[SLP]Do not require external uses for roots and single use for other instructions in computeMinimumValueSizes. #72679

alexey-bataev · 2023-11-17T17:08:27Z

After changes, that does not require support from InstCombine, we can
drop some extra requirements for values-to-be-demoted. No need to check
for external uses for roots/other instructions, just check that the
no non-vectorized insertelement instruction, which may require
widening.

llvmbot · 2023-11-17T17:08:58Z

@llvm/pr-subscribers-llvm-transforms

Author: Alexey Bataev (alexey-bataev)

Changes

After changes, that does not require support from InstCombine, we can
drop some extra requirements for values-to-be-demoted. No need to check
for external uses for roots/other instructions, just check that the
no non-vectorized insertelement instruction, which may require
widening.

Full diff: https://github.com/llvm/llvm-project/pull/72679.diff

2 Files Affected:

(modified) llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp (+6-8)
(modified) llvm/test/Transforms/SLPVectorizer/X86/root-trunc-extract-reuse.ll (+10-12)

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index bff3e49b14bc6bc..42e4540eb8ecbd7 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -13078,10 +13078,13 @@ bool BoUpSLP::collectValuesToDemote(Value *V,
     return true;
   }
 
-  // If the value is not a vectorized instruction in the expression with only
-  // one use, it cannot be demoted.
+  // If the value is not a vectorized instruction in the expression and not the
+  // used by the insertelement instruction, it cannot be demoted.
   auto *I = dyn_cast<Instruction>(V);
-  if (!I || !I->hasOneUse() || !getTreeEntry(I) || !Visited.insert(I).second)
+  if (!I || !getTreeEntry(I) || !Visited.insert(I).second ||
+      all_of(I->users(), [&](User *U) {
+        return isa<InsertElementInst>(U) && !getTreeEntry(U);
+      }))
     return false;
 
   switch (I->getOpcode()) {
@@ -13140,11 +13143,6 @@ bool BoUpSLP::collectValuesToDemote(Value *V,
 }
 
 void BoUpSLP::computeMinimumValueSizes() {
-  // If there are no external uses, the expression tree must be rooted by a
-  // store. We can't demote in-memory values, so there is nothing to do here.
-  if (ExternalUses.empty())
-    return;
-
   // We only attempt to truncate integer expressions.
   auto &TreeRoot = VectorizableTree[0]->Scalars;
   auto *TreeRootIT = dyn_cast<IntegerType>(TreeRoot[0]->getType());
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/root-trunc-extract-reuse.ll b/llvm/test/Transforms/SLPVectorizer/X86/root-trunc-extract-reuse.ll
index f48528e502b8cf1..af46b4f576234b2 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/root-trunc-extract-reuse.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/root-trunc-extract-reuse.ll
@@ -8,20 +8,18 @@ define i1 @test() {
 ; CHECK:       then:
 ; CHECK-NEXT:    br label [[ELSE]]
 ; CHECK:       else:
-; CHECK-NEXT:    [[TMP0:%.*]] = phi <2 x i1> [ zeroinitializer, [[THEN]] ], [ zeroinitializer, [[ENTRY:%.*]] ]
-; CHECK-NEXT:    [[TMP1:%.*]] = zext <2 x i1> [[TMP0]] to <2 x i32>
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i1> [[TMP0]], i32 0
-; CHECK-NEXT:    [[TMP3:%.*]] = zext i1 [[TMP2]] to i32
-; CHECK-NEXT:    [[BF_CAST162:%.*]] = and i32 [[TMP3]], 0
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> zeroinitializer, <2 x i32> [[TMP1]], <2 x i32> <i32 3, i32 1>
-; CHECK-NEXT:    [[T13:%.*]] = and <2 x i32> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP0:%.*]] = phi <2 x i32> [ zeroinitializer, [[THEN]] ], [ zeroinitializer, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x i32> [[TMP0]], i32 0
+; CHECK-NEXT:    [[BF_CAST162:%.*]] = and i32 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> zeroinitializer, <2 x i32> [[TMP0]], <2 x i32> <i32 3, i32 1>
+; CHECK-NEXT:    [[T13:%.*]] = and <2 x i32> [[TMP2]], zeroinitializer
 ; CHECK-NEXT:    br label [[ELSE1:%.*]]
 ; CHECK:       else1:
-; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x i32> [[T13]], <2 x i32> poison, <2 x i32> <i32 poison, i32 0>
-; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x i32> [[TMP5]], i32 [[BF_CAST162]], i32 0
-; CHECK-NEXT:    [[TMP7:%.*]] = icmp ugt <2 x i32> [[TMP6]], zeroinitializer
-; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1
-; CHECK-NEXT:    ret i1 [[TMP8]]
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <2 x i32> [[T13]], <2 x i32> poison, <2 x i32> <i32 poison, i32 0>
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> [[TMP3]], i32 [[BF_CAST162]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ugt <2 x i32> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
+; CHECK-NEXT:    ret i1 [[TMP6]]
 ;
 entry:
   br i1 false, label %then, label %else

