| Issue |
174545
|
| Summary |
[MLIR]Inconsistent results between `-convert-linalg-to-affine-loops` and `-convert-linalg-to-loops` for `linalg.matmul`
|
| Labels |
mlir
|
| Assignees |
|
| Reporter |
BealZephyr
|
**test commit**: [5cfd02](https://github.com/llvm/llvm-project/commit/5cfd02f44a43a2e2a085a633b022a62f64ba2b93)
## Description:
I encountered an issue with the linalg.matmul operation where the results differ when using two different conversion passes.
`-convert-linalg-to-affine-loops -lower-affine` produces the expected result.
`-convert-linalg-to-loops` produces an unexpected output, which suggests that the issue is tied to the loop lowering mechanism.
## test case
```
module {
func.func private @printMemrefI32(tensor<*xi32>)
func.func @main() {
%0 = "tosa.const"() <{values = dense<-6.693750e+01> : tensor<16x8xf16>}> : () -> tensor<16x8xf16>
%1 = "tosa.const"() <{values = dense<-1.133700e+02> : tensor<8x32xf64>}> : () -> tensor<8x32xf64>
%2 = "tosa.const"() <{values = dense<22> : tensor<16x32xi32>}> : () -> tensor<16x32xi32>
%3 = linalg.matmul {cast = #linalg.type_fn<cast_unsigned>} ins(%0, %1 : tensor<16x8xf16>, tensor<8x32xf64>) outs(%2 : tensor<16x32xi32>) -> tensor<16x32xi32>
%cast = tensor.cast %3 : tensor<16x32xi32> to tensor<*xi32>
call @printMemrefI32(%cast) : (tensor<*xi32>) -> ()
return
}
}
```
## Command:
### 1. Use `-convert-linalg-to-affine-loops -lower-affine`
#### cmd:
```
mlir-opt test.mlir -tosa-to-arith -one-shot-bufferize="bufferize-function-boundaries=1" -convert-linalg-to-affine-loops -lower-affine \
-convert-scf-to-cf -convert-to-llvm | mlir-runner -e main -entry-point-result=void -shared-libs=/home/workdir/llvm-project-latest/build/lib/libmlir_runner_utils.so
```
#### output:
```
[[22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, ...]]
```
### 2. Use `-convert-linalg-to-loops`:
#### cmd:
```
mlir-opt test.mlir -tosa-to-arith -one-shot-bufferize="bufferize-function-boundaries=1" -convert-linalg-to-loops \
-convert-scf-to-cf -convert-to-llvm | mlir-runner -e main -entry-point-result=void -shared-libs=/home/workdir/llvm-project-latest/build/lib/libmlir_runner_utils.so
```
#### output:
```
[[30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, ...]]
```
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
