Bug Summary

File:lib/Transforms/Scalar/SROA.cpp
Warning:line 2144, column 3
Called C++ object pointer is null

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name SROA.cpp -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=cplusplus -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -analyzer-config-compatibility-mode=true -mrelocation-model pic -pic-level 2 -mthread-model posix -mframe-pointer=none -fmath-errno -masm-verbose -mconstructor-aliases -munwind-tables -fuse-init-array -target-cpu x86-64 -dwarf-column-info -debugger-tuning=gdb -ffunction-sections -fdata-sections -resource-dir /usr/lib/llvm-10/lib/clang/10.0.0 -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-10~svn374877/build-llvm/lib/Transforms/Scalar -I /build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar -I /build/llvm-toolchain-snapshot-10~svn374877/build-llvm/include -I /build/llvm-toolchain-snapshot-10~svn374877/include -U NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/x86_64-linux-gnu/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/x86_64-linux-gnu/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/c++/6.3.0/backward -internal-isystem /usr/local/include -internal-isystem /usr/lib/llvm-10/lib/clang/10.0.0/include -internal-externc-isystem /usr/include/x86_64-linux-gnu -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -Wno-unused-parameter -Wwrite-strings -Wno-missing-field-initializers -Wno-long-long -Wno-maybe-uninitialized -Wno-comment -std=c++14 -fdeprecated-macro -fdebug-compilation-dir /build/llvm-toolchain-snapshot-10~svn374877/build-llvm/lib/Transforms/Scalar -fdebug-prefix-map=/build/llvm-toolchain-snapshot-10~svn374877=. -ferror-limit 19 -fmessage-length 0 -fvisibility-inlines-hidden -stack-protector 2 -fgnuc-version=4.2.1 -fobjc-runtime=gcc -fdiagnostics-show-option -vectorize-loops -vectorize-slp -analyzer-output=html -analyzer-config stable-report-filename=true -faddrsig -o /tmp/scan-build-2019-10-15-233810-7101-1 -x c++ /build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp

/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp

1//===- SROA.cpp - Scalar Replacement Of Aggregates ------------------------===//
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/// \file
9/// This transformation implements the well known scalar replacement of
10/// aggregates transformation. It tries to identify promotable elements of an
11/// aggregate alloca, and promote them to registers. It will also try to
12/// convert uses of an element (or set of elements) of an alloca into a vector
13/// or bitfield-style integer scalar if appropriate.
14///
15/// It works to do this with minimal slicing of the alloca so that regions
16/// which are merely transferred in and out of external memory remain unchanged
17/// and are not decomposed to scalar code.
18///
19/// Because this also performs alloca promotion, it can be thought of as also
20/// serving the purpose of SSA formation. The algorithm iterates on the
21/// function until all opportunities for promotion have been realized.
22///
23//===----------------------------------------------------------------------===//
24
25#include "llvm/Transforms/Scalar/SROA.h"
26#include "llvm/ADT/APInt.h"
27#include "llvm/ADT/ArrayRef.h"
28#include "llvm/ADT/DenseMap.h"
29#include "llvm/ADT/PointerIntPair.h"
30#include "llvm/ADT/STLExtras.h"
31#include "llvm/ADT/SetVector.h"
32#include "llvm/ADT/SmallBitVector.h"
33#include "llvm/ADT/SmallPtrSet.h"
34#include "llvm/ADT/SmallVector.h"
35#include "llvm/ADT/Statistic.h"
36#include "llvm/ADT/StringRef.h"
37#include "llvm/ADT/Twine.h"
38#include "llvm/ADT/iterator.h"
39#include "llvm/ADT/iterator_range.h"
40#include "llvm/Analysis/AssumptionCache.h"
41#include "llvm/Analysis/GlobalsModRef.h"
42#include "llvm/Analysis/Loads.h"
43#include "llvm/Analysis/PtrUseVisitor.h"
44#include "llvm/Transforms/Utils/Local.h"
45#include "llvm/Config/llvm-config.h"
46#include "llvm/IR/BasicBlock.h"
47#include "llvm/IR/Constant.h"
48#include "llvm/IR/ConstantFolder.h"
49#include "llvm/IR/Constants.h"
50#include "llvm/IR/DIBuilder.h"
51#include "llvm/IR/DataLayout.h"
52#include "llvm/IR/DebugInfoMetadata.h"
53#include "llvm/IR/DerivedTypes.h"
54#include "llvm/IR/Dominators.h"
55#include "llvm/IR/Function.h"
56#include "llvm/IR/GetElementPtrTypeIterator.h"
57#include "llvm/IR/GlobalAlias.h"
58#include "llvm/IR/IRBuilder.h"
59#include "llvm/IR/InstVisitor.h"
60#include "llvm/IR/InstrTypes.h"
61#include "llvm/IR/Instruction.h"
62#include "llvm/IR/Instructions.h"
63#include "llvm/IR/IntrinsicInst.h"
64#include "llvm/IR/Intrinsics.h"
65#include "llvm/IR/LLVMContext.h"
66#include "llvm/IR/Metadata.h"
67#include "llvm/IR/Module.h"
68#include "llvm/IR/Operator.h"
69#include "llvm/IR/PassManager.h"
70#include "llvm/IR/Type.h"
71#include "llvm/IR/Use.h"
72#include "llvm/IR/User.h"
73#include "llvm/IR/Value.h"
74#include "llvm/Pass.h"
75#include "llvm/Support/Casting.h"
76#include "llvm/Support/CommandLine.h"
77#include "llvm/Support/Compiler.h"
78#include "llvm/Support/Debug.h"
79#include "llvm/Support/ErrorHandling.h"
80#include "llvm/Support/MathExtras.h"
81#include "llvm/Support/raw_ostream.h"
82#include "llvm/Transforms/Scalar.h"
83#include "llvm/Transforms/Utils/PromoteMemToReg.h"
84#include <algorithm>
85#include <cassert>
86#include <chrono>
87#include <cstddef>
88#include <cstdint>
89#include <cstring>
90#include <iterator>
91#include <string>
92#include <tuple>
93#include <utility>
94#include <vector>
95
96#ifndef NDEBUG
97// We only use this for a debug check.
98#include <random>
99#endif
100
101using namespace llvm;
102using namespace llvm::sroa;
103
104#define DEBUG_TYPE"sroa" "sroa"
105
106STATISTIC(NumAllocasAnalyzed, "Number of allocas analyzed for replacement")static llvm::Statistic NumAllocasAnalyzed = {"sroa", "NumAllocasAnalyzed"
, "Number of allocas analyzed for replacement"}
;
107STATISTIC(NumAllocaPartitions, "Number of alloca partitions formed")static llvm::Statistic NumAllocaPartitions = {"sroa", "NumAllocaPartitions"
, "Number of alloca partitions formed"}
;
108STATISTIC(MaxPartitionsPerAlloca, "Maximum number of partitions per alloca")static llvm::Statistic MaxPartitionsPerAlloca = {"sroa", "MaxPartitionsPerAlloca"
, "Maximum number of partitions per alloca"}
;
109STATISTIC(NumAllocaPartitionUses, "Number of alloca partition uses rewritten")static llvm::Statistic NumAllocaPartitionUses = {"sroa", "NumAllocaPartitionUses"
, "Number of alloca partition uses rewritten"}
;
110STATISTIC(MaxUsesPerAllocaPartition, "Maximum number of uses of a partition")static llvm::Statistic MaxUsesPerAllocaPartition = {"sroa", "MaxUsesPerAllocaPartition"
, "Maximum number of uses of a partition"}
;
111STATISTIC(NumNewAllocas, "Number of new, smaller allocas introduced")static llvm::Statistic NumNewAllocas = {"sroa", "NumNewAllocas"
, "Number of new, smaller allocas introduced"}
;
112STATISTIC(NumPromoted, "Number of allocas promoted to SSA values")static llvm::Statistic NumPromoted = {"sroa", "NumPromoted", "Number of allocas promoted to SSA values"
}
;
113STATISTIC(NumLoadsSpeculated, "Number of loads speculated to allow promotion")static llvm::Statistic NumLoadsSpeculated = {"sroa", "NumLoadsSpeculated"
, "Number of loads speculated to allow promotion"}
;
114STATISTIC(NumDeleted, "Number of instructions deleted")static llvm::Statistic NumDeleted = {"sroa", "NumDeleted", "Number of instructions deleted"
}
;
115STATISTIC(NumVectorized, "Number of vectorized aggregates")static llvm::Statistic NumVectorized = {"sroa", "NumVectorized"
, "Number of vectorized aggregates"}
;
116
117/// Hidden option to enable randomly shuffling the slices to help uncover
118/// instability in their order.
119static cl::opt<bool> SROARandomShuffleSlices("sroa-random-shuffle-slices",
120 cl::init(false), cl::Hidden);
121
122/// Hidden option to experiment with completely strict handling of inbounds
123/// GEPs.
124static cl::opt<bool> SROAStrictInbounds("sroa-strict-inbounds", cl::init(false),
125 cl::Hidden);
126
127namespace {
128
129/// A custom IRBuilder inserter which prefixes all names, but only in
130/// Assert builds.
131class IRBuilderPrefixedInserter : public IRBuilderDefaultInserter {
132 std::string Prefix;
133
134 const Twine getNameWithPrefix(const Twine &Name) const {
135 return Name.isTriviallyEmpty() ? Name : Prefix + Name;
46
'?' condition is false
47
Calling 'operator+'
50
Returning from 'operator+'
136 }
137
138public:
139 void SetNamePrefix(const Twine &P) { Prefix = P.str(); }
140
141protected:
142 void InsertHelper(Instruction *I, const Twine &Name, BasicBlock *BB,
143 BasicBlock::iterator InsertPt) const {
144 IRBuilderDefaultInserter::InsertHelper(I, getNameWithPrefix(Name), BB,
45
Calling 'IRBuilderPrefixedInserter::getNameWithPrefix'
51
Returning from 'IRBuilderPrefixedInserter::getNameWithPrefix'
145 InsertPt);
146 }
147};
148
149/// Provide a type for IRBuilder that drops names in release builds.
150using IRBuilderTy = IRBuilder<ConstantFolder, IRBuilderPrefixedInserter>;
151
152/// A used slice of an alloca.
153///
154/// This structure represents a slice of an alloca used by some instruction. It
155/// stores both the begin and end offsets of this use, a pointer to the use
156/// itself, and a flag indicating whether we can classify the use as splittable
157/// or not when forming partitions of the alloca.
158class Slice {
159 /// The beginning offset of the range.
160 uint64_t BeginOffset = 0;
161
162 /// The ending offset, not included in the range.
163 uint64_t EndOffset = 0;
164
165 /// Storage for both the use of this slice and whether it can be
166 /// split.
167 PointerIntPair<Use *, 1, bool> UseAndIsSplittable;
168
169public:
170 Slice() = default;
171
172 Slice(uint64_t BeginOffset, uint64_t EndOffset, Use *U, bool IsSplittable)
173 : BeginOffset(BeginOffset), EndOffset(EndOffset),
174 UseAndIsSplittable(U, IsSplittable) {}
175
176 uint64_t beginOffset() const { return BeginOffset; }
177 uint64_t endOffset() const { return EndOffset; }
178
179 bool isSplittable() const { return UseAndIsSplittable.getInt(); }
180 void makeUnsplittable() { UseAndIsSplittable.setInt(false); }
181
182 Use *getUse() const { return UseAndIsSplittable.getPointer(); }
183
184 bool isDead() const { return getUse() == nullptr; }
185 void kill() { UseAndIsSplittable.setPointer(nullptr); }
186
187 /// Support for ordering ranges.
188 ///
189 /// This provides an ordering over ranges such that start offsets are
190 /// always increasing, and within equal start offsets, the end offsets are
191 /// decreasing. Thus the spanning range comes first in a cluster with the
192 /// same start position.
193 bool operator<(const Slice &RHS) const {
194 if (beginOffset() < RHS.beginOffset())
195 return true;
196 if (beginOffset() > RHS.beginOffset())
197 return false;
198 if (isSplittable() != RHS.isSplittable())
199 return !isSplittable();
200 if (endOffset() > RHS.endOffset())
201 return true;
202 return false;
203 }
204
205 /// Support comparison with a single offset to allow binary searches.
206 friend LLVM_ATTRIBUTE_UNUSED__attribute__((__unused__)) bool operator<(const Slice &LHS,
207 uint64_t RHSOffset) {
208 return LHS.beginOffset() < RHSOffset;
209 }
210 friend LLVM_ATTRIBUTE_UNUSED__attribute__((__unused__)) bool operator<(uint64_t LHSOffset,
211 const Slice &RHS) {
212 return LHSOffset < RHS.beginOffset();
213 }
214
215 bool operator==(const Slice &RHS) const {
216 return isSplittable() == RHS.isSplittable() &&
217 beginOffset() == RHS.beginOffset() && endOffset() == RHS.endOffset();
218 }
219 bool operator!=(const Slice &RHS) const { return !operator==(RHS); }
220};
221
222} // end anonymous namespace
223
224/// Representation of the alloca slices.
225///
226/// This class represents the slices of an alloca which are formed by its
227/// various uses. If a pointer escapes, we can't fully build a representation
228/// for the slices used and we reflect that in this structure. The uses are
229/// stored, sorted by increasing beginning offset and with unsplittable slices
230/// starting at a particular offset before splittable slices.
231class llvm::sroa::AllocaSlices {
232public:
233 /// Construct the slices of a particular alloca.
234 AllocaSlices(const DataLayout &DL, AllocaInst &AI);
235
236 /// Test whether a pointer to the allocation escapes our analysis.
237 ///
238 /// If this is true, the slices are never fully built and should be
239 /// ignored.
240 bool isEscaped() const { return PointerEscapingInstr; }
241
242 /// Support for iterating over the slices.
243 /// @{
244 using iterator = SmallVectorImpl<Slice>::iterator;
245 using range = iterator_range<iterator>;
246
247 iterator begin() { return Slices.begin(); }
248 iterator end() { return Slices.end(); }
249
250 using const_iterator = SmallVectorImpl<Slice>::const_iterator;
251 using const_range = iterator_range<const_iterator>;
252
253 const_iterator begin() const { return Slices.begin(); }
254 const_iterator end() const { return Slices.end(); }
255 /// @}
256
257 /// Erase a range of slices.
258 void erase(iterator Start, iterator Stop) { Slices.erase(Start, Stop); }
259
260 /// Insert new slices for this alloca.
261 ///
262 /// This moves the slices into the alloca's slices collection, and re-sorts
263 /// everything so that the usual ordering properties of the alloca's slices
264 /// hold.
265 void insert(ArrayRef<Slice> NewSlices) {
266 int OldSize = Slices.size();
267 Slices.append(NewSlices.begin(), NewSlices.end());
268 auto SliceI = Slices.begin() + OldSize;
269 llvm::sort(SliceI, Slices.end());
270 std::inplace_merge(Slices.begin(), SliceI, Slices.end());
271 }
272
273 // Forward declare the iterator and range accessor for walking the
274 // partitions.
275 class partition_iterator;
276 iterator_range<partition_iterator> partitions();
277
278 /// Access the dead users for this alloca.
279 ArrayRef<Instruction *> getDeadUsers() const { return DeadUsers; }
280
281 /// Access the dead operands referring to this alloca.
282 ///
283 /// These are operands which have cannot actually be used to refer to the
284 /// alloca as they are outside its range and the user doesn't correct for
285 /// that. These mostly consist of PHI node inputs and the like which we just
286 /// need to replace with undef.
287 ArrayRef<Use *> getDeadOperands() const { return DeadOperands; }
288
289#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
290 void print(raw_ostream &OS, const_iterator I, StringRef Indent = " ") const;
291 void printSlice(raw_ostream &OS, const_iterator I,
292 StringRef Indent = " ") const;
293 void printUse(raw_ostream &OS, const_iterator I,
294 StringRef Indent = " ") const;
295 void print(raw_ostream &OS) const;
296 void dump(const_iterator I) const;
297 void dump() const;
298#endif
299
300private:
301 template <typename DerivedT, typename RetT = void> class BuilderBase;
302 class SliceBuilder;
303
304 friend class AllocaSlices::SliceBuilder;
305
306#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
307 /// Handle to alloca instruction to simplify method interfaces.
308 AllocaInst &AI;
309#endif
310
311 /// The instruction responsible for this alloca not having a known set
312 /// of slices.
313 ///
314 /// When an instruction (potentially) escapes the pointer to the alloca, we
315 /// store a pointer to that here and abort trying to form slices of the
316 /// alloca. This will be null if the alloca slices are analyzed successfully.
317 Instruction *PointerEscapingInstr;
318
319 /// The slices of the alloca.
320 ///
321 /// We store a vector of the slices formed by uses of the alloca here. This
322 /// vector is sorted by increasing begin offset, and then the unsplittable
323 /// slices before the splittable ones. See the Slice inner class for more
324 /// details.
325 SmallVector<Slice, 8> Slices;
326
327 /// Instructions which will become dead if we rewrite the alloca.
328 ///
329 /// Note that these are not separated by slice. This is because we expect an
330 /// alloca to be completely rewritten or not rewritten at all. If rewritten,
331 /// all these instructions can simply be removed and replaced with undef as
332 /// they come from outside of the allocated space.
333 SmallVector<Instruction *, 8> DeadUsers;
334
335 /// Operands which will become dead if we rewrite the alloca.
336 ///
337 /// These are operands that in their particular use can be replaced with
338 /// undef when we rewrite the alloca. These show up in out-of-bounds inputs
339 /// to PHI nodes and the like. They aren't entirely dead (there might be
340 /// a GEP back into the bounds using it elsewhere) and nor is the PHI, but we
341 /// want to swap this particular input for undef to simplify the use lists of
342 /// the alloca.
343 SmallVector<Use *, 8> DeadOperands;
344};
345
346/// A partition of the slices.
347///
348/// An ephemeral representation for a range of slices which can be viewed as
349/// a partition of the alloca. This range represents a span of the alloca's
350/// memory which cannot be split, and provides access to all of the slices
351/// overlapping some part of the partition.
352///
353/// Objects of this type are produced by traversing the alloca's slices, but
354/// are only ephemeral and not persistent.
355class llvm::sroa::Partition {
356private:
357 friend class AllocaSlices;
358 friend class AllocaSlices::partition_iterator;
359
360 using iterator = AllocaSlices::iterator;
361
362 /// The beginning and ending offsets of the alloca for this
363 /// partition.
364 uint64_t BeginOffset, EndOffset;
365
366 /// The start and end iterators of this partition.
367 iterator SI, SJ;
368
369 /// A collection of split slice tails overlapping the partition.
370 SmallVector<Slice *, 4> SplitTails;
371
372 /// Raw constructor builds an empty partition starting and ending at
373 /// the given iterator.
374 Partition(iterator SI) : SI(SI), SJ(SI) {}
375
376public:
377 /// The start offset of this partition.
378 ///
379 /// All of the contained slices start at or after this offset.
380 uint64_t beginOffset() const { return BeginOffset; }
381
382 /// The end offset of this partition.
383 ///
384 /// All of the contained slices end at or before this offset.
385 uint64_t endOffset() const { return EndOffset; }
386
387 /// The size of the partition.
388 ///
389 /// Note that this can never be zero.
390 uint64_t size() const {
391 assert(BeginOffset < EndOffset && "Partitions must span some bytes!")((BeginOffset < EndOffset && "Partitions must span some bytes!"
) ? static_cast<void> (0) : __assert_fail ("BeginOffset < EndOffset && \"Partitions must span some bytes!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 391, __PRETTY_FUNCTION__))
;
392 return EndOffset - BeginOffset;
393 }
394
395 /// Test whether this partition contains no slices, and merely spans
396 /// a region occupied by split slices.
397 bool empty() const { return SI == SJ; }
398
399 /// \name Iterate slices that start within the partition.
400 /// These may be splittable or unsplittable. They have a begin offset >= the
401 /// partition begin offset.
402 /// @{
403 // FIXME: We should probably define a "concat_iterator" helper and use that
404 // to stitch together pointee_iterators over the split tails and the
405 // contiguous iterators of the partition. That would give a much nicer
406 // interface here. We could then additionally expose filtered iterators for
407 // split, unsplit, and unsplittable splices based on the usage patterns.
408 iterator begin() const { return SI; }
409 iterator end() const { return SJ; }
410 /// @}
411
412 /// Get the sequence of split slice tails.
413 ///
414 /// These tails are of slices which start before this partition but are
415 /// split and overlap into the partition. We accumulate these while forming
416 /// partitions.
417 ArrayRef<Slice *> splitSliceTails() const { return SplitTails; }
418};
419
420/// An iterator over partitions of the alloca's slices.
421///
422/// This iterator implements the core algorithm for partitioning the alloca's
423/// slices. It is a forward iterator as we don't support backtracking for
424/// efficiency reasons, and re-use a single storage area to maintain the
425/// current set of split slices.
426///
427/// It is templated on the slice iterator type to use so that it can operate
428/// with either const or non-const slice iterators.
429class AllocaSlices::partition_iterator
430 : public iterator_facade_base<partition_iterator, std::forward_iterator_tag,
431 Partition> {
432 friend class AllocaSlices;
433
434 /// Most of the state for walking the partitions is held in a class
435 /// with a nice interface for examining them.
436 Partition P;
437
438 /// We need to keep the end of the slices to know when to stop.
439 AllocaSlices::iterator SE;
440
441 /// We also need to keep track of the maximum split end offset seen.
442 /// FIXME: Do we really?
443 uint64_t MaxSplitSliceEndOffset = 0;
444
445 /// Sets the partition to be empty at given iterator, and sets the
446 /// end iterator.
447 partition_iterator(AllocaSlices::iterator SI, AllocaSlices::iterator SE)
448 : P(SI), SE(SE) {
449 // If not already at the end, advance our state to form the initial
450 // partition.
451 if (SI != SE)
452 advance();
453 }
454
455 /// Advance the iterator to the next partition.
456 ///
457 /// Requires that the iterator not be at the end of the slices.
458 void advance() {
459 assert((P.SI != SE || !P.SplitTails.empty()) &&(((P.SI != SE || !P.SplitTails.empty()) && "Cannot advance past the end of the slices!"
) ? static_cast<void> (0) : __assert_fail ("(P.SI != SE || !P.SplitTails.empty()) && \"Cannot advance past the end of the slices!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 460, __PRETTY_FUNCTION__))
460 "Cannot advance past the end of the slices!")(((P.SI != SE || !P.SplitTails.empty()) && "Cannot advance past the end of the slices!"
) ? static_cast<void> (0) : __assert_fail ("(P.SI != SE || !P.SplitTails.empty()) && \"Cannot advance past the end of the slices!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 460, __PRETTY_FUNCTION__))
;
461
462 // Clear out any split uses which have ended.
463 if (!P.SplitTails.empty()) {
464 if (P.EndOffset >= MaxSplitSliceEndOffset) {
465 // If we've finished all splits, this is easy.
466 P.SplitTails.clear();
467 MaxSplitSliceEndOffset = 0;
468 } else {
469 // Remove the uses which have ended in the prior partition. This
470 // cannot change the max split slice end because we just checked that
471 // the prior partition ended prior to that max.
472 P.SplitTails.erase(llvm::remove_if(P.SplitTails,
473 [&](Slice *S) {
474 return S->endOffset() <=
475 P.EndOffset;
476 }),
477 P.SplitTails.end());
478 assert(llvm::any_of(P.SplitTails,((llvm::any_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() == MaxSplitSliceEndOffset; }) && "Could not find the current max split slice offset!"
) ? static_cast<void> (0) : __assert_fail ("llvm::any_of(P.SplitTails, [&](Slice *S) { return S->endOffset() == MaxSplitSliceEndOffset; }) && \"Could not find the current max split slice offset!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 482, __PRETTY_FUNCTION__))
479 [&](Slice *S) {((llvm::any_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() == MaxSplitSliceEndOffset; }) && "Could not find the current max split slice offset!"
) ? static_cast<void> (0) : __assert_fail ("llvm::any_of(P.SplitTails, [&](Slice *S) { return S->endOffset() == MaxSplitSliceEndOffset; }) && \"Could not find the current max split slice offset!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 482, __PRETTY_FUNCTION__))
480 return S->endOffset() == MaxSplitSliceEndOffset;((llvm::any_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() == MaxSplitSliceEndOffset; }) && "Could not find the current max split slice offset!"
) ? static_cast<void> (0) : __assert_fail ("llvm::any_of(P.SplitTails, [&](Slice *S) { return S->endOffset() == MaxSplitSliceEndOffset; }) && \"Could not find the current max split slice offset!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 482, __PRETTY_FUNCTION__))
481 }) &&((llvm::any_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() == MaxSplitSliceEndOffset; }) && "Could not find the current max split slice offset!"
) ? static_cast<void> (0) : __assert_fail ("llvm::any_of(P.SplitTails, [&](Slice *S) { return S->endOffset() == MaxSplitSliceEndOffset; }) && \"Could not find the current max split slice offset!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 482, __PRETTY_FUNCTION__))
482 "Could not find the current max split slice offset!")((llvm::any_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() == MaxSplitSliceEndOffset; }) && "Could not find the current max split slice offset!"
) ? static_cast<void> (0) : __assert_fail ("llvm::any_of(P.SplitTails, [&](Slice *S) { return S->endOffset() == MaxSplitSliceEndOffset; }) && \"Could not find the current max split slice offset!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 482, __PRETTY_FUNCTION__))
;
483 assert(llvm::all_of(P.SplitTails,((llvm::all_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() <= MaxSplitSliceEndOffset; }) && "Max split slice end offset is not actually the max!"
) ? static_cast<void> (0) : __assert_fail ("llvm::all_of(P.SplitTails, [&](Slice *S) { return S->endOffset() <= MaxSplitSliceEndOffset; }) && \"Max split slice end offset is not actually the max!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 487, __PRETTY_FUNCTION__))
484 [&](Slice *S) {((llvm::all_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() <= MaxSplitSliceEndOffset; }) && "Max split slice end offset is not actually the max!"
) ? static_cast<void> (0) : __assert_fail ("llvm::all_of(P.SplitTails, [&](Slice *S) { return S->endOffset() <= MaxSplitSliceEndOffset; }) && \"Max split slice end offset is not actually the max!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 487, __PRETTY_FUNCTION__))
485 return S->endOffset() <= MaxSplitSliceEndOffset;((llvm::all_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() <= MaxSplitSliceEndOffset; }) && "Max split slice end offset is not actually the max!"
) ? static_cast<void> (0) : __assert_fail ("llvm::all_of(P.SplitTails, [&](Slice *S) { return S->endOffset() <= MaxSplitSliceEndOffset; }) && \"Max split slice end offset is not actually the max!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 487, __PRETTY_FUNCTION__))
486 }) &&((llvm::all_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() <= MaxSplitSliceEndOffset; }) && "Max split slice end offset is not actually the max!"
) ? static_cast<void> (0) : __assert_fail ("llvm::all_of(P.SplitTails, [&](Slice *S) { return S->endOffset() <= MaxSplitSliceEndOffset; }) && \"Max split slice end offset is not actually the max!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 487, __PRETTY_FUNCTION__))
487 "Max split slice end offset is not actually the max!")((llvm::all_of(P.SplitTails, [&](Slice *S) { return S->
endOffset() <= MaxSplitSliceEndOffset; }) && "Max split slice end offset is not actually the max!"
) ? static_cast<void> (0) : __assert_fail ("llvm::all_of(P.SplitTails, [&](Slice *S) { return S->endOffset() <= MaxSplitSliceEndOffset; }) && \"Max split slice end offset is not actually the max!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 487, __PRETTY_FUNCTION__))
;
488 }
489 }
490
491 // If P.SI is already at the end, then we've cleared the split tail and
492 // now have an end iterator.
493 if (P.SI == SE) {
494 assert(P.SplitTails.empty() && "Failed to clear the split slices!")((P.SplitTails.empty() && "Failed to clear the split slices!"
) ? static_cast<void> (0) : __assert_fail ("P.SplitTails.empty() && \"Failed to clear the split slices!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 494, __PRETTY_FUNCTION__))
;
495 return;
496 }
497
498 // If we had a non-empty partition previously, set up the state for
499 // subsequent partitions.
500 if (P.SI != P.SJ) {
501 // Accumulate all the splittable slices which started in the old
502 // partition into the split list.
503 for (Slice &S : P)
504 if (S.isSplittable() && S.endOffset() > P.EndOffset) {
505 P.SplitTails.push_back(&S);
506 MaxSplitSliceEndOffset =
507 std::max(S.endOffset(), MaxSplitSliceEndOffset);
508 }
509
510 // Start from the end of the previous partition.
511 P.SI = P.SJ;
512
513 // If P.SI is now at the end, we at most have a tail of split slices.
514 if (P.SI == SE) {
515 P.BeginOffset = P.EndOffset;
516 P.EndOffset = MaxSplitSliceEndOffset;
517 return;
518 }
519
520 // If the we have split slices and the next slice is after a gap and is
521 // not splittable immediately form an empty partition for the split
522 // slices up until the next slice begins.
523 if (!P.SplitTails.empty() && P.SI->beginOffset() != P.EndOffset &&
524 !P.SI->isSplittable()) {
525 P.BeginOffset = P.EndOffset;
526 P.EndOffset = P.SI->beginOffset();
527 return;
528 }
529 }
530
531 // OK, we need to consume new slices. Set the end offset based on the
532 // current slice, and step SJ past it. The beginning offset of the
533 // partition is the beginning offset of the next slice unless we have
534 // pre-existing split slices that are continuing, in which case we begin
535 // at the prior end offset.
536 P.BeginOffset = P.SplitTails.empty() ? P.SI->beginOffset() : P.EndOffset;
537 P.EndOffset = P.SI->endOffset();
538 ++P.SJ;
539
540 // There are two strategies to form a partition based on whether the
541 // partition starts with an unsplittable slice or a splittable slice.
542 if (!P.SI->isSplittable()) {
543 // When we're forming an unsplittable region, it must always start at
544 // the first slice and will extend through its end.
545 assert(P.BeginOffset == P.SI->beginOffset())((P.BeginOffset == P.SI->beginOffset()) ? static_cast<void
> (0) : __assert_fail ("P.BeginOffset == P.SI->beginOffset()"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 545, __PRETTY_FUNCTION__))
;
546
547 // Form a partition including all of the overlapping slices with this
548 // unsplittable slice.
549 while (P.SJ != SE && P.SJ->beginOffset() < P.EndOffset) {
550 if (!P.SJ->isSplittable())
551 P.EndOffset = std::max(P.EndOffset, P.SJ->endOffset());
552 ++P.SJ;
553 }
554
555 // We have a partition across a set of overlapping unsplittable
556 // partitions.
557 return;
558 }
559
560 // If we're starting with a splittable slice, then we need to form
561 // a synthetic partition spanning it and any other overlapping splittable
562 // splices.
563 assert(P.SI->isSplittable() && "Forming a splittable partition!")((P.SI->isSplittable() && "Forming a splittable partition!"
) ? static_cast<void> (0) : __assert_fail ("P.SI->isSplittable() && \"Forming a splittable partition!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 563, __PRETTY_FUNCTION__))
;
564
565 // Collect all of the overlapping splittable slices.
566 while (P.SJ != SE && P.SJ->beginOffset() < P.EndOffset &&
567 P.SJ->isSplittable()) {
568 P.EndOffset = std::max(P.EndOffset, P.SJ->endOffset());
569 ++P.SJ;
570 }
571
572 // Back upiP.EndOffset if we ended the span early when encountering an
573 // unsplittable slice. This synthesizes the early end offset of
574 // a partition spanning only splittable slices.
575 if (P.SJ != SE && P.SJ->beginOffset() < P.EndOffset) {
576 assert(!P.SJ->isSplittable())((!P.SJ->isSplittable()) ? static_cast<void> (0) : __assert_fail
("!P.SJ->isSplittable()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 576, __PRETTY_FUNCTION__))
;
577 P.EndOffset = P.SJ->beginOffset();
578 }
579 }
580
581public:
582 bool operator==(const partition_iterator &RHS) const {
583 assert(SE == RHS.SE &&((SE == RHS.SE && "End iterators don't match between compared partition iterators!"
) ? static_cast<void> (0) : __assert_fail ("SE == RHS.SE && \"End iterators don't match between compared partition iterators!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 584, __PRETTY_FUNCTION__))
584 "End iterators don't match between compared partition iterators!")((SE == RHS.SE && "End iterators don't match between compared partition iterators!"
) ? static_cast<void> (0) : __assert_fail ("SE == RHS.SE && \"End iterators don't match between compared partition iterators!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 584, __PRETTY_FUNCTION__))
;
585
586 // The observed positions of partitions is marked by the P.SI iterator and
587 // the emptiness of the split slices. The latter is only relevant when
588 // P.SI == SE, as the end iterator will additionally have an empty split
589 // slices list, but the prior may have the same P.SI and a tail of split
590 // slices.
591 if (P.SI == RHS.P.SI && P.SplitTails.empty() == RHS.P.SplitTails.empty()) {
592 assert(P.SJ == RHS.P.SJ &&((P.SJ == RHS.P.SJ && "Same set of slices formed two different sized partitions!"
) ? static_cast<void> (0) : __assert_fail ("P.SJ == RHS.P.SJ && \"Same set of slices formed two different sized partitions!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 593, __PRETTY_FUNCTION__))
593 "Same set of slices formed two different sized partitions!")((P.SJ == RHS.P.SJ && "Same set of slices formed two different sized partitions!"
) ? static_cast<void> (0) : __assert_fail ("P.SJ == RHS.P.SJ && \"Same set of slices formed two different sized partitions!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 593, __PRETTY_FUNCTION__))
;
594 assert(P.SplitTails.size() == RHS.P.SplitTails.size() &&((P.SplitTails.size() == RHS.P.SplitTails.size() && "Same slice position with differently sized non-empty split "
"slice tails!") ? static_cast<void> (0) : __assert_fail
("P.SplitTails.size() == RHS.P.SplitTails.size() && \"Same slice position with differently sized non-empty split \" \"slice tails!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 596, __PRETTY_FUNCTION__))
595 "Same slice position with differently sized non-empty split "((P.SplitTails.size() == RHS.P.SplitTails.size() && "Same slice position with differently sized non-empty split "
"slice tails!") ? static_cast<void> (0) : __assert_fail
("P.SplitTails.size() == RHS.P.SplitTails.size() && \"Same slice position with differently sized non-empty split \" \"slice tails!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 596, __PRETTY_FUNCTION__))
596 "slice tails!")((P.SplitTails.size() == RHS.P.SplitTails.size() && "Same slice position with differently sized non-empty split "
"slice tails!") ? static_cast<void> (0) : __assert_fail
("P.SplitTails.size() == RHS.P.SplitTails.size() && \"Same slice position with differently sized non-empty split \" \"slice tails!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 596, __PRETTY_FUNCTION__))
;
597 return true;
598 }
599 return false;
600 }
601
602 partition_iterator &operator++() {
603 advance();
604 return *this;
605 }
606
607 Partition &operator*() { return P; }
608};
609
610/// A forward range over the partitions of the alloca's slices.
611///
612/// This accesses an iterator range over the partitions of the alloca's
613/// slices. It computes these partitions on the fly based on the overlapping
614/// offsets of the slices and the ability to split them. It will visit "empty"
615/// partitions to cover regions of the alloca only accessed via split
616/// slices.
617iterator_range<AllocaSlices::partition_iterator> AllocaSlices::partitions() {
618 return make_range(partition_iterator(begin(), end()),
619 partition_iterator(end(), end()));
620}
621
622static Value *foldSelectInst(SelectInst &SI) {
623 // If the condition being selected on is a constant or the same value is
624 // being selected between, fold the select. Yes this does (rarely) happen
625 // early on.
626 if (ConstantInt *CI = dyn_cast<ConstantInt>(SI.getCondition()))
627 return SI.getOperand(1 + CI->isZero());
628 if (SI.getOperand(1) == SI.getOperand(2))
629 return SI.getOperand(1);
630
631 return nullptr;
632}
633
634/// A helper that folds a PHI node or a select.
635static Value *foldPHINodeOrSelectInst(Instruction &I) {
636 if (PHINode *PN = dyn_cast<PHINode>(&I)) {
637 // If PN merges together the same value, return that value.
638 return PN->hasConstantValue();
639 }
640 return foldSelectInst(cast<SelectInst>(I));
641}
642
643/// Builder for the alloca slices.
644///
645/// This class builds a set of alloca slices by recursively visiting the uses
646/// of an alloca and making a slice for each load and store at each offset.
647class AllocaSlices::SliceBuilder : public PtrUseVisitor<SliceBuilder> {
648 friend class PtrUseVisitor<SliceBuilder>;
649 friend class InstVisitor<SliceBuilder>;
650
651 using Base = PtrUseVisitor<SliceBuilder>;
652
653 const uint64_t AllocSize;
654 AllocaSlices &AS;
655
656 SmallDenseMap<Instruction *, unsigned> MemTransferSliceMap;
657 SmallDenseMap<Instruction *, uint64_t> PHIOrSelectSizes;
658
659 /// Set to de-duplicate dead instructions found in the use walk.