alexey-bataev · 2023-11-30T14:39:51Z

Ping!

RKSimon · 2023-11-30T15:18:25Z

@alexey-bataev please can you rebase?

alexey-bataev · 2023-12-07T16:50:38Z

Ping!

alexey-bataev · 2024-01-03T21:19:23Z

Ping!

RKSimon

LGTM

RKSimon · 2023-11-17T17:24:03Z

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

-  // If the value is not a vectorized instruction in the expression with only
-  // one use, it cannot be demoted.
+  // If the value is not a vectorized instruction in the expression and not the
+  // used by the insertelement instruction, it cannot be demoted.


"and not the used by the insertelement instruction" -> "and not used by the insertelement instruction"?

Is this comment correct? It still doesn't seem right.

Missed your comments, for some reason did not get the updates in e-mails.
Fixed this message.

RKSimon · 2024-01-05T11:29:43Z

llvm/test/Transforms/SLPVectorizer/X86/int-bitcast-minbitwidth.ll

+; CHECK-NEXT:    [[TMP0:%.*]] = call i16 @llvm.vector.reduce.or.v4i16(<4 x i16> zeroinitializer)
+; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[TMP0]] to i32
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[TMP1]], 65535
+; CHECK-NEXT:    store i32 [[TMP2]], ptr null, align 4
 ; CHECK-NEXT:    ret void


Do we have better tests than these cases that will all constant fold away?

There are a few. I'll try to improve this one.

instructions in computeMinimumValueSizes. After changes, that does not require support from InstCombine, we can drop some extra requirements for values-to-be-demoted. No need to check for external uses for roots/other instructions, just check that the no non-vectorized insertelement instruction, which may require widening.

mstorsjo · 2024-01-11T09:09:55Z

This triggers failed asserts:

char a[];
short *b;
int c, d, e, f;
void g() {
  char *h;
  for (;;) {
    for (; f; ++f) {
      h[f] = b[0] * a[e] + b[c] * a[1] >> 7;
      ++b;
    }
    h += d;
  }
}

$ clang -target x86_64-linux-gnu -c repro.c -O2
clang: ../lib/IR/Instructions.cpp:3335: static llvm::CastInst* llvm::CastInst::Create(llvm::Instruction::CastOps, llvm::Value*, llvm::Type*, const llvm::Twine&, llvm::Instruction*): Assertion `castIsValid(op, S, Ty) && "Invalid cast!"' failed.

This comes up in compiling libvpx and openh264.

I'll push a revert soon.

…r other instructions in computeMinimumValueSizes. (#72679)" This reverts commit 408dce8. This triggered failed asserts with code like this: char a[]; short *b; int c, d, e, f; void g() { char *h; for (;;) { for (; f; ++f) { h[f] = b[0] * a[e] + b[c] * a[1] >> 7; ++b; } h += d; } } Compiled like this: $ clang -target x86_64-linux-gnu -c repro.c -O2 clang: ../lib/IR/Instructions.cpp:3335: static llvm::CastInst* llvm::CastInst::Create(llvm::Instruction::CastOps, llvm::Value*, llvm::Type*, const llvm::Twine&, llvm::Instruction*): Assertion `castIsValid(op, S, Ty) && "Invalid cast!"' failed.

