Prepare functions for generalization of codegen/cost estimation.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp | ||
---|---|---|
7126–7133 | "Adds another" suggests we can combine these, but all it does it overwrite the mask/vector list - should we assert(CommonMask.empty() && InVectors.empty()) ? |
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp | ||
---|---|---|
7126–7133 | Yes, will combine in the next patches, currently this functionality is not used, so just added a simple version. Will add assertions |
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp | ||
---|---|---|
7312 | (style) unnecessary control-flow (please excuse the bad indentation) if (all_of(GatheredScalars, PoisonValue ::classof)) return Estimator.finalize(E->ReuseShuffleIndices); return Estimator.finalize(E->ReuseShuffleIndices, E->Scalars.size(), [&](Value *&Vec, SmallVectorImpl<int> &Mask) { Vec = Estimator.gather( GatheredScalars, Constant::getNullValue(FixedVectorType::get( GatheredScalars.front()->getType(), GatheredScalars.size()))); }); |
LGTM with one minor
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp | ||
---|---|---|
7123 | maybe add the assertion in these other add() methods as well (in a separate commit). |
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp | ||
---|---|---|
7123 | Most probably the next patch will make these functions more smart. |
Hi @ABataev, this commit causes assertion failure on the following test:
; ModuleID = 'reduced.bc' target datalayout = "e-i64:64-i128:128-v16:16-v32:32-n16:32:64" target triple = "nvptx64-unknown-unknown" define <2 x float> @baz() { bb: %extractelement = extractelement <2 x float> zeroinitializer, i64 0 %fcmp = fcmp uno float %extractelement, 0.000000e+00 %fcmp1 = fcmp uno float 0.000000e+00, 0.000000e+00 %or = or i1 %fcmp, %fcmp1 %fcmp2 = fcmp oeq float 0.000000e+00, 0.000000e+00 %or3 = or i1 %fcmp2, %or %fcmp4 = fcmp oeq float 0.000000e+00, 0.000000e+00 %or5 = or i1 %fcmp4, %or3 br i1 %or5, label %bb6, label %bb7 bb6: ; preds = %bb ret <2 x float> zeroinitializer bb7: ; preds = %bb ret <2 x float> zeroinitializer }
Could you please take a look?
$ opt -passes=slp-vectorizer -disable-output test.ll opt: /iusers/againull/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp:7128: void llvm::slpvectorizer::BoUpSLP::ShuffleCostEstimator::add(llvm::Value*, llvm::ArrayRef<int>): Assertion `CommonMask.empty() && InVectors.empty() && "Expected empty input mask/vectors."' failed. PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace. Stack dump: 0. Program arguments: opt -passes=slp-vectorizer -disable-output test.ll #0 0x00007f593ce67301 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMSupport.so.17git+0x1aa301) #1 0x00007f593ce64d24 SignalHandler(int) Signals.cpp:0:0 #2 0x00007f593c7d5db0 __restore_rt (/lib64/libc.so.6+0x59db0) #3 0x00007f593c82242c __pthread_kill_implementation (/lib64/libc.so.6+0xa642c) #4 0x00007f593c7d5d06 gsignal (/lib64/libc.so.6+0x59d06) #5 0x00007f593c7a87d3 abort (/lib64/libc.so.6+0x2c7d3) #6 0x00007f593c7a86fb _nl_load_domain.cold (/lib64/libc.so.6+0x2c6fb) #7 0x00007f593c7cec86 (/lib64/libc.so.6+0x52c86) #8 0x00007f593e20aa60 llvm::slpvectorizer::BoUpSLP::getEntryCost(llvm::slpvectorizer::BoUpSLP::TreeEntry const*, llvm::ArrayRef<llvm::Value*>, llvm::SmallPtrSetImpl<llvm::Value*>&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMVectorize.so.17git+0x116a60) #9 0x00007f593e23188b llvm::slpvectorizer::BoUpSLP::getTreeCost(llvm::ArrayRef<llvm::Value*>) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMVectorize.so.17git+0x13d88b) #10 0x00007f593e23513e (anonymous namespace)::HorizontalReduction::tryToReduce(llvm::slpvectorizer::BoUpSLP&, llvm::TargetTransformInfo*, llvm::TargetLibraryInfo const&) SLPVectorizer.cpp:0:0 #11 0x00007f593e237e03 llvm::SLPVectorizerPass::vectorizeHorReduction(llvm::PHINode*, llvm::Instruction*, llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&, llvm::TargetTransformInfo*, llvm::SmallVectorImpl<llvm::WeakTrackingVH>&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMVectorize.so.17git+0x143e03) #12 0x00007f593e23c0f4 llvm::SLPVectorizerPass::vectorizeRootInstruction(llvm::PHINode*, llvm::Instruction*, llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&, llvm::TargetTransformInfo*) (.constprop.0) SLPVectorizer.cpp:0:0 #13 0x00007f593e23f740 llvm::SLPVectorizerPass::vectorizeChainsInBlock(llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMVectorize.so.17git+0x14b740) #14 0x00007f593e2426db llvm::SLPVectorizerPass::runImpl(llvm::Function&, llvm::ScalarEvolution*, llvm::TargetTransformInfo*, llvm::TargetLibraryInfo*, llvm::AAResults*, llvm::LoopInfo*, llvm::DominatorTree*, llvm::AssumptionCache*, llvm::DemandedBits*, llvm::OptimizationRemarkEmitter*) (.part.0) SLPVectorizer.cpp:0:0 #15 0x00007f593e242fe2 llvm::SLPVectorizerPass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMVectorize.so.17git+0x14efe2) #16 0x00007f593fbb50ce llvm::detail::PassModel<llvm::Function, llvm::SLPVectorizerPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMPasses.so.17git+0x5e0ce) #17 0x00007f593feb79fb 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>&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMNVPTXCodeGen.so.17git+0x9b9fb) #18 0x00007f593d2674dd llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMCore.so.17git+0x2c54dd) #19 0x00007f593fea00ee llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMNVPTXCodeGen.so.17git+0x840ee) #20 0x00007f593d26538c llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/localdisk2/againull/llvm-project/build/bin/../lib/libLLVMCore.so.17git+0x2c338c) #21 0x0000000000422c9d 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) (/localdisk2/againull/llvm-project/build/bin/opt+0x422c9d) #22 0x0000000000415baa main (/localdisk2/againull/llvm-project/build/bin/opt+0x415baa) #23 0x00007f593c7c0e50 __libc_start_call_main (/lib64/libc.so.6+0x44e50) #24 0x00007f593c7c0efc __libc_start_main@GLIBC_2.2.5 (/lib64/libc.so.6+0x44efc) #25 0x00000000004167e5 _start (/localdisk2/againull/llvm-project/build/bin/opt+0x4167e5) Aborted (core dumped)
maybe add the assertion in these other add() methods as well (in a separate commit).