| Issue |
177833
|
| Summary |
mlir-opt --convert-amdgpu-to-rocdl crashes on vector.mask with division by zero
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
Subway2023
|
# Description
mlir-opt version: 22.1.0-rc1
Running mlir-opt with the --convert-amdgpu-to-rocdl pass on a module containing vector.mask, vector.vscale, and a division by zero (arith.divsi) causes a crash in vector::MaskOp::fold
# Reproduce
location: If the **vector.mask** is commented out, the crash no longer occurs.
```
module {
func.func @main() {
%0 = arith.constant 0 : i32
%1 = vector.constant_mask [1] : vector<1xi1>
%2 = vector.vscale
%3 = vector.reduction <add>, %1 : vector<1xi1> into i1
%4 = arith.divsi %0, %0 : i32
vector.mask %1, %0 { vector.yield %0 : i32 } : vector<1xi1> -> i32
return
}
}
```
```
mlir-opt --convert-amdgpu-to-rocdl test.mlir
```
```
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace and instructions to reproduce the bug.
Stack dump:
0. Program arguments: /mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt --convert-amdgpu-to-rocdl /mnt/sdd1/sbw/mlir/data/genProgram_1218_morning/181617_7f2f/mlir_steps/181617_7f2f_5.mlir
#0 0x000062475a761e32 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt+0x252fe32)
#1 0x000062475a75e5af llvm::sys::RunSignalHandlers() (/mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt+0x252c5af)
#2 0x000062475a75e6fc SignalHandler(int, siginfo_t*, void*) Signals.cpp:0:0
#3 0x000077498ca45330 (/lib/x86_64-linux-gnu/libc.so.6+0x45330)
#4 0x0000624760d5ce28 mlir::vector::MaskOp::fold(mlir::vector::MaskOpGenericAdaptor<llvm::ArrayRef<mlir::Attribute>>, llvm::SmallVectorImpl<mlir::OpFoldResult>&) (/mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt+0x8b2ae28)
#5 0x0000624760d5d0e9 llvm::LogicalResult llvm::detail::UniqueFunctionBase<llvm::LogicalResult, mlir::Operation*, llvm::ArrayRef<mlir::Attribute>, llvm::SmallVectorImpl<mlir::OpFoldResult>&>::CallImpl<mlir::Op<mlir::vector::MaskOp, mlir::OpTrait::OneRegion, mlir::OpTrait::VariadicResults, mlir::OpTrait::ZeroSuccessors, mlir::OpTrait::AtLeastNOperands<1u>::Impl, mlir::OpTrait::SingleBlock, mlir::OpTrait::SingleBlockImplicitTerminator<mlir::vector::YieldOp>::Impl, mlir::OpTrait::NoRegionArguments, mlir::OpTrait::OpInvariants, mlir::vector::MaskingOpInterface::Trait, mlir::OpTrait::HasRecursiveMemoryEffects>::getFoldHookFn()::'lambda'(mlir::Operation*, llvm::ArrayRef<mlir::Attribute>, llvm::SmallVectorImpl<mlir::OpFoldResult>&) const>(void*, mlir::Operation*, llvm::ArrayRef<mlir::Attribute>, llvm::SmallVectorImpl<mlir::OpFoldResult>&) (/mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt+0x8b2b0e9)
#6 0x0000624760df1301 mlir::RegisteredOperationName::Model<mlir::vector::MaskOp>::foldHook(mlir::Operation*, llvm::ArrayRef<mlir::Attribute>, llvm::SmallVectorImpl<mlir::OpFoldResult>&) (/mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt+0x8bbf301)
#7 0x0000624762377074 mlir::Operation::fold(llvm::ArrayRef<mlir::Attribute>, llvm::SmallVectorImpl<mlir::OpFoldResult>&) (/mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt+0xa145074)
#8 0x00006247623774e3 mlir::Operation::fold(llvm::SmallVectorImpl<mlir::OpFoldResult>&) (/mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt+0xa1454e3)
#9 0x00006247622ad951 mlir::OpBuilder::tryFold(mlir::Operation*, llvm::SmallVectorImpl<mlir::Value>&, llvm::SmallVectorImpl<mlir::Operation*>*) (/mnt/sdd1/sbw/mlir/tool/llvmorg-22.1.0-install/bin/mlir-opt+0xa07b951)
#10 0x0000624761f399b0 (anonymous namespace)::OperationLegalizer::legalizeWithFold(mlir::Operation*) DialectConversion.cpp:0:0
```
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
