BB); |
1077 | } |
1078 | |
1079 | //===----------------------------------------------------------------------===// |
1080 | // CatchSwitchInst Implementation |
1081 | //===----------------------------------------------------------------------===// |
1082 | |
1083 | CatchSwitchInst::CatchSwitchInst(Value *ParentPad, BasicBlock *UnwindDest, |
1084 | unsigned NumReservedValues, |
1085 | const Twine &NameStr, |
1086 | InsertPosition InsertBefore) |
1087 | : Instruction(ParentPad->getType(), Instruction::CatchSwitch, AllocMarker, |
1088 | InsertBefore) { |
1089 | if (UnwindDest) |
1090 | ++NumReservedValues; |
1091 | init(ParentPad, UnwindDest, NumReserved: NumReservedValues + 1); |
1092 | setName(NameStr); |
1093 | } |
1094 | |
1095 | CatchSwitchInst::CatchSwitchInst(const CatchSwitchInst &CSI) |
1096 | : Instruction(CSI.getType(), Instruction::CatchSwitch, AllocMarker) { |
1097 | NumUserOperands = CSI.NumUserOperands; |
1098 | init(ParentPad: CSI.getParentPad(), UnwindDest: CSI.getUnwindDest(), NumReserved: CSI.getNumOperands()); |
1099 | setNumHungOffUseOperands(ReservedSpace); |
1100 | Use *OL = getOperandList(); |
1101 | const Use *InOL = CSI.getOperandList(); |
1102 | for (unsigned I = 1, E = ReservedSpace; I != E; ++I) |
1103 | OL[I] = InOL[I]; |
1104 | } |
1105 | |
1106 | void CatchSwitchInst::init(Value *ParentPad, BasicBlock *UnwindDest, |
1107 | unsigned NumReservedValues) { |
1108 | assert(ParentPad && NumReservedValues); |
1109 | |
1110 | ReservedSpace = NumReservedValues; |
1111 | setNumHungOffUseOperands(UnwindDest ? 2 : 1); |
1112 | allocHungoffUses(N: ReservedSpace); |
1113 | |
1114 | Op<0>() = ParentPad; |
1115 | if (UnwindDest) { |
1116 | setSubclassData<UnwindDestField>(true); |
1117 | setUnwindDest(UnwindDest); |
1118 | } |
1119 | } |
1120 | |
1121 | /// growOperands - grow operands - This grows the operand list in response to a |
1122 | /// push_back style of operation. This grows the number of ops by 2 times. |
1123 | void CatchSwitchInst::growOperands(unsigned Size) { |
1124 | unsigned NumOperands = getNumOperands(); |
1125 | assert(NumOperands >= 1); |
1126 | if (ReservedSpace >= NumOperands + Size) |
1127 | return; |
1128 | ReservedSpace = (NumOperands + Size / 2) * 2; |
1129 | growHungoffUses(N: ReservedSpace); |
1130 | } |
1131 | |
1132 | void CatchSwitchInst::addHandler(BasicBlock *Handler) { |
1133 | unsigned OpNo = getNumOperands(); |
1134 | growOperands(Size: 1); |
1135 | assert(OpNo < ReservedSpace && "Growing didn't work!" ); |
1136 | setNumHungOffUseOperands(getNumOperands() + 1); |
1137 | getOperandList()[OpNo] = Handler; |
1138 | } |
1139 | |
1140 | void CatchSwitchInst::removeHandler(handler_iterator HI) { |
1141 | // Move all subsequent handlers up one. |
1142 | Use *EndDst = op_end() - 1; |
1143 | for (Use *CurDst = HI.getCurrent(); CurDst != EndDst; ++CurDst) |
1144 | *CurDst = *(CurDst + 1); |
1145 | // Null out the last handler use. |
1146 | *EndDst = nullptr; |
1147 | |
1148 | setNumHungOffUseOperands(getNumOperands() - 1); |
1149 | } |
1150 | |
1151 | //===----------------------------------------------------------------------===// |
1152 | // FuncletPadInst Implementation |
1153 | //===----------------------------------------------------------------------===// |
1154 | void FuncletPadInst::init(Value *ParentPad, ArrayRef<Value *> Args, |
1155 | const Twine &NameStr) { |
1156 | assert(getNumOperands() == 1 + Args.size() && "NumOperands not set up?" ); |
1157 | llvm::copy(Range&: Args, Out: op_begin()); |
1158 | setParentPad(ParentPad); |
1159 | setName(NameStr); |
1160 | } |
1161 | |
1162 | FuncletPadInst::FuncletPadInst(const FuncletPadInst &FPI, AllocInfo AllocInfo) |
1163 | : Instruction(FPI.getType(), FPI.getOpcode(), AllocInfo) { |
1164 | assert(getNumOperands() == FPI.getNumOperands() && |
1165 | "Wrong number of operands allocated" ); |
1166 | std::copy(first: FPI.op_begin(), last: FPI.op_end(), result: op_begin()); |
1167 | setParentPad(FPI.getParentPad()); |
1168 | } |
1169 | |
1170 | FuncletPadInst::FuncletPadInst(Instruction::FuncletPadOps Op, Value *ParentPad, |
1171 | ArrayRef<Value *> Args, AllocInfo AllocInfo, |
1172 | const Twine &NameStr, |
1173 | InsertPosition InsertBefore) |
1174 | : Instruction(ParentPad->getType(), Op, AllocInfo, InsertBefore) { |
1175 | init(ParentPad, Args, NameStr); |
1176 | } |
1177 | |
1178 | //===----------------------------------------------------------------------===// |
1179 | // UnreachableInst Implementation |
1180 | //===----------------------------------------------------------------------===// |
1181 | |
1182 | UnreachableInst::UnreachableInst(LLVMContext &Context, |
1183 | InsertPosition InsertBefore) |
1184 | : Instruction(Type::getVoidTy(C&: Context), Instruction::Unreachable, |
1185 | AllocMarker, InsertBefore) {} |
1186 | |
1187 | //===----------------------------------------------------------------------===// |
1188 | // BranchInst Implementation |
1189 | //===----------------------------------------------------------------------===// |
1190 | |
1191 | void BranchInst::AssertOK() { |
1192 | if (isConditional()) |
1193 | assert(getCondition()->getType()->isIntegerTy(1) && |
1194 | "May only branch on boolean predicates!" ); |
1195 | } |
1196 | |
1197 | BranchInst::BranchInst(BasicBlock *IfTrue, AllocInfo AllocInfo, |
1198 | InsertPosition InsertBefore) |
1199 | : Instruction(Type::getVoidTy(C&: IfTrue->getContext()), Instruction::Br, |
1200 | AllocInfo, InsertBefore) { |
1201 | assert(IfTrue && "Branch destination may not be null!" ); |
1202 | Op<-1>() = IfTrue; |
1203 | } |
1204 | |
1205 | BranchInst::BranchInst(BasicBlock *IfTrue, BasicBlock *IfFalse, Value *Cond, |
1206 | AllocInfo AllocInfo, InsertPosition InsertBefore) |
1207 | : Instruction(Type::getVoidTy(C&: IfTrue->getContext()), Instruction::Br, |
1208 | AllocInfo, InsertBefore) { |
1209 | // Assign in order of operand index to make use-list order predictable. |
1210 | Op<-3>() = Cond; |
1211 | Op<-2>() = IfFalse; |
1212 | Op<-1>() = IfTrue; |
1213 | #ifndef NDEBUG |
1214 | AssertOK(); |
1215 | #endif |
1216 | } |
1217 | |
1218 | BranchInst::BranchInst(const BranchInst &BI, AllocInfo AllocInfo) |
1219 | : Instruction(Type::getVoidTy(C&: BI.getContext()), Instruction::Br, |
1220 | AllocInfo) { |
1221 | assert(getNumOperands() == BI.getNumOperands() && |
1222 | "Wrong number of operands allocated" ); |
1223 | // Assign in order of operand index to make use-list order predictable. |
1224 | if (BI.getNumOperands() != 1) { |
1225 | assert(BI.getNumOperands() == 3 && "BR can have 1 or 3 operands!" ); |
1226 | Op<-3>() = BI.Op<-3>(); |
1227 | Op<-2>() = BI.Op<-2>(); |
1228 | } |
1229 | Op<-1>() = BI.Op<-1>(); |
1230 | SubclassOptionalData = BI.SubclassOptionalData; |
1231 | } |
1232 | |
1233 | void BranchInst::swapSuccessors() { |
1234 | assert(isConditional() && |
1235 | "Cannot swap successors of an unconditional branch" ); |
1236 | Op<-1>().swap(RHS&: Op<-2>()); |
1237 | |
1238 | // Update profile metadata if present and it matches our structural |
1239 | // expectations. |
1240 | swapProfMetadata(); |
1241 | } |
1242 | |
1243 | //===----------------------------------------------------------------------===// |
1244 | // AllocaInst Implementation |
1245 | //===----------------------------------------------------------------------===// |
1246 | |
1247 | static Value *getAISize(LLVMContext &Context, Value *Amt) { |
1248 | if (!Amt) |
1249 | Amt = ConstantInt::get(Ty: Type::getInt32Ty(C&: Context), V: 1); |
1250 | else { |
1251 | assert(!isa<BasicBlock>(Amt) && |
1252 | "Passed basic block into allocation size parameter! Use other ctor" ); |
1253 | assert(Amt->getType()->isIntegerTy() && |
1254 | "Allocation array size is not an integer!" ); |
1255 | } |
1256 | return Amt; |
1257 | } |
1258 | |
1259 | static Align computeAllocaDefaultAlign(Type *Ty, InsertPosition Pos) { |
1260 | assert(Pos.isValid() && |
1261 | "Insertion position cannot be null when alignment not provided!" ); |
1262 | BasicBlock *BB = Pos.getBasicBlock(); |
1263 | assert(BB->getParent() && |
1264 | "BB must be in a Function when alignment not provided!" ); |
1265 | const DataLayout &DL = BB->getDataLayout(); |
1266 | return DL.getPrefTypeAlign(Ty); |
1267 | } |
1268 | |
1269 | AllocaInst::AllocaInst(Type *Ty, unsigned AddrSpace, const Twine &Name, |
1270 | InsertPosition InsertBefore) |
1271 | : AllocaInst(Ty, AddrSpace, /*ArraySize=*/nullptr, Name, InsertBefore) {} |
1272 | |
1273 | AllocaInst::AllocaInst(Type *Ty, unsigned AddrSpace, Value *ArraySize, |
1274 | const Twine &Name, InsertPosition InsertBefore) |
1275 | : AllocaInst(Ty, AddrSpace, ArraySize, |
1276 | computeAllocaDefaultAlign(Ty, Pos: InsertBefore), Name, |
1277 | InsertBefore) {} |
1278 | |
1279 | AllocaInst::AllocaInst(Type *Ty, unsigned AddrSpace, Value *ArraySize, |
1280 | Align Align, const Twine &Name, |
1281 | InsertPosition InsertBefore) |
1282 | : UnaryInstruction(PointerType::get(C&: Ty->getContext(), AddressSpace: AddrSpace), Alloca, |
1283 | getAISize(Context&: Ty->getContext(), Amt: ArraySize), InsertBefore), |
1284 | AllocatedType(Ty) { |
1285 | setAlignment(Align); |
1286 | assert(!Ty->isVoidTy() && "Cannot allocate void!" ); |
1287 | setName(Name); |
1288 | } |
1289 | |
1290 | bool AllocaInst::isArrayAllocation() const { |
1291 | if (ConstantInt *CI = dyn_cast<ConstantInt>(Val: getOperand(i_nocapture: 0))) |
1292 | return !CI->isOne(); |
1293 | return true; |
1294 | } |
1295 | |
1296 | /// isStaticAlloca - Return true if this alloca is in the entry block of the |
1297 | /// function and is a constant size. If so, the code generator will fold it |
1298 | /// into the prolog/epilog code, so it is basically free. |
1299 | bool AllocaInst::isStaticAlloca() const { |
1300 | // Must be constant size. |
1301 | if (!isa<ConstantInt>(Val: getArraySize())) return false; |
1302 | |
1303 | // Must be in the entry block. |
1304 | const BasicBlock *Parent = getParent(); |
1305 | return Parent->isEntryBlock() && !isUsedWithInAlloca(); |
1306 | } |
1307 | |
1308 | //===----------------------------------------------------------------------===// |
1309 | // LoadInst Implementation |
1310 | //===----------------------------------------------------------------------===// |
1311 | |
1312 | void LoadInst::AssertOK() { |
1313 | assert(getOperand(0)->getType()->isPointerTy() && |
1314 | "Ptr must have pointer type." ); |
1315 | } |
1316 | |
1317 | static Align computeLoadStoreDefaultAlign(Type *Ty, InsertPosition Pos) { |
1318 | assert(Pos.isValid() && |
1319 | "Insertion position cannot be null when alignment not provided!" ); |
1320 | BasicBlock *BB = Pos.getBasicBlock(); |
1321 | assert(BB->getParent() && |
1322 | "BB must be in a Function when alignment not provided!" ); |
1323 | const DataLayout &DL = BB->getDataLayout(); |
1324 | return DL.getABITypeAlign(Ty); |
1325 | } |
1326 | |
1327 | LoadInst::LoadInst(Type *Ty, Value *Ptr, const Twine &Name, |
1328 | InsertPosition InsertBef) |
1329 | : LoadInst(Ty, Ptr, Name, /*isVolatile=*/false, InsertBef) {} |
1330 | |
1331 | LoadInst::LoadInst(Type *Ty, Value *Ptr, const Twine &Name, bool isVolatile, |
1332 | InsertPosition InsertBef) |
1333 | : LoadInst(Ty, Ptr, Name, isVolatile, |
1334 | computeLoadStoreDefaultAlign(Ty, Pos: InsertBef), InsertBef) {} |
1335 | |
1336 | LoadInst::LoadInst(Type *Ty, Value *Ptr, const Twine &Name, bool isVolatile, |
1337 | Align Align, InsertPosition InsertBef) |
1338 | : LoadInst(Ty, Ptr, Name, isVolatile, Align, AtomicOrdering::NotAtomic, |
1339 | SyncScope::System, InsertBef) {} |
1340 | |
1341 | LoadInst::LoadInst(Type *Ty, Value *Ptr, const Twine &Name, bool isVolatile, |
1342 | Align Align, AtomicOrdering Order, SyncScope::ID SSID, |
1343 | InsertPosition InsertBef) |
1344 | : UnaryInstruction(Ty, Load, Ptr, InsertBef) { |
1345 | setVolatile(isVolatile); |
1346 | setAlignment(Align); |
1347 | setAtomic(Ordering: Order, SSID); |
1348 | AssertOK(); |
1349 | setName(Name); |
1350 | } |
1351 | |
1352 | //===----------------------------------------------------------------------===// |
1353 | // StoreInst Implementation |
1354 | //===----------------------------------------------------------------------===// |
1355 | |
1356 | void StoreInst::AssertOK() { |
1357 | assert(getOperand(0) && getOperand(1) && "Both operands must be non-null!" ); |
1358 | assert(getOperand(1)->getType()->isPointerTy() && |
1359 | "Ptr must have pointer type!" ); |
1360 | } |
1361 | |
1362 | StoreInst::StoreInst(Value *val, Value *addr, InsertPosition InsertBefore) |
1363 | : StoreInst(val, addr, /*isVolatile=*/false, InsertBefore) {} |
1364 | |
1365 | StoreInst::StoreInst(Value *val, Value *addr, bool isVolatile, |
1366 | InsertPosition InsertBefore) |
1367 | : StoreInst(val, addr, isVolatile, |
1368 | computeLoadStoreDefaultAlign(Ty: val->getType(), Pos: InsertBefore), |
1369 | InsertBefore) {} |
1370 | |
1371 | StoreInst::StoreInst(Value *val, Value *addr, bool isVolatile, Align Align, |
1372 | InsertPosition InsertBefore) |
1373 | : StoreInst(val, addr, isVolatile, Align, AtomicOrdering::NotAtomic, |
1374 | SyncScope::System, InsertBefore) {} |
1375 | |
1376 | StoreInst::StoreInst(Value *val, Value *addr, bool isVolatile, Align Align, |
1377 | AtomicOrdering Order, SyncScope::ID SSID, |
1378 | InsertPosition InsertBefore) |
1379 | : Instruction(Type::getVoidTy(C&: val->getContext()), Store, AllocMarker, |
1380 | InsertBefore) { |
1381 | Op<0>() = val; |
1382 | Op<1>() = addr; |
1383 | setVolatile(isVolatile); |
1384 | setAlignment(Align); |
1385 | setAtomic(Ordering: Order, SSID); |
1386 | AssertOK(); |
1387 | } |
1388 | |
1389 | //===----------------------------------------------------------------------===// |
1390 | // AtomicCmpXchgInst Implementation |
1391 | //===----------------------------------------------------------------------===// |
1392 | |
1393 | void AtomicCmpXchgInst::Init(Value *Ptr, Value *Cmp, Value *NewVal, |
1394 | Align Alignment, AtomicOrdering SuccessOrdering, |
1395 | AtomicOrdering FailureOrdering, |
1396 | SyncScope::ID SSID) { |
1397 | Op<0>() = Ptr; |
1398 | Op<1>() = Cmp; |
1399 | Op<2>() = NewVal; |
1400 | setSuccessOrdering(SuccessOrdering); |
1401 | setFailureOrdering(FailureOrdering); |
1402 | setSyncScopeID(SSID); |
1403 | setAlignment(Alignment); |
1404 | |
1405 | assert(getOperand(0) && getOperand(1) && getOperand(2) && |
1406 | "All operands must be non-null!" ); |
1407 | assert(getOperand(0)->getType()->isPointerTy() && |
1408 | "Ptr must have pointer type!" ); |
1409 | assert(getOperand(1)->getType() == getOperand(2)->getType() && |
1410 | "Cmp type and NewVal type must be same!" ); |
1411 | } |
1412 | |
1413 | AtomicCmpXchgInst::AtomicCmpXchgInst(Value *Ptr, Value *Cmp, Value *NewVal, |
1414 | Align Alignment, |
1415 | AtomicOrdering SuccessOrdering, |
1416 | AtomicOrdering FailureOrdering, |
1417 | SyncScope::ID SSID, |
1418 | InsertPosition InsertBefore) |
1419 | : Instruction( |
1420 | StructType::get(elt1: Cmp->getType(), elts: Type::getInt1Ty(C&: Cmp->getContext())), |
1421 | AtomicCmpXchg, AllocMarker, InsertBefore) { |
1422 | Init(Ptr, Cmp, NewVal, Alignment, SuccessOrdering, FailureOrdering, SSID); |
1423 | } |
1424 | |
1425 | //===----------------------------------------------------------------------===// |
1426 | // AtomicRMWInst Implementation |
1427 | //===----------------------------------------------------------------------===// |
1428 | |
1429 | void AtomicRMWInst::Init(BinOp Operation, Value *Ptr, Value *Val, |
1430 | Align Alignment, AtomicOrdering Ordering, |
1431 | SyncScope::ID SSID) { |
1432 | assert(Ordering != AtomicOrdering::NotAtomic && |
1433 | "atomicrmw instructions can only be atomic." ); |
1434 | assert(Ordering != AtomicOrdering::Unordered && |
1435 | "atomicrmw instructions cannot be unordered." ); |
1436 | Op<0>() = Ptr; |
1437 | Op<1>() = Val; |
1438 | setOperation(Operation); |
1439 | setOrdering(Ordering); |
1440 | setSyncScopeID(SSID); |
1441 | setAlignment(Alignment); |
1442 | |
1443 | assert(getOperand(0) && getOperand(1) && "All operands must be non-null!" ); |
1444 | assert(getOperand(0)->getType()->isPointerTy() && |
1445 | "Ptr must have pointer type!" ); |
1446 | assert(Ordering != AtomicOrdering::NotAtomic && |
1447 | "AtomicRMW instructions must be atomic!" ); |
1448 | } |
1449 | |
1450 | AtomicRMWInst::AtomicRMWInst(BinOp Operation, Value *Ptr, Value *Val, |
1451 | Align Alignment, AtomicOrdering Ordering, |
1452 | SyncScope::ID SSID, InsertPosition InsertBefore) |
1453 | : Instruction(Val->getType(), AtomicRMW, AllocMarker, InsertBefore) { |
1454 | Init(Operation, Ptr, Val, Alignment, Ordering, SSID); |
1455 | } |
1456 | |
1457 | StringRef AtomicRMWInst::getOperationName(BinOp Op) { |
1458 | switch (Op) { |
1459 | case AtomicRMWInst::Xchg: |
1460 | return "xchg" ; |
1461 | case AtomicRMWInst::Add: |
1462 | return "add" ; |
1463 | case AtomicRMWInst::Sub: |
1464 | return "sub" ; |
1465 | case AtomicRMWInst::And: |
1466 | return "and" ; |
1467 | case AtomicRMWInst::Nand: |
1468 | return "nand" ; |
1469 | case AtomicRMWInst::Or: |
1470 | return "or" ; |
1471 | case AtomicRMWInst::Xor: |
1472 | return "xor" ; |
1473 | case AtomicRMWInst::Max: |
1474 | return "max" ; |
1475 | case AtomicRMWInst::Min: |
1476 | return "min" ; |
1477 | case AtomicRMWInst::UMax: |
1478 | return "umax" ; |
1479 | case AtomicRMWInst::UMin: |
1480 | return "umin" ; |
1481 | case AtomicRMWInst::FAdd: |
1482 | return "fadd" ; |
1483 | case AtomicRMWInst::FSub: |
1484 | return "fsub" ; |
1485 | case AtomicRMWInst::FMax: |
1486 | return "fmax" ; |
1487 | case AtomicRMWInst::FMin: |
1488 | return "fmin" ; |
1489 | case AtomicRMWInst::FMaximum: |
1490 | return "fmaximum" ; |
1491 | case AtomicRMWInst::FMinimum: |
1492 | return "fminimum" ; |
1493 | case AtomicRMWInst::UIncWrap: |
1494 | return "uinc_wrap" ; |
1495 | case AtomicRMWInst::UDecWrap: |
1496 | return "udec_wrap" ; |
1497 | case AtomicRMWInst::USubCond: |
1498 | return "usub_cond" ; |
1499 | case AtomicRMWInst::USubSat: |
