1//===- InferAddressSpace.cpp - --------------------------------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// CUDA C/C++ includes memory space designation as variable type qualifers (such
10// as __global__ and __shared__). Knowing the space of a memory access allows
11// CUDA compilers to emit faster PTX loads and stores. For example, a load from
12// shared memory can be translated to `ld.shared` which is roughly 10% faster
13// than a generic `ld` on an NVIDIA Tesla K40c.
14//
15// Unfortunately, type qualifiers only apply to variable declarations, so CUDA
16// compilers must infer the memory space of an address expression from
17// type-qualified variables.
18//
19// LLVM IR uses non-zero (so-called) specific address spaces to represent memory
20// spaces (e.g. addrspace(3) means shared memory). The Clang frontend
21// places only type-qualified variables in specific address spaces, and then
22// conservatively `addrspacecast`s each type-qualified variable to addrspace(0)
23// (so-called the generic address space) for other instructions to use.
24//
25// For example, the Clang translates the following CUDA code
26// __shared__ float a[10];
27// float v = a[i];
28// to
29// %0 = addrspacecast [10 x float] addrspace(3)* @a to [10 x float]*
30// %1 = gep [10 x float], [10 x float]* %0, i64 0, i64 %i
31// %v = load float, float* %1 ; emits ld.f32
32// @a is in addrspace(3) since it's type-qualified, but its use from %1 is
33// redirected to %0 (the generic version of @a).
34//
35// The optimization implemented in this file propagates specific address spaces
36// from type-qualified variable declarations to its users. For example, it
37// optimizes the above IR to
38// %1 = gep [10 x float] addrspace(3)* @a, i64 0, i64 %i
39// %v = load float addrspace(3)* %1 ; emits ld.shared.f32
40// propagating the addrspace(3) from @a to %1. As the result, the NVPTX
41// codegen is able to emit ld.shared.f32 for %v.
42//
43// Address space inference works in two steps. First, it uses a data-flow
44// analysis to infer as many generic pointers as possible to point to only one
45// specific address space. In the above example, it can prove that %1 only
46// points to addrspace(3). This algorithm was published in
47// CUDA: Compiling and optimizing for a GPU platform
48// Chakrabarti, Grover, Aarts, Kong, Kudlur, Lin, Marathe, Murphy, Wang
49// ICCS 2012
50//
51// Then, address space inference replaces all refinable generic pointers with
52// equivalent specific pointers.
53//
54// The major challenge of implementing this optimization is handling PHINodes,
55// which may create loops in the data flow graph. This brings two complications.
56//
57// First, the data flow analysis in Step 1 needs to be circular. For example,
58// %generic.input = addrspacecast float addrspace(3)* %input to float*
59// loop:
60// %y = phi [ %generic.input, %y2 ]
61// %y2 = getelementptr %y, 1
62// %v = load %y2
63// br ..., label %loop, ...
64// proving %y specific requires proving both %generic.input and %y2 specific,
65// but proving %y2 specific circles back to %y. To address this complication,
66// the data flow analysis operates on a lattice:
67// uninitialized > specific address spaces > generic.
68// All address expressions (our implementation only considers phi, bitcast,
69// addrspacecast, and getelementptr) start with the uninitialized address space.
70// The monotone transfer function moves the address space of a pointer down a
71// lattice path from uninitialized to specific and then to generic. A join
72// operation of two different specific address spaces pushes the expression down
73// to the generic address space. The analysis completes once it reaches a fixed
74// point.
75//
76// Second, IR rewriting in Step 2 also needs to be circular. For example,
77// converting %y to addrspace(3) requires the compiler to know the converted
78// %y2, but converting %y2 needs the converted %y. To address this complication,
79// we break these cycles using "poison" placeholders. When converting an
80// instruction `I` to a new address space, if its operand `Op` is not converted
81// yet, we let `I` temporarily use `poison` and fix all the uses later.
82// For instance, our algorithm first converts %y to
83// %y' = phi float addrspace(3)* [ %input, poison ]
84// Then, it converts %y2 to
85// %y2' = getelementptr %y', 1
86// Finally, it fixes the poison in %y' so that
87// %y' = phi float addrspace(3)* [ %input, %y2' ]
88//
89//===----------------------------------------------------------------------===//
90
91#include "llvm/Transforms/Scalar/InferAddressSpaces.h"
92#include "llvm/ADT/ArrayRef.h"
93#include "llvm/ADT/DenseMap.h"
94#include "llvm/ADT/DenseSet.h"
95#include "llvm/ADT/SetVector.h"
96#include "llvm/ADT/SmallVector.h"
97#include "llvm/Analysis/AssumptionCache.h"
98#include "llvm/Analysis/TargetTransformInfo.h"
99#include "llvm/Analysis/ValueTracking.h"
100#include "llvm/IR/Argument.h"
101#include "llvm/IR/BasicBlock.h"
102#include "llvm/IR/Constant.h"
103#include "llvm/IR/Constants.h"
104#include "llvm/IR/Dominators.h"
105#include "llvm/IR/Function.h"
106#include "llvm/IR/IRBuilder.h"
107#include "llvm/IR/InstIterator.h"
108#include "llvm/IR/Instruction.h"
109#include "llvm/IR/Instructions.h"
110#include "llvm/IR/IntrinsicInst.h"
111#include "llvm/IR/Intrinsics.h"
112#include "llvm/IR/LLVMContext.h"
113#include "llvm/IR/Operator.h"
114#include "llvm/IR/PassManager.h"
115#include "llvm/IR/PatternMatch.h"
116#include "llvm/IR/Type.h"
117#include "llvm/IR/Use.h"
118#include "llvm/IR/User.h"
119#include "llvm/IR/Value.h"
120#include "llvm/IR/ValueHandle.h"
121#include "llvm/InitializePasses.h"
122#include "llvm/Pass.h"
123#include "llvm/Support/Casting.h"
124#include "llvm/Support/Debug.h"
125#include "llvm/Support/ErrorHandling.h"
126#include "llvm/Support/KnownBits.h"
127#include "llvm/Support/raw_ostream.h"
128#include "llvm/Transforms/Scalar.h"
129#include "llvm/Transforms/Utils/Local.h"
130#include "llvm/Transforms/Utils/ValueMapper.h"
131#include <cassert>
132#include <iterator>
133#include <limits>
134#include <optional>
135#include <utility>
136#include <vector>
137
138#define DEBUG_TYPE "infer-address-spaces"
139
140using namespace llvm;
141using namespace llvm::PatternMatch;
142
143static const unsigned UninitializedAddressSpace =
144 std::numeric_limits<unsigned>::max();
145
146namespace {
147
148using ValueToAddrSpaceMapTy = DenseMap<const Value *, unsigned>;
149// Different from ValueToAddrSpaceMapTy, where a new addrspace is inferred on
150// the *def* of a value, PredicatedAddrSpaceMapTy is map where a new
151// addrspace is inferred on the *use* of a pointer. This map is introduced to
152// infer addrspace from the addrspace predicate assumption built from assume
153// intrinsic. In that scenario, only specific uses (under valid assumption
154// context) could be inferred with a new addrspace.
155using PredicatedAddrSpaceMapTy =
156 DenseMap<std::pair<const Value *, const Value *>, unsigned>;
157using PostorderStackTy = llvm::SmallVector<PointerIntPair<Value *, 1, bool>, 4>;
158
159class InferAddressSpaces : public FunctionPass {
160 unsigned FlatAddrSpace = 0;
161
162public:
163 static char ID;
164
165 InferAddressSpaces()
166 : FunctionPass(ID), FlatAddrSpace(UninitializedAddressSpace) {
167 initializeInferAddressSpacesPass(*PassRegistry::getPassRegistry());
168 }
169 InferAddressSpaces(unsigned AS) : FunctionPass(ID), FlatAddrSpace(AS) {
170 initializeInferAddressSpacesPass(*PassRegistry::getPassRegistry());
171 }
172
173 void getAnalysisUsage(AnalysisUsage &AU) const override {
174 AU.setPreservesCFG();
175 AU.addRequired<AssumptionCacheTracker>();
176 AU.addRequired<TargetTransformInfoWrapperPass>();
177 }
178
179 bool runOnFunction(Function &F) override;
180};
181
182class InferAddressSpacesImpl {
183 AssumptionCache &AC;
184 Function *F = nullptr;
185 const DominatorTree *DT = nullptr;
186 const TargetTransformInfo *TTI = nullptr;
187 const DataLayout *DL = nullptr;
188
189 /// Target specific address space which uses of should be replaced if
190 /// possible.
191 unsigned FlatAddrSpace = 0;
192
193 /// The default address space is assumed as the flat address space. This is
194 /// mainly for test purpose.
195 const bool AssumeDefaultIsFlatAddressSpace = false;
196
197 DenseMap<const Value *, Value *> PtrIntCastPairs;
198
199 // Tries to find if the inttoptr instruction is derived from an pointer have
200 // specific address space, and is safe to propagate the address space to the
201 // new pointer that inttoptr produces.
202 Value *getIntToPtrPointerOperand(const Operator *I2P) const;
203 // Tries to find if the inttoptr instruction is derived from an pointer have
204 // specific address space, and is safe to propagate the address space to the
205 // new pointer that inttoptr produces. If the old pointer is found, cache the
206 // <OldPtr, inttoptr> pairs to a map.
207 void collectIntToPtrPointerOperand();
208 // Check if an old pointer is found ahead of time. The safety has been checked
209 // when collecting the inttoptr original pointer and the result is cached in
210 // PtrIntCastPairs.
211 bool isSafeToCastIntToPtrAddrSpace(const Operator *I2P) const {
212 return PtrIntCastPairs.contains(Val: I2P);
213 }
214 bool isAddressExpression(const Value &V, const DataLayout &DL,
215 const TargetTransformInfo *TTI) const;
216 Value *cloneConstantExprWithNewAddressSpace(
217 ConstantExpr *CE, unsigned NewAddrSpace,
218 const ValueToValueMapTy &ValueWithNewAddrSpace, const DataLayout *DL,
219 const TargetTransformInfo *TTI) const;
220
221 SmallVector<Value *, 2>
222 getPointerOperands(const Value &V, const DataLayout &DL,
223 const TargetTransformInfo *TTI) const;
224
225 // Try to update the address space of V. If V is updated, returns true and
226 // false otherwise.
227 bool updateAddressSpace(const Value &V,
228 ValueToAddrSpaceMapTy &InferredAddrSpace,
229 PredicatedAddrSpaceMapTy &PredicatedAS) const;
230
231 // Adds the users of V whose address space may still change to Worklist.