660 SmallPtrSet<Instruction *, 4> VisitedDeadInsts;
661
662public:
663 SliceBuilder(const DataLayout &DL, AllocaInst &AI, AllocaSlices &AS)
664 : PtrUseVisitor<SliceBuilder>(DL),
665 AllocSize(DL.getTypeAllocSize(AI.getAllocatedType())), AS(AS) {}
666
667private:
668 void markAsDead(Instruction &I) {
669 if (VisitedDeadInsts.insert(&I).second)
670 AS.DeadUsers.push_back(&I);
671 }
672
673 void insertUse(Instruction &I, const APInt &Offset, uint64_t Size,
674 bool IsSplittable = false) {
675 // Completely skip uses which have a zero size or start either before or
676 // past the end of the allocation.
677 if (Size == 0 || Offset.uge(AllocSize)) {
678 LLVM_DEBUG(dbgs() << "WARNING: Ignoring " << Size << " byte use @"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte use @" << Offset << " which has zero size or starts outside of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
679 << Offsetdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte use @" << Offset << " which has zero size or starts outside of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
680 << " which has zero size or starts outside of the "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte use @" << Offset << " which has zero size or starts outside of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
681 << AllocSize << " byte alloca:\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte use @" << Offset << " which has zero size or starts outside of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
682 << " alloca: " << AS.AI << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte use @" << Offset << " which has zero size or starts outside of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
683 << " use: " << I << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte use @" << Offset << " which has zero size or starts outside of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
;
684 return markAsDead(I);
685 }
686
687 uint64_t BeginOffset = Offset.getZExtValue();
688 uint64_t EndOffset = BeginOffset + Size;
689
690 // Clamp the end offset to the end of the allocation. Note that this is
691 // formulated to handle even the case where "BeginOffset + Size" overflows.
692 // This may appear superficially to be something we could ignore entirely,
693 // but that is not so! There may be widened loads or PHI-node uses where
694 // some instructions are dead but not others. We can't completely ignore
695 // them, and so have to record at least the information here.
696 assert(AllocSize >= BeginOffset)((AllocSize >= BeginOffset) ? static_cast<void> (0) :
__assert_fail ("AllocSize >= BeginOffset", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 696, __PRETTY_FUNCTION__))
; // Established above.
697 if (Size > AllocSize - BeginOffset) {
698 LLVM_DEBUG(dbgs() << "WARNING: Clamping a " << Size << " byte use @"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Clamping a " << Size
<< " byte use @" << Offset << " to remain within the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
699 << Offset << " to remain within the " << AllocSizedo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Clamping a " << Size
<< " byte use @" << Offset << " to remain within the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
700 << " byte alloca:\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Clamping a " << Size
<< " byte use @" << Offset << " to remain within the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
701 << " alloca: " << AS.AI << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Clamping a " << Size
<< " byte use @" << Offset << " to remain within the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
702 << " use: " << I << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Clamping a " << Size
<< " byte use @" << Offset << " to remain within the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
I << "\n"; } } while (false)
;
703 EndOffset = AllocSize;
704 }
705
706 AS.Slices.push_back(Slice(BeginOffset, EndOffset, U, IsSplittable));
707 }
708
709 void visitBitCastInst(BitCastInst &BC) {
710 if (BC.use_empty())
711 return markAsDead(BC);
712
713 return Base::visitBitCastInst(BC);
714 }
715
716 void visitAddrSpaceCastInst(AddrSpaceCastInst &ASC) {
717 if (ASC.use_empty())
718 return markAsDead(ASC);
719
720 return Base::visitAddrSpaceCastInst(ASC);
721 }
722
723 void visitGetElementPtrInst(GetElementPtrInst &GEPI) {
724 if (GEPI.use_empty())
725 return markAsDead(GEPI);
726
727 if (SROAStrictInbounds && GEPI.isInBounds()) {
728 // FIXME: This is a manually un-factored variant of the basic code inside
729 // of GEPs with checking of the inbounds invariant specified in the
730 // langref in a very strict sense. If we ever want to enable
731 // SROAStrictInbounds, this code should be factored cleanly into
732 // PtrUseVisitor, but it is easier to experiment with SROAStrictInbounds
733 // by writing out the code here where we have the underlying allocation
734 // size readily available.
735 APInt GEPOffset = Offset;
736 const DataLayout &DL = GEPI.getModule()->getDataLayout();
737 for (gep_type_iterator GTI = gep_type_begin(GEPI),
738 GTE = gep_type_end(GEPI);
739 GTI != GTE; ++GTI) {
740 ConstantInt *OpC = dyn_cast<ConstantInt>(GTI.getOperand());
741 if (!OpC)
742 break;
743
744 // Handle a struct index, which adds its field offset to the pointer.
745 if (StructType *STy = GTI.getStructTypeOrNull()) {
746 unsigned ElementIdx = OpC->getZExtValue();
747 const StructLayout *SL = DL.getStructLayout(STy);
748 GEPOffset +=
749 APInt(Offset.getBitWidth(), SL->getElementOffset(ElementIdx));
750 } else {
751 // For array or vector indices, scale the index by the size of the
752 // type.
753 APInt Index = OpC->getValue().sextOrTrunc(Offset.getBitWidth());
754 GEPOffset += Index * APInt(Offset.getBitWidth(),
755 DL.getTypeAllocSize(GTI.getIndexedType()));
756 }
757
758 // If this index has computed an intermediate pointer which is not
759 // inbounds, then the result of the GEP is a poison value and we can
760 // delete it and all uses.
761 if (GEPOffset.ugt(AllocSize))
762 return markAsDead(GEPI);
763 }
764 }
765
766 return Base::visitGetElementPtrInst(GEPI);
767 }
768
769 void handleLoadOrStore(Type *Ty, Instruction &I, const APInt &Offset,
770 uint64_t Size, bool IsVolatile) {
771 // We allow splitting of non-volatile loads and stores where the type is an
772 // integer type. These may be used to implement 'memcpy' or other "transfer
773 // of bits" patterns.
774 bool IsSplittable = Ty->isIntegerTy() && !IsVolatile;
775
776 insertUse(I, Offset, Size, IsSplittable);
777 }
778
779 void visitLoadInst(LoadInst &LI) {
780 assert((!LI.isSimple() || LI.getType()->isSingleValueType()) &&(((!LI.isSimple() || LI.getType()->isSingleValueType()) &&
"All simple FCA loads should have been pre-split") ? static_cast
<void> (0) : __assert_fail ("(!LI.isSimple() || LI.getType()->isSingleValueType()) && \"All simple FCA loads should have been pre-split\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 781, __PRETTY_FUNCTION__))
781 "All simple FCA loads should have been pre-split")(((!LI.isSimple() || LI.getType()->isSingleValueType()) &&
"All simple FCA loads should have been pre-split") ? static_cast
<void> (0) : __assert_fail ("(!LI.isSimple() || LI.getType()->isSingleValueType()) && \"All simple FCA loads should have been pre-split\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 781, __PRETTY_FUNCTION__))
;
782
783 if (!IsOffsetKnown)
784 return PI.setAborted(&LI);
785
786 if (LI.isVolatile() &&
787 LI.getPointerAddressSpace() != DL.getAllocaAddrSpace())
788 return PI.setAborted(&LI);
789
790 uint64_t Size = DL.getTypeStoreSize(LI.getType());
791 return handleLoadOrStore(LI.getType(), LI, Offset, Size, LI.isVolatile());
792 }
793
794 void visitStoreInst(StoreInst &SI) {
795 Value *ValOp = SI.getValueOperand();
796 if (ValOp == *U)
797 return PI.setEscapedAndAborted(&SI);
798 if (!IsOffsetKnown)
799 return PI.setAborted(&SI);
800
801 if (SI.isVolatile() &&
802 SI.getPointerAddressSpace() != DL.getAllocaAddrSpace())
803 return PI.setAborted(&SI);
804
805 uint64_t Size = DL.getTypeStoreSize(ValOp->getType());
806
807 // If this memory access can be shown to *statically* extend outside the
808 // bounds of the allocation, it's behavior is undefined, so simply
809 // ignore it. Note that this is more strict than the generic clamping
810 // behavior of insertUse. We also try to handle cases which might run the
811 // risk of overflow.
812 // FIXME: We should instead consider the pointer to have escaped if this
813 // function is being instrumented for addressing bugs or race conditions.
814 if (Size > AllocSize || Offset.ugt(AllocSize - Size)) {
815 LLVM_DEBUG(dbgs() << "WARNING: Ignoring " << Size << " byte store @"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte store @" << Offset << " which extends past the end of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
SI << "\n"; } } while (false)
816 << Offset << " which extends past the end of the "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte store @" << Offset << " which extends past the end of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
SI << "\n"; } } while (false)
817 << AllocSize << " byte alloca:\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte store @" << Offset << " which extends past the end of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
SI << "\n"; } } while (false)
818 << " alloca: " << AS.AI << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte store @" << Offset << " which extends past the end of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
SI << "\n"; } } while (false)
819 << " use: " << SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "WARNING: Ignoring " << Size
<< " byte store @" << Offset << " which extends past the end of the "
<< AllocSize << " byte alloca:\n" << " alloca: "
<< AS.AI << "\n" << " use: " <<
SI << "\n"; } } while (false)
;
820 return markAsDead(SI);
821 }
822
823 assert((!SI.isSimple() || ValOp->getType()->isSingleValueType()) &&(((!SI.isSimple() || ValOp->getType()->isSingleValueType
()) && "All simple FCA stores should have been pre-split"
) ? static_cast<void> (0) : __assert_fail ("(!SI.isSimple() || ValOp->getType()->isSingleValueType()) && \"All simple FCA stores should have been pre-split\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 824, __PRETTY_FUNCTION__))
824 "All simple FCA stores should have been pre-split")(((!SI.isSimple() || ValOp->getType()->isSingleValueType
()) && "All simple FCA stores should have been pre-split"
) ? static_cast<void> (0) : __assert_fail ("(!SI.isSimple() || ValOp->getType()->isSingleValueType()) && \"All simple FCA stores should have been pre-split\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 824, __PRETTY_FUNCTION__))
;
825 handleLoadOrStore(ValOp->getType(), SI, Offset, Size, SI.isVolatile());
826 }
827
828 void visitMemSetInst(MemSetInst &II) {
829 assert(II.getRawDest() == *U && "Pointer use is not the destination?")((II.getRawDest() == *U && "Pointer use is not the destination?"
) ? static_cast<void> (0) : __assert_fail ("II.getRawDest() == *U && \"Pointer use is not the destination?\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 829, __PRETTY_FUNCTION__))
;
830 ConstantInt *Length = dyn_cast<ConstantInt>(II.getLength());
831 if ((Length && Length->getValue() == 0) ||
832 (IsOffsetKnown && Offset.uge(AllocSize)))
833 // Zero-length mem transfer intrinsics can be ignored entirely.
834 return markAsDead(II);
835
836 if (!IsOffsetKnown)
837 return PI.setAborted(&II);
838
839 // Don't replace this with a store with a different address space. TODO:
840 // Use a store with the casted new alloca?
841 if (II.isVolatile() && II.getDestAddressSpace() != DL.getAllocaAddrSpace())
842 return PI.setAborted(&II);
843
844 insertUse(II, Offset, Length ? Length->getLimitedValue()
845 : AllocSize - Offset.getLimitedValue(),
846 (bool)Length);
847 }
848
849 void visitMemTransferInst(MemTransferInst &II) {
850 ConstantInt *Length = dyn_cast<ConstantInt>(II.getLength());
851 if (Length && Length->getValue() == 0)
852 // Zero-length mem transfer intrinsics can be ignored entirely.
853 return markAsDead(II);
854
855 // Because we can visit these intrinsics twice, also check to see if the
856 // first time marked this instruction as dead. If so, skip it.
857 if (VisitedDeadInsts.count(&II))
858 return;
859
860 if (!IsOffsetKnown)
861 return PI.setAborted(&II);
862
863 // Don't replace this with a load/store with a different address space.
864 // TODO: Use a store with the casted new alloca?
865 if (II.isVolatile() &&
866 (II.getDestAddressSpace() != DL.getAllocaAddrSpace() ||
867 II.getSourceAddressSpace() != DL.getAllocaAddrSpace()))
868 return PI.setAborted(&II);
869
870 // This side of the transfer is completely out-of-bounds, and so we can
871 // nuke the entire transfer. However, we also need to nuke the other side
872 // if already added to our partitions.
873 // FIXME: Yet another place we really should bypass this when
874 // instrumenting for ASan.
875 if (Offset.uge(AllocSize)) {
876 SmallDenseMap<Instruction *, unsigned>::iterator MTPI =
877 MemTransferSliceMap.find(&II);
878 if (MTPI != MemTransferSliceMap.end())
879 AS.Slices[MTPI->second].kill();
880 return markAsDead(II);
881 }
882
883 uint64_t RawOffset = Offset.getLimitedValue();
884 uint64_t Size = Length ? Length->getLimitedValue() : AllocSize - RawOffset;
885
886 // Check for the special case where the same exact value is used for both
887 // source and dest.
888 if (*U == II.getRawDest() && *U == II.getRawSource()) {
889 // For non-volatile transfers this is a no-op.
890 if (!II.isVolatile())
891 return markAsDead(II);
892
893 return insertUse(II, Offset, Size, /*IsSplittable=*/false);
894 }
895
896 // If we have seen both source and destination for a mem transfer, then
897 // they both point to the same alloca.
898 bool Inserted;
899 SmallDenseMap<Instruction *, unsigned>::iterator MTPI;
900 std::tie(MTPI, Inserted) =
901 MemTransferSliceMap.insert(std::make_pair(&II, AS.Slices.size()));
902 unsigned PrevIdx = MTPI->second;
903 if (!Inserted) {
904 Slice &PrevP = AS.Slices[PrevIdx];
905
906 // Check if the begin offsets match and this is a non-volatile transfer.
907 // In that case, we can completely elide the transfer.
908 if (!II.isVolatile() && PrevP.beginOffset() == RawOffset) {
909 PrevP.kill();
910 return markAsDead(II);
911 }
912
913 // Otherwise we have an offset transfer within the same alloca. We can't
914 // split those.
915 PrevP.makeUnsplittable();
916 }
917
918 // Insert the use now that we've fixed up the splittable nature.
919 insertUse(II, Offset, Size, /*IsSplittable=*/Inserted && Length);
920
921 // Check that we ended up with a valid index in the map.
922 assert(AS.Slices[PrevIdx].getUse()->getUser() == &II &&((AS.Slices[PrevIdx].getUse()->getUser() == &II &&
"Map index doesn't point back to a slice with this user.") ?
static_cast<void> (0) : __assert_fail ("AS.Slices[PrevIdx].getUse()->getUser() == &II && \"Map index doesn't point back to a slice with this user.\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 923, __PRETTY_FUNCTION__))
923 "Map index doesn't point back to a slice with this user.")((AS.Slices[PrevIdx].getUse()->getUser() == &II &&
"Map index doesn't point back to a slice with this user.") ?
static_cast<void> (0) : __assert_fail ("AS.Slices[PrevIdx].getUse()->getUser() == &II && \"Map index doesn't point back to a slice with this user.\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 923, __PRETTY_FUNCTION__))
;
924 }
925
926 // Disable SRoA for any intrinsics except for lifetime invariants.
927 // FIXME: What about debug intrinsics? This matches old behavior, but
928 // doesn't make sense.
929 void visitIntrinsicInst(IntrinsicInst &II) {
930 if (!IsOffsetKnown)
931 return PI.setAborted(&II);
932
933 if (II.isLifetimeStartOrEnd()) {
934 ConstantInt *Length = cast<ConstantInt>(II.getArgOperand(0));
935 uint64_t Size = std::min(AllocSize - Offset.getLimitedValue(),
936 Length->getLimitedValue());
937 insertUse(II, Offset, Size, true);
938 return;
939 }
940
941 Base::visitIntrinsicInst(II);
942 }
943
944 Instruction *hasUnsafePHIOrSelectUse(Instruction *Root, uint64_t &Size) {
945 // We consider any PHI or select that results in a direct load or store of
946 // the same offset to be a viable use for slicing purposes. These uses
947 // are considered unsplittable and the size is the maximum loaded or stored
948 // size.
949 SmallPtrSet<Instruction *, 4> Visited;
950 SmallVector<std::pair<Instruction *, Instruction *>, 4> Uses;
951 Visited.insert(Root);
952 Uses.push_back(std::make_pair(cast<Instruction>(*U), Root));
953 const DataLayout &DL = Root->getModule()->getDataLayout();
954 // If there are no loads or stores, the access is dead. We mark that as
955 // a size zero access.
956 Size = 0;
957 do {
958 Instruction *I, *UsedI;
959 std::tie(UsedI, I) = Uses.pop_back_val();
960
961 if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
962 Size = std::max(Size,
963 DL.getTypeStoreSize(LI->getType()).getFixedSize());
964 continue;
965 }
966 if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
967 Value *Op = SI->getOperand(0);
968 if (Op == UsedI)
969 return SI;
970 Size = std::max(Size,
971 DL.getTypeStoreSize(Op->getType()).getFixedSize());
972 continue;
973 }
974
975 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
976 if (!GEP->hasAllZeroIndices())
977 return GEP;
978 } else if (!isa<BitCastInst>(I) && !isa<PHINode>(I) &&
979 !isa<SelectInst>(I) && !isa<AddrSpaceCastInst>(I)) {
980 return I;
981 }
982
983 for (User *U : I->users())
984 if (Visited.insert(cast<Instruction>(U)).second)
985 Uses.push_back(std::make_pair(I, cast<Instruction>(U)));
986 } while (!Uses.empty());
987
988 return nullptr;
989 }
990
991 void visitPHINodeOrSelectInst(Instruction &I) {
992 assert(isa<PHINode>(I) || isa<SelectInst>(I))((isa<PHINode>(I) || isa<SelectInst>(I)) ? static_cast
<void> (0) : __assert_fail ("isa<PHINode>(I) || isa<SelectInst>(I)"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 992, __PRETTY_FUNCTION__))
;
993 if (I.use_empty())
994 return markAsDead(I);
995
996 // TODO: We could use SimplifyInstruction here to fold PHINodes and
997 // SelectInsts. However, doing so requires to change the current
998 // dead-operand-tracking mechanism. For instance, suppose neither loading
999 // from %U nor %other traps. Then "load (select undef, %U, %other)" does not
1000 // trap either. However, if we simply replace %U with undef using the
1001 // current dead-operand-tracking mechanism, "load (select undef, undef,
1002 // %other)" may trap because the select may return the first operand
1003 // "undef".
1004 if (Value *Result = foldPHINodeOrSelectInst(I)) {
1005 if (Result == *U)
1006 // If the result of the constant fold will be the pointer, recurse
1007 // through the PHI/select as if we had RAUW'ed it.
1008 enqueueUsers(I);
1009 else
1010 // Otherwise the operand to the PHI/select is dead, and we can replace
1011 // it with undef.
1012 AS.DeadOperands.push_back(U);
1013
1014 return;
1015 }
1016
1017 if (!IsOffsetKnown)
1018 return PI.setAborted(&I);
1019
1020 // See if we already have computed info on this node.
1021 uint64_t &Size = PHIOrSelectSizes[&I];
1022 if (!Size) {
1023 // This is a new PHI/Select, check for an unsafe use of it.
1024 if (Instruction *UnsafeI = hasUnsafePHIOrSelectUse(&I, Size))
1025 return PI.setAborted(UnsafeI);
1026 }
1027
1028 // For PHI and select operands outside the alloca, we can't nuke the entire
1029 // phi or select -- the other side might still be relevant, so we special
1030 // case them here and use a separate structure to track the operands
1031 // themselves which should be replaced with undef.
1032 // FIXME: This should instead be escaped in the event we're instrumenting
1033 // for address sanitization.
1034 if (Offset.uge(AllocSize)) {
1035 AS.DeadOperands.push_back(U);
1036 return;
1037 }
1038
1039 insertUse(I, Offset, Size);
1040 }
1041
1042 void visitPHINode(PHINode &PN) { visitPHINodeOrSelectInst(PN); }
1043
1044 void visitSelectInst(SelectInst &SI) { visitPHINodeOrSelectInst(SI); }
1045
1046 /// Disable SROA entirely if there are unhandled users of the alloca.
1047 void visitInstruction(Instruction &I) { PI.setAborted(&I); }
1048};
1049
1050AllocaSlices::AllocaSlices(const DataLayout &DL, AllocaInst &AI)
1051 :
1052#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
1053 AI(AI),
1054#endif
1055 PointerEscapingInstr(nullptr) {
1056 SliceBuilder PB(DL, AI, *this);
1057 SliceBuilder::PtrInfo PtrI = PB.visitPtr(AI);
1058 if (PtrI.isEscaped() || PtrI.isAborted()) {
1059 // FIXME: We should sink the escape vs. abort info into the caller nicely,
1060 // possibly by just storing the PtrInfo in the AllocaSlices.
1061 PointerEscapingInstr = PtrI.getEscapingInst() ? PtrI.getEscapingInst()
1062 : PtrI.getAbortingInst();
1063 assert(PointerEscapingInstr && "Did not track a bad instruction")((PointerEscapingInstr && "Did not track a bad instruction"
) ? static_cast<void> (0) : __assert_fail ("PointerEscapingInstr && \"Did not track a bad instruction\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1063, __PRETTY_FUNCTION__))
;
1064 return;
1065 }
1066
1067 Slices.erase(
1068 llvm::remove_if(Slices, [](const Slice &S) { return S.isDead(); }),
1069 Slices.end());
1070
1071#ifndef NDEBUG
1072 if (SROARandomShuffleSlices) {
1073 std::mt19937 MT(static_cast<unsigned>(
1074 std::chrono::system_clock::now().time_since_epoch().count()));
1075 std::shuffle(Slices.begin(), Slices.end(), MT);
1076 }
1077#endif
1078
1079 // Sort the uses. This arranges for the offsets to be in ascending order,
1080 // and the sizes to be in descending order.
1081 llvm::sort(Slices);
1082}
1083
1084#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
1085
1086void AllocaSlices::print(raw_ostream &OS, const_iterator I,
1087 StringRef Indent) const {
1088 printSlice(OS, I, Indent);
1089 OS << "\n";
1090 printUse(OS, I, Indent);
1091}
1092
1093void AllocaSlices::printSlice(raw_ostream &OS, const_iterator I,
1094 StringRef Indent) const {
1095 OS << Indent << "[" << I->beginOffset() << "," << I->endOffset() << ")"
1096 << " slice #" << (I - begin())
1097 << (I->isSplittable() ? " (splittable)" : "");
1098}
1099
1100void AllocaSlices::printUse(raw_ostream &OS, const_iterator I,
1101 StringRef Indent) const {
1102 OS << Indent << " used by: " << *I->getUse()->getUser() << "\n";
1103}
1104
1105void AllocaSlices::print(raw_ostream &OS) const {
1106 if (PointerEscapingInstr) {
1107 OS << "Can't analyze slices for alloca: " << AI << "\n"
1108 << " A pointer to this alloca escaped by:\n"
1109 << " " << *PointerEscapingInstr << "\n";
1110 return;
1111 }
1112
1113 OS << "Slices of alloca: " << AI << "\n";
1114 for (const_iterator I = begin(), E = end(); I != E; ++I)
1115 print(OS, I);
1116}
1117
1118LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void AllocaSlices::dump(const_iterator I) const {
1119 print(dbgs(), I);
1120}
1121LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void AllocaSlices::dump() const { print(dbgs()); }
1122
1123#endif // !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
1124
1125/// Walk the range of a partitioning looking for a common type to cover this
1126/// sequence of slices.
1127static Type *findCommonType(AllocaSlices::const_iterator B,
1128 AllocaSlices::const_iterator E,
1129 uint64_t EndOffset) {
1130 Type *Ty = nullptr;
1131 bool TyIsCommon = true;
1132 IntegerType *ITy = nullptr;
1133
1134 // Note that we need to look at *every* alloca slice's Use to ensure we
1135 // always get consistent results regardless of the order of slices.
1136 for (AllocaSlices::const_iterator I = B; I != E; ++I) {
1137 Use *U = I->getUse();
1138 if (isa<IntrinsicInst>(*U->getUser()))
1139 continue;
1140 if (I->beginOffset() != B->beginOffset() || I->endOffset() != EndOffset)
1141 continue;
1142
1143 Type *UserTy = nullptr;
1144 if (LoadInst *LI = dyn_cast<LoadInst>(U->getUser())) {
1145 UserTy = LI->getType();
1146 } else if (StoreInst *SI = dyn_cast<StoreInst>(U->getUser())) {
1147 UserTy = SI->getValueOperand()->getType();
1148 }
1149
1150 if (IntegerType *UserITy = dyn_cast_or_null<IntegerType>(UserTy)) {
1151 // If the type is larger than the partition, skip it. We only encounter
1152 // this for split integer operations where we want to use the type of the
1153 // entity causing the split. Also skip if the type is not a byte width
1154 // multiple.
1155 if (UserITy->getBitWidth() % 8 != 0 ||
1156 UserITy->getBitWidth() / 8 > (EndOffset - B->beginOffset()))
1157 continue;
1158
1159 // Track the largest bitwidth integer type used in this way in case there
1160 // is no common type.
1161 if (!ITy || ITy->getBitWidth() < UserITy->getBitWidth())
1162 ITy = UserITy;
1163 }
1164
1165 // To avoid depending on the order of slices, Ty and TyIsCommon must not
1166 // depend on types skipped above.
1167 if (!UserTy || (Ty && Ty != UserTy))
1168 TyIsCommon = false; // Give up on anything but an iN type.
1169 else
1170 Ty = UserTy;
1171 }
1172
1173 return TyIsCommon ? Ty : ITy;
1174}
1175
1176/// PHI instructions that use an alloca and are subsequently loaded can be
1177/// rewritten to load both input pointers in the pred blocks and then PHI the
1178/// results, allowing the load of the alloca to be promoted.
1179/// From this:
1180/// %P2 = phi [i32* %Alloca, i32* %Other]
1181/// %V = load i32* %P2
1182/// to:
1183/// %V1 = load i32* %Alloca -> will be mem2reg'd
1184/// ...
1185/// %V2 = load i32* %Other
1186/// ...
1187/// %V = phi [i32 %V1, i32 %V2]
1188///
1189/// We can do this to a select if its only uses are loads and if the operands
1190/// to the select can be loaded unconditionally.
1191///
1192/// FIXME: This should be hoisted into a generic utility, likely in
1193/// Transforms/Util/Local.h
1194static bool isSafePHIToSpeculate(PHINode &PN) {
1195 const DataLayout &DL = PN.getModule()->getDataLayout();
1196
1197 // For now, we can only do this promotion if the load is in the same block
1198 // as the PHI, and if there are no stores between the phi and load.
1199 // TODO: Allow recursive phi users.
1200 // TODO: Allow stores.
1201 BasicBlock *BB = PN.getParent();
1202 unsigned MaxAlign = 0;
1203 uint64_t APWidth = DL.getIndexTypeSizeInBits(PN.getType());
1204 APInt MaxSize(APWidth, 0);
1205 bool HaveLoad = false;
1206 for (User *U : PN.users()) {
1207 LoadInst *LI = dyn_cast<LoadInst>(U);
1208 if (!LI || !LI->isSimple())
1209 return false;
1210
1211 // For now we only allow loads in the same block as the PHI. This is
1212 // a common case that happens when instcombine merges two loads through
1213 // a PHI.
1214 if (LI->getParent() != BB)
1215 return false;
1216
1217 // Ensure that there are no instructions between the PHI and the load that
1218 // could store.
1219 for (BasicBlock::iterator BBI(PN); &*BBI != LI; ++BBI)
1220 if (BBI->mayWriteToMemory())
1221 return false;
1222
1223 uint64_t Size = DL.getTypeStoreSize(LI->getType());
1224 MaxAlign = std::max(MaxAlign, LI->getAlignment());
1225 MaxSize = MaxSize.ult(Size) ? APInt(APWidth, Size) : MaxSize;
1226 HaveLoad = true;
1227 }
1228
1229 if (!HaveLoad)
1230 return false;
1231
1232 // We can only transform this if it is safe to push the loads into the
1233 // predecessor blocks. The only thing to watch out for is that we can't put
1234 // a possibly trapping load in the predecessor if it is a critical edge.
1235 for (unsigned Idx = 0, Num = PN.getNumIncomingValues(); Idx != Num; ++Idx) {
1236 Instruction *TI = PN.getIncomingBlock(Idx)->getTerminator();
1237 Value *InVal = PN.getIncomingValue(Idx);
1238
1239 // If the value is produced by the terminator of the predecessor (an
1240 // invoke) or it has side-effects, there is no valid place to put a load
1241 // in the predecessor.
1242 if (TI == InVal || TI->mayHaveSideEffects())
1243 return false;
1244
1245 // If the predecessor has a single successor, then the edge isn't
1246 // critical.
1247 if (TI->getNumSuccessors() == 1)
1248 continue;
1249
1250 // If this pointer is always safe to load, or if we can prove that there
1251 // is already a load in the block, then we can move the load to the pred
1252 // block.
1253 if (isSafeToLoadUnconditionally(InVal, MaxAlign, MaxSize, DL, TI))
1254 continue;
1255
1256 return false;
1257 }
1258
1259 return true;
1260}
1261
1262static void speculatePHINodeLoads(PHINode &PN) {
1263 LLVM_DEBUG(dbgs() << " original: " << PN << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << PN <<
"\n"; } } while (false)
;
1264
1265 LoadInst *SomeLoad = cast<LoadInst>(PN.user_back());
1266 Type *LoadTy = SomeLoad->getType();
1267 IRBuilderTy PHIBuilder(&PN);
1268 PHINode *NewPN = PHIBuilder.CreatePHI(LoadTy, PN.getNumIncomingValues(),
1269 PN.getName() + ".sroa.speculated");
1270
1271 // Get the AA tags and alignment to use from one of the loads. It does not
1272 // matter which one we get and if any differ.
1273 AAMDNodes AATags;
1274 SomeLoad->getAAMetadata(AATags);
1275 const MaybeAlign Align = MaybeAlign(SomeLoad->getAlignment());
1276
1277 // Rewrite all loads of the PN to use the new PHI.
1278 while (!PN.use_empty()) {
1279 LoadInst *LI = cast<LoadInst>(PN.user_back());
1280 LI->replaceAllUsesWith(NewPN);
1281 LI->eraseFromParent();
1282 }
1283
1284 // Inject loads into all of the pred blocks.
1285 DenseMap<BasicBlock*, Value*> InjectedLoads;
1286 for (unsigned Idx = 0, Num = PN.getNumIncomingValues(); Idx != Num; ++Idx) {
1287 BasicBlock *Pred = PN.getIncomingBlock(Idx);
1288 Value *InVal = PN.getIncomingValue(Idx);
1289
1290 // A PHI node is allowed to have multiple (duplicated) entries for the same
1291 // basic block, as long as the value is the same. So if we already injected
1292 // a load in the predecessor, then we should reuse the same load for all
1293 // duplicated entries.
1294 if (Value* V = InjectedLoads.lookup(Pred)) {
1295 NewPN->addIncoming(V, Pred);
1296 continue;
1297 }
1298
1299 Instruction *TI = Pred->getTerminator();
1300 IRBuilderTy PredBuilder(TI);
1301
1302 LoadInst *Load = PredBuilder.CreateLoad(
1303 LoadTy, InVal,
1304 (PN.getName() + ".sroa.speculate.load." + Pred->getName()));
1305 ++NumLoadsSpeculated;
1306 Load->setAlignment(Align);
1307 if (AATags)
1308 Load->setAAMetadata(AATags);
1309 NewPN->addIncoming(Load, Pred);
1310 InjectedLoads[Pred] = Load;
1311 }
1312
1313 LLVM_DEBUG(dbgs() << " speculated to: " << *NewPN << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " speculated to: " <<
*NewPN << "\n"; } } while (false)
;
1314 PN.eraseFromParent();
1315}
1316
1317/// Select instructions that use an alloca and are subsequently loaded can be
1318/// rewritten to load both input pointers and then select between the result,
1319/// allowing the load of the alloca to be promoted.
1320/// From this:
1321/// %P2 = select i1 %cond, i32* %Alloca, i32* %Other
1322/// %V = load i32* %P2
1323/// to:
1324/// %V1 = load i32* %Alloca -> will be mem2reg'd
1325/// %V2 = load i32* %Other
1326/// %V = select i1 %cond, i32 %V1, i32 %V2
1327///
1328/// We can do this to a select if its only uses are loads and if the operand
1329/// to the select can be loaded unconditionally.
1330static bool isSafeSelectToSpeculate(SelectInst &SI) {
1331 Value *TValue = SI.getTrueValue();
1332 Value *FValue = SI.getFalseValue();
1333 const DataLayout &DL = SI.getModule()->getDataLayout();
1334
1335 for (User *U : SI.users()) {
1336 LoadInst *LI = dyn_cast<LoadInst>(U);
1337 if (!LI || !LI->isSimple())
1338 return false;
1339
1340 // Both operands to the select need to be dereferenceable, either
1341 // absolutely (e.g. allocas) or at this point because we can see other
1342 // accesses to it.
1343 if (!isSafeToLoadUnconditionally(TValue, LI->getType(), LI->getAlignment(),
1344 DL, LI))
1345 return false;
1346 if (!isSafeToLoadUnconditionally(FValue, LI->getType(), LI->getAlignment(),
1347 DL, LI))
1348 return false;
1349 }
1350
1351 return true;
1352}
1353
1354static void speculateSelectInstLoads(SelectInst &SI) {
1355 LLVM_DEBUG(dbgs() << " original: " << SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << SI <<
"\n"; } } while (false)
;
1356
1357 IRBuilderTy IRB(&SI);
1358 Value *TV = SI.getTrueValue();
1359 Value *FV = SI.getFalseValue();
1360 // Replace the loads of the select with a select of two loads.
1361 while (!SI.use_empty()) {
1362 LoadInst *LI = cast<LoadInst>(SI.user_back());
1363 assert(LI->isSimple() && "We only speculate simple loads")((LI->isSimple() && "We only speculate simple loads"
) ? static_cast<void> (0) : __assert_fail ("LI->isSimple() && \"We only speculate simple loads\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1363, __PRETTY_FUNCTION__))
;
1364
1365 IRB.SetInsertPoint(LI);
1366 LoadInst *TL = IRB.CreateLoad(LI->getType(), TV,
1367 LI->getName() + ".sroa.speculate.load.true");
1368 LoadInst *FL = IRB.CreateLoad(LI->getType(), FV,
1369 LI->getName() + ".sroa.speculate.load.false");
1370 NumLoadsSpeculated += 2;
1371
1372 // Transfer alignment and AA info if present.
1373 TL->setAlignment(MaybeAlign(LI->getAlignment()));
1374 FL->setAlignment(MaybeAlign(LI->getAlignment()));
1375
1376 AAMDNodes Tags;
1377 LI->getAAMetadata(Tags);
1378 if (Tags) {
1379 TL->setAAMetadata(Tags);
1380 FL->setAAMetadata(Tags);
1381 }
1382
1383 Value *V = IRB.CreateSelect(SI.getCondition(), TL, FL,
1384 LI->getName() + ".sroa.speculated");
1385
1386 LLVM_DEBUG(dbgs() << " speculated to: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " speculated to: " <<
*V << "\n"; } } while (false)
;
1387 LI->replaceAllUsesWith(V);
1388 LI->eraseFromParent();
1389 }
1390 SI.eraseFromParent();
1391}
1392
1393/// Build a GEP out of a base pointer and indices.
1394///
1395/// This will return the BasePtr if that is valid, or build a new GEP
1396/// instruction using the IRBuilder if GEP-ing is needed.
1397static Value *buildGEP(IRBuilderTy &IRB, Value *BasePtr,
1398 SmallVectorImpl<Value *> &Indices, Twine NamePrefix) {
1399 if (Indices.empty())
1400 return BasePtr;
1401
1402 // A single zero index is a no-op, so check for this and avoid building a GEP
1403 // in that case.
1404 if (Indices.size() == 1 && cast<ConstantInt>(Indices.back())->isZero())
1405 return BasePtr;
1406
1407 return IRB.CreateInBoundsGEP(BasePtr->getType()->getPointerElementType(),
1408 BasePtr, Indices, NamePrefix + "sroa_idx");
1409}
1410
1411/// Get a natural GEP off of the BasePtr walking through Ty toward
1412/// TargetTy without changing the offset of the pointer.
1413///
1414/// This routine assumes we've already established a properly offset GEP with
1415/// Indices, and arrived at the Ty type. The goal is to continue to GEP with
1416/// zero-indices down through type layers until we find one the same as
1417/// TargetTy. If we can't find one with the same type, we at least try to use
1418/// one with the same size. If none of that works, we just produce the GEP as
1419/// indicated by Indices to have the correct offset.
1420static Value *getNaturalGEPWithType(IRBuilderTy &IRB, const DataLayout &DL,
1421 Value *BasePtr, Type *Ty, Type *TargetTy,
1422 SmallVectorImpl<Value *> &Indices,
1423 Twine NamePrefix) {
1424 if (Ty == TargetTy)
1425 return buildGEP(IRB, BasePtr, Indices, NamePrefix);
1426
1427 // Offset size to use for the indices.