1500 | return "usub_sat" ; |
1501 | case AtomicRMWInst::BAD_BINOP: |
1502 | return "<invalid operation>" ; |
1503 | } |
1504 | |
1505 | llvm_unreachable("invalid atomicrmw operation" ); |
1506 | } |
1507 | |
1508 | //===----------------------------------------------------------------------===// |
1509 | // FenceInst Implementation |
1510 | //===----------------------------------------------------------------------===// |
1511 | |
1512 | FenceInst::FenceInst(LLVMContext &C, AtomicOrdering Ordering, |
1513 | SyncScope::ID SSID, InsertPosition InsertBefore) |
1514 | : Instruction(Type::getVoidTy(C), Fence, AllocMarker, InsertBefore) { |
1515 | setOrdering(Ordering); |
1516 | setSyncScopeID(SSID); |
1517 | } |
1518 | |
1519 | //===----------------------------------------------------------------------===// |
1520 | // GetElementPtrInst Implementation |
1521 | //===----------------------------------------------------------------------===// |
1522 | |
1523 | void GetElementPtrInst::init(Value *Ptr, ArrayRef<Value *> IdxList, |
1524 | const Twine &Name) { |
1525 | assert(getNumOperands() == 1 + IdxList.size() && |
1526 | "NumOperands not initialized?" ); |
1527 | Op<0>() = Ptr; |
1528 | llvm::copy(Range&: IdxList, Out: op_begin() + 1); |
1529 | setName(Name); |
1530 | } |
1531 | |
1532 | GetElementPtrInst::GetElementPtrInst(const GetElementPtrInst &GEPI, |
1533 | AllocInfo AllocInfo) |
1534 | : Instruction(GEPI.getType(), GetElementPtr, AllocInfo), |
1535 | SourceElementType(GEPI.SourceElementType), |
1536 | ResultElementType(GEPI.ResultElementType) { |
1537 | assert(getNumOperands() == GEPI.getNumOperands() && |
1538 | "Wrong number of operands allocated" ); |
1539 | std::copy(first: GEPI.op_begin(), last: GEPI.op_end(), result: op_begin()); |
1540 | SubclassOptionalData = GEPI.SubclassOptionalData; |
1541 | } |
1542 | |
1543 | Type *GetElementPtrInst::getTypeAtIndex(Type *Ty, Value *Idx) { |
1544 | if (auto *Struct = dyn_cast<StructType>(Val: Ty)) { |
1545 | if (!Struct->indexValid(V: Idx)) |
1546 | return nullptr; |
1547 | return Struct->getTypeAtIndex(V: Idx); |
1548 | } |
1549 | if (!Idx->getType()->isIntOrIntVectorTy()) |
1550 | return nullptr; |
1551 | if (auto *Array = dyn_cast<ArrayType>(Val: Ty)) |
1552 | return Array->getElementType(); |
1553 | if (auto *Vector = dyn_cast<VectorType>(Val: Ty)) |
1554 | return Vector->getElementType(); |
1555 | return nullptr; |
1556 | } |
1557 | |
1558 | Type *GetElementPtrInst::getTypeAtIndex(Type *Ty, uint64_t Idx) { |
1559 | if (auto *Struct = dyn_cast<StructType>(Val: Ty)) { |
1560 | if (Idx >= Struct->getNumElements()) |
1561 | return nullptr; |
1562 | return Struct->getElementType(N: Idx); |
1563 | } |
1564 | if (auto *Array = dyn_cast<ArrayType>(Val: Ty)) |
1565 | return Array->getElementType(); |
1566 | if (auto *Vector = dyn_cast<VectorType>(Val: Ty)) |
1567 | return Vector->getElementType(); |
1568 | return nullptr; |
1569 | } |
1570 | |
1571 | template <typename IndexTy> |
1572 | static Type *getIndexedTypeInternal(Type *Ty, ArrayRef<IndexTy> IdxList) { |
1573 | if (IdxList.empty()) |
1574 | return Ty; |
1575 | for (IndexTy V : IdxList.slice(1)) { |
1576 | Ty = GetElementPtrInst::getTypeAtIndex(Ty, V); |
1577 | if (!Ty) |
1578 | return Ty; |
1579 | } |
1580 | return Ty; |
1581 | } |
1582 | |
1583 | Type *GetElementPtrInst::getIndexedType(Type *Ty, ArrayRef<Value *> IdxList) { |
1584 | return getIndexedTypeInternal(Ty, IdxList); |
1585 | } |
1586 | |
1587 | Type *GetElementPtrInst::getIndexedType(Type *Ty, |
1588 | ArrayRef<Constant *> IdxList) { |
1589 | return getIndexedTypeInternal(Ty, IdxList); |
1590 | } |
1591 | |
1592 | Type *GetElementPtrInst::getIndexedType(Type *Ty, ArrayRef<uint64_t> IdxList) { |
1593 | return getIndexedTypeInternal(Ty, IdxList); |
1594 | } |
1595 | |
1596 | /// hasAllZeroIndices - Return true if all of the indices of this GEP are |
1597 | /// zeros. If so, the result pointer and the first operand have the same |
1598 | /// value, just potentially different types. |
1599 | bool GetElementPtrInst::hasAllZeroIndices() const { |
1600 | for (unsigned i = 1, e = getNumOperands(); i != e; ++i) { |
1601 | if (ConstantInt *CI = dyn_cast<ConstantInt>(Val: getOperand(i_nocapture: i))) { |
1602 | if (!CI->isZero()) return false; |
1603 | } else { |
1604 | return false; |
1605 | } |
1606 | } |
1607 | return true; |
1608 | } |
1609 | |
1610 | /// hasAllConstantIndices - Return true if all of the indices of this GEP are |
1611 | /// constant integers. If so, the result pointer and the first operand have |
1612 | /// a constant offset between them. |
1613 | bool GetElementPtrInst::hasAllConstantIndices() const { |
1614 | for (unsigned i = 1, e = getNumOperands(); i != e; ++i) { |
1615 | if (!isa<ConstantInt>(Val: getOperand(i_nocapture: i))) |
1616 | return false; |
1617 | } |
1618 | return true; |
1619 | } |
1620 | |
1621 | void GetElementPtrInst::setNoWrapFlags(GEPNoWrapFlags NW) { |
1622 | SubclassOptionalData = NW.getRaw(); |
1623 | } |
1624 | |
1625 | void GetElementPtrInst::setIsInBounds(bool B) { |
1626 | GEPNoWrapFlags NW = cast<GEPOperator>(Val: this)->getNoWrapFlags(); |
1627 | if (B) |
1628 | NW |= GEPNoWrapFlags::inBounds(); |
1629 | else |
1630 | NW = NW.withoutInBounds(); |
1631 | setNoWrapFlags(NW); |
1632 | } |
1633 | |
1634 | GEPNoWrapFlags GetElementPtrInst::getNoWrapFlags() const { |
1635 | return cast<GEPOperator>(Val: this)->getNoWrapFlags(); |
1636 | } |
1637 | |
1638 | bool GetElementPtrInst::isInBounds() const { |
1639 | return cast<GEPOperator>(Val: this)->isInBounds(); |
1640 | } |
1641 | |
1642 | bool GetElementPtrInst::hasNoUnsignedSignedWrap() const { |
1643 | return cast<GEPOperator>(Val: this)->hasNoUnsignedSignedWrap(); |
1644 | } |
1645 | |
1646 | bool GetElementPtrInst::hasNoUnsignedWrap() const { |
1647 | return cast<GEPOperator>(Val: this)->hasNoUnsignedWrap(); |
1648 | } |
1649 | |
1650 | bool GetElementPtrInst::accumulateConstantOffset(const DataLayout &DL, |
1651 | APInt &Offset) const { |
1652 | // Delegate to the generic GEPOperator implementation. |
1653 | return cast<GEPOperator>(Val: this)->accumulateConstantOffset(DL, Offset); |
1654 | } |
1655 | |
1656 | bool GetElementPtrInst::collectOffset( |
1657 | const DataLayout &DL, unsigned BitWidth, |
1658 | SmallMapVector<Value *, APInt, 4> &VariableOffsets, |
1659 | APInt &ConstantOffset) const { |
1660 | // Delegate to the generic GEPOperator implementation. |
1661 | return cast<GEPOperator>(Val: this)->collectOffset(DL, BitWidth, VariableOffsets, |
1662 | ConstantOffset); |
1663 | } |
1664 | |
1665 | //===----------------------------------------------------------------------===// |
1666 | // ExtractElementInst Implementation |
1667 | //===----------------------------------------------------------------------===// |
1668 | |
1669 | ExtractElementInst::(Value *Val, Value *Index, |
1670 | const Twine &Name, |
1671 | InsertPosition InsertBef) |
1672 | : Instruction(cast<VectorType>(Val: Val->getType())->getElementType(), |
1673 | ExtractElement, AllocMarker, InsertBef) { |
1674 | assert(isValidOperands(Val, Index) && |
1675 | "Invalid extractelement instruction operands!" ); |
1676 | Op<0>() = Val; |
1677 | Op<1>() = Index; |
1678 | setName(Name); |
1679 | } |
1680 | |
1681 | bool ExtractElementInst::isValidOperands(const Value *Val, const Value *Index) { |
1682 | if (!Val->getType()->isVectorTy() || !Index->getType()->isIntegerTy()) |
1683 | return false; |
1684 | return true; |
1685 | } |
1686 | |
1687 | //===----------------------------------------------------------------------===// |
1688 | // InsertElementInst Implementation |
1689 | //===----------------------------------------------------------------------===// |
1690 | |
1691 | InsertElementInst::InsertElementInst(Value *Vec, Value *Elt, Value *Index, |
1692 | const Twine &Name, |
1693 | InsertPosition InsertBef) |
1694 | : Instruction(Vec->getType(), InsertElement, AllocMarker, InsertBef) { |
1695 | assert(isValidOperands(Vec, Elt, Index) && |
1696 | "Invalid insertelement instruction operands!" ); |
1697 | Op<0>() = Vec; |
1698 | Op<1>() = Elt; |
1699 | Op<2>() = Index; |
1700 | setName(Name); |
1701 | } |
1702 | |
1703 | bool InsertElementInst::isValidOperands(const Value *Vec, const Value *Elt, |
1704 | const Value *Index) { |
1705 | if (!Vec->getType()->isVectorTy()) |
1706 | return false; // First operand of insertelement must be vector type. |
1707 | |
1708 | if (Elt->getType() != cast<VectorType>(Val: Vec->getType())->getElementType()) |
1709 | return false;// Second operand of insertelement must be vector element type. |
1710 | |
1711 | if (!Index->getType()->isIntegerTy()) |
1712 | return false; // Third operand of insertelement must be i32. |
1713 | return true; |
1714 | } |
1715 | |
1716 | //===----------------------------------------------------------------------===// |
1717 | // ShuffleVectorInst Implementation |
1718 | //===----------------------------------------------------------------------===// |
1719 | |
1720 | static Value *createPlaceholderForShuffleVector(Value *V) { |
1721 | assert(V && "Cannot create placeholder of nullptr V" ); |
1722 | return PoisonValue::get(T: V->getType()); |
1723 | } |
1724 | |
1725 | ShuffleVectorInst::ShuffleVectorInst(Value *V1, Value *Mask, const Twine &Name, |
1726 | InsertPosition InsertBefore) |
1727 | : ShuffleVectorInst(V1, createPlaceholderForShuffleVector(V: V1), Mask, Name, |
1728 | InsertBefore) {} |
1729 | |
1730 | ShuffleVectorInst::ShuffleVectorInst(Value *V1, ArrayRef<int> Mask, |
1731 | const Twine &Name, |
1732 | InsertPosition InsertBefore) |
1733 | : ShuffleVectorInst(V1, createPlaceholderForShuffleVector(V: V1), Mask, Name, |
1734 | InsertBefore) {} |
1735 | |
1736 | ShuffleVectorInst::ShuffleVectorInst(Value *V1, Value *V2, Value *Mask, |
1737 | const Twine &Name, |
1738 | InsertPosition InsertBefore) |
1739 | : Instruction( |
1740 | VectorType::get(ElementType: cast<VectorType>(Val: V1->getType())->getElementType(), |
1741 | EC: cast<VectorType>(Val: Mask->getType())->getElementCount()), |
1742 | ShuffleVector, AllocMarker, InsertBefore) { |
1743 | assert(isValidOperands(V1, V2, Mask) && |
1744 | "Invalid shuffle vector instruction operands!" ); |
1745 | |
1746 | Op<0>() = V1; |
1747 | Op<1>() = V2; |
1748 | SmallVector<int, 16> MaskArr; |
1749 | getShuffleMask(Mask: cast<Constant>(Val: Mask), Result&: MaskArr); |
1750 | setShuffleMask(MaskArr); |
1751 | setName(Name); |
1752 | } |
1753 | |
1754 | ShuffleVectorInst::ShuffleVectorInst(Value *V1, Value *V2, ArrayRef<int> Mask, |
1755 | const Twine &Name, |
1756 | InsertPosition InsertBefore) |
1757 | : Instruction( |
1758 | VectorType::get(ElementType: cast<VectorType>(Val: V1->getType())->getElementType(), |
1759 | NumElements: Mask.size(), Scalable: isa<ScalableVectorType>(Val: V1->getType())), |
1760 | ShuffleVector, AllocMarker, InsertBefore) { |
1761 | assert(isValidOperands(V1, V2, Mask) && |
1762 | "Invalid shuffle vector instruction operands!" ); |
1763 | Op<0>() = V1; |
1764 | Op<1>() = V2; |
1765 | setShuffleMask(Mask); |
1766 | setName(Name); |
1767 | } |
1768 | |
1769 | void ShuffleVectorInst::commute() { |
1770 | int NumOpElts = cast<FixedVectorType>(Val: Op<0>()->getType())->getNumElements(); |
1771 | int NumMaskElts = ShuffleMask.size(); |
1772 | SmallVector<int, 16> NewMask(NumMaskElts); |
1773 | for (int i = 0; i != NumMaskElts; ++i) { |
1774 | int MaskElt = getMaskValue(Elt: i); |
1775 | if (MaskElt == PoisonMaskElem) { |
1776 | NewMask[i] = PoisonMaskElem; |
1777 | continue; |
1778 | } |
1779 | assert(MaskElt >= 0 && MaskElt < 2 * NumOpElts && "Out-of-range mask" ); |
1780 | MaskElt = (MaskElt < NumOpElts) ? MaskElt + NumOpElts : MaskElt - NumOpElts; |
1781 | NewMask[i] = MaskElt; |
1782 | } |
1783 | setShuffleMask(NewMask); |
1784 | Op<0>().swap(RHS&: Op<1>()); |
1785 | } |
1786 | |
1787 | bool ShuffleVectorInst::isValidOperands(const Value *V1, const Value *V2, |
1788 | ArrayRef<int> Mask) { |
1789 | // V1 and V2 must be vectors of the same type. |
1790 | if (!isa<VectorType>(Val: V1->getType()) || V1->getType() != V2->getType()) |
1791 | return false; |
1792 | |
1793 | // Make sure the mask elements make sense. |
1794 | int V1Size = |
1795 | cast<VectorType>(Val: V1->getType())->getElementCount().getKnownMinValue(); |
1796 | for (int Elem : Mask) |
1797 | if (Elem != PoisonMaskElem && Elem >= V1Size * 2) |
1798 | return false; |
1799 | |
1800 | if (isa<ScalableVectorType>(Val: V1->getType())) |
1801 | if ((Mask[0] != 0 && Mask[0] != PoisonMaskElem) || !all_equal(Range&: Mask)) |
1802 | return false; |
1803 | |
1804 | return true; |
1805 | } |
1806 | |
1807 | bool ShuffleVectorInst::isValidOperands(const Value *V1, const Value *V2, |
1808 | const Value *Mask) { |
1809 | // V1 and V2 must be vectors of the same type. |
1810 | if (!V1->getType()->isVectorTy() || V1->getType() != V2->getType()) |
1811 | return false; |
1812 | |
1813 | // Mask must be vector of i32, and must be the same kind of vector as the |
1814 | // input vectors |
1815 | auto *MaskTy = dyn_cast<VectorType>(Val: Mask->getType()); |
1816 | if (!MaskTy || !MaskTy->getElementType()->isIntegerTy(Bitwidth: 32) || |
1817 | isa<ScalableVectorType>(Val: MaskTy) != isa<ScalableVectorType>(Val: V1->getType())) |
1818 | return false; |
1819 | |
1820 | // Check to see if Mask is valid. |
1821 | if (isa<UndefValue>(Val: Mask) || isa<ConstantAggregateZero>(Val: Mask)) |
1822 | return true; |
1823 | |
1824 | // NOTE: Through vector ConstantInt we have the potential to support more |
1825 | // than just zero splat masks but that requires a LangRef change. |
1826 | if (isa<ScalableVectorType>(Val: MaskTy)) |
1827 | return false; |
1828 | |
1829 | unsigned V1Size = cast<FixedVectorType>(Val: V1->getType())->getNumElements(); |
1830 | |
1831 | if (const auto *CI = dyn_cast<ConstantInt>(Val: Mask)) |
1832 | return !CI->uge(Num: V1Size * 2); |
1833 | |
1834 | if (const auto *MV = dyn_cast<ConstantVector>(Val: Mask)) { |
1835 | for (Value *Op : MV->operands()) { |
1836 | if (auto *CI = dyn_cast<ConstantInt>(Val: Op)) { |
1837 | if (CI->uge(Num: V1Size*2)) |
1838 | return false; |
1839 | } else if (!isa<UndefValue>(Val: Op)) { |
1840 | return false; |
1841 | } |
1842 | } |
1843 | return true; |
1844 | } |
1845 | |
1846 | if (const auto *CDS = dyn_cast<ConstantDataSequential>(Val: Mask)) { |
1847 | for (unsigned i = 0, e = cast<FixedVectorType>(Val: MaskTy)->getNumElements(); |
1848 | i != e; ++i) |
1849 | if (CDS->getElementAsInteger(i) >= V1Size*2) |
1850 | return false; |
1851 | return true; |
1852 | } |
1853 | |
1854 | return false; |
1855 | } |
1856 | |
1857 | void ShuffleVectorInst::getShuffleMask(const Constant *Mask, |
1858 | SmallVectorImpl<int> &Result) { |
1859 | ElementCount EC = cast<VectorType>(Val: Mask->getType())->getElementCount(); |
1860 | |
1861 | if (isa<ConstantAggregateZero>(Val: Mask) || isa<UndefValue>(Val: Mask)) { |
1862 | int MaskVal = isa<UndefValue>(Val: Mask) ? -1 : 0; |
1863 | Result.append(NumInputs: EC.getKnownMinValue(), Elt: MaskVal); |
1864 | return; |
1865 | } |
1866 | |
1867 | assert(!EC.isScalable() && |
1868 | "Scalable vector shuffle mask must be undef or zeroinitializer" ); |
1869 | |
1870 | unsigned NumElts = EC.getFixedValue(); |
1871 | |
1872 | Result.reserve(N: NumElts); |
1873 | |
1874 | if (auto *CDS = dyn_cast<ConstantDataSequential>(Val: Mask)) { |
1875 | for (unsigned i = 0; i != NumElts; ++i) |
1876 | Result.push_back(Elt: CDS->getElementAsInteger(i)); |
1877 | return; |
1878 | } |
1879 | for (unsigned i = 0; i != NumElts; ++i) { |
1880 | Constant *C = Mask->getAggregateElement(Elt: i); |
1881 | Result.push_back(Elt: isa<UndefValue>(Val: C) ? -1 : |
1882 | cast<ConstantInt>(Val: C)->getZExtValue()); |
1883 | } |
1884 | } |
1885 | |
1886 | void ShuffleVectorInst::setShuffleMask(ArrayRef<int> Mask) { |
1887 | ShuffleMask.assign(in_start: Mask.begin(), in_end: Mask.end()); |
1888 | ShuffleMaskForBitcode = convertShuffleMaskForBitcode(Mask, ResultTy: getType()); |
1889 | } |
1890 | |
1891 | Constant *ShuffleVectorInst::convertShuffleMaskForBitcode(ArrayRef<int> Mask, |
1892 | Type *ResultTy) { |
1893 | Type *Int32Ty = Type::getInt32Ty(C&: ResultTy->getContext()); |
1894 | if (isa<ScalableVectorType>(Val: ResultTy)) { |
1895 | assert(all_equal(Mask) && "Unexpected shuffle" ); |
1896 | Type *VecTy = VectorType::get(ElementType: Int32Ty, NumElements: Mask.size(), Scalable: true); |
1897 | if (Mask[0] == 0) |
1898 | return Constant::getNullValue(Ty: VecTy); |
1899 | return PoisonValue::get(T: VecTy); |
1900 | } |
1901 | SmallVector<Constant *, 16> MaskConst; |
1902 | for (int Elem : Mask) { |
1903 | if (Elem == PoisonMaskElem) |
1904 | MaskConst.push_back(Elt: PoisonValue::get(T: Int32Ty)); |
1905 | else |
1906 | MaskConst.push_back(Elt: ConstantInt::get(Ty: Int32Ty, V: Elem)); |
1907 | } |
1908 | return ConstantVector::get(V: MaskConst); |
1909 | } |
1910 | |
1911 | static bool isSingleSourceMaskImpl(ArrayRef<int> Mask, int NumOpElts) { |
1912 | assert(!Mask.empty() && "Shuffle mask must contain elements" ); |
1913 | bool UsesLHS = false; |
1914 | bool UsesRHS = false; |
1915 | for (int I : Mask) { |
1916 | if (I == -1) |
1917 | continue; |
1918 | assert(I >= 0 && I < (NumOpElts * 2) && |
1919 | "Out-of-bounds shuffle mask element" ); |
1920 | UsesLHS |= (I < NumOpElts); |
1921 | UsesRHS |= (I >= NumOpElts); |
1922 | if (UsesLHS && UsesRHS) |
1923 | return false; |
1924 | } |
1925 | // Allow for degenerate case: completely undef mask means neither source is used. |
1926 | return UsesLHS || UsesRHS; |
1927 | } |
1928 | |
1929 | bool ShuffleVectorInst::isSingleSourceMask(ArrayRef<int> Mask, int NumSrcElts) { |
1930 | // We don't have vector operand size information, so assume operands are the |
1931 | // same size as the mask. |
1932 | return isSingleSourceMaskImpl(Mask, NumOpElts: NumSrcElts); |
1933 | } |
1934 | |
1935 | static bool isIdentityMaskImpl(ArrayRef<int> Mask, int NumOpElts) { |
1936 | if (!isSingleSourceMaskImpl(Mask, NumOpElts)) |
1937 | return false; |
