Reapplying PR #1641 [AG and RS channel tuning] Add thread work threshold to tuning models and precompute reg index in LL128 (#1713)

* Reapply "[AG and RS channel tuning] Add thread work threshold to tuning models and precompute reg index in LL128 (#1641)"

This reverts commit 943ad6f7820739385a0b54e81f823d0df1dbf71c.

* Decreasing NCCL_LL128_SHMEM_ELEMS_PER_THREAD from 16 to 8

[ROCm/rccl commit: 3f7c08648f]
This commit is contained in:
Pedram Alizadeh authored and GitHub committed 2025-06-04 13:22:11 -04:00
1 parent a50ff2c3d3
commit 1ace5d05ed
7 files changed
+50 -17

No files matched your search

+1
View File
@@ -1773,6 +1773,7 @@ static ncclResult_t topoGetAlgoInfo(
// NVLS should not need more than 16 channels to get peak BW.
nc = comm->nvlsChannels;
} else {
rcclUpdateThreadThreshold(comm, nBytes, info, threadThreshold);
// Ring/Tree channel tuning
while (nBytes < nc * nt * threadThreshold) {
if (nc >= 2) nc--;