…instructions in computeMinimumValueSizes. (#72679) After changes, that does not require support from InstCombine, we can drop some extra requirements for values-to-be-demoted. No need to check for external uses for roots/other instructions, just check that the no non-vectorized insertelement instruction, which may require widening. Review: #72679

mikaelholmen · 2024-01-12T07:27:17Z

Hi @alexey-bataev
The following starts crashing with this patch (also the re-applied one)

opt -passes=slp-vectorizer bbi-90943.ll -o /dev/nul

It crashes with

opt: ../lib/IR/Instructions.cpp:2956: static llvm::BinaryOperator *llvm::BinaryOperator::Create(llvm::Instruction::BinaryOps, llvm::Value *, llvm::Value *, const llvm::Twine &, llvm::Instruction *): Assertion `S1->getType() == S2->getType() && "Cannot create binary operator with two operands of differing type!"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.	Program arguments: ../../main-github/llvm/build-all/bin/opt -passes=slp-vectorizer bbi-90943.ll -o /dev/null
 #0 0x00005652edde3e67 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (../../main-github/llvm/build-all/bin/opt+0x63a8e67)
 #1 0x00005652edde19be llvm::sys::RunSignalHandlers() (../../main-github/llvm/build-all/bin/opt+0x63a69be)
 #2 0x00005652edde452f SignalHandler(int) (../../main-github/llvm/build-all/bin/opt+0x63a952f)
 #3 0x00007f6c391a4630 __restore_rt (/lib64/libpthread.so.0+0xf630)
 #4 0x00007f6c368eb387 raise (/lib64/libc.so.6+0x36387)
 #5 0x00007f6c368eca78 abort (/lib64/libc.so.6+0x37a78)
 #6 0x00007f6c368e41a6 __assert_fail_base (/lib64/libc.so.6+0x2f1a6)
 #7 0x00007f6c368e4252 (/lib64/libc.so.6+0x2f252)
 #8 0x00005652ed7b39e0 llvm::BinaryOperator::Create(llvm::Instruction::BinaryOps, llvm::Value*, llvm::Value*, llvm::Twine const&, llvm::BasicBlock*) (../../main-github/llvm/build-all/bin/opt+0x5d789e0)
 #9 0x00005652ee4d11a0 (anonymous namespace)::HorizontalReduction::tryToReduce(llvm::slpvectorizer::BoUpSLP&, llvm::TargetTransformInfo*, llvm::TargetLibraryInfo const&) (../../main-github/llvm/build-all/bin/opt+0x6a961a0)
#10 0x00005652ee4b4f18 llvm::SLPVectorizerPass::vectorizeHorReduction(llvm::PHINode*, llvm::Instruction*, llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&, llvm::TargetTransformInfo*, llvm::SmallVectorImpl<llvm::WeakTrackingVH>&) (../../main-github/llvm/build-all/bin/opt+0x6a79f18)
#11 0x00005652ee4e0250 bool llvm::SLPVectorizerPass::vectorizeCmpInsts<std::reverse_iterator<llvm::CmpInst* const*> >(llvm::iterator_range<std::reverse_iterator<llvm::CmpInst* const*> >, llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&) (../../main-github/llvm/build-all/bin/opt+0x6aa5250)
#12 0x00005652ee4b6917 llvm::SLPVectorizerPass::vectorizeChainsInBlock(llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&)::$_146::operator()(bool) const (../../main-github/llvm/build-all/bin/opt+0x6a7b917)
#13 0x00005652ee4abf76 llvm::SLPVectorizerPass::vectorizeChainsInBlock(llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&) (../../main-github/llvm/build-all/bin/opt+0x6a70f76)
#14 0x00005652ee4a98d0 llvm::SLPVectorizerPass::runImpl(llvm::Function&, llvm::ScalarEvolution*, llvm::TargetTransformInfo*, llvm::TargetLibraryInfo*, llvm::AAResults*, llvm::LoopInfo*, llvm::DominatorTree*, llvm::AssumptionCache*, llvm::DemandedBits*, llvm::OptimizationRemarkEmitter*) (../../main-github/llvm/build-all/bin/opt+0x6a6e8d0)
#15 0x00005652ee4a8e51 llvm::SLPVectorizerPass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x6a6de51)
#16 0x00005652edffb7fd llvm::detail::PassModel<llvm::Function, llvm::SLPVectorizerPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function> >::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x65c07fd)
#17 0x00005652ed815634 llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function> >::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x5dda634)
#18 0x00005652ebbcbe3d llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function> >, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function> >::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x4190e3d)
#19 0x00005652ed819a1e llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0x5ddea1e)
#20 0x00005652ebbcbbdd llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module> >::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0x4190bdd)
#21 0x00005652ed814756 llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module> >::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0x5dd9756)
#22 0x00005652eb6e3d43 llvm::runPassPipeline(llvm::StringRef, llvm::Module&, llvm::TargetMachine*, llvm::TargetLibraryInfoImpl*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::StringRef, llvm::ArrayRef<llvm::PassPlugin>, llvm::opt_tool::OutputKind, llvm::opt_tool::VerifierKind, bool, bool, bool, bool, bool, bool, bool) (../../main-github/llvm/build-all/bin/opt+0x3ca8d43)
#23 0x00005652eb6f2444 main (../../main-github/llvm/build-all/bin/opt+0x3cb7444)
#24 0x00007f6c368d7555 __libc_start_main (/lib64/libc.so.6+0x22555)
#25 0x00005652eb6ddee9 _start (../../main-github/llvm/build-all/bin/opt+0x3ca2ee9)
Abort (core dumped)