1938 | for (int i = 0, NumMaskElts = Mask.size(); i < NumMaskElts; ++i) { |
1939 | if (Mask[i] == -1) |
1940 | continue; |
1941 | if (Mask[i] != i && Mask[i] != (NumOpElts + i)) |
1942 | return false; |
1943 | } |
1944 | return true; |
1945 | } |
1946 | |
1947 | bool ShuffleVectorInst::isIdentityMask(ArrayRef<int> Mask, int NumSrcElts) { |
1948 | if (Mask.size() != static_cast<unsigned>(NumSrcElts)) |
1949 | return false; |
1950 | // We don't have vector operand size information, so assume operands are the |
1951 | // same size as the mask. |
1952 | return isIdentityMaskImpl(Mask, NumOpElts: NumSrcElts); |
1953 | } |
1954 | |
1955 | bool ShuffleVectorInst::isReverseMask(ArrayRef<int> Mask, int NumSrcElts) { |
1956 | if (Mask.size() != static_cast<unsigned>(NumSrcElts)) |
1957 | return false; |
1958 | if (!isSingleSourceMask(Mask, NumSrcElts)) |
1959 | return false; |
1960 | |
1961 | // The number of elements in the mask must be at least 2. |
1962 | if (NumSrcElts < 2) |
1963 | return false; |
1964 | |
1965 | for (int I = 0, E = Mask.size(); I < E; ++I) { |
1966 | if (Mask[I] == -1) |
1967 | continue; |
1968 | if (Mask[I] != (NumSrcElts - 1 - I) && |
1969 | Mask[I] != (NumSrcElts + NumSrcElts - 1 - I)) |
1970 | return false; |
1971 | } |
1972 | return true; |
1973 | } |
1974 | |
1975 | bool ShuffleVectorInst::isZeroEltSplatMask(ArrayRef<int> Mask, int NumSrcElts) { |
1976 | if (Mask.size() != static_cast<unsigned>(NumSrcElts)) |
1977 | return false; |
1978 | if (!isSingleSourceMask(Mask, NumSrcElts)) |
1979 | return false; |
1980 | for (int I = 0, E = Mask.size(); I < E; ++I) { |
1981 | if (Mask[I] == -1) |
1982 | continue; |
1983 | if (Mask[I] != 0 && Mask[I] != NumSrcElts) |
1984 | return false; |
1985 | } |
1986 | return true; |
1987 | } |
1988 | |
1989 | bool ShuffleVectorInst::isSelectMask(ArrayRef<int> Mask, int NumSrcElts) { |
1990 | if (Mask.size() != static_cast<unsigned>(NumSrcElts)) |
1991 | return false; |
1992 | // Select is differentiated from identity. It requires using both sources. |
1993 | if (isSingleSourceMask(Mask, NumSrcElts)) |
1994 | return false; |
1995 | for (int I = 0, E = Mask.size(); I < E; ++I) { |
1996 | if (Mask[I] == -1) |
1997 | continue; |
1998 | if (Mask[I] != I && Mask[I] != (NumSrcElts + I)) |
1999 | return false; |
2000 | } |
2001 | return true; |
2002 | } |
2003 | |
2004 | bool ShuffleVectorInst::isTransposeMask(ArrayRef<int> Mask, int NumSrcElts) { |
2005 | // Example masks that will return true: |
2006 | // v1 = <a, b, c, d> |
2007 | // v2 = <e, f, g, h> |
2008 | // trn1 = shufflevector v1, v2 <0, 4, 2, 6> = <a, e, c, g> |
2009 | // trn2 = shufflevector v1, v2 <1, 5, 3, 7> = <b, f, d, h> |
2010 | |
2011 | if (Mask.size() != static_cast<unsigned>(NumSrcElts)) |
2012 | return false; |
2013 | // 1. The number of elements in the mask must be a power-of-2 and at least 2. |
2014 | int Sz = Mask.size(); |
2015 | if (Sz < 2 || !isPowerOf2_32(Value: Sz)) |
2016 | return false; |
2017 | |
2018 | // 2. The first element of the mask must be either a 0 or a 1. |
2019 | if (Mask[0] != 0 && Mask[0] != 1) |
2020 | return false; |
2021 | |
2022 | // 3. The difference between the first 2 elements must be equal to the |
2023 | // number of elements in the mask. |
2024 | if ((Mask[1] - Mask[0]) != NumSrcElts) |
2025 | return false; |
2026 | |
2027 | // 4. The difference between consecutive even-numbered and odd-numbered |
2028 | // elements must be equal to 2. |
2029 | for (int I = 2; I < Sz; ++I) { |
2030 | int MaskEltVal = Mask[I]; |
2031 | if (MaskEltVal == -1) |
2032 | return false; |
2033 | int MaskEltPrevVal = Mask[I - 2]; |
2034 | if (MaskEltVal - MaskEltPrevVal != 2) |
2035 | return false; |
2036 | } |
2037 | return true; |
2038 | } |
2039 | |
2040 | bool ShuffleVectorInst::isSpliceMask(ArrayRef<int> Mask, int NumSrcElts, |
2041 | int &Index) { |
2042 | if (Mask.size() != static_cast<unsigned>(NumSrcElts)) |
2043 | return false; |
2044 | // Example: shufflevector <4 x n> A, <4 x n> B, <1,2,3,4> |
2045 | int StartIndex = -1; |
2046 | for (int I = 0, E = Mask.size(); I != E; ++I) { |
2047 | int MaskEltVal = Mask[I]; |
2048 | if (MaskEltVal == -1) |
2049 | continue; |
2050 | |
2051 | if (StartIndex == -1) { |
2052 | // Don't support a StartIndex that begins in the second input, or if the |
2053 | // first non-undef index would access below the StartIndex. |
2054 | if (MaskEltVal < I || NumSrcElts <= (MaskEltVal - I)) |
2055 | return false; |
2056 | |
2057 | StartIndex = MaskEltVal - I; |
2058 | continue; |
2059 | } |
2060 | |
2061 | // Splice is sequential starting from StartIndex. |
2062 | if (MaskEltVal != (StartIndex + I)) |
2063 | return false; |
2064 | } |
2065 | |
2066 | if (StartIndex == -1) |
2067 | return false; |
2068 | |
2069 | // NOTE: This accepts StartIndex == 0 (COPY). |
2070 | Index = StartIndex; |
2071 | return true; |
2072 | } |
2073 | |
2074 | bool ShuffleVectorInst::(ArrayRef<int> Mask, |
2075 | int NumSrcElts, int &Index) { |
2076 | // Must extract from a single source. |
2077 | if (!isSingleSourceMaskImpl(Mask, NumOpElts: NumSrcElts)) |
2078 | return false; |
2079 | |
2080 | // Must be smaller (else this is an Identity shuffle). |
2081 | if (NumSrcElts <= (int)Mask.size()) |
2082 | return false; |
2083 | |
2084 | // Find start of extraction, accounting that we may start with an UNDEF. |
2085 | int SubIndex = -1; |
2086 | for (int i = 0, e = Mask.size(); i != e; ++i) { |
2087 | int M = Mask[i]; |
2088 | if (M < 0) |
2089 | continue; |
2090 | int Offset = (M % NumSrcElts) - i; |
2091 | if (0 <= SubIndex && SubIndex != Offset) |
2092 | return false; |
2093 | SubIndex = Offset; |
2094 | } |
2095 | |
2096 | if (0 <= SubIndex && SubIndex + (int)Mask.size() <= NumSrcElts) { |
2097 | Index = SubIndex; |
2098 | return true; |
2099 | } |
2100 | return false; |
2101 | } |
2102 | |
2103 | bool ShuffleVectorInst::isInsertSubvectorMask(ArrayRef<int> Mask, |
2104 | int NumSrcElts, int &NumSubElts, |
2105 | int &Index) { |
2106 | int NumMaskElts = Mask.size(); |
2107 | |
2108 | // Don't try to match if we're shuffling to a smaller size. |
2109 | if (NumMaskElts < NumSrcElts) |
2110 | return false; |
2111 | |
2112 | // TODO: We don't recognize self-insertion/widening. |
2113 | if (isSingleSourceMaskImpl(Mask, NumOpElts: NumSrcElts)) |
2114 | return false; |
2115 | |
2116 | // Determine which mask elements are attributed to which source. |
2117 | APInt UndefElts = APInt::getZero(numBits: NumMaskElts); |
2118 | APInt Src0Elts = APInt::getZero(numBits: NumMaskElts); |
2119 | APInt Src1Elts = APInt::getZero(numBits: NumMaskElts); |
2120 | bool Src0Identity = true; |
2121 | bool Src1Identity = true; |
2122 | |
2123 | for (int i = 0; i != NumMaskElts; ++i) { |
2124 | int M = Mask[i]; |
2125 | if (M < 0) { |
2126 | UndefElts.setBit(i); |
2127 | continue; |
2128 | } |
2129 | if (M < NumSrcElts) { |
2130 | Src0Elts.setBit(i); |
2131 | Src0Identity &= (M == i); |
2132 | continue; |
2133 | } |
2134 | Src1Elts.setBit(i); |
2135 | Src1Identity &= (M == (i + NumSrcElts)); |
2136 | } |
2137 | assert((Src0Elts | Src1Elts | UndefElts).isAllOnes() && |
2138 | "unknown shuffle elements" ); |
2139 | assert(!Src0Elts.isZero() && !Src1Elts.isZero() && |
2140 | "2-source shuffle not found" ); |
2141 | |
2142 | // Determine lo/hi span ranges. |
2143 | // TODO: How should we handle undefs at the start of subvector insertions? |
2144 | int Src0Lo = Src0Elts.countr_zero(); |
2145 | int Src1Lo = Src1Elts.countr_zero(); |
2146 | int Src0Hi = NumMaskElts - Src0Elts.countl_zero(); |
2147 | int Src1Hi = NumMaskElts - Src1Elts.countl_zero(); |
2148 | |
2149 | // If src0 is in place, see if the src1 elements is inplace within its own |
2150 | // span. |
2151 | if (Src0Identity) { |
2152 | int NumSub1Elts = Src1Hi - Src1Lo; |
2153 | ArrayRef<int> Sub1Mask = Mask.slice(N: Src1Lo, M: NumSub1Elts); |
2154 | if (isIdentityMaskImpl(Mask: Sub1Mask, NumOpElts: NumSrcElts)) { |
2155 | NumSubElts = NumSub1Elts; |
2156 | Index = Src1Lo; |
2157 | return true; |
2158 | } |
2159 | } |
2160 | |
2161 | // If src1 is in place, see if the src0 elements is inplace within its own |
2162 | // span. |
2163 | if (Src1Identity) { |
2164 | int NumSub0Elts = Src0Hi - Src0Lo; |
2165 | ArrayRef<int> Sub0Mask = Mask.slice(N: Src0Lo, M: NumSub0Elts); |
2166 | if (isIdentityMaskImpl(Mask: Sub0Mask, NumOpElts: NumSrcElts)) { |
2167 | NumSubElts = NumSub0Elts; |
2168 | Index = Src0Lo; |
2169 | return true; |
2170 | } |
2171 | } |
2172 | |
2173 | return false; |
2174 | } |
2175 | |
2176 | bool ShuffleVectorInst::isIdentityWithPadding() const { |
2177 | // FIXME: Not currently possible to express a shuffle mask for a scalable |
2178 | // vector for this case. |
2179 | if (isa<ScalableVectorType>(Val: getType())) |
2180 | return false; |
2181 | |
2182 | int NumOpElts = cast<FixedVectorType>(Val: Op<0>()->getType())->getNumElements(); |
2183 | int NumMaskElts = cast<FixedVectorType>(Val: getType())->getNumElements(); |
2184 | if (NumMaskElts <= NumOpElts) |
2185 | return false; |
2186 | |
2187 | // The first part of the mask must choose elements from exactly 1 source op. |
2188 | ArrayRef<int> Mask = getShuffleMask(); |
2189 | if (!isIdentityMaskImpl(Mask, NumOpElts)) |
2190 | return false; |
2191 | |
2192 | // All extending must be with undef elements. |
2193 | for (int i = NumOpElts; i < NumMaskElts; ++i) |
2194 | if (Mask[i] != -1) |
2195 | return false; |
2196 | |
2197 | return true; |
2198 | } |
2199 | |
2200 | bool ShuffleVectorInst::() const { |
2201 | // FIXME: Not currently possible to express a shuffle mask for a scalable |
2202 | // vector for this case. |
2203 | if (isa<ScalableVectorType>(Val: getType())) |
2204 | return false; |
2205 | |
2206 | int NumOpElts = cast<FixedVectorType>(Val: Op<0>()->getType())->getNumElements(); |
2207 | int NumMaskElts = cast<FixedVectorType>(Val: getType())->getNumElements(); |
2208 | if (NumMaskElts >= NumOpElts) |
2209 | return false; |
2210 | |
2211 | return isIdentityMaskImpl(Mask: getShuffleMask(), NumOpElts); |
2212 | } |
2213 | |
2214 | bool ShuffleVectorInst::isConcat() const { |
2215 | // Vector concatenation is differentiated from identity with padding. |
2216 | if (isa<UndefValue>(Val: Op<0>()) || isa<UndefValue>(Val: Op<1>())) |
2217 | return false; |
2218 | |
2219 | // FIXME: Not currently possible to express a shuffle mask for a scalable |
2220 | // vector for this case. |
2221 | if (isa<ScalableVectorType>(Val: getType())) |
2222 | return false; |
2223 | |
2224 | int NumOpElts = cast<FixedVectorType>(Val: Op<0>()->getType())->getNumElements(); |
2225 | int NumMaskElts = cast<FixedVectorType>(Val: getType())->getNumElements(); |
2226 | if (NumMaskElts != NumOpElts * 2) |
2227 | return false; |
2228 | |
2229 | // Use the mask length rather than the operands' vector lengths here. We |
2230 | // already know that the shuffle returns a vector twice as long as the inputs, |
2231 | // and neither of the inputs are undef vectors. If the mask picks consecutive |
2232 | // elements from both inputs, then this is a concatenation of the inputs. |
2233 | return isIdentityMaskImpl(Mask: getShuffleMask(), NumOpElts: NumMaskElts); |
2234 | } |
2235 | |
2236 | static bool isReplicationMaskWithParams(ArrayRef<int> Mask, |
2237 | int ReplicationFactor, int VF) { |
2238 | assert(Mask.size() == (unsigned)ReplicationFactor * VF && |
2239 | "Unexpected mask size." ); |
2240 | |
2241 | for (int CurrElt : seq(Size: VF)) { |
2242 | ArrayRef<int> CurrSubMask = Mask.take_front(N: ReplicationFactor); |
2243 | assert(CurrSubMask.size() == (unsigned)ReplicationFactor && |
2244 | "Run out of mask?" ); |
2245 | Mask = Mask.drop_front(N: ReplicationFactor); |
2246 | if (!all_of(Range&: CurrSubMask, P: [CurrElt](int MaskElt) { |
2247 | return MaskElt == PoisonMaskElem || MaskElt == CurrElt; |
2248 | })) |
2249 | return false; |
2250 | } |
2251 | assert(Mask.empty() && "Did not consume the whole mask?" ); |
2252 | |
2253 | return true; |
2254 | } |
2255 | |
2256 | bool ShuffleVectorInst::isReplicationMask(ArrayRef<int> Mask, |
2257 | int &ReplicationFactor, int &VF) { |
2258 | // undef-less case is trivial. |
2259 | if (!llvm::is_contained(Range&: Mask, Element: PoisonMaskElem)) { |
2260 | ReplicationFactor = |
2261 | Mask.take_while(Pred: [](int MaskElt) { return MaskElt == 0; }).size(); |
2262 | if (ReplicationFactor == 0 || Mask.size() % ReplicationFactor != 0) |
2263 | return false; |
2264 | VF = Mask.size() / ReplicationFactor; |
2265 | return isReplicationMaskWithParams(Mask, ReplicationFactor, VF); |
2266 | } |
2267 | |
2268 | // However, if the mask contains undef's, we have to enumerate possible tuples |
2269 | // and pick one. There are bounds on replication factor: [1, mask size] |
2270 | // (where RF=1 is an identity shuffle, RF=mask size is a broadcast shuffle) |
2271 | // Additionally, mask size is a replication factor multiplied by vector size, |
2272 | // which further significantly reduces the search space. |
2273 | |
2274 | // Before doing that, let's perform basic correctness checking first. |
2275 | int Largest = -1; |
2276 | for (int MaskElt : Mask) { |
2277 | if (MaskElt == PoisonMaskElem) |
2278 | continue; |
2279 | // Elements must be in non-decreasing order. |
2280 | if (MaskElt < Largest) |
2281 | return false; |
2282 | Largest = std::max(a: Largest, b: MaskElt); |
2283 | } |
2284 | |
2285 | // Prefer larger replication factor if all else equal. |
2286 | for (int PossibleReplicationFactor : |
2287 | reverse(C: seq_inclusive<unsigned>(Begin: 1, End: Mask.size()))) { |
2288 | if (Mask.size() % PossibleReplicationFactor != 0) |
2289 | continue; |
2290 | int PossibleVF = Mask.size() / PossibleReplicationFactor; |
2291 | if (!isReplicationMaskWithParams(Mask, ReplicationFactor: PossibleReplicationFactor, |
2292 | VF: PossibleVF)) |
2293 | continue; |
2294 | ReplicationFactor = PossibleReplicationFactor; |
2295 | VF = PossibleVF; |
2296 | return true; |
2297 | } |
2298 | |
2299 | return false; |
2300 | } |
2301 | |
2302 | bool ShuffleVectorInst::isReplicationMask(int &ReplicationFactor, |
2303 | int &VF) const { |
2304 | // Not possible to express a shuffle mask for a scalable vector for this |
2305 | // case. |
2306 | if (isa<ScalableVectorType>(Val: getType())) |
2307 | return false; |
2308 | |
2309 | VF = cast<FixedVectorType>(Val: Op<0>()->getType())->getNumElements(); |
2310 | if (ShuffleMask.size() % VF != 0) |
2311 | return false; |
2312 | ReplicationFactor = ShuffleMask.size() / VF; |
2313 | |
2314 | return isReplicationMaskWithParams(Mask: ShuffleMask, ReplicationFactor, VF); |
2315 | } |
2316 | |
2317 | bool ShuffleVectorInst::isOneUseSingleSourceMask(ArrayRef<int> Mask, int VF) { |
2318 | if (VF <= 0 || Mask.size() < static_cast<unsigned>(VF) || |
2319 | Mask.size() % VF != 0) |
2320 | return false; |
2321 | for (unsigned K = 0, Sz = Mask.size(); K < Sz; K += VF) { |
2322 | ArrayRef<int> SubMask = Mask.slice(N: K, M: VF); |
2323 | if (all_of(Range&: SubMask, P: [](int Idx) { return Idx == PoisonMaskElem; })) |
2324 | continue; |
2325 | SmallBitVector Used(VF, false); |
2326 | for (int Idx : SubMask) { |
2327 | if (Idx != PoisonMaskElem && Idx < VF) |
2328 | Used.set(Idx); |
2329 | } |
2330 | if (!Used.all()) |
2331 | return false; |
2332 | } |
2333 | return true; |
2334 | } |
2335 | |
2336 | /// Return true if this shuffle mask is a replication mask. |
2337 | bool ShuffleVectorInst::isOneUseSingleSourceMask(int VF) const { |
2338 | // Not possible to express a shuffle mask for a scalable vector for this |
2339 | // case. |
2340 | if (isa<ScalableVectorType>(Val: getType())) |
2341 | return false; |
2342 | if (!isSingleSourceMask(Mask: ShuffleMask, NumSrcElts: VF)) |
2343 | return false; |
2344 | |
2345 | return isOneUseSingleSourceMask(Mask: ShuffleMask, VF); |
2346 | } |
2347 | |
2348 | bool ShuffleVectorInst::isInterleave(unsigned Factor) { |
2349 | FixedVectorType *OpTy = dyn_cast<FixedVectorType>(Val: getOperand(i_nocapture: 0)->getType()); |
2350 | // shuffle_vector can only interleave fixed length vectors - for scalable |
2351 | // vectors, see the @llvm.vector.interleave2 intrinsic |
2352 | if (!OpTy) |
2353 | return false; |
2354 | unsigned OpNumElts = OpTy->getNumElements(); |
2355 | |
2356 | return isInterleaveMask(Mask: ShuffleMask, Factor, NumInputElts: OpNumElts * 2); |
2357 | } |
2358 | |
2359 | bool ShuffleVectorInst::isInterleaveMask( |
2360 | ArrayRef<int> Mask, unsigned Factor, unsigned NumInputElts, |
2361 | SmallVectorImpl<unsigned> &StartIndexes) { |
2362 | unsigned NumElts = Mask.size(); |
2363 | if (NumElts % Factor) |
2364 | return false; |
2365 | |
2366 | unsigned LaneLen = NumElts / Factor; |
2367 | if (!isPowerOf2_32(Value: LaneLen)) |
2368 | return false; |
2369 | |
2370 | StartIndexes.resize(N: Factor); |
2371 | |
2372 | // Check whether each element matches the general interleaved rule. |
2373 | // Ignore undef elements, as long as the defined elements match the rule. |
2374 | // Outer loop processes all factors (x, y, z in the above example) |
2375 | unsigned I = 0, J; |
2376 | for (; I < Factor; I++) { |
2377 | unsigned SavedLaneValue; |
2378 | unsigned SavedNoUndefs = 0; |
2379 | |
2380 | // Inner loop processes consecutive accesses (x, x+1... in the example) |
2381 | for (J = 0; J < LaneLen - 1; J++) { |
2382 | // Lane computes x's position in the Mask |
2383 | unsigned Lane = J * Factor + I; |
2384 | unsigned NextLane = Lane + Factor; |
2385 | int LaneValue = Mask[Lane]; |
2386 | int NextLaneValue = Mask[NextLane]; |
2387 | |
2388 | // If both are defined, values must be sequential |
2389 | if (LaneValue >= 0 && NextLaneValue >= 0 && |
2390 | LaneValue + 1 != NextLaneValue) |
2391 | break; |
2392 | |
2393 | // If the next value is undef, save the current one as reference |
2394 | if (LaneValue >= 0 && NextLaneValue < 0) { |
2395 | SavedLaneValue = LaneValue; |
2396 | SavedNoUndefs = 1; |
2397 | } |
2398 | |
2399 | // Undefs are allowed, but defined elements must still be consecutive: |