232 void enqueueUsers(Value &V, const ValueToAddrSpaceMapTy &InferredAddrSpace,
233 SetVector<Value *> &Worklist) const;
234
235 // Propagates address spaces out of Worklist until nothing changes.
236 void runToFixPoint(SetVector<Value *> &Worklist,
237 ValueToAddrSpaceMapTy &InferredAddrSpace,
238 PredicatedAddrSpaceMapTy &PredicatedAS) const;
239
240 // Tries to infer the specific address space of each address expression in
241 // Postorder.
242 void inferAddressSpaces(ArrayRef<WeakTrackingVH> Postorder,
243 ValueToAddrSpaceMapTy &InferredAddrSpace,
244 PredicatedAddrSpaceMapTy &PredicatedAS) const;
245
246 bool isSafeToCastConstAddrSpace(Constant *C, unsigned NewAS) const;
247
248 Value *clonePtrMaskWithNewAddressSpace(
249 IntrinsicInst *I, unsigned NewAddrSpace,
250 const ValueToValueMapTy &ValueWithNewAddrSpace,
251 const PredicatedAddrSpaceMapTy &PredicatedAS,
252 SmallVectorImpl<const Use *> *PoisonUsesToFix) const;
253
254 Value *cloneInstructionWithNewAddressSpace(
255 Instruction *I, unsigned NewAddrSpace,
256 const ValueToValueMapTy &ValueWithNewAddrSpace,
257 const PredicatedAddrSpaceMapTy &PredicatedAS,
258 SmallVectorImpl<const Use *> *PoisonUsesToFix) const;
259
260 void performPointerReplacement(
261 Value *V, Value *NewV, Use &U, ValueToValueMapTy &ValueWithNewAddrSpace,
262 SmallVectorImpl<Instruction *> &DeadInstructions) const;
263
264 // Changes the flat address expressions in function F to point to specific
265 // address spaces if InferredAddrSpace says so. Postorder is the postorder of
266 // all flat expressions in the use-def graph of function F.
267 bool rewriteWithNewAddressSpaces(
268 ArrayRef<WeakTrackingVH> Postorder,
269 const ValueToAddrSpaceMapTy &InferredAddrSpace,
270 const PredicatedAddrSpaceMapTy &PredicatedAS) const;
271
272 void appendsFlatAddressExpressionToPostorderStack(
273 Value *V, PostorderStackTy &PostorderStack,
274 DenseSet<Value *> &Visited) const;
275
276 bool rewriteIntrinsicOperands(IntrinsicInst *II, Value *OldV,
277 Value *NewV) const;
278 void collectRewritableIntrinsicOperands(IntrinsicInst *II,
279 PostorderStackTy &PostorderStack,
280 DenseSet<Value *> &Visited) const;
281
282 std::vector<WeakTrackingVH> collectFlatAddressExpressions(Function &F) const;
283
284 Value *cloneValueWithNewAddressSpace(
285 Value *V, unsigned NewAddrSpace,
286 const ValueToValueMapTy &ValueWithNewAddrSpace,
287 const PredicatedAddrSpaceMapTy &PredicatedAS,
288 SmallVectorImpl<const Use *> *PoisonUsesToFix) const;
289 unsigned joinAddressSpaces(unsigned AS1, unsigned AS2) const;
290
291 unsigned getPredicatedAddrSpace(const Value &PtrV,
292 const Value *UserCtx) const;
293
294public:
295 InferAddressSpacesImpl(AssumptionCache &AC, const DominatorTree *DT,
296 const TargetTransformInfo *TTI, unsigned FlatAddrSpace,
297 bool AssumeDefaultIsFlatAddressSpace)
298 : AC(AC), DT(DT), TTI(TTI), FlatAddrSpace(FlatAddrSpace),
299 AssumeDefaultIsFlatAddressSpace(AssumeDefaultIsFlatAddressSpace) {}
300 bool run(Function &F);
301};
302
303} // end anonymous namespace
304
305char InferAddressSpaces::ID = 0;
306
307INITIALIZE_PASS_BEGIN(InferAddressSpaces, DEBUG_TYPE, "Infer address spaces",
308 false, false)
309INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
310INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
311INITIALIZE_PASS_END(InferAddressSpaces, DEBUG_TYPE, "Infer address spaces",
312 false, false)
313
314static Type *getPtrOrVecOfPtrsWithNewAS(Type *Ty, unsigned NewAddrSpace) {
315 assert(Ty->isPtrOrPtrVectorTy());
316 PointerType *NPT = PointerType::get(C&: Ty->getContext(), AddressSpace: NewAddrSpace);
317 return Ty->getWithNewType(EltTy: NPT);
318}
319
320// Check whether that's no-op pointer bitcast using a pair of
321// `ptrtoint`/`inttoptr` due to the missing no-op pointer bitcast over
322// different address spaces.
323static bool isNoopPtrIntCastPair(const Operator *I2P, const DataLayout &DL,
324 const TargetTransformInfo *TTI) {
325 assert(I2P->getOpcode() == Instruction::IntToPtr);
326 auto *P2I = dyn_cast<Operator>(Val: I2P->getOperand(i: 0));
327 if (!P2I || P2I->getOpcode() != Instruction::PtrToInt)
328 return false;
329 // Check it's really safe to treat that pair of `ptrtoint`/`inttoptr` as a
330 // no-op cast. Besides checking both of them are no-op casts, as the
331 // reinterpreted pointer may be used in other pointer arithmetic, we also
332 // need to double-check that through the target-specific hook. That ensures
333 // the underlying target also agrees that's a no-op address space cast and
334 // pointer bits are preserved.
335 // The current IR spec doesn't have clear rules on address space casts,
336 // especially a clear definition for pointer bits in non-default address
337 // spaces. It would be undefined if that pointer is dereferenced after an
338 // invalid reinterpret cast. Also, due to the unclearness for the meaning of
339 // bits in non-default address spaces in the current spec, the pointer
340 // arithmetic may also be undefined after invalid pointer reinterpret cast.
341 // However, as we confirm through the target hooks that it's a no-op
342 // addrspacecast, it doesn't matter since the bits should be the same.
343 unsigned P2IOp0AS = P2I->getOperand(i: 0)->getType()->getPointerAddressSpace();
344 unsigned I2PAS = I2P->getType()->getPointerAddressSpace();
345 return CastInst::isNoopCast(Opcode: Instruction::CastOps(I2P->getOpcode()),
346 SrcTy: I2P->getOperand(i: 0)->getType(), DstTy: I2P->getType(),
347 DL) &&
348 CastInst::isNoopCast(Opcode: Instruction::CastOps(P2I->getOpcode()),
349 SrcTy: P2I->getOperand(i: 0)->getType(), DstTy: P2I->getType(),
350 DL) &&
351 (P2IOp0AS == I2PAS || TTI->isNoopAddrSpaceCast(FromAS: P2IOp0AS, ToAS: I2PAS));
352}
353
354// Returns true if V is an address expression.
355// TODO: Currently, we only consider:
356// - arguments
357// - phi, bitcast, addrspacecast, and getelementptr operators
358bool InferAddressSpacesImpl::isAddressExpression(
359 const Value &V, const DataLayout &DL,
360 const TargetTransformInfo *TTI) const {
361
362 if (const Argument *Arg = dyn_cast<Argument>(Val: &V))
363 return Arg->getType()->isPointerTy() &&
364 TTI->getAssumedAddrSpace(V: &V) != UninitializedAddressSpace;
365
366 const Operator *Op = dyn_cast<Operator>(Val: &V);
367 if (!Op)
368 return false;
369
370 switch (Op->getOpcode()) {
371 case Instruction::PHI:
372 assert(Op->getType()->isPtrOrPtrVectorTy());
373 return true;
374 case Instruction::BitCast:
375 case Instruction::AddrSpaceCast:
376 case Instruction::GetElementPtr:
377 return true;
378 case Instruction::Select:
379 return Op->getType()->isPtrOrPtrVectorTy();
380 case Instruction::Call: {
381 const IntrinsicInst *II = dyn_cast<IntrinsicInst>(Val: &V);
382 return II && II->getIntrinsicID() == Intrinsic::ptrmask;
383 }
384 case Instruction::IntToPtr:
385 return isNoopPtrIntCastPair(I2P: Op, DL, TTI) ||
386 isSafeToCastIntToPtrAddrSpace(I2P: Op);
387 default:
388 // That value is an address expression if it has an assumed address space.
389 return TTI->getAssumedAddrSpace(V: &V) != UninitializedAddressSpace;
390 }
391}
392
393// Returns the pointer operands of V.
394//
395// Precondition: V is an address expression.
396SmallVector<Value *, 2> InferAddressSpacesImpl::getPointerOperands(
397 const Value &V, const DataLayout &DL,
398 const TargetTransformInfo *TTI) const {
399 if (isa<Argument>(Val: &V))
400 return {};
401
402 const Operator &Op = cast<Operator>(Val: V);
403 switch (Op.getOpcode()) {
404 case Instruction::PHI: {
405 auto IncomingValues = cast<PHINode>(Val: Op).incoming_values();
406 return {IncomingValues.begin(), IncomingValues.end()};
407 }
408 case Instruction::BitCast:
409 case Instruction::AddrSpaceCast:
410 case Instruction::GetElementPtr:
411 return {Op.getOperand(i: 0)};
412 case Instruction::Select:
413 return {Op.getOperand(i: 1), Op.getOperand(i: 2)};
414 case Instruction::Call: {
415 const IntrinsicInst &II = cast<IntrinsicInst>(Val: Op);
416 assert(II.getIntrinsicID() == Intrinsic::ptrmask &&
417 "unexpected intrinsic call");
418 return {II.getArgOperand(i: 0)};
419 }
420 case Instruction::IntToPtr: {
421 if (isNoopPtrIntCastPair(I2P: &Op, DL, TTI)) {
422 auto *P2I = cast<Operator>(Val: Op.getOperand(i: 0));
423 return {P2I->getOperand(i: 0)};
424 }
425 assert(isSafeToCastIntToPtrAddrSpace(&Op));
426 return {getIntToPtrPointerOperand(I2P: &Op)};
427 }
428 default:
429 llvm_unreachable("Unexpected instruction type.");
430 }
431}
432
433// Return mask. The 1 in mask indicate the bit is changed.
434// This helper function is to compute the max know changed bits for ptr1 and
435// ptr2 after the operation `ptr2 = ptr1 Op Mask`.
