def __init__(self, pcidev):
"""
check if the pci device is a GPU device and privided by vendor,
if yes, fill in self.nvh
"""
self.nvh = None
self.display = ':0'
self.gpu_flag = False
INFO_PATH = "/proc/driver/nvidia/gpus/"
super(GpuNV, self).__init__(pcidev)
nv.nvmlInit()
self.nvh = None
self.name = 'NV '
try:
self.nvh = nv.nvmlDeviceGetHandleByPciBusId(
pcidev.slot_name.encode())
self.nv_id = nv.nvmlDeviceGetIndex(self.nvh)
self.gpu_flag = True
self.pcidev = pcidev
except Exception as e:
if e.value == nv.NVML_ERROR_GPU_IS_LOST:
self.gpu_flag = True
# for each GPU, it have audio and vidoe pci device, supress warning for audio device
if e.value != nv.NVML_ERROR_INVALID_ARGUMENT:
logger.debug(
f'[{self.pci_dev.slot_name}/{self.name}] {str(e)}')
self.curve = GpuNV.CURVE
self.temp_delta = GpuNV.TEMP_DELTA
def is_working(self):
rv = True
try:
h = nv.nvmlDeviceGetHandleByPciBusId(
self.pcidev.slot_name.encode())
except Exception as e:
rv = False
return rv
def test_nvmlDeviceGetHandleByPciBusId(ngpus, pci_info):
handles = [
pynvml.nvmlDeviceGetHandleByPciBusId(pci_info[i].busId) for i in range(ngpus)
]
assert len(handles) == ngpus