1428 unsigned OffsetSize = DL.getIndexTypeSizeInBits(BasePtr->getType());
1429
1430 // See if we can descend into a struct and locate a field with the correct
1431 // type.
1432 unsigned NumLayers = 0;
1433 Type *ElementTy = Ty;
1434 do {
1435 if (ElementTy->isPointerTy())
1436 break;
1437
1438 if (ArrayType *ArrayTy = dyn_cast<ArrayType>(ElementTy)) {
1439 ElementTy = ArrayTy->getElementType();
1440 Indices.push_back(IRB.getIntN(OffsetSize, 0));
1441 } else if (VectorType *VectorTy = dyn_cast<VectorType>(ElementTy)) {
1442 ElementTy = VectorTy->getElementType();
1443 Indices.push_back(IRB.getInt32(0));
1444 } else if (StructType *STy = dyn_cast<StructType>(ElementTy)) {
1445 if (STy->element_begin() == STy->element_end())
1446 break; // Nothing left to descend into.
1447 ElementTy = *STy->element_begin();
1448 Indices.push_back(IRB.getInt32(0));
1449 } else {
1450 break;
1451 }
1452 ++NumLayers;
1453 } while (ElementTy != TargetTy);
1454 if (ElementTy != TargetTy)
1455 Indices.erase(Indices.end() - NumLayers, Indices.end());
1456
1457 return buildGEP(IRB, BasePtr, Indices, NamePrefix);
1458}
1459
1460/// Recursively compute indices for a natural GEP.
1461///
1462/// This is the recursive step for getNaturalGEPWithOffset that walks down the
1463/// element types adding appropriate indices for the GEP.
1464static Value *getNaturalGEPRecursively(IRBuilderTy &IRB, const DataLayout &DL,
1465 Value *Ptr, Type *Ty, APInt &Offset,
1466 Type *TargetTy,
1467 SmallVectorImpl<Value *> &Indices,
1468 Twine NamePrefix) {
1469 if (Offset == 0)
1470 return getNaturalGEPWithType(IRB, DL, Ptr, Ty, TargetTy, Indices,
1471 NamePrefix);
1472
1473 // We can't recurse through pointer types.
1474 if (Ty->isPointerTy())
1475 return nullptr;
1476
1477 // We try to analyze GEPs over vectors here, but note that these GEPs are
1478 // extremely poorly defined currently. The long-term goal is to remove GEPing
1479 // over a vector from the IR completely.
1480 if (VectorType *VecTy = dyn_cast<VectorType>(Ty)) {
1481 unsigned ElementSizeInBits = DL.getTypeSizeInBits(VecTy->getScalarType());
1482 if (ElementSizeInBits % 8 != 0) {
1483 // GEPs over non-multiple of 8 size vector elements are invalid.
1484 return nullptr;
1485 }
1486 APInt ElementSize(Offset.getBitWidth(), ElementSizeInBits / 8);
1487 APInt NumSkippedElements = Offset.sdiv(ElementSize);
1488 if (NumSkippedElements.ugt(VecTy->getNumElements()))
1489 return nullptr;
1490 Offset -= NumSkippedElements * ElementSize;
1491 Indices.push_back(IRB.getInt(NumSkippedElements));
1492 return getNaturalGEPRecursively(IRB, DL, Ptr, VecTy->getElementType(),
1493 Offset, TargetTy, Indices, NamePrefix);
1494 }
1495
1496 if (ArrayType *ArrTy = dyn_cast<ArrayType>(Ty)) {
1497 Type *ElementTy = ArrTy->getElementType();
1498 APInt ElementSize(Offset.getBitWidth(), DL.getTypeAllocSize(ElementTy));
1499 APInt NumSkippedElements = Offset.sdiv(ElementSize);
1500 if (NumSkippedElements.ugt(ArrTy->getNumElements()))
1501 return nullptr;
1502
1503 Offset -= NumSkippedElements * ElementSize;
1504 Indices.push_back(IRB.getInt(NumSkippedElements));
1505 return getNaturalGEPRecursively(IRB, DL, Ptr, ElementTy, Offset, TargetTy,
1506 Indices, NamePrefix);
1507 }
1508
1509 StructType *STy = dyn_cast<StructType>(Ty);
1510 if (!STy)
1511 return nullptr;
1512
1513 const StructLayout *SL = DL.getStructLayout(STy);
1514 uint64_t StructOffset = Offset.getZExtValue();
1515 if (StructOffset >= SL->getSizeInBytes())
1516 return nullptr;
1517 unsigned Index = SL->getElementContainingOffset(StructOffset);
1518 Offset -= APInt(Offset.getBitWidth(), SL->getElementOffset(Index));
1519 Type *ElementTy = STy->getElementType(Index);
1520 if (Offset.uge(DL.getTypeAllocSize(ElementTy)))
1521 return nullptr; // The offset points into alignment padding.
1522
1523 Indices.push_back(IRB.getInt32(Index));
1524 return getNaturalGEPRecursively(IRB, DL, Ptr, ElementTy, Offset, TargetTy,
1525 Indices, NamePrefix);
1526}
1527
1528/// Get a natural GEP from a base pointer to a particular offset and
1529/// resulting in a particular type.
1530///
1531/// The goal is to produce a "natural" looking GEP that works with the existing
1532/// composite types to arrive at the appropriate offset and element type for
1533/// a pointer. TargetTy is the element type the returned GEP should point-to if
1534/// possible. We recurse by decreasing Offset, adding the appropriate index to
1535/// Indices, and setting Ty to the result subtype.
1536///
1537/// If no natural GEP can be constructed, this function returns null.
1538static Value *getNaturalGEPWithOffset(IRBuilderTy &IRB, const DataLayout &DL,
1539 Value *Ptr, APInt Offset, Type *TargetTy,
1540 SmallVectorImpl<Value *> &Indices,
1541 Twine NamePrefix) {
1542 PointerType *Ty = cast<PointerType>(Ptr->getType());
1543
1544 // Don't consider any GEPs through an i8* as natural unless the TargetTy is
1545 // an i8.
1546 if (Ty == IRB.getInt8PtrTy(Ty->getAddressSpace()) && TargetTy->isIntegerTy(8))
1547 return nullptr;
1548
1549 Type *ElementTy = Ty->getElementType();
1550 if (!ElementTy->isSized())
1551 return nullptr; // We can't GEP through an unsized element.
1552 APInt ElementSize(Offset.getBitWidth(), DL.getTypeAllocSize(ElementTy));
1553 if (ElementSize == 0)
1554 return nullptr; // Zero-length arrays can't help us build a natural GEP.
1555 APInt NumSkippedElements = Offset.sdiv(ElementSize);
1556
1557 Offset -= NumSkippedElements * ElementSize;
1558 Indices.push_back(IRB.getInt(NumSkippedElements));
1559 return getNaturalGEPRecursively(IRB, DL, Ptr, ElementTy, Offset, TargetTy,
1560 Indices, NamePrefix);
1561}
1562
1563/// Compute an adjusted pointer from Ptr by Offset bytes where the
1564/// resulting pointer has PointerTy.
1565///
1566/// This tries very hard to compute a "natural" GEP which arrives at the offset
1567/// and produces the pointer type desired. Where it cannot, it will try to use
1568/// the natural GEP to arrive at the offset and bitcast to the type. Where that
1569/// fails, it will try to use an existing i8* and GEP to the byte offset and
1570/// bitcast to the type.
1571///
1572/// The strategy for finding the more natural GEPs is to peel off layers of the
1573/// pointer, walking back through bit casts and GEPs, searching for a base
1574/// pointer from which we can compute a natural GEP with the desired
1575/// properties. The algorithm tries to fold as many constant indices into
1576/// a single GEP as possible, thus making each GEP more independent of the
1577/// surrounding code.
1578static Value *getAdjustedPtr(IRBuilderTy &IRB, const DataLayout &DL, Value *Ptr,
1579 APInt Offset, Type *PointerTy, Twine NamePrefix) {
1580 // Even though we don't look through PHI nodes, we could be called on an
1581 // instruction in an unreachable block, which may be on a cycle.
1582 SmallPtrSet<Value *, 4> Visited;
1583 Visited.insert(Ptr);
1584 SmallVector<Value *, 4> Indices;
1585
1586 // We may end up computing an offset pointer that has the wrong type. If we
1587 // never are able to compute one directly that has the correct type, we'll
1588 // fall back to it, so keep it and the base it was computed from around here.
1589 Value *OffsetPtr = nullptr;
1590 Value *OffsetBasePtr;
1591
1592 // Remember any i8 pointer we come across to re-use if we need to do a raw
1593 // byte offset.
1594 Value *Int8Ptr = nullptr;
1595 APInt Int8PtrOffset(Offset.getBitWidth(), 0);
1596
1597 PointerType *TargetPtrTy = cast<PointerType>(PointerTy);
26
'PointerTy' is a 'PointerType'
1598 Type *TargetTy = TargetPtrTy->getElementType();
1599
1600 // As `addrspacecast` is , `Ptr` (the storage pointer) may have different
1601 // address space from the expected `PointerTy` (the pointer to be used).
1602 // Adjust the pointer type based the original storage pointer.
1603 auto AS = cast<PointerType>(Ptr->getType())->getAddressSpace();
27
The object is a 'PointerType'
1604 PointerTy = TargetTy->getPointerTo(AS);
1605
1606 do {
1607 // First fold any existing GEPs into the offset.
1608 while (GEPOperator *GEP = dyn_cast<GEPOperator>(Ptr)) {
28
Assuming 'Ptr' is not a 'GEPOperator'
29
Loop condition is false. Execution continues on line 1619
1609 APInt GEPOffset(Offset.getBitWidth(), 0);
1610 if (!GEP->accumulateConstantOffset(DL, GEPOffset))
1611 break;
1612 Offset += GEPOffset;
1613 Ptr = GEP->getPointerOperand();
1614 if (!Visited.insert(Ptr).second)
1615 break;
1616 }
1617
1618 // See if we can perform a natural GEP here.
1619 Indices.clear();
1620 if (Value *P
29.1
'P' is null
29.1
'P' is null
29.1
'P' is null
= getNaturalGEPWithOffset(IRB, DL, Ptr, Offset, TargetTy,
30
Taking false branch
1621 Indices, NamePrefix)) {
1622 // If we have a new natural pointer at the offset, clear out any old
1623 // offset pointer we computed. Unless it is the base pointer or
1624 // a non-instruction, we built a GEP we don't need. Zap it.
1625 if (OffsetPtr && OffsetPtr != OffsetBasePtr)
1626 if (Instruction *I = dyn_cast<Instruction>(OffsetPtr)) {
1627 assert(I->use_empty() && "Built a GEP with uses some how!")((I->use_empty() && "Built a GEP with uses some how!"
) ? static_cast<void> (0) : __assert_fail ("I->use_empty() && \"Built a GEP with uses some how!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1627, __PRETTY_FUNCTION__))
;
1628 I->eraseFromParent();
1629 }
1630 OffsetPtr = P;
1631 OffsetBasePtr = Ptr;
1632 // If we also found a pointer of the right type, we're done.
1633 if (P->getType() == PointerTy)
1634 break;
1635 }
1636
1637 // Stash this pointer if we've found an i8*.
1638 if (Ptr->getType()->isIntegerTy(8)) {
31
Assuming the condition is false
32
Taking false branch
1639 Int8Ptr = Ptr;
1640 Int8PtrOffset = Offset;
1641 }
1642
1643 // Peel off a layer of the pointer and update the offset appropriately.
1644 if (Operator::getOpcode(Ptr) == Instruction::BitCast) {
33
Taking false branch
1645 Ptr = cast<Operator>(Ptr)->getOperand(0);
1646 } else if (GlobalAlias *GA
34.1
'GA' is null
34.1
'GA' is null
34.1
'GA' is null
= dyn_cast<GlobalAlias>(Ptr)) {
34
Assuming 'Ptr' is not a 'GlobalAlias'
35
Taking false branch
1647 if (GA->isInterposable())
1648 break;
1649 Ptr = GA->getAliasee();
1650 } else {
1651 break;
36
Execution continues on line 1656
1652 }
1653 assert(Ptr->getType()->isPointerTy() && "Unexpected operand type!")((Ptr->getType()->isPointerTy() && "Unexpected operand type!"
) ? static_cast<void> (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Unexpected operand type!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1653, __PRETTY_FUNCTION__))
;
1654 } while (Visited.insert(Ptr).second);
1655
1656 if (!OffsetPtr
36.1
'OffsetPtr' is null
36.1
'OffsetPtr' is null
36.1
'OffsetPtr' is null
) {
37
Taking true branch
1657 if (!Int8Ptr
37.1
'Int8Ptr' is null
37.1
'Int8Ptr' is null
37.1
'Int8Ptr' is null
) {
38
Taking true branch
1658 Int8Ptr = IRB.CreateBitCast(
1659 Ptr, IRB.getInt8PtrTy(PointerTy->getPointerAddressSpace()),
1660 NamePrefix + "sroa_raw_cast");
1661 Int8PtrOffset = Offset;
1662 }
1663
1664 OffsetPtr = Int8PtrOffset == 0
39
'?' condition is false
1665 ? Int8Ptr
1666 : IRB.CreateInBoundsGEP(IRB.getInt8Ty(), Int8Ptr,
40
Calling 'IRBuilder::CreateInBoundsGEP'
54
Returning from 'IRBuilder::CreateInBoundsGEP'
1667 IRB.getInt(Int8PtrOffset),
1668 NamePrefix + "sroa_raw_idx");
1669 }
1670 Ptr = OffsetPtr;
1671
1672 // On the off chance we were targeting i8*, guard the bitcast here.
1673 if (cast<PointerType>(Ptr->getType()) != TargetPtrTy) {
55
The object is a 'PointerType'
56
Assuming the condition is false
57
Taking false branch
1674 Ptr = IRB.CreatePointerBitCastOrAddrSpaceCast(Ptr,
1675 TargetPtrTy,
1676 NamePrefix + "sroa_cast");
1677 }
1678
1679 return Ptr;
1680}
1681
1682/// Compute the adjusted alignment for a load or store from an offset.
1683static unsigned getAdjustedAlignment(Instruction *I, uint64_t Offset,
1684 const DataLayout &DL) {
1685 unsigned Alignment;
1686 Type *Ty;
1687 if (auto *LI = dyn_cast<LoadInst>(I)) {
1688 Alignment = LI->getAlignment();
1689 Ty = LI->getType();
1690 } else if (auto *SI = dyn_cast<StoreInst>(I)) {
1691 Alignment = SI->getAlignment();
1692 Ty = SI->getValueOperand()->getType();
1693 } else {
1694 llvm_unreachable("Only loads and stores are allowed!")::llvm::llvm_unreachable_internal("Only loads and stores are allowed!"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1694)
;
1695 }
1696
1697 if (!Alignment)
1698 Alignment = DL.getABITypeAlignment(Ty);
1699
1700 return MinAlign(Alignment, Offset);
1701}
1702
1703/// Test whether we can convert a value from the old to the new type.
1704///
1705/// This predicate should be used to guard calls to convertValue in order to
1706/// ensure that we only try to convert viable values. The strategy is that we
1707/// will peel off single element struct and array wrappings to get to an
1708/// underlying value, and convert that value.
1709static bool canConvertValue(const DataLayout &DL, Type *OldTy, Type *NewTy) {
1710 if (OldTy == NewTy)
1711 return true;
1712
1713 // For integer types, we can't handle any bit-width differences. This would
1714 // break both vector conversions with extension and introduce endianness
1715 // issues when in conjunction with loads and stores.
1716 if (isa<IntegerType>(OldTy) && isa<IntegerType>(NewTy)) {
1717 assert(cast<IntegerType>(OldTy)->getBitWidth() !=((cast<IntegerType>(OldTy)->getBitWidth() != cast<
IntegerType>(NewTy)->getBitWidth() && "We can't have the same bitwidth for different int types"
) ? static_cast<void> (0) : __assert_fail ("cast<IntegerType>(OldTy)->getBitWidth() != cast<IntegerType>(NewTy)->getBitWidth() && \"We can't have the same bitwidth for different int types\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1719, __PRETTY_FUNCTION__))
1718 cast<IntegerType>(NewTy)->getBitWidth() &&((cast<IntegerType>(OldTy)->getBitWidth() != cast<
IntegerType>(NewTy)->getBitWidth() && "We can't have the same bitwidth for different int types"
) ? static_cast<void> (0) : __assert_fail ("cast<IntegerType>(OldTy)->getBitWidth() != cast<IntegerType>(NewTy)->getBitWidth() && \"We can't have the same bitwidth for different int types\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1719, __PRETTY_FUNCTION__))
1719 "We can't have the same bitwidth for different int types")((cast<IntegerType>(OldTy)->getBitWidth() != cast<
IntegerType>(NewTy)->getBitWidth() && "We can't have the same bitwidth for different int types"
) ? static_cast<void> (0) : __assert_fail ("cast<IntegerType>(OldTy)->getBitWidth() != cast<IntegerType>(NewTy)->getBitWidth() && \"We can't have the same bitwidth for different int types\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1719, __PRETTY_FUNCTION__))
;
1720 return false;
1721 }
1722
1723 if (DL.getTypeSizeInBits(NewTy) != DL.getTypeSizeInBits(OldTy))
1724 return false;
1725 if (!NewTy->isSingleValueType() || !OldTy->isSingleValueType())
1726 return false;
1727
1728 // We can convert pointers to integers and vice-versa. Same for vectors
1729 // of pointers and integers.
1730 OldTy = OldTy->getScalarType();
1731 NewTy = NewTy->getScalarType();
1732 if (NewTy->isPointerTy() || OldTy->isPointerTy()) {
1733 if (NewTy->isPointerTy() && OldTy->isPointerTy()) {
1734 return cast<PointerType>(NewTy)->getPointerAddressSpace() ==
1735 cast<PointerType>(OldTy)->getPointerAddressSpace();
1736 }
1737
1738 // We can convert integers to integral pointers, but not to non-integral
1739 // pointers.
1740 if (OldTy->isIntegerTy())
1741 return !DL.isNonIntegralPointerType(NewTy);
1742
1743 // We can convert integral pointers to integers, but non-integral pointers
1744 // need to remain pointers.
1745 if (!DL.isNonIntegralPointerType(OldTy))
1746 return NewTy->isIntegerTy();
1747
1748 return false;
1749 }
1750
1751 return true;
1752}
1753
1754/// Generic routine to convert an SSA value to a value of a different
1755/// type.
1756///
1757/// This will try various different casting techniques, such as bitcasts,
1758/// inttoptr, and ptrtoint casts. Use the \c canConvertValue predicate to test
1759/// two types for viability with this routine.
1760static Value *convertValue(const DataLayout &DL, IRBuilderTy &IRB, Value *V,
1761 Type *NewTy) {
1762 Type *OldTy = V->getType();
1763 assert(canConvertValue(DL, OldTy, NewTy) && "Value not convertable to type")((canConvertValue(DL, OldTy, NewTy) && "Value not convertable to type"
) ? static_cast<void> (0) : __assert_fail ("canConvertValue(DL, OldTy, NewTy) && \"Value not convertable to type\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1763, __PRETTY_FUNCTION__))
;
1764
1765 if (OldTy == NewTy)
1766 return V;
1767
1768 assert(!(isa<IntegerType>(OldTy) && isa<IntegerType>(NewTy)) &&((!(isa<IntegerType>(OldTy) && isa<IntegerType
>(NewTy)) && "Integer types must be the exact same to convert."
) ? static_cast<void> (0) : __assert_fail ("!(isa<IntegerType>(OldTy) && isa<IntegerType>(NewTy)) && \"Integer types must be the exact same to convert.\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1769, __PRETTY_FUNCTION__))
1769 "Integer types must be the exact same to convert.")((!(isa<IntegerType>(OldTy) && isa<IntegerType
>(NewTy)) && "Integer types must be the exact same to convert."
) ? static_cast<void> (0) : __assert_fail ("!(isa<IntegerType>(OldTy) && isa<IntegerType>(NewTy)) && \"Integer types must be the exact same to convert.\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1769, __PRETTY_FUNCTION__))
;
1770
1771 // See if we need inttoptr for this type pair. A cast involving both scalars
1772 // and vectors requires and additional bitcast.
1773 if (OldTy->isIntOrIntVectorTy() && NewTy->isPtrOrPtrVectorTy()) {
1774 // Expand <2 x i32> to i8* --> <2 x i32> to i64 to i8*
1775 if (OldTy->isVectorTy() && !NewTy->isVectorTy())
1776 return IRB.CreateIntToPtr(IRB.CreateBitCast(V, DL.getIntPtrType(NewTy)),
1777 NewTy);
1778
1779 // Expand i128 to <2 x i8*> --> i128 to <2 x i64> to <2 x i8*>
1780 if (!OldTy->isVectorTy() && NewTy->isVectorTy())
1781 return IRB.CreateIntToPtr(IRB.CreateBitCast(V, DL.getIntPtrType(NewTy)),
1782 NewTy);
1783
1784 return IRB.CreateIntToPtr(V, NewTy);
1785 }
1786
1787 // See if we need ptrtoint for this type pair. A cast involving both scalars
1788 // and vectors requires and additional bitcast.
1789 if (OldTy->isPtrOrPtrVectorTy() && NewTy->isIntOrIntVectorTy()) {
1790 // Expand <2 x i8*> to i128 --> <2 x i8*> to <2 x i64> to i128
1791 if (OldTy->isVectorTy() && !NewTy->isVectorTy())
1792 return IRB.CreateBitCast(IRB.CreatePtrToInt(V, DL.getIntPtrType(OldTy)),
1793 NewTy);
1794
1795 // Expand i8* to <2 x i32> --> i8* to i64 to <2 x i32>
1796 if (!OldTy->isVectorTy() && NewTy->isVectorTy())
1797 return IRB.CreateBitCast(IRB.CreatePtrToInt(V, DL.getIntPtrType(OldTy)),
1798 NewTy);
1799
1800 return IRB.CreatePtrToInt(V, NewTy);
1801 }
1802
1803 return IRB.CreateBitCast(V, NewTy);
1804}
1805
1806/// Test whether the given slice use can be promoted to a vector.
1807///
1808/// This function is called to test each entry in a partition which is slated
1809/// for a single slice.
1810static bool isVectorPromotionViableForSlice(Partition &P, const Slice &S,
1811 VectorType *Ty,
1812 uint64_t ElementSize,
1813 const DataLayout &DL) {
1814 // First validate the slice offsets.
1815 uint64_t BeginOffset =
1816 std::max(S.beginOffset(), P.beginOffset()) - P.beginOffset();
1817 uint64_t BeginIndex = BeginOffset / ElementSize;
1818 if (BeginIndex * ElementSize != BeginOffset ||
1819 BeginIndex >= Ty->getNumElements())
1820 return false;
1821 uint64_t EndOffset =
1822 std::min(S.endOffset(), P.endOffset()) - P.beginOffset();
1823 uint64_t EndIndex = EndOffset / ElementSize;
1824 if (EndIndex * ElementSize != EndOffset || EndIndex > Ty->getNumElements())
1825 return false;
1826
1827 assert(EndIndex > BeginIndex && "Empty vector!")((EndIndex > BeginIndex && "Empty vector!") ? static_cast
<void> (0) : __assert_fail ("EndIndex > BeginIndex && \"Empty vector!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1827, __PRETTY_FUNCTION__))
;
1828 uint64_t NumElements = EndIndex - BeginIndex;
1829 Type *SliceTy = (NumElements == 1)
1830 ? Ty->getElementType()
1831 : VectorType::get(Ty->getElementType(), NumElements);
1832
1833 Type *SplitIntTy =
1834 Type::getIntNTy(Ty->getContext(), NumElements * ElementSize * 8);
1835
1836 Use *U = S.getUse();
1837
1838 if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(U->getUser())) {
1839 if (MI->isVolatile())
1840 return false;
1841 if (!S.isSplittable())
1842 return false; // Skip any unsplittable intrinsics.
1843 } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U->getUser())) {
1844 if (!II->isLifetimeStartOrEnd())
1845 return false;
1846 } else if (U->get()->getType()->getPointerElementType()->isStructTy()) {
1847 // Disable vector promotion when there are loads or stores of an FCA.
1848 return false;
1849 } else if (LoadInst *LI = dyn_cast<LoadInst>(U->getUser())) {
1850 if (LI->isVolatile())
1851 return false;
1852 Type *LTy = LI->getType();
1853 if (P.beginOffset() > S.beginOffset() || P.endOffset() < S.endOffset()) {
1854 assert(LTy->isIntegerTy())((LTy->isIntegerTy()) ? static_cast<void> (0) : __assert_fail
("LTy->isIntegerTy()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1854, __PRETTY_FUNCTION__))
;
1855 LTy = SplitIntTy;
1856 }
1857 if (!canConvertValue(DL, SliceTy, LTy))
1858 return false;
1859 } else if (StoreInst *SI = dyn_cast<StoreInst>(U->getUser())) {
1860 if (SI->isVolatile())
1861 return false;
1862 Type *STy = SI->getValueOperand()->getType();
1863 if (P.beginOffset() > S.beginOffset() || P.endOffset() < S.endOffset()) {
1864 assert(STy->isIntegerTy())((STy->isIntegerTy()) ? static_cast<void> (0) : __assert_fail
("STy->isIntegerTy()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1864, __PRETTY_FUNCTION__))
;
1865 STy = SplitIntTy;
1866 }
1867 if (!canConvertValue(DL, STy, SliceTy))
1868 return false;
1869 } else {
1870 return false;
1871 }
1872
1873 return true;
1874}
1875
1876/// Test whether the given alloca partitioning and range of slices can be
1877/// promoted to a vector.
1878///
1879/// This is a quick test to check whether we can rewrite a particular alloca
1880/// partition (and its newly formed alloca) into a vector alloca with only
1881/// whole-vector loads and stores such that it could be promoted to a vector
1882/// SSA value. We only can ensure this for a limited set of operations, and we
1883/// don't want to do the rewrites unless we are confident that the result will
1884/// be promotable, so we have an early test here.
1885static VectorType *isVectorPromotionViable(Partition &P, const DataLayout &DL) {
1886 // Collect the candidate types for vector-based promotion. Also track whether
1887 // we have different element types.
1888 SmallVector<VectorType *, 4> CandidateTys;
1889 Type *CommonEltTy = nullptr;
1890 bool HaveCommonEltTy = true;
1891 auto CheckCandidateType = [&](Type *Ty) {
1892 if (auto *VTy = dyn_cast<VectorType>(Ty)) {
1893 // Return if bitcast to vectors is different for total size in bits.
1894 if (!CandidateTys.empty()) {
1895 VectorType *V = CandidateTys[0];
1896 if (DL.getTypeSizeInBits(VTy) != DL.getTypeSizeInBits(V)) {
1897 CandidateTys.clear();
1898 return;
1899 }
1900 }
1901 CandidateTys.push_back(VTy);
1902 if (!CommonEltTy)
1903 CommonEltTy = VTy->getElementType();
1904 else if (CommonEltTy != VTy->getElementType())
1905 HaveCommonEltTy = false;
1906 }
1907 };
1908 // Consider any loads or stores that are the exact size of the slice.
1909 for (const Slice &S : P)
1910 if (S.beginOffset() == P.beginOffset() &&
1911 S.endOffset() == P.endOffset()) {
1912 if (auto *LI = dyn_cast<LoadInst>(S.getUse()->getUser()))
1913 CheckCandidateType(LI->getType());
1914 else if (auto *SI = dyn_cast<StoreInst>(S.getUse()->getUser()))
1915 CheckCandidateType(SI->getValueOperand()->getType());
1916 }
1917
1918 // If we didn't find a vector type, nothing to do here.
1919 if (CandidateTys.empty())
1920 return nullptr;
1921
1922 // Remove non-integer vector types if we had multiple common element types.
1923 // FIXME: It'd be nice to replace them with integer vector types, but we can't
1924 // do that until all the backends are known to produce good code for all
1925 // integer vector types.
1926 if (!HaveCommonEltTy) {
1927 CandidateTys.erase(
1928 llvm::remove_if(CandidateTys,
1929 [](VectorType *VTy) {
1930 return !VTy->getElementType()->isIntegerTy();
1931 }),
1932 CandidateTys.end());
1933
1934 // If there were no integer vector types, give up.
1935 if (CandidateTys.empty())
1936 return nullptr;
1937
1938 // Rank the remaining candidate vector types. This is easy because we know
1939 // they're all integer vectors. We sort by ascending number of elements.
1940 auto RankVectorTypes = [&DL](VectorType *RHSTy, VectorType *LHSTy) {
1941 (void)DL;
1942 assert(DL.getTypeSizeInBits(RHSTy) == DL.getTypeSizeInBits(LHSTy) &&((DL.getTypeSizeInBits(RHSTy) == DL.getTypeSizeInBits(LHSTy) &&
"Cannot have vector types of different sizes!") ? static_cast
<void> (0) : __assert_fail ("DL.getTypeSizeInBits(RHSTy) == DL.getTypeSizeInBits(LHSTy) && \"Cannot have vector types of different sizes!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1943, __PRETTY_FUNCTION__))
1943 "Cannot have vector types of different sizes!")((DL.getTypeSizeInBits(RHSTy) == DL.getTypeSizeInBits(LHSTy) &&
"Cannot have vector types of different sizes!") ? static_cast
<void> (0) : __assert_fail ("DL.getTypeSizeInBits(RHSTy) == DL.getTypeSizeInBits(LHSTy) && \"Cannot have vector types of different sizes!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1943, __PRETTY_FUNCTION__))
;
1944 assert(RHSTy->getElementType()->isIntegerTy() &&((RHSTy->getElementType()->isIntegerTy() && "All non-integer types eliminated!"
) ? static_cast<void> (0) : __assert_fail ("RHSTy->getElementType()->isIntegerTy() && \"All non-integer types eliminated!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1945, __PRETTY_FUNCTION__))
1945 "All non-integer types eliminated!")((RHSTy->getElementType()->isIntegerTy() && "All non-integer types eliminated!"
) ? static_cast<void> (0) : __assert_fail ("RHSTy->getElementType()->isIntegerTy() && \"All non-integer types eliminated!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1945, __PRETTY_FUNCTION__))
;
1946 assert(LHSTy->getElementType()->isIntegerTy() &&((LHSTy->getElementType()->isIntegerTy() && "All non-integer types eliminated!"
) ? static_cast<void> (0) : __assert_fail ("LHSTy->getElementType()->isIntegerTy() && \"All non-integer types eliminated!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1947, __PRETTY_FUNCTION__))
1947 "All non-integer types eliminated!")((LHSTy->getElementType()->isIntegerTy() && "All non-integer types eliminated!"
) ? static_cast<void> (0) : __assert_fail ("LHSTy->getElementType()->isIntegerTy() && \"All non-integer types eliminated!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1947, __PRETTY_FUNCTION__))
;
1948 return RHSTy->getNumElements() < LHSTy->getNumElements();
1949 };
1950 llvm::sort(CandidateTys, RankVectorTypes);
1951 CandidateTys.erase(
1952 std::unique(CandidateTys.begin(), CandidateTys.end(), RankVectorTypes),
1953 CandidateTys.end());
1954 } else {
1955// The only way to have the same element type in every vector type is to
1956// have the same vector type. Check that and remove all but one.
1957#ifndef NDEBUG
1958 for (VectorType *VTy : CandidateTys) {
1959 assert(VTy->getElementType() == CommonEltTy &&((VTy->getElementType() == CommonEltTy && "Unaccounted for element type!"
) ? static_cast<void> (0) : __assert_fail ("VTy->getElementType() == CommonEltTy && \"Unaccounted for element type!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1960, __PRETTY_FUNCTION__))
1960 "Unaccounted for element type!")((VTy->getElementType() == CommonEltTy && "Unaccounted for element type!"
) ? static_cast<void> (0) : __assert_fail ("VTy->getElementType() == CommonEltTy && \"Unaccounted for element type!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1960, __PRETTY_FUNCTION__))
;
1961 assert(VTy == CandidateTys[0] &&((VTy == CandidateTys[0] && "Different vector types with the same element type!"
) ? static_cast<void> (0) : __assert_fail ("VTy == CandidateTys[0] && \"Different vector types with the same element type!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1962, __PRETTY_FUNCTION__))
1962 "Different vector types with the same element type!")((VTy == CandidateTys[0] && "Different vector types with the same element type!"
) ? static_cast<void> (0) : __assert_fail ("VTy == CandidateTys[0] && \"Different vector types with the same element type!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1962, __PRETTY_FUNCTION__))
;
1963 }
1964#endif
1965 CandidateTys.resize(1);
1966 }
1967
1968 // Try each vector type, and return the one which works.
1969 auto CheckVectorTypeForPromotion = [&](VectorType *VTy) {
1970 uint64_t ElementSize = DL.getTypeSizeInBits(VTy->getElementType());
1971
1972 // While the definition of LLVM vectors is bitpacked, we don't support sizes
1973 // that aren't byte sized.
1974 if (ElementSize % 8)
1975 return false;
1976 assert((DL.getTypeSizeInBits(VTy) % 8) == 0 &&(((DL.getTypeSizeInBits(VTy) % 8) == 0 && "vector size not a multiple of element size?"
) ? static_cast<void> (0) : __assert_fail ("(DL.getTypeSizeInBits(VTy) % 8) == 0 && \"vector size not a multiple of element size?\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1977, __PRETTY_FUNCTION__))
1977 "vector size not a multiple of element size?")(((DL.getTypeSizeInBits(VTy) % 8) == 0 && "vector size not a multiple of element size?"
) ? static_cast<void> (0) : __assert_fail ("(DL.getTypeSizeInBits(VTy) % 8) == 0 && \"vector size not a multiple of element size?\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 1977, __PRETTY_FUNCTION__))
;
1978 ElementSize /= 8;
1979
1980 for (const Slice &S : P)
1981 if (!isVectorPromotionViableForSlice(P, S, VTy, ElementSize, DL))
1982 return false;
1983
1984 for (const Slice *S : P.splitSliceTails())
1985 if (!isVectorPromotionViableForSlice(P, *S, VTy, ElementSize, DL))
1986 return false;
1987
1988 return true;
1989 };
1990 for (VectorType *VTy : CandidateTys)
1991 if (CheckVectorTypeForPromotion(VTy))
1992 return VTy;
1993
1994 return nullptr;
1995}
1996
1997/// Test whether a slice of an alloca is valid for integer widening.
1998///
1999/// This implements the necessary checking for the \c isIntegerWideningViable
2000/// test below on a single slice of the alloca.
2001static bool isIntegerWideningViableForSlice(const Slice &S,
2002 uint64_t AllocBeginOffset,
2003 Type *AllocaTy,
2004 const DataLayout &DL,
2005 bool &WholeAllocaOp) {
2006 uint64_t Size = DL.getTypeStoreSize(AllocaTy);
2007
2008 uint64_t RelBegin = S.beginOffset() - AllocBeginOffset;
2009 uint64_t RelEnd = S.endOffset() - AllocBeginOffset;
2010
2011 // We can't reasonably handle cases where the load or store extends past
2012 // the end of the alloca's type and into its padding.
2013 if (RelEnd > Size)
2014 return false;
2015
2016 Use *U = S.getUse();
2017
2018 if (LoadInst *LI = dyn_cast<LoadInst>(U->getUser())) {
2019 if (LI->isVolatile())
2020 return false;
2021 // We can't handle loads that extend past the allocated memory.
2022 if (DL.getTypeStoreSize(LI->getType()) > Size)
2023 return false;
2024 // So far, AllocaSliceRewriter does not support widening split slice tails
2025 // in rewriteIntegerLoad.
2026 if (S.beginOffset() < AllocBeginOffset)
2027 return false;
2028 // Note that we don't count vector loads or stores as whole-alloca
2029 // operations which enable integer widening because we would prefer to use
2030 // vector widening instead.
2031 if (!isa<VectorType>(LI->getType()) && RelBegin == 0 && RelEnd == Size)
2032 WholeAllocaOp = true;
2033 if (IntegerType *ITy = dyn_cast<IntegerType>(LI->getType())) {
2034 if (ITy->getBitWidth() < DL.getTypeStoreSizeInBits(ITy))
2035 return false;
2036 } else if (RelBegin != 0 || RelEnd != Size ||
2037 !canConvertValue(DL, AllocaTy, LI->getType())) {
2038 // Non-integer loads need to be convertible from the alloca type so that
2039 // they are promotable.
2040 return false;
2041 }
2042 } else if (StoreInst *SI = dyn_cast<StoreInst>(U->getUser())) {
2043 Type *ValueTy = SI->getValueOperand()->getType();
2044 if (SI->isVolatile())
2045 return false;
2046 // We can't handle stores that extend past the allocated memory.
2047 if (DL.getTypeStoreSize(ValueTy) > Size)
2048 return false;
2049 // So far, AllocaSliceRewriter does not support widening split slice tails
2050 // in rewriteIntegerStore.
2051 if (S.beginOffset() < AllocBeginOffset)
2052 return false;
2053 // Note that we don't count vector loads or stores as whole-alloca
2054 // operations which enable integer widening because we would prefer to use
2055 // vector widening instead.