436static APInt computeMaxChangedPtrBits(const Operator *Op, const Value *Mask,
437 const DataLayout &DL, AssumptionCache *AC,
438 const DominatorTree *DT) {
439 KnownBits Known = computeKnownBits(V: Mask, DL, AC, CxtI: nullptr, DT);
440 switch (Op->getOpcode()) {
441 case Instruction::Xor:
442 case Instruction::Or:
443 return ~Known.Zero;
444 case Instruction::And:
445 return ~Known.One;
446 default:
447 return APInt::getAllOnes(numBits: Known.getBitWidth());
448 }
449}
450
451Value *
452InferAddressSpacesImpl::getIntToPtrPointerOperand(const Operator *I2P) const {
453 assert(I2P->getOpcode() == Instruction::IntToPtr);
454 if (I2P->getType()->isVectorTy())
455 return nullptr;
456
457 // If I2P has been accessed and has the corresponding old pointer value, just
458 // return true.
459 if (auto *OldPtr = PtrIntCastPairs.lookup(Val: I2P))
460 return OldPtr;
461
462 Value *LogicalOp = I2P->getOperand(i: 0);
463 Value *OldPtr, *Mask;
464 if (!match(V: LogicalOp,
465 P: m_c_BitwiseLogic(L: m_PtrToInt(Op: m_Value(V&: OldPtr)), R: m_Value(V&: Mask))))
466 return nullptr;
467
468 Operator *AsCast = dyn_cast<AddrSpaceCastOperator>(Val: OldPtr);
469 if (!AsCast)
470 return nullptr;
471
472 unsigned SrcAS = I2P->getType()->getPointerAddressSpace();
473 unsigned DstAS = AsCast->getOperand(i: 0)->getType()->getPointerAddressSpace();
474 APInt PreservedPtrMask = TTI->getAddrSpaceCastPreservedPtrMask(SrcAS, DstAS);
475 if (PreservedPtrMask.isZero())
476 return nullptr;
477 APInt ChangedPtrBits =
478 computeMaxChangedPtrBits(Op: cast<Operator>(Val: LogicalOp), Mask, DL: *DL, AC: &AC, DT);
479 // Check if the address bits change is within the preserved mask. If the bits
480 // change is not preserved, it is not safe to perform address space cast.
481 // The following pattern is not safe to cast address space.
482 // %1 = ptrtoint ptr addrspace(3) %sp to i32
483 // %2 = zext i32 %1 to i64
484 // %gp = inttoptr i64 %2 to ptr
485 assert(ChangedPtrBits.getBitWidth() == PreservedPtrMask.getBitWidth());
486 if (ChangedPtrBits.isSubsetOf(RHS: PreservedPtrMask))
487 return OldPtr;
488
489 return nullptr;
490}
491
492void InferAddressSpacesImpl::collectIntToPtrPointerOperand() {
493 // Only collect inttoptr instruction.
494 // TODO: We need to collect inttoptr constant expression as well.
495 for (Instruction &I : instructions(F)) {
496 if (!dyn_cast<IntToPtrInst>(Val: &I))
497 continue;
498 if (auto *OldPtr = getIntToPtrPointerOperand(I2P: cast<Operator>(Val: &I)))
499 PtrIntCastPairs.insert(KV: {&I, OldPtr});
500 }
501}
502
503bool InferAddressSpacesImpl::rewriteIntrinsicOperands(IntrinsicInst *II,
504 Value *OldV,
505 Value *NewV) const {
506 Module *M = II->getParent()->getParent()->getParent();
507 Intrinsic::ID IID = II->getIntrinsicID();
508 switch (IID) {
509 case Intrinsic::objectsize:
510 case Intrinsic::masked_load: {
511 Type *DestTy = II->getType();
512 Type *SrcTy = NewV->getType();
513 Function *NewDecl =
514 Intrinsic::getOrInsertDeclaration(M, id: IID, OverloadTys: {DestTy, SrcTy});
515 II->setArgOperand(i: 0, v: NewV);
516 II->setCalledFunction(NewDecl);
517 return true;
518 }
519 case Intrinsic::ptrmask:
520 // This is handled as an address expression, not as a use memory operation.
521 return false;
522 case Intrinsic::masked_gather: {
523 Type *RetTy = II->getType();
524 Type *NewPtrTy = NewV->getType();
525 Function *NewDecl =
526 Intrinsic::getOrInsertDeclaration(M, id: IID, OverloadTys: {RetTy, NewPtrTy});
527 II->setArgOperand(i: 0, v: NewV);
528 II->setCalledFunction(NewDecl);
529 return true;
530 }
531 case Intrinsic::masked_store:
532 case Intrinsic::masked_scatter: {
533 Type *ValueTy = II->getOperand(i_nocapture: 0)->getType();
534 Type *NewPtrTy = NewV->getType();
535 Function *NewDecl = Intrinsic::getOrInsertDeclaration(
536 M, id: II->getIntrinsicID(), OverloadTys: {ValueTy, NewPtrTy});
537 II->setArgOperand(i: 1, v: NewV);
538 II->setCalledFunction(NewDecl);
539 return true;
540 }
541 case Intrinsic::prefetch:
542 case Intrinsic::is_constant: {
543 Function *NewDecl = Intrinsic::getOrInsertDeclaration(
544 M, id: II->getIntrinsicID(), OverloadTys: {NewV->getType()});
545 II->setArgOperand(i: 0, v: NewV);
546 II->setCalledFunction(NewDecl);
547 return true;
548 }
549 case Intrinsic::fake_use: {
550 II->replaceUsesOfWith(From: OldV, To: NewV);
551 return true;
552 }
553 case Intrinsic::lifetime_start:
554 case Intrinsic::lifetime_end: {
555 // Always force lifetime markers to work directly on the alloca.
556 NewV = NewV->stripPointerCasts();
557 Function *NewDecl = Intrinsic::getOrInsertDeclaration(
558 M, id: II->getIntrinsicID(), OverloadTys: {NewV->getType()});
559 II->setArgOperand(i: 0, v: NewV);
560 II->setCalledFunction(NewDecl);
561 return true;
562 }
563 default: {
564 Value *Rewrite = TTI->rewriteIntrinsicWithAddressSpace(II, OldV, NewV);
565 if (!Rewrite)
566 return false;
567 if (Rewrite != II)
568 II->replaceAllUsesWith(V: Rewrite);
569 return true;
570 }
571 }
572}
573
574void InferAddressSpacesImpl::collectRewritableIntrinsicOperands(
575 IntrinsicInst *II, PostorderStackTy &PostorderStack,
576 DenseSet<Value *> &Visited) const {
577 auto IID = II->getIntrinsicID();
578 switch (IID) {
579 case Intrinsic::ptrmask:
580 case Intrinsic::objectsize:
581 appendsFlatAddressExpressionToPostorderStack(V: II->getArgOperand(i: 0),
582 PostorderStack, Visited);
583 break;
584 case Intrinsic::is_constant: {
585 Value *Ptr = II->getArgOperand(i: 0);
586 if (Ptr->getType()->isPtrOrPtrVectorTy()) {
587 appendsFlatAddressExpressionToPostorderStack(V: Ptr, PostorderStack,
588 Visited);
589 }
590
591 break;
592 }
593 case Intrinsic::masked_load:
594 case Intrinsic::masked_gather:
595 case Intrinsic::prefetch:
596 appendsFlatAddressExpressionToPostorderStack(V: II->getArgOperand(i: 0),
597 PostorderStack, Visited);
598 break;
599 case Intrinsic::masked_store:
600 case Intrinsic::masked_scatter:
601 appendsFlatAddressExpressionToPostorderStack(V: II->getArgOperand(i: 1),
602 PostorderStack, Visited);
603 break;
604 case Intrinsic::fake_use: {
605 for (Value *Op : II->operands()) {
606 if (Op->getType()->isPtrOrPtrVectorTy()) {
607 appendsFlatAddressExpressionToPostorderStack(V: Op, PostorderStack,
608 Visited);
609 }
610 }
611
612 break;
613 }
614 case Intrinsic::lifetime_start:
615 case Intrinsic::lifetime_end: {
616 appendsFlatAddressExpressionToPostorderStack(V: II->getArgOperand(i: 0),
617 PostorderStack, Visited);
618 break;
619 }
620 default:
621 SmallVector<int, 2> OpIndexes;
622 if (TTI->collectFlatAddressOperands(OpIndexes, IID)) {
623 for (int Idx : OpIndexes) {
624 appendsFlatAddressExpressionToPostorderStack(V: II->getArgOperand(i: Idx),
625 PostorderStack, Visited);
626 }
627 }
628 break;
629 }
630}
631
632// Returns all flat address expressions in function F. The elements are
633// If V is an unvisited flat address expression, appends V to PostorderStack
634// and marks it as visited.
635void InferAddressSpacesImpl::appendsFlatAddressExpressionToPostorderStack(
636 Value *V, PostorderStackTy &PostorderStack,
637 DenseSet<Value *> &Visited) const {
638 assert(V->getType()->isPtrOrPtrVectorTy());
639
640 // Generic addressing expressions may be hidden in nested constant
641 // expressions.
642 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Val: V)) {
643 // TODO: Look in non-address parts, like icmp operands.
644 if (isAddressExpression(V: *CE, DL: *DL, TTI) && Visited.insert(V: CE).second)
645 PostorderStack.emplace_back(Args&: CE, Args: false);
646
647 return;
648 }
649
650 if (V->getType()->getPointerAddressSpace() == FlatAddrSpace &&
651 isAddressExpression(V: *V, DL: *DL, TTI)) {
652 if (Visited.insert(V).second) {
653 PostorderStack.emplace_back(Args&: V, Args: false);
654
655 if (auto *Op = dyn_cast<Operator>(Val: V))
656 for (auto &O : Op->operands())
657 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Val&: O))
658 if (isAddressExpression(V: *CE, DL: *DL, TTI) && Visited.insert(V: CE).second)
659 PostorderStack.emplace_back(Args&: CE, Args: false);
660 }
661 }
662}
663
664// Returns all flat address expressions in function F. The elements are ordered
665// in postorder.
666std::vector<WeakTrackingVH>
667InferAddressSpacesImpl::collectFlatAddressExpressions(Function &F) const {
668 // This function implements a non-recursive postorder traversal of a partial
669 // use-def graph of function F.
670 PostorderStackTy PostorderStack;
671 // The set of visited expressions.
672 DenseSet<Value *> Visited;
673
674 auto PushPtrOperand = [&](Value *Ptr) {
675 appendsFlatAddressExpressionToPostorderStack(V: Ptr, PostorderStack, Visited);
676 };
677
678 // Look at operations that may be interesting accelerate by moving to a known
679 // address space. We aim at generating after loads and stores, but pure
680 // addressing calculations may also be faster.
681 for (Instruction &I : instructions(F)) {
682 if (auto *GEP = dyn_cast<GetElementPtrInst>(Val: &I)) {
683 PushPtrOperand(GEP->getPointerOperand());
684 } else if (auto *LI = dyn_cast<LoadInst>(Val: &I))
685 PushPtrOperand(LI->getPointerOperand());
686 else if (auto *SI = dyn_cast<StoreInst>(Val: &I))
687 PushPtrOperand(SI->getPointerOperand());
688 else if (auto *RMW = dyn_cast<AtomicRMWInst>(Val: &I))
689 PushPtrOperand(RMW->getPointerOperand());
690 else if (auto *CmpX = dyn_cast<AtomicCmpXchgInst>(Val: &I))
691 PushPtrOperand(CmpX->getPointerOperand());
692 else if (auto *MI = dyn_cast<MemIntrinsic>(Val: &I)) {
693 // For memset/memcpy/memmove, any pointer operand can be replaced.
694 PushPtrOperand(MI->getRawDest());
695
696 // Handle 2nd operand for memcpy/memmove.
697 if (auto *MTI = dyn_cast<MemTransferInst>(Val: MI))
698 PushPtrOperand(MTI->getRawSource());
699 } else if (auto *II = dyn_cast<IntrinsicInst>(Val: &I))
700 collectRewritableIntrinsicOperands(II, PostorderStack, Visited);
701 else if (ICmpInst *Cmp = dyn_cast<ICmpInst>(Val: &I)) {
702 if (Cmp->getOperand(i_nocapture: 0)->getType()->isPtrOrPtrVectorTy()) {
703 PushPtrOperand(Cmp->getOperand(i_nocapture: 0));
704 PushPtrOperand(Cmp->getOperand(i_nocapture: 1));
705 }
706 } else if (auto *ASC = dyn_cast<AddrSpaceCastInst>(Val: &I)) {
707 PushPtrOperand(ASC->getPointerOperand());
708 } else if (auto *I2P = dyn_cast<IntToPtrInst>(Val: &I)) {
709 if (isNoopPtrIntCastPair(I2P: cast<Operator>(Val: I2P), DL: *DL, TTI))
710 PushPtrOperand(cast<Operator>(Val: I2P->getOperand(i_nocapture: 0))->getOperand(i: 0));
711 else if (isSafeToCastIntToPtrAddrSpace(I2P: cast<Operator>(Val: I2P)))
712 PushPtrOperand(getIntToPtrPointerOperand(I2P: cast<Operator>(Val: I2P)));
713 } else if (auto *RI = dyn_cast<ReturnInst>(Val: &I)) {
714 if (auto *RV = RI->getReturnValue();
715 RV && RV->getType()->isPtrOrPtrVectorTy())
716 PushPtrOperand(RV);
717 }
718 }
719
720 std::vector<WeakTrackingVH> Postorder; // The resultant postorder.
721 while (!PostorderStack.empty()) {
722 Value *TopVal = PostorderStack.back().getPointer();
723 // If the operands of the expression on the top are already explored,
724 // adds that expression to the resultant postorder.
725 if (PostorderStack.back().getInt()) {
726 if (TopVal->getType()->getPointerAddressSpace() == FlatAddrSpace)
727 Postorder.push_back(x: TopVal);
728 PostorderStack.pop_back();
729 continue;
730 }
731 // Otherwise, adds its operands to the stack and explores them.
732 PostorderStack.back().setInt(true);
733 // Skip values with an assumed address space.
734 if (TTI->getAssumedAddrSpace(V: TopVal) == UninitializedAddressSpace) {
735 for (Value *PtrOperand : getPointerOperands(V: *TopVal, DL: *DL, TTI)) {
736 appendsFlatAddressExpressionToPostorderStack(V: PtrOperand, PostorderStack,
737 Visited);
738 }
739 }
740 }
741 return Postorder;
742}
743
744// Inserts an addrspacecast for a phi node operand, handling the proper
745// insertion position based on the operand type.
746static Value *phiNodeOperandWithNewAddressSpace(AddrSpaceCastInst *NewI,
747 Value *Operand) {
748 auto InsertBefore = [NewI](auto It) {
749 NewI->insertBefore(It);
750 NewI->setDebugLoc(It->getDebugLoc());
751 return NewI;
752 };
753
754 if (auto *Arg = dyn_cast<Argument>(Val: Operand)) {
755 // For arguments, insert the cast at the beginning of entry block.
756 // Consider inserting at the dominating block for better placement.
757 Function *F = Arg->getParent();
758 auto InsertI = F->getEntryBlock().getFirstNonPHIIt();
759 return InsertBefore(InsertI);
760 }
761
762 // No check for Constant here, as constants are already handled.
763 assert(isa<Instruction>(Operand));
764
765 Instruction *OpInst = cast<Instruction>(Val: Operand);
766 if (LLVM_UNLIKELY(OpInst->getOpcode() == Instruction::PHI)) {
767 // If the operand is defined by another PHI node, insert after the first
768 // non-PHI instruction at the corresponding basic block.
769 auto InsertI = OpInst->getParent()->getFirstNonPHIIt();
770 return InsertBefore(InsertI);
771 }
772
773 // Otherwise, insert immediately after the operand definition.
774 NewI->insertAfter(InsertPos: OpInst->getIterator());
775 NewI->setDebugLoc(OpInst->getDebugLoc());
776 return NewI;
777}
778
779// A helper function for cloneInstructionWithNewAddressSpace. Returns the clone
780// of OperandUse.get() in the new address space. If the clone is not ready yet,
781// returns poison in the new address space as a placeholder.
782static Value *operandWithNewAddressSpaceOrCreatePoison(
783 const Use &OperandUse, unsigned NewAddrSpace,
784 const ValueToValueMapTy &ValueWithNewAddrSpace,
785 const PredicatedAddrSpaceMapTy &PredicatedAS,
786 SmallVectorImpl<const Use *> *PoisonUsesToFix) {
787 Value *Operand = OperandUse.get();
788
789 Type *NewPtrTy = getPtrOrVecOfPtrsWithNewAS(Ty: Operand->getType(), NewAddrSpace);
790
791 if (Constant *C = dyn_cast<Constant>(Val: Operand))
792 return ConstantExpr::getAddrSpaceCast(C, Ty: NewPtrTy);
793
794 if (Value *NewOperand = ValueWithNewAddrSpace.lookup(Val: Operand))
795 return NewOperand;
796
797 Instruction *Inst = cast<Instruction>(Val: OperandUse.getUser());
798 auto I = PredicatedAS.find(Val: std::make_pair(x&: Inst, y&: Operand));
799 if (I != PredicatedAS.end()) {
800 // Insert an addrspacecast on that operand before the user.
801 unsigned NewAS = I->second;
802 Type *NewPtrTy = getPtrOrVecOfPtrsWithNewAS(Ty: Operand->getType(), NewAddrSpace: NewAS);
803 auto *NewI = new AddrSpaceCastInst(Operand, NewPtrTy);
804
805 if (LLVM_UNLIKELY(Inst->getOpcode() == Instruction::PHI))
806 return phiNodeOperandWithNewAddressSpace(NewI, Operand);
807
808 NewI->insertBefore(InsertPos: Inst->getIterator());
809 NewI->setDebugLoc(Inst->getDebugLoc());
810 return NewI;
811 }
812
813 PoisonUsesToFix->push_back(Elt: &OperandUse);
814 return PoisonValue::get(T: NewPtrTy);
815}
816
817// A helper function for cloneInstructionWithNewAddressSpace. Handles the
818// conversion of a ptrmask intrinsic instruction.
819Value *InferAddressSpacesImpl::clonePtrMaskWithNewAddressSpace(
820 IntrinsicInst *I, unsigned NewAddrSpace,
821 const ValueToValueMapTy &ValueWithNewAddrSpace,
822 const PredicatedAddrSpaceMapTy &PredicatedAS,
823 SmallVectorImpl<const Use *> *PoisonUsesToFix) const {
824 const Use &PtrOpUse = I->getArgOperandUse(i: 0);
825 unsigned OldAddrSpace = PtrOpUse->getType()->getPointerAddressSpace();
826 Value *MaskOp = I->getArgOperand(i: 1);
827 Type *MaskTy = MaskOp->getType();
828
829 KnownBits OldPtrBits{DL->getPointerSizeInBits(AS: OldAddrSpace)};
830 KnownBits NewPtrBits{DL->getPointerSizeInBits(AS: NewAddrSpace)};
831 if (!TTI->isNoopAddrSpaceCast(FromAS: OldAddrSpace, ToAS: NewAddrSpace)) {
832 std::tie(args&: OldPtrBits, args&: NewPtrBits) =
833 TTI->computeKnownBitsAddrSpaceCast(ToAS: NewAddrSpace, PtrOp: *PtrOpUse.get());
834 }
835
836 // If the pointers in both addrspaces have a bitwise representation and if the
837 // representation of the new pointer is smaller (fewer bits) than the old one,
838 // check if the mask is applicable to the ptr in the new addrspace. Any
839 // masking only clearing the low bits will also apply in the new addrspace
840 // Note: checking if the mask clears high bits is not sufficient as those
841 // might have already been 0 in the old ptr.
842 if (OldPtrBits.getBitWidth() > NewPtrBits.getBitWidth()) {
843 KnownBits MaskBits =
844 computeKnownBits(V: MaskOp, DL: *DL, /*AssumptionCache=*/AC: nullptr, CxtI: I);
845 // Set all unknown bits of the old ptr to 1, so that we are conservative in
846 // checking which bits are cleared by the mask.
847 OldPtrBits.One |= ~OldPtrBits.Zero;
848 // Check which bits are cleared by the mask in the old ptr.
849 KnownBits ClearedBits = KnownBits::sub(LHS: OldPtrBits, RHS: OldPtrBits & MaskBits);
850
851 // If the mask isn't applicable to the new ptr, leave the ptrmask as-is and
852 // insert an addrspacecast after it.
