search

intel / pti-gpu

Profiling Tools Interfaces for GPU (PTI for GPU) is a set of Getting Started Documentation and Tools Library to start performance analysis on Intel(R) Processor Graphics easily
MIT License
202 stars 57 forks source link

Add suport for `intel_get_dual_subslice_id` #9

Closed TApplencourt closed 3 years ago

TApplencourt commented 3 years ago

Hi Anton,

Maybe you already know, but IGC just added support for intel_get_dual_subslice_id (https://github.com/intel/intel-graphics-compiler/commit/cc63edd997606f0b32c5f1648d047e1461821bba). You may want to update cl_gemm_inst and co at some point.

If you are short on time, I can submit a PR.

Thanks again for this repo!

Regards, Thomas

anton-v-gorshkov commented 3 years ago

Thanks, Thomas! I'll definitely add this, just need to wait for official IGC and compute-runtime releases to keep things testable. It will probably take like a week or two (current IGC release is based on 16 days last commit). Will notify you when it will be done.

anton-v-gorshkov commented 3 years ago

Added in version 0.32.0

TApplencourt commented 3 years ago

Thanks a lot! My cluster is on maintenance, so I can't try it yet :( Maybe a stupid remark but did you forget to update the computation of "Total number of HW thread" (https://github.com/intel/pti-gpu/blob/360b835279fdb3f279221dc0143ef2593e9e3641/samples/cl_gemm_inst/main.cc#L192) with the DSS count?

anton-v-gorshkov commented 3 years ago

Thanks Thomas! Indeed it's a bug, fixed in newest version.