Benjamin Ulmer

Good catch. Looks like this README hasn't been updated in quite some time. In the meantime, the [tuning doc](https://github.com/ROCmSoftwarePlatform/Tensile/tree/develop/tuning_docs) has more up-to-date information on how to use these scripts.

Hi, I'm not sure I understand the question. By input size, do you mean the GEMM problem dimensions (M, N, K) you want to tune?

Hi @littlewu2508 Sorry this has taken so long to get to. I updated the PR to address the conflicts with the addition of gfx11 series support. Please look this over...

What are you trying to accomplish by setting the value for ISA here? Setting the ISA in the Tensile benchmark config yaml is not something we support.

Those sizes are all going to be memory limited as opposed to compute limited, so unfortunately, there's a pretty hard performance ceiling. Larger batch counts can help improve compute unit...

+1 to everything @cgmb said. I'm going to try and get the ball rolling on this and Tensile #1511 this week. > I have a technical question about this tuning,...

No open is fine as long as CI doesn't get to bogged down. We just won't merge until after 5.4 FC merges are done.

> I'll add them via the normal channels. It would probably be easiest to do this in the same PR that pulls the changes back into develop after this PR...

The CI failures look like the normals ones to me. After the force push, the commit in the PR once again matches the one used for the CQE tests, which...

A colleague suggested we change the longitude domain from [-pi, pi] to [0, 2pi]. ~Doing this seems to fix the issue (at least in this example). Is this the expected...