853 if (ClearedBits.countMaxActiveBits() > NewPtrBits.countMaxActiveBits()) {
854 std::optional<BasicBlock::iterator> InsertPoint =
855 I->getInsertionPointAfterDef();
856 assert(InsertPoint && "insertion after ptrmask should be possible");
857 Type *NewPtrType = getPtrOrVecOfPtrsWithNewAS(Ty: I->getType(), NewAddrSpace);
858 Instruction *AddrSpaceCast =
859 new AddrSpaceCastInst(I, NewPtrType, "", *InsertPoint);
860 AddrSpaceCast->setDebugLoc(I->getDebugLoc());
861 return AddrSpaceCast;
862 }
863 }
864
865 IRBuilder<> B(I);
866 if (NewPtrBits.getBitWidth() < MaskTy->getScalarSizeInBits()) {
867 MaskTy = MaskTy->getWithNewBitWidth(NewBitWidth: NewPtrBits.getBitWidth());
868 MaskOp = B.CreateTrunc(V: MaskOp, DestTy: MaskTy);
869 }
870 Value *NewPtr = operandWithNewAddressSpaceOrCreatePoison(
871 OperandUse: PtrOpUse, NewAddrSpace, ValueWithNewAddrSpace, PredicatedAS,
872 PoisonUsesToFix);
873 return B.CreateIntrinsic(ID: Intrinsic::ptrmask, OverloadTypes: {NewPtr->getType(), MaskTy},
874 Args: {NewPtr, MaskOp});
875}
876
877// Returns a clone of `I` with its operands converted to those specified in
878// ValueWithNewAddrSpace. Due to potential cycles in the data flow graph, an
879// operand whose address space needs to be modified might not exist in
880// ValueWithNewAddrSpace. In that case, uses poison as a placeholder operand and
881// adds that operand use to PoisonUsesToFix so that caller can fix them later.
882//
883// Note that we do not necessarily clone `I`, e.g., if it is an addrspacecast
884// from a pointer whose type already matches. Therefore, this function returns a
885// Value* instead of an Instruction*.
886Value *InferAddressSpacesImpl::cloneInstructionWithNewAddressSpace(
887 Instruction *I, unsigned NewAddrSpace,
888 const ValueToValueMapTy &ValueWithNewAddrSpace,
889 const PredicatedAddrSpaceMapTy &PredicatedAS,
890 SmallVectorImpl<const Use *> *PoisonUsesToFix) const {
891 Type *NewPtrType = getPtrOrVecOfPtrsWithNewAS(Ty: I->getType(), NewAddrSpace);
892
893 if (I->getOpcode() == Instruction::AddrSpaceCast) {
894 Value *Src = I->getOperand(i: 0);
895 // Because `I` is flat, the source address space must be specific.
896 // Therefore, the inferred address space must be the source space, according
897 // to our algorithm.
898 assert(Src->getType()->getPointerAddressSpace() == NewAddrSpace);
899 return Src;
900 }
901
902 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Val: I)) {
903 // Technically the intrinsic ID is a pointer typed argument, so specially
904 // handle calls early.
905 assert(II->getIntrinsicID() == Intrinsic::ptrmask);
906 return clonePtrMaskWithNewAddressSpace(
907 I: II, NewAddrSpace, ValueWithNewAddrSpace, PredicatedAS, PoisonUsesToFix);
908 }
909
910 unsigned AS = TTI->getAssumedAddrSpace(V: I);
911 if (AS != UninitializedAddressSpace) {
912 // For the assumed address space, insert an `addrspacecast` to make that
913 // explicit.
914 Type *NewPtrTy = getPtrOrVecOfPtrsWithNewAS(Ty: I->getType(), NewAddrSpace: AS);
915 auto *NewI = new AddrSpaceCastInst(I, NewPtrTy);
916 NewI->insertAfter(InsertPos: I->getIterator());
917 NewI->setDebugLoc(I->getDebugLoc());
918 return NewI;
919 }
920
921 // Computes the converted pointer operands.
922 SmallVector<Value *, 4> NewPointerOperands;
923 for (const Use &OperandUse : I->operands()) {
924 if (!OperandUse.get()->getType()->isPtrOrPtrVectorTy())
925 NewPointerOperands.push_back(Elt: nullptr);
926 else
927 NewPointerOperands.push_back(Elt: operandWithNewAddressSpaceOrCreatePoison(
928 OperandUse, NewAddrSpace, ValueWithNewAddrSpace, PredicatedAS,
929 PoisonUsesToFix));
930 }
931
932 switch (I->getOpcode()) {
933 case Instruction::BitCast:
934 return new BitCastInst(NewPointerOperands[0], NewPtrType);
935 case Instruction::PHI: {
936 assert(I->getType()->isPtrOrPtrVectorTy());
937 PHINode *PHI = cast<PHINode>(Val: I);
938 PHINode *NewPHI = PHINode::Create(Ty: NewPtrType, NumReservedValues: PHI->getNumIncomingValues());
939 for (unsigned Index = 0; Index < PHI->getNumIncomingValues(); ++Index) {
940 unsigned OperandNo = PHINode::getOperandNumForIncomingValue(i: Index);
941 NewPHI->addIncoming(V: NewPointerOperands[OperandNo],
942 BB: PHI->getIncomingBlock(i: Index));
943 }
944 return NewPHI;
945 }
946 case Instruction::GetElementPtr: {
947 GetElementPtrInst *GEP = cast<GetElementPtrInst>(Val: I);
948 GetElementPtrInst *NewGEP = GetElementPtrInst::Create(
949 PointeeType: GEP->getSourceElementType(), Ptr: NewPointerOperands[0],
950 IdxList: SmallVector<Value *, 4>(GEP->indices()));
951 NewGEP->setIsInBounds(GEP->isInBounds());
952 return NewGEP;
953 }
954 case Instruction::Select:
955 assert(I->getType()->isPtrOrPtrVectorTy());
956 return SelectInst::Create(C: I->getOperand(i: 0), S1: NewPointerOperands[1],
957 S2: NewPointerOperands[2], NameStr: "", InsertBefore: nullptr, MDFrom: I);
958 case Instruction::IntToPtr: {
959 if (isNoopPtrIntCastPair(I2P: cast<Operator>(Val: I), DL: *DL, TTI)) {
960 Value *Src = cast<Operator>(Val: I->getOperand(i: 0))->getOperand(i: 0);
961 if (Src->getType() == NewPtrType)
962 return Src;
963
964 // If we had a no-op inttoptr/ptrtoint pair, we may still have inferred a
965 // source address space from a generic pointer source need to insert a
966 // cast back.
967 return new AddrSpaceCastInst(Src, NewPtrType);
968 }
969 assert(isSafeToCastIntToPtrAddrSpace(cast<Operator>(I)));
970 AddrSpaceCastInst *AsCast = new AddrSpaceCastInst(I, NewPtrType);
971 AsCast->insertAfter(InsertPos: I);
972 return AsCast;
973 }
974 default:
975 llvm_unreachable("Unexpected opcode");
976 }
977}
978
979// Similar to cloneInstructionWithNewAddressSpace, returns a clone of the
980// constant expression `CE` with its operands replaced as specified in
981// ValueWithNewAddrSpace.
982Value *InferAddressSpacesImpl::cloneConstantExprWithNewAddressSpace(
983 ConstantExpr *CE, unsigned NewAddrSpace,
984 const ValueToValueMapTy &ValueWithNewAddrSpace, const DataLayout *DL,
985 const TargetTransformInfo *TTI) const {
986 Type *TargetType =
987 CE->getType()->isPtrOrPtrVectorTy()
988 ? getPtrOrVecOfPtrsWithNewAS(Ty: CE->getType(), NewAddrSpace)
989 : CE->getType();
990
991 if (CE->getOpcode() == Instruction::AddrSpaceCast) {
992 // Because CE is flat, the source address space must be specific.
993 // Therefore, the inferred address space must be the source space according
994 // to our algorithm.
995 assert(CE->getOperand(0)->getType()->getPointerAddressSpace() ==
996 NewAddrSpace);
997 return CE->getOperand(i_nocapture: 0);
998 }
999
1000 if (CE->getOpcode() == Instruction::BitCast) {
1001 if (Value *NewOperand = ValueWithNewAddrSpace.lookup(Val: CE->getOperand(i_nocapture: 0)))
1002 return ConstantExpr::getBitCast(C: cast<Constant>(Val: NewOperand), Ty: TargetType);
1003 return ConstantExpr::getAddrSpaceCast(C: CE, Ty: TargetType);
1004 }
1005
1006 if (CE->getOpcode() == Instruction::IntToPtr) {
1007 if (isNoopPtrIntCastPair(I2P: cast<Operator>(Val: CE), DL: *DL, TTI)) {
1008 Constant *Src = cast<ConstantExpr>(Val: CE->getOperand(i_nocapture: 0))->getOperand(i_nocapture: 0);
1009 assert(Src->getType()->getPointerAddressSpace() == NewAddrSpace);
1010 return Src;
1011 }
1012 assert(isSafeToCastIntToPtrAddrSpace(cast<Operator>(CE)));
1013 return ConstantExpr::getAddrSpaceCast(C: CE, Ty: TargetType);
1014 }
1015
1016 // Computes the operands of the new constant expression.
1017 bool IsNew = false;
1018 SmallVector<Constant *, 4> NewOperands;
1019 for (unsigned Index = 0; Index < CE->getNumOperands(); ++Index) {
1020 Constant *Operand = CE->getOperand(i_nocapture: Index);
1021 // If the address space of `Operand` needs to be modified, the new operand
1022 // with the new address space should already be in ValueWithNewAddrSpace
1023 // because (1) the constant expressions we consider (i.e. addrspacecast,
1024 // bitcast, and getelementptr) do not incur cycles in the data flow graph
1025 // and (2) this function is called on constant expressions in postorder.
1026 if (Value *NewOperand = ValueWithNewAddrSpace.lookup(Val: Operand)) {
1027 IsNew = true;
1028 NewOperands.push_back(Elt: cast<Constant>(Val: NewOperand));
1029 continue;
1030 }
1031 if (auto *CExpr = dyn_cast<ConstantExpr>(Val: Operand))
1032 if (Value *NewOperand = cloneConstantExprWithNewAddressSpace(
1033 CE: CExpr, NewAddrSpace, ValueWithNewAddrSpace, DL, TTI)) {
1034 IsNew = true;
1035 NewOperands.push_back(Elt: cast<Constant>(Val: NewOperand));
1036 continue;
1037 }
1038 // Otherwise, reuses the old operand.
1039 NewOperands.push_back(Elt: Operand);
1040 }
1041
1042 // If !IsNew, we will replace the Value with itself. However, replaced values
1043 // are assumed to wrapped in an addrspacecast cast later so drop it now.
1044 if (!IsNew)
1045 return nullptr;
1046
1047 if (CE->getOpcode() == Instruction::GetElementPtr) {
1048 // Needs to specify the source type while constructing a getelementptr
1049 // constant expression.
1050 return CE->getWithOperands(Ops: NewOperands, Ty: TargetType, /*OnlyIfReduced=*/false,
1051 SrcTy: cast<GEPOperator>(Val: CE)->getSourceElementType());
1052 }
1053
1054 return CE->getWithOperands(Ops: NewOperands, Ty: TargetType);
1055}
1056
1057// Returns a clone of the value `V`, with its operands replaced as specified in
1058// ValueWithNewAddrSpace. This function is called on every flat address
1059// expression whose address space needs to be modified, in postorder.
1060//
1061// See cloneInstructionWithNewAddressSpace for the meaning of PoisonUsesToFix.
1062Value *InferAddressSpacesImpl::cloneValueWithNewAddressSpace(
1063 Value *V, unsigned NewAddrSpace,
1064 const ValueToValueMapTy &ValueWithNewAddrSpace,
1065 const PredicatedAddrSpaceMapTy &PredicatedAS,
1066 SmallVectorImpl<const Use *> *PoisonUsesToFix) const {
1067 // All values in Postorder are flat address expressions.
1068 assert(V->getType()->getPointerAddressSpace() == FlatAddrSpace &&
1069 isAddressExpression(*V, *DL, TTI));
1070
1071 if (auto *Arg = dyn_cast<Argument>(Val: V)) {
1072 // Arguments are address space casted in the function body, as we do not
1073 // want to change the function signature.
1074 Function *F = Arg->getParent();
1075 BasicBlock::iterator Insert = F->getEntryBlock().getFirstNonPHIIt();
1076
1077 Type *NewPtrTy = PointerType::get(C&: Arg->getContext(), AddressSpace: NewAddrSpace);
1078 auto *NewI = new AddrSpaceCastInst(Arg, NewPtrTy);
1079 NewI->insertBefore(InsertPos: Insert);
1080 return NewI;
1081 }
1082
1083 if (Instruction *I = dyn_cast<Instruction>(Val: V)) {
1084 Value *NewV = cloneInstructionWithNewAddressSpace(
1085 I, NewAddrSpace, ValueWithNewAddrSpace, PredicatedAS, PoisonUsesToFix);
1086 if (Instruction *NewI = dyn_cast_or_null<Instruction>(Val: NewV)) {
1087 if (NewI->getParent() == nullptr) {
1088 NewI->insertBefore(InsertPos: I->getIterator());
1089 NewI->takeName(V: I);
1090 NewI->setDebugLoc(I->getDebugLoc());
1091 }
1092 }
1093 return NewV;
1094 }
1095
1096 return cloneConstantExprWithNewAddressSpace(
1097 CE: cast<ConstantExpr>(Val: V), NewAddrSpace, ValueWithNewAddrSpace, DL, TTI);
1098}
1099
1100// Defines the join operation on the address space lattice (see the file header
1101// comments).
1102unsigned InferAddressSpacesImpl::joinAddressSpaces(unsigned AS1,
1103 unsigned AS2) const {
1104 if (AS1 == FlatAddrSpace || AS2 == FlatAddrSpace)
1105 return FlatAddrSpace;
1106
1107 if (AS1 == UninitializedAddressSpace)
1108 return AS2;
1109 if (AS2 == UninitializedAddressSpace)
1110 return AS1;
1111
1112 // The join of two different specific address spaces is flat.
1113 return (AS1 == AS2) ? AS1 : FlatAddrSpace;
1114}
1115
1116bool InferAddressSpacesImpl::run(Function &CurFn) {
1117 F = &CurFn;
1118 DL = &F->getDataLayout();
1119 PtrIntCastPairs.clear();
1120
1121 if (AssumeDefaultIsFlatAddressSpace)
1122 FlatAddrSpace = 0;
1123
1124 if (FlatAddrSpace == UninitializedAddressSpace) {
1125 FlatAddrSpace = TTI->getFlatAddressSpace();
1126 if (FlatAddrSpace == UninitializedAddressSpace)
1127 return false;
1128 }
1129
1130 collectIntToPtrPointerOperand();
1131 // Collects all flat address expressions in postorder.
1132 std::vector<WeakTrackingVH> Postorder = collectFlatAddressExpressions(F&: *F);
1133
1134 // Runs a data-flow analysis to refine the address spaces of every expression
1135 // in Postorder.
1136 ValueToAddrSpaceMapTy InferredAddrSpace;
1137 PredicatedAddrSpaceMapTy PredicatedAS;
1138 inferAddressSpaces(Postorder, InferredAddrSpace, PredicatedAS);
1139
1140 // Changes the address spaces of the flat address expressions who are inferred
1141 // to point to a specific address space.
1142 return rewriteWithNewAddressSpaces(Postorder, InferredAddrSpace,
1143 PredicatedAS);
1144}
1145
1146void InferAddressSpacesImpl::enqueueUsers(
1147 Value &V, const ValueToAddrSpaceMapTy &InferredAddrSpace,
1148 SetVector<Value *> &Worklist) const {
1149 for (Value *User : V.users()) {
1150 // Skip if User is already in the worklist.
1151 if (Worklist.count(key: User))
1152 continue;
1153
1154 ValueToAddrSpaceMapTy::const_iterator Pos = InferredAddrSpace.find(Val: User);
1155 // Our algorithm only updates the address spaces of flat address
1156 // expressions, which are those in InferredAddrSpace.
1157 if (Pos == InferredAddrSpace.end())
1158 continue;
1159
1160 // Function updateAddressSpace moves the address space down a lattice path.
1161 // Therefore, nothing to do if User is already inferred as flat (the bottom
1162 // element in the lattice).
1163 if (Pos->second == FlatAddrSpace)
1164 continue;
1165
1166 Worklist.insert(X: User);
1167 }
1168}
1169
1170void InferAddressSpacesImpl::runToFixPoint(
1171 SetVector<Value *> &Worklist, ValueToAddrSpaceMapTy &InferredAddrSpace,
1172 PredicatedAddrSpaceMapTy &PredicatedAS) const {
1173 while (!Worklist.empty()) {
1174 Value *V = Worklist.pop_back_val();
1175
1176 // Try to update the address space of the stack top according to the
1177 // address spaces of its operands.
1178 if (!updateAddressSpace(V: *V, InferredAddrSpace, PredicatedAS))
1179 continue;
1180
1181 enqueueUsers(V&: *V, InferredAddrSpace, Worklist);
1182 }
1183}
1184
1185// Constants need to be tracked through RAUW to handle cases with nested
1186// constant expressions, so wrap values in WeakTrackingVH.
1187void InferAddressSpacesImpl::inferAddressSpaces(
1188 ArrayRef<WeakTrackingVH> Postorder,
1189 ValueToAddrSpaceMapTy &InferredAddrSpace,
1190 PredicatedAddrSpaceMapTy &PredicatedAS) const {
1191 SetVector<Value *> Worklist(llvm::from_range, Postorder);
1192 // Initially, all expressions are in the uninitialized address space.
1193 for (Value *V : Postorder)
1194 InferredAddrSpace[V] = UninitializedAddressSpace;
1195
1196 runToFixPoint(Worklist, InferredAddrSpace, PredicatedAS);
1197
1198 // A value still uninitialized here is stuck in a cycle of uninitialized
1199 // values and carries no address space information. Lower it to flat so its
1200 // users join to flat, instead of being rewritten to reference an operand
1201 // that rewriteWithNewAddressSpaces() never converts.
1202 SmallVector<Value *, 4> Lowered;
1203 for (Value *V : Postorder) {
1204 ValueToAddrSpaceMapTy::iterator I = InferredAddrSpace.find(Val: V);
1205 if (I->second == UninitializedAddressSpace) {
1206 I->second = FlatAddrSpace;
1207 Lowered.push_back(Elt: V);
1208 }
1209 }
1210
1211 for (Value *V : Lowered)
1212 enqueueUsers(V&: *V, InferredAddrSpace, Worklist);
1213
1214 runToFixPoint(Worklist, InferredAddrSpace, PredicatedAS);
1215}
1216
1217unsigned
1218InferAddressSpacesImpl::getPredicatedAddrSpace(const Value &Ptr,
1219 const Value *UserCtx) const {
1220 const Instruction *UserCtxI = dyn_cast<Instruction>(Val: UserCtx);
1221 if (!UserCtxI)
1222 return UninitializedAddressSpace;
1223
1224 const Value *StrippedPtr = Ptr.stripInBoundsOffsets();
1225 for (auto &AssumeVH : AC.assumptionsFor(V: StrippedPtr)) {
1226 if (!AssumeVH)
1227 continue;
1228 CallInst *CI = cast<CallInst>(Val&: AssumeVH);
1229 if (!isValidAssumeForContext(I: CI, CxtI: UserCtxI, DT))
1230 continue;
1231
1232 const Value *Ptr;
1233 unsigned AS;
1234 std::tie(args&: Ptr, args&: AS) = TTI->getPredicatedAddrSpace(V: CI->getArgOperand(i: 0));
1235 if (Ptr)
1236 return AS;
1237 }
1238
1239 return UninitializedAddressSpace;
1240}
1241
1242bool InferAddressSpacesImpl::updateAddressSpace(
1243 const Value &V, ValueToAddrSpaceMapTy &InferredAddrSpace,
1244 PredicatedAddrSpaceMapTy &PredicatedAS) const {
1245 assert(InferredAddrSpace.count(&V));
1246
1247 LLVM_DEBUG(dbgs() << "Updating the address space of\n " << V << '\n');
1248
1249 // The new inferred address space equals the join of the address spaces
1250 // of all its pointer operands.
1251 unsigned NewAS = UninitializedAddressSpace;
1252
1253 // isAddressExpression should guarantee that V is an operator or an argument.
1254 assert(isa<Operator>(V) || isa<Argument>(V));
1255
1256 unsigned AS = TTI->getAssumedAddrSpace(V: &V);
1257 if (AS != UninitializedAddressSpace) {
1258 // Use the assumed address space directly.
1259 NewAS = AS;
1260 } else {
1261 // Otherwise, infer the address space from its pointer operands.
1262 SmallVector<Constant *, 2> ConstantPtrOps;
1263 SmallVector<Value *, 2> PtrOps = getPointerOperands(V, DL: *DL, TTI);
1264 for (Value *PtrOperand : PtrOps) {
1265 auto I = InferredAddrSpace.find(Val: PtrOperand);
1266 unsigned OperandAS;
1267 if (I == InferredAddrSpace.end()) {
1268 OperandAS = PtrOperand->getType()->getPointerAddressSpace();
1269 if (auto *C = dyn_cast<Constant>(Val: PtrOperand);
1270 C && OperandAS == FlatAddrSpace) {
1271 // Defer joining the address space of constant pointer operands.
1272 ConstantPtrOps.push_back(Elt: C);
1273 continue;
1274 }
1275 if (OperandAS == FlatAddrSpace) {
1276 // Check AC for assumption dominating V.
1277 unsigned AS = getPredicatedAddrSpace(Ptr: *PtrOperand, UserCtx: &V);
1278 if (AS != UninitializedAddressSpace) {
1279 LLVM_DEBUG(dbgs()
1280 << " deduce operand AS from the predicate addrspace "
1281 << AS << '\n');
1282 OperandAS = AS;
1283 // Record this use with the predicated AS.
1284 PredicatedAS[std::make_pair(x: &V, y&: PtrOperand)] = OperandAS;
1285 }
1286 }
1287 } else
1288 OperandAS = I->second;
1289
1290 // join(flat, *) = flat. So we can break if NewAS is already flat.
1291 NewAS = joinAddressSpaces(AS1: NewAS, AS2: OperandAS);
1292 if (NewAS == FlatAddrSpace)
1293 break;
1294 }
1295
1296 if (NewAS != FlatAddrSpace && NewAS != UninitializedAddressSpace) {
1297 if (any_of(Range&: ConstantPtrOps, P: [=](Constant *C) {
1298 return !isSafeToCastConstAddrSpace(C, NewAS);
1299 }))
1300 NewAS = FlatAddrSpace;
1301 }
1302
1303 // operator(flat const, flat const, ...) -> flat
1304 if (NewAS == UninitializedAddressSpace &&
1305 PtrOps.size() == ConstantPtrOps.size())
1306 NewAS = FlatAddrSpace;
1307 }
1308
1309 unsigned OldAS = InferredAddrSpace.lookup(Val: &V);
1310 assert(OldAS != FlatAddrSpace);
1311 if (OldAS == NewAS)
1312 return false;
1313
1314 // If any updates are made, grabs its users to the worklist because
1315 // their address spaces can also be possibly updated.
1316 LLVM_DEBUG(dbgs() << " to " << NewAS << '\n');
1317 InferredAddrSpace[&V] = NewAS;
1318 return true;
1319}
1320
1321/// Replace operand \p OpIdx in \p Inst, if the value is the same as \p OldVal
1322/// with \p NewVal.
1323static bool replaceOperandIfSame(Instruction *Inst, unsigned OpIdx,
1324 Value *OldVal, Value *NewVal) {
1325 Use &U = Inst->getOperandUse(i: OpIdx);
1326 if (U.get() == OldVal) {
1327 U.set(NewVal);
1328 return true;
1329 }
1330
1331 return false;
1332}
1333
1334template <typename InstrType>
1335static bool replaceSimplePointerUse(const TargetTransformInfo &TTI,
1336 InstrType *MemInstr, unsigned AddrSpace,
1337 Value *OldV, Value *NewV) {
1338 if (!MemInstr->isVolatile() || TTI.hasVolatileVariant(I: MemInstr, AddrSpace)) {
1339 return replaceOperandIfSame(MemInstr, InstrType::getPointerOperandIndex(),
1340 OldV, NewV);
1341 }
1342
1343 return false;
1344}
1345
1346/// If \p OldV is used as the pointer operand of a compatible memory operation
1347/// \p Inst, replaces the pointer operand with NewV.
1348///
1349/// This covers memory instructions with a single pointer operand that can have
1350/// its address space changed by simply mutating the use to a new value.
1351///
1352/// \p returns true the user replacement was made.
1353static bool replaceIfSimplePointerUse(const TargetTransformInfo &TTI,
1354 User *Inst, unsigned AddrSpace,
1355 Value *OldV, Value *NewV) {
1356 if (auto *LI = dyn_cast<LoadInst>(Val: Inst))
1357 return replaceSimplePointerUse(TTI, MemInstr: LI, AddrSpace, OldV, NewV);
1358
1359 if (auto *SI = dyn_cast<StoreInst>(Val: Inst))
1360 return replaceSimplePointerUse(TTI, MemInstr: SI, AddrSpace, OldV, NewV);
1361
1362 if (auto *RMW = dyn_cast<AtomicRMWInst>(Val: Inst))
1363 return replaceSimplePointerUse(TTI, MemInstr: RMW, AddrSpace, OldV, NewV);
1364
1365 if (auto *CmpX = dyn_cast<AtomicCmpXchgInst>(Val: Inst))
1366 return replaceSimplePointerUse(TTI, MemInstr: CmpX, AddrSpace, OldV, NewV);
1367
1368 return false;
1369}
1370
1371/// Update memory intrinsic uses that require more complex processing than
1372/// simple memory instructions. These require re-mangling and may have multiple
1373/// pointer operands.
1374static bool handleMemIntrinsicPtrUse(MemIntrinsic *MI, Value *OldV,
1375 Value *NewV) {
1376 IRBuilder<> B(MI);
1377 if (auto *MSI = dyn_cast<MemSetInst>(Val: MI)) {
1378 B.CreateMemSet(Ptr: NewV, Val: MSI->getValue(), Size: MSI->getLength(), Align: MSI->getDestAlign(),
1379 isVolatile: false, // isVolatile
1380 AAInfo: MI->getAAMetadata());
1381 } else if (auto *MTI = dyn_cast<MemTransferInst>(Val: MI)) {
1382 Value *Src = MTI->getRawSource();
1383 Value *Dest = MTI->getRawDest();
1384
1385 // Be careful in case this is a self-to-self copy.
1386 if (Src == OldV)
1387 Src = NewV;
1388
1389 if (Dest == OldV)
1390 Dest = NewV;
1391
1392 if (auto *MCI = dyn_cast<MemCpyInst>(Val: MTI)) {
1393 if (MCI->isForceInlined())
1394 B.CreateMemCpyInline(Dst: Dest, DstAlign: MTI->getDestAlign(), Src,
1395 SrcAlign: MTI->getSourceAlign(), Size: MTI->getLength(),
1396 isVolatile: false, // isVolatile
1397 AAInfo: MI->getAAMetadata());
1398 else
1399 B.CreateMemCpy(Dst: Dest, DstAlign: MTI->getDestAlign(), Src, SrcAlign: MTI->getSourceAlign(),
1400 Size: MTI->getLength(),
1401 isVolatile: false, // isVolatile
1402 AAInfo: MI->getAAMetadata());
1403 } else {
1404 assert(isa<MemMoveInst>(MTI));
1405 B.CreateMemMove(Dst: Dest, DstAlign: MTI->getDestAlign(), Src, SrcAlign: MTI->getSourceAlign(),
1406 Size: MTI->getLength(),
1407 isVolatile: false, // isVolatile
1408 AAInfo: MI->getAAMetadata());
1409 }
1410 } else
1411 llvm_unreachable("unhandled MemIntrinsic");
1412
1413 MI->eraseFromParent();
1414 return true;
1415}
1416
1417// \p returns true if it is OK to change the address space of constant \p C with
1418// a ConstantExpr addrspacecast.
1419bool InferAddressSpacesImpl::isSafeToCastConstAddrSpace(Constant *C,
1420 unsigned NewAS) const {
1421 assert(NewAS != UninitializedAddressSpace);
1422
1423 unsigned SrcAS = C->getType()->getPointerAddressSpace();
1424 if (SrcAS == NewAS || isa<UndefValue>(Val: C))
1425 return true;
1426
1427 // Prevent illegal casts between different non-flat address spaces.
1428 if (SrcAS != FlatAddrSpace && NewAS != FlatAddrSpace)
1429 return false;
1430
1431 if (isa<ConstantPointerNull>(Val: C) || isa<ConstantAggregateZero>(Val: C))
1432 return true;
1433
1434 if (auto *Op = dyn_cast<Operator>(Val: C)) {
1435 // If we already have a constant addrspacecast, it should be safe to cast it
1436 // off.
1437 if (Op->getOpcode() == Instruction::AddrSpaceCast)
1438 return isSafeToCastConstAddrSpace(C: cast<Constant>(Val: Op->getOperand(i: 0)),
1439 NewAS);
1440
1441 if (Op->getOpcode() == Instruction::IntToPtr &&
1442 Op->getType()->getPointerAddressSpace() == FlatAddrSpace)
1443 return true;
1444 }
1445
1446 return false;
1447}
1448
1449static Value::use_iterator skipToNextUser(Value::use_iterator I,
1450 Value::use_iterator End) {
1451 User *CurUser = I->getUser();
1452 ++I;
1453
1454 while (I != End && I->getUser() == CurUser)
1455 ++I;
1456
1457 return I;
1458}
1459
1460void InferAddressSpacesImpl::performPointerReplacement(
1461 Value *V, Value *NewV, Use &U, ValueToValueMapTy &ValueWithNewAddrSpace,
1462 SmallVectorImpl<Instruction *> &DeadInstructions) const {
1463
1464 User *CurUser = U.getUser();
1465
1466 unsigned AddrSpace = V->getType()->getPointerAddressSpace();
1467 if (replaceIfSimplePointerUse(TTI: *TTI, Inst: CurUser, AddrSpace, OldV: V, NewV))
1468 return;
1469
1470 // Skip if the current user is the new value itself.
1471 if (CurUser == NewV)
1472 return;
1473
1474 auto *CurUserI = dyn_cast<Instruction>(Val: CurUser);
1475 if (!CurUserI || CurUserI->getFunction() != F)
1476 return;
1477
1478 // Handle more complex cases like intrinsic that need to be remangled.
1479 if (auto *MI = dyn_cast<MemIntrinsic>(Val: CurUser)) {
1480 if (!MI->isVolatile() && handleMemIntrinsicPtrUse(MI, OldV: V, NewV))
1481 return;
1482 }
1483
1484 if (auto *II = dyn_cast<IntrinsicInst>(Val: CurUser)) {
1485 if (rewriteIntrinsicOperands(II, OldV: V, NewV))
1486 return;
1487 }
1488
1489 if (ICmpInst *Cmp = dyn_cast<ICmpInst>(Val: CurUserI)) {
1490 // If we can infer that both pointers are in the same addrspace,
1491 // transform e.g.
1492 // %cmp = icmp eq float* %p, %q
1493 // into
1494 // %cmp = icmp eq float addrspace(3)* %new_p, %new_q
1495
1496 unsigned NewAS = NewV->getType()->getPointerAddressSpace();
1497 int SrcIdx = U.getOperandNo();
1498 int OtherIdx = (SrcIdx == 0) ? 1 : 0;
1499 Value *OtherSrc = Cmp->getOperand(i_nocapture: OtherIdx);
1500
1501 if (Value *OtherNewV = ValueWithNewAddrSpace.lookup(Val: OtherSrc)) {
1502 if (OtherNewV->getType()->getPointerAddressSpace() == NewAS) {
1503 Cmp->setOperand(i_nocapture: OtherIdx, Val_nocapture: OtherNewV);
1504 Cmp->setOperand(i_nocapture: SrcIdx, Val_nocapture: NewV);
1505 return;
1506 }
1507 }
1508
1509 // Even if the type mismatches, we can cast the constant.