2056 if (!isa<VectorType>(ValueTy) && RelBegin == 0 && RelEnd == Size)
2057 WholeAllocaOp = true;
2058 if (IntegerType *ITy = dyn_cast<IntegerType>(ValueTy)) {
2059 if (ITy->getBitWidth() < DL.getTypeStoreSizeInBits(ITy))
2060 return false;
2061 } else if (RelBegin != 0 || RelEnd != Size ||
2062 !canConvertValue(DL, ValueTy, AllocaTy)) {
2063 // Non-integer stores need to be convertible to the alloca type so that
2064 // they are promotable.
2065 return false;
2066 }
2067 } else if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(U->getUser())) {
2068 if (MI->isVolatile() || !isa<Constant>(MI->getLength()))
2069 return false;
2070 if (!S.isSplittable())
2071 return false; // Skip any unsplittable intrinsics.
2072 } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U->getUser())) {
2073 if (!II->isLifetimeStartOrEnd())
2074 return false;
2075 } else {
2076 return false;
2077 }
2078
2079 return true;
2080}
2081
2082/// Test whether the given alloca partition's integer operations can be
2083/// widened to promotable ones.
2084///
2085/// This is a quick test to check whether we can rewrite the integer loads and
2086/// stores to a particular alloca into wider loads and stores and be able to
2087/// promote the resulting alloca.
2088static bool isIntegerWideningViable(Partition &P, Type *AllocaTy,
2089 const DataLayout &DL) {
2090 uint64_t SizeInBits = DL.getTypeSizeInBits(AllocaTy);
2091 // Don't create integer types larger than the maximum bitwidth.
2092 if (SizeInBits > IntegerType::MAX_INT_BITS)
2093 return false;
2094
2095 // Don't try to handle allocas with bit-padding.
2096 if (SizeInBits != DL.getTypeStoreSizeInBits(AllocaTy))
2097 return false;
2098
2099 // We need to ensure that an integer type with the appropriate bitwidth can
2100 // be converted to the alloca type, whatever that is. We don't want to force
2101 // the alloca itself to have an integer type if there is a more suitable one.
2102 Type *IntTy = Type::getIntNTy(AllocaTy->getContext(), SizeInBits);
2103 if (!canConvertValue(DL, AllocaTy, IntTy) ||
2104 !canConvertValue(DL, IntTy, AllocaTy))
2105 return false;
2106
2107 // While examining uses, we ensure that the alloca has a covering load or
2108 // store. We don't want to widen the integer operations only to fail to
2109 // promote due to some other unsplittable entry (which we may make splittable
2110 // later). However, if there are only splittable uses, go ahead and assume
2111 // that we cover the alloca.
2112 // FIXME: We shouldn't consider split slices that happen to start in the
2113 // partition here...
2114 bool WholeAllocaOp =
2115 P.begin() != P.end() ? false : DL.isLegalInteger(SizeInBits);
2116
2117 for (const Slice &S : P)
2118 if (!isIntegerWideningViableForSlice(S, P.beginOffset(), AllocaTy, DL,
2119 WholeAllocaOp))
2120 return false;
2121
2122 for (const Slice *S : P.splitSliceTails())
2123 if (!isIntegerWideningViableForSlice(*S, P.beginOffset(), AllocaTy, DL,
2124 WholeAllocaOp))
2125 return false;
2126
2127 return WholeAllocaOp;
2128}
2129
2130static Value *extractInteger(const DataLayout &DL, IRBuilderTy &IRB, Value *V,
2131 IntegerType *Ty, uint64_t Offset,
2132 const Twine &Name) {
2133 LLVM_DEBUG(dbgs() << " start: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " start: " << *V <<
"\n"; } } while (false)
;
65
Assuming 'DebugFlag' is false
66
Loop condition is false. Exiting loop
2134 IntegerType *IntTy = cast<IntegerType>(V->getType());
67
The object is a 'IntegerType'
2135 assert(DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) &&((DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(
IntTy) && "Element extends past full value") ? static_cast
<void> (0) : __assert_fail ("DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) && \"Element extends past full value\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2136, __PRETTY_FUNCTION__))
68
Assuming the condition is true
69
'?' condition is true
2136 "Element extends past full value")((DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(
IntTy) && "Element extends past full value") ? static_cast
<void> (0) : __assert_fail ("DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) && \"Element extends past full value\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2136, __PRETTY_FUNCTION__))
;
2137 uint64_t ShAmt = 8 * Offset;
2138 if (DL.isBigEndian())
70
Assuming the condition is false
71
Taking false branch
2139 ShAmt = 8 * (DL.getTypeStoreSize(IntTy) - DL.getTypeStoreSize(Ty) - Offset);
2140 if (ShAmt) {
72
Assuming 'ShAmt' is 0
73
Taking false branch
2141 V = IRB.CreateLShr(V, ShAmt, Name + ".shift");
2142 LLVM_DEBUG(dbgs() << " shifted: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " shifted: " << *V <<
"\n"; } } while (false)
;
2143 }
2144 assert(Ty->getBitWidth() <= IntTy->getBitWidth() &&((Ty->getBitWidth() <= IntTy->getBitWidth() &&
"Cannot extract to a larger integer!") ? static_cast<void
> (0) : __assert_fail ("Ty->getBitWidth() <= IntTy->getBitWidth() && \"Cannot extract to a larger integer!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2145, __PRETTY_FUNCTION__))
74
Called C++ object pointer is null
2145 "Cannot extract to a larger integer!")((Ty->getBitWidth() <= IntTy->getBitWidth() &&
"Cannot extract to a larger integer!") ? static_cast<void
> (0) : __assert_fail ("Ty->getBitWidth() <= IntTy->getBitWidth() && \"Cannot extract to a larger integer!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2145, __PRETTY_FUNCTION__))
;
2146 if (Ty != IntTy) {
2147 V = IRB.CreateTrunc(V, Ty, Name + ".trunc");
2148 LLVM_DEBUG(dbgs() << " trunced: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " trunced: " << *V <<
"\n"; } } while (false)
;
2149 }
2150 return V;
2151}
2152
2153static Value *insertInteger(const DataLayout &DL, IRBuilderTy &IRB, Value *Old,
2154 Value *V, uint64_t Offset, const Twine &Name) {
2155 IntegerType *IntTy = cast<IntegerType>(Old->getType());
2156 IntegerType *Ty = cast<IntegerType>(V->getType());
2157 assert(Ty->getBitWidth() <= IntTy->getBitWidth() &&((Ty->getBitWidth() <= IntTy->getBitWidth() &&
"Cannot insert a larger integer!") ? static_cast<void>
(0) : __assert_fail ("Ty->getBitWidth() <= IntTy->getBitWidth() && \"Cannot insert a larger integer!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2158, __PRETTY_FUNCTION__))
2158 "Cannot insert a larger integer!")((Ty->getBitWidth() <= IntTy->getBitWidth() &&
"Cannot insert a larger integer!") ? static_cast<void>
(0) : __assert_fail ("Ty->getBitWidth() <= IntTy->getBitWidth() && \"Cannot insert a larger integer!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2158, __PRETTY_FUNCTION__))
;
2159 LLVM_DEBUG(dbgs() << " start: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " start: " << *V <<
"\n"; } } while (false)
;
2160 if (Ty != IntTy) {
2161 V = IRB.CreateZExt(V, IntTy, Name + ".ext");
2162 LLVM_DEBUG(dbgs() << " extended: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " extended: " << *V <<
"\n"; } } while (false)
;
2163 }
2164 assert(DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) &&((DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(
IntTy) && "Element store outside of alloca store") ? static_cast
<void> (0) : __assert_fail ("DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) && \"Element store outside of alloca store\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2165, __PRETTY_FUNCTION__))
2165 "Element store outside of alloca store")((DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(
IntTy) && "Element store outside of alloca store") ? static_cast
<void> (0) : __assert_fail ("DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) && \"Element store outside of alloca store\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2165, __PRETTY_FUNCTION__))
;
2166 uint64_t ShAmt = 8 * Offset;
2167 if (DL.isBigEndian())
2168 ShAmt = 8 * (DL.getTypeStoreSize(IntTy) - DL.getTypeStoreSize(Ty) - Offset);
2169 if (ShAmt) {
2170 V = IRB.CreateShl(V, ShAmt, Name + ".shift");
2171 LLVM_DEBUG(dbgs() << " shifted: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " shifted: " << *V <<
"\n"; } } while (false)
;
2172 }
2173
2174 if (ShAmt || Ty->getBitWidth() < IntTy->getBitWidth()) {
2175 APInt Mask = ~Ty->getMask().zext(IntTy->getBitWidth()).shl(ShAmt);
2176 Old = IRB.CreateAnd(Old, Mask, Name + ".mask");
2177 LLVM_DEBUG(dbgs() << " masked: " << *Old << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " masked: " << *Old <<
"\n"; } } while (false)
;
2178 V = IRB.CreateOr(Old, V, Name + ".insert");
2179 LLVM_DEBUG(dbgs() << " inserted: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " inserted: " << *V <<
"\n"; } } while (false)
;
2180 }
2181 return V;
2182}
2183
2184static Value *extractVector(IRBuilderTy &IRB, Value *V, unsigned BeginIndex,
2185 unsigned EndIndex, const Twine &Name) {
2186 VectorType *VecTy = cast<VectorType>(V->getType());
2187 unsigned NumElements = EndIndex - BeginIndex;
2188 assert(NumElements <= VecTy->getNumElements() && "Too many elements!")((NumElements <= VecTy->getNumElements() && "Too many elements!"
) ? static_cast<void> (0) : __assert_fail ("NumElements <= VecTy->getNumElements() && \"Too many elements!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2188, __PRETTY_FUNCTION__))
;
2189
2190 if (NumElements == VecTy->getNumElements())
2191 return V;
2192
2193 if (NumElements == 1) {
2194 V = IRB.CreateExtractElement(V, IRB.getInt32(BeginIndex),
2195 Name + ".extract");
2196 LLVM_DEBUG(dbgs() << " extract: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " extract: " << *V <<
"\n"; } } while (false)
;
2197 return V;
2198 }
2199
2200 SmallVector<Constant *, 8> Mask;
2201 Mask.reserve(NumElements);
2202 for (unsigned i = BeginIndex; i != EndIndex; ++i)
2203 Mask.push_back(IRB.getInt32(i));
2204 V = IRB.CreateShuffleVector(V, UndefValue::get(V->getType()),
2205 ConstantVector::get(Mask), Name + ".extract");
2206 LLVM_DEBUG(dbgs() << " shuffle: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " shuffle: " << *V <<
"\n"; } } while (false)
;
2207 return V;
2208}
2209
2210static Value *insertVector(IRBuilderTy &IRB, Value *Old, Value *V,
2211 unsigned BeginIndex, const Twine &Name) {
2212 VectorType *VecTy = cast<VectorType>(Old->getType());
2213 assert(VecTy && "Can only insert a vector into a vector")((VecTy && "Can only insert a vector into a vector") ?
static_cast<void> (0) : __assert_fail ("VecTy && \"Can only insert a vector into a vector\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2213, __PRETTY_FUNCTION__))
;
2214
2215 VectorType *Ty = dyn_cast<VectorType>(V->getType());
2216 if (!Ty) {
2217 // Single element to insert.
2218 V = IRB.CreateInsertElement(Old, V, IRB.getInt32(BeginIndex),
2219 Name + ".insert");
2220 LLVM_DEBUG(dbgs() << " insert: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " insert: " << *V <<
"\n"; } } while (false)
;
2221 return V;
2222 }
2223
2224 assert(Ty->getNumElements() <= VecTy->getNumElements() &&((Ty->getNumElements() <= VecTy->getNumElements() &&
"Too many elements!") ? static_cast<void> (0) : __assert_fail
("Ty->getNumElements() <= VecTy->getNumElements() && \"Too many elements!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2225, __PRETTY_FUNCTION__))
2225 "Too many elements!")((Ty->getNumElements() <= VecTy->getNumElements() &&
"Too many elements!") ? static_cast<void> (0) : __assert_fail
("Ty->getNumElements() <= VecTy->getNumElements() && \"Too many elements!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2225, __PRETTY_FUNCTION__))
;
2226 if (Ty->getNumElements() == VecTy->getNumElements()) {
2227 assert(V->getType() == VecTy && "Vector type mismatch")((V->getType() == VecTy && "Vector type mismatch")
? static_cast<void> (0) : __assert_fail ("V->getType() == VecTy && \"Vector type mismatch\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2227, __PRETTY_FUNCTION__))
;
2228 return V;
2229 }
2230 unsigned EndIndex = BeginIndex + Ty->getNumElements();
2231
2232 // When inserting a smaller vector into the larger to store, we first
2233 // use a shuffle vector to widen it with undef elements, and then
2234 // a second shuffle vector to select between the loaded vector and the
2235 // incoming vector.
2236 SmallVector<Constant *, 8> Mask;
2237 Mask.reserve(VecTy->getNumElements());
2238 for (unsigned i = 0; i != VecTy->getNumElements(); ++i)
2239 if (i >= BeginIndex && i < EndIndex)
2240 Mask.push_back(IRB.getInt32(i - BeginIndex));
2241 else
2242 Mask.push_back(UndefValue::get(IRB.getInt32Ty()));
2243 V = IRB.CreateShuffleVector(V, UndefValue::get(V->getType()),
2244 ConstantVector::get(Mask), Name + ".expand");
2245 LLVM_DEBUG(dbgs() << " shuffle: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " shuffle: " << *V <<
"\n"; } } while (false)
;
2246
2247 Mask.clear();
2248 for (unsigned i = 0; i != VecTy->getNumElements(); ++i)
2249 Mask.push_back(IRB.getInt1(i >= BeginIndex && i < EndIndex));
2250
2251 V = IRB.CreateSelect(ConstantVector::get(Mask), V, Old, Name + "blend");
2252
2253 LLVM_DEBUG(dbgs() << " blend: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " blend: " << *V <<
"\n"; } } while (false)
;
2254 return V;
2255}
2256
2257/// Visitor to rewrite instructions using p particular slice of an alloca
2258/// to use a new alloca.
2259///
2260/// Also implements the rewriting to vector-based accesses when the partition
2261/// passes the isVectorPromotionViable predicate. Most of the rewriting logic
2262/// lives here.
2263class llvm::sroa::AllocaSliceRewriter
2264 : public InstVisitor<AllocaSliceRewriter, bool> {
2265 // Befriend the base class so it can delegate to private visit methods.
2266 friend class InstVisitor<AllocaSliceRewriter, bool>;
2267
2268 using Base = InstVisitor<AllocaSliceRewriter, bool>;
2269
2270 const DataLayout &DL;
2271 AllocaSlices &AS;
2272 SROA &Pass;
2273 AllocaInst &OldAI, &NewAI;
2274 const uint64_t NewAllocaBeginOffset, NewAllocaEndOffset;
2275 Type *NewAllocaTy;
2276
2277 // This is a convenience and flag variable that will be null unless the new
2278 // alloca's integer operations should be widened to this integer type due to
2279 // passing isIntegerWideningViable above. If it is non-null, the desired
2280 // integer type will be stored here for easy access during rewriting.
2281 IntegerType *IntTy;
2282
2283 // If we are rewriting an alloca partition which can be written as pure
2284 // vector operations, we stash extra information here. When VecTy is
2285 // non-null, we have some strict guarantees about the rewritten alloca:
2286 // - The new alloca is exactly the size of the vector type here.
2287 // - The accesses all either map to the entire vector or to a single
2288 // element.
2289 // - The set of accessing instructions is only one of those handled above
2290 // in isVectorPromotionViable. Generally these are the same access kinds
2291 // which are promotable via mem2reg.
2292 VectorType *VecTy;
2293 Type *ElementTy;
2294 uint64_t ElementSize;
2295
2296 // The original offset of the slice currently being rewritten relative to
2297 // the original alloca.
2298 uint64_t BeginOffset = 0;
2299 uint64_t EndOffset = 0;
2300
2301 // The new offsets of the slice currently being rewritten relative to the
2302 // original alloca.
2303 uint64_t NewBeginOffset, NewEndOffset;
2304
2305 uint64_t SliceSize;
2306 bool IsSplittable = false;
2307 bool IsSplit = false;
2308 Use *OldUse = nullptr;
2309 Instruction *OldPtr = nullptr;
2310
2311 // Track post-rewrite users which are PHI nodes and Selects.
2312 SmallSetVector<PHINode *, 8> &PHIUsers;
2313 SmallSetVector<SelectInst *, 8> &SelectUsers;
2314
2315 // Utility IR builder, whose name prefix is setup for each visited use, and
2316 // the insertion point is set to point to the user.
2317 IRBuilderTy IRB;
2318
2319public:
2320 AllocaSliceRewriter(const DataLayout &DL, AllocaSlices &AS, SROA &Pass,
2321 AllocaInst &OldAI, AllocaInst &NewAI,
2322 uint64_t NewAllocaBeginOffset,
2323 uint64_t NewAllocaEndOffset, bool IsIntegerPromotable,
2324 VectorType *PromotableVecTy,
2325 SmallSetVector<PHINode *, 8> &PHIUsers,
2326 SmallSetVector<SelectInst *, 8> &SelectUsers)
2327 : DL(DL), AS(AS), Pass(Pass), OldAI(OldAI), NewAI(NewAI),
2328 NewAllocaBeginOffset(NewAllocaBeginOffset),
2329 NewAllocaEndOffset(NewAllocaEndOffset),
2330 NewAllocaTy(NewAI.getAllocatedType()),
2331 IntTy(IsIntegerPromotable
2332 ? Type::getIntNTy(
2333 NewAI.getContext(),
2334 DL.getTypeSizeInBits(NewAI.getAllocatedType()))
2335 : nullptr),
2336 VecTy(PromotableVecTy),
2337 ElementTy(VecTy ? VecTy->getElementType() : nullptr),
2338 ElementSize(VecTy ? DL.getTypeSizeInBits(ElementTy) / 8 : 0),
2339 PHIUsers(PHIUsers), SelectUsers(SelectUsers),
2340 IRB(NewAI.getContext(), ConstantFolder()) {
2341 if (VecTy) {
2342 assert((DL.getTypeSizeInBits(ElementTy) % 8) == 0 &&(((DL.getTypeSizeInBits(ElementTy) % 8) == 0 && "Only multiple-of-8 sized vector elements are viable"
) ? static_cast<void> (0) : __assert_fail ("(DL.getTypeSizeInBits(ElementTy) % 8) == 0 && \"Only multiple-of-8 sized vector elements are viable\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2343, __PRETTY_FUNCTION__))
2343 "Only multiple-of-8 sized vector elements are viable")(((DL.getTypeSizeInBits(ElementTy) % 8) == 0 && "Only multiple-of-8 sized vector elements are viable"
) ? static_cast<void> (0) : __assert_fail ("(DL.getTypeSizeInBits(ElementTy) % 8) == 0 && \"Only multiple-of-8 sized vector elements are viable\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2343, __PRETTY_FUNCTION__))
;
2344 ++NumVectorized;
2345 }
2346 assert((!IntTy && !VecTy) || (IntTy && !VecTy) || (!IntTy && VecTy))(((!IntTy && !VecTy) || (IntTy && !VecTy) || (
!IntTy && VecTy)) ? static_cast<void> (0) : __assert_fail
("(!IntTy && !VecTy) || (IntTy && !VecTy) || (!IntTy && VecTy)"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2346, __PRETTY_FUNCTION__))
;
2347 }
2348
2349 bool visit(AllocaSlices::const_iterator I) {
2350 bool CanSROA = true;
2351 BeginOffset = I->beginOffset();
2352 EndOffset = I->endOffset();
2353 IsSplittable = I->isSplittable();
2354 IsSplit =
2355 BeginOffset < NewAllocaBeginOffset || EndOffset > NewAllocaEndOffset;
2356 LLVM_DEBUG(dbgs() << " rewriting " << (IsSplit ? "split " : ""))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " rewriting " << (IsSplit ?
"split " : ""); } } while (false)
;
2357 LLVM_DEBUG(AS.printSlice(dbgs(), I, ""))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { AS.printSlice(dbgs(), I, ""); } } while (false)
;
2358 LLVM_DEBUG(dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "\n"; } } while (false)
;
2359
2360 // Compute the intersecting offset range.
2361 assert(BeginOffset < NewAllocaEndOffset)((BeginOffset < NewAllocaEndOffset) ? static_cast<void>
(0) : __assert_fail ("BeginOffset < NewAllocaEndOffset", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2361, __PRETTY_FUNCTION__))
;
2362 assert(EndOffset > NewAllocaBeginOffset)((EndOffset > NewAllocaBeginOffset) ? static_cast<void>
(0) : __assert_fail ("EndOffset > NewAllocaBeginOffset", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2362, __PRETTY_FUNCTION__))
;
2363 NewBeginOffset = std::max(BeginOffset, NewAllocaBeginOffset);
2364 NewEndOffset = std::min(EndOffset, NewAllocaEndOffset);
2365
2366 SliceSize = NewEndOffset - NewBeginOffset;
2367
2368 OldUse = I->getUse();
2369 OldPtr = cast<Instruction>(OldUse->get());
2370
2371 Instruction *OldUserI = cast<Instruction>(OldUse->getUser());
2372 IRB.SetInsertPoint(OldUserI);
2373 IRB.SetCurrentDebugLocation(OldUserI->getDebugLoc());
2374 IRB.SetNamePrefix(Twine(NewAI.getName()) + "." + Twine(BeginOffset) + ".");
2375
2376 CanSROA &= visit(cast<Instruction>(OldUse->getUser()));
2377 if (VecTy || IntTy)
2378 assert(CanSROA)((CanSROA) ? static_cast<void> (0) : __assert_fail ("CanSROA"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2378, __PRETTY_FUNCTION__))
;
2379 return CanSROA;
2380 }
2381
2382private:
2383 // Make sure the other visit overloads are visible.
2384 using Base::visit;
2385
2386 // Every instruction which can end up as a user must have a rewrite rule.
2387 bool visitInstruction(Instruction &I) {
2388 LLVM_DEBUG(dbgs() << " !!!! Cannot rewrite: " << I << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " !!!! Cannot rewrite: " <<
I << "\n"; } } while (false)
;
2389 llvm_unreachable("No rewrite rule for this instruction!")::llvm::llvm_unreachable_internal("No rewrite rule for this instruction!"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2389)
;
2390 }
2391
2392 Value *getNewAllocaSlicePtr(IRBuilderTy &IRB, Type *PointerTy) {
2393 // Note that the offset computation can use BeginOffset or NewBeginOffset
2394 // interchangeably for unsplit slices.
2395 assert(IsSplit || BeginOffset == NewBeginOffset)((IsSplit || BeginOffset == NewBeginOffset) ? static_cast<
void> (0) : __assert_fail ("IsSplit || BeginOffset == NewBeginOffset"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2395, __PRETTY_FUNCTION__))
;
2396 uint64_t Offset = NewBeginOffset - NewAllocaBeginOffset;
2397
2398#ifndef NDEBUG
2399 StringRef OldName = OldPtr->getName();
2400 // Skip through the last '.sroa.' component of the name.
2401 size_t LastSROAPrefix = OldName.rfind(".sroa.");
2402 if (LastSROAPrefix != StringRef::npos) {
2403 OldName = OldName.substr(LastSROAPrefix + strlen(".sroa."));
2404 // Look for an SROA slice index.
2405 size_t IndexEnd = OldName.find_first_not_of("0123456789");
2406 if (IndexEnd != StringRef::npos && OldName[IndexEnd] == '.') {
2407 // Strip the index and look for the offset.
2408 OldName = OldName.substr(IndexEnd + 1);
2409 size_t OffsetEnd = OldName.find_first_not_of("0123456789");
2410 if (OffsetEnd != StringRef::npos && OldName[OffsetEnd] == '.')
2411 // Strip the offset.
2412 OldName = OldName.substr(OffsetEnd + 1);
2413 }
2414 }
2415 // Strip any SROA suffixes as well.
2416 OldName = OldName.substr(0, OldName.find(".sroa_"));
2417#endif
2418
2419 return getAdjustedPtr(IRB, DL, &NewAI,
2420 APInt(DL.getIndexTypeSizeInBits(PointerTy), Offset),
2421 PointerTy,
2422#ifndef NDEBUG
2423 Twine(OldName) + "."
2424#else
2425 Twine()
2426#endif
2427 );
2428 }
2429
2430 /// Compute suitable alignment to access this slice of the *new*
2431 /// alloca.
2432 ///
2433 /// You can optionally pass a type to this routine and if that type's ABI
2434 /// alignment is itself suitable, this will return zero.
2435 unsigned getSliceAlign(Type *Ty = nullptr) {
2436 unsigned NewAIAlign = NewAI.getAlignment();
2437 if (!NewAIAlign)
2438 NewAIAlign = DL.getABITypeAlignment(NewAI.getAllocatedType());
2439 unsigned Align =
2440 MinAlign(NewAIAlign, NewBeginOffset - NewAllocaBeginOffset);
2441 return (Ty && Align == DL.getABITypeAlignment(Ty)) ? 0 : Align;
2442 }
2443
2444 unsigned getIndex(uint64_t Offset) {
2445 assert(VecTy && "Can only call getIndex when rewriting a vector")((VecTy && "Can only call getIndex when rewriting a vector"
) ? static_cast<void> (0) : __assert_fail ("VecTy && \"Can only call getIndex when rewriting a vector\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2445, __PRETTY_FUNCTION__))
;
2446 uint64_t RelOffset = Offset - NewAllocaBeginOffset;
2447 assert(RelOffset / ElementSize < UINT32_MAX && "Index out of bounds")((RelOffset / ElementSize < (4294967295U) && "Index out of bounds"
) ? static_cast<void> (0) : __assert_fail ("RelOffset / ElementSize < UINT32_MAX && \"Index out of bounds\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2447, __PRETTY_FUNCTION__))
;
2448 uint32_t Index = RelOffset / ElementSize;
2449 assert(Index * ElementSize == RelOffset)((Index * ElementSize == RelOffset) ? static_cast<void>
(0) : __assert_fail ("Index * ElementSize == RelOffset", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2449, __PRETTY_FUNCTION__))
;
2450 return Index;
2451 }
2452
2453 void deleteIfTriviallyDead(Value *V) {
2454 Instruction *I = cast<Instruction>(V);
2455 if (isInstructionTriviallyDead(I))
2456 Pass.DeadInsts.insert(I);
2457 }
2458
2459 Value *rewriteVectorizedLoadInst() {
2460 unsigned BeginIndex = getIndex(NewBeginOffset);
2461 unsigned EndIndex = getIndex(NewEndOffset);
2462 assert(EndIndex > BeginIndex && "Empty vector!")((EndIndex > BeginIndex && "Empty vector!") ? static_cast
<void> (0) : __assert_fail ("EndIndex > BeginIndex && \"Empty vector!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2462, __PRETTY_FUNCTION__))
;
2463
2464 Value *V = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
2465 NewAI.getAlignment(), "load");
2466 return extractVector(IRB, V, BeginIndex, EndIndex, "vec");
2467 }
2468
2469 Value *rewriteIntegerLoad(LoadInst &LI) {
2470 assert(IntTy && "We cannot insert an integer to the alloca")((IntTy && "We cannot insert an integer to the alloca"
) ? static_cast<void> (0) : __assert_fail ("IntTy && \"We cannot insert an integer to the alloca\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2470, __PRETTY_FUNCTION__))
;
2471 assert(!LI.isVolatile())((!LI.isVolatile()) ? static_cast<void> (0) : __assert_fail
("!LI.isVolatile()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2471, __PRETTY_FUNCTION__))
;
2472 Value *V = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
2473 NewAI.getAlignment(), "load");
2474 V = convertValue(DL, IRB, V, IntTy);
2475 assert(NewBeginOffset >= NewAllocaBeginOffset && "Out of bounds offset")((NewBeginOffset >= NewAllocaBeginOffset && "Out of bounds offset"
) ? static_cast<void> (0) : __assert_fail ("NewBeginOffset >= NewAllocaBeginOffset && \"Out of bounds offset\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2475, __PRETTY_FUNCTION__))
;
2476 uint64_t Offset = NewBeginOffset - NewAllocaBeginOffset;
2477 if (Offset > 0 || NewEndOffset < NewAllocaEndOffset) {
2478 IntegerType *ExtractTy = Type::getIntNTy(LI.getContext(), SliceSize * 8);
2479 V = extractInteger(DL, IRB, V, ExtractTy, Offset, "extract");
2480 }
2481 // It is possible that the extracted type is not the load type. This
2482 // happens if there is a load past the end of the alloca, and as
2483 // a consequence the slice is narrower but still a candidate for integer
2484 // lowering. To handle this case, we just zero extend the extracted
2485 // integer.
2486 assert(cast<IntegerType>(LI.getType())->getBitWidth() >= SliceSize * 8 &&((cast<IntegerType>(LI.getType())->getBitWidth() >=
SliceSize * 8 && "Can only handle an extract for an overly wide load"
) ? static_cast<void> (0) : __assert_fail ("cast<IntegerType>(LI.getType())->getBitWidth() >= SliceSize * 8 && \"Can only handle an extract for an overly wide load\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2487, __PRETTY_FUNCTION__))
2487 "Can only handle an extract for an overly wide load")((cast<IntegerType>(LI.getType())->getBitWidth() >=
SliceSize * 8 && "Can only handle an extract for an overly wide load"
) ? static_cast<void> (0) : __assert_fail ("cast<IntegerType>(LI.getType())->getBitWidth() >= SliceSize * 8 && \"Can only handle an extract for an overly wide load\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2487, __PRETTY_FUNCTION__))
;
2488 if (cast<IntegerType>(LI.getType())->getBitWidth() > SliceSize * 8)
2489 V = IRB.CreateZExt(V, LI.getType());
2490 return V;
2491 }
2492
2493 bool visitLoadInst(LoadInst &LI) {
2494 LLVM_DEBUG(dbgs() << " original: " << LI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << LI <<
"\n"; } } while (false)
;
2495 Value *OldOp = LI.getOperand(0);
2496 assert(OldOp == OldPtr)((OldOp == OldPtr) ? static_cast<void> (0) : __assert_fail
("OldOp == OldPtr", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2496, __PRETTY_FUNCTION__))
;
2497
2498 AAMDNodes AATags;
2499 LI.getAAMetadata(AATags);
2500
2501 unsigned AS = LI.getPointerAddressSpace();
2502
2503 Type *TargetTy = IsSplit ? Type::getIntNTy(LI.getContext(), SliceSize * 8)
2504 : LI.getType();
2505 const bool IsLoadPastEnd = DL.getTypeStoreSize(TargetTy) > SliceSize;
2506 bool IsPtrAdjusted = false;
2507 Value *V;
2508 if (VecTy) {
2509 V = rewriteVectorizedLoadInst();
2510 } else if (IntTy && LI.getType()->isIntegerTy()) {
2511 V = rewriteIntegerLoad(LI);
2512 } else if (NewBeginOffset == NewAllocaBeginOffset &&
2513 NewEndOffset == NewAllocaEndOffset &&
2514 (canConvertValue(DL, NewAllocaTy, TargetTy) ||
2515 (IsLoadPastEnd && NewAllocaTy->isIntegerTy() &&
2516 TargetTy->isIntegerTy()))) {
2517 LoadInst *NewLI = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
2518 NewAI.getAlignment(),
2519 LI.isVolatile(), LI.getName());
2520 if (AATags)
2521 NewLI->setAAMetadata(AATags);
2522 if (LI.isVolatile())
2523 NewLI->setAtomic(LI.getOrdering(), LI.getSyncScopeID());
2524
2525 // Any !nonnull metadata or !range metadata on the old load is also valid
2526 // on the new load. This is even true in some cases even when the loads
2527 // are different types, for example by mapping !nonnull metadata to
2528 // !range metadata by modeling the null pointer constant converted to the
2529 // integer type.
2530 // FIXME: Add support for range metadata here. Currently the utilities
2531 // for this don't propagate range metadata in trivial cases from one
2532 // integer load to another, don't handle non-addrspace-0 null pointers
2533 // correctly, and don't have any support for mapping ranges as the
2534 // integer type becomes winder or narrower.
2535 if (MDNode *N = LI.getMetadata(LLVMContext::MD_nonnull))
2536 copyNonnullMetadata(LI, N, *NewLI);
2537
2538 // Try to preserve nonnull metadata
2539 V = NewLI;
2540
2541 // If this is an integer load past the end of the slice (which means the
2542 // bytes outside the slice are undef or this load is dead) just forcibly
2543 // fix the integer size with correct handling of endianness.
2544 if (auto *AITy = dyn_cast<IntegerType>(NewAllocaTy))
2545 if (auto *TITy = dyn_cast<IntegerType>(TargetTy))
2546 if (AITy->getBitWidth() < TITy->getBitWidth()) {
2547 V = IRB.CreateZExt(V, TITy, "load.ext");
2548 if (DL.isBigEndian())
2549 V = IRB.CreateShl(V, TITy->getBitWidth() - AITy->getBitWidth(),
2550 "endian_shift");
2551 }
2552 } else {
2553 Type *LTy = TargetTy->getPointerTo(AS);
2554 LoadInst *NewLI = IRB.CreateAlignedLoad(
2555 TargetTy, getNewAllocaSlicePtr(IRB, LTy), getSliceAlign(TargetTy),
2556 LI.isVolatile(), LI.getName());
2557 if (AATags)
2558 NewLI->setAAMetadata(AATags);
2559 if (LI.isVolatile())
2560 NewLI->setAtomic(LI.getOrdering(), LI.getSyncScopeID());
2561
2562 V = NewLI;
2563 IsPtrAdjusted = true;
2564 }
2565 V = convertValue(DL, IRB, V, TargetTy);
2566
2567 if (IsSplit) {
2568 assert(!LI.isVolatile())((!LI.isVolatile()) ? static_cast<void> (0) : __assert_fail
("!LI.isVolatile()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2568, __PRETTY_FUNCTION__))
;
2569 assert(LI.getType()->isIntegerTy() &&((LI.getType()->isIntegerTy() && "Only integer type loads and stores are split"
) ? static_cast<void> (0) : __assert_fail ("LI.getType()->isIntegerTy() && \"Only integer type loads and stores are split\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2570, __PRETTY_FUNCTION__))
2570 "Only integer type loads and stores are split")((LI.getType()->isIntegerTy() && "Only integer type loads and stores are split"
) ? static_cast<void> (0) : __assert_fail ("LI.getType()->isIntegerTy() && \"Only integer type loads and stores are split\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2570, __PRETTY_FUNCTION__))
;
2571 assert(SliceSize < DL.getTypeStoreSize(LI.getType()) &&((SliceSize < DL.getTypeStoreSize(LI.getType()) &&
"Split load isn't smaller than original load") ? static_cast
<void> (0) : __assert_fail ("SliceSize < DL.getTypeStoreSize(LI.getType()) && \"Split load isn't smaller than original load\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2572, __PRETTY_FUNCTION__))
2572 "Split load isn't smaller than original load")((SliceSize < DL.getTypeStoreSize(LI.getType()) &&
"Split load isn't smaller than original load") ? static_cast
<void> (0) : __assert_fail ("SliceSize < DL.getTypeStoreSize(LI.getType()) && \"Split load isn't smaller than original load\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2572, __PRETTY_FUNCTION__))
;
2573 assert(DL.typeSizeEqualsStoreSize(LI.getType()) &&((DL.typeSizeEqualsStoreSize(LI.getType()) && "Non-byte-multiple bit width"
) ? static_cast<void> (0) : __assert_fail ("DL.typeSizeEqualsStoreSize(LI.getType()) && \"Non-byte-multiple bit width\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2574, __PRETTY_FUNCTION__))
2574 "Non-byte-multiple bit width")((DL.typeSizeEqualsStoreSize(LI.getType()) && "Non-byte-multiple bit width"
) ? static_cast<void> (0) : __assert_fail ("DL.typeSizeEqualsStoreSize(LI.getType()) && \"Non-byte-multiple bit width\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2574, __PRETTY_FUNCTION__))
;
2575 // Move the insertion point just past the load so that we can refer to it.
2576 IRB.SetInsertPoint(&*std::next(BasicBlock::iterator(&LI)));
2577 // Create a placeholder value with the same type as LI to use as the
2578 // basis for the new value. This allows us to replace the uses of LI with
2579 // the computed value, and then replace the placeholder with LI, leaving
2580 // LI only used for this computation.
2581 Value *Placeholder = new LoadInst(
2582 LI.getType(), UndefValue::get(LI.getType()->getPointerTo(AS)));
2583 V = insertInteger(DL, IRB, Placeholder, V, NewBeginOffset - BeginOffset,
2584 "insert");
2585 LI.replaceAllUsesWith(V);
2586 Placeholder->replaceAllUsesWith(&LI);
2587 Placeholder->deleteValue();
2588 } else {
2589 LI.replaceAllUsesWith(V);
2590 }
2591
2592 Pass.DeadInsts.insert(&LI);
2593 deleteIfTriviallyDead(OldOp);
2594 LLVM_DEBUG(dbgs() << " to: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *V <<
"\n"; } } while (false)
;
2595 return !LI.isVolatile() && !IsPtrAdjusted;
2596 }
2597
2598 bool rewriteVectorizedStoreInst(Value *V, StoreInst &SI, Value *OldOp,
2599 AAMDNodes AATags) {
2600 if (V->getType() != VecTy) {
2601 unsigned BeginIndex = getIndex(NewBeginOffset);
2602 unsigned EndIndex = getIndex(NewEndOffset);
2603 assert(EndIndex > BeginIndex && "Empty vector!")((EndIndex > BeginIndex && "Empty vector!") ? static_cast
<void> (0) : __assert_fail ("EndIndex > BeginIndex && \"Empty vector!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2603, __PRETTY_FUNCTION__))
;
2604 unsigned NumElements = EndIndex - BeginIndex;
2605 assert(NumElements <= VecTy->getNumElements() && "Too many elements!")((NumElements <= VecTy->getNumElements() && "Too many elements!"
