sirhcm/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests 2 Actions 8 Activity

amd: pmc generic arch for gfx10+ (#13407)

Browse Source
This commit is contained in:
nimlgen
2025-11-22 12:31:23 +03:00
committed by GitHub
parent 1f8b24a6b9
commit 3a42680e22
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tinygrad/runtime/support/amd.py
+3 -1
View File
@@ -65,7 +65,9 @@ def import_ip_offsets(ip): return type("IPOFF", (object,), import_header(f"inclu
def import_pmc(ip) -> dict[str, tuple[str, int]]:
res:dict[str, tuple[str, int]] = {}
arch = f"gfx{ip[0]}{ip[1]:x}{ip[2]:x}"
# NOTE: precise arch for mi300+, generic for others, since rocm headers lack some archs
arch = f"gfx{ip[0]}{ip[1]:x}{ip[2]:x}" if ip[0] == 9 else f"gfx{ip[0]}"
for sec in header_download("rocprofiler-compute/src/rocprof_compute_soc/profile_configs/counter_defs.yaml", url=ROCM_URL).split('- name: ')[1:]:
for arch_spec in sec.split('- architectures:')[1:]: