diff options
| author | Bjorn Pettersson <bjorn.a.pettersson@ericsson.com> | 2017-10-26 13:42:55 +0000 |
|---|---|---|
| committer | Bjorn Pettersson <bjorn.a.pettersson@ericsson.com> | 2017-10-26 13:42:55 +0000 |
| commit | 22a2282da19ee2ad212c313a3a2da568c9e8d284 (patch) | |
| tree | bb31f292a778f938266320bfb4ddc2ec95f8394d /llvm/test/Transforms/LoadStoreVectorizer | |
| parent | 809df34efc653c6a471f951305a88bd5e675b522 (diff) | |
| download | bcm5719-llvm-22a2282da19ee2ad212c313a3a2da568c9e8d284.tar.gz bcm5719-llvm-22a2282da19ee2ad212c313a3a2da568c9e8d284.zip | |
[LSV] Skip all non-byte sizes, not only less than eight bits
Summary:
The code comments indicate that no effort has been spent on
handling load/stores when the size isn't a multiple of the
byte size correctly. However, the code only avoided types
smaller than 8 bits. So for example a load of an i28 could
still be considered as a candidate for vectorization.
This patch adjusts the code to behave according to the code
comment.
The test case used to hit the following assert when
trying to use "cast" an i32 to i28 using CreateBitOrPointerCast:
opt: ../lib/IR/Instructions.cpp:2565: Assertion `castIsValid(op, S, Ty) && "Invalid cast!"' failed.
#0 PrintStackTraceSignalHandler(void*)
#1 SignalHandler(int)
#2 __restore_rt
#3 __GI_raise
#4 __GI_abort
#5 __GI___assert_fail
#6 llvm::CastInst::Create(llvm::Instruction::CastOps, llvm::Value*, llvm::Type*, llvm::Twine const&, llvm::Instruction*)
#7 llvm::IRBuilder<llvm::ConstantFolder, llvm::IRBuilderDefaultInserter>::CreateBitOrPointerCast(llvm::Value*, llvm::Type*, llvm::Twine const&)
#8 (anonymous namespace)::Vectorizer::vectorizeLoadChain(llvm::ArrayRef<llvm::Instruction*>, llvm::SmallPtrSet<llvm::Instruction*, 16u>*)
Reviewers: arsenm
Reviewed By: arsenm
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D39295
llvm-svn: 316663
Diffstat (limited to 'llvm/test/Transforms/LoadStoreVectorizer')
| -rw-r--r-- | llvm/test/Transforms/LoadStoreVectorizer/X86/non-byte-size.ll | 30 |
1 files changed, 30 insertions, 0 deletions
diff --git a/llvm/test/Transforms/LoadStoreVectorizer/X86/non-byte-size.ll b/llvm/test/Transforms/LoadStoreVectorizer/X86/non-byte-size.ll new file mode 100644 index 00000000000..1f00f980eac --- /dev/null +++ b/llvm/test/Transforms/LoadStoreVectorizer/X86/non-byte-size.ll @@ -0,0 +1,30 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; RUN: opt < %s -load-store-vectorizer -S -mtriple=x86_64-unknown-linux-gnu | FileCheck %s + +%rec = type { i32, i28 } + +; We currently do not optimize this scenario. +; But we verify that we no longer crash when compiling this. +define void @test1(%rec* %out, %rec* %in) { +; CHECK-LABEL: @test1( +; CHECK-NEXT: [[IN1:%.*]] = getelementptr [[REC:%.*]], %rec* [[IN:%.*]], i16 0, i32 0 +; CHECK-NEXT: [[IN2:%.*]] = getelementptr [[REC]], %rec* [[IN]], i16 0, i32 1 +; CHECK-NEXT: [[VAL1:%.*]] = load i32, i32* [[IN1]], align 8 +; CHECK-NEXT: [[VAL2:%.*]] = load i28, i28* [[IN2]] +; CHECK-NEXT: [[OUT1:%.*]] = getelementptr [[REC]], %rec* [[OUT:%.*]], i16 0, i32 0 +; CHECK-NEXT: [[OUT2:%.*]] = getelementptr [[REC]], %rec* [[OUT]], i16 0, i32 1 +; CHECK-NEXT: store i32 [[VAL1]], i32* [[OUT1]], align 8 +; CHECK-NEXT: store i28 [[VAL2]], i28* [[OUT2]] +; CHECK-NEXT: ret void +; + %in1 = getelementptr %rec, %rec* %in, i16 0, i32 0 + %in2 = getelementptr %rec, %rec* %in, i16 0, i32 1 + %val1 = load i32, i32* %in1, align 8 + %val2 = load i28, i28* %in2 + %out1 = getelementptr %rec, %rec* %out, i16 0, i32 0 + %out2 = getelementptr %rec, %rec* %out, i16 0, i32 1 + store i32 %val1, i32* %out1, align 8 + store i28 %val2, i28* %out2 + ret void +} + |

