| Issue |
180413
|
| Summary |
mlir-opt crash: hasRank() assertion failure in tosa.max_pool2d shape inference
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
xvv4402
|
**Bug Op**: tosa.max_pool2d
**Bug Pass**: --tosa-infer-shapes
**Detailed Description**: The mlir-opt tool crashes during the --tosa-infer-shapes pass when encountering a tosa.max_pool2d operation with an unranked input tensor (e.g., tensor<*xi32>).
```mlir
module {
spirv.EntryPoint "GLCompute" @main
func.func @main() {
%0 = "tosa.const"() <{values = dense<0> : tensor<1xi32>}> : () -> tensor<1xi32>
%1 = tosa.bitwise_not %0 : (tensor<1xi32>) -> tensor<1xi32>
%2 = tosa.logical_right_shift %1, %0 : (tensor<1xi32>, tensor<1xi32>) -> tensor<1xi32>
%3 = tosa.reduce_product %2 {axis = 0 : i32} : (tensor<1xi32>) -> tensor<1xi32>
%4 = tensor.cast %3 : tensor<1xi32> to tensor<*xi32>
%5 = tosa.max_pool2d %4 {kernel = array<i64: 1, 1>, pad = array<i64: 0, 0, 0, 0>, stride = array<i64: 1, 1>} : (tensor<*xi32>) -> tensor<*xi32>
return
}
}
```
```
mlir-opt --tosa-infer-shapes temp.mlir
```
```
mlir-opt: llvm-project-llvmorg-22.1.0-rc1/mlir/lib/Interfaces/InferTypeOpInterface.cpp:121: int64_t mlir::ShapeAdaptor::getDimSize(int) const: Assertion `hasRank()' failed.
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: llvmorg-22.1.0-install/bin/mlir-opt --tosa-infer-shapes temp.mlir
#0 0x0000628e3b20de32 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (llvmorg-22.1.0-install/bin/mlir-opt+0x252fe32)
#1 0x0000628e3b20a5af llvm::sys::RunSignalHandlers() (llvmorg-22.1.0-install/bin/mlir-opt+0x252c5af)
#2 0x0000628e3b20a6fc SignalHandler(int, siginfo_t*, void*) Signals.cpp:0:0
#3 0x00007aab5dc45330 (/lib/x86_64-linux-gnu/libc.so.6+0x45330)
#4 0x00007aab5dc9eb2c __pthread_kill_implementation ./nptl/pthread_kill.c:44:76
#5 0x00007aab5dc9eb2c __pthread_kill_internal ./nptl/pthread_kill.c:78:10
#6 0x00007aab5dc9eb2c pthread_kill ./nptl/pthread_kill.c:89:10
#7 0x00007aab5dc4527e raise ./signal/../sysdeps/posix/raise.c:27:6
#8 0x00007aab5dc288ff abort ./stdlib/abort.c:81:7
#9 0x00007aab5dc2881b _nl_load_domain ./intl/loadmsgcat.c:1177:9
#10 0x00007aab5dc3b517 (/lib/x86_64-linux-gnu/libc.so.6+0x3b517)
#11 0x0000628e42c58621 mlir::ShapeAdaptor::getDimSize(int) const (llvmorg-22.1.0-install/bin/mlir-opt+0x9f7a621)
#12 0x0000628e40e4ae95 poolingInferReturnTypes(mlir::ShapeAdaptor, llvm::ArrayRef<long>, llvm::ArrayRef<long>, llvm::ArrayRef<long>, llvm::SmallVectorImpl<mlir::ShapedTypeComponents>&) TosaOps.cpp:0:0
#13 0x0000628e40e4b539 mlir::tosa::MaxPool2dOp::inferReturnTypeComponents(mlir::MLIRContext*, std::optional<mlir::Location>, mlir::tosa::MaxPool2dOpAdaptor, llvm::SmallVectorImpl<mlir::ShapedTypeComponents>&) (llvmorg-22.1.0-install/bin/mlir-opt+0x816d539)
#14 0x0000628e40e4b835 mlir::detail::InferShapedTypeOpInterfaceInterfaceTraits::Model<mlir::tosa::MaxPool2dOp>::inferReturnTypeComponents(mlir::MLIRContext*, std::optional<mlir::Location>, mlir::ValueShapeRange, mlir::DictionaryAttr, mlir::OpaqueProperties, mlir::RegionRange, llvm::SmallVectorImpl<mlir::ShapedTypeComponents>&) (llvmorg-22.1.0-install/bin/mlir-opt+0x816d835)
#15 0x0000628e3eb5e3ab (anonymous namespace)::propagateShapesInRegion(mlir::Region&, (anonymous namespace)::TypeModificationState&) TosaInferShapes.cpp:0:0
```
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
