Add chip specs (#681)

* Add perfmon config spec, enhance memory partition info.

* Add gfx950 perfmon config.

* Add High Freq variants in gfx942.

* Add backup detection methods for gpu model.

* Improve get_num_xcds logic by adding detection of 1to1 arch-to-compute_partition logic.

* Add default compute partition settings spx:8 for when gpu_model=None.

* Update gpu spec tests.

* Add backup compute partition detection.

---------

Signed-off-by: xuchen-amd <xuchen@amd.com>

[ROCm/rocprofiler-compute commit: f0fad19e8b]
This commit is contained in:
xuchen-amd
2025-05-29 16:35:34 -04:00
committed by GitHub
parent 7fd93cfed3
commit 6c40056aa9
13 changed files with 527 additions and 177 deletions
@@ -756,7 +756,9 @@ def run_prof(
if new_env and not using_v3() and not using_v1():
# flatten tcc for applicable mi300 input
f = path(workload_dir + "/out/pmc_1/results_" + fbase + ".csv")
xcds = mi_gpu_specs.get_num_xcds(mspec.gpu_model, mspec.compute_partition)
xcds = mi_gpu_specs.get_num_xcds(
mspec.gpu_arch, mspec.gpu_model, mspec.compute_partition
)
df = flatten_tcc_info_across_xcds(f, xcds, int(mspec._l2_banks))
df.to_csv(f, index=False)