Merge pull request #125 from wenkaidu/fix_nvml_id

Assign unused nmvlDev to avoid random number

[ROCm/rccl commit: 6759660529]
이 커밋은 다음에 포함됨:
Wenkai Du 작성하고 GitHub가 2019-08-19 09:08:13 -07:00를 커밋함
커밋 4df1defc3b
1 파일이 변경됨
+5
+5
파일 보기
@@ -17,6 +17,10 @@
// Convert a logical cudaDev index to the NVML device minor number
ncclResult_t getNvmlDevice(int cudaDev, int *nvmlDev) {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
// assign nmvlDev to be same as cudaDev to avoid garbage numbers
*nvmlDev = cudaDev;
#else
char busId[NVML_DEVICE_PCI_BUS_ID_BUFFER_SIZE];
nvmlDevice_t nvmlDevice;
unsigned int dev;
@@ -26,6 +30,7 @@ ncclResult_t getNvmlDevice(int cudaDev, int *nvmlDev) {
NCCLCHECK(wrapNvmlDeviceGetMinorNumber(nvmlDevice, &dev));
*nvmlDev = dev;
#endif
return ncclSuccess;
}