2400 | // i.e.: x,..., undef,..., x + 2,..., undef,..., undef,..., x + 5, .... |
2401 | // Verify this by storing the last non-undef followed by an undef |
2402 | // Check that following non-undef masks are incremented with the |
2403 | // corresponding distance. |
2404 | if (SavedNoUndefs > 0 && LaneValue < 0) { |
2405 | SavedNoUndefs++; |
2406 | if (NextLaneValue >= 0 && |
2407 | SavedLaneValue + SavedNoUndefs != (unsigned)NextLaneValue) |
2408 | break; |
2409 | } |
2410 | } |
2411 | |
2412 | if (J < LaneLen - 1) |
2413 | return false; |
2414 | |
2415 | int StartMask = 0; |
2416 | if (Mask[I] >= 0) { |
2417 | // Check that the start of the I range (J=0) is greater than 0 |
2418 | StartMask = Mask[I]; |
2419 | } else if (Mask[(LaneLen - 1) * Factor + I] >= 0) { |
2420 | // StartMask defined by the last value in lane |
2421 | StartMask = Mask[(LaneLen - 1) * Factor + I] - J; |
2422 | } else if (SavedNoUndefs > 0) { |
2423 | // StartMask defined by some non-zero value in the j loop |
2424 | StartMask = SavedLaneValue - (LaneLen - 1 - SavedNoUndefs); |
2425 | } |
2426 | // else StartMask remains set to 0, i.e. all elements are undefs |
2427 | |
2428 | if (StartMask < 0) |
2429 | return false; |
2430 | // We must stay within the vectors; This case can happen with undefs. |
2431 | if (StartMask + LaneLen > NumInputElts) |
2432 | return false; |
2433 | |
2434 | StartIndexes[I] = StartMask; |
2435 | } |
2436 | |
2437 | return true; |
2438 | } |
2439 | |
2440 | /// Check if the mask is a DE-interleave mask of the given factor |
2441 | /// \p Factor like: |
2442 | /// <Index, Index+Factor, ..., Index+(NumElts-1)*Factor> |
2443 | bool ShuffleVectorInst::isDeInterleaveMaskOfFactor(ArrayRef<int> Mask, |
2444 | unsigned Factor, |
2445 | unsigned &Index) { |
2446 | // Check all potential start indices from 0 to (Factor - 1). |
2447 | for (unsigned Idx = 0; Idx < Factor; Idx++) { |
2448 | unsigned I = 0; |
2449 | |
2450 | // Check that elements are in ascending order by Factor. Ignore undef |
2451 | // elements. |
2452 | for (; I < Mask.size(); I++) |
2453 | if (Mask[I] >= 0 && static_cast<unsigned>(Mask[I]) != Idx + I * Factor) |
2454 | break; |
2455 | |
2456 | if (I == Mask.size()) { |
2457 | Index = Idx; |
2458 | return true; |
2459 | } |
2460 | } |
2461 | |
2462 | return false; |
2463 | } |
2464 | |
2465 | /// Try to lower a vector shuffle as a bit rotation. |
2466 | /// |
2467 | /// Look for a repeated rotation pattern in each sub group. |
2468 | /// Returns an element-wise left bit rotation amount or -1 if failed. |
2469 | static int matchShuffleAsBitRotate(ArrayRef<int> Mask, int NumSubElts) { |
2470 | int NumElts = Mask.size(); |
2471 | assert((NumElts % NumSubElts) == 0 && "Illegal shuffle mask" ); |
2472 | |
2473 | int RotateAmt = -1; |
2474 | for (int i = 0; i != NumElts; i += NumSubElts) { |
2475 | for (int j = 0; j != NumSubElts; ++j) { |
2476 | int M = Mask[i + j]; |
2477 | if (M < 0) |
2478 | continue; |
2479 | if (M < i || M >= i + NumSubElts) |
2480 | return -1; |
2481 | int Offset = (NumSubElts - (M - (i + j))) % NumSubElts; |
2482 | if (0 <= RotateAmt && Offset != RotateAmt) |
2483 | return -1; |
2484 | RotateAmt = Offset; |
2485 | } |
2486 | } |
2487 | return RotateAmt; |
2488 | } |
2489 | |
2490 | bool ShuffleVectorInst::isBitRotateMask( |
2491 | ArrayRef<int> Mask, unsigned EltSizeInBits, unsigned MinSubElts, |
2492 | unsigned MaxSubElts, unsigned &NumSubElts, unsigned &RotateAmt) { |
2493 | for (NumSubElts = MinSubElts; NumSubElts <= MaxSubElts; NumSubElts *= 2) { |
2494 | int EltRotateAmt = matchShuffleAsBitRotate(Mask, NumSubElts); |
2495 | if (EltRotateAmt < 0) |
2496 | continue; |
2497 | RotateAmt = EltRotateAmt * EltSizeInBits; |
2498 | return true; |
2499 | } |
2500 | |
2501 | return false; |
2502 | } |
2503 | |
2504 | //===----------------------------------------------------------------------===// |
2505 | // InsertValueInst Class |
2506 | //===----------------------------------------------------------------------===// |
2507 | |
2508 | void InsertValueInst::init(Value *Agg, Value *Val, ArrayRef<unsigned> Idxs, |
2509 | const Twine &Name) { |
2510 | assert(getNumOperands() == 2 && "NumOperands not initialized?" ); |
2511 | |
2512 | // There's no fundamental reason why we require at least one index |
2513 | // (other than weirdness with &*IdxBegin being invalid; see |
2514 | // getelementptr's init routine for example). But there's no |
2515 | // present need to support it. |
2516 | assert(!Idxs.empty() && "InsertValueInst must have at least one index" ); |
2517 | |
2518 | assert(ExtractValueInst::getIndexedType(Agg->getType(), Idxs) == |
2519 | Val->getType() && "Inserted value must match indexed type!" ); |
2520 | Op<0>() = Agg; |
2521 | Op<1>() = Val; |
2522 | |
2523 | Indices.append(in_start: Idxs.begin(), in_end: Idxs.end()); |
2524 | setName(Name); |
2525 | } |
2526 | |
2527 | InsertValueInst::InsertValueInst(const InsertValueInst &IVI) |
2528 | : Instruction(IVI.getType(), InsertValue, AllocMarker), |
2529 | Indices(IVI.Indices) { |
2530 | Op<0>() = IVI.getOperand(i_nocapture: 0); |
2531 | Op<1>() = IVI.getOperand(i_nocapture: 1); |
2532 | SubclassOptionalData = IVI.SubclassOptionalData; |
2533 | } |
2534 | |
2535 | //===----------------------------------------------------------------------===// |
2536 | // ExtractValueInst Class |
2537 | //===----------------------------------------------------------------------===// |
2538 | |
2539 | void ExtractValueInst::(ArrayRef<unsigned> Idxs, const Twine &Name) { |
2540 | assert(getNumOperands() == 1 && "NumOperands not initialized?" ); |
2541 | |
2542 | // There's no fundamental reason why we require at least one index. |
2543 | // But there's no present need to support it. |
2544 | assert(!Idxs.empty() && "ExtractValueInst must have at least one index" ); |
2545 | |
2546 | Indices.append(in_start: Idxs.begin(), in_end: Idxs.end()); |
2547 | setName(Name); |
2548 | } |
2549 | |
2550 | ExtractValueInst::(const ExtractValueInst &EVI) |
2551 | : UnaryInstruction(EVI.getType(), ExtractValue, EVI.getOperand(i_nocapture: 0), |
2552 | (BasicBlock *)nullptr), |
2553 | Indices(EVI.Indices) { |
2554 | SubclassOptionalData = EVI.SubclassOptionalData; |
2555 | } |
2556 | |
2557 | // getIndexedType - Returns the type of the element that would be extracted |
2558 | // with an extractvalue instruction with the specified parameters. |
2559 | // |
2560 | // A null type is returned if the indices are invalid for the specified |
2561 | // pointer type. |
2562 | // |
2563 | Type *ExtractValueInst::(Type *Agg, |
2564 | ArrayRef<unsigned> Idxs) { |
2565 | for (unsigned Index : Idxs) { |
2566 | // We can't use CompositeType::indexValid(Index) here. |
2567 | // indexValid() always returns true for arrays because getelementptr allows |
2568 | // out-of-bounds indices. Since we don't allow those for extractvalue and |
2569 | // insertvalue we need to check array indexing manually. |
2570 | // Since the only other types we can index into are struct types it's just |
2571 | // as easy to check those manually as well. |
2572 | if (ArrayType *AT = dyn_cast<ArrayType>(Val: Agg)) { |
2573 | if (Index >= AT->getNumElements()) |
2574 | return nullptr; |
2575 | Agg = AT->getElementType(); |
2576 | } else if (StructType *ST = dyn_cast<StructType>(Val: Agg)) { |
2577 | if (Index >= ST->getNumElements()) |
2578 | return nullptr; |
2579 | Agg = ST->getElementType(N: Index); |
2580 | } else { |
2581 | // Not a valid type to index into. |
2582 | return nullptr; |
2583 | } |
2584 | } |
2585 | return Agg; |
2586 | } |
2587 | |
2588 | //===----------------------------------------------------------------------===// |
2589 | // UnaryOperator Class |
2590 | //===----------------------------------------------------------------------===// |
2591 | |
2592 | UnaryOperator::UnaryOperator(UnaryOps iType, Value *S, Type *Ty, |
2593 | const Twine &Name, InsertPosition InsertBefore) |
2594 | : UnaryInstruction(Ty, iType, S, InsertBefore) { |
2595 | Op<0>() = S; |
2596 | setName(Name); |
2597 | AssertOK(); |
2598 | } |
2599 | |
2600 | UnaryOperator *UnaryOperator::Create(UnaryOps Op, Value *S, const Twine &Name, |
2601 | InsertPosition InsertBefore) { |
2602 | return new UnaryOperator(Op, S, S->getType(), Name, InsertBefore); |
2603 | } |
2604 | |
2605 | void UnaryOperator::AssertOK() { |
2606 | Value *LHS = getOperand(i_nocapture: 0); |
2607 | (void)LHS; // Silence warnings. |
2608 | #ifndef NDEBUG |
2609 | switch (getOpcode()) { |
2610 | case FNeg: |
2611 | assert(getType() == LHS->getType() && |
2612 | "Unary operation should return same type as operand!" ); |
2613 | assert(getType()->isFPOrFPVectorTy() && |
2614 | "Tried to create a floating-point operation on a " |
2615 | "non-floating-point type!" ); |
2616 | break; |
2617 | default: llvm_unreachable("Invalid opcode provided" ); |
2618 | } |
2619 | #endif |
2620 | } |
2621 | |
2622 | //===----------------------------------------------------------------------===// |
2623 | // BinaryOperator Class |
2624 | //===----------------------------------------------------------------------===// |
2625 | |
2626 | BinaryOperator::BinaryOperator(BinaryOps iType, Value *S1, Value *S2, Type *Ty, |
2627 | const Twine &Name, InsertPosition InsertBefore) |
2628 | : Instruction(Ty, iType, AllocMarker, InsertBefore) { |
2629 | Op<0>() = S1; |
2630 | Op<1>() = S2; |
2631 | setName(Name); |
2632 | AssertOK(); |
2633 | } |
2634 | |
2635 | void BinaryOperator::AssertOK() { |
2636 | Value *LHS = getOperand(i_nocapture: 0), *RHS = getOperand(i_nocapture: 1); |
2637 | (void)LHS; (void)RHS; // Silence warnings. |
2638 | assert(LHS->getType() == RHS->getType() && |
2639 | "Binary operator operand types must match!" ); |
2640 | #ifndef NDEBUG |
2641 | switch (getOpcode()) { |
2642 | case Add: case Sub: |
2643 | case Mul: |
2644 | assert(getType() == LHS->getType() && |
2645 | "Arithmetic operation should return same type as operands!" ); |
2646 | assert(getType()->isIntOrIntVectorTy() && |
2647 | "Tried to create an integer operation on a non-integer type!" ); |
2648 | break; |
2649 | case FAdd: case FSub: |
2650 | case FMul: |
2651 | assert(getType() == LHS->getType() && |
2652 | "Arithmetic operation should return same type as operands!" ); |
2653 | assert(getType()->isFPOrFPVectorTy() && |
2654 | "Tried to create a floating-point operation on a " |
2655 | "non-floating-point type!" ); |
2656 | break; |
2657 | case UDiv: |
2658 | case SDiv: |
2659 | assert(getType() == LHS->getType() && |
2660 | "Arithmetic operation should return same type as operands!" ); |
2661 | assert(getType()->isIntOrIntVectorTy() && |
2662 | "Incorrect operand type (not integer) for S/UDIV" ); |
2663 | break; |
2664 | case FDiv: |
2665 | assert(getType() == LHS->getType() && |
2666 | "Arithmetic operation should return same type as operands!" ); |
2667 | assert(getType()->isFPOrFPVectorTy() && |
2668 | "Incorrect operand type (not floating point) for FDIV" ); |
2669 | break; |
2670 | case URem: |
2671 | case SRem: |
2672 | assert(getType() == LHS->getType() && |
2673 | "Arithmetic operation should return same type as operands!" ); |
2674 | assert(getType()->isIntOrIntVectorTy() && |
2675 | "Incorrect operand type (not integer) for S/UREM" ); |
2676 | break; |
2677 | case FRem: |
2678 | assert(getType() == LHS->getType() && |
2679 | "Arithmetic operation should return same type as operands!" ); |
2680 | assert(getType()->isFPOrFPVectorTy() && |
2681 | "Incorrect operand type (not floating point) for FREM" ); |
2682 | break; |
2683 | case Shl: |
2684 | case LShr: |
2685 | case AShr: |
2686 | assert(getType() == LHS->getType() && |
2687 | "Shift operation should return same type as operands!" ); |
2688 | assert(getType()->isIntOrIntVectorTy() && |
2689 | "Tried to create a shift operation on a non-integral type!" ); |
2690 | break; |
2691 | case And: case Or: |
2692 | case Xor: |
2693 | assert(getType() == LHS->getType() && |
2694 | "Logical operation should return same type as operands!" ); |
2695 | assert(getType()->isIntOrIntVectorTy() && |
2696 | "Tried to create a logical operation on a non-integral type!" ); |
2697 | break; |
2698 | default: llvm_unreachable("Invalid opcode provided" ); |
2699 | } |
2700 | #endif |
2701 | } |
2702 | |
2703 | BinaryOperator *BinaryOperator::Create(BinaryOps Op, Value *S1, Value *S2, |
2704 | const Twine &Name, |
2705 | InsertPosition InsertBefore) { |
2706 | assert(S1->getType() == S2->getType() && |
2707 | "Cannot create binary operator with two operands of differing type!" ); |
2708 | return new BinaryOperator(Op, S1, S2, S1->getType(), Name, InsertBefore); |
2709 | } |
2710 | |
2711 | BinaryOperator *BinaryOperator::CreateNeg(Value *Op, const Twine &Name, |
2712 | InsertPosition InsertBefore) { |
2713 | Value *Zero = ConstantInt::get(Ty: Op->getType(), V: 0); |
2714 | return new BinaryOperator(Instruction::Sub, Zero, Op, Op->getType(), Name, |
2715 | InsertBefore); |
2716 | } |
2717 | |
2718 | BinaryOperator *BinaryOperator::CreateNSWNeg(Value *Op, const Twine &Name, |
2719 | InsertPosition InsertBefore) { |
2720 | Value *Zero = ConstantInt::get(Ty: Op->getType(), V: 0); |
2721 | return BinaryOperator::CreateNSWSub(V1: Zero, V2: Op, Name, InsertBefore); |
2722 | } |
2723 | |
2724 | BinaryOperator *BinaryOperator::CreateNot(Value *Op, const Twine &Name, |
2725 | InsertPosition InsertBefore) { |
2726 | Constant *C = Constant::getAllOnesValue(Ty: Op->getType()); |
2727 | return new BinaryOperator(Instruction::Xor, Op, C, |
2728 | Op->getType(), Name, InsertBefore); |
2729 | } |
2730 | |
2731 | // Exchange the two operands to this instruction. This instruction is safe to |
2732 | // use on any binary instruction and does not modify the semantics of the |
2733 | // instruction. |
2734 | bool BinaryOperator::swapOperands() { |
2735 | if (!isCommutative()) |
2736 | return true; // Can't commute operands |
2737 | Op<0>().swap(RHS&: Op<1>()); |
2738 | return false; |
2739 | } |
2740 | |
2741 | //===----------------------------------------------------------------------===// |
2742 | // FPMathOperator Class |
2743 | //===----------------------------------------------------------------------===// |
2744 | |
2745 | float FPMathOperator::getFPAccuracy() const { |
2746 | const MDNode *MD = |
2747 | cast<Instruction>(Val: this)->getMetadata(KindID: LLVMContext::MD_fpmath); |
2748 | if (!MD) |
2749 | return 0.0; |
2750 | ConstantFP *Accuracy = mdconst::extract<ConstantFP>(MD: MD->getOperand(I: 0)); |
2751 | return Accuracy->getValueAPF().convertToFloat(); |
2752 | } |
2753 | |
2754 | //===----------------------------------------------------------------------===// |
2755 | // CastInst Class |
2756 | //===----------------------------------------------------------------------===// |
2757 | |
2758 | // Just determine if this cast only deals with integral->integral conversion. |
2759 | bool CastInst::isIntegerCast() const { |
2760 | switch (getOpcode()) { |
2761 | default: return false; |
2762 | case Instruction::ZExt: |
2763 | case Instruction::SExt: |
2764 | case Instruction::Trunc: |
2765 | return true; |
2766 | case Instruction::BitCast: |
2767 | return getOperand(i_nocapture: 0)->getType()->isIntegerTy() && |
2768 | getType()->isIntegerTy(); |
2769 | } |
2770 | } |
2771 | |
2772 | /// This function determines if the CastInst does not require any bits to be |
2773 | /// changed in order to effect the cast. Essentially, it identifies cases where |
2774 | /// no code gen is necessary for the cast, hence the name no-op cast. For |
2775 | /// example, the following are all no-op casts: |
2776 | /// # bitcast i32* %x to i8* |
2777 | /// # bitcast <2 x i32> %x to <4 x i16> |
2778 | /// # ptrtoint i32* %x to i32 ; on 32-bit plaforms only |
2779 | /// Determine if the described cast is a no-op. |
2780 | bool CastInst::isNoopCast(Instruction::CastOps Opcode, |
2781 | Type *SrcTy, |
2782 | Type *DestTy, |
2783 | const DataLayout &DL) { |
2784 | assert(castIsValid(Opcode, SrcTy, DestTy) && "method precondition" ); |
2785 | switch (Opcode) { |
2786 | default: llvm_unreachable("Invalid CastOp" ); |
2787 | case Instruction::Trunc: |
2788 | case Instruction::ZExt: |
2789 | case Instruction::SExt: |
2790 | case Instruction::FPTrunc: |
2791 | case Instruction::FPExt: |
2792 | case Instruction::UIToFP: |
2793 | case Instruction::SIToFP: |
2794 | case Instruction::FPToUI: |
2795 | case Instruction::FPToSI: |
2796 | case Instruction::AddrSpaceCast: |
2797 | // TODO: Target informations may give a more accurate answer here. |
2798 | return false; |
2799 | case Instruction::BitCast: |
2800 | return true; // BitCast never modifies bits. |
2801 | case Instruction::PtrToInt: |
2802 | return DL.getIntPtrType(SrcTy)->getScalarSizeInBits() == |
2803 | DestTy->getScalarSizeInBits(); |
2804 | case Instruction::IntToPtr: |
2805 | return DL.getIntPtrType(DestTy)->getScalarSizeInBits() == |
2806 | SrcTy->getScalarSizeInBits(); |
2807 | } |
2808 | } |
2809 | |
2810 | bool CastInst::isNoopCast(const DataLayout &DL) const { |
2811 | return isNoopCast(Opcode: getOpcode(), SrcTy: getOperand(i_nocapture: 0)->getType(), DestTy: getType(), DL); |
2812 | } |
2813 | |
2814 | /// This function determines if a pair of casts can be eliminated and what |
2815 | /// opcode should be used in the elimination. This assumes that there are two |
2816 | /// instructions like this: |
2817 | /// * %F = firstOpcode SrcTy %x to MidTy |
2818 | /// * %S = secondOpcode MidTy %F to DstTy |
2819 | /// The function returns a resultOpcode so these two casts can be replaced with: |
2820 | /// * %Replacement = resultOpcode %SrcTy %x to DstTy |
2821 | /// If no such cast is permitted, the function returns 0. |
2822 | unsigned CastInst::isEliminableCastPair( |
2823 | Instruction::CastOps firstOp, Instruction::CastOps secondOp, |
2824 | Type *SrcTy, Type *MidTy, Type *DstTy, Type *SrcIntPtrTy, Type *MidIntPtrTy, |
2825 | Type *DstIntPtrTy) { |
2826 | // Define the 144 possibilities for these two cast instructions. The values |