bbi-90943.ll.gz

alexey-bataev · 2024-01-12T11:08:02Z

Hi @alexey-bataev The following starts crashing with this patch (also the re-applied one)

opt -passes=slp-vectorizer bbi-90943.ll -o /dev/nul

It crashes with

opt: ../lib/IR/Instructions.cpp:2956: static llvm::BinaryOperator *llvm::BinaryOperator::Create(llvm::Instruction::BinaryOps, llvm::Value *, llvm::Value *, const llvm::Twine &, llvm::Instruction *): Assertion `S1->getType() == S2->getType() && "Cannot create binary operator with two operands of differing type!"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.	Program arguments: ../../main-github/llvm/build-all/bin/opt -passes=slp-vectorizer bbi-90943.ll -o /dev/null
 #0 0x00005652edde3e67 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (../../main-github/llvm/build-all/bin/opt+0x63a8e67)
 #1 0x00005652edde19be llvm::sys::RunSignalHandlers() (../../main-github/llvm/build-all/bin/opt+0x63a69be)
 #2 0x00005652edde452f SignalHandler(int) (../../main-github/llvm/build-all/bin/opt+0x63a952f)
 #3 0x00007f6c391a4630 __restore_rt (/lib64/libpthread.so.0+0xf630)
 #4 0x00007f6c368eb387 raise (/lib64/libc.so.6+0x36387)
 #5 0x00007f6c368eca78 abort (/lib64/libc.so.6+0x37a78)
 #6 0x00007f6c368e41a6 __assert_fail_base (/lib64/libc.so.6+0x2f1a6)
 #7 0x00007f6c368e4252 (/lib64/libc.so.6+0x2f252)
 #8 0x00005652ed7b39e0 llvm::BinaryOperator::Create(llvm::Instruction::BinaryOps, llvm::Value*, llvm::Value*, llvm::Twine const&, llvm::BasicBlock*) (../../main-github/llvm/build-all/bin/opt+0x5d789e0)
 #9 0x00005652ee4d11a0 (anonymous namespace)::HorizontalReduction::tryToReduce(llvm::slpvectorizer::BoUpSLP&, llvm::TargetTransformInfo*, llvm::TargetLibraryInfo const&) (../../main-github/llvm/build-all/bin/opt+0x6a961a0)
#10 0x00005652ee4b4f18 llvm::SLPVectorizerPass::vectorizeHorReduction(llvm::PHINode*, llvm::Instruction*, llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&, llvm::TargetTransformInfo*, llvm::SmallVectorImpl<llvm::WeakTrackingVH>&) (../../main-github/llvm/build-all/bin/opt+0x6a79f18)
#11 0x00005652ee4e0250 bool llvm::SLPVectorizerPass::vectorizeCmpInsts<std::reverse_iterator<llvm::CmpInst* const*> >(llvm::iterator_range<std::reverse_iterator<llvm::CmpInst* const*> >, llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&) (../../main-github/llvm/build-all/bin/opt+0x6aa5250)
#12 0x00005652ee4b6917 llvm::SLPVectorizerPass::vectorizeChainsInBlock(llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&)::$_146::operator()(bool) const (../../main-github/llvm/build-all/bin/opt+0x6a7b917)
#13 0x00005652ee4abf76 llvm::SLPVectorizerPass::vectorizeChainsInBlock(llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&) (../../main-github/llvm/build-all/bin/opt+0x6a70f76)
#14 0x00005652ee4a98d0 llvm::SLPVectorizerPass::runImpl(llvm::Function&, llvm::ScalarEvolution*, llvm::TargetTransformInfo*, llvm::TargetLibraryInfo*, llvm::AAResults*, llvm::LoopInfo*, llvm::DominatorTree*, llvm::AssumptionCache*, llvm::DemandedBits*, llvm::OptimizationRemarkEmitter*) (../../main-github/llvm/build-all/bin/opt+0x6a6e8d0)
