oec-hardware/oec-hardware-1.1.5-6-fix-bug-gpu.patch
2024-08-30 16:37:22 +08:00

37 lines
1.5 KiB
Diff

diff -Naur rpm/tests/compatible/gpu/nvidia_gpu.py oech/tests/compatible/gpu/nvidia_gpu.py
--- rpm/tests/compatible/gpu/nvidia_gpu.py 2024-08-29 19:48:19.472522063 +0800
+++ oech/tests/compatible/gpu/nvidia_gpu.py 2024-08-30 16:21:51.821261220 +0800
@@ -73,6 +73,11 @@
self.logger.info("Set default test gpu as %s." % id_num)
+ def clean_default_gpu(self):
+ if 'CUDA_VISIBLE_DEVICES' in os.environ:
+ del os.environ['CUDA_VISIBLE_DEVICES']
+ self.logger.info("Clean default test gpu.")
+
def test_pressure(self):
"""
Set pressure for gpu to test
@@ -87,10 +92,7 @@
os.chdir("/opt/gpu-burn")
cmd = self.command.run_cmd(
- "nvidia-smi -q | grep -i -A1 '%s' | grep 'Product Name' | cut -d ':' -f 2" % pci_num)
- device_name = cmd[0].strip()
- cmd = self.command.run_cmd(
- "./gpu_burn -l | grep -i '%s' | cut -d ':' -f 1 | awk '{print $2}'" % device_name)
+ "nvidia-smi -q | grep -i -A20 '%s' | grep 'Minor Number' | cut -d ':' -f 2" % pci_num)
run_id = cmd[0].strip()
cmd = getstatusoutput(
'nohup ./gpu_burn -i%s 10 &> %s &' % (run_id, self.gpu_burn))
@@ -185,6 +187,8 @@
result = False
self.logger.error("Test Vulkan failed.")
+ self.clean_default_gpu()
+
except Exception as e:
self.logger.error(
"Failed to run the script because compiling or setting variables: %s" % e)