2827 | // in this matrix determine what to do in a given situation and select the |
2828 | // case in the switch below. The rows correspond to firstOp, the columns |
2829 | // correspond to secondOp. In looking at the table below, keep in mind |
2830 | // the following cast properties: |
2831 | // |
2832 | // Size Compare Source Destination |
2833 | // Operator Src ? Size Type Sign Type Sign |
2834 | // -------- ------------ ------------------- --------------------- |
2835 | // TRUNC > Integer Any Integral Any |
2836 | // ZEXT < Integral Unsigned Integer Any |
2837 | // SEXT < Integral Signed Integer Any |
2838 | // FPTOUI n/a FloatPt n/a Integral Unsigned |
2839 | // FPTOSI n/a FloatPt n/a Integral Signed |
2840 | // UITOFP n/a Integral Unsigned FloatPt n/a |
2841 | // SITOFP n/a Integral Signed FloatPt n/a |
2842 | // FPTRUNC > FloatPt n/a FloatPt n/a |
2843 | // FPEXT < FloatPt n/a FloatPt n/a |
2844 | // PTRTOINT n/a Pointer n/a Integral Unsigned |
2845 | // INTTOPTR n/a Integral Unsigned Pointer n/a |
2846 | // BITCAST = FirstClass n/a FirstClass n/a |
2847 | // ADDRSPCST n/a Pointer n/a Pointer n/a |
2848 | // |
2849 | // NOTE: some transforms are safe, but we consider them to be non-profitable. |
2850 | // For example, we could merge "fptoui double to i32" + "zext i32 to i64", |
2851 | // into "fptoui double to i64", but this loses information about the range |
2852 | // of the produced value (we no longer know the top-part is all zeros). |
2853 | // Further this conversion is often much more expensive for typical hardware, |
2854 | // and causes issues when building libgcc. We disallow fptosi+sext for the |
2855 | // same reason. |
2856 | const unsigned numCastOps = |
2857 | Instruction::CastOpsEnd - Instruction::CastOpsBegin; |
2858 | static const uint8_t CastResults[numCastOps][numCastOps] = { |
2859 | // T F F U S F F P I B A -+ |
2860 | // R Z S P P I I T P 2 N T S | |
2861 | // U E E 2 2 2 2 R E I T C C +- secondOp |
2862 | // N X X U S F F N X N 2 V V | |
2863 | // C T T I I P P C T T P T T -+ |
2864 | { 1, 0, 0,99,99, 0, 0,99,99,99, 0, 3, 0}, // Trunc -+ |
2865 | { 8, 1, 9,99,99, 2,17,99,99,99, 2, 3, 0}, // ZExt | |
2866 | { 8, 0, 1,99,99, 0, 2,99,99,99, 0, 3, 0}, // SExt | |
2867 | { 0, 0, 0,99,99, 0, 0,99,99,99, 0, 3, 0}, // FPToUI | |
2868 | { 0, 0, 0,99,99, 0, 0,99,99,99, 0, 3, 0}, // FPToSI | |
2869 | { 99,99,99, 0, 0,99,99, 0, 0,99,99, 4, 0}, // UIToFP +- firstOp |
2870 | { 99,99,99, 0, 0,99,99, 0, 0,99,99, 4, 0}, // SIToFP | |
2871 | { 99,99,99, 0, 0,99,99, 0, 0,99,99, 4, 0}, // FPTrunc | |
2872 | { 99,99,99, 2, 2,99,99, 8, 2,99,99, 4, 0}, // FPExt | |
2873 | { 1, 0, 0,99,99, 0, 0,99,99,99, 7, 3, 0}, // PtrToInt | |
2874 | { 99,99,99,99,99,99,99,99,99,11,99,15, 0}, // IntToPtr | |
2875 | { 5, 5, 5, 0, 0, 5, 5, 0, 0,16, 5, 1,14}, // BitCast | |
2876 | { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,13,12}, // AddrSpaceCast -+ |
2877 | }; |
2878 | |
2879 | // TODO: This logic could be encoded into the table above and handled in the |
2880 | // switch below. |
2881 | // If either of the casts are a bitcast from scalar to vector, disallow the |
2882 | // merging. However, any pair of bitcasts are allowed. |
2883 | bool IsFirstBitcast = (firstOp == Instruction::BitCast); |
2884 | bool IsSecondBitcast = (secondOp == Instruction::BitCast); |
2885 | bool AreBothBitcasts = IsFirstBitcast && IsSecondBitcast; |
2886 | |
2887 | // Check if any of the casts convert scalars <-> vectors. |
2888 | if ((IsFirstBitcast && isa<VectorType>(Val: SrcTy) != isa<VectorType>(Val: MidTy)) || |
2889 | (IsSecondBitcast && isa<VectorType>(Val: MidTy) != isa<VectorType>(Val: DstTy))) |
2890 | if (!AreBothBitcasts) |
2891 | return 0; |
2892 | |
2893 | int ElimCase = CastResults[firstOp-Instruction::CastOpsBegin] |
2894 | [secondOp-Instruction::CastOpsBegin]; |
2895 | switch (ElimCase) { |
2896 | case 0: |
2897 | // Categorically disallowed. |
2898 | return 0; |
2899 | case 1: |
2900 | // Allowed, use first cast's opcode. |
2901 | return firstOp; |
2902 | case 2: |
2903 | // Allowed, use second cast's opcode. |
2904 | return secondOp; |
2905 | case 3: |
2906 | // No-op cast in second op implies firstOp as long as the DestTy |
2907 | // is integer and we are not converting between a vector and a |
2908 | // non-vector type. |
2909 | if (!SrcTy->isVectorTy() && DstTy->isIntegerTy()) |
2910 | return firstOp; |
2911 | return 0; |
2912 | case 4: |
2913 | // No-op cast in second op implies firstOp as long as the DestTy |
2914 | // matches MidTy. |
2915 | if (DstTy == MidTy) |
2916 | return firstOp; |
2917 | return 0; |
2918 | case 5: |
2919 | // No-op cast in first op implies secondOp as long as the SrcTy |
2920 | // is an integer. |
2921 | if (SrcTy->isIntegerTy()) |
2922 | return secondOp; |
2923 | return 0; |
2924 | case 7: { |
2925 | // Disable inttoptr/ptrtoint optimization if enabled. |
2926 | if (DisableI2pP2iOpt) |
2927 | return 0; |
2928 | |
2929 | // Cannot simplify if address spaces are different! |
2930 | if (SrcTy->getPointerAddressSpace() != DstTy->getPointerAddressSpace()) |
2931 | return 0; |
2932 | |
2933 | unsigned MidSize = MidTy->getScalarSizeInBits(); |
2934 | // We can still fold this without knowing the actual sizes as long we |
2935 | // know that the intermediate pointer is the largest possible |
2936 | // pointer size. |
2937 | // FIXME: Is this always true? |
2938 | if (MidSize == 64) |
2939 | return Instruction::BitCast; |
2940 | |
2941 | // ptrtoint, inttoptr -> bitcast (ptr -> ptr) if int size is >= ptr size. |
2942 | if (!SrcIntPtrTy || DstIntPtrTy != SrcIntPtrTy) |
2943 | return 0; |
2944 | unsigned PtrSize = SrcIntPtrTy->getScalarSizeInBits(); |
2945 | if (MidSize >= PtrSize) |
2946 | return Instruction::BitCast; |
2947 | return 0; |
2948 | } |
2949 | case 8: { |
2950 | // ext, trunc -> bitcast, if the SrcTy and DstTy are the same |
2951 | // ext, trunc -> ext, if sizeof(SrcTy) < sizeof(DstTy) |
2952 | // ext, trunc -> trunc, if sizeof(SrcTy) > sizeof(DstTy) |
2953 | unsigned SrcSize = SrcTy->getScalarSizeInBits(); |
2954 | unsigned DstSize = DstTy->getScalarSizeInBits(); |
2955 | if (SrcTy == DstTy) |
2956 | return Instruction::BitCast; |
2957 | if (SrcSize < DstSize) |
2958 | return firstOp; |
2959 | if (SrcSize > DstSize) |
2960 | return secondOp; |
2961 | return 0; |
2962 | } |
2963 | case 9: |
2964 | // zext, sext -> zext, because sext can't sign extend after zext |
2965 | return Instruction::ZExt; |
2966 | case 11: { |
2967 | // inttoptr, ptrtoint -> bitcast if SrcSize<=PtrSize and SrcSize==DstSize |
2968 | if (!MidIntPtrTy) |
2969 | return 0; |
2970 | unsigned PtrSize = MidIntPtrTy->getScalarSizeInBits(); |
2971 | unsigned SrcSize = SrcTy->getScalarSizeInBits(); |
2972 | unsigned DstSize = DstTy->getScalarSizeInBits(); |
2973 | if (SrcSize <= PtrSize && SrcSize == DstSize) |
2974 | return Instruction::BitCast; |
2975 | return 0; |
2976 | } |
2977 | case 12: |
2978 | // addrspacecast, addrspacecast -> bitcast, if SrcAS == DstAS |
2979 | // addrspacecast, addrspacecast -> addrspacecast, if SrcAS != DstAS |
2980 | if (SrcTy->getPointerAddressSpace() != DstTy->getPointerAddressSpace()) |
2981 | return Instruction::AddrSpaceCast; |
2982 | return Instruction::BitCast; |
2983 | case 13: |
2984 | // FIXME: this state can be merged with (1), but the following assert |
2985 | // is useful to check the correcteness of the sequence due to semantic |
2986 | // change of bitcast. |
2987 | assert( |
2988 | SrcTy->isPtrOrPtrVectorTy() && |
2989 | MidTy->isPtrOrPtrVectorTy() && |
2990 | DstTy->isPtrOrPtrVectorTy() && |
2991 | SrcTy->getPointerAddressSpace() != MidTy->getPointerAddressSpace() && |
2992 | MidTy->getPointerAddressSpace() == DstTy->getPointerAddressSpace() && |
2993 | "Illegal addrspacecast, bitcast sequence!" ); |
2994 | // Allowed, use first cast's opcode |
2995 | return firstOp; |
2996 | case 14: |
2997 | // bitcast, addrspacecast -> addrspacecast |
2998 | return Instruction::AddrSpaceCast; |
2999 | case 15: |
3000 | // FIXME: this state can be merged with (1), but the following assert |
3001 | // is useful to check the correcteness of the sequence due to semantic |
3002 | // change of bitcast. |
3003 | assert( |
3004 | SrcTy->isIntOrIntVectorTy() && |
3005 | MidTy->isPtrOrPtrVectorTy() && |
3006 | DstTy->isPtrOrPtrVectorTy() && |
3007 | MidTy->getPointerAddressSpace() == DstTy->getPointerAddressSpace() && |
3008 | "Illegal inttoptr, bitcast sequence!" ); |
3009 | // Allowed, use first cast's opcode |
3010 | return firstOp; |
3011 | case 16: |
3012 | // FIXME: this state can be merged with (2), but the following assert |
3013 | // is useful to check the correcteness of the sequence due to semantic |
3014 | // change of bitcast. |
3015 | assert( |
3016 | SrcTy->isPtrOrPtrVectorTy() && |
3017 | MidTy->isPtrOrPtrVectorTy() && |
3018 | DstTy->isIntOrIntVectorTy() && |
3019 | SrcTy->getPointerAddressSpace() == MidTy->getPointerAddressSpace() && |
3020 | "Illegal bitcast, ptrtoint sequence!" ); |
3021 | // Allowed, use second cast's opcode |
3022 | return secondOp; |
3023 | case 17: |
3024 | // (sitofp (zext x)) -> (uitofp x) |
3025 | return Instruction::UIToFP; |
3026 | case 99: |
3027 | // Cast combination can't happen (error in input). This is for all cases |
3028 | // where the MidTy is not the same for the two cast instructions. |
3029 | llvm_unreachable("Invalid Cast Combination" ); |
3030 | default: |
3031 | llvm_unreachable("Error in CastResults table!!!" ); |
3032 | } |
3033 | } |
3034 | |
3035 | CastInst *CastInst::Create(Instruction::CastOps op, Value *S, Type *Ty, |
3036 | const Twine &Name, InsertPosition InsertBefore) { |
3037 | assert(castIsValid(op, S, Ty) && "Invalid cast!" ); |
3038 | // Construct and return the appropriate CastInst subclass |
3039 | switch (op) { |
3040 | case Trunc: return new TruncInst (S, Ty, Name, InsertBefore); |
3041 | case ZExt: return new ZExtInst (S, Ty, Name, InsertBefore); |
3042 | case SExt: return new SExtInst (S, Ty, Name, InsertBefore); |
3043 | case FPTrunc: return new FPTruncInst (S, Ty, Name, InsertBefore); |
3044 | case FPExt: return new FPExtInst (S, Ty, Name, InsertBefore); |
3045 | case UIToFP: return new UIToFPInst (S, Ty, Name, InsertBefore); |
3046 | case SIToFP: return new SIToFPInst (S, Ty, Name, InsertBefore); |
3047 | case FPToUI: return new FPToUIInst (S, Ty, Name, InsertBefore); |
3048 | case FPToSI: return new FPToSIInst (S, Ty, Name, InsertBefore); |
3049 | case PtrToInt: return new PtrToIntInst (S, Ty, Name, InsertBefore); |
3050 | case IntToPtr: return new IntToPtrInst (S, Ty, Name, InsertBefore); |
3051 | case BitCast: |
3052 | return new BitCastInst(S, Ty, Name, InsertBefore); |
3053 | case AddrSpaceCast: |
3054 | return new AddrSpaceCastInst(S, Ty, Name, InsertBefore); |
3055 | default: |
3056 | llvm_unreachable("Invalid opcode provided" ); |
3057 | } |
3058 | } |
3059 | |
3060 | CastInst *CastInst::CreateZExtOrBitCast(Value *S, Type *Ty, const Twine &Name, |
3061 | InsertPosition InsertBefore) { |
3062 | if (S->getType()->getScalarSizeInBits() == Ty->getScalarSizeInBits()) |
3063 | return Create(op: Instruction::BitCast, S, Ty, Name, InsertBefore); |
3064 | return Create(op: Instruction::ZExt, S, Ty, Name, InsertBefore); |
3065 | } |
3066 | |
3067 | CastInst *CastInst::CreateSExtOrBitCast(Value *S, Type *Ty, const Twine &Name, |
3068 | InsertPosition InsertBefore) { |
3069 | if (S->getType()->getScalarSizeInBits() == Ty->getScalarSizeInBits()) |
3070 | return Create(op: Instruction::BitCast, S, Ty, Name, InsertBefore); |
3071 | return Create(op: Instruction::SExt, S, Ty, Name, InsertBefore); |
3072 | } |
3073 | |
3074 | CastInst *CastInst::CreateTruncOrBitCast(Value *S, Type *Ty, const Twine &Name, |
3075 | InsertPosition InsertBefore) { |
3076 | if (S->getType()->getScalarSizeInBits() == Ty->getScalarSizeInBits()) |
3077 | return Create(op: Instruction::BitCast, S, Ty, Name, InsertBefore); |
3078 | return Create(op: Instruction::Trunc, S, Ty, Name, InsertBefore); |
3079 | } |
3080 | |
3081 | /// Create a BitCast or a PtrToInt cast instruction |
3082 | CastInst *CastInst::CreatePointerCast(Value *S, Type *Ty, const Twine &Name, |
3083 | InsertPosition InsertBefore) { |
3084 | assert(S->getType()->isPtrOrPtrVectorTy() && "Invalid cast" ); |
3085 | assert((Ty->isIntOrIntVectorTy() || Ty->isPtrOrPtrVectorTy()) && |
3086 | "Invalid cast" ); |
3087 | assert(Ty->isVectorTy() == S->getType()->isVectorTy() && "Invalid cast" ); |
3088 | assert((!Ty->isVectorTy() || |
3089 | cast<VectorType>(Ty)->getElementCount() == |
3090 | cast<VectorType>(S->getType())->getElementCount()) && |
3091 | "Invalid cast" ); |
3092 | |
3093 | if (Ty->isIntOrIntVectorTy()) |
3094 | return Create(op: Instruction::PtrToInt, S, Ty, Name, InsertBefore); |
3095 | |
3096 | return CreatePointerBitCastOrAddrSpaceCast(S, Ty, Name, InsertBefore); |
3097 | } |
3098 | |
3099 | CastInst *CastInst::CreatePointerBitCastOrAddrSpaceCast( |
3100 | Value *S, Type *Ty, const Twine &Name, InsertPosition InsertBefore) { |
3101 | assert(S->getType()->isPtrOrPtrVectorTy() && "Invalid cast" ); |
3102 | assert(Ty->isPtrOrPtrVectorTy() && "Invalid cast" ); |
3103 | |
3104 | if (S->getType()->getPointerAddressSpace() != Ty->getPointerAddressSpace()) |
3105 | return Create(op: Instruction::AddrSpaceCast, S, Ty, Name, InsertBefore); |
3106 | |
3107 | return Create(op: Instruction::BitCast, S, Ty, Name, InsertBefore); |
3108 | } |
3109 | |
3110 | CastInst *CastInst::CreateBitOrPointerCast(Value *S, Type *Ty, |
3111 | const Twine &Name, |
3112 | InsertPosition InsertBefore) { |
3113 | if (S->getType()->isPointerTy() && Ty->isIntegerTy()) |
3114 | return Create(op: Instruction::PtrToInt, S, Ty, Name, InsertBefore); |
3115 | if (S->getType()->isIntegerTy() && Ty->isPointerTy()) |
3116 | return Create(op: Instruction::IntToPtr, S, Ty, Name, InsertBefore); |
3117 | |
3118 | return Create(op: Instruction::BitCast, S, Ty, Name, InsertBefore); |
3119 | } |
3120 | |
3121 | CastInst *CastInst::CreateIntegerCast(Value *C, Type *Ty, bool isSigned, |
3122 | const Twine &Name, |
3123 | InsertPosition InsertBefore) { |
3124 | assert(C->getType()->isIntOrIntVectorTy() && Ty->isIntOrIntVectorTy() && |
3125 | "Invalid integer cast" ); |
3126 | unsigned SrcBits = C->getType()->getScalarSizeInBits(); |
3127 | unsigned DstBits = Ty->getScalarSizeInBits(); |
3128 | Instruction::CastOps opcode = |
3129 | (SrcBits == DstBits ? Instruction::BitCast : |
3130 | (SrcBits > DstBits ? Instruction::Trunc : |
3131 | (isSigned ? Instruction::SExt : Instruction::ZExt))); |
3132 | return Create(op: opcode, S: C, Ty, Name, InsertBefore); |
3133 | } |
3134 | |
3135 | CastInst *CastInst::CreateFPCast(Value *C, Type *Ty, const Twine &Name, |
3136 | InsertPosition InsertBefore) { |
3137 | assert(C->getType()->isFPOrFPVectorTy() && Ty->isFPOrFPVectorTy() && |
3138 | "Invalid cast" ); |
3139 | unsigned SrcBits = C->getType()->getScalarSizeInBits(); |
3140 | unsigned DstBits = Ty->getScalarSizeInBits(); |
3141 | assert((C->getType() == Ty || SrcBits != DstBits) && "Invalid cast" ); |
3142 | Instruction::CastOps opcode = |
3143 | (SrcBits == DstBits ? Instruction::BitCast : |
3144 | (SrcBits > DstBits ? Instruction::FPTrunc : Instruction::FPExt)); |
3145 | return Create(op: opcode, S: C, Ty, Name, InsertBefore); |
3146 | } |
3147 | |
3148 | bool CastInst::isBitCastable(Type *SrcTy, Type *DestTy) { |
3149 | if (!SrcTy->isFirstClassType() || !DestTy->isFirstClassType()) |
3150 | return false; |
3151 | |
3152 | if (SrcTy == DestTy) |
3153 | return true; |
3154 | |
3155 | if (VectorType *SrcVecTy = dyn_cast<VectorType>(Val: SrcTy)) { |
3156 | if (VectorType *DestVecTy = dyn_cast<VectorType>(Val: DestTy)) { |
3157 | if (SrcVecTy->getElementCount() == DestVecTy->getElementCount()) { |
3158 | // An element by element cast. Valid if casting the elements is valid. |
3159 | SrcTy = SrcVecTy->getElementType(); |
3160 | DestTy = DestVecTy->getElementType(); |
3161 | } |
3162 | } |
3163 | } |
3164 | |
3165 | if (PointerType *DestPtrTy = dyn_cast<PointerType>(Val: DestTy)) { |
3166 | if (PointerType *SrcPtrTy = dyn_cast<PointerType>(Val: SrcTy)) { |
3167 | return SrcPtrTy->getAddressSpace() == DestPtrTy->getAddressSpace(); |
3168 | } |
3169 | } |
3170 | |
3171 | TypeSize SrcBits = SrcTy->getPrimitiveSizeInBits(); // 0 for ptr |
3172 | TypeSize DestBits = DestTy->getPrimitiveSizeInBits(); // 0 for ptr |
3173 | |
3174 | // Could still have vectors of pointers if the number of elements doesn't |
3175 | // match |
3176 | if (SrcBits.getKnownMinValue() == 0 || DestBits.getKnownMinValue() == 0) |
3177 | return false; |
3178 | |
3179 | if (SrcBits != DestBits) |
3180 | return false; |
3181 | |
3182 | return true; |
3183 | } |
3184 | |
3185 | bool CastInst::isBitOrNoopPointerCastable(Type *SrcTy, Type *DestTy, |
3186 | const DataLayout &DL) { |
3187 | // ptrtoint and inttoptr are not allowed on non-integral pointers |
3188 | if (auto *PtrTy = dyn_cast<PointerType>(Val: SrcTy)) |
3189 | if (auto *IntTy = dyn_cast<IntegerType>(Val: DestTy)) |
3190 | return (IntTy->getBitWidth() == DL.getPointerTypeSizeInBits(PtrTy) && |
3191 | !DL.isNonIntegralPointerType(PT: PtrTy)); |
3192 | if (auto *PtrTy = dyn_cast<PointerType>(Val: DestTy)) |
3193 | if (auto *IntTy = dyn_cast<IntegerType>(Val: SrcTy)) |
3194 | return (IntTy->getBitWidth() == DL.getPointerTypeSizeInBits(PtrTy) && |
3195 | !DL.isNonIntegralPointerType(PT: PtrTy)); |
3196 | |
3197 | return isBitCastable(SrcTy, DestTy); |
3198 | } |
3199 | |
3200 | // Provide a way to get a "cast" where the cast opcode is inferred from the |
3201 | // types and size of the operand. This, basically, is a parallel of the |
3202 | // logic in the castIsValid function below. This axiom should hold: |
3203 | // castIsValid( getCastOpcode(Val, Ty), Val, Ty) |
3204 | // should not assert in castIsValid. In other words, this produces a "correct" |
3205 | // casting opcode for the arguments passed to it. |
3206 | Instruction::CastOps |
3207 | CastInst::getCastOpcode( |
3208 | const Value *Src, bool SrcIsSigned, Type *DestTy, bool DestIsSigned) { |
