[Benchmark] Run xetla streamk gemm in benchmark

[ESI-SYD]

No description provided.

[whitneywhtsang]

@ESI-SYD Performance from https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/11230502839 is 240TFlops, the reported performance was 262TFlops, can you investigate why?
There are a lot of printouts like below, can they be cleaned up?

 problem size: (3072,3072), tiled_shape: (12,12), tiles: 144, dp_tiles: 96, sk_tiles: 48, iters_per_tile: 128, num_workgroups: 128, dp_workgroups: 96, dp_waves: 3, sk_groups_per_region: 32, sk_regions: 1, sk_waves: 1, sk_iters_per_normal_group: 192, sk_big_groups_per_region: 0, avail_xecores: 32

Local range: {1, 8, 4} 
SK Score: 51

[ESI-SYD]

@ESI-SYD Performance from https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/11230502839 is 240TFlops, the reported performance was 262TFlops, can you investigate why? There are a lot of printouts like below, can they be cleaned up?

 problem size: (3072,3072), tiled_shape: (12,12), tiles: 144, dp_tiles: 96, sk_tiles: 48, iters_per_tile: 128, num_workgroups: 128, dp_workgroups: 96, dp_waves: 3, sk_groups_per_region: 32, sk_regions: 1, sk_waves: 1, sk_iters_per_normal_group: 192, sk_big_groups_per_region: 0, avail_xecores: 32

Local range: {1, 8, 4} 
SK Score: 51

Let me check, previously I can get ~252 locally

[ESI-SYD]

Try to disable prints intel/xetla#54

[whitneywhtsang]

What has changed? Is performance good now? Print disabled?

[ESI-SYD]

What has changed? Is performance good now? Print disabled?

xecores increased, 251.6 now, 96% , print disable change not landed (looks like not activate in their public repo).

https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/11338429336/job/31531618981

[ESI-SYD]

Note: Pre-commit checks failure not releated to this PR.

[whitneywhtsang]

In https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/11338429336, I see stream-k performance is only 103TFlops.

[ESI-SYD]

In https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/11338429336, I see stream-k performance is only 103TFlops.

        M       K       N   Triton-GB/s    XeTLA-GB/s  Triton-GB/s-min  XeTLA-GB/s-min  Triton-GB/s-max  XeTLA-GB/s-max  Triton-TFlops  XeTLA-TFlops  Triton-TFlops-min  XeTLA-TFlops-min  Triton-TFlops-max  XeTLA-TFlops-max  Triton-CV  XeTLA-CV
0  3072.0  4096.0  3072.0  5.033170e+07  5.033177e+07     5.033169e+07    5.033173e+07     5.033170e+07    5.033179e+07     103.746892    251.641847          93.622131        168.239498         110.720402        280.920817   0.054009  0.174175

[whitneywhtsang]

In https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/11338429336, I see stream-k performance is only 103TFlops.

        M       K       N   Triton-GB/s    XeTLA-GB/s  Triton-GB/s-min  XeTLA-GB/s-min  Triton-GB/s-max  XeTLA-GB/s-max  Triton-TFlops  XeTLA-TFlops  Triton-TFlops-min  XeTLA-TFlops-min  Triton-TFlops-max  XeTLA-TFlops-max  Triton-CV  XeTLA-CV
0  3072.0  4096.0  3072.0  5.033170e+07  5.033177e+07     5.033169e+07    5.033173e+07     5.033170e+07    5.033179e+07     103.746892    251.641847          93.622131        168.239498         110.720402        280.920817   0.054009  0.174175

Opps, I was looking at the wrong column.

[whitneywhtsang]

Please create an issue to track the removal of the prints.

[ESI-SYD]

Please create an issue to track the removal of the prints.

#2489

[etiotto]

@ESI-SYD reminder, the pre-commit is failing.