[Mlir-commits] [mlir] [mlir][tosa] Fix tosa-reduce-transposes to handle large constants better (PR #148755)

[llvmlistbot at llvm.org]

Hanumanth04 wrote:

> Thanks for this nice optimization!
> 
> Regarding testing, since this is a performance optimization, we couldn't accomplish anything further than the existing LIT tests in LLVM/MLIR.
> 
> Could you also enable this as default within TorchToTosa in Torch-MLIR, assuming that's your frontend ?

Hi Suraj,

Thanks for the clarification regarding testing.

Re: Could you also enable this as default within TorchToTosa in Torch-MLIR, assuming that's your frontend?

This pass has already been added to the TorchToTosa backend as part of the PR below; I'm not sure if this is what you mean. Please let me know if you meant something else.

https://github.com/llvm/torch-mlir/pull/4165

https://github.com/llvm/llvm-project/pull/148755