3209 | Type *SrcTy = Src->getType(); |
3210 | |
3211 | assert(SrcTy->isFirstClassType() && DestTy->isFirstClassType() && |
3212 | "Only first class types are castable!" ); |
3213 | |
3214 | if (SrcTy == DestTy) |
3215 | return BitCast; |
3216 | |
3217 | // FIXME: Check address space sizes here |
3218 | if (VectorType *SrcVecTy = dyn_cast<VectorType>(Val: SrcTy)) |
3219 | if (VectorType *DestVecTy = dyn_cast<VectorType>(Val: DestTy)) |
3220 | if (SrcVecTy->getElementCount() == DestVecTy->getElementCount()) { |
3221 | // An element by element cast. Find the appropriate opcode based on the |
3222 | // element types. |
3223 | SrcTy = SrcVecTy->getElementType(); |
3224 | DestTy = DestVecTy->getElementType(); |
3225 | } |
3226 | |
3227 | // Get the bit sizes, we'll need these |
3228 | unsigned SrcBits = SrcTy->getPrimitiveSizeInBits(); // 0 for ptr |
3229 | unsigned DestBits = DestTy->getPrimitiveSizeInBits(); // 0 for ptr |
3230 | |
3231 | // Run through the possibilities ... |
3232 | if (DestTy->isIntegerTy()) { // Casting to integral |
3233 | if (SrcTy->isIntegerTy()) { // Casting from integral |
3234 | if (DestBits < SrcBits) |
3235 | return Trunc; // int -> smaller int |
3236 | else if (DestBits > SrcBits) { // its an extension |
3237 | if (SrcIsSigned) |
3238 | return SExt; // signed -> SEXT |
3239 | else |
3240 | return ZExt; // unsigned -> ZEXT |
3241 | } else { |
3242 | return BitCast; // Same size, No-op cast |
3243 | } |
3244 | } else if (SrcTy->isFloatingPointTy()) { // Casting from floating pt |
3245 | if (DestIsSigned) |
3246 | return FPToSI; // FP -> sint |
3247 | else |
3248 | return FPToUI; // FP -> uint |
3249 | } else if (SrcTy->isVectorTy()) { |
3250 | assert(DestBits == SrcBits && |
3251 | "Casting vector to integer of different width" ); |
3252 | return BitCast; // Same size, no-op cast |
3253 | } else { |
3254 | assert(SrcTy->isPointerTy() && |
3255 | "Casting from a value that is not first-class type" ); |
3256 | return PtrToInt; // ptr -> int |
3257 | } |
3258 | } else if (DestTy->isFloatingPointTy()) { // Casting to floating pt |
3259 | if (SrcTy->isIntegerTy()) { // Casting from integral |
3260 | if (SrcIsSigned) |
3261 | return SIToFP; // sint -> FP |
3262 | else |
3263 | return UIToFP; // uint -> FP |
3264 | } else if (SrcTy->isFloatingPointTy()) { // Casting from floating pt |
3265 | if (DestBits < SrcBits) { |
3266 | return FPTrunc; // FP -> smaller FP |
3267 | } else if (DestBits > SrcBits) { |
3268 | return FPExt; // FP -> larger FP |
3269 | } else { |
3270 | return BitCast; // same size, no-op cast |
3271 | } |
3272 | } else if (SrcTy->isVectorTy()) { |
3273 | assert(DestBits == SrcBits && |
3274 | "Casting vector to floating point of different width" ); |
3275 | return BitCast; // same size, no-op cast |
3276 | } |
3277 | llvm_unreachable("Casting pointer or non-first class to float" ); |
3278 | } else if (DestTy->isVectorTy()) { |
3279 | assert(DestBits == SrcBits && |
3280 | "Illegal cast to vector (wrong type or size)" ); |
3281 | return BitCast; |
3282 | } else if (DestTy->isPointerTy()) { |
3283 | if (SrcTy->isPointerTy()) { |
3284 | if (DestTy->getPointerAddressSpace() != SrcTy->getPointerAddressSpace()) |
3285 | return AddrSpaceCast; |
3286 | return BitCast; // ptr -> ptr |
3287 | } else if (SrcTy->isIntegerTy()) { |
3288 | return IntToPtr; // int -> ptr |
3289 | } |
3290 | llvm_unreachable("Casting pointer to other than pointer or int" ); |
3291 | } |
3292 | llvm_unreachable("Casting to type that is not first-class" ); |
3293 | } |
3294 | |
3295 | //===----------------------------------------------------------------------===// |
3296 | // CastInst SubClass Constructors |
3297 | //===----------------------------------------------------------------------===// |
3298 | |
3299 | /// Check that the construction parameters for a CastInst are correct. This |
3300 | /// could be broken out into the separate constructors but it is useful to have |
3301 | /// it in one place and to eliminate the redundant code for getting the sizes |
3302 | /// of the types involved. |
3303 | bool |
3304 | CastInst::castIsValid(Instruction::CastOps op, Type *SrcTy, Type *DstTy) { |
3305 | if (!SrcTy->isFirstClassType() || !DstTy->isFirstClassType() || |
3306 | SrcTy->isAggregateType() || DstTy->isAggregateType()) |
3307 | return false; |
3308 | |
3309 | // Get the size of the types in bits, and whether we are dealing |
3310 | // with vector types, we'll need this later. |
3311 | bool SrcIsVec = isa<VectorType>(Val: SrcTy); |
3312 | bool DstIsVec = isa<VectorType>(Val: DstTy); |
3313 | unsigned SrcScalarBitSize = SrcTy->getScalarSizeInBits(); |
3314 | unsigned DstScalarBitSize = DstTy->getScalarSizeInBits(); |
3315 | |
3316 | // If these are vector types, get the lengths of the vectors (using zero for |
3317 | // scalar types means that checking that vector lengths match also checks that |
3318 | // scalars are not being converted to vectors or vectors to scalars). |
3319 | ElementCount SrcEC = SrcIsVec ? cast<VectorType>(Val: SrcTy)->getElementCount() |
3320 | : ElementCount::getFixed(MinVal: 0); |
3321 | ElementCount DstEC = DstIsVec ? cast<VectorType>(Val: DstTy)->getElementCount() |
3322 | : ElementCount::getFixed(MinVal: 0); |
3323 | |
3324 | // Switch on the opcode provided |
3325 | switch (op) { |
3326 | default: return false; // This is an input error |
3327 | case Instruction::Trunc: |
3328 | return SrcTy->isIntOrIntVectorTy() && DstTy->isIntOrIntVectorTy() && |
3329 | SrcEC == DstEC && SrcScalarBitSize > DstScalarBitSize; |
3330 | case Instruction::ZExt: |
3331 | return SrcTy->isIntOrIntVectorTy() && DstTy->isIntOrIntVectorTy() && |
3332 | SrcEC == DstEC && SrcScalarBitSize < DstScalarBitSize; |
3333 | case Instruction::SExt: |
3334 | return SrcTy->isIntOrIntVectorTy() && DstTy->isIntOrIntVectorTy() && |
3335 | SrcEC == DstEC && SrcScalarBitSize < DstScalarBitSize; |
3336 | case Instruction::FPTrunc: |
3337 | return SrcTy->isFPOrFPVectorTy() && DstTy->isFPOrFPVectorTy() && |
3338 | SrcEC == DstEC && SrcScalarBitSize > DstScalarBitSize; |
3339 | case Instruction::FPExt: |
3340 | return SrcTy->isFPOrFPVectorTy() && DstTy->isFPOrFPVectorTy() && |
3341 | SrcEC == DstEC && SrcScalarBitSize < DstScalarBitSize; |
3342 | case Instruction::UIToFP: |
3343 | case Instruction::SIToFP: |
3344 | return SrcTy->isIntOrIntVectorTy() && DstTy->isFPOrFPVectorTy() && |
3345 | SrcEC == DstEC; |
3346 | case Instruction::FPToUI: |
3347 | case Instruction::FPToSI: |
3348 | return SrcTy->isFPOrFPVectorTy() && DstTy->isIntOrIntVectorTy() && |
3349 | SrcEC == DstEC; |
3350 | case Instruction::PtrToInt: |
3351 | if (SrcEC != DstEC) |
3352 | return false; |
3353 | return SrcTy->isPtrOrPtrVectorTy() && DstTy->isIntOrIntVectorTy(); |
3354 | case Instruction::IntToPtr: |
3355 | if (SrcEC != DstEC) |
3356 | return false; |
3357 | return SrcTy->isIntOrIntVectorTy() && DstTy->isPtrOrPtrVectorTy(); |
3358 | case Instruction::BitCast: { |
3359 | PointerType *SrcPtrTy = dyn_cast<PointerType>(Val: SrcTy->getScalarType()); |
3360 | PointerType *DstPtrTy = dyn_cast<PointerType>(Val: DstTy->getScalarType()); |
3361 | |
3362 | // BitCast implies a no-op cast of type only. No bits change. |
3363 | // However, you can't cast pointers to anything but pointers. |
3364 | if (!SrcPtrTy != !DstPtrTy) |
3365 | return false; |
3366 | |
3367 | // For non-pointer cases, the cast is okay if the source and destination bit |
3368 | // widths are identical. |
3369 | if (!SrcPtrTy) |
3370 | return SrcTy->getPrimitiveSizeInBits() == DstTy->getPrimitiveSizeInBits(); |
3371 | |
3372 | // If both are pointers then the address spaces must match. |
3373 | if (SrcPtrTy->getAddressSpace() != DstPtrTy->getAddressSpace()) |
3374 | return false; |
3375 | |
3376 | // A vector of pointers must have the same number of elements. |
3377 | if (SrcIsVec && DstIsVec) |
3378 | return SrcEC == DstEC; |
3379 | if (SrcIsVec) |
3380 | return SrcEC == ElementCount::getFixed(MinVal: 1); |
3381 | if (DstIsVec) |
3382 | return DstEC == ElementCount::getFixed(MinVal: 1); |
3383 | |
3384 | return true; |
3385 | } |
3386 | case Instruction::AddrSpaceCast: { |
3387 | PointerType *SrcPtrTy = dyn_cast<PointerType>(Val: SrcTy->getScalarType()); |
3388 | if (!SrcPtrTy) |
3389 | return false; |
3390 | |
3391 | PointerType *DstPtrTy = dyn_cast<PointerType>(Val: DstTy->getScalarType()); |
3392 | if (!DstPtrTy) |
3393 | return false; |
3394 | |
3395 | if (SrcPtrTy->getAddressSpace() == DstPtrTy->getAddressSpace()) |
3396 | return false; |
3397 | |
3398 | return SrcEC == DstEC; |
3399 | } |
3400 | } |
3401 | } |
3402 | |
3403 | TruncInst::TruncInst(Value *S, Type *Ty, const Twine &Name, |
3404 | InsertPosition InsertBefore) |
3405 | : CastInst(Ty, Trunc, S, Name, InsertBefore) { |
3406 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal Trunc" ); |
3407 | } |
3408 | |
3409 | ZExtInst::ZExtInst(Value *S, Type *Ty, const Twine &Name, |
3410 | InsertPosition InsertBefore) |
3411 | : CastInst(Ty, ZExt, S, Name, InsertBefore) { |
3412 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal ZExt" ); |
3413 | } |
3414 | |
3415 | SExtInst::SExtInst(Value *S, Type *Ty, const Twine &Name, |
3416 | InsertPosition InsertBefore) |
3417 | : CastInst(Ty, SExt, S, Name, InsertBefore) { |
3418 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal SExt" ); |
3419 | } |
3420 | |
3421 | FPTruncInst::FPTruncInst(Value *S, Type *Ty, const Twine &Name, |
3422 | InsertPosition InsertBefore) |
3423 | : CastInst(Ty, FPTrunc, S, Name, InsertBefore) { |
3424 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal FPTrunc" ); |
3425 | } |
3426 | |
3427 | FPExtInst::FPExtInst(Value *S, Type *Ty, const Twine &Name, |
3428 | InsertPosition InsertBefore) |
3429 | : CastInst(Ty, FPExt, S, Name, InsertBefore) { |
3430 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal FPExt" ); |
3431 | } |
3432 | |
3433 | UIToFPInst::UIToFPInst(Value *S, Type *Ty, const Twine &Name, |
3434 | InsertPosition InsertBefore) |
3435 | : CastInst(Ty, UIToFP, S, Name, InsertBefore) { |
3436 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal UIToFP" ); |
3437 | } |
3438 | |
3439 | SIToFPInst::SIToFPInst(Value *S, Type *Ty, const Twine &Name, |
3440 | InsertPosition InsertBefore) |
3441 | : CastInst(Ty, SIToFP, S, Name, InsertBefore) { |
3442 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal SIToFP" ); |
3443 | } |
3444 | |
3445 | FPToUIInst::FPToUIInst(Value *S, Type *Ty, const Twine &Name, |
3446 | InsertPosition InsertBefore) |
3447 | : CastInst(Ty, FPToUI, S, Name, InsertBefore) { |
3448 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal FPToUI" ); |
3449 | } |
3450 | |
3451 | FPToSIInst::FPToSIInst(Value *S, Type *Ty, const Twine &Name, |
3452 | InsertPosition InsertBefore) |
3453 | : CastInst(Ty, FPToSI, S, Name, InsertBefore) { |
3454 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal FPToSI" ); |
3455 | } |
3456 | |
3457 | PtrToIntInst::PtrToIntInst(Value *S, Type *Ty, const Twine &Name, |
3458 | InsertPosition InsertBefore) |
3459 | : CastInst(Ty, PtrToInt, S, Name, InsertBefore) { |
3460 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal PtrToInt" ); |
3461 | } |
3462 | |
3463 | IntToPtrInst::IntToPtrInst(Value *S, Type *Ty, const Twine &Name, |
3464 | InsertPosition InsertBefore) |
3465 | : CastInst(Ty, IntToPtr, S, Name, InsertBefore) { |
3466 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal IntToPtr" ); |
3467 | } |
3468 | |
3469 | BitCastInst::BitCastInst(Value *S, Type *Ty, const Twine &Name, |
3470 | InsertPosition InsertBefore) |
3471 | : CastInst(Ty, BitCast, S, Name, InsertBefore) { |
3472 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal BitCast" ); |
3473 | } |
3474 | |
3475 | AddrSpaceCastInst::AddrSpaceCastInst(Value *S, Type *Ty, const Twine &Name, |
3476 | InsertPosition InsertBefore) |
3477 | : CastInst(Ty, AddrSpaceCast, S, Name, InsertBefore) { |
3478 | assert(castIsValid(getOpcode(), S, Ty) && "Illegal AddrSpaceCast" ); |
3479 | } |
3480 | |
3481 | //===----------------------------------------------------------------------===// |
3482 | // CmpInst Classes |
3483 | //===----------------------------------------------------------------------===// |
3484 | |
3485 | CmpInst::CmpInst(Type *ty, OtherOps op, Predicate predicate, Value *LHS, |
3486 | Value *RHS, const Twine &Name, InsertPosition InsertBefore, |
3487 | Instruction *FlagsSource) |
3488 | : Instruction(ty, op, AllocMarker, InsertBefore) { |
3489 | Op<0>() = LHS; |
3490 | Op<1>() = RHS; |
3491 | setPredicate((Predicate)predicate); |
3492 | setName(Name); |
3493 | if (FlagsSource) |
3494 | copyIRFlags(V: FlagsSource); |
3495 | } |
3496 | |
3497 | CmpInst *CmpInst::Create(OtherOps Op, Predicate predicate, Value *S1, Value *S2, |
3498 | const Twine &Name, InsertPosition InsertBefore) { |
3499 | if (Op == Instruction::ICmp) { |
3500 | if (InsertBefore.isValid()) |
3501 | return new ICmpInst(InsertBefore, CmpInst::Predicate(predicate), |
3502 | S1, S2, Name); |
3503 | else |
3504 | return new ICmpInst(CmpInst::Predicate(predicate), |
3505 | S1, S2, Name); |
3506 | } |
3507 | |
3508 | if (InsertBefore.isValid()) |
3509 | return new FCmpInst(InsertBefore, CmpInst::Predicate(predicate), |
3510 | S1, S2, Name); |
3511 | else |
3512 | return new FCmpInst(CmpInst::Predicate(predicate), |
3513 | S1, S2, Name); |
3514 | } |
3515 | |
3516 | CmpInst *CmpInst::CreateWithCopiedFlags(OtherOps Op, Predicate Pred, Value *S1, |
3517 | Value *S2, |
3518 | const Instruction *FlagsSource, |
3519 | const Twine &Name, |
3520 | InsertPosition InsertBefore) { |
3521 | CmpInst *Inst = Create(Op, predicate: Pred, S1, S2, Name, InsertBefore); |
3522 | Inst->copyIRFlags(V: FlagsSource); |
3523 | return Inst; |
3524 | } |
3525 | |
3526 | void CmpInst::swapOperands() { |
3527 | if (ICmpInst *IC = dyn_cast<ICmpInst>(Val: this)) |
3528 | IC->swapOperands(); |
3529 | else |
3530 | cast<FCmpInst>(Val: this)->swapOperands(); |
3531 | } |
3532 | |
3533 | bool CmpInst::isCommutative() const { |
3534 | if (const ICmpInst *IC = dyn_cast<ICmpInst>(Val: this)) |
3535 | return IC->isCommutative(); |
3536 | return cast<FCmpInst>(Val: this)->isCommutative(); |
3537 | } |
3538 | |
3539 | bool CmpInst::isEquality(Predicate P) { |
3540 | if (ICmpInst::isIntPredicate(P)) |
3541 | return ICmpInst::isEquality(P); |
3542 | if (FCmpInst::isFPPredicate(P)) |
3543 | return FCmpInst::isEquality(Pred: P); |
3544 | llvm_unreachable("Unsupported predicate kind" ); |
3545 | } |
3546 | |
3547 | // Returns true if either operand of CmpInst is a provably non-zero |
3548 | // floating-point constant. |
3549 | static bool hasNonZeroFPOperands(const CmpInst *Cmp) { |
3550 | auto *LHS = dyn_cast<Constant>(Val: Cmp->getOperand(i_nocapture: 0)); |
3551 | auto *RHS = dyn_cast<Constant>(Val: Cmp->getOperand(i_nocapture: 1)); |
3552 | if (auto *Const = LHS ? LHS : RHS) { |
3553 | using namespace llvm::PatternMatch; |
3554 | return match(V: Const, P: m_NonZeroNotDenormalFP()); |
3555 | } |
3556 | return false; |
3557 | } |
3558 | |
3559 | // Floating-point equality is not an equivalence when comparing +0.0 with |
3560 | // -0.0, when comparing NaN with another value, or when flushing |
3561 | // denormals-to-zero. |
3562 | bool CmpInst::isEquivalence(bool Invert) const { |
3563 | switch (Invert ? getInversePredicate() : getPredicate()) { |
3564 | case CmpInst::Predicate::ICMP_EQ: |
3565 | return true; |
3566 | case CmpInst::Predicate::FCMP_UEQ: |
3567 | if (!hasNoNaNs()) |
3568 | return false; |
3569 | [[fallthrough]]; |
3570 | case CmpInst::Predicate::FCMP_OEQ: |
3571 | return hasNonZeroFPOperands(Cmp: this); |
3572 | default: |
3573 | return false; |
3574 | } |
3575 | } |
3576 | |
3577 | CmpInst::Predicate CmpInst::getInversePredicate(Predicate pred) { |
3578 | switch (pred) { |
3579 | default: llvm_unreachable("Unknown cmp predicate!" ); |
3580 | case ICMP_EQ: return ICMP_NE; |
3581 | case ICMP_NE: return ICMP_EQ; |
3582 | case ICMP_UGT: return ICMP_ULE; |
3583 | case ICMP_ULT: return ICMP_UGE; |
3584 | case ICMP_UGE: return ICMP_ULT; |
3585 | case ICMP_ULE: return ICMP_UGT; |
3586 | case ICMP_SGT: return ICMP_SLE; |
3587 | case ICMP_SLT: return ICMP_SGE; |
3588 | case ICMP_SGE: return ICMP_SLT; |
3589 | case ICMP_SLE: return ICMP_SGT; |
3590 | |
3591 | case FCMP_OEQ: return FCMP_UNE; |
3592 | case FCMP_ONE: return FCMP_UEQ; |
3593 | case FCMP_OGT: return FCMP_ULE; |
3594 | case FCMP_OLT: return FCMP_UGE; |
3595 | case FCMP_OGE: return FCMP_ULT; |
3596 | case FCMP_OLE: return FCMP_UGT; |
3597 | case FCMP_UEQ: return FCMP_ONE; |
3598 | case FCMP_UNE: return FCMP_OEQ; |
3599 | case FCMP_UGT: return FCMP_OLE; |
3600 | case FCMP_ULT: return FCMP_OGE; |
3601 | case FCMP_UGE: return FCMP_OLT; |
3602 | case FCMP_ULE: return FCMP_OGT; |
3603 | case FCMP_ORD: return FCMP_UNO; |
3604 | case FCMP_UNO: return FCMP_ORD; |
3605 | case FCMP_TRUE: return FCMP_FALSE; |
3606 | case FCMP_FALSE: return FCMP_TRUE; |
3607 | } |
3608 | } |
3609 | |
3610 | StringRef CmpInst::getPredicateName(Predicate Pred) { |
3611 | switch (Pred) { |
3612 | default: return "unknown" ; |
3613 | case FCmpInst::FCMP_FALSE: return "false" ; |
3614 | case FCmpInst::FCMP_OEQ: return "oeq" ; |
3615 | case FCmpInst::FCMP_OGT: return "ogt" ; |
3616 | case FCmpInst::FCMP_OGE: return "oge" ; |
3617 | case FCmpInst::FCMP_OLT: return "olt" ; |
3618 | case FCmpInst::FCMP_OLE: return "ole" ; |
3619 | case FCmpInst::FCMP_ONE: return "one" ; |
3620 | case FCmpInst::FCMP_ORD: return "ord" ; |
3621 | case FCmpInst::FCMP_UNO: return "uno" ; |
3622 | case FCmpInst::FCMP_UEQ: return "ueq" ; |
3623 | case FCmpInst::FCMP_UGT: return "ugt" ; |
3624 | case FCmpInst::FCMP_UGE: return "uge" ; |
3625 | case FCmpInst::FCMP_ULT: return "ult" ; |
3626 | case FCmpInst::FCMP_ULE: return "ule" ; |
3627 | case FCmpInst::FCMP_UNE: return "une" ; |
3628 | case FCmpInst::FCMP_TRUE: return "true" ; |
3629 | case ICmpInst::ICMP_EQ: return "eq" ; |
3630 | case ICmpInst::ICMP_NE: return "ne" ; |
