[Mlir-commits] [mlir] [MLIR][NVGPU] Use `gpu.dynamic_shared_memory` in tests (PR #133051)

[Karlo Basioli]

basioli-k wrote:

It seems that tests `mlir/test/Integration/GPU/CUDA:sm90/gemm_f32_f16_f16_128x128x128.mlir.test` and `mlir/test/Integration/GPU/CUDA:sm90/gemm_pred_f32_f16_f16_128x128x128.mlir.test` are failing now with the following:

`mlir/test/Integration/GPU/CUDA:sm90/gemm_f32_f16_f16_128x128x128.mlir.test`
```
Input was:

<<<<<<

          1: Correct Results :0 

          2: Incorrect Results :16384 
```

`mlir/test/Integration/GPU/CUDA:sm90/gemm_pred_f32_f16_f16_128x128x128.mlir.test`
```
Input was:

<<<<<<

          1: Correct Results :0 

          2: Incorrect Results :16384 
```

https://github.com/llvm/llvm-project/pull/133051