1510 if (auto *KOtherSrc = dyn_cast<Constant>(Val: OtherSrc)) {
1511 if (isSafeToCastConstAddrSpace(C: KOtherSrc, NewAS)) {
1512 Cmp->setOperand(i_nocapture: SrcIdx, Val_nocapture: NewV);
1513 Cmp->setOperand(i_nocapture: OtherIdx, Val_nocapture: ConstantExpr::getAddrSpaceCast(
1514 C: KOtherSrc, Ty: NewV->getType()));
1515 return;
1516 }
1517 }
1518 }
1519
1520 if (AddrSpaceCastInst *ASC = dyn_cast<AddrSpaceCastInst>(Val: CurUserI)) {
1521 unsigned NewAS = NewV->getType()->getPointerAddressSpace();
1522 if (ASC->getDestAddressSpace() == NewAS) {
1523 ASC->replaceAllUsesWith(V: NewV);
1524 DeadInstructions.push_back(Elt: ASC);
1525 return;
1526 }
1527 }
1528
1529 // Otherwise, replaces the use with flat(NewV).
1530 if (isa<Instruction>(Val: V) || isa<Instruction>(Val: NewV)) {
1531 // Don't create a copy of the original addrspacecast.
1532 if (U == V && isa<AddrSpaceCastInst>(Val: V))
1533 return;
1534
1535 // Insert the addrspacecast after NewV.
1536 BasicBlock::iterator InsertPos;
1537 if (Instruction *NewVInst = dyn_cast<Instruction>(Val: NewV))
1538 InsertPos = std::next(x: NewVInst->getIterator());
1539 else
1540 InsertPos = std::next(x: cast<Instruction>(Val: V)->getIterator());
1541
1542 while (isa<PHINode>(Val: InsertPos))
1543 ++InsertPos;
1544 // This instruction may contain multiple uses of V, update them all.
1545 CurUser->replaceUsesOfWith(
1546 From: V, To: new AddrSpaceCastInst(NewV, V->getType(), "", InsertPos));
1547 } else {
1548 CurUserI->replaceUsesOfWith(
1549 From: V, To: ConstantExpr::getAddrSpaceCast(C: cast<Constant>(Val: NewV), Ty: V->getType()));
1550 }
1551}
1552
1553bool InferAddressSpacesImpl::rewriteWithNewAddressSpaces(
1554 ArrayRef<WeakTrackingVH> Postorder,
1555 const ValueToAddrSpaceMapTy &InferredAddrSpace,
1556 const PredicatedAddrSpaceMapTy &PredicatedAS) const {
1557 // For each address expression to be modified, creates a clone of it with its
1558 // pointer operands converted to the new address space. Since the pointer
1559 // operands are converted, the clone is naturally in the new address space by
1560 // construction.
1561 ValueToValueMapTy ValueWithNewAddrSpace;
1562 SmallVector<const Use *, 32> PoisonUsesToFix;
1563 for (Value *V : Postorder) {
1564 unsigned NewAddrSpace = InferredAddrSpace.lookup(Val: V);
1565
1566 // In some degenerate cases (e.g. invalid IR in unreachable code), we may
1567 // not even infer the value to have its original address space.
1568 if (NewAddrSpace == UninitializedAddressSpace)
1569 continue;
1570
1571 if (V->getType()->getPointerAddressSpace() != NewAddrSpace) {
1572 Value *New =
1573 cloneValueWithNewAddressSpace(V, NewAddrSpace, ValueWithNewAddrSpace,
1574 PredicatedAS, PoisonUsesToFix: &PoisonUsesToFix);
1575 if (New)
1576 ValueWithNewAddrSpace[V] = New;
1577 }
1578 }
1579
1580 if (ValueWithNewAddrSpace.empty())
1581 return false;
1582
1583 // Fixes all the poison uses generated by cloneInstructionWithNewAddressSpace.
1584 for (const Use *PoisonUse : PoisonUsesToFix) {
1585 User *V = PoisonUse->getUser();
1586 User *NewV = cast_or_null<User>(Val: ValueWithNewAddrSpace.lookup(Val: V));
1587 if (!NewV)
1588 continue;
1589
1590 unsigned OperandNo = PoisonUse->getOperandNo();
1591 assert(isa<PoisonValue>(NewV->getOperand(OperandNo)));
1592 WeakTrackingVH NewOp = ValueWithNewAddrSpace.lookup(Val: PoisonUse->get());
1593 assert(NewOp &&
1594 "poison replacements in ValueWithNewAddrSpace shouldn't be null");
1595 NewV->setOperand(i: OperandNo, Val: NewOp);
1596 }
1597
1598 SmallVector<Instruction *, 16> DeadInstructions;
1599 ValueToValueMapTy VMap;
1600 ValueMapper VMapper(VMap, RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
1601
1602 // Replaces the uses of the old address expressions with the new ones.
1603 for (const WeakTrackingVH &WVH : Postorder) {
1604 assert(WVH && "value was unexpectedly deleted");
1605 Value *V = WVH;
1606 Value *NewV = ValueWithNewAddrSpace.lookup(Val: V);
1607 if (NewV == nullptr)
1608 continue;
1609
1610 LLVM_DEBUG(dbgs() << "Replacing the uses of " << *V << "\n with\n "
1611 << *NewV << '\n');
1612
1613 if (Constant *C = dyn_cast<Constant>(Val: V)) {
1614 Constant *Replace =
1615 ConstantExpr::getAddrSpaceCast(C: cast<Constant>(Val: NewV), Ty: C->getType());
1616 if (C != Replace) {
1617 LLVM_DEBUG(dbgs() << "Inserting replacement const cast: " << Replace
1618 << ": " << *Replace << '\n');
1619 SmallVector<User *, 16> WorkList;
1620 for (User *U : make_early_inc_range(Range: C->users())) {
1621 if (auto *I = dyn_cast<Instruction>(Val: U)) {
1622 if (I->getFunction() == F)
1623 I->replaceUsesOfWith(From: C, To: Replace);
1624 } else {
1625 WorkList.append(in_start: U->user_begin(), in_end: U->user_end());
1626 }
1627 }
1628 if (!WorkList.empty()) {
1629 VMap[C] = Replace;
1630 DenseSet<User *> Visited{WorkList.begin(), WorkList.end()};
1631 while (!WorkList.empty()) {
1632 User *U = WorkList.pop_back_val();
1633 if (auto *I = dyn_cast<Instruction>(Val: U)) {
1634 if (I->getFunction() == F)
1635 VMapper.remapInstruction(I&: *I);
1636 continue;
1637 }
1638 for (User *U2 : U->users())
1639 if (Visited.insert(V: U2).second)
1640 WorkList.push_back(Elt: U2);
1641 }
1642 }
1643 V = Replace;
1644 }
1645 }
1646
1647 Value::use_iterator I, E, Next;
1648 for (I = V->use_begin(), E = V->use_end(); I != E;) {
1649 Use &U = *I;
1650
1651 // Some users may see the same pointer operand in multiple operands. Skip
1652 // to the next instruction.
1653 I = skipToNextUser(I, End: E);
1654
1655 performPointerReplacement(V, NewV, U, ValueWithNewAddrSpace,
1656 DeadInstructions);
1657 }
1658
1659 if (V->use_empty()) {
1660 if (Instruction *I = dyn_cast<Instruction>(Val: V))
1661 DeadInstructions.push_back(Elt: I);
1662 }
1663 }
1664
1665 // Deleting one instruction may recursively delete another queued
1666 // instruction. Create handles before the first deletion so overlapping
1667 // entries are nulled instead of leaving dangling pointers.
1668 auto DeadInstructionHandles =
1669 to_vector_of<WeakTrackingVH, 16>(Range&: DeadInstructions);
1670 RecursivelyDeleteTriviallyDeadInstructions(DeadInsts&: DeadInstructionHandles);
1671
1672 return true;
1673}
1674
1675bool InferAddressSpaces::runOnFunction(Function &F) {
1676 if (skipFunction(F))
1677 return false;
1678
1679 auto *DTWP = getAnalysisIfAvailable<DominatorTreeWrapperPass>();
1680 DominatorTree *DT = DTWP ? &DTWP->getDomTree() : nullptr;
1681 return InferAddressSpacesImpl(
1682 getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F), DT,
1683 &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F),
1684 FlatAddrSpace, /*AssumeDefaultIsFlatAddressSpace=*/false)
1685 .run(CurFn&: F);
1686}
1687
1688FunctionPass *llvm::createInferAddressSpacesPass(unsigned AddressSpace) {
1689 return new InferAddressSpaces(AddressSpace);
1690}
1691
1692InferAddressSpacesPass::InferAddressSpacesPass(
1693 bool AssumeDefaultIsFlatAddressSpace)
1694 : FlatAddrSpace(UninitializedAddressSpace),
1695 AssumeDefaultIsFlatAddressSpace(AssumeDefaultIsFlatAddressSpace) {}
1696InferAddressSpacesPass::InferAddressSpacesPass(
1697 unsigned AddressSpace, bool AssumeDefaultIsFlatAddressSpace)
1698 : FlatAddrSpace(AddressSpace),
1699 AssumeDefaultIsFlatAddressSpace(AssumeDefaultIsFlatAddressSpace) {}
1700
1701PreservedAnalyses InferAddressSpacesPass::run(Function &F,
1702 FunctionAnalysisManager &AM) {
1703 bool Changed =
1704 InferAddressSpacesImpl(AM.getResult<AssumptionAnalysis>(IR&: F),
1705 AM.getCachedResult<DominatorTreeAnalysis>(IR&: F),
1706 &AM.getResult<TargetIRAnalysis>(IR&: F), FlatAddrSpace,
1707 AssumeDefaultIsFlatAddressSpace)
1708 .run(CurFn&: F);
1709 if (Changed) {
1710 PreservedAnalyses PA;
1711 PA.preserveSet<CFGAnalyses>();
1712 return PA;
1713 }
1714 return PreservedAnalyses::all();
1715}
1716
1717void InferAddressSpacesPass::printPipeline(
1718 raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) {
1719 static_cast<PassInfoMixin<InferAddressSpacesPass> *>(this)->printPipeline(
1720 OS, MapClassName2PassName);
1721 if (AssumeDefaultIsFlatAddressSpace)
1722 OS << "<assume-default-is-flat-addrspace>";
1723}
1724