#15 0x00005652ee4a8e51 llvm::SLPVectorizerPass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x6a6de51)
#16 0x00005652edffb7fd llvm::detail::PassModel<llvm::Function, llvm::SLPVectorizerPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function> >::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x65c07fd)
#17 0x00005652ed815634 llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function> >::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x5dda634)
#18 0x00005652ebbcbe3d llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function> >, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function> >::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x4190e3d)
#19 0x00005652ed819a1e llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0x5ddea1e)
#20 0x00005652ebbcbbdd llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module> >::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0x4190bdd)
#21 0x00005652ed814756 llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module> >::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0x5dd9756)
#22 0x00005652eb6e3d43 llvm::runPassPipeline(llvm::StringRef, llvm::Module&, llvm::TargetMachine*, llvm::TargetLibraryInfoImpl*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::StringRef, llvm::ArrayRef<llvm::PassPlugin>, llvm::opt_tool::OutputKind, llvm::opt_tool::VerifierKind, bool, bool, bool, bool, bool, bool, bool) (../../main-github/llvm/build-all/bin/opt+0x3ca8d43)
#23 0x00005652eb6f2444 main (../../main-github/llvm/build-all/bin/opt+0x3cb7444)
#24 0x00007f6c368d7555 __libc_start_main (/lib64/libc.so.6+0x22555)
#25 0x00005652eb6ddee9 _start (../../main-github/llvm/build-all/bin/opt+0x3ca2ee9)
Abort (core dumped)

bbi-90943.ll.gz

Thanks, will fix it ASAP

alexey-bataev · 2024-01-12T12:50:45Z

Fixed in 39b2104

mikaelholmen · 2024-01-12T13:44:25Z

Fixed in 39b2104

Thanks!

…instructions in computeMinimumValueSizes. (llvm#72679) After changes, that does not require support from InstCombine, we can drop some extra requirements for values-to-be-demoted. No need to check for external uses for roots/other instructions, just check that the no non-vectorized insertelement instruction, which may require widening.

…r other instructions in computeMinimumValueSizes. (llvm#72679)" This reverts commit 408dce8. This triggered failed asserts with code like this: char a[]; short *b; int c, d, e, f; void g() { char *h; for (;;) { for (; f; ++f) { h[f] = b[0] * a[e] + b[c] * a[1] >> 7; ++b; } h += d; } } Compiled like this: $ clang -target x86_64-linux-gnu -c repro.c -O2 clang: ../lib/IR/Instructions.cpp:3335: static llvm::CastInst* llvm::CastInst::Create(llvm::Instruction::CastOps, llvm::Value*, llvm::Type*, const llvm::Twine&, llvm::Instruction*): Assertion `castIsValid(op, S, Ty) && "Invalid cast!"' failed.