) ? static_cast<void> (0) : __assert_fail ("NumElements <= VecTy->getNumElements() && \"Too many elements!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2605, __PRETTY_FUNCTION__))
;
2606 Type *SliceTy = (NumElements == 1)
2607 ? ElementTy
2608 : VectorType::get(ElementTy, NumElements);
2609 if (V->getType() != SliceTy)
2610 V = convertValue(DL, IRB, V, SliceTy);
2611
2612 // Mix in the existing elements.
2613 Value *Old = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
2614 NewAI.getAlignment(), "load");
2615 V = insertVector(IRB, Old, V, BeginIndex, "vec");
2616 }
2617 StoreInst *Store = IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlignment());
2618 if (AATags)
2619 Store->setAAMetadata(AATags);
2620 Pass.DeadInsts.insert(&SI);
2621
2622 LLVM_DEBUG(dbgs() << " to: " << *Store << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *Store <<
"\n"; } } while (false)
;
2623 return true;
2624 }
2625
2626 bool rewriteIntegerStore(Value *V, StoreInst &SI, AAMDNodes AATags) {
2627 assert(IntTy && "We cannot extract an integer from the alloca")((IntTy && "We cannot extract an integer from the alloca"
) ? static_cast<void> (0) : __assert_fail ("IntTy && \"We cannot extract an integer from the alloca\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2627, __PRETTY_FUNCTION__))
;
2628 assert(!SI.isVolatile())((!SI.isVolatile()) ? static_cast<void> (0) : __assert_fail
("!SI.isVolatile()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2628, __PRETTY_FUNCTION__))
;
2629 if (DL.getTypeSizeInBits(V->getType()) != IntTy->getBitWidth()) {
2630 Value *Old = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
2631 NewAI.getAlignment(), "oldload");
2632 Old = convertValue(DL, IRB, Old, IntTy);
2633 assert(BeginOffset >= NewAllocaBeginOffset && "Out of bounds offset")((BeginOffset >= NewAllocaBeginOffset && "Out of bounds offset"
) ? static_cast<void> (0) : __assert_fail ("BeginOffset >= NewAllocaBeginOffset && \"Out of bounds offset\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2633, __PRETTY_FUNCTION__))
;
2634 uint64_t Offset = BeginOffset - NewAllocaBeginOffset;
2635 V = insertInteger(DL, IRB, Old, SI.getValueOperand(), Offset, "insert");
2636 }
2637 V = convertValue(DL, IRB, V, NewAllocaTy);
2638 StoreInst *Store = IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlignment());
2639 Store->copyMetadata(SI, {LLVMContext::MD_mem_parallel_loop_access,
2640 LLVMContext::MD_access_group});
2641 if (AATags)
2642 Store->setAAMetadata(AATags);
2643 Pass.DeadInsts.insert(&SI);
2644 LLVM_DEBUG(dbgs() << " to: " << *Store << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *Store <<
"\n"; } } while (false)
;
2645 return true;
2646 }
2647
2648 bool visitStoreInst(StoreInst &SI) {
2649 LLVM_DEBUG(dbgs() << " original: " << SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << SI <<
"\n"; } } while (false)
;
2650 Value *OldOp = SI.getOperand(1);
2651 assert(OldOp == OldPtr)((OldOp == OldPtr) ? static_cast<void> (0) : __assert_fail
("OldOp == OldPtr", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2651, __PRETTY_FUNCTION__))
;
2652
2653 AAMDNodes AATags;
2654 SI.getAAMetadata(AATags);
2655
2656 Value *V = SI.getValueOperand();
2657
2658 // Strip all inbounds GEPs and pointer casts to try to dig out any root
2659 // alloca that should be re-examined after promoting this alloca.
2660 if (V->getType()->isPointerTy())
2661 if (AllocaInst *AI = dyn_cast<AllocaInst>(V->stripInBoundsOffsets()))
2662 Pass.PostPromotionWorklist.insert(AI);
2663
2664 if (SliceSize < DL.getTypeStoreSize(V->getType())) {
2665 assert(!SI.isVolatile())((!SI.isVolatile()) ? static_cast<void> (0) : __assert_fail
("!SI.isVolatile()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2665, __PRETTY_FUNCTION__))
;
2666 assert(V->getType()->isIntegerTy() &&((V->getType()->isIntegerTy() && "Only integer type loads and stores are split"
) ? static_cast<void> (0) : __assert_fail ("V->getType()->isIntegerTy() && \"Only integer type loads and stores are split\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2667, __PRETTY_FUNCTION__))
2667 "Only integer type loads and stores are split")((V->getType()->isIntegerTy() && "Only integer type loads and stores are split"
) ? static_cast<void> (0) : __assert_fail ("V->getType()->isIntegerTy() && \"Only integer type loads and stores are split\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2667, __PRETTY_FUNCTION__))
;
2668 assert(DL.typeSizeEqualsStoreSize(V->getType()) &&((DL.typeSizeEqualsStoreSize(V->getType()) && "Non-byte-multiple bit width"
) ? static_cast<void> (0) : __assert_fail ("DL.typeSizeEqualsStoreSize(V->getType()) && \"Non-byte-multiple bit width\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2669, __PRETTY_FUNCTION__))
2669 "Non-byte-multiple bit width")((DL.typeSizeEqualsStoreSize(V->getType()) && "Non-byte-multiple bit width"
) ? static_cast<void> (0) : __assert_fail ("DL.typeSizeEqualsStoreSize(V->getType()) && \"Non-byte-multiple bit width\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2669, __PRETTY_FUNCTION__))
;
2670 IntegerType *NarrowTy = Type::getIntNTy(SI.getContext(), SliceSize * 8);
2671 V = extractInteger(DL, IRB, V, NarrowTy, NewBeginOffset - BeginOffset,
2672 "extract");
2673 }
2674
2675 if (VecTy)
2676 return rewriteVectorizedStoreInst(V, SI, OldOp, AATags);
2677 if (IntTy && V->getType()->isIntegerTy())
2678 return rewriteIntegerStore(V, SI, AATags);
2679
2680 const bool IsStorePastEnd = DL.getTypeStoreSize(V->getType()) > SliceSize;
2681 StoreInst *NewSI;
2682 if (NewBeginOffset == NewAllocaBeginOffset &&
2683 NewEndOffset == NewAllocaEndOffset &&
2684 (canConvertValue(DL, V->getType(), NewAllocaTy) ||
2685 (IsStorePastEnd && NewAllocaTy->isIntegerTy() &&
2686 V->getType()->isIntegerTy()))) {
2687 // If this is an integer store past the end of slice (and thus the bytes
2688 // past that point are irrelevant or this is unreachable), truncate the
2689 // value prior to storing.
2690 if (auto *VITy = dyn_cast<IntegerType>(V->getType()))
2691 if (auto *AITy = dyn_cast<IntegerType>(NewAllocaTy))
2692 if (VITy->getBitWidth() > AITy->getBitWidth()) {
2693 if (DL.isBigEndian())
2694 V = IRB.CreateLShr(V, VITy->getBitWidth() - AITy->getBitWidth(),
2695 "endian_shift");
2696 V = IRB.CreateTrunc(V, AITy, "load.trunc");
2697 }
2698
2699 V = convertValue(DL, IRB, V, NewAllocaTy);
2700 NewSI = IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlignment(),
2701 SI.isVolatile());
2702 } else {
2703 unsigned AS = SI.getPointerAddressSpace();
2704 Value *NewPtr = getNewAllocaSlicePtr(IRB, V->getType()->getPointerTo(AS));
2705 NewSI = IRB.CreateAlignedStore(V, NewPtr, getSliceAlign(V->getType()),
2706 SI.isVolatile());
2707 }
2708 NewSI->copyMetadata(SI, {LLVMContext::MD_mem_parallel_loop_access,
2709 LLVMContext::MD_access_group});
2710 if (AATags)
2711 NewSI->setAAMetadata(AATags);
2712 if (SI.isVolatile())
2713 NewSI->setAtomic(SI.getOrdering(), SI.getSyncScopeID());
2714 Pass.DeadInsts.insert(&SI);
2715 deleteIfTriviallyDead(OldOp);
2716
2717 LLVM_DEBUG(dbgs() << " to: " << *NewSI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *NewSI <<
"\n"; } } while (false)
;
2718 return NewSI->getPointerOperand() == &NewAI && !SI.isVolatile();
2719 }
2720
2721 /// Compute an integer value from splatting an i8 across the given
2722 /// number of bytes.
2723 ///
2724 /// Note that this routine assumes an i8 is a byte. If that isn't true, don't
2725 /// call this routine.
2726 /// FIXME: Heed the advice above.
2727 ///
2728 /// \param V The i8 value to splat.
2729 /// \param Size The number of bytes in the output (assuming i8 is one byte)
2730 Value *getIntegerSplat(Value *V, unsigned Size) {
2731 assert(Size > 0 && "Expected a positive number of bytes.")((Size > 0 && "Expected a positive number of bytes."
) ? static_cast<void> (0) : __assert_fail ("Size > 0 && \"Expected a positive number of bytes.\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2731, __PRETTY_FUNCTION__))
;
2732 IntegerType *VTy = cast<IntegerType>(V->getType());
2733 assert(VTy->getBitWidth() == 8 && "Expected an i8 value for the byte")((VTy->getBitWidth() == 8 && "Expected an i8 value for the byte"
) ? static_cast<void> (0) : __assert_fail ("VTy->getBitWidth() == 8 && \"Expected an i8 value for the byte\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2733, __PRETTY_FUNCTION__))
;
2734 if (Size == 1)
2735 return V;
2736
2737 Type *SplatIntTy = Type::getIntNTy(VTy->getContext(), Size * 8);
2738 V = IRB.CreateMul(
2739 IRB.CreateZExt(V, SplatIntTy, "zext"),
2740 ConstantExpr::getUDiv(
2741 Constant::getAllOnesValue(SplatIntTy),
2742 ConstantExpr::getZExt(Constant::getAllOnesValue(V->getType()),
2743 SplatIntTy)),
2744 "isplat");
2745 return V;
2746 }
2747
2748 /// Compute a vector splat for a given element value.
2749 Value *getVectorSplat(Value *V, unsigned NumElements) {
2750 V = IRB.CreateVectorSplat(NumElements, V, "vsplat");
2751 LLVM_DEBUG(dbgs() << " splat: " << *V << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " splat: " << *V <<
"\n"; } } while (false)
;
2752 return V;
2753 }
2754
2755 bool visitMemSetInst(MemSetInst &II) {
2756 LLVM_DEBUG(dbgs() << " original: " << II << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << II <<
"\n"; } } while (false)
;
2757 assert(II.getRawDest() == OldPtr)((II.getRawDest() == OldPtr) ? static_cast<void> (0) : __assert_fail
("II.getRawDest() == OldPtr", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2757, __PRETTY_FUNCTION__))
;
2758
2759 AAMDNodes AATags;
2760 II.getAAMetadata(AATags);
2761
2762 // If the memset has a variable size, it cannot be split, just adjust the
2763 // pointer to the new alloca.
2764 if (!isa<Constant>(II.getLength())) {
2765 assert(!IsSplit)((!IsSplit) ? static_cast<void> (0) : __assert_fail ("!IsSplit"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2765, __PRETTY_FUNCTION__))
;
2766 assert(NewBeginOffset == BeginOffset)((NewBeginOffset == BeginOffset) ? static_cast<void> (0
) : __assert_fail ("NewBeginOffset == BeginOffset", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2766, __PRETTY_FUNCTION__))
;
2767 II.setDest(getNewAllocaSlicePtr(IRB, OldPtr->getType()));
2768 II.setDestAlignment(getSliceAlign());
2769
2770 deleteIfTriviallyDead(OldPtr);
2771 return false;
2772 }
2773
2774 // Record this instruction for deletion.
2775 Pass.DeadInsts.insert(&II);
2776
2777 Type *AllocaTy = NewAI.getAllocatedType();
2778 Type *ScalarTy = AllocaTy->getScalarType();
2779
2780 const bool CanContinue = [&]() {
2781 if (VecTy || IntTy)
2782 return true;
2783 if (BeginOffset > NewAllocaBeginOffset ||
2784 EndOffset < NewAllocaEndOffset)
2785 return false;
2786 auto *C = cast<ConstantInt>(II.getLength());
2787 if (C->getBitWidth() > 64)
2788 return false;
2789 const auto Len = C->getZExtValue();
2790 auto *Int8Ty = IntegerType::getInt8Ty(NewAI.getContext());
2791 auto *SrcTy = VectorType::get(Int8Ty, Len);
2792 return canConvertValue(DL, SrcTy, AllocaTy) &&
2793 DL.isLegalInteger(DL.getTypeSizeInBits(ScalarTy));
2794 }();
2795
2796 // If this doesn't map cleanly onto the alloca type, and that type isn't
2797 // a single value type, just emit a memset.
2798 if (!CanContinue) {
2799 Type *SizeTy = II.getLength()->getType();
2800 Constant *Size = ConstantInt::get(SizeTy, NewEndOffset - NewBeginOffset);
2801 CallInst *New = IRB.CreateMemSet(
2802 getNewAllocaSlicePtr(IRB, OldPtr->getType()), II.getValue(), Size,
2803 getSliceAlign(), II.isVolatile());
2804 if (AATags)
2805 New->setAAMetadata(AATags);
2806 LLVM_DEBUG(dbgs() << " to: " << *New << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *New <<
"\n"; } } while (false)
;
2807 return false;
2808 }
2809
2810 // If we can represent this as a simple value, we have to build the actual
2811 // value to store, which requires expanding the byte present in memset to
2812 // a sensible representation for the alloca type. This is essentially
2813 // splatting the byte to a sufficiently wide integer, splatting it across
2814 // any desired vector width, and bitcasting to the final type.
2815 Value *V;
2816
2817 if (VecTy) {
2818 // If this is a memset of a vectorized alloca, insert it.
2819 assert(ElementTy == ScalarTy)((ElementTy == ScalarTy) ? static_cast<void> (0) : __assert_fail
("ElementTy == ScalarTy", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2819, __PRETTY_FUNCTION__))
;
2820
2821 unsigned BeginIndex = getIndex(NewBeginOffset);
2822 unsigned EndIndex = getIndex(NewEndOffset);
2823 assert(EndIndex > BeginIndex && "Empty vector!")((EndIndex > BeginIndex && "Empty vector!") ? static_cast
<void> (0) : __assert_fail ("EndIndex > BeginIndex && \"Empty vector!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2823, __PRETTY_FUNCTION__))
;
2824 unsigned NumElements = EndIndex - BeginIndex;
2825 assert(NumElements <= VecTy->getNumElements() && "Too many elements!")((NumElements <= VecTy->getNumElements() && "Too many elements!"
) ? static_cast<void> (0) : __assert_fail ("NumElements <= VecTy->getNumElements() && \"Too many elements!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2825, __PRETTY_FUNCTION__))
;
2826
2827 Value *Splat =
2828 getIntegerSplat(II.getValue(), DL.getTypeSizeInBits(ElementTy) / 8);
2829 Splat = convertValue(DL, IRB, Splat, ElementTy);
2830 if (NumElements > 1)
2831 Splat = getVectorSplat(Splat, NumElements);
2832
2833 Value *Old = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
2834 NewAI.getAlignment(), "oldload");
2835 V = insertVector(IRB, Old, Splat, BeginIndex, "vec");
2836 } else if (IntTy) {
2837 // If this is a memset on an alloca where we can widen stores, insert the
2838 // set integer.
2839 assert(!II.isVolatile())((!II.isVolatile()) ? static_cast<void> (0) : __assert_fail
("!II.isVolatile()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2839, __PRETTY_FUNCTION__))
;
2840
2841 uint64_t Size = NewEndOffset - NewBeginOffset;
2842 V = getIntegerSplat(II.getValue(), Size);
2843
2844 if (IntTy && (BeginOffset != NewAllocaBeginOffset ||
2845 EndOffset != NewAllocaBeginOffset)) {
2846 Value *Old = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
2847 NewAI.getAlignment(), "oldload");
2848 Old = convertValue(DL, IRB, Old, IntTy);
2849 uint64_t Offset = NewBeginOffset - NewAllocaBeginOffset;
2850 V = insertInteger(DL, IRB, Old, V, Offset, "insert");
2851 } else {
2852 assert(V->getType() == IntTy &&((V->getType() == IntTy && "Wrong type for an alloca wide integer!"
) ? static_cast<void> (0) : __assert_fail ("V->getType() == IntTy && \"Wrong type for an alloca wide integer!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2853, __PRETTY_FUNCTION__))
2853 "Wrong type for an alloca wide integer!")((V->getType() == IntTy && "Wrong type for an alloca wide integer!"
) ? static_cast<void> (0) : __assert_fail ("V->getType() == IntTy && \"Wrong type for an alloca wide integer!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2853, __PRETTY_FUNCTION__))
;
2854 }
2855 V = convertValue(DL, IRB, V, AllocaTy);
2856 } else {
2857 // Established these invariants above.
2858 assert(NewBeginOffset == NewAllocaBeginOffset)((NewBeginOffset == NewAllocaBeginOffset) ? static_cast<void
> (0) : __assert_fail ("NewBeginOffset == NewAllocaBeginOffset"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2858, __PRETTY_FUNCTION__))
;
2859 assert(NewEndOffset == NewAllocaEndOffset)((NewEndOffset == NewAllocaEndOffset) ? static_cast<void>
(0) : __assert_fail ("NewEndOffset == NewAllocaEndOffset", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2859, __PRETTY_FUNCTION__))
;
2860
2861 V = getIntegerSplat(II.getValue(), DL.getTypeSizeInBits(ScalarTy) / 8);
2862 if (VectorType *AllocaVecTy = dyn_cast<VectorType>(AllocaTy))
2863 V = getVectorSplat(V, AllocaVecTy->getNumElements());
2864
2865 V = convertValue(DL, IRB, V, AllocaTy);
2866 }
2867
2868 StoreInst *New = IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlignment(),
2869 II.isVolatile());
2870 if (AATags)
2871 New->setAAMetadata(AATags);
2872 LLVM_DEBUG(dbgs() << " to: " << *New << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *New <<
"\n"; } } while (false)
;
2873 return !II.isVolatile();
2874 }
2875
2876 bool visitMemTransferInst(MemTransferInst &II) {
2877 // Rewriting of memory transfer instructions can be a bit tricky. We break
2878 // them into two categories: split intrinsics and unsplit intrinsics.
2879
2880 LLVM_DEBUG(dbgs() << " original: " << II << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << II <<
"\n"; } } while (false)
;
1
Assuming 'DebugFlag' is false
2
Loop condition is false. Exiting loop
2881
2882 AAMDNodes AATags;
2883 II.getAAMetadata(AATags);
2884
2885 bool IsDest = &II.getRawDestUse() == OldUse;
2886 assert
3.1
'IsDest' is false
3.1
'IsDest' is false
3.1
'IsDest' is false
((IsDest && II.getRawDest() == OldPtr) ||(((IsDest && II.getRawDest() == OldPtr) || (!IsDest &&
II.getRawSource() == OldPtr)) ? static_cast<void> (0) :
__assert_fail ("(IsDest && II.getRawDest() == OldPtr) || (!IsDest && II.getRawSource() == OldPtr)"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2887, __PRETTY_FUNCTION__))
3
Assuming 'IsDest' is false
4
Assuming the condition is true
5
'?' condition is true
2887 (!IsDest && II.getRawSource() == OldPtr))(((IsDest && II.getRawDest() == OldPtr) || (!IsDest &&
II.getRawSource() == OldPtr)) ? static_cast<void> (0) :
__assert_fail ("(IsDest && II.getRawDest() == OldPtr) || (!IsDest && II.getRawSource() == OldPtr)"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2887, __PRETTY_FUNCTION__))
;
2888
2889 unsigned SliceAlign = getSliceAlign();
2890
2891 // For unsplit intrinsics, we simply modify the source and destination
2892 // pointers in place. This isn't just an optimization, it is a matter of
2893 // correctness. With unsplit intrinsics we may be dealing with transfers
2894 // within a single alloca before SROA ran, or with transfers that have
2895 // a variable length. We may also be dealing with memmove instead of
2896 // memcpy, and so simply updating the pointers is the necessary for us to
2897 // update both source and dest of a single call.
2898 if (!IsSplittable) {
6
Assuming field 'IsSplittable' is true
7
Taking false branch
2899 Value *AdjustedPtr = getNewAllocaSlicePtr(IRB, OldPtr->getType());
2900 if (IsDest) {
2901 II.setDest(AdjustedPtr);
2902 II.setDestAlignment(SliceAlign);
2903 }
2904 else {
2905 II.setSource(AdjustedPtr);
2906 II.setSourceAlignment(SliceAlign);
2907 }
2908
2909 LLVM_DEBUG(dbgs() << " to: " << II << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << II <<
"\n"; } } while (false)
;
2910 deleteIfTriviallyDead(OldPtr);
2911 return false;
2912 }
2913 // For split transfer intrinsics we have an incredibly useful assurance:
2914 // the source and destination do not reside within the same alloca, and at
2915 // least one of them does not escape. This means that we can replace
2916 // memmove with memcpy, and we don't need to worry about all manner of
2917 // downsides to splitting and transforming the operations.
2918
2919 // If this doesn't map cleanly onto the alloca type, and that type isn't
2920 // a single value type, just emit a memcpy.
2921 bool EmitMemCpy =
2922 !VecTy && !IntTy &&
8
Assuming field 'VecTy' is non-null
2923 (BeginOffset > NewAllocaBeginOffset || EndOffset < NewAllocaEndOffset ||
2924 SliceSize != DL.getTypeStoreSize(NewAI.getAllocatedType()) ||
2925 !NewAI.getAllocatedType()->isSingleValueType());
2926
2927 // If we're just going to emit a memcpy, the alloca hasn't changed, and the
2928 // size hasn't been shrunk based on analysis of the viable range, this is
2929 // a no-op.
2930 if (EmitMemCpy
8.1
'EmitMemCpy' is false
8.1
'EmitMemCpy' is false
8.1
'EmitMemCpy' is false
&& &OldAI == &NewAI) {
2931 // Ensure the start lines up.
2932 assert(NewBeginOffset == BeginOffset)((NewBeginOffset == BeginOffset) ? static_cast<void> (0
) : __assert_fail ("NewBeginOffset == BeginOffset", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2932, __PRETTY_FUNCTION__))
;
2933
2934 // Rewrite the size as needed.
2935 if (NewEndOffset != EndOffset)
2936 II.setLength(ConstantInt::get(II.getLength()->getType(),
2937 NewEndOffset - NewBeginOffset));
2938 return false;
2939 }
2940 // Record this instruction for deletion.
2941 Pass.DeadInsts.insert(&II);
2942
2943 // Strip all inbounds GEPs and pointer casts to try to dig out any root
2944 // alloca that should be re-examined after rewriting this instruction.
2945 Value *OtherPtr = IsDest
8.2
'IsDest' is false
8.2
'IsDest' is false
8.2
'IsDest' is false
? II.getRawSource() : II.getRawDest();
9
'?' condition is false
2946 if (AllocaInst *AI
10.1
'AI' is null
10.1
'AI' is null
10.1
'AI' is null
=
11
Taking false branch
2947 dyn_cast<AllocaInst>(OtherPtr->stripInBoundsOffsets())) {
10
Assuming the object is not a 'AllocaInst'
2948 assert(AI != &OldAI && AI != &NewAI &&((AI != &OldAI && AI != &NewAI && "Splittable transfers cannot reach the same alloca on both ends."
) ? static_cast<void> (0) : __assert_fail ("AI != &OldAI && AI != &NewAI && \"Splittable transfers cannot reach the same alloca on both ends.\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2949, __PRETTY_FUNCTION__))
2949 "Splittable transfers cannot reach the same alloca on both ends.")((AI != &OldAI && AI != &NewAI && "Splittable transfers cannot reach the same alloca on both ends."
) ? static_cast<void> (0) : __assert_fail ("AI != &OldAI && AI != &NewAI && \"Splittable transfers cannot reach the same alloca on both ends.\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 2949, __PRETTY_FUNCTION__))
;
2950 Pass.Worklist.insert(AI);
2951 }
2952
2953 Type *OtherPtrTy = OtherPtr->getType();
2954 unsigned OtherAS = OtherPtrTy->getPointerAddressSpace();
2955
2956 // Compute the relative offset for the other pointer within the transfer.
2957 unsigned OffsetWidth = DL.getIndexSizeInBits(OtherAS);
2958 APInt OtherOffset(OffsetWidth, NewBeginOffset - BeginOffset);
2959 unsigned OtherAlign =
2960 IsDest
11.1
'IsDest' is false
11.1
'IsDest' is false
11.1
'IsDest' is false
? II.getSourceAlignment() : II.getDestAlignment();
12
'?' condition is false
2961 OtherAlign = MinAlign(OtherAlign ? OtherAlign : 1,
13
Assuming 'OtherAlign' is 0
14
'?' condition is false
2962 OtherOffset.zextOrTrunc(64).getZExtValue());
2963
2964 if (EmitMemCpy
14.1
'EmitMemCpy' is false
14.1
'EmitMemCpy' is false
14.1
'EmitMemCpy' is false
) {
15
Taking false branch
2965 // Compute the other pointer, folding as much as possible to produce
2966 // a single, simple GEP in most cases.
2967 OtherPtr = getAdjustedPtr(IRB, DL, OtherPtr, OtherOffset, OtherPtrTy,
2968 OtherPtr->getName() + ".");
2969
2970 Value *OurPtr = getNewAllocaSlicePtr(IRB, OldPtr->getType());
2971 Type *SizeTy = II.getLength()->getType();
2972 Constant *Size = ConstantInt::get(SizeTy, NewEndOffset - NewBeginOffset);
2973
2974 Value *DestPtr, *SrcPtr;
2975 unsigned DestAlign, SrcAlign;
2976 // Note: IsDest is true iff we're copying into the new alloca slice
2977 if (IsDest) {
2978 DestPtr = OurPtr;
2979 DestAlign = SliceAlign;
2980 SrcPtr = OtherPtr;
2981 SrcAlign = OtherAlign;
2982 } else {
2983 DestPtr = OtherPtr;
2984 DestAlign = OtherAlign;
2985 SrcPtr = OurPtr;
2986 SrcAlign = SliceAlign;
2987 }
2988 CallInst *New = IRB.CreateMemCpy(DestPtr, DestAlign, SrcPtr, SrcAlign,
2989 Size, II.isVolatile());
2990 if (AATags)
2991 New->setAAMetadata(AATags);
2992 LLVM_DEBUG(dbgs() << " to: " << *New << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *New <<
"\n"; } } while (false)
;
2993 return false;
2994 }
2995
2996 bool IsWholeAlloca = NewBeginOffset == NewAllocaBeginOffset &&
16
Assuming field 'NewBeginOffset' is not equal to field 'NewAllocaBeginOffset'
2997 NewEndOffset == NewAllocaEndOffset;
2998 uint64_t Size = NewEndOffset - NewBeginOffset;
2999 unsigned BeginIndex = VecTy
16.1
Field 'VecTy' is non-null
16.1
Field 'VecTy' is non-null
16.1
Field 'VecTy' is non-null
? getIndex(NewBeginOffset) : 0;
17
'?' condition is true
3000 unsigned EndIndex = VecTy
17.1
Field 'VecTy' is non-null
17.1
Field 'VecTy' is non-null
17.1
Field 'VecTy' is non-null
? getIndex(NewEndOffset) : 0;
18
'?' condition is true
3001 unsigned NumElements = EndIndex - BeginIndex;
3002 IntegerType *SubIntTy =
21
'SubIntTy' initialized to a null pointer value
3003 IntTy ? Type::getIntNTy(IntTy->getContext(), Size * 8) : nullptr;
19
Assuming field 'IntTy' is null
20
'?' condition is false
3004
3005 // Reset the other pointer type to match the register type we're going to
3006 // use, but using the address space of the original other pointer.
3007 Type *OtherTy;
3008 if (VecTy
21.1
Field 'VecTy' is non-null
21.1
Field 'VecTy' is non-null
21.1
Field 'VecTy' is non-null
&& !IsWholeAlloca
21.2
'IsWholeAlloca' is false
21.2
'IsWholeAlloca' is false
21.2
'IsWholeAlloca' is false
) {
22
Taking true branch
3009 if (NumElements == 1)
23
Assuming 'NumElements' is not equal to 1
24
Taking false branch
3010 OtherTy = VecTy->getElementType();
3011 else
3012 OtherTy = VectorType::get(VecTy->getElementType(), NumElements);
3013 } else if (IntTy && !IsWholeAlloca) {
3014 OtherTy = SubIntTy;
3015 } else {
3016 OtherTy = NewAllocaTy;
3017 }
3018 OtherPtrTy = OtherTy->getPointerTo(OtherAS);
3019
3020 Value *SrcPtr = getAdjustedPtr(IRB, DL, OtherPtr, OtherOffset, OtherPtrTy,
25
Calling 'getAdjustedPtr'
58
Returning from 'getAdjustedPtr'
3021 OtherPtr->getName() + ".");
3022 unsigned SrcAlign = OtherAlign;
3023 Value *DstPtr = &NewAI;
3024 unsigned DstAlign = SliceAlign;
3025 if (!IsDest
58.1
'IsDest' is false
58.1
'IsDest' is false
58.1
'IsDest' is false
) {
59
Taking true branch
3026 std::swap(SrcPtr, DstPtr);
3027 std::swap(SrcAlign, DstAlign);
3028 }
3029
3030 Value *Src;
3031 if (VecTy && !IsWholeAlloca && !IsDest) {
60
Assuming field 'VecTy' is null
3032 Src = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
3033 NewAI.getAlignment(), "load");
3034 Src = extractVector(IRB, Src, BeginIndex, EndIndex, "vec");
3035 } else if (IntTy && !IsWholeAlloca
61.1
'IsWholeAlloca' is false
61.1
'IsWholeAlloca' is false
61.1
'IsWholeAlloca' is false
&& !IsDest
61.2
'IsDest' is false
61.2
'IsDest' is false
61.2
'IsDest' is false
) {
61
Assuming field 'IntTy' is non-null
62
Taking true branch
3036 Src = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
3037 NewAI.getAlignment(), "load");
3038 Src = convertValue(DL, IRB, Src, IntTy);
3039 uint64_t Offset = NewBeginOffset - NewAllocaBeginOffset;
3040 Src = extractInteger(DL, IRB, Src, SubIntTy, Offset, "extract");
63
Passing null pointer value via 4th parameter 'Ty'
64
Calling 'extractInteger'
3041 } else {
3042 LoadInst *Load = IRB.CreateAlignedLoad(OtherTy, SrcPtr, SrcAlign,
3043 II.isVolatile(), "copyload");
3044 if (AATags)
3045 Load->setAAMetadata(AATags);
3046 Src = Load;
3047 }
3048
3049 if (VecTy && !IsWholeAlloca && IsDest) {
3050 Value *Old = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
3051 NewAI.getAlignment(), "oldload");
3052 Src = insertVector(IRB, Old, Src, BeginIndex, "vec");
3053 } else if (IntTy && !IsWholeAlloca && IsDest) {
3054 Value *Old = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
3055 NewAI.getAlignment(), "oldload");
3056 Old = convertValue(DL, IRB, Old, IntTy);
3057 uint64_t Offset = NewBeginOffset - NewAllocaBeginOffset;
3058 Src = insertInteger(DL, IRB, Old, Src, Offset, "insert");
3059 Src = convertValue(DL, IRB, Src, NewAllocaTy);
3060 }
3061
3062 StoreInst *Store = cast<StoreInst>(
3063 IRB.CreateAlignedStore(Src, DstPtr, DstAlign, II.isVolatile()));
3064 if (AATags)
3065 Store->setAAMetadata(AATags);
3066 LLVM_DEBUG(dbgs() << " to: " << *Store << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *Store <<
"\n"; } } while (false)
;
3067 return !II.isVolatile();
3068 }
3069
3070 bool visitIntrinsicInst(IntrinsicInst &II) {
3071 assert(II.isLifetimeStartOrEnd())((II.isLifetimeStartOrEnd()) ? static_cast<void> (0) : __assert_fail
("II.isLifetimeStartOrEnd()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3071, __PRETTY_FUNCTION__))
;
3072 LLVM_DEBUG(dbgs() << " original: " << II << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << II <<
"\n"; } } while (false)
;
3073 assert(II.getArgOperand(1) == OldPtr)((II.getArgOperand(1) == OldPtr) ? static_cast<void> (0
) : __assert_fail ("II.getArgOperand(1) == OldPtr", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3073, __PRETTY_FUNCTION__))
;
3074
3075 // Record this instruction for deletion.
3076 Pass.DeadInsts.insert(&II);
3077
3078 // Lifetime intrinsics are only promotable if they cover the whole alloca.
3079 // Therefore, we drop lifetime intrinsics which don't cover the whole
3080 // alloca.
3081 // (In theory, intrinsics which partially cover an alloca could be
3082 // promoted, but PromoteMemToReg doesn't handle that case.)
3083 // FIXME: Check whether the alloca is promotable before dropping the
3084 // lifetime intrinsics?
3085 if (NewBeginOffset != NewAllocaBeginOffset ||
3086 NewEndOffset != NewAllocaEndOffset)
3087 return true;
3088
3089 ConstantInt *Size =
3090 ConstantInt::get(cast<IntegerType>(II.getArgOperand(0)->getType()),
3091 NewEndOffset - NewBeginOffset);
3092 // Lifetime intrinsics always expect an i8* so directly get such a pointer
3093 // for the new alloca slice.
3094 Type *PointerTy = IRB.getInt8PtrTy(OldPtr->getType()->getPointerAddressSpace());
3095 Value *Ptr = getNewAllocaSlicePtr(IRB, PointerTy);
3096 Value *New;
3097 if (II.getIntrinsicID() == Intrinsic::lifetime_start)
3098 New = IRB.CreateLifetimeStart(Ptr, Size);
3099 else
3100 New = IRB.CreateLifetimeEnd(Ptr, Size);
3101
3102 (void)New;
3103 LLVM_DEBUG(dbgs() << " to: " << *New << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *New <<
"\n"; } } while (false)
;
3104
3105 return true;
3106 }
3107
3108 void fixLoadStoreAlign(Instruction &Root) {
3109 // This algorithm implements the same visitor loop as
3110 // hasUnsafePHIOrSelectUse, and fixes the alignment of each load
3111 // or store found.
3112 SmallPtrSet<Instruction *, 4> Visited;
3113 SmallVector<Instruction *, 4> Uses;
3114 Visited.insert(&Root);
3115 Uses.push_back(&Root);
3116 do {
3117 Instruction *I = Uses.pop_back_val();
3118
3119 if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
3120 unsigned LoadAlign = LI->getAlignment();
3121 if (!LoadAlign)
3122 LoadAlign = DL.getABITypeAlignment(LI->getType());
3123 LI->setAlignment(MaybeAlign(std::min(LoadAlign, getSliceAlign())));
3124 continue;
3125 }
3126 if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
3127 unsigned StoreAlign = SI->getAlignment();
3128 if (!StoreAlign) {
3129 Value *Op = SI->getOperand(0);
3130 StoreAlign = DL.getABITypeAlignment(Op->getType());
3131 }
3132 SI->setAlignment(MaybeAlign(std::min(StoreAlign, getSliceAlign())));
3133 continue;
3134 }
3135
3136 assert(isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I) ||((isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I
) || isa<PHINode>(I) || isa<SelectInst>(I) || isa
<GetElementPtrInst>(I)) ? static_cast<void> (0) :
__assert_fail ("isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I) || isa<PHINode>(I) || isa<SelectInst>(I) || isa<GetElementPtrInst>(I)"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3138, __PRETTY_FUNCTION__))
3137 isa<PHINode>(I) || isa<SelectInst>(I) ||((isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I
) || isa<PHINode>(I) || isa<SelectInst>(I) || isa
<GetElementPtrInst>(I)) ? static_cast<void> (0) :
__assert_fail ("isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I) || isa<PHINode>(I) || isa<SelectInst>(I) || isa<GetElementPtrInst>(I)"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3138, __PRETTY_FUNCTION__))
3138 isa<GetElementPtrInst>(I))((isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I
) || isa<PHINode>(I) || isa<SelectInst>(I) || isa
<GetElementPtrInst>(I)) ? static_cast<void> (0) :
__assert_fail ("isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I) || isa<PHINode>(I) || isa<SelectInst>(I) || isa<GetElementPtrInst>(I)"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3138, __PRETTY_FUNCTION__))
;
3139 for (User *U : I->users())
3140 if (Visited.insert(cast<Instruction>(U)).second)
3141 Uses.push_back(cast<Instruction>(U));
3142 } while (!Uses.empty());
3143 }
3144
3145 bool visitPHINode(PHINode &PN) {
3146 LLVM_DEBUG(dbgs() << " original: " << PN << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << PN <<
"\n"; } } while (false)
;
3147 assert(BeginOffset >= NewAllocaBeginOffset && "PHIs are unsplittable")((BeginOffset >= NewAllocaBeginOffset && "PHIs are unsplittable"
) ? static_cast<void> (0) : __assert_fail ("BeginOffset >= NewAllocaBeginOffset && \"PHIs are unsplittable\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3147, __PRETTY_FUNCTION__))
;
3148 assert(EndOffset <= NewAllocaEndOffset && "PHIs are unsplittable")((EndOffset <= NewAllocaEndOffset && "PHIs are unsplittable"
) ? static_cast<void> (0) : __assert_fail ("EndOffset <= NewAllocaEndOffset && \"PHIs are unsplittable\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3148, __PRETTY_FUNCTION__))
;
3149
3150 // We would like to compute a new pointer in only one place, but have it be
3151 // as local as possible to the PHI. To do that, we re-use the location of
3152 // the old pointer, which necessarily must be in the right position to
3153 // dominate the PHI.
3154 IRBuilderTy PtrBuilder(IRB);
3155 if (isa<PHINode>(OldPtr))
3156 PtrBuilder.SetInsertPoint(&*OldPtr->getParent()->getFirstInsertionPt());
3157 else
3158 PtrBuilder.SetInsertPoint(OldPtr);
3159 PtrBuilder.SetCurrentDebugLocation(OldPtr->getDebugLoc());
3160
3161 Value *NewPtr = getNewAllocaSlicePtr(PtrBuilder, OldPtr->getType());
3162 // Replace the operands which were using the old pointer.
3163 std::replace(PN.op_begin(), PN.op_end(), cast<Value>(OldPtr), NewPtr);
3164
3165 LLVM_DEBUG(dbgs() << " to: " << PN << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << PN <<
"\n"; } } while (false)
;
3166 deleteIfTriviallyDead(OldPtr);
3167
3168 // Fix the alignment of any loads or stores using this PHI node.
3169 fixLoadStoreAlign(PN);
3170
3171 // PHIs can't be promoted on their own, but often can be speculated. We
3172 // check the speculation outside of the rewriter so that we see the
3173 // fully-rewritten alloca.
3174 PHIUsers.insert(&PN);
3175 return true;
3176 }
3177
3178 bool visitSelectInst(SelectInst &SI) {
3179 LLVM_DEBUG(dbgs() << " original: " << SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << SI <<
"\n"; } } while (false)
;
3180 assert((SI.getTrueValue() == OldPtr || SI.getFalseValue() == OldPtr) &&(((SI.getTrueValue() == OldPtr || SI.getFalseValue() == OldPtr
) && "Pointer isn't an operand!") ? static_cast<void
> (0) : __assert_fail ("(SI.getTrueValue() == OldPtr || SI.getFalseValue() == OldPtr) && \"Pointer isn't an operand!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3181, __PRETTY_FUNCTION__))
3181 "Pointer isn't an operand!")(((SI.getTrueValue() == OldPtr || SI.getFalseValue() == OldPtr
) && "Pointer isn't an operand!") ? static_cast<void
> (0) : __assert_fail ("(SI.getTrueValue() == OldPtr || SI.getFalseValue() == OldPtr) && \"Pointer isn't an operand!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3181, __PRETTY_FUNCTION__))
;
3182 assert(BeginOffset >= NewAllocaBeginOffset && "Selects are unsplittable")((BeginOffset >= NewAllocaBeginOffset && "Selects are unsplittable"
) ? static_cast<void> (0) : __assert_fail ("BeginOffset >= NewAllocaBeginOffset && \"Selects are unsplittable\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3182, __PRETTY_FUNCTION__))
;
3183 assert(EndOffset <= NewAllocaEndOffset && "Selects are unsplittable")((EndOffset <= NewAllocaEndOffset && "Selects are unsplittable"
) ? static_cast<void> (0) : __assert_fail ("EndOffset <= NewAllocaEndOffset && \"Selects are unsplittable\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3183, __PRETTY_FUNCTION__))
;
3184
3185 Value *NewPtr = getNewAllocaSlicePtr(IRB, OldPtr->getType());
3186 // Replace the operands which were using the old pointer.
3187 if (SI.getOperand(1) == OldPtr)
3188 SI.setOperand(1, NewPtr);
3189 if (SI.getOperand(2) == OldPtr)
3190 SI.setOperand(2, NewPtr);
3191
3192 LLVM_DEBUG(dbgs() << " to: " << SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << SI <<
"\n"; } } while (false)
;
3193 deleteIfTriviallyDead(OldPtr);
3194
3195 // Fix the alignment of any loads or stores using this select.
3196 fixLoadStoreAlign(SI);
3197
3198 // Selects can't be promoted on their own, but often can be speculated. We
3199 // check the speculation outside of the rewriter so that we see the
3200 // fully-rewritten alloca.
3201 SelectUsers.insert(&SI);
3202 return true;
3203 }
3204};
3205
3206namespace {
3207
3208/// Visitor to rewrite aggregate loads and stores as scalar.
3209///
3210/// This pass aggressively rewrites all aggregate loads and stores on
3211/// a particular pointer (or any pointer derived from it which we can identify)
3212/// with scalar loads and stores.
3213class AggLoadStoreRewriter : public InstVisitor<AggLoadStoreRewriter, bool> {
3214 // Befriend the base class so it can delegate to private visit methods.
3215 friend class InstVisitor<AggLoadStoreRewriter, bool>;
3216
3217 /// Queue of pointer uses to analyze and potentially rewrite.
3218 SmallVector<Use *, 8> Queue;
3219
3220 /// Set to prevent us from cycling with phi nodes and loops.
3221 SmallPtrSet<User *, 8> Visited;
3222
3223 /// The current pointer use being rewritten. This is used to dig up the used
3224 /// value (as opposed to the user).
3225 Use *U;
3226
3227 /// Used to calculate offsets, and hence alignment, of subobjects.
3228 const DataLayout &DL;
3229
3230public:
3231 AggLoadStoreRewriter(const DataLayout &DL) : DL(DL) {}
3232
3233 /// Rewrite loads and stores through a pointer and all pointers derived from
3234 /// it.
3235 bool rewrite(Instruction &I) {
3236 LLVM_DEBUG(dbgs() << " Rewriting FCA loads and stores...\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Rewriting FCA loads and stores...\n"
; } } while (false)
;
3237 enqueueUsers(I);
3238 bool Changed = false;
3239 while (!Queue.empty()) {
3240 U = Queue.pop_back_val();
3241 Changed |= visit(cast<Instruction>(U->getUser()));
3242 }
3243 return Changed;
3244 }
3245
3246private:
3247 /// Enqueue all the users of the given instruction for further processing.
3248 /// This uses a set to de-duplicate users.
3249 void enqueueUsers(Instruction &I) {
3250 for (Use &U : I.uses())
3251 if (Visited.insert(U.getUser()).second)
3252 Queue.push_back(&U);
3253 }
3254
3255 // Conservative default is to not rewrite anything.
3256 bool visitInstruction(Instruction &I) { return false; }
3257
3258 /// Generic recursive split emission class.
3259 template <typename Derived> class OpSplitter {
3260 protected:
3261 /// The builder used to form new instructions.
3262 IRBuilderTy IRB;
3263
3264 /// The indices which to be used with insert- or extractvalue to select the
3265 /// appropriate value within the aggregate.
3266 SmallVector<unsigned, 4> Indices;
3267
3268 /// The indices to a GEP instruction which will move Ptr to the correct slot
3269 /// within the aggregate.
3270 SmallVector<Value *, 4> GEPIndices;
3271
3272 /// The base pointer of the original op, used as a base for GEPing the
3273 /// split operations.
3274 Value *Ptr;
3275
3276 /// The base pointee type being GEPed into.
3277 Type *BaseTy;
3278
3279 /// Known alignment of the base pointer.
3280 unsigned BaseAlign;
3281
3282 /// To calculate offset of each component so we can correctly deduce
3283 /// alignments.
3284 const DataLayout &DL;
3285
3286 /// Initialize the splitter with an insertion point, Ptr and start with a
3287 /// single zero GEP index.
3288 OpSplitter(Instruction *InsertionPoint, Value *Ptr, Type *BaseTy,
3289 unsigned BaseAlign, const DataLayout &DL)
3290 : IRB(InsertionPoint), GEPIndices(1, IRB.getInt32(0)), Ptr(Ptr),
3291 BaseTy(BaseTy), BaseAlign(BaseAlign), DL(DL) {}
3292
3293 public:
3294 /// Generic recursive split emission routine.
3295 ///
3296 /// This method recursively splits an aggregate op (load or store) into
3297 /// scalar or vector ops. It splits recursively until it hits a single value
3298 /// and emits that single value operation via the template argument.
3299 ///
3300 /// The logic of this routine relies on GEPs and insertvalue and
3301 /// extractvalue all operating with the same fundamental index list, merely
3302 /// formatted differently (GEPs need actual values).
3303 ///
3304 /// \param Ty The type being split recursively into smaller ops.
3305 /// \param Agg The aggregate value being built up or stored, depending on
3306 /// whether this is splitting a load or a store respectively.
3307 void emitSplitOps(Type *Ty, Value *&Agg, const Twine &Name) {
3308 if (Ty->isSingleValueType()) {
3309 unsigned Offset = DL.getIndexedOffsetInType(BaseTy, GEPIndices);
3310 return static_cast<Derived *>(this)->emitFunc(
3311 Ty, Agg, MinAlign(BaseAlign, Offset), Name);
3312 }
3313
3314 if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
3315 unsigned OldSize = Indices.size();
3316 (void)OldSize;
3317 for (unsigned Idx = 0, Size = ATy->getNumElements(); Idx != Size;
3318 ++Idx) {
3319 assert(Indices.size() == OldSize && "Did not return to the old size")((Indices.size() == OldSize && "Did not return to the old size"
) ? static_cast<void> (0) : __assert_fail ("Indices.size() == OldSize && \"Did not return to the old size\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3319, __PRETTY_FUNCTION__))
;
3320 Indices.push_back(Idx);
3321 GEPIndices.push_back(IRB.getInt32(Idx));
3322 emitSplitOps(ATy->getElementType(), Agg, Name + "." + Twine(Idx));
3323 GEPIndices.pop_back();
3324 Indices.pop_back();
3325 }
3326 return;
3327 }
3328
3329 if (StructType *STy = dyn_cast<StructType>(Ty)) {
3330 unsigned OldSize = Indices.size();
3331 (void)OldSize;
3332 for (unsigned Idx = 0, Size = STy->getNumElements(); Idx != Size;
3333 ++Idx) {
3334 assert(Indices.size() == OldSize && "Did not return to the old size")((Indices.size() == OldSize && "Did not return to the old size"
) ? static_cast<void> (0) : __assert_fail ("Indices.size() == OldSize && \"Did not return to the old size\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3334, __PRETTY_FUNCTION__))
;
3335 Indices.push_back(Idx);
3336 GEPIndices.push_back(IRB.getInt32(Idx));
3337 emitSplitOps(STy->getElementType(Idx), Agg, Name + "." + Twine(Idx));
3338 GEPIndices.pop_back();
3339 Indices.pop_back();
3340 }
3341 return;
3342 }
3343
3344 llvm_unreachable("Only arrays and structs are aggregate loadable types")::llvm::llvm_unreachable_internal("Only arrays and structs are aggregate loadable types"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3344)
;
3345 }
3346 };
3347
3348 struct LoadOpSplitter : public OpSplitter<LoadOpSplitter> {
3349 AAMDNodes AATags;
3350
3351 LoadOpSplitter(Instruction *InsertionPoint, Value *Ptr, Type *BaseTy,
3352 AAMDNodes AATags, unsigned BaseAlign, const DataLayout &DL)
3353 : OpSplitter<LoadOpSplitter>(InsertionPoint, Ptr, BaseTy, BaseAlign,
3354 DL), AATags(AATags) {}
3355
3356 /// Emit a leaf load of a single value. This is called at the leaves of the
3357 /// recursive emission to actually load values.
3358 void emitFunc(Type *Ty, Value *&Agg, unsigned Align, const Twine &Name) {
3359 assert(Ty->isSingleValueType())((Ty->isSingleValueType()) ? static_cast<void> (0) :
__assert_fail ("Ty->isSingleValueType()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3359, __PRETTY_FUNCTION__))
;
3360 // Load the single value and insert it using the indices.
3361 Value *GEP =
3362 IRB.CreateInBoundsGEP(BaseTy, Ptr, GEPIndices, Name + ".gep");
3363 LoadInst *Load = IRB.CreateAlignedLoad(Ty, GEP, Align, Name + ".load");
3364 if (AATags)
3365 Load->setAAMetadata(AATags);
3366 Agg = IRB.CreateInsertValue(Agg, Load, Indices, Name + ".insert");
3367 LLVM_DEBUG(dbgs() << " to: " << *Load << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *Load <<
"\n"; } } while (false)
;
3368 }
3369 };
3370
3371 bool visitLoadInst(LoadInst &LI) {
3372 assert(LI.getPointerOperand() == *U)((LI.getPointerOperand() == *U) ? static_cast<void> (0)
: __assert_fail ("LI.getPointerOperand() == *U", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3372, __PRETTY_FUNCTION__))
;
3373 if (!LI.isSimple() || LI.getType()->isSingleValueType())
3374 return false;
3375
3376 // We have an aggregate being loaded, split it apart.
3377 LLVM_DEBUG(dbgs() << " original: " << LI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << LI <<
"\n"; } } while (false)
;
3378 AAMDNodes AATags;
3379 LI.getAAMetadata(AATags);
3380 LoadOpSplitter Splitter(&LI, *U, LI.getType(), AATags,
3381 getAdjustedAlignment(&LI, 0, DL), DL);
3382 Value *V = UndefValue::get(LI.getType());
3383 Splitter.emitSplitOps(LI.getType(), V, LI.getName() + ".fca");
3384 LI.replaceAllUsesWith(V);
3385 LI.eraseFromParent();
3386 return true;
3387 }
3388
3389 struct StoreOpSplitter : public OpSplitter<StoreOpSplitter> {
3390 StoreOpSplitter(Instruction *InsertionPoint, Value *Ptr, Type *BaseTy,
3391 AAMDNodes AATags, unsigned BaseAlign, const DataLayout &DL)
3392 : OpSplitter<StoreOpSplitter>(InsertionPoint, Ptr, BaseTy, BaseAlign,
3393 DL),
3394 AATags(AATags) {}
3395 AAMDNodes AATags;
3396 /// Emit a leaf store of a single value. This is called at the leaves of the
3397 /// recursive emission to actually produce stores.
3398 void emitFunc(Type *Ty, Value *&Agg, unsigned Align, const Twine &Name) {
3399 assert(Ty->isSingleValueType())((Ty->isSingleValueType()) ? static_cast<void> (0) :
__assert_fail ("Ty->isSingleValueType()", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3399, __PRETTY_FUNCTION__))
;
3400 // Extract the single value and store it using the indices.
3401 //
3402 // The gep and extractvalue values are factored out of the CreateStore
3403 // call to make the output independent of the argument evaluation order.
3404 Value *ExtractValue =
3405 IRB.CreateExtractValue(Agg, Indices, Name + ".extract");
3406 Value *InBoundsGEP =
3407 IRB.CreateInBoundsGEP(BaseTy, Ptr, GEPIndices, Name + ".gep");
3408 StoreInst *Store =
3409 IRB.CreateAlignedStore(ExtractValue, InBoundsGEP, Align);
3410 if (AATags)
3411 Store->setAAMetadata(AATags);
3412 LLVM_DEBUG(dbgs() << " to: " << *Store << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " to: " << *Store <<
"\n"; } } while (false)
;
3413 }
3414 };
3415
3416 bool visitStoreInst(StoreInst &SI) {
3417 if (!SI.isSimple() || SI.getPointerOperand() != *U)
3418 return false;
3419 Value *V = SI.getValueOperand();
3420 if (V->getType()->isSingleValueType())
3421 return false;
3422
3423 // We have an aggregate being stored, split it apart.
3424 LLVM_DEBUG(dbgs() << " original: " << SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " original: " << SI <<
"\n"; } } while (false)
;
3425 AAMDNodes AATags;
3426 SI.getAAMetadata(AATags);
3427 StoreOpSplitter Splitter(&SI, *U, V->getType(), AATags,
3428 getAdjustedAlignment(&SI, 0, DL), DL);
3429 Splitter.emitSplitOps(V->getType(), V, V->getName() + ".fca");
3430 SI.eraseFromParent();
3431 return true;
3432 }
3433
3434 bool visitBitCastInst(BitCastInst &BC) {
3435 enqueueUsers(BC);
3436 return false;
3437 }
3438
3439 bool visitAddrSpaceCastInst(AddrSpaceCastInst &ASC) {
3440 enqueueUsers(ASC);
3441 return false;
3442 }
3443
3444 bool visitGetElementPtrInst(GetElementPtrInst &GEPI) {
3445 enqueueUsers(GEPI);
3446 return false;
3447 }
3448
3449 bool visitPHINode(PHINode &PN) {
3450 enqueueUsers(PN);
3451 return false;
3452 }
3453
3454 bool visitSelectInst(SelectInst &SI) {
3455 enqueueUsers(SI);
3456 return false;
3457 }
3458};
3459
3460} // end anonymous namespace
3461
3462/// Strip aggregate type wrapping.
3463///
3464/// This removes no-op aggregate types wrapping an underlying type. It will
3465/// strip as many layers of types as it can without changing either the type
3466/// size or the allocated size.
3467static Type *stripAggregateTypeWrapping(const DataLayout &DL, Type *Ty) {
3468 if (Ty->isSingleValueType())
3469 return Ty;
3470
3471 uint64_t AllocSize = DL.getTypeAllocSize(Ty);
3472 uint64_t TypeSize = DL.getTypeSizeInBits(Ty);
3473
3474 Type *InnerTy;
3475 if (ArrayType *ArrTy = dyn_cast<ArrayType>(Ty)) {
3476 InnerTy = ArrTy->getElementType();
3477 } else if (StructType *STy = dyn_cast<StructType>(Ty)) {
3478 const StructLayout *SL = DL.getStructLayout(STy);
3479 unsigned Index = SL->getElementContainingOffset(0);
3480 InnerTy = STy->getElementType(Index);
3481 } else {
3482 return Ty;
3483 }
3484
3485 if (AllocSize > DL.getTypeAllocSize(InnerTy) ||
3486 TypeSize > DL.getTypeSizeInBits(InnerTy))
3487 return Ty;
3488
3489 return stripAggregateTypeWrapping(DL, InnerTy);
3490}
3491
3492/// Try to find a partition of the aggregate type passed in for a given
3493/// offset and size.
3494///
3495/// This recurses through the aggregate type and tries to compute a subtype
3496/// based on the offset and size. When the offset and size span a sub-section
3497/// of an array, it will even compute a new array type for that sub-section,
3498/// and the same for structs.
3499///
3500/// Note that this routine is very strict and tries to find a partition of the
3501/// type which produces the *exact* right offset and size. It is not forgiving
3502/// when the size or offset cause either end of type-based partition to be off.
3503/// Also, this is a best-effort routine. It is reasonable to give up and not
3504/// return a type if necessary.
3505static Type *getTypePartition(const DataLayout &DL, Type *Ty, uint64_t Offset,
3506 uint64_t Size) {
3507 if (Offset == 0 && DL.getTypeAllocSize(Ty) == Size)
3508 return stripAggregateTypeWrapping(DL, Ty);
3509 if (Offset > DL.getTypeAllocSize(Ty) ||
3510 (DL.getTypeAllocSize(Ty) - Offset) < Size)
3511 return nullptr;
3512
3513 if (SequentialType *SeqTy = dyn_cast<SequentialType>(Ty)) {
3514 Type *ElementTy = SeqTy->getElementType();
3515 uint64_t ElementSize = DL.getTypeAllocSize(ElementTy);
3516 uint64_t NumSkippedElements = Offset / ElementSize;
3517 if (NumSkippedElements >= SeqTy->getNumElements())
3518 return nullptr;
3519 Offset -= NumSkippedElements * ElementSize;
3520
3521 // First check if we need to recurse.
3522 if (Offset > 0 || Size < ElementSize) {
3523 // Bail if the partition ends in a different array element.
3524 if ((Offset + Size) > ElementSize)
3525 return nullptr;
3526 // Recurse through the element type trying to peel off offset bytes.
3527 return getTypePartition(DL, ElementTy, Offset, Size);
3528 }
3529 assert(Offset == 0)((Offset == 0) ? static_cast<void> (0) : __assert_fail (
"Offset == 0", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3529, __PRETTY_FUNCTION__))
;
3530
3531 if (Size == ElementSize)
3532 return stripAggregateTypeWrapping(DL, ElementTy);
3533 assert(Size > ElementSize)((Size > ElementSize) ? static_cast<void> (0) : __assert_fail
("Size > ElementSize", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3533, __PRETTY_FUNCTION__))
;
3534 uint64_t NumElements = Size / ElementSize;
3535 if (NumElements * ElementSize != Size)
3536 return nullptr;
3537 return ArrayType::get(ElementTy, NumElements);
3538 }
3539
3540 StructType *STy = dyn_cast<StructType>(Ty);
3541 if (!STy)
3542 return nullptr;
3543
3544 const StructLayout *SL = DL.getStructLayout(STy);
3545 if (Offset >= SL->getSizeInBytes())
3546 return nullptr;
3547 uint64_t EndOffset = Offset + Size;
3548 if (EndOffset > SL->getSizeInBytes())
3549 return nullptr;
3550
3551 unsigned Index = SL->getElementContainingOffset(Offset);
3552 Offset -= SL->getElementOffset(Index);
3553
3554 Type *ElementTy = STy->getElementType(Index);
3555 uint64_t ElementSize = DL.getTypeAllocSize(ElementTy);
3556 if (Offset >= ElementSize)
3557 return nullptr; // The offset points into alignment padding.
3558
3559 // See if any partition must be contained by the element.
3560 if (Offset > 0 || Size < ElementSize) {
3561 if ((Offset + Size) > ElementSize)
3562 return nullptr;
3563 return getTypePartition(DL, ElementTy, Offset, Size);
3564 }
3565 assert(Offset == 0)((Offset == 0) ? static_cast<void> (0) : __assert_fail (
"Offset == 0", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3565, __PRETTY_FUNCTION__))
;
3566
3567 if (Size == ElementSize)
3568 return stripAggregateTypeWrapping(DL, ElementTy);
3569
3570 StructType::element_iterator EI = STy->element_begin() + Index,
3571 EE = STy->element_end();
3572 if (EndOffset < SL->getSizeInBytes()) {
3573 unsigned EndIndex = SL->getElementContainingOffset(EndOffset);
3574 if (Index == EndIndex)
3575 return nullptr; // Within a single element and its padding.
3576
3577 // Don't try to form "natural" types if the elements don't line up with the
3578 // expected size.
3579 // FIXME: We could potentially recurse down through the last element in the
3580 // sub-struct to find a natural end point.
3581 if (SL->getElementOffset(EndIndex) != EndOffset)
3582 return nullptr;
3583
3584 assert(Index < EndIndex)((Index < EndIndex) ? static_cast<void> (0) : __assert_fail
("Index < EndIndex", "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3584, __PRETTY_FUNCTION__))
;
3585 EE = STy->element_begin() + EndIndex;
3586 }
3587
3588 // Try to build up a sub-structure.
3589 StructType *SubTy =
3590 StructType::get(STy->getContext(), makeArrayRef(EI, EE), STy->isPacked());
3591 const StructLayout *SubSL = DL.getStructLayout(SubTy);
3592 if (Size != SubSL->getSizeInBytes())
3593 return nullptr; // The sub-struct doesn't have quite the size needed.
3594
3595 return SubTy;
3596}
3597
3598/// Pre-split loads and stores to simplify rewriting.
3599///
3600/// We want to break up the splittable load+store pairs as much as
3601/// possible. This is important to do as a preprocessing step, as once we
3602/// start rewriting the accesses to partitions of the alloca we lose the
3603/// necessary information to correctly split apart paired loads and stores
3604/// which both point into this alloca. The case to consider is something like
3605/// the following:
3606///
3607/// %a = alloca [12 x i8]
3608/// %gep1 = getelementptr [12 x i8]* %a, i32 0, i32 0
3609/// %gep2 = getelementptr [12 x i8]* %a, i32 0, i32 4
3610/// %gep3 = getelementptr [12 x i8]* %a, i32 0, i32 8
3611/// %iptr1 = bitcast i8* %gep1 to i64*
3612/// %iptr2 = bitcast i8* %gep2 to i64*
3613/// %fptr1 = bitcast i8* %gep1 to float*
3614/// %fptr2 = bitcast i8* %gep2 to float*
3615/// %fptr3 = bitcast i8* %gep3 to float*
3616/// store float 0.0, float* %fptr1
3617/// store float 1.0, float* %fptr2
3618/// %v = load i64* %iptr1
3619/// store i64 %v, i64* %iptr2
3620/// %f1 = load float* %fptr2
3621/// %f2 = load float* %fptr3
3622///
3623/// Here we want to form 3 partitions of the alloca, each 4 bytes large, and
3624/// promote everything so we recover the 2 SSA values that should have been
3625/// there all along.
3626///
3627/// \returns true if any changes are made.
3628bool SROA::presplitLoadsAndStores(AllocaInst &AI, AllocaSlices &AS) {
3629 LLVM_DEBUG(dbgs() << "Pre-splitting loads and stores\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "Pre-splitting loads and stores\n"
; } } while (false)
;
3630
3631 // Track the loads and stores which are candidates for pre-splitting here, in
3632 // the order they first appear during the partition scan. These give stable
3633 // iteration order and a basis for tracking which loads and stores we
3634 // actually split.
3635 SmallVector<LoadInst *, 4> Loads;
3636 SmallVector<StoreInst *, 4> Stores;
3637
3638 // We need to accumulate the splits required of each load or store where we
3639 // can find them via a direct lookup. This is important to cross-check loads
3640 // and stores against each other. We also track the slice so that we can kill
3641 // all the slices that end up split.
3642 struct SplitOffsets {
3643 Slice *S;
3644 std::vector<uint64_t> Splits;
3645 };
3646 SmallDenseMap<Instruction *, SplitOffsets, 8> SplitOffsetsMap;
3647
3648 // Track loads out of this alloca which cannot, for any reason, be pre-split.
3649 // This is important as we also cannot pre-split stores of those loads!
3650 // FIXME: This is all pretty gross. It means that we can be more aggressive
3651 // in pre-splitting when the load feeding the store happens to come from
3652 // a separate alloca. Put another way, the effectiveness of SROA would be
3653 // decreased by a frontend which just concatenated all of its local allocas
3654 // into one big flat alloca. But defeating such patterns is exactly the job
3655 // SROA is tasked with! Sadly, to not have this discrepancy we would have
3656 // change store pre-splitting to actually force pre-splitting of the load
3657 // that feeds it *and all stores*. That makes pre-splitting much harder, but
3658 // maybe it would make it more principled?
3659 SmallPtrSet<LoadInst *, 8> UnsplittableLoads;
3660
3661 LLVM_DEBUG(dbgs() << " Searching for candidate loads and stores\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Searching for candidate loads and stores\n"
; } } while (false)
;
3662 for (auto &P : AS.partitions()) {
3663 for (Slice &S : P) {
3664 Instruction *I = cast<Instruction>(S.getUse()->getUser());
3665 if (!S.isSplittable() || S.endOffset() <= P.endOffset()) {
3666 // If this is a load we have to track that it can't participate in any
3667 // pre-splitting. If this is a store of a load we have to track that
3668 // that load also can't participate in any pre-splitting.
3669 if (auto *LI = dyn_cast<LoadInst>(I))
3670 UnsplittableLoads.insert(LI);
3671 else if (auto *SI = dyn_cast<StoreInst>(I))
3672 if (auto *LI = dyn_cast<LoadInst>(SI->getValueOperand()))
3673 UnsplittableLoads.insert(LI);
3674 continue;
3675 }
3676 assert(P.endOffset() > S.beginOffset() &&((P.endOffset() > S.beginOffset() && "Empty or backwards partition!"
) ? static_cast<void> (0) : __assert_fail ("P.endOffset() > S.beginOffset() && \"Empty or backwards partition!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3677, __PRETTY_FUNCTION__))
3677 "Empty or backwards partition!")((P.endOffset() > S.beginOffset() && "Empty or backwards partition!"
) ? static_cast<void> (0) : __assert_fail ("P.endOffset() > S.beginOffset() && \"Empty or backwards partition!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3677, __PRETTY_FUNCTION__))
;
3678
3679 // Determine if this is a pre-splittable slice.
3680 if (auto *LI = dyn_cast<LoadInst>(I)) {
3681 assert(!LI->isVolatile() && "Cannot split volatile loads!")((!LI->isVolatile() && "Cannot split volatile loads!"
) ? static_cast<void> (0) : __assert_fail ("!LI->isVolatile() && \"Cannot split volatile loads!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3681, __PRETTY_FUNCTION__))
;
3682
3683 // The load must be used exclusively to store into other pointers for
3684 // us to be able to arbitrarily pre-split it. The stores must also be
3685 // simple to avoid changing semantics.
3686 auto IsLoadSimplyStored = [](LoadInst *LI) {
3687 for (User *LU : LI->users()) {
3688 auto *SI = dyn_cast<StoreInst>(LU);
3689 if (!SI || !SI->isSimple())
3690 return false;
3691 }
3692 return true;
3693 };
3694 if (!IsLoadSimplyStored(LI)) {
3695 UnsplittableLoads.insert(LI);
3696 continue;
3697 }
3698
3699 Loads.push_back(LI);
3700 } else if (auto *SI = dyn_cast<StoreInst>(I)) {
3701 if (S.getUse() != &SI->getOperandUse(SI->getPointerOperandIndex()))
3702 // Skip stores *of* pointers. FIXME: This shouldn't even be possible!
3703 continue;
3704 auto *StoredLoad = dyn_cast<LoadInst>(SI->getValueOperand());
3705 if (!StoredLoad || !StoredLoad->isSimple())
3706 continue;
3707 assert(!SI->isVolatile() && "Cannot split volatile stores!")((!SI->isVolatile() && "Cannot split volatile stores!"
) ? static_cast<void> (0) : __assert_fail ("!SI->isVolatile() && \"Cannot split volatile stores!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3707, __PRETTY_FUNCTION__))
;
3708
3709 Stores.push_back(SI);
3710 } else {
3711 // Other uses cannot be pre-split.
3712 continue;
3713 }
3714
3715 // Record the initial split.
3716 LLVM_DEBUG(dbgs() << " Candidate: " << *I << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Candidate: " << *I <<
"\n"; } } while (false)
;
3717 auto &Offsets = SplitOffsetsMap[I];
3718 assert(Offsets.Splits.empty() &&((Offsets.Splits.empty() && "Should not have splits the first time we see an instruction!"
) ? static_cast<void> (0) : __assert_fail ("Offsets.Splits.empty() && \"Should not have splits the first time we see an instruction!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3719, __PRETTY_FUNCTION__))
3719 "Should not have splits the first time we see an instruction!")((Offsets.Splits.empty() && "Should not have splits the first time we see an instruction!"
) ? static_cast<void> (0) : __assert_fail ("Offsets.Splits.empty() && \"Should not have splits the first time we see an instruction!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3719, __PRETTY_FUNCTION__))
;
3720 Offsets.S = &S;
3721 Offsets.Splits.push_back(P.endOffset() - S.beginOffset());
3722 }
3723
3724 // Now scan the already split slices, and add a split for any of them which
3725 // we're going to pre-split.
3726 for (Slice *S : P.splitSliceTails()) {
3727 auto SplitOffsetsMapI =
3728 SplitOffsetsMap.find(cast<Instruction>(S->getUse()->getUser()));
3729 if (SplitOffsetsMapI == SplitOffsetsMap.end())
3730 continue;
3731 auto &Offsets = SplitOffsetsMapI->second;
3732
3733 assert(Offsets.S == S && "Found a mismatched slice!")((Offsets.S == S && "Found a mismatched slice!") ? static_cast
<void> (0) : __assert_fail ("Offsets.S == S && \"Found a mismatched slice!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3733, __PRETTY_FUNCTION__))
;
3734 assert(!Offsets.Splits.empty() &&((!Offsets.Splits.empty() && "Cannot have an empty set of splits on the second partition!"
) ? static_cast<void> (0) : __assert_fail ("!Offsets.Splits.empty() && \"Cannot have an empty set of splits on the second partition!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3735, __PRETTY_FUNCTION__))
3735 "Cannot have an empty set of splits on the second partition!")((!Offsets.Splits.empty() && "Cannot have an empty set of splits on the second partition!"
) ? static_cast<void> (0) : __assert_fail ("!Offsets.Splits.empty() && \"Cannot have an empty set of splits on the second partition!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3735, __PRETTY_FUNCTION__))
;
3736 assert(Offsets.Splits.back() ==((Offsets.Splits.back() == P.beginOffset() - Offsets.S->beginOffset
() && "Previous split does not end where this one begins!"
) ? static_cast<void> (0) : __assert_fail ("Offsets.Splits.back() == P.beginOffset() - Offsets.S->beginOffset() && \"Previous split does not end where this one begins!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3738, __PRETTY_FUNCTION__))
3737 P.beginOffset() - Offsets.S->beginOffset() &&((Offsets.Splits.back() == P.beginOffset() - Offsets.S->beginOffset
() && "Previous split does not end where this one begins!"
) ? static_cast<void> (0) : __assert_fail ("Offsets.Splits.back() == P.beginOffset() - Offsets.S->beginOffset() && \"Previous split does not end where this one begins!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3738, __PRETTY_FUNCTION__))
3738 "Previous split does not end where this one begins!")((Offsets.Splits.back() == P.beginOffset() - Offsets.S->beginOffset
() && "Previous split does not end where this one begins!"
) ? static_cast<void> (0) : __assert_fail ("Offsets.Splits.back() == P.beginOffset() - Offsets.S->beginOffset() && \"Previous split does not end where this one begins!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3738, __PRETTY_FUNCTION__))
;
3739
3740 // Record each split. The last partition's end isn't needed as the size
3741 // of the slice dictates that.
3742 if (S->endOffset() > P.endOffset())
3743 Offsets.Splits.push_back(P.endOffset() - Offsets.S->beginOffset());
3744 }
3745 }
3746
3747 // We may have split loads where some of their stores are split stores. For
3748 // such loads and stores, we can only pre-split them if their splits exactly
3749 // match relative to their starting offset. We have to verify this prior to
3750 // any rewriting.
3751 Stores.erase(
3752 llvm::remove_if(Stores,
3753 [&UnsplittableLoads, &SplitOffsetsMap](StoreInst *SI) {
3754 // Lookup the load we are storing in our map of split
3755 // offsets.
3756 auto *LI = cast<LoadInst>(SI->getValueOperand());
3757 // If it was completely unsplittable, then we're done,
3758 // and this store can't be pre-split.
3759 if (UnsplittableLoads.count(LI))
3760 return true;
3761
3762 auto LoadOffsetsI = SplitOffsetsMap.find(LI);
3763 if (LoadOffsetsI == SplitOffsetsMap.end())
3764 return false; // Unrelated loads are definitely safe.
3765 auto &LoadOffsets = LoadOffsetsI->second;
3766
3767 // Now lookup the store's offsets.
3768 auto &StoreOffsets = SplitOffsetsMap[SI];
3769
3770 // If the relative offsets of each split in the load and
3771 // store match exactly, then we can split them and we
3772 // don't need to remove them here.
3773 if (LoadOffsets.Splits == StoreOffsets.Splits)
3774 return false;
3775
3776 LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Mismatched splits for load and store:\n"
<< " " << *LI << "\n" << " "
<< *SI << "\n"; } } while (false)
3777 dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Mismatched splits for load and store:\n"
<< " " << *LI << "\n" << " "
<< *SI << "\n"; } } while (false)
3778 << " Mismatched splits for load and store:\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Mismatched splits for load and store:\n"
<< " " << *LI << "\n" << " "
<< *SI << "\n"; } } while (false)
3779 << " " << *LI << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Mismatched splits for load and store:\n"
<< " " << *LI << "\n" << " "
<< *SI << "\n"; } } while (false)
3780 << " " << *SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Mismatched splits for load and store:\n"
<< " " << *LI << "\n" << " "
<< *SI << "\n"; } } while (false)
;
3781
3782 // We've found a store and load that we need to split
3783 // with mismatched relative splits. Just give up on them
3784 // and remove both instructions from our list of
3785 // candidates.
3786 UnsplittableLoads.insert(LI);
3787 return true;
3788 }),
3789 Stores.end());
3790 // Now we have to go *back* through all the stores, because a later store may
3791 // have caused an earlier store's load to become unsplittable and if it is
3792 // unsplittable for the later store, then we can't rely on it being split in
3793 // the earlier store either.