3631 | case ICmpInst::ICMP_SGT: return "sgt" ; |
3632 | case ICmpInst::ICMP_SGE: return "sge" ; |
3633 | case ICmpInst::ICMP_SLT: return "slt" ; |
3634 | case ICmpInst::ICMP_SLE: return "sle" ; |
3635 | case ICmpInst::ICMP_UGT: return "ugt" ; |
3636 | case ICmpInst::ICMP_UGE: return "uge" ; |
3637 | case ICmpInst::ICMP_ULT: return "ult" ; |
3638 | case ICmpInst::ICMP_ULE: return "ule" ; |
3639 | } |
3640 | } |
3641 | |
3642 | raw_ostream &llvm::operator<<(raw_ostream &OS, CmpInst::Predicate Pred) { |
3643 | OS << CmpInst::getPredicateName(Pred); |
3644 | return OS; |
3645 | } |
3646 | |
3647 | ICmpInst::Predicate ICmpInst::getSignedPredicate(Predicate pred) { |
3648 | switch (pred) { |
3649 | default: llvm_unreachable("Unknown icmp predicate!" ); |
3650 | case ICMP_EQ: case ICMP_NE: |
3651 | case ICMP_SGT: case ICMP_SLT: case ICMP_SGE: case ICMP_SLE: |
3652 | return pred; |
3653 | case ICMP_UGT: return ICMP_SGT; |
3654 | case ICMP_ULT: return ICMP_SLT; |
3655 | case ICMP_UGE: return ICMP_SGE; |
3656 | case ICMP_ULE: return ICMP_SLE; |
3657 | } |
3658 | } |
3659 | |
3660 | ICmpInst::Predicate ICmpInst::getUnsignedPredicate(Predicate pred) { |
3661 | switch (pred) { |
3662 | default: llvm_unreachable("Unknown icmp predicate!" ); |
3663 | case ICMP_EQ: case ICMP_NE: |
3664 | case ICMP_UGT: case ICMP_ULT: case ICMP_UGE: case ICMP_ULE: |
3665 | return pred; |
3666 | case ICMP_SGT: return ICMP_UGT; |
3667 | case ICMP_SLT: return ICMP_ULT; |
3668 | case ICMP_SGE: return ICMP_UGE; |
3669 | case ICMP_SLE: return ICMP_ULE; |
3670 | } |
3671 | } |
3672 | |
3673 | CmpInst::Predicate CmpInst::getSwappedPredicate(Predicate pred) { |
3674 | switch (pred) { |
3675 | default: llvm_unreachable("Unknown cmp predicate!" ); |
3676 | case ICMP_EQ: case ICMP_NE: |
3677 | return pred; |
3678 | case ICMP_SGT: return ICMP_SLT; |
3679 | case ICMP_SLT: return ICMP_SGT; |
3680 | case ICMP_SGE: return ICMP_SLE; |
3681 | case ICMP_SLE: return ICMP_SGE; |
3682 | case ICMP_UGT: return ICMP_ULT; |
3683 | case ICMP_ULT: return ICMP_UGT; |
3684 | case ICMP_UGE: return ICMP_ULE; |
3685 | case ICMP_ULE: return ICMP_UGE; |
3686 | |
3687 | case FCMP_FALSE: case FCMP_TRUE: |
3688 | case FCMP_OEQ: case FCMP_ONE: |
3689 | case FCMP_UEQ: case FCMP_UNE: |
3690 | case FCMP_ORD: case FCMP_UNO: |
3691 | return pred; |
3692 | case FCMP_OGT: return FCMP_OLT; |
3693 | case FCMP_OLT: return FCMP_OGT; |
3694 | case FCMP_OGE: return FCMP_OLE; |
3695 | case FCMP_OLE: return FCMP_OGE; |
3696 | case FCMP_UGT: return FCMP_ULT; |
3697 | case FCMP_ULT: return FCMP_UGT; |
3698 | case FCMP_UGE: return FCMP_ULE; |
3699 | case FCMP_ULE: return FCMP_UGE; |
3700 | } |
3701 | } |
3702 | |
3703 | bool CmpInst::isNonStrictPredicate(Predicate pred) { |
3704 | switch (pred) { |
3705 | case ICMP_SGE: |
3706 | case ICMP_SLE: |
3707 | case ICMP_UGE: |
3708 | case ICMP_ULE: |
3709 | case FCMP_OGE: |
3710 | case FCMP_OLE: |
3711 | case FCMP_UGE: |
3712 | case FCMP_ULE: |
3713 | return true; |
3714 | default: |
3715 | return false; |
3716 | } |
3717 | } |
3718 | |
3719 | bool CmpInst::isStrictPredicate(Predicate pred) { |
3720 | switch (pred) { |
3721 | case ICMP_SGT: |
3722 | case ICMP_SLT: |
3723 | case ICMP_UGT: |
3724 | case ICMP_ULT: |
3725 | case FCMP_OGT: |
3726 | case FCMP_OLT: |
3727 | case FCMP_UGT: |
3728 | case FCMP_ULT: |
3729 | return true; |
3730 | default: |
3731 | return false; |
3732 | } |
3733 | } |
3734 | |
3735 | CmpInst::Predicate CmpInst::getStrictPredicate(Predicate pred) { |
3736 | switch (pred) { |
3737 | case ICMP_SGE: |
3738 | return ICMP_SGT; |
3739 | case ICMP_SLE: |
3740 | return ICMP_SLT; |
3741 | case ICMP_UGE: |
3742 | return ICMP_UGT; |
3743 | case ICMP_ULE: |
3744 | return ICMP_ULT; |
3745 | case FCMP_OGE: |
3746 | return FCMP_OGT; |
3747 | case FCMP_OLE: |
3748 | return FCMP_OLT; |
3749 | case FCMP_UGE: |
3750 | return FCMP_UGT; |
3751 | case FCMP_ULE: |
3752 | return FCMP_ULT; |
3753 | default: |
3754 | return pred; |
3755 | } |
3756 | } |
3757 | |
3758 | CmpInst::Predicate CmpInst::getNonStrictPredicate(Predicate pred) { |
3759 | switch (pred) { |
3760 | case ICMP_SGT: |
3761 | return ICMP_SGE; |
3762 | case ICMP_SLT: |
3763 | return ICMP_SLE; |
3764 | case ICMP_UGT: |
3765 | return ICMP_UGE; |
3766 | case ICMP_ULT: |
3767 | return ICMP_ULE; |
3768 | case FCMP_OGT: |
3769 | return FCMP_OGE; |
3770 | case FCMP_OLT: |
3771 | return FCMP_OLE; |
3772 | case FCMP_UGT: |
3773 | return FCMP_UGE; |
3774 | case FCMP_ULT: |
3775 | return FCMP_ULE; |
3776 | default: |
3777 | return pred; |
3778 | } |
3779 | } |
3780 | |
3781 | CmpInst::Predicate CmpInst::getFlippedStrictnessPredicate(Predicate pred) { |
3782 | assert(CmpInst::isRelational(pred) && "Call only with relational predicate!" ); |
3783 | |
3784 | if (isStrictPredicate(pred)) |
3785 | return getNonStrictPredicate(pred); |
3786 | if (isNonStrictPredicate(pred)) |
3787 | return getStrictPredicate(pred); |
3788 | |
3789 | llvm_unreachable("Unknown predicate!" ); |
3790 | } |
3791 | |
3792 | bool CmpInst::isUnsigned(Predicate predicate) { |
3793 | switch (predicate) { |
3794 | default: return false; |
3795 | case ICmpInst::ICMP_ULT: case ICmpInst::ICMP_ULE: case ICmpInst::ICMP_UGT: |
3796 | case ICmpInst::ICMP_UGE: return true; |
3797 | } |
3798 | } |
3799 | |
3800 | bool CmpInst::isSigned(Predicate predicate) { |
3801 | switch (predicate) { |
3802 | default: return false; |
3803 | case ICmpInst::ICMP_SLT: case ICmpInst::ICMP_SLE: case ICmpInst::ICMP_SGT: |
3804 | case ICmpInst::ICMP_SGE: return true; |
3805 | } |
3806 | } |
3807 | |
3808 | bool ICmpInst::compare(const APInt &LHS, const APInt &RHS, |
3809 | ICmpInst::Predicate Pred) { |
3810 | assert(ICmpInst::isIntPredicate(Pred) && "Only for integer predicates!" ); |
3811 | switch (Pred) { |
3812 | case ICmpInst::Predicate::ICMP_EQ: |
3813 | return LHS.eq(RHS); |
3814 | case ICmpInst::Predicate::ICMP_NE: |
3815 | return LHS.ne(RHS); |
3816 | case ICmpInst::Predicate::ICMP_UGT: |
3817 | return LHS.ugt(RHS); |
3818 | case ICmpInst::Predicate::ICMP_UGE: |
3819 | return LHS.uge(RHS); |
3820 | case ICmpInst::Predicate::ICMP_ULT: |
3821 | return LHS.ult(RHS); |
3822 | case ICmpInst::Predicate::ICMP_ULE: |
3823 | return LHS.ule(RHS); |
3824 | case ICmpInst::Predicate::ICMP_SGT: |
3825 | return LHS.sgt(RHS); |
3826 | case ICmpInst::Predicate::ICMP_SGE: |
3827 | return LHS.sge(RHS); |
3828 | case ICmpInst::Predicate::ICMP_SLT: |
3829 | return LHS.slt(RHS); |
3830 | case ICmpInst::Predicate::ICMP_SLE: |
3831 | return LHS.sle(RHS); |
3832 | default: |
3833 | llvm_unreachable("Unexpected non-integer predicate." ); |
3834 | }; |
3835 | } |
3836 | |
3837 | bool FCmpInst::compare(const APFloat &LHS, const APFloat &RHS, |
3838 | FCmpInst::Predicate Pred) { |
3839 | APFloat::cmpResult R = LHS.compare(RHS); |
3840 | switch (Pred) { |
3841 | default: |
3842 | llvm_unreachable("Invalid FCmp Predicate" ); |
3843 | case FCmpInst::FCMP_FALSE: |
3844 | return false; |
3845 | case FCmpInst::FCMP_TRUE: |
3846 | return true; |
3847 | case FCmpInst::FCMP_UNO: |
3848 | return R == APFloat::cmpUnordered; |
3849 | case FCmpInst::FCMP_ORD: |
3850 | return R != APFloat::cmpUnordered; |
3851 | case FCmpInst::FCMP_UEQ: |
3852 | return R == APFloat::cmpUnordered || R == APFloat::cmpEqual; |
3853 | case FCmpInst::FCMP_OEQ: |
3854 | return R == APFloat::cmpEqual; |
3855 | case FCmpInst::FCMP_UNE: |
3856 | return R != APFloat::cmpEqual; |
3857 | case FCmpInst::FCMP_ONE: |
3858 | return R == APFloat::cmpLessThan || R == APFloat::cmpGreaterThan; |
3859 | case FCmpInst::FCMP_ULT: |
3860 | return R == APFloat::cmpUnordered || R == APFloat::cmpLessThan; |
3861 | case FCmpInst::FCMP_OLT: |
3862 | return R == APFloat::cmpLessThan; |
3863 | case FCmpInst::FCMP_UGT: |
3864 | return R == APFloat::cmpUnordered || R == APFloat::cmpGreaterThan; |
3865 | case FCmpInst::FCMP_OGT: |
3866 | return R == APFloat::cmpGreaterThan; |
3867 | case FCmpInst::FCMP_ULE: |
3868 | return R != APFloat::cmpGreaterThan; |
3869 | case FCmpInst::FCMP_OLE: |
3870 | return R == APFloat::cmpLessThan || R == APFloat::cmpEqual; |
3871 | case FCmpInst::FCMP_UGE: |
3872 | return R != APFloat::cmpLessThan; |
3873 | case FCmpInst::FCMP_OGE: |
3874 | return R == APFloat::cmpGreaterThan || R == APFloat::cmpEqual; |
3875 | } |
3876 | } |
3877 | |
3878 | std::optional<bool> ICmpInst::compare(const KnownBits &LHS, |
3879 | const KnownBits &RHS, |
3880 | ICmpInst::Predicate Pred) { |
3881 | switch (Pred) { |
3882 | case ICmpInst::ICMP_EQ: |
3883 | return KnownBits::eq(LHS, RHS); |
3884 | case ICmpInst::ICMP_NE: |
3885 | return KnownBits::ne(LHS, RHS); |
3886 | case ICmpInst::ICMP_UGE: |
3887 | return KnownBits::uge(LHS, RHS); |
3888 | case ICmpInst::ICMP_UGT: |
3889 | return KnownBits::ugt(LHS, RHS); |
3890 | case ICmpInst::ICMP_ULE: |
3891 | return KnownBits::ule(LHS, RHS); |
3892 | case ICmpInst::ICMP_ULT: |
3893 | return KnownBits::ult(LHS, RHS); |
3894 | case ICmpInst::ICMP_SGE: |
3895 | return KnownBits::sge(LHS, RHS); |
3896 | case ICmpInst::ICMP_SGT: |
3897 | return KnownBits::sgt(LHS, RHS); |
3898 | case ICmpInst::ICMP_SLE: |
3899 | return KnownBits::sle(LHS, RHS); |
3900 | case ICmpInst::ICMP_SLT: |
3901 | return KnownBits::slt(LHS, RHS); |
3902 | default: |
3903 | llvm_unreachable("Unexpected non-integer predicate." ); |
3904 | } |
3905 | } |
3906 | |
3907 | CmpInst::Predicate ICmpInst::getFlippedSignednessPredicate(Predicate pred) { |
3908 | if (CmpInst::isEquality(P: pred)) |
3909 | return pred; |
3910 | if (isSigned(predicate: pred)) |
3911 | return getUnsignedPredicate(pred); |
3912 | if (isUnsigned(predicate: pred)) |
3913 | return getSignedPredicate(pred); |
3914 | |
3915 | llvm_unreachable("Unknown predicate!" ); |
3916 | } |
3917 | |
3918 | bool CmpInst::isOrdered(Predicate predicate) { |
3919 | switch (predicate) { |
3920 | default: return false; |
3921 | case FCmpInst::FCMP_OEQ: case FCmpInst::FCMP_ONE: case FCmpInst::FCMP_OGT: |
3922 | case FCmpInst::FCMP_OLT: case FCmpInst::FCMP_OGE: case FCmpInst::FCMP_OLE: |
3923 | case FCmpInst::FCMP_ORD: return true; |
3924 | } |
3925 | } |
3926 | |
3927 | bool CmpInst::isUnordered(Predicate predicate) { |
3928 | switch (predicate) { |
3929 | default: return false; |
3930 | case FCmpInst::FCMP_UEQ: case FCmpInst::FCMP_UNE: case FCmpInst::FCMP_UGT: |
3931 | case FCmpInst::FCMP_ULT: case FCmpInst::FCMP_UGE: case FCmpInst::FCMP_ULE: |
3932 | case FCmpInst::FCMP_UNO: return true; |
3933 | } |
3934 | } |
3935 | |
3936 | bool CmpInst::isTrueWhenEqual(Predicate predicate) { |
3937 | switch(predicate) { |
3938 | default: return false; |
3939 | case ICMP_EQ: case ICMP_UGE: case ICMP_ULE: case ICMP_SGE: case ICMP_SLE: |
3940 | case FCMP_TRUE: case FCMP_UEQ: case FCMP_UGE: case FCMP_ULE: return true; |
3941 | } |
3942 | } |
3943 | |
3944 | bool CmpInst::isFalseWhenEqual(Predicate predicate) { |
3945 | switch(predicate) { |
3946 | case ICMP_NE: case ICMP_UGT: case ICMP_ULT: case ICMP_SGT: case ICMP_SLT: |
3947 | case FCMP_FALSE: case FCMP_ONE: case FCMP_OGT: case FCMP_OLT: return true; |
3948 | default: return false; |
3949 | } |
3950 | } |
3951 | |
3952 | static bool isImpliedTrueByMatchingCmp(CmpPredicate Pred1, CmpPredicate Pred2) { |
3953 | // If the predicates match, then we know the first condition implies the |
3954 | // second is true. |
3955 | if (CmpPredicate::getMatching(A: Pred1, B: Pred2)) |
3956 | return true; |
3957 | |
3958 | if (Pred1.hasSameSign() && CmpInst::isSigned(predicate: Pred2)) |
3959 | Pred1 = ICmpInst::getFlippedSignednessPredicate(pred: Pred1); |
3960 | else if (Pred2.hasSameSign() && CmpInst::isSigned(predicate: Pred1)) |
3961 | Pred2 = ICmpInst::getFlippedSignednessPredicate(pred: Pred2); |
3962 | |
3963 | switch (Pred1) { |
3964 | default: |
3965 | break; |
3966 | case CmpInst::ICMP_EQ: |
3967 | // A == B implies A >=u B, A <=u B, A >=s B, and A <=s B are true. |
3968 | return Pred2 == CmpInst::ICMP_UGE || Pred2 == CmpInst::ICMP_ULE || |
3969 | Pred2 == CmpInst::ICMP_SGE || Pred2 == CmpInst::ICMP_SLE; |
3970 | case CmpInst::ICMP_UGT: // A >u B implies A != B and A >=u B are true. |
3971 | return Pred2 == CmpInst::ICMP_NE || Pred2 == CmpInst::ICMP_UGE; |
3972 | case CmpInst::ICMP_ULT: // A <u B implies A != B and A <=u B are true. |
3973 | return Pred2 == CmpInst::ICMP_NE || Pred2 == CmpInst::ICMP_ULE; |
3974 | case CmpInst::ICMP_SGT: // A >s B implies A != B and A >=s B are true. |
3975 | return Pred2 == CmpInst::ICMP_NE || Pred2 == CmpInst::ICMP_SGE; |
3976 | case CmpInst::ICMP_SLT: // A <s B implies A != B and A <=s B are true. |
3977 | return Pred2 == CmpInst::ICMP_NE || Pred2 == CmpInst::ICMP_SLE; |
3978 | } |
3979 | return false; |
3980 | } |
3981 | |
3982 | static bool isImpliedFalseByMatchingCmp(CmpPredicate Pred1, |
3983 | CmpPredicate Pred2) { |
3984 | return isImpliedTrueByMatchingCmp(Pred1, |
3985 | Pred2: ICmpInst::getInverseCmpPredicate(Pred: Pred2)); |
3986 | } |
3987 | |
3988 | std::optional<bool> ICmpInst::isImpliedByMatchingCmp(CmpPredicate Pred1, |
3989 | CmpPredicate Pred2) { |
3990 | if (isImpliedTrueByMatchingCmp(Pred1, Pred2)) |
3991 | return true; |
3992 | if (isImpliedFalseByMatchingCmp(Pred1, Pred2)) |
3993 | return false; |
3994 | return std::nullopt; |
3995 | } |
3996 | |
3997 | //===----------------------------------------------------------------------===// |
3998 | // CmpPredicate Implementation |
3999 | //===----------------------------------------------------------------------===// |
4000 | |
4001 | std::optional<CmpPredicate> CmpPredicate::getMatching(CmpPredicate A, |
4002 | CmpPredicate B) { |
4003 | if (A.Pred == B.Pred) |
4004 | return A.HasSameSign == B.HasSameSign ? A : CmpPredicate(A.Pred); |
4005 | if (CmpInst::isFPPredicate(P: A) || CmpInst::isFPPredicate(P: B)) |
4006 | return {}; |
4007 | if (A.HasSameSign && |
4008 | A.Pred == ICmpInst::getFlippedSignednessPredicate(pred: B.Pred)) |
4009 | return B.Pred; |
4010 | if (B.HasSameSign && |
4011 | B.Pred == ICmpInst::getFlippedSignednessPredicate(pred: A.Pred)) |
4012 | return A.Pred; |
4013 | return {}; |
4014 | } |
4015 | |
4016 | CmpInst::Predicate CmpPredicate::getPreferredSignedPredicate() const { |
4017 | return HasSameSign ? ICmpInst::getSignedPredicate(pred: Pred) : Pred; |
4018 | } |
4019 | |
4020 | CmpPredicate CmpPredicate::get(const CmpInst *Cmp) { |
4021 | if (auto *ICI = dyn_cast<ICmpInst>(Val: Cmp)) |
4022 | return ICI->getCmpPredicate(); |
4023 | return Cmp->getPredicate(); |
4024 | } |
4025 | |
4026 | CmpPredicate CmpPredicate::getSwapped(CmpPredicate P) { |
4027 | return {CmpInst::getSwappedPredicate(pred: P), P.hasSameSign()}; |
4028 | } |
4029 | |
4030 | CmpPredicate CmpPredicate::getSwapped(const CmpInst *Cmp) { |
4031 | return getSwapped(P: get(Cmp)); |
4032 | } |
4033 | |
4034 | //===----------------------------------------------------------------------===// |
4035 | // SwitchInst Implementation |
4036 | //===----------------------------------------------------------------------===// |
4037 | |
4038 | void SwitchInst::init(Value *Value, BasicBlock *Default, unsigned NumReserved) { |
4039 | assert(Value && Default && NumReserved); |
4040 | ReservedSpace = NumReserved; |
4041 | setNumHungOffUseOperands(2); |
4042 | allocHungoffUses(N: ReservedSpace); |
4043 | |
4044 | Op<0>() = Value; |
4045 | Op<1>() = Default; |
4046 | } |
4047 | |
4048 | /// SwitchInst ctor - Create a new switch instruction, specifying a value to |
4049 | /// switch on and a default destination. The number of additional cases can |
4050 | /// be specified here to make memory allocation more efficient. This |
4051 | /// constructor can also autoinsert before another instruction. |
4052 | SwitchInst::SwitchInst(Value *Value, BasicBlock *Default, unsigned NumCases, |
4053 | InsertPosition InsertBefore) |
4054 | : Instruction(Type::getVoidTy(C&: Value->getContext()), Instruction::Switch, |
4055 | AllocMarker, InsertBefore) { |
4056 | init(Value, Default, NumReserved: 2+NumCases*2); |
4057 | } |
4058 | |
4059 | SwitchInst::SwitchInst(const SwitchInst &SI) |
4060 | : Instruction(SI.getType(), Instruction::Switch, AllocMarker) { |
4061 | init(Value: SI.getCondition(), Default: SI.getDefaultDest(), NumReserved: SI.getNumOperands()); |
4062 | setNumHungOffUseOperands(SI.getNumOperands()); |
4063 | Use *OL = getOperandList(); |
4064 | const Use *InOL = SI.getOperandList(); |
4065 | for (unsigned i = 2, E = SI.getNumOperands(); i != E; i += 2) { |
4066 | OL[i] = InOL[i]; |
4067 | OL[i+1] = InOL[i+1]; |
4068 | } |
4069 | SubclassOptionalData = SI.SubclassOptionalData; |
4070 | } |
4071 | |
4072 | /// addCase - Add an entry to the switch instruction... |
4073 | /// |
4074 | void SwitchInst::addCase(ConstantInt *OnVal, BasicBlock *Dest) { |
4075 | unsigned NewCaseIdx = getNumCases(); |
4076 | unsigned OpNo = getNumOperands(); |
4077 | if (OpNo+2 > ReservedSpace) |
4078 | growOperands(); // Get more space! |
4079 | // Initialize some new operands. |
4080 | assert(OpNo+1 < ReservedSpace && "Growing didn't work!" ); |
4081 | setNumHungOffUseOperands(OpNo+2); |
4082 | CaseHandle Case(this, NewCaseIdx); |
4083 | Case.setValue(OnVal); |
4084 | Case.setSuccessor(Dest); |
4085 | } |
4086 | |
4087 | /// removeCase - This method removes the specified case and its successor |
4088 | /// from the switch instruction. |
4089 | SwitchInst::CaseIt SwitchInst::removeCase(CaseIt I) { |
4090 | unsigned idx = I->getCaseIndex(); |
4091 | |
4092 | assert(2 + idx*2 < getNumOperands() && "Case index out of range!!!" ); |
4093 | |
4094 | unsigned NumOps = getNumOperands(); |
4095 | Use *OL = getOperandList(); |
4096 | |
4097 | // Overwrite this case with the end of the list. |
4098 | if (2 + (idx + 1) * 2 != NumOps) { |
4099 | OL[2 + idx * 2] = OL[NumOps - 2]; |
4100 | OL[2 + idx * 2 + 1] = OL[NumOps - 1]; |
4101 | } |
4102 | |
4103 | // Nuke the last value. |
4104 | OL[NumOps-2].set(nullptr); |
4105 | OL[NumOps-2+1].set(nullptr); |
4106 | setNumHungOffUseOperands(NumOps-2); |
4107 | |
4108 | return CaseIt(this, idx); |
4109 | } |
4110 | |
4111 | /// growOperands - grow operands - This grows the operand list in response |
