Add get_device_uuid for rocm (#37694)

Signed-off-by: Tiffany Mintz <Tiffany.Mintz@amd.com>
This commit is contained in:
tmm77
2026-03-20 23:33:16 -04:00
committed by GitHub
parent c57d38d603
commit 1c472f8fe1

View File

@@ -28,6 +28,7 @@ try:
from amdsmi import (
AmdSmiException,
amdsmi_get_gpu_asic_info,
amdsmi_get_gpu_device_uuid,
amdsmi_get_processor_handles,
amdsmi_init,
amdsmi_shut_down,
@@ -608,6 +609,20 @@ class RocmPlatform(Platform):
return _ROCM_DEVICE_ID_NAME_MAP[device_name]
return asic_info["market_name"]
@classmethod
@with_amdsmi_context
def get_device_uuid(cls, device_id: int = 0) -> str:
try:
device = amdsmi_get_processor_handles()[device_id]
except AmdSmiException as error:
logger.error("amdsmi device query failed ", exc_info=error)
return ""
try:
device_uuid = amdsmi_get_gpu_device_uuid(device)
except AmdSmiException as error:
logger.error("amdsmi device uuid query failed ", exc_info=error)
return device_uuid
@classmethod
def get_device_total_memory(cls, device_id: int = 0) -> int:
device_props = torch.cuda.get_device_properties(device_id)