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

Fix GPU to NIC mapping for peertopeer (#994)

Browse Source
This commit is contained in:
pradeep-ramanna
2023-12-01 10:00:17 -06:00
committed by GitHub
parent e44e112a17
commit 0b53f79196
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/paths.cc
+2 -2
View File
@@ -548,7 +548,7 @@ static bool rcclPathOverride(struct ncclTopoSystem* system, uint64_t distance) {
for (i = 0; i < system->nodes[GPU].count; i++) {
for (j = 0; j < system->nodes[NET].count; j++) {
if (system->nodes[NET].nodes[j].net.busId - system->nodes[GPU].nodes[i].id == distance)
if ((system->nodes[NET].nodes[j].net.busId - system->nodes[GPU].nodes[i].id == distance) || (system->nodes[GPU].nodes[i].id - system->nodes[NET].nodes[j].net.busId == distance))
break;
}
if (j >= system->nodes[NET].count)
@@ -557,7 +557,7 @@ static bool rcclPathOverride(struct ncclTopoSystem* system, uint64_t distance) {
if (i >= system->nodes[GPU].count) {
for (i = 0; i < system->nodes[GPU].count; i++) {
for (j = 0; j < system->nodes[NET].count; j++) {
if (system->nodes[NET].nodes[j].net.busId - system->nodes[GPU].nodes[i].id == distance)
if ((system->nodes[NET].nodes[j].net.busId - system->nodes[GPU].nodes[i].id == distance) || (system->nodes[GPU].nodes[i].id - system->nodes[NET].nodes[j].net.busId == distance))
system->nodes[GPU].nodes[i].paths[NET][j].type = PATH_PXB;
}
}