4112 | /// to a push_back style of operation. This grows the number of ops by 3 times. |
4113 | /// |
4114 | void SwitchInst::growOperands() { |
4115 | unsigned e = getNumOperands(); |
4116 | unsigned NumOps = e*3; |
4117 | |
4118 | ReservedSpace = NumOps; |
4119 | growHungoffUses(N: ReservedSpace); |
4120 | } |
4121 | |
4122 | MDNode *SwitchInstProfUpdateWrapper::buildProfBranchWeightsMD() { |
4123 | assert(Changed && "called only if metadata has changed" ); |
4124 | |
4125 | if (!Weights) |
4126 | return nullptr; |
4127 | |
4128 | assert(SI.getNumSuccessors() == Weights->size() && |
4129 | "num of prof branch_weights must accord with num of successors" ); |
4130 | |
4131 | bool AllZeroes = all_of(Range&: *Weights, P: [](uint32_t W) { return W == 0; }); |
4132 | |
4133 | if (AllZeroes || Weights->size() < 2) |
4134 | return nullptr; |
4135 | |
4136 | return MDBuilder(SI.getParent()->getContext()).createBranchWeights(Weights: *Weights); |
4137 | } |
4138 | |
4139 | void SwitchInstProfUpdateWrapper::init() { |
4140 | MDNode *ProfileData = getBranchWeightMDNode(I: SI); |
4141 | if (!ProfileData) |
4142 | return; |
4143 | |
4144 | if (getNumBranchWeights(ProfileData: *ProfileData) != SI.getNumSuccessors()) { |
4145 | llvm_unreachable("number of prof branch_weights metadata operands does " |
4146 | "not correspond to number of succesors" ); |
4147 | } |
4148 | |
4149 | SmallVector<uint32_t, 8> Weights; |
4150 | if (!extractBranchWeights(ProfileData, Weights)) |
4151 | return; |
4152 | this->Weights = std::move(Weights); |
4153 | } |
4154 | |
4155 | SwitchInst::CaseIt |
4156 | SwitchInstProfUpdateWrapper::removeCase(SwitchInst::CaseIt I) { |
4157 | if (Weights) { |
4158 | assert(SI.getNumSuccessors() == Weights->size() && |
4159 | "num of prof branch_weights must accord with num of successors" ); |
4160 | Changed = true; |
4161 | // Copy the last case to the place of the removed one and shrink. |
4162 | // This is tightly coupled with the way SwitchInst::removeCase() removes |
4163 | // the cases in SwitchInst::removeCase(CaseIt). |
4164 | (*Weights)[I->getCaseIndex() + 1] = Weights->back(); |
4165 | Weights->pop_back(); |
4166 | } |
4167 | return SI.removeCase(I); |
4168 | } |
4169 | |
4170 | void SwitchInstProfUpdateWrapper::addCase( |
4171 | ConstantInt *OnVal, BasicBlock *Dest, |
4172 | SwitchInstProfUpdateWrapper::CaseWeightOpt W) { |
4173 | SI.addCase(OnVal, Dest); |
4174 | |
4175 | if (!Weights && W && *W) { |
4176 | Changed = true; |
4177 | Weights = SmallVector<uint32_t, 8>(SI.getNumSuccessors(), 0); |
4178 | (*Weights)[SI.getNumSuccessors() - 1] = *W; |
4179 | } else if (Weights) { |
4180 | Changed = true; |
4181 | Weights->push_back(Elt: W.value_or(u: 0)); |
4182 | } |
4183 | if (Weights) |
4184 | assert(SI.getNumSuccessors() == Weights->size() && |
4185 | "num of prof branch_weights must accord with num of successors" ); |
4186 | } |
4187 | |
4188 | Instruction::InstListType::iterator |
4189 | SwitchInstProfUpdateWrapper::eraseFromParent() { |
4190 | // Instruction is erased. Mark as unchanged to not touch it in the destructor. |
4191 | Changed = false; |
4192 | if (Weights) |
4193 | Weights->resize(N: 0); |
4194 | return SI.eraseFromParent(); |
4195 | } |
4196 | |
4197 | SwitchInstProfUpdateWrapper::CaseWeightOpt |
4198 | SwitchInstProfUpdateWrapper::getSuccessorWeight(unsigned idx) { |
4199 | if (!Weights) |
4200 | return std::nullopt; |
4201 | return (*Weights)[idx]; |
4202 | } |
4203 | |
4204 | void SwitchInstProfUpdateWrapper::setSuccessorWeight( |
4205 | unsigned idx, SwitchInstProfUpdateWrapper::CaseWeightOpt W) { |
4206 | if (!W) |
4207 | return; |
4208 | |
4209 | if (!Weights && *W) |
4210 | Weights = SmallVector<uint32_t, 8>(SI.getNumSuccessors(), 0); |
4211 | |
4212 | if (Weights) { |
4213 | auto &OldW = (*Weights)[idx]; |
4214 | if (*W != OldW) { |
4215 | Changed = true; |
4216 | OldW = *W; |
4217 | } |
4218 | } |
4219 | } |
4220 | |
4221 | SwitchInstProfUpdateWrapper::CaseWeightOpt |
4222 | SwitchInstProfUpdateWrapper::getSuccessorWeight(const SwitchInst &SI, |
4223 | unsigned idx) { |
4224 | if (MDNode *ProfileData = getBranchWeightMDNode(I: SI)) |
4225 | if (ProfileData->getNumOperands() == SI.getNumSuccessors() + 1) |
4226 | return mdconst::extract<ConstantInt>(MD: ProfileData->getOperand(I: idx + 1)) |
4227 | ->getValue() |
4228 | .getZExtValue(); |
4229 | |
4230 | return std::nullopt; |
4231 | } |
4232 | |
4233 | //===----------------------------------------------------------------------===// |
4234 | // IndirectBrInst Implementation |
4235 | //===----------------------------------------------------------------------===// |
4236 | |
4237 | void IndirectBrInst::init(Value *Address, unsigned NumDests) { |
4238 | assert(Address && Address->getType()->isPointerTy() && |
4239 | "Address of indirectbr must be a pointer" ); |
4240 | ReservedSpace = 1+NumDests; |
4241 | setNumHungOffUseOperands(1); |
4242 | allocHungoffUses(N: ReservedSpace); |
4243 | |
4244 | Op<0>() = Address; |
4245 | } |
4246 | |
4247 | |
4248 | /// growOperands - grow operands - This grows the operand list in response |
4249 | /// to a push_back style of operation. This grows the number of ops by 2 times. |
4250 | /// |
4251 | void IndirectBrInst::growOperands() { |
4252 | unsigned e = getNumOperands(); |
4253 | unsigned NumOps = e*2; |
4254 | |
4255 | ReservedSpace = NumOps; |
4256 | growHungoffUses(N: ReservedSpace); |
4257 | } |
4258 | |
4259 | IndirectBrInst::IndirectBrInst(Value *Address, unsigned NumCases, |
4260 | InsertPosition InsertBefore) |
4261 | : Instruction(Type::getVoidTy(C&: Address->getContext()), |
4262 | Instruction::IndirectBr, AllocMarker, InsertBefore) { |
4263 | init(Address, NumDests: NumCases); |
4264 | } |
4265 | |
4266 | IndirectBrInst::IndirectBrInst(const IndirectBrInst &IBI) |
4267 | : Instruction(Type::getVoidTy(C&: IBI.getContext()), Instruction::IndirectBr, |
4268 | AllocMarker) { |
4269 | NumUserOperands = IBI.NumUserOperands; |
4270 | allocHungoffUses(N: IBI.getNumOperands()); |
4271 | Use *OL = getOperandList(); |
4272 | const Use *InOL = IBI.getOperandList(); |
4273 | for (unsigned i = 0, E = IBI.getNumOperands(); i != E; ++i) |
4274 | OL[i] = InOL[i]; |
4275 | SubclassOptionalData = IBI.SubclassOptionalData; |
4276 | } |
4277 | |
4278 | /// addDestination - Add a destination. |
4279 | /// |
4280 | void IndirectBrInst::addDestination(BasicBlock *DestBB) { |
4281 | unsigned OpNo = getNumOperands(); |
4282 | if (OpNo+1 > ReservedSpace) |
4283 | growOperands(); // Get more space! |
4284 | // Initialize some new operands. |
4285 | assert(OpNo < ReservedSpace && "Growing didn't work!" ); |
4286 | setNumHungOffUseOperands(OpNo+1); |
4287 | getOperandList()[OpNo] = DestBB; |
4288 | } |
4289 | |
4290 | /// removeDestination - This method removes the specified successor from the |
4291 | /// indirectbr instruction. |
4292 | void IndirectBrInst::removeDestination(unsigned idx) { |
4293 | assert(idx < getNumOperands()-1 && "Successor index out of range!" ); |
4294 | |
4295 | unsigned NumOps = getNumOperands(); |
4296 | Use *OL = getOperandList(); |
4297 | |
4298 | // Replace this value with the last one. |
4299 | OL[idx+1] = OL[NumOps-1]; |
4300 | |
4301 | // Nuke the last value. |
4302 | OL[NumOps-1].set(nullptr); |
4303 | setNumHungOffUseOperands(NumOps-1); |
4304 | } |
4305 | |
4306 | //===----------------------------------------------------------------------===// |
4307 | // FreezeInst Implementation |
4308 | //===----------------------------------------------------------------------===// |
4309 | |
4310 | FreezeInst::FreezeInst(Value *S, const Twine &Name, InsertPosition InsertBefore) |
4311 | : UnaryInstruction(S->getType(), Freeze, S, InsertBefore) { |
4312 | setName(Name); |
4313 | } |
4314 | |
4315 | //===----------------------------------------------------------------------===// |
4316 | // cloneImpl() implementations |
4317 | //===----------------------------------------------------------------------===// |
4318 | |
4319 | // Define these methods here so vtables don't get emitted into every translation |
4320 | // unit that uses these classes. |
4321 | |
4322 | GetElementPtrInst *GetElementPtrInst::cloneImpl() const { |
4323 | IntrusiveOperandsAllocMarker AllocMarker{.NumOps: getNumOperands()}; |
4324 | return new (AllocMarker) GetElementPtrInst(*this, AllocMarker); |
4325 | } |
4326 | |
4327 | UnaryOperator *UnaryOperator::cloneImpl() const { |
4328 | return Create(Op: getOpcode(), S: Op<0>()); |
4329 | } |
4330 | |
4331 | BinaryOperator *BinaryOperator::cloneImpl() const { |
4332 | return Create(Op: getOpcode(), S1: Op<0>(), S2: Op<1>()); |
4333 | } |
4334 | |
4335 | FCmpInst *FCmpInst::cloneImpl() const { |
4336 | return new FCmpInst(getPredicate(), Op<0>(), Op<1>()); |
4337 | } |
4338 | |
4339 | ICmpInst *ICmpInst::cloneImpl() const { |
4340 | return new ICmpInst(getPredicate(), Op<0>(), Op<1>()); |
4341 | } |
4342 | |
4343 | ExtractValueInst *ExtractValueInst::() const { |
4344 | return new ExtractValueInst(*this); |
4345 | } |
4346 | |
4347 | InsertValueInst *InsertValueInst::cloneImpl() const { |
4348 | return new InsertValueInst(*this); |
4349 | } |
4350 | |
4351 | AllocaInst *AllocaInst::cloneImpl() const { |
4352 | AllocaInst *Result = new AllocaInst(getAllocatedType(), getAddressSpace(), |
4353 | getOperand(i_nocapture: 0), getAlign()); |
4354 | Result->setUsedWithInAlloca(isUsedWithInAlloca()); |
4355 | Result->setSwiftError(isSwiftError()); |
4356 | return Result; |
4357 | } |
4358 | |
4359 | LoadInst *LoadInst::cloneImpl() const { |
4360 | return new LoadInst(getType(), getOperand(i_nocapture: 0), Twine(), isVolatile(), |
4361 | getAlign(), getOrdering(), getSyncScopeID()); |
4362 | } |
4363 | |
4364 | StoreInst *StoreInst::cloneImpl() const { |
4365 | return new StoreInst(getOperand(i_nocapture: 0), getOperand(i_nocapture: 1), isVolatile(), getAlign(), |
4366 | getOrdering(), getSyncScopeID()); |
4367 | } |
4368 | |
4369 | AtomicCmpXchgInst *AtomicCmpXchgInst::cloneImpl() const { |
4370 | AtomicCmpXchgInst *Result = new AtomicCmpXchgInst( |
4371 | getOperand(i_nocapture: 0), getOperand(i_nocapture: 1), getOperand(i_nocapture: 2), getAlign(), |
4372 | getSuccessOrdering(), getFailureOrdering(), getSyncScopeID()); |
4373 | Result->setVolatile(isVolatile()); |
4374 | Result->setWeak(isWeak()); |
4375 | return Result; |
4376 | } |
4377 | |
4378 | AtomicRMWInst *AtomicRMWInst::cloneImpl() const { |
4379 | AtomicRMWInst *Result = |
4380 | new AtomicRMWInst(getOperation(), getOperand(i_nocapture: 0), getOperand(i_nocapture: 1), |
4381 | getAlign(), getOrdering(), getSyncScopeID()); |
4382 | Result->setVolatile(isVolatile()); |
4383 | return Result; |
4384 | } |
4385 | |
4386 | FenceInst *FenceInst::cloneImpl() const { |
4387 | return new FenceInst(getContext(), getOrdering(), getSyncScopeID()); |
4388 | } |
4389 | |
4390 | TruncInst *TruncInst::cloneImpl() const { |
4391 | return new TruncInst(getOperand(i_nocapture: 0), getType()); |
4392 | } |
4393 | |
4394 | ZExtInst *ZExtInst::cloneImpl() const { |
4395 | return new ZExtInst(getOperand(i_nocapture: 0), getType()); |
4396 | } |
4397 | |
4398 | SExtInst *SExtInst::cloneImpl() const { |
4399 | return new SExtInst(getOperand(i_nocapture: 0), getType()); |
4400 | } |
4401 | |
4402 | FPTruncInst *FPTruncInst::cloneImpl() const { |
4403 | return new FPTruncInst(getOperand(i_nocapture: 0), getType()); |
4404 | } |
4405 | |
4406 | FPExtInst *FPExtInst::cloneImpl() const { |
4407 | return new FPExtInst(getOperand(i_nocapture: 0), getType()); |
4408 | } |
4409 | |
4410 | UIToFPInst *UIToFPInst::cloneImpl() const { |
4411 | return new UIToFPInst(getOperand(i_nocapture: 0), getType()); |
4412 | } |
4413 | |
4414 | SIToFPInst *SIToFPInst::cloneImpl() const { |
4415 | return new SIToFPInst(getOperand(i_nocapture: 0), getType()); |
4416 | } |
4417 | |
4418 | FPToUIInst *FPToUIInst::cloneImpl() const { |
4419 | return new FPToUIInst(getOperand(i_nocapture: 0), getType()); |
4420 | } |
4421 | |
4422 | FPToSIInst *FPToSIInst::cloneImpl() const { |
4423 | return new FPToSIInst(getOperand(i_nocapture: 0), getType()); |
4424 | } |
4425 | |
4426 | PtrToIntInst *PtrToIntInst::cloneImpl() const { |
4427 | return new PtrToIntInst(getOperand(i_nocapture: 0), getType()); |
4428 | } |
4429 | |
4430 | IntToPtrInst *IntToPtrInst::cloneImpl() const { |
4431 | return new IntToPtrInst(getOperand(i_nocapture: 0), getType()); |
4432 | } |
4433 | |
4434 | BitCastInst *BitCastInst::cloneImpl() const { |
4435 | return new BitCastInst(getOperand(i_nocapture: 0), getType()); |
4436 | } |
4437 | |
4438 | AddrSpaceCastInst *AddrSpaceCastInst::cloneImpl() const { |
4439 | return new AddrSpaceCastInst(getOperand(i_nocapture: 0), getType()); |
4440 | } |
4441 | |
4442 | CallInst *CallInst::cloneImpl() const { |
4443 | if (hasOperandBundles()) { |
4444 | IntrusiveOperandsAndDescriptorAllocMarker AllocMarker{ |
4445 | .NumOps: getNumOperands(), |
4446 | .DescBytes: getNumOperandBundles() * unsigned(sizeof(BundleOpInfo))}; |
4447 | return new (AllocMarker) CallInst(*this, AllocMarker); |
4448 | } |
4449 | IntrusiveOperandsAllocMarker AllocMarker{.NumOps: getNumOperands()}; |
4450 | return new (AllocMarker) CallInst(*this, AllocMarker); |
4451 | } |
4452 | |
4453 | SelectInst *SelectInst::cloneImpl() const { |
4454 | return SelectInst::Create(C: getOperand(i_nocapture: 0), S1: getOperand(i_nocapture: 1), S2: getOperand(i_nocapture: 2)); |
4455 | } |
4456 | |
4457 | VAArgInst *VAArgInst::cloneImpl() const { |
4458 | return new VAArgInst(getOperand(i_nocapture: 0), getType()); |
4459 | } |
4460 | |
4461 | ExtractElementInst *ExtractElementInst::() const { |
4462 | return ExtractElementInst::Create(Vec: getOperand(i_nocapture: 0), Idx: getOperand(i_nocapture: 1)); |
4463 | } |
4464 | |
4465 | InsertElementInst *InsertElementInst::cloneImpl() const { |
4466 | return InsertElementInst::Create(Vec: getOperand(i_nocapture: 0), NewElt: getOperand(i_nocapture: 1), Idx: getOperand(i_nocapture: 2)); |
4467 | } |
4468 | |
4469 | ShuffleVectorInst *ShuffleVectorInst::cloneImpl() const { |
4470 | return new ShuffleVectorInst(getOperand(i_nocapture: 0), getOperand(i_nocapture: 1), getShuffleMask()); |
4471 | } |
4472 | |
4473 | PHINode *PHINode::cloneImpl() const { return new (AllocMarker) PHINode(*this); } |
4474 | |
4475 | LandingPadInst *LandingPadInst::cloneImpl() const { |
4476 | return new LandingPadInst(*this); |
4477 | } |
4478 | |
4479 | ReturnInst *ReturnInst::cloneImpl() const { |
4480 | IntrusiveOperandsAllocMarker AllocMarker{.NumOps: getNumOperands()}; |
4481 | return new (AllocMarker) ReturnInst(*this, AllocMarker); |
4482 | } |
4483 | |
4484 | BranchInst *BranchInst::cloneImpl() const { |
4485 | IntrusiveOperandsAllocMarker AllocMarker{.NumOps: getNumOperands()}; |
4486 | return new (AllocMarker) BranchInst(*this, AllocMarker); |
4487 | } |
4488 | |
4489 | SwitchInst *SwitchInst::cloneImpl() const { return new SwitchInst(*this); } |
4490 | |
4491 | IndirectBrInst *IndirectBrInst::cloneImpl() const { |
4492 | return new IndirectBrInst(*this); |
4493 | } |
4494 | |
4495 | InvokeInst *InvokeInst::cloneImpl() const { |
4496 | if (hasOperandBundles()) { |
4497 | IntrusiveOperandsAndDescriptorAllocMarker AllocMarker{ |
4498 | .NumOps: getNumOperands(), |
4499 | .DescBytes: getNumOperandBundles() * unsigned(sizeof(BundleOpInfo))}; |
4500 | return new (AllocMarker) InvokeInst(*this, AllocMarker); |
4501 | } |
4502 | IntrusiveOperandsAllocMarker AllocMarker{.NumOps: getNumOperands()}; |
4503 | return new (AllocMarker) InvokeInst(*this, AllocMarker); |
4504 | } |
4505 | |
4506 | CallBrInst *CallBrInst::cloneImpl() const { |
4507 | if (hasOperandBundles()) { |
4508 | IntrusiveOperandsAndDescriptorAllocMarker AllocMarker{ |
4509 | .NumOps: getNumOperands(), |
4510 | .DescBytes: getNumOperandBundles() * unsigned(sizeof(BundleOpInfo))}; |
4511 | return new (AllocMarker) CallBrInst(*this, AllocMarker); |
4512 | } |
4513 | IntrusiveOperandsAllocMarker AllocMarker{.NumOps: getNumOperands()}; |
4514 | return new (AllocMarker) CallBrInst(*this, AllocMarker); |
4515 | } |
4516 | |
4517 | ResumeInst *ResumeInst::cloneImpl() const { |
4518 | return new (AllocMarker) ResumeInst(*this); |
4519 | } |
4520 | |
4521 | CleanupReturnInst *CleanupReturnInst::cloneImpl() const { |
4522 | IntrusiveOperandsAllocMarker AllocMarker{.NumOps: getNumOperands()}; |
4523 | return new (AllocMarker) CleanupReturnInst(*this, AllocMarker); |
4524 | } |
4525 | |
4526 | CatchReturnInst *CatchReturnInst::cloneImpl() const { |
4527 | return new (AllocMarker) CatchReturnInst(*this); |
4528 | } |
4529 | |
4530 | CatchSwitchInst *CatchSwitchInst::cloneImpl() const { |
4531 | return new CatchSwitchInst(*this); |
4532 | } |
4533 | |
4534 | FuncletPadInst *FuncletPadInst::cloneImpl() const { |
4535 | IntrusiveOperandsAllocMarker AllocMarker{.NumOps: getNumOperands()}; |
4536 | return new (AllocMarker) FuncletPadInst(*this, AllocMarker); |
4537 | } |
4538 | |
4539 | UnreachableInst *UnreachableInst::cloneImpl() const { |
4540 | LLVMContext &Context = getContext(); |
4541 | return new UnreachableInst(Context); |
4542 | } |
4543 | |
4544 | bool UnreachableInst::shouldLowerToTrap(bool TrapUnreachable, |
4545 | bool NoTrapAfterNoreturn) const { |
4546 | if (!TrapUnreachable) |
4547 | return false; |
4548 | |
4549 | // We may be able to ignore unreachable behind a noreturn call. |
4550 | if (const CallInst *Call = dyn_cast_or_null<CallInst>(Val: getPrevNode()); |
4551 | Call && Call->doesNotReturn()) { |
4552 | if (NoTrapAfterNoreturn) |
4553 | return false; |
4554 | // Do not emit an additional trap instruction. |
4555 | if (Call->isNonContinuableTrap()) |
4556 | return false; |
4557 | } |
4558 | |
4559 | if (getFunction()->hasFnAttribute(Kind: Attribute::Naked)) |
4560 | return false; |
4561 | |
4562 | return true; |
4563 | } |
4564 | |
4565 | FreezeInst *FreezeInst::cloneImpl() const { |
4566 | return new FreezeInst(getOperand(i_nocapture: 0)); |
4567 | } |
4568 | |