AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Query XGMI link count through rocm_smi_lib API (#442)

Browse Source 
[ROCm/rccl commit: 14a184eb67]
This commit is contained in:
Wenkai Du
2021-10-26 10:30:20 -07:00
committed by GitHub
parent 0afd607328
commit 15143b1cfb
4 changed files with 67 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/graph/xml.cc
+3 -3
View File
@@ -671,8 +671,8 @@ ncclResult_t ncclTopoGetXmlFromGpu(struct ncclXmlNode* pciNode, nvmlDevice_t nvm
for (int i=0; i<deviceCnt; i++) {
if (i != dev) {
RSMI_IO_LINK_TYPE rsmi_type;
int hops, bw;
if (rocm_smi_getLinkInfo(dev, i, &rsmi_type, &hops, &bw) == ncclSuccess) {
int hops, count;
if (rocm_smi_getLinkInfo(dev, i, &rsmi_type, &hops, &count) == ncclSuccess) {
if (rsmi_type == RSMI_IOLINK_TYPE_XGMI && hops == 1) {
char busIdStr[] = "00000000:00:00.0";
CUDACHECK(hipDeviceGetPCIBusId(busIdStr, sizeof(busIdStr), i));
@@ -685,7 +685,7 @@ ncclResult_t ncclTopoGetXmlFromGpu(struct ncclXmlNode* pciNode, nvmlDevice_t nvm
if (nvlNode == NULL) {
NCCLCHECK(xmlAddNode(xml, gpuNode, "xgmi", &nvlNode));
NCCLCHECK(xmlSetAttr(nvlNode, "target", lowerId));
NCCLCHECK(xmlSetAttrInt(nvlNode, "count", 1));
NCCLCHECK(xmlSetAttrInt(nvlNode, "count", count));
}
}
}