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

Implement NIC identification and remapping (#420)

Browse Source 
* Add 1H16P GPU model

* Implement NIC identification and remapping

* Revert "Sort IB devices based on device name (#413)"

This reverts commit 2d0ed8dff6.

* Fix permute and check order

* Correction on IB speed reporting

* Revert "Allow user to link layer with RCCL_IB_HCA_SKIP_LINK_LAYER (#361)"

This reverts commit caf5c9992a.
This commit is contained in:
Wenkai Du
2021-08-24 09:42:04 -07:00
committed by GitHub
parent 5f15ed6e3e
commit 5c8380ff5b
11 changed files with 623 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/search.cc
+4 -1
View File
@@ -795,7 +795,7 @@ ncclResult_t ncclTopoCompute(ncclTopoSystem* system, struct ncclTopoGraph* graph
str = getenv("NCCL_RINGS");
if (str) {
// user supplied topo
NCCLCHECK(parseGraph(str, system, graph, NULL));
NCCLCHECK(parseGraph(str, system, graph, NULL, NULL));
if (graph->nChannels) {
system->type |= RCCL_TOPO_4P2H_ROME;
}
@@ -805,6 +805,9 @@ ncclResult_t ncclTopoCompute(ncclTopoSystem* system, struct ncclTopoGraph* graph
if (graph->nChannels) return ncclSuccess;
// try to match Rome 4P2H
NCCLCHECK(parseRome4P2H(system, graph));
if (graph->nChannels) return ncclSuccess;
// try to match 1H16P
NCCLCHECK(parse1H16P(system, graph));
}
if (graph->nChannels) return ncclSuccess;