…instructions in computeMinimumValueSizes. (llvm#72679) After changes, that does not require support from InstCombine, we can drop some extra requirements for values-to-be-demoted. No need to check for external uses for roots/other instructions, just check that the no non-vectorized insertelement instruction, which may require widening. Review: llvm#72679

alexey-bataev requested a review from RKSimon November 17, 2023 17:08

llvmbot added vectorizers llvm:transforms labels Nov 17, 2023

alexey-bataev force-pushed the SLPImproveCollectValuesToDemote branch 3 times, most recently from 0b9adf8 to 2c9bc62 Compare November 27, 2023 13:24

alexey-bataev force-pushed the SLPImproveCollectValuesToDemote branch 2 times, most recently from 1c699bc to 0a526cc Compare December 1, 2023 17:53

alexey-bataev force-pushed the SLPImproveCollectValuesToDemote branch from 0a526cc to 708edc7 Compare January 3, 2024 20:41

alexey-bataev force-pushed the SLPImproveCollectValuesToDemote branch from 708edc7 to 9f174f5 Compare January 10, 2024 15:15

RKSimon approved these changes Jan 10, 2024

View reviewed changes

alexey-bataev force-pushed the SLPImproveCollectValuesToDemote branch from 9f174f5 to 486f4bb Compare January 10, 2024 19:06

alexey-bataev merged commit 408dce8 into llvm:main Jan 10, 2024

alexey-bataev deleted the SLPImproveCollectValuesToDemote branch January 10, 2024 19:07

hiraditya mentioned this pull request Apr 22, 2024

error: Invalid cast (Producer: 'LLVM18.0.1' Reader: 'LLVM 18.0.1' #89673

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[SLP]Do not require external uses for roots and single use for other instructions in computeMinimumValueSizes. #72679

[SLP]Do not require external uses for roots and single use for other instructions in computeMinimumValueSizes. #72679

alexey-bataev commented Nov 17, 2023

llvmbot commented Nov 17, 2023

alexey-bataev commented Nov 30, 2023

RKSimon commented Nov 30, 2023

alexey-bataev commented Dec 7, 2023

alexey-bataev commented Jan 3, 2024

RKSimon left a comment

RKSimon Nov 17, 2023

RKSimon Nov 30, 2023

RKSimon Dec 7, 2023

alexey-bataev Jan 10, 2024

RKSimon Jan 5, 2024

alexey-bataev Jan 10, 2024

mstorsjo commented Jan 11, 2024 •

edited

Loading

mikaelholmen commented Jan 12, 2024

alexey-bataev commented Jan 12, 2024

alexey-bataev commented Jan 12, 2024

mikaelholmen commented Jan 12, 2024

[SLP]Do not require external uses for roots and single use for other instructions in computeMinimumValueSizes. #72679

[SLP]Do not require external uses for roots and single use for other instructions in computeMinimumValueSizes. #72679

Conversation

alexey-bataev commented Nov 17, 2023

llvmbot commented Nov 17, 2023

alexey-bataev commented Nov 30, 2023

RKSimon commented Nov 30, 2023

alexey-bataev commented Dec 7, 2023

alexey-bataev commented Jan 3, 2024

RKSimon left a comment

Choose a reason for hiding this comment

RKSimon Nov 17, 2023

Choose a reason for hiding this comment

RKSimon Nov 30, 2023

Choose a reason for hiding this comment

RKSimon Dec 7, 2023

Choose a reason for hiding this comment

alexey-bataev Jan 10, 2024

Choose a reason for hiding this comment

RKSimon Jan 5, 2024

Choose a reason for hiding this comment

alexey-bataev Jan 10, 2024

Choose a reason for hiding this comment

mstorsjo commented Jan 11, 2024 • edited Loading

mikaelholmen commented Jan 12, 2024

alexey-bataev commented Jan 12, 2024

alexey-bataev commented Jan 12, 2024

mikaelholmen commented Jan 12, 2024

mstorsjo commented Jan 11, 2024 •

edited

Loading