3794 Stores.erase(llvm::remove_if(Stores,
3795 [&UnsplittableLoads](StoreInst *SI) {
3796 auto *LI =
3797 cast<LoadInst>(SI->getValueOperand());
3798 return UnsplittableLoads.count(LI);
3799 }),
3800 Stores.end());
3801 // Once we've established all the loads that can't be split for some reason,
3802 // filter any that made it into our list out.
3803 Loads.erase(llvm::remove_if(Loads,
3804 [&UnsplittableLoads](LoadInst *LI) {
3805 return UnsplittableLoads.count(LI);
3806 }),
3807 Loads.end());
3808
3809 // If no loads or stores are left, there is no pre-splitting to be done for
3810 // this alloca.
3811 if (Loads.empty() && Stores.empty())
3812 return false;
3813
3814 // From here on, we can't fail and will be building new accesses, so rig up
3815 // an IR builder.
3816 IRBuilderTy IRB(&AI);
3817
3818 // Collect the new slices which we will merge into the alloca slices.
3819 SmallVector<Slice, 4> NewSlices;
3820
3821 // Track any allocas we end up splitting loads and stores for so we iterate
3822 // on them.
3823 SmallPtrSet<AllocaInst *, 4> ResplitPromotableAllocas;
3824
3825 // At this point, we have collected all of the loads and stores we can
3826 // pre-split, and the specific splits needed for them. We actually do the
3827 // splitting in a specific order in order to handle when one of the loads in
3828 // the value operand to one of the stores.
3829 //
3830 // First, we rewrite all of the split loads, and just accumulate each split
3831 // load in a parallel structure. We also build the slices for them and append
3832 // them to the alloca slices.
3833 SmallDenseMap<LoadInst *, std::vector<LoadInst *>, 1> SplitLoadsMap;
3834 std::vector<LoadInst *> SplitLoads;
3835 const DataLayout &DL = AI.getModule()->getDataLayout();
3836 for (LoadInst *LI : Loads) {
3837 SplitLoads.clear();
3838
3839 IntegerType *Ty = cast<IntegerType>(LI->getType());
3840 uint64_t LoadSize = Ty->getBitWidth() / 8;
3841 assert(LoadSize > 0 && "Cannot have a zero-sized integer load!")((LoadSize > 0 && "Cannot have a zero-sized integer load!"
) ? static_cast<void> (0) : __assert_fail ("LoadSize > 0 && \"Cannot have a zero-sized integer load!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3841, __PRETTY_FUNCTION__))
;
3842
3843 auto &Offsets = SplitOffsetsMap[LI];
3844 assert(LoadSize == Offsets.S->endOffset() - Offsets.S->beginOffset() &&((LoadSize == Offsets.S->endOffset() - Offsets.S->beginOffset
() && "Slice size should always match load size exactly!"
) ? static_cast<void> (0) : __assert_fail ("LoadSize == Offsets.S->endOffset() - Offsets.S->beginOffset() && \"Slice size should always match load size exactly!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3845, __PRETTY_FUNCTION__))
3845 "Slice size should always match load size exactly!")((LoadSize == Offsets.S->endOffset() - Offsets.S->beginOffset
() && "Slice size should always match load size exactly!"
) ? static_cast<void> (0) : __assert_fail ("LoadSize == Offsets.S->endOffset() - Offsets.S->beginOffset() && \"Slice size should always match load size exactly!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3845, __PRETTY_FUNCTION__))
;
3846 uint64_t BaseOffset = Offsets.S->beginOffset();
3847 assert(BaseOffset + LoadSize > BaseOffset &&((BaseOffset + LoadSize > BaseOffset && "Cannot represent alloca access size using 64-bit integers!"
) ? static_cast<void> (0) : __assert_fail ("BaseOffset + LoadSize > BaseOffset && \"Cannot represent alloca access size using 64-bit integers!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3848, __PRETTY_FUNCTION__))
3848 "Cannot represent alloca access size using 64-bit integers!")((BaseOffset + LoadSize > BaseOffset && "Cannot represent alloca access size using 64-bit integers!"
) ? static_cast<void> (0) : __assert_fail ("BaseOffset + LoadSize > BaseOffset && \"Cannot represent alloca access size using 64-bit integers!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3848, __PRETTY_FUNCTION__))
;
3849
3850 Instruction *BasePtr = cast<Instruction>(LI->getPointerOperand());
3851 IRB.SetInsertPoint(LI);
3852
3853 LLVM_DEBUG(dbgs() << " Splitting load: " << *LI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Splitting load: " << *LI
<< "\n"; } } while (false)
;
3854
3855 uint64_t PartOffset = 0, PartSize = Offsets.Splits.front();
3856 int Idx = 0, Size = Offsets.Splits.size();
3857 for (;;) {
3858 auto *PartTy = Type::getIntNTy(Ty->getContext(), PartSize * 8);
3859 auto AS = LI->getPointerAddressSpace();
3860 auto *PartPtrTy = PartTy->getPointerTo(AS);
3861 LoadInst *PLoad = IRB.CreateAlignedLoad(
3862 PartTy,
3863 getAdjustedPtr(IRB, DL, BasePtr,
3864 APInt(DL.getIndexSizeInBits(AS), PartOffset),
3865 PartPtrTy, BasePtr->getName() + "."),
3866 getAdjustedAlignment(LI, PartOffset, DL), /*IsVolatile*/ false,
3867 LI->getName());
3868 PLoad->copyMetadata(*LI, {LLVMContext::MD_mem_parallel_loop_access,
3869 LLVMContext::MD_access_group});
3870
3871 // Append this load onto the list of split loads so we can find it later
3872 // to rewrite the stores.
3873 SplitLoads.push_back(PLoad);
3874
3875 // Now build a new slice for the alloca.
3876 NewSlices.push_back(
3877 Slice(BaseOffset + PartOffset, BaseOffset + PartOffset + PartSize,
3878 &PLoad->getOperandUse(PLoad->getPointerOperandIndex()),
3879 /*IsSplittable*/ false));
3880 LLVM_DEBUG(dbgs() << " new slice [" << NewSlices.back().beginOffset()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " new slice [" << NewSlices
.back().beginOffset() << ", " << NewSlices.back()
.endOffset() << "): " << *PLoad << "\n"; } }
while (false)
3881 << ", " << NewSlices.back().endOffset()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " new slice [" << NewSlices
.back().beginOffset() << ", " << NewSlices.back()
.endOffset() << "): " << *PLoad << "\n"; } }
while (false)
3882 << "): " << *PLoad << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " new slice [" << NewSlices
.back().beginOffset() << ", " << NewSlices.back()
.endOffset() << "): " << *PLoad << "\n"; } }
while (false)
;
3883
3884 // See if we've handled all the splits.
3885 if (Idx >= Size)
3886 break;
3887
3888 // Setup the next partition.
3889 PartOffset = Offsets.Splits[Idx];
3890 ++Idx;
3891 PartSize = (Idx < Size ? Offsets.Splits[Idx] : LoadSize) - PartOffset;
3892 }
3893
3894 // Now that we have the split loads, do the slow walk over all uses of the
3895 // load and rewrite them as split stores, or save the split loads to use
3896 // below if the store is going to be split there anyways.
3897 bool DeferredStores = false;
3898 for (User *LU : LI->users()) {
3899 StoreInst *SI = cast<StoreInst>(LU);
3900 if (!Stores.empty() && SplitOffsetsMap.count(SI)) {
3901 DeferredStores = true;
3902 LLVM_DEBUG(dbgs() << " Deferred splitting of store: " << *SIdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Deferred splitting of store: "
<< *SI << "\n"; } } while (false)
3903 << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Deferred splitting of store: "
<< *SI << "\n"; } } while (false)
;
3904 continue;
3905 }
3906
3907 Value *StoreBasePtr = SI->getPointerOperand();
3908 IRB.SetInsertPoint(SI);
3909
3910 LLVM_DEBUG(dbgs() << " Splitting store of load: " << *SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Splitting store of load: " <<
*SI << "\n"; } } while (false)
;
3911
3912 for (int Idx = 0, Size = SplitLoads.size(); Idx < Size; ++Idx) {
3913 LoadInst *PLoad = SplitLoads[Idx];
3914 uint64_t PartOffset = Idx == 0 ? 0 : Offsets.Splits[Idx - 1];
3915 auto *PartPtrTy =
3916 PLoad->getType()->getPointerTo(SI->getPointerAddressSpace());
3917
3918 auto AS = SI->getPointerAddressSpace();
3919 StoreInst *PStore = IRB.CreateAlignedStore(
3920 PLoad,
3921 getAdjustedPtr(IRB, DL, StoreBasePtr,
3922 APInt(DL.getIndexSizeInBits(AS), PartOffset),
3923 PartPtrTy, StoreBasePtr->getName() + "."),
3924 getAdjustedAlignment(SI, PartOffset, DL), /*IsVolatile*/ false);
3925 PStore->copyMetadata(*LI, {LLVMContext::MD_mem_parallel_loop_access,
3926 LLVMContext::MD_access_group});
3927 LLVM_DEBUG(dbgs() << " +" << PartOffset << ":" << *PStore << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " +" << PartOffset <<
":" << *PStore << "\n"; } } while (false)
;
3928 }
3929
3930 // We want to immediately iterate on any allocas impacted by splitting
3931 // this store, and we have to track any promotable alloca (indicated by
3932 // a direct store) as needing to be resplit because it is no longer
3933 // promotable.
3934 if (AllocaInst *OtherAI = dyn_cast<AllocaInst>(StoreBasePtr)) {
3935 ResplitPromotableAllocas.insert(OtherAI);
3936 Worklist.insert(OtherAI);
3937 } else if (AllocaInst *OtherAI = dyn_cast<AllocaInst>(
3938 StoreBasePtr->stripInBoundsOffsets())) {
3939 Worklist.insert(OtherAI);
3940 }
3941
3942 // Mark the original store as dead.
3943 DeadInsts.insert(SI);
3944 }
3945
3946 // Save the split loads if there are deferred stores among the users.
3947 if (DeferredStores)
3948 SplitLoadsMap.insert(std::make_pair(LI, std::move(SplitLoads)));
3949
3950 // Mark the original load as dead and kill the original slice.
3951 DeadInsts.insert(LI);
3952 Offsets.S->kill();
3953 }
3954
3955 // Second, we rewrite all of the split stores. At this point, we know that
3956 // all loads from this alloca have been split already. For stores of such
3957 // loads, we can simply look up the pre-existing split loads. For stores of
3958 // other loads, we split those loads first and then write split stores of
3959 // them.
3960 for (StoreInst *SI : Stores) {
3961 auto *LI = cast<LoadInst>(SI->getValueOperand());
3962 IntegerType *Ty = cast<IntegerType>(LI->getType());
3963 uint64_t StoreSize = Ty->getBitWidth() / 8;
3964 assert(StoreSize > 0 && "Cannot have a zero-sized integer store!")((StoreSize > 0 && "Cannot have a zero-sized integer store!"
) ? static_cast<void> (0) : __assert_fail ("StoreSize > 0 && \"Cannot have a zero-sized integer store!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3964, __PRETTY_FUNCTION__))
;
3965
3966 auto &Offsets = SplitOffsetsMap[SI];
3967 assert(StoreSize == Offsets.S->endOffset() - Offsets.S->beginOffset() &&((StoreSize == Offsets.S->endOffset() - Offsets.S->beginOffset
() && "Slice size should always match load size exactly!"
) ? static_cast<void> (0) : __assert_fail ("StoreSize == Offsets.S->endOffset() - Offsets.S->beginOffset() && \"Slice size should always match load size exactly!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3968, __PRETTY_FUNCTION__))
3968 "Slice size should always match load size exactly!")((StoreSize == Offsets.S->endOffset() - Offsets.S->beginOffset
() && "Slice size should always match load size exactly!"
) ? static_cast<void> (0) : __assert_fail ("StoreSize == Offsets.S->endOffset() - Offsets.S->beginOffset() && \"Slice size should always match load size exactly!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3968, __PRETTY_FUNCTION__))
;
3969 uint64_t BaseOffset = Offsets.S->beginOffset();
3970 assert(BaseOffset + StoreSize > BaseOffset &&((BaseOffset + StoreSize > BaseOffset && "Cannot represent alloca access size using 64-bit integers!"
) ? static_cast<void> (0) : __assert_fail ("BaseOffset + StoreSize > BaseOffset && \"Cannot represent alloca access size using 64-bit integers!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3971, __PRETTY_FUNCTION__))
3971 "Cannot represent alloca access size using 64-bit integers!")((BaseOffset + StoreSize > BaseOffset && "Cannot represent alloca access size using 64-bit integers!"
) ? static_cast<void> (0) : __assert_fail ("BaseOffset + StoreSize > BaseOffset && \"Cannot represent alloca access size using 64-bit integers!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3971, __PRETTY_FUNCTION__))
;
3972
3973 Value *LoadBasePtr = LI->getPointerOperand();
3974 Instruction *StoreBasePtr = cast<Instruction>(SI->getPointerOperand());
3975
3976 LLVM_DEBUG(dbgs() << " Splitting store: " << *SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Splitting store: " << *SI
<< "\n"; } } while (false)
;
3977
3978 // Check whether we have an already split load.
3979 auto SplitLoadsMapI = SplitLoadsMap.find(LI);
3980 std::vector<LoadInst *> *SplitLoads = nullptr;
3981 if (SplitLoadsMapI != SplitLoadsMap.end()) {
3982 SplitLoads = &SplitLoadsMapI->second;
3983 assert(SplitLoads->size() == Offsets.Splits.size() + 1 &&((SplitLoads->size() == Offsets.Splits.size() + 1 &&
"Too few split loads for the number of splits in the store!"
) ? static_cast<void> (0) : __assert_fail ("SplitLoads->size() == Offsets.Splits.size() + 1 && \"Too few split loads for the number of splits in the store!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3984, __PRETTY_FUNCTION__))
3984 "Too few split loads for the number of splits in the store!")((SplitLoads->size() == Offsets.Splits.size() + 1 &&
"Too few split loads for the number of splits in the store!"
) ? static_cast<void> (0) : __assert_fail ("SplitLoads->size() == Offsets.Splits.size() + 1 && \"Too few split loads for the number of splits in the store!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 3984, __PRETTY_FUNCTION__))
;
3985 } else {
3986 LLVM_DEBUG(dbgs() << " of load: " << *LI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " of load: " << *LI
<< "\n"; } } while (false)
;
3987 }
3988
3989 uint64_t PartOffset = 0, PartSize = Offsets.Splits.front();
3990 int Idx = 0, Size = Offsets.Splits.size();
3991 for (;;) {
3992 auto *PartTy = Type::getIntNTy(Ty->getContext(), PartSize * 8);
3993 auto *LoadPartPtrTy = PartTy->getPointerTo(LI->getPointerAddressSpace());
3994 auto *StorePartPtrTy = PartTy->getPointerTo(SI->getPointerAddressSpace());
3995
3996 // Either lookup a split load or create one.
3997 LoadInst *PLoad;
3998 if (SplitLoads) {
3999 PLoad = (*SplitLoads)[Idx];
4000 } else {
4001 IRB.SetInsertPoint(LI);
4002 auto AS = LI->getPointerAddressSpace();
4003 PLoad = IRB.CreateAlignedLoad(
4004 PartTy,
4005 getAdjustedPtr(IRB, DL, LoadBasePtr,
4006 APInt(DL.getIndexSizeInBits(AS), PartOffset),
4007 LoadPartPtrTy, LoadBasePtr->getName() + "."),
4008 getAdjustedAlignment(LI, PartOffset, DL), /*IsVolatile*/ false,
4009 LI->getName());
4010 }
4011
4012 // And store this partition.
4013 IRB.SetInsertPoint(SI);
4014 auto AS = SI->getPointerAddressSpace();
4015 StoreInst *PStore = IRB.CreateAlignedStore(
4016 PLoad,
4017 getAdjustedPtr(IRB, DL, StoreBasePtr,
4018 APInt(DL.getIndexSizeInBits(AS), PartOffset),
4019 StorePartPtrTy, StoreBasePtr->getName() + "."),
4020 getAdjustedAlignment(SI, PartOffset, DL), /*IsVolatile*/ false);
4021
4022 // Now build a new slice for the alloca.
4023 NewSlices.push_back(
4024 Slice(BaseOffset + PartOffset, BaseOffset + PartOffset + PartSize,
4025 &PStore->getOperandUse(PStore->getPointerOperandIndex()),
4026 /*IsSplittable*/ false));
4027 LLVM_DEBUG(dbgs() << " new slice [" << NewSlices.back().beginOffset()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " new slice [" << NewSlices
.back().beginOffset() << ", " << NewSlices.back()
.endOffset() << "): " << *PStore << "\n"; }
} while (false)
4028 << ", " << NewSlices.back().endOffset()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " new slice [" << NewSlices
.back().beginOffset() << ", " << NewSlices.back()
.endOffset() << "): " << *PStore << "\n"; }
} while (false)
4029 << "): " << *PStore << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " new slice [" << NewSlices
.back().beginOffset() << ", " << NewSlices.back()
.endOffset() << "): " << *PStore << "\n"; }
} while (false)
;
4030 if (!SplitLoads) {
4031 LLVM_DEBUG(dbgs() << " of split load: " << *PLoad << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " of split load: " << *
PLoad << "\n"; } } while (false)
;
4032 }
4033
4034 // See if we've finished all the splits.
4035 if (Idx >= Size)
4036 break;
4037
4038 // Setup the next partition.
4039 PartOffset = Offsets.Splits[Idx];
4040 ++Idx;
4041 PartSize = (Idx < Size ? Offsets.Splits[Idx] : StoreSize) - PartOffset;
4042 }
4043
4044 // We want to immediately iterate on any allocas impacted by splitting
4045 // this load, which is only relevant if it isn't a load of this alloca and
4046 // thus we didn't already split the loads above. We also have to keep track
4047 // of any promotable allocas we split loads on as they can no longer be
4048 // promoted.
4049 if (!SplitLoads) {
4050 if (AllocaInst *OtherAI = dyn_cast<AllocaInst>(LoadBasePtr)) {
4051 assert(OtherAI != &AI && "We can't re-split our own alloca!")((OtherAI != &AI && "We can't re-split our own alloca!"
) ? static_cast<void> (0) : __assert_fail ("OtherAI != &AI && \"We can't re-split our own alloca!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 4051, __PRETTY_FUNCTION__))
;
4052 ResplitPromotableAllocas.insert(OtherAI);
4053 Worklist.insert(OtherAI);
4054 } else if (AllocaInst *OtherAI = dyn_cast<AllocaInst>(
4055 LoadBasePtr->stripInBoundsOffsets())) {
4056 assert(OtherAI != &AI && "We can't re-split our own alloca!")((OtherAI != &AI && "We can't re-split our own alloca!"
) ? static_cast<void> (0) : __assert_fail ("OtherAI != &AI && \"We can't re-split our own alloca!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 4056, __PRETTY_FUNCTION__))
;
4057 Worklist.insert(OtherAI);
4058 }
4059 }
4060
4061 // Mark the original store as dead now that we've split it up and kill its
4062 // slice. Note that we leave the original load in place unless this store
4063 // was its only use. It may in turn be split up if it is an alloca load
4064 // for some other alloca, but it may be a normal load. This may introduce
4065 // redundant loads, but where those can be merged the rest of the optimizer
4066 // should handle the merging, and this uncovers SSA splits which is more
4067 // important. In practice, the original loads will almost always be fully
4068 // split and removed eventually, and the splits will be merged by any
4069 // trivial CSE, including instcombine.
4070 if (LI->hasOneUse()) {
4071 assert(*LI->user_begin() == SI && "Single use isn't this store!")((*LI->user_begin() == SI && "Single use isn't this store!"
) ? static_cast<void> (0) : __assert_fail ("*LI->user_begin() == SI && \"Single use isn't this store!\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 4071, __PRETTY_FUNCTION__))
;
4072 DeadInsts.insert(LI);
4073 }
4074 DeadInsts.insert(SI);
4075 Offsets.S->kill();
4076 }
4077
4078 // Remove the killed slices that have ben pre-split.
4079 AS.erase(llvm::remove_if(AS, [](const Slice &S) { return S.isDead(); }),
4080 AS.end());
4081
4082 // Insert our new slices. This will sort and merge them into the sorted
4083 // sequence.
4084 AS.insert(NewSlices);
4085
4086 LLVM_DEBUG(dbgs() << " Pre-split slices:\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Pre-split slices:\n"; } } while
(false)
;
4087#ifndef NDEBUG
4088 for (auto I = AS.begin(), E = AS.end(); I != E; ++I)
4089 LLVM_DEBUG(AS.print(dbgs(), I, " "))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { AS.print(dbgs(), I, " "); } } while (false)
;
4090#endif
4091
4092 // Finally, don't try to promote any allocas that new require re-splitting.
4093 // They have already been added to the worklist above.
4094 PromotableAllocas.erase(
4095 llvm::remove_if(
4096 PromotableAllocas,
4097 [&](AllocaInst *AI) { return ResplitPromotableAllocas.count(AI); }),
4098 PromotableAllocas.end());
4099
4100 return true;
4101}
4102
4103/// Rewrite an alloca partition's users.
4104///
4105/// This routine drives both of the rewriting goals of the SROA pass. It tries
4106/// to rewrite uses of an alloca partition to be conducive for SSA value
4107/// promotion. If the partition needs a new, more refined alloca, this will
4108/// build that new alloca, preserving as much type information as possible, and
4109/// rewrite the uses of the old alloca to point at the new one and have the
4110/// appropriate new offsets. It also evaluates how successful the rewrite was
4111/// at enabling promotion and if it was successful queues the alloca to be
4112/// promoted.
4113AllocaInst *SROA::rewritePartition(AllocaInst &AI, AllocaSlices &AS,
4114 Partition &P) {
4115 // Try to compute a friendly type for this partition of the alloca. This
4116 // won't always succeed, in which case we fall back to a legal integer type
4117 // or an i8 array of an appropriate size.
4118 Type *SliceTy = nullptr;
4119 const DataLayout &DL = AI.getModule()->getDataLayout();
4120 if (Type *CommonUseTy = findCommonType(P.begin(), P.end(), P.endOffset()))
4121 if (DL.getTypeAllocSize(CommonUseTy) >= P.size())
4122 SliceTy = CommonUseTy;
4123 if (!SliceTy)
4124 if (Type *TypePartitionTy = getTypePartition(DL, AI.getAllocatedType(),
4125 P.beginOffset(), P.size()))
4126 SliceTy = TypePartitionTy;
4127 if ((!SliceTy || (SliceTy->isArrayTy() &&
4128 SliceTy->getArrayElementType()->isIntegerTy())) &&
4129 DL.isLegalInteger(P.size() * 8))
4130 SliceTy = Type::getIntNTy(*C, P.size() * 8);
4131 if (!SliceTy)
4132 SliceTy = ArrayType::get(Type::getInt8Ty(*C), P.size());
4133 assert(DL.getTypeAllocSize(SliceTy) >= P.size())((DL.getTypeAllocSize(SliceTy) >= P.size()) ? static_cast<
void> (0) : __assert_fail ("DL.getTypeAllocSize(SliceTy) >= P.size()"
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 4133, __PRETTY_FUNCTION__))
;
4134
4135 bool IsIntegerPromotable = isIntegerWideningViable(P, SliceTy, DL);
4136
4137 VectorType *VecTy =
4138 IsIntegerPromotable ? nullptr : isVectorPromotionViable(P, DL);
4139 if (VecTy)
4140 SliceTy = VecTy;
4141
4142 // Check for the case where we're going to rewrite to a new alloca of the
4143 // exact same type as the original, and with the same access offsets. In that
4144 // case, re-use the existing alloca, but still run through the rewriter to
4145 // perform phi and select speculation.
4146 // P.beginOffset() can be non-zero even with the same type in a case with
4147 // out-of-bounds access (e.g. @PR35657 function in SROA/basictest.ll).
4148 AllocaInst *NewAI;
4149 if (SliceTy == AI.getAllocatedType() && P.beginOffset() == 0) {
4150 NewAI = &AI;
4151 // FIXME: We should be able to bail at this point with "nothing changed".
4152 // FIXME: We might want to defer PHI speculation until after here.
4153 // FIXME: return nullptr;
4154 } else {
4155 unsigned Alignment = AI.getAlignment();
4156 if (!Alignment) {
4157 // The minimum alignment which users can rely on when the explicit
4158 // alignment is omitted or zero is that required by the ABI for this
4159 // type.
4160 Alignment = DL.getABITypeAlignment(AI.getAllocatedType());
4161 }
4162 Alignment = MinAlign(Alignment, P.beginOffset());
4163 // If we will get at least this much alignment from the type alone, leave
4164 // the alloca's alignment unconstrained.
4165 if (Alignment <= DL.getABITypeAlignment(SliceTy))
4166 Alignment = 0;
4167 NewAI = new AllocaInst(
4168 SliceTy, AI.getType()->getAddressSpace(), nullptr, Alignment,
4169 AI.getName() + ".sroa." + Twine(P.begin() - AS.begin()), &AI);
4170 // Copy the old AI debug location over to the new one.
4171 NewAI->setDebugLoc(AI.getDebugLoc());
4172 ++NumNewAllocas;
4173 }
4174
4175 LLVM_DEBUG(dbgs() << "Rewriting alloca partition "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "Rewriting alloca partition " <<
"[" << P.beginOffset() << "," << P.endOffset
() << ") to: " << *NewAI << "\n"; } } while
(false)
4176 << "[" << P.beginOffset() << "," << P.endOffset()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "Rewriting alloca partition " <<
"[" << P.beginOffset() << "," << P.endOffset
() << ") to: " << *NewAI << "\n"; } } while
(false)
4177 << ") to: " << *NewAI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "Rewriting alloca partition " <<
"[" << P.beginOffset() << "," << P.endOffset
() << ") to: " << *NewAI << "\n"; } } while
(false)
;
4178
4179 // Track the high watermark on the worklist as it is only relevant for
4180 // promoted allocas. We will reset it to this point if the alloca is not in
4181 // fact scheduled for promotion.
4182 unsigned PPWOldSize = PostPromotionWorklist.size();
4183 unsigned NumUses = 0;
4184 SmallSetVector<PHINode *, 8> PHIUsers;
4185 SmallSetVector<SelectInst *, 8> SelectUsers;
4186
4187 AllocaSliceRewriter Rewriter(DL, AS, *this, AI, *NewAI, P.beginOffset(),
4188 P.endOffset(), IsIntegerPromotable, VecTy,
4189 PHIUsers, SelectUsers);
4190 bool Promotable = true;
4191 for (Slice *S : P.splitSliceTails()) {
4192 Promotable &= Rewriter.visit(S);
4193 ++NumUses;
4194 }
4195 for (Slice &S : P) {
4196 Promotable &= Rewriter.visit(&S);
4197 ++NumUses;
4198 }
4199
4200 NumAllocaPartitionUses += NumUses;
4201 MaxUsesPerAllocaPartition.updateMax(NumUses);
4202
4203 // Now that we've processed all the slices in the new partition, check if any
4204 // PHIs or Selects would block promotion.
4205 for (PHINode *PHI : PHIUsers)
4206 if (!isSafePHIToSpeculate(*PHI)) {
4207 Promotable = false;
4208 PHIUsers.clear();
4209 SelectUsers.clear();
4210 break;
4211 }
4212
4213 for (SelectInst *Sel : SelectUsers)
4214 if (!isSafeSelectToSpeculate(*Sel)) {
4215 Promotable = false;
4216 PHIUsers.clear();
4217 SelectUsers.clear();
4218 break;
4219 }
4220
4221 if (Promotable) {
4222 if (PHIUsers.empty() && SelectUsers.empty()) {
4223 // Promote the alloca.
4224 PromotableAllocas.push_back(NewAI);
4225 } else {
4226 // If we have either PHIs or Selects to speculate, add them to those
4227 // worklists and re-queue the new alloca so that we promote in on the
4228 // next iteration.
4229 for (PHINode *PHIUser : PHIUsers)
4230 SpeculatablePHIs.insert(PHIUser);
4231 for (SelectInst *SelectUser : SelectUsers)
4232 SpeculatableSelects.insert(SelectUser);
4233 Worklist.insert(NewAI);
4234 }
4235 } else {
4236 // Drop any post-promotion work items if promotion didn't happen.
4237 while (PostPromotionWorklist.size() > PPWOldSize)
4238 PostPromotionWorklist.pop_back();
4239
4240 // We couldn't promote and we didn't create a new partition, nothing
4241 // happened.
4242 if (NewAI == &AI)
4243 return nullptr;
4244
4245 // If we can't promote the alloca, iterate on it to check for new
4246 // refinements exposed by splitting the current alloca. Don't iterate on an
4247 // alloca which didn't actually change and didn't get promoted.
4248 Worklist.insert(NewAI);
4249 }
4250
4251 return NewAI;
4252}
4253
4254/// Walks the slices of an alloca and form partitions based on them,
4255/// rewriting each of their uses.
4256bool SROA::splitAlloca(AllocaInst &AI, AllocaSlices &AS) {
4257 if (AS.begin() == AS.end())
4258 return false;
4259
4260 unsigned NumPartitions = 0;
4261 bool Changed = false;
4262 const DataLayout &DL = AI.getModule()->getDataLayout();
4263
4264 // First try to pre-split loads and stores.
4265 Changed |= presplitLoadsAndStores(AI, AS);
4266
4267 // Now that we have identified any pre-splitting opportunities,
4268 // mark loads and stores unsplittable except for the following case.
4269 // We leave a slice splittable if all other slices are disjoint or fully
4270 // included in the slice, such as whole-alloca loads and stores.
4271 // If we fail to split these during pre-splitting, we want to force them
4272 // to be rewritten into a partition.
4273 bool IsSorted = true;
4274
4275 uint64_t AllocaSize = DL.getTypeAllocSize(AI.getAllocatedType());
4276 const uint64_t MaxBitVectorSize = 1024;
4277 if (AllocaSize <= MaxBitVectorSize) {
4278 // If a byte boundary is included in any load or store, a slice starting or
4279 // ending at the boundary is not splittable.
4280 SmallBitVector SplittableOffset(AllocaSize + 1, true);
4281 for (Slice &S : AS)
4282 for (unsigned O = S.beginOffset() + 1;
4283 O < S.endOffset() && O < AllocaSize; O++)
4284 SplittableOffset.reset(O);
4285
4286 for (Slice &S : AS) {
4287 if (!S.isSplittable())
4288 continue;
4289
4290 if ((S.beginOffset() > AllocaSize || SplittableOffset[S.beginOffset()]) &&
4291 (S.endOffset() > AllocaSize || SplittableOffset[S.endOffset()]))
4292 continue;
4293
4294 if (isa<LoadInst>(S.getUse()->getUser()) ||
4295 isa<StoreInst>(S.getUse()->getUser())) {
4296 S.makeUnsplittable();
4297 IsSorted = false;
4298 }
4299 }
4300 }
4301 else {
4302 // We only allow whole-alloca splittable loads and stores
4303 // for a large alloca to avoid creating too large BitVector.
4304 for (Slice &S : AS) {
4305 if (!S.isSplittable())
4306 continue;
4307
4308 if (S.beginOffset() == 0 && S.endOffset() >= AllocaSize)
4309 continue;
4310
4311 if (isa<LoadInst>(S.getUse()->getUser()) ||
4312 isa<StoreInst>(S.getUse()->getUser())) {
4313 S.makeUnsplittable();
4314 IsSorted = false;
4315 }
4316 }
4317 }
4318
4319 if (!IsSorted)
4320 llvm::sort(AS);
4321
4322 /// Describes the allocas introduced by rewritePartition in order to migrate
4323 /// the debug info.
4324 struct Fragment {
4325 AllocaInst *Alloca;
4326 uint64_t Offset;
4327 uint64_t Size;
4328 Fragment(AllocaInst *AI, uint64_t O, uint64_t S)
4329 : Alloca(AI), Offset(O), Size(S) {}
4330 };
4331 SmallVector<Fragment, 4> Fragments;
4332
4333 // Rewrite each partition.
4334 for (auto &P : AS.partitions()) {
4335 if (AllocaInst *NewAI = rewritePartition(AI, AS, P)) {
4336 Changed = true;
4337 if (NewAI != &AI) {
4338 uint64_t SizeOfByte = 8;
4339 uint64_t AllocaSize = DL.getTypeSizeInBits(NewAI->getAllocatedType());
4340 // Don't include any padding.
4341 uint64_t Size = std::min(AllocaSize, P.size() * SizeOfByte);
4342 Fragments.push_back(Fragment(NewAI, P.beginOffset() * SizeOfByte, Size));
4343 }
4344 }
4345 ++NumPartitions;
4346 }
4347
4348 NumAllocaPartitions += NumPartitions;
4349 MaxPartitionsPerAlloca.updateMax(NumPartitions);
4350
4351 // Migrate debug information from the old alloca to the new alloca(s)
4352 // and the individual partitions.
4353 TinyPtrVector<DbgVariableIntrinsic *> DbgDeclares = FindDbgAddrUses(&AI);
4354 if (!DbgDeclares.empty()) {
4355 auto *Var = DbgDeclares.front()->getVariable();
4356 auto *Expr = DbgDeclares.front()->getExpression();
4357 auto VarSize = Var->getSizeInBits();
4358 DIBuilder DIB(*AI.getModule(), /*AllowUnresolved*/ false);
4359 uint64_t AllocaSize = DL.getTypeSizeInBits(AI.getAllocatedType());
4360 for (auto Fragment : Fragments) {
4361 // Create a fragment expression describing the new partition or reuse AI's
4362 // expression if there is only one partition.
4363 auto *FragmentExpr = Expr;
4364 if (Fragment.Size < AllocaSize || Expr->isFragment()) {
4365 // If this alloca is already a scalar replacement of a larger aggregate,
4366 // Fragment.Offset describes the offset inside the scalar.
4367 auto ExprFragment = Expr->getFragmentInfo();
4368 uint64_t Offset = ExprFragment ? ExprFragment->OffsetInBits : 0;
4369 uint64_t Start = Offset + Fragment.Offset;
4370 uint64_t Size = Fragment.Size;
4371 if (ExprFragment) {
4372 uint64_t AbsEnd =
4373 ExprFragment->OffsetInBits + ExprFragment->SizeInBits;
4374 if (Start >= AbsEnd)
4375 // No need to describe a SROAed padding.
4376 continue;
4377 Size = std::min(Size, AbsEnd - Start);
4378 }
4379 // The new, smaller fragment is stenciled out from the old fragment.
4380 if (auto OrigFragment = FragmentExpr->getFragmentInfo()) {
4381 assert(Start >= OrigFragment->OffsetInBits &&((Start >= OrigFragment->OffsetInBits && "new fragment is outside of original fragment"
) ? static_cast<void> (0) : __assert_fail ("Start >= OrigFragment->OffsetInBits && \"new fragment is outside of original fragment\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 4382, __PRETTY_FUNCTION__))
4382 "new fragment is outside of original fragment")((Start >= OrigFragment->OffsetInBits && "new fragment is outside of original fragment"
) ? static_cast<void> (0) : __assert_fail ("Start >= OrigFragment->OffsetInBits && \"new fragment is outside of original fragment\""
, "/build/llvm-toolchain-snapshot-10~svn374877/lib/Transforms/Scalar/SROA.cpp"
, 4382, __PRETTY_FUNCTION__))
;
4383 Start -= OrigFragment->OffsetInBits;
4384 }
4385
4386 // The alloca may be larger than the variable.
4387 if (VarSize) {
4388 if (Size > *VarSize)
4389 Size = *VarSize;
4390 if (Size == 0 || Start + Size > *VarSize)
4391 continue;
4392 }
4393
4394 // Avoid creating a fragment expression that covers the entire variable.
4395 if (!VarSize || *VarSize != Size) {
4396 if (auto E =
4397 DIExpression::createFragmentExpression(Expr, Start, Size))
4398 FragmentExpr = *E;
4399 else
4400 continue;
4401 }
4402 }
4403
4404 // Remove any existing intrinsics describing the same alloca.
4405 for (DbgVariableIntrinsic *OldDII : FindDbgAddrUses(Fragment.Alloca))
4406 OldDII->eraseFromParent();
4407
4408 DIB.insertDeclare(Fragment.Alloca, Var, FragmentExpr,
4409 DbgDeclares.front()->getDebugLoc(), &AI);
4410 }
4411 }
4412 return Changed;
4413}
4414
4415/// Clobber a use with undef, deleting the used value if it becomes dead.
4416void SROA::clobberUse(Use &U) {
4417 Value *OldV = U;
4418 // Replace the use with an undef value.
4419 U = UndefValue::get(OldV->getType());
4420
4421 // Check for this making an instruction dead. We have to garbage collect
4422 // all the dead instructions to ensure the uses of any alloca end up being
4423 // minimal.
4424 if (Instruction *OldI = dyn_cast<Instruction>(OldV))
4425 if (isInstructionTriviallyDead(OldI)) {
4426 DeadInsts.insert(OldI);
4427 }
4428}
4429
4430/// Analyze an alloca for SROA.
4431///
4432/// This analyzes the alloca to ensure we can reason about it, builds
4433/// the slices of the alloca, and then hands it off to be split and
4434/// rewritten as needed.
