microsoft/mscclpp
1
0
Fork 0
mirror of https://github.com/microsoft/mscclpp.git synced 2026-05-24 23:06:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Tighten NVML IPC domain hash lookup

Browse Source 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This commit is contained in:
Qinghua Zhou
2026-05-16 23:23:30 +00:00
parent 594dc79657
commit 18d37379d2
1 changed files with 5 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
src/core/utils_internal.cc
View File

@@ -208,30 +208,18 @@ uint64_t getFabricHash(const nvmlGpuFabricInfo_t& fabricInfo) {
bool tryGetNvmlIpcDomainHash(uint64_t& ipcDomainHash) {
// Use the current CUDA device; callers must set the rank's device before querying.
int deviceId;
if (cudaGetDevice(&deviceId) != cudaSuccess) {
return false;
}
char pciBusId[] = "00000000:00:00.0";
if (cudaDeviceGetPCIBusId(pciBusId, sizeof(pciBusId), deviceId) != cudaSuccess) {
if (cudaGetDevice(&deviceId) != cudaSuccess ||
cudaDeviceGetPCIBusId(pciBusId, sizeof(pciBusId), deviceId) != cudaSuccess) {
return false;
}
static NvmlState nvml;
if (!nvml.isInitialized()) {
return false;
}
nvmlDevice_t nvmlDevice;
if (nvmlDeviceGetHandleByPciBusId_v2(pciBusId, &nvmlDevice) != NVML_SUCCESS) {
return false;
}
nvmlGpuFabricInfo_t fabricInfo = {};
if (nvmlDeviceGetGpuFabricInfo(nvmlDevice, &fabricInfo) != NVML_SUCCESS) {
return false;
}
if (fabricInfo.state != NVML_GPU_FABRIC_STATE_COMPLETED || fabricInfo.status != NVML_SUCCESS) {
if (!nvml.isInitialized() || nvmlDeviceGetHandleByPciBusId_v2(pciBusId, &nvmlDevice) != NVML_SUCCESS ||
nvmlDeviceGetGpuFabricInfo(nvmlDevice, &fabricInfo) != NVML_SUCCESS ||
fabricInfo.state != NVML_GPU_FABRIC_STATE_COMPLETED || fabricInfo.status != NVML_SUCCESS) {
return false;
}