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

Fix LL128 proto selection to respect user setting (#1822)

Browse Source
This commit is contained in:
Rahul Vaidya
2025-10-09 14:08:03 -05:00
committed by GitHub
parent d22a39e954
commit 6b200ee6c5
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/tuning.cc
+1 -1
View File
@@ -736,7 +736,7 @@ ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCom
for (int c=0; c<NCCL_NUM_FUNCTIONS; c++) for (int a=0; a<NCCL_NUM_ALGORITHMS; a++) for (int p=0; p<NCCL_NUM_PROTOCOLS; p++) {
// Disable LL protocol on gfx12xx
int pEnable = (p == NCCL_PROTO_LL && IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx12")) ? 0 : protoEnable[c*NCCL_NUM_PROTOCOLS+p];
if (p == NCCL_PROTO_LL128) {
if (pEnable != 0 && p == NCCL_PROTO_LL128) {
#if defined(__HIP_PLATFORM_AMD__) || defined(__HIPCC__)
#if defined(ENABLE_LL128)
// Enable LL128 by default only on gfx90a with available tuning table