4435bool SROA::runOnAlloca(AllocaInst &AI) {
4436 LLVM_DEBUG(dbgs() << "SROA alloca: " << AI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "SROA alloca: " << AI <<
"\n"; } } while (false)
;
4437 ++NumAllocasAnalyzed;
4438
4439 // Special case dead allocas, as they're trivial.
4440 if (AI.use_empty()) {
4441 AI.eraseFromParent();
4442 return true;
4443 }
4444 const DataLayout &DL = AI.getModule()->getDataLayout();
4445
4446 // Skip alloca forms that this analysis can't handle.
4447 if (AI.isArrayAllocation() || !AI.getAllocatedType()->isSized() ||
4448 DL.getTypeAllocSize(AI.getAllocatedType()) == 0)
4449 return false;
4450
4451 bool Changed = false;
4452
4453 // First, split any FCA loads and stores touching this alloca to promote
4454 // better splitting and promotion opportunities.
4455 AggLoadStoreRewriter AggRewriter(DL);
4456 Changed |= AggRewriter.rewrite(AI);
4457
4458 // Build the slices using a recursive instruction-visiting builder.
4459 AllocaSlices AS(DL, AI);
4460 LLVM_DEBUG(AS.print(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { AS.print(dbgs()); } } while (false)
;
4461 if (AS.isEscaped())
4462 return Changed;
4463
4464 // Delete all the dead users of this alloca before splitting and rewriting it.
4465 for (Instruction *DeadUser : AS.getDeadUsers()) {
4466 // Free up everything used by this instruction.
4467 for (Use &DeadOp : DeadUser->operands())
4468 clobberUse(DeadOp);
4469
4470 // Now replace the uses of this instruction.
4471 DeadUser->replaceAllUsesWith(UndefValue::get(DeadUser->getType()));
4472
4473 // And mark it for deletion.
4474 DeadInsts.insert(DeadUser);
4475 Changed = true;
4476 }
4477 for (Use *DeadOp : AS.getDeadOperands()) {
4478 clobberUse(*DeadOp);
4479 Changed = true;
4480 }
4481
4482 // No slices to split. Leave the dead alloca for a later pass to clean up.
4483 if (AS.begin() == AS.end())
4484 return Changed;
4485
4486 Changed |= splitAlloca(AI, AS);
4487
4488 LLVM_DEBUG(dbgs() << " Speculating PHIs\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Speculating PHIs\n"; } } while
(false)
;
4489 while (!SpeculatablePHIs.empty())
4490 speculatePHINodeLoads(*SpeculatablePHIs.pop_back_val());
4491
4492 LLVM_DEBUG(dbgs() << " Speculating Selects\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << " Speculating Selects\n"; } } while
(false)
;
4493 while (!SpeculatableSelects.empty())
4494 speculateSelectInstLoads(*SpeculatableSelects.pop_back_val());
4495
4496 return Changed;
4497}
4498
4499/// Delete the dead instructions accumulated in this run.
4500///
4501/// Recursively deletes the dead instructions we've accumulated. This is done
4502/// at the very end to maximize locality of the recursive delete and to
4503/// minimize the problems of invalidated instruction pointers as such pointers
4504/// are used heavily in the intermediate stages of the algorithm.
4505///
4506/// We also record the alloca instructions deleted here so that they aren't
4507/// subsequently handed to mem2reg to promote.
4508bool SROA::deleteDeadInstructions(
4509 SmallPtrSetImpl<AllocaInst *> &DeletedAllocas) {
4510 bool Changed = false;
4511 while (!DeadInsts.empty()) {
4512 Instruction *I = DeadInsts.pop_back_val();
4513 LLVM_DEBUG(dbgs() << "Deleting dead instruction: " << *I << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "Deleting dead instruction: " <<
*I << "\n"; } } while (false)
;
4514
4515 // If the instruction is an alloca, find the possible dbg.declare connected
4516 // to it, and remove it too. We must do this before calling RAUW or we will
4517 // not be able to find it.
4518 if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
4519 DeletedAllocas.insert(AI);
4520 for (DbgVariableIntrinsic *OldDII : FindDbgAddrUses(AI))
4521 OldDII->eraseFromParent();
4522 }
4523
4524 I->replaceAllUsesWith(UndefValue::get(I->getType()));
4525
4526 for (Use &Operand : I->operands())
4527 if (Instruction *U = dyn_cast<Instruction>(Operand)) {
4528 // Zero out the operand and see if it becomes trivially dead.
4529 Operand = nullptr;
4530 if (isInstructionTriviallyDead(U))
4531 DeadInsts.insert(U);
4532 }
4533
4534 ++NumDeleted;
4535 I->eraseFromParent();
4536 Changed = true;
4537 }
4538 return Changed;
4539}
4540
4541/// Promote the allocas, using the best available technique.
4542///
4543/// This attempts to promote whatever allocas have been identified as viable in
4544/// the PromotableAllocas list. If that list is empty, there is nothing to do.
4545/// This function returns whether any promotion occurred.
4546bool SROA::promoteAllocas(Function &F) {
4547 if (PromotableAllocas.empty())
4548 return false;
4549
4550 NumPromoted += PromotableAllocas.size();
4551
4552 LLVM_DEBUG(dbgs() << "Promoting allocas with mem2reg...\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "Promoting allocas with mem2reg...\n"
; } } while (false)
;
4553 PromoteMemToReg(PromotableAllocas, *DT, AC);
4554 PromotableAllocas.clear();
4555 return true;
4556}
4557
4558PreservedAnalyses SROA::runImpl(Function &F, DominatorTree &RunDT,
4559 AssumptionCache &RunAC) {
4560 LLVM_DEBUG(dbgs() << "SROA function: " << F.getName() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sroa")) { dbgs() << "SROA function: " << F.getName
() << "\n"; } } while (false)
;
4561 C = &F.getContext();
4562 DT = &RunDT;
4563 AC = &RunAC;
4564
4565 BasicBlock &EntryBB = F.getEntryBlock();
4566 for (BasicBlock::iterator I = EntryBB.begin(), E = std::prev(EntryBB.end());
4567 I != E; ++I) {
4568 if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
4569 Worklist.insert(AI);
4570 }
4571
4572 bool Changed = false;
4573 // A set of deleted alloca instruction pointers which should be removed from
4574 // the list of promotable allocas.
4575 SmallPtrSet<AllocaInst *, 4> DeletedAllocas;
4576
4577 do {
4578 while (!Worklist.empty()) {
4579 Changed |= runOnAlloca(*Worklist.pop_back_val());
4580 Changed |= deleteDeadInstructions(DeletedAllocas);
4581
4582 // Remove the deleted allocas from various lists so that we don't try to
4583 // continue processing them.
4584 if (!DeletedAllocas.empty()) {
4585 auto IsInSet = [&](AllocaInst *AI) { return DeletedAllocas.count(AI); };
4586 Worklist.remove_if(IsInSet);
4587 PostPromotionWorklist.remove_if(IsInSet);
4588 PromotableAllocas.erase(llvm::remove_if(PromotableAllocas, IsInSet),
4589 PromotableAllocas.end());
4590 DeletedAllocas.clear();
4591 }
4592 }
4593
4594 Changed |= promoteAllocas(F);
4595
4596 Worklist = PostPromotionWorklist;
4597 PostPromotionWorklist.clear();
4598 } while (!Worklist.empty());
4599
4600 if (!Changed)
4601 return PreservedAnalyses::all();
4602
4603 PreservedAnalyses PA;
4604 PA.preserveSet<CFGAnalyses>();
4605 PA.preserve<GlobalsAA>();
4606 return PA;
4607}
4608
4609PreservedAnalyses SROA::run(Function &F, FunctionAnalysisManager &AM) {
4610 return runImpl(F, AM.getResult<DominatorTreeAnalysis>(F),
4611 AM.getResult<AssumptionAnalysis>(F));
4612}
4613
4614/// A legacy pass for the legacy pass manager that wraps the \c SROA pass.
4615///
4616/// This is in the llvm namespace purely to allow it to be a friend of the \c
4617/// SROA pass.
4618class llvm::sroa::SROALegacyPass : public FunctionPass {
4619 /// The SROA implementation.
4620 SROA Impl;
4621
4622public:
4623 static char ID;
4624
4625 SROALegacyPass() : FunctionPass(ID) {
4626 initializeSROALegacyPassPass(*PassRegistry::getPassRegistry());
4627 }
4628
4629 bool runOnFunction(Function &F) override {
4630 if (skipFunction(F))
4631 return false;
4632
4633 auto PA = Impl.runImpl(
4634 F, getAnalysis<DominatorTreeWrapperPass>().getDomTree(),
4635 getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F));
4636 return !PA.areAllPreserved();
4637 }
4638
4639 void getAnalysisUsage(AnalysisUsage &AU) const override {
4640 AU.addRequired<AssumptionCacheTracker>();
4641 AU.addRequired<DominatorTreeWrapperPass>();
4642 AU.addPreserved<GlobalsAAWrapperPass>();
4643 AU.setPreservesCFG();
4644 }
4645
4646 StringRef getPassName() const override { return "SROA"; }
4647};
4648
4649char SROALegacyPass::ID = 0;
4650
4651FunctionPass *llvm::createSROAPass() { return new SROALegacyPass(); }
4652
4653INITIALIZE_PASS_BEGIN(SROALegacyPass, "sroa",static void *initializeSROALegacyPassPassOnce(PassRegistry &
Registry) {
4654 "Scalar Replacement Of Aggregates", false, false)static void *initializeSROALegacyPassPassOnce(PassRegistry &
Registry) {
4655INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)initializeAssumptionCacheTrackerPass(Registry);
4656INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)initializeDominatorTreeWrapperPassPass(Registry);
4657INITIALIZE_PASS_END(SROALegacyPass, "sroa", "Scalar Replacement Of Aggregates",PassInfo *PI = new PassInfo( "Scalar Replacement Of Aggregates"
, "sroa", &SROALegacyPass::ID, PassInfo::NormalCtor_t(callDefaultCtor
<SROALegacyPass>), false, false); Registry.registerPass
(*PI, true); return PI; } static llvm::once_flag InitializeSROALegacyPassPassFlag
; void llvm::initializeSROALegacyPassPass(PassRegistry &Registry
) { llvm::call_once(InitializeSROALegacyPassPassFlag, initializeSROALegacyPassPassOnce
, std::ref(Registry)); }
4658 false, false)PassInfo *PI = new PassInfo( "Scalar Replacement Of Aggregates"
, "sroa", &SROALegacyPass::ID, PassInfo::NormalCtor_t(callDefaultCtor
<SROALegacyPass>), false, false); Registry.registerPass
(*PI, true); return PI; } static llvm::once_flag InitializeSROALegacyPassPassFlag
; void llvm::initializeSROALegacyPassPass(PassRegistry &Registry
) { llvm::call_once(InitializeSROALegacyPassPassFlag, initializeSROALegacyPassPassOnce
, std::ref(Registry)); }

/build/llvm-toolchain-snapshot-10~svn374877/include/llvm/IR/IRBuilder.h

1//===- llvm/IRBuilder.h - Builder for LLVM Instructions ---------*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file defines the IRBuilder class, which is used as a convenient way
10// to create LLVM instructions with a consistent and simplified interface.
11//
12//===----------------------------------------------------------------------===//
13
14#ifndef LLVM_IR_IRBUILDER_H
15#define LLVM_IR_IRBUILDER_H
16
17#include "llvm-c/Types.h"
18#include "llvm/ADT/ArrayRef.h"
19#include "llvm/ADT/None.h"
20#include "llvm/ADT/StringRef.h"
21#include "llvm/ADT/Twine.h"
22#include "llvm/IR/BasicBlock.h"
23#include "llvm/IR/Constant.h"
24#include "llvm/IR/ConstantFolder.h"
25#include "llvm/IR/Constants.h"
26#include "llvm/IR/DataLayout.h"
27#include "llvm/IR/DebugLoc.h"
28#include "llvm/IR/DerivedTypes.h"
29#include "llvm/IR/Function.h"
30#include "llvm/IR/GlobalVariable.h"
31#include "llvm/IR/InstrTypes.h"
32#include "llvm/IR/Instruction.h"
33#include "llvm/IR/Instructions.h"
34#include "llvm/IR/IntrinsicInst.h"
35#include "llvm/IR/LLVMContext.h"
36#include "llvm/IR/Module.h"
37#include "llvm/IR/Operator.h"
38#include "llvm/IR/Type.h"
39#include "llvm/IR/Value.h"
40#include "llvm/IR/ValueHandle.h"
41#include "llvm/Support/AtomicOrdering.h"
42#include "llvm/Support/CBindingWrapping.h"
43#include "llvm/Support/Casting.h"
44#include <cassert>
45#include <cstddef>
46#include <cstdint>
47#include <functional>
48#include <utility>
49
50namespace llvm {
51
52class APInt;
53class MDNode;
54class Use;
55
56/// This provides the default implementation of the IRBuilder
57/// 'InsertHelper' method that is called whenever an instruction is created by
58/// IRBuilder and needs to be inserted.
59///
60/// By default, this inserts the instruction at the insertion point.
61class IRBuilderDefaultInserter {
62protected:
63 void InsertHelper(Instruction *I, const Twine &Name,
64 BasicBlock *BB, BasicBlock::iterator InsertPt) const {
65 if (BB) BB->getInstList().insert(InsertPt, I);
66 I->setName(Name);
67 }
68};
69
70/// Provides an 'InsertHelper' that calls a user-provided callback after
71/// performing the default insertion.
72class IRBuilderCallbackInserter : IRBuilderDefaultInserter {
73 std::function<void(Instruction *)> Callback;
74
75public:
76 IRBuilderCallbackInserter(std::function<void(Instruction *)> Callback)
77 : Callback(std::move(Callback)) {}
78
79protected:
80 void InsertHelper(Instruction *I, const Twine &Name,
81 BasicBlock *BB, BasicBlock::iterator InsertPt) const {
82 IRBuilderDefaultInserter::InsertHelper(I, Name, BB, InsertPt);
83 Callback(I);
84 }
85};
86
87/// Common base class shared among various IRBuilders.
88class IRBuilderBase {
89 DebugLoc CurDbgLocation;
90
91protected:
92 BasicBlock *BB;
93 BasicBlock::iterator InsertPt;
94 LLVMContext &Context;
95
96 MDNode *DefaultFPMathTag;
97 FastMathFlags FMF;
98
99 bool IsFPConstrained;
100 ConstrainedFPIntrinsic::ExceptionBehavior DefaultConstrainedExcept;
101 ConstrainedFPIntrinsic::RoundingMode DefaultConstrainedRounding;
102
103 ArrayRef<OperandBundleDef> DefaultOperandBundles;
104
105public:
106 IRBuilderBase(LLVMContext &context, MDNode *FPMathTag = nullptr,
107 ArrayRef<OperandBundleDef> OpBundles = None)
108 : Context(context), DefaultFPMathTag(FPMathTag), IsFPConstrained(false),
109 DefaultConstrainedExcept(ConstrainedFPIntrinsic::ebStrict),
110 DefaultConstrainedRounding(ConstrainedFPIntrinsic::rmDynamic),
111 DefaultOperandBundles(OpBundles) {
112 ClearInsertionPoint();
113 }
114
115 //===--------------------------------------------------------------------===//
116 // Builder configuration methods
117 //===--------------------------------------------------------------------===//
118
119 /// Clear the insertion point: created instructions will not be
120 /// inserted into a block.
121 void ClearInsertionPoint() {
122 BB = nullptr;
123 InsertPt = BasicBlock::iterator();
124 }
125
126 BasicBlock *GetInsertBlock() const { return BB; }
127 BasicBlock::iterator GetInsertPoint() const { return InsertPt; }
128 LLVMContext &getContext() const { return Context; }
129
130 /// This specifies that created instructions should be appended to the
131 /// end of the specified block.
132 void SetInsertPoint(BasicBlock *TheBB) {
133 BB = TheBB;
134 InsertPt = BB->end();
135 }
136
137 /// This specifies that created instructions should be inserted before
138 /// the specified instruction.
139 void SetInsertPoint(Instruction *I) {
140 BB = I->getParent();
141 InsertPt = I->getIterator();
142 assert(InsertPt != BB->end() && "Can't read debug loc from end()")((InsertPt != BB->end() && "Can't read debug loc from end()"
) ? static_cast<void> (0) : __assert_fail ("InsertPt != BB->end() && \"Can't read debug loc from end()\""
, "/build/llvm-toolchain-snapshot-10~svn374877/include/llvm/IR/IRBuilder.h"
, 142, __PRETTY_FUNCTION__))
;
143 SetCurrentDebugLocation(I->getDebugLoc());
144 }
145
146 /// This specifies that created instructions should be inserted at the
147 /// specified point.
148 void SetInsertPoint(BasicBlock *TheBB, BasicBlock::iterator IP) {
149 BB = TheBB;
150 InsertPt = IP;
151 if (IP != TheBB->end())
152 SetCurrentDebugLocation(IP->getDebugLoc());
153 }
154
155 /// Set location information used by debugging information.
156 void SetCurrentDebugLocation(DebugLoc L) { CurDbgLocation = std::move(L); }
157
158 /// Get location information used by debugging information.
159 const DebugLoc &getCurrentDebugLocation() const { return CurDbgLocation; }
160
161 /// If this builder has a current debug location, set it on the
162 /// specified instruction.
163 void SetInstDebugLocation(Instruction *I) const {
164 if (CurDbgLocation)
165 I->setDebugLoc(CurDbgLocation);
166 }
167
168 /// Get the return type of the current function that we're emitting
169 /// into.
170 Type *getCurrentFunctionReturnType() const;
171
172 /// InsertPoint - A saved insertion point.
173 class InsertPoint {
174 BasicBlock *Block = nullptr;
175 BasicBlock::iterator Point;
176
177 public:
178 /// Creates a new insertion point which doesn't point to anything.
179 InsertPoint() = default;
180
181 /// Creates a new insertion point at the given location.
182 InsertPoint(BasicBlock *InsertBlock, BasicBlock::iterator InsertPoint)
183 : Block(InsertBlock), Point(InsertPoint) {}
184
185 /// Returns true if this insert point is set.
186 bool isSet() const { return (Block != nullptr); }
187
188 BasicBlock *getBlock() const { return Block; }
189 BasicBlock::iterator getPoint() const { return Point; }
190 };
191
192 /// Returns the current insert point.
193 InsertPoint saveIP() const {
194 return InsertPoint(GetInsertBlock(), GetInsertPoint());
195 }
196
197 /// Returns the current insert point, clearing it in the process.
198 InsertPoint saveAndClearIP() {
199 InsertPoint IP(GetInsertBlock(), GetInsertPoint());
200 ClearInsertionPoint();
201 return IP;
202 }
203
204 /// Sets the current insert point to a previously-saved location.
205 void restoreIP(InsertPoint IP) {
206 if (IP.isSet())
207 SetInsertPoint(IP.getBlock(), IP.getPoint());
208 else
209 ClearInsertionPoint();
210 }
211
212 /// Get the floating point math metadata being used.
213 MDNode *getDefaultFPMathTag() const { return DefaultFPMathTag; }
214
215 /// Get the flags to be applied to created floating point ops
216 FastMathFlags getFastMathFlags() const { return FMF; }
217
218 /// Clear the fast-math flags.
219 void clearFastMathFlags() { FMF.clear(); }
220
221 /// Set the floating point math metadata to be used.
222 void setDefaultFPMathTag(MDNode *FPMathTag) { DefaultFPMathTag = FPMathTag; }
223
224 /// Set the fast-math flags to be used with generated fp-math operators
225 void setFastMathFlags(FastMathFlags NewFMF) { FMF = NewFMF; }
226
227 /// Enable/Disable use of constrained floating point math. When
228 /// enabled the CreateF<op>() calls instead create constrained
229 /// floating point intrinsic calls. Fast math flags are unaffected
230 /// by this setting.
231 void setIsFPConstrained(bool IsCon) { IsFPConstrained = IsCon; }
232
233 /// Query for the use of constrained floating point math
234 bool getIsFPConstrained() { return IsFPConstrained; }
235
236 /// Set the exception handling to be used with constrained floating point
237 void setDefaultConstrainedExcept(
238 ConstrainedFPIntrinsic::ExceptionBehavior NewExcept) {
239 DefaultConstrainedExcept = NewExcept;
240 }
241
242 /// Set the rounding mode handling to be used with constrained floating point
243 void setDefaultConstrainedRounding(
244 ConstrainedFPIntrinsic::RoundingMode NewRounding) {
245 DefaultConstrainedRounding = NewRounding;
246 }
247
248 /// Get the exception handling used with constrained floating point
249 ConstrainedFPIntrinsic::ExceptionBehavior getDefaultConstrainedExcept() {
250 return DefaultConstrainedExcept;
251 }
252
253 /// Get the rounding mode handling used with constrained floating point
254 ConstrainedFPIntrinsic::RoundingMode getDefaultConstrainedRounding() {
255 return DefaultConstrainedRounding;
256 }
257
258 //===--------------------------------------------------------------------===//
259 // RAII helpers.
260 //===--------------------------------------------------------------------===//
261
262 // RAII object that stores the current insertion point and restores it
263 // when the object is destroyed. This includes the debug location.
264 class InsertPointGuard {
265 IRBuilderBase &Builder;
266 AssertingVH<BasicBlock> Block;
267 BasicBlock::iterator Point;
268 DebugLoc DbgLoc;
269
270 public:
271 InsertPointGuard(IRBuilderBase &B)
272 : Builder(B), Block(B.GetInsertBlock()), Point(B.GetInsertPoint()),
273 DbgLoc(B.getCurrentDebugLocation()) {}
274
275 InsertPointGuard(const InsertPointGuard &) = delete;
276 InsertPointGuard &operator=(const InsertPointGuard &) = delete;
277
278 ~InsertPointGuard() {
279 Builder.restoreIP(InsertPoint(Block, Point));
280 Builder.SetCurrentDebugLocation(DbgLoc);
281 }
282 };
283
284 // RAII object that stores the current fast math settings and restores
285 // them when the object is destroyed.
286 class FastMathFlagGuard {
287 IRBuilderBase &Builder;
288 FastMathFlags FMF;
289 MDNode *FPMathTag;
290
291 public:
292 FastMathFlagGuard(IRBuilderBase &B)
293 : Builder(B), FMF(B.FMF), FPMathTag(B.DefaultFPMathTag) {}
294
295 FastMathFlagGuard(const FastMathFlagGuard &) = delete;
296 FastMathFlagGuard &operator=(const FastMathFlagGuard &) = delete;
297
298 ~FastMathFlagGuard() {
299 Builder.FMF = FMF;
300 Builder.DefaultFPMathTag = FPMathTag;
301 }
302 };
303
304 //===--------------------------------------------------------------------===//
305 // Miscellaneous creation methods.
306 //===--------------------------------------------------------------------===//
307
308 /// Make a new global variable with initializer type i8*
309 ///
310 /// Make a new global variable with an initializer that has array of i8 type
311 /// filled in with the null terminated string value specified. The new global
312 /// variable will be marked mergable with any others of the same contents. If
313 /// Name is specified, it is the name of the global variable created.
314 GlobalVariable *CreateGlobalString(StringRef Str, const Twine &Name = "",
315 unsigned AddressSpace = 0);
316
317 /// Get a constant value representing either true or false.
318 ConstantInt *getInt1(bool V) {
319 return ConstantInt::get(getInt1Ty(), V);
320 }
321
322 /// Get the constant value for i1 true.
323 ConstantInt *getTrue() {
324 return ConstantInt::getTrue(Context);
325 }
326
327 /// Get the constant value for i1 false.
328 ConstantInt *getFalse() {
329 return ConstantInt::getFalse(Context);
330 }
331
332 /// Get a constant 8-bit value.
333 ConstantInt *getInt8(uint8_t C) {
334 return ConstantInt::get(getInt8Ty(), C);
335 }
336
337 /// Get a constant 16-bit value.
338 ConstantInt *getInt16(uint16_t C) {
339 return ConstantInt::get(getInt16Ty(), C);
340 }
341
342 /// Get a constant 32-bit value.
343 ConstantInt *getInt32(uint32_t C) {
344 return ConstantInt::get(getInt32Ty(), C);
345 }
346
347 /// Get a constant 64-bit value.
348 ConstantInt *getInt64(uint64_t C) {
349 return ConstantInt::get(getInt64Ty(), C);
350 }
351
352 /// Get a constant N-bit value, zero extended or truncated from
353 /// a 64-bit value.
354 ConstantInt *getIntN(unsigned N, uint64_t C) {
355 return ConstantInt::get(getIntNTy(N), C);
356 }
357
358 /// Get a constant integer value.
359 ConstantInt *getInt(const APInt &AI) {
360 return ConstantInt::get(Context, AI);
361 }
362
363 //===--------------------------------------------------------------------===//
364 // Type creation methods
365 //===--------------------------------------------------------------------===//
366
367 /// Fetch the type representing a single bit
368 IntegerType *getInt1Ty() {
369 return Type::getInt1Ty(Context);
370 }
371
372 /// Fetch the type representing an 8-bit integer.
373 IntegerType *getInt8Ty() {
374 return Type::getInt8Ty(Context);
375 }
376
377 /// Fetch the type representing a 16-bit integer.
378 IntegerType *getInt16Ty() {
379 return Type::getInt16Ty(Context);
380 }
381
382 /// Fetch the type representing a 32-bit integer.
383 IntegerType *getInt32Ty() {
384 return Type::getInt32Ty(Context);
385 }
386
387 /// Fetch the type representing a 64-bit integer.
388 IntegerType *getInt64Ty() {
389 return Type::getInt64Ty(Context);
390 }
391
392 /// Fetch the type representing a 128-bit integer.
393 IntegerType *getInt128Ty() { return Type::getInt128Ty(Context); }
394
395 /// Fetch the type representing an N-bit integer.
396 IntegerType *getIntNTy(unsigned N) {
397 return Type::getIntNTy(Context, N);
398 }
399
400 /// Fetch the type representing a 16-bit floating point value.
401 Type *getHalfTy() {
402 return Type::getHalfTy(Context);
403 }
404
405 /// Fetch the type representing a 32-bit floating point value.
406 Type *getFloatTy() {
407 return Type::getFloatTy(Context);
408 }
409
410 /// Fetch the type representing a 64-bit floating point value.
411 Type *getDoubleTy() {
412 return Type::getDoubleTy(Context);
413 }
414
415 /// Fetch the type representing void.
416 Type *getVoidTy() {
417 return Type::getVoidTy(Context);
418 }
419
420 /// Fetch the type representing a pointer to an 8-bit integer value.
421 PointerType *getInt8PtrTy(unsigned AddrSpace = 0) {
422 return Type::getInt8PtrTy(Context, AddrSpace);
423 }
424
425 /// Fetch the type representing a pointer to an integer value.
426 IntegerType *getIntPtrTy(const DataLayout &DL, unsigned AddrSpace = 0) {
427 return DL.getIntPtrType(Context, AddrSpace);
428 }
429
430 //===--------------------------------------------------------------------===//
431 // Intrinsic creation methods
432 //===--------------------------------------------------------------------===//
433
434 /// Create and insert a memset to the specified pointer and the
435 /// specified value.
436 ///
437 /// If the pointer isn't an i8*, it will be converted. If a TBAA tag is
438 /// specified, it will be added to the instruction. Likewise with alias.scope
439 /// and noalias tags.
440 CallInst *CreateMemSet(Value *Ptr, Value *Val, uint64_t Size, unsigned Align,
441 bool isVolatile = false, MDNode *TBAATag = nullptr,
442 MDNode *ScopeTag = nullptr,
443 MDNode *NoAliasTag = nullptr) {
444 return CreateMemSet(Ptr, Val, getInt64(Size), Align, isVolatile,
445 TBAATag, ScopeTag, NoAliasTag);
446 }
447
448 CallInst *CreateMemSet(Value *Ptr, Value *Val, Value *Size, unsigned Align,
449 bool isVolatile = false, MDNode *TBAATag = nullptr,
450 MDNode *ScopeTag = nullptr,
451 MDNode *NoAliasTag = nullptr);
452
453 /// Create and insert an element unordered-atomic memset of the region of
454 /// memory starting at the given pointer to the given value.
455 ///
456 /// If the pointer isn't an i8*, it will be converted. If a TBAA tag is
457 /// specified, it will be added to the instruction. Likewise with alias.scope
458 /// and noalias tags.
459 CallInst *CreateElementUnorderedAtomicMemSet(Value *Ptr, Value *Val,
460 uint64_t Size, unsigned Align,
461 uint32_t ElementSize,
462 MDNode *TBAATag = nullptr,
463 MDNode *ScopeTag = nullptr,
464 MDNode *NoAliasTag = nullptr) {
465 return CreateElementUnorderedAtomicMemSet(Ptr, Val, getInt64(Size), Align,
466 ElementSize, TBAATag, ScopeTag,
467 NoAliasTag);
468 }
469
470 CallInst *CreateElementUnorderedAtomicMemSet(Value *Ptr, Value *Val,
471 Value *Size, unsigned Align,
472 uint32_t ElementSize,
473 MDNode *TBAATag = nullptr,
474 MDNode *ScopeTag = nullptr,
475 MDNode *NoAliasTag = nullptr);
476
477 /// Create and insert a memcpy between the specified pointers.
478 ///
479 /// If the pointers aren't i8*, they will be converted. If a TBAA tag is
480 /// specified, it will be added to the instruction. Likewise with alias.scope
481 /// and noalias tags.
482 CallInst *CreateMemCpy(Value *Dst, unsigned DstAlign, Value *Src,
483 unsigned SrcAlign, uint64_t Size,
484 bool isVolatile = false, MDNode *TBAATag = nullptr,
485 MDNode *TBAAStructTag = nullptr,
486 MDNode *ScopeTag = nullptr,
487 MDNode *NoAliasTag = nullptr) {
488 return CreateMemCpy(Dst, DstAlign, Src, SrcAlign, getInt64(Size),
489 isVolatile, TBAATag, TBAAStructTag, ScopeTag,
490 NoAliasTag);
491 }
492
493 CallInst *CreateMemCpy(Value *Dst, unsigned DstAlign, Value *Src,
494 unsigned SrcAlign, Value *Size,
495 bool isVolatile = false, MDNode *TBAATag = nullptr,
496 MDNode *TBAAStructTag = nullptr,
497 MDNode *ScopeTag = nullptr,
498 MDNode *NoAliasTag = nullptr);
499
500 /// Create and insert an element unordered-atomic memcpy between the
501 /// specified pointers.
502 ///
503 /// DstAlign/SrcAlign are the alignments of the Dst/Src pointers, respectively.
504 ///
505 /// If the pointers aren't i8*, they will be converted. If a TBAA tag is
506 /// specified, it will be added to the instruction. Likewise with alias.scope
507 /// and noalias tags.
508 CallInst *CreateElementUnorderedAtomicMemCpy(
509 Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign,
510 uint64_t Size, uint32_t ElementSize, MDNode *TBAATag = nullptr,
511 MDNode *TBAAStructTag = nullptr, MDNode *ScopeTag = nullptr,
512 MDNode *NoAliasTag = nullptr) {
513 return CreateElementUnorderedAtomicMemCpy(
514 Dst, DstAlign, Src, SrcAlign, getInt64(Size), ElementSize, TBAATag,
515 TBAAStructTag, ScopeTag, NoAliasTag);
516 }
517
518 CallInst *CreateElementUnorderedAtomicMemCpy(
519 Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign, Value *Size,
520 uint32_t ElementSize, MDNode *TBAATag = nullptr,
521 MDNode *TBAAStructTag = nullptr, MDNode *ScopeTag = nullptr,
522 MDNode *NoAliasTag = nullptr);
523
524 /// Create and insert a memmove between the specified
525 /// pointers.
526 ///
527 /// If the pointers aren't i8*, they will be converted. If a TBAA tag is
528 /// specified, it will be added to the instruction. Likewise with alias.scope
529 /// and noalias tags.
530 CallInst *CreateMemMove(Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign,
531 uint64_t Size, bool isVolatile = false,
532 MDNode *TBAATag = nullptr, MDNode *ScopeTag = nullptr,
533 MDNode *NoAliasTag = nullptr) {
534 return CreateMemMove(Dst, DstAlign, Src, SrcAlign, getInt64(Size), isVolatile,
535 TBAATag, ScopeTag, NoAliasTag);
536 }
537
538 CallInst *CreateMemMove(Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign,
539 Value *Size, bool isVolatile = false, MDNode *TBAATag = nullptr,
540 MDNode *ScopeTag = nullptr,
541 MDNode *NoAliasTag = nullptr);
542
543 /// \brief Create and insert an element unordered-atomic memmove between the
544 /// specified pointers.
545 ///
546 /// DstAlign/SrcAlign are the alignments of the Dst/Src pointers,
547 /// respectively.
548 ///
549 /// If the pointers aren't i8*, they will be converted. If a TBAA tag is
550 /// specified, it will be added to the instruction. Likewise with alias.scope
551 /// and noalias tags.
552 CallInst *CreateElementUnorderedAtomicMemMove(
553 Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign,
554 uint64_t Size, uint32_t ElementSize, MDNode *TBAATag = nullptr,
555 MDNode *TBAAStructTag = nullptr, MDNode *ScopeTag = nullptr,
556 MDNode *NoAliasTag = nullptr) {
557 return CreateElementUnorderedAtomicMemMove(
558 Dst, DstAlign, Src, SrcAlign, getInt64(Size), ElementSize, TBAATag,
559 TBAAStructTag, ScopeTag, NoAliasTag);
560 }
561
562 CallInst *CreateElementUnorderedAtomicMemMove(
563 Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign, Value *Size,
564 uint32_t ElementSize, MDNode *TBAATag = nullptr,
565 MDNode *TBAAStructTag = nullptr, MDNode *ScopeTag = nullptr,
566 MDNode *NoAliasTag = nullptr);
567
568 /// Create a vector fadd reduction intrinsic of the source vector.
569 /// The first parameter is a scalar accumulator value for ordered reductions.
570 CallInst *CreateFAddReduce(Value *Acc, Value *Src);
571
572 /// Create a vector fmul reduction intrinsic of the source vector.
573 /// The first parameter is a scalar accumulator value for ordered reductions.
574 CallInst *CreateFMulReduce(Value *Acc, Value *Src);
575
576 /// Create a vector int add reduction intrinsic of the source vector.
577 CallInst *CreateAddReduce(Value *Src);
578
579 /// Create a vector int mul reduction intrinsic of the source vector.
580 CallInst *CreateMulReduce(Value *Src);
581
582 /// Create a vector int AND reduction intrinsic of the source vector.
583 CallInst *CreateAndReduce(Value *Src);
584
585 /// Create a vector int OR reduction intrinsic of the source vector.
586 CallInst *CreateOrReduce(Value *Src);
587
588 /// Create a vector int XOR reduction intrinsic of the source vector.
589 CallInst *CreateXorReduce(Value *Src);
590
591 /// Create a vector integer max reduction intrinsic of the source
592 /// vector.
593 CallInst *CreateIntMaxReduce(Value *Src, bool IsSigned = false);
594
595 /// Create a vector integer min reduction intrinsic of the source
596 /// vector.
597 CallInst *CreateIntMinReduce(Value *Src, bool IsSigned = false);
598
599 /// Create a vector float max reduction intrinsic of the source
600 /// vector.
601 CallInst *CreateFPMaxReduce(Value *Src, bool NoNaN = false);
602
603 /// Create a vector float min reduction intrinsic of the source
604 /// vector.
605 CallInst *CreateFPMinReduce(Value *Src, bool NoNaN = false);
606
607 /// Create a lifetime.start intrinsic.
608 ///
609 /// If the pointer isn't i8* it will be converted.
610 CallInst *CreateLifetimeStart(Value *Ptr, ConstantInt *Size = nullptr);
611
612 /// Create a lifetime.end intrinsic.
613 ///
614 /// If the pointer isn't i8* it will be converted.
615 CallInst *CreateLifetimeEnd(Value *Ptr, ConstantInt *Size = nullptr);
616
617 /// Create a call to invariant.start intrinsic.
618 ///
619 /// If the pointer isn't i8* it will be converted.
620 CallInst *CreateInvariantStart(Value *Ptr, ConstantInt *Size = nullptr);
621
622 /// Create a call to Masked Load intrinsic
623 CallInst *CreateMaskedLoad(Value *Ptr, unsigned Align, Value *Mask,
624 Value *PassThru = nullptr, const Twine &Name = "");
625
626 /// Create a call to Masked Store intrinsic
627 CallInst *CreateMaskedStore(Value *Val, Value *Ptr, unsigned Align,
628 Value *Mask);
629
630 /// Create a call to Masked Gather intrinsic
631 CallInst *CreateMaskedGather(Value *Ptrs, unsigned Align,
632 Value *Mask = nullptr,
633 Value *PassThru = nullptr,
634 const Twine& Name = "");
635
636 /// Create a call to Masked Scatter intrinsic
637 CallInst *CreateMaskedScatter(Value *Val, Value *Ptrs, unsigned Align,
638 Value *Mask = nullptr);
639
640 /// Create an assume intrinsic call that allows the optimizer to
641 /// assume that the provided condition will be true.
642 CallInst *CreateAssumption(Value *Cond);
643
644 /// Create a call to the experimental.gc.statepoint intri