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

enable building rccl for gfx950 (#1571)

Browse Source
This commit is contained in:
Pedram Alizadeh
2025-02-25 16:13:48 -05:00
committed by GitHub
parent daaa6e155f
commit f268553ee4
24 changed files with 66 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/connect.cc
+3 -3
View File
@@ -286,7 +286,7 @@ static ncclResult_t setTreeDown(struct ncclTree* tree, int* indexes, int d) {
static ncclResult_t connectTrees(struct ncclComm* comm, int* treeToParent, int* treeToChild0, int* treeToChild1, int* treePatterns) {
const int channelLimit = IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx94") ? 2*CHANNEL_LIMIT : CHANNEL_LIMIT;
const int channelLimit = (IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx94") || IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx950")) ? 2*CHANNEL_LIMIT : CHANNEL_LIMIT;
const int nChannels = (comm->nChannels > channelLimit) ? comm->nChannels / 2 : comm->nChannels;
const int nNodes = comm->nNodes, node = comm->node;
@@ -803,8 +803,8 @@ ncclResult_t ncclTopoPostset(struct ncclComm* comm, int* firstRanks, int* treePa
}
}
// Only use full MAXCHANNELS for gfx94x
int maxChannels = IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx94") ?
// Only use full MAXCHANNELS for gfx94x and gfx950
int maxChannels = (IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx94") || IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx950")) ?
((comm->topo->nodes[GPU].nodes[0].gpu.cu == 80 || comm->topo->nodes[GPU].nodes[0].gpu.cu == 20 || comm->topo->nodes[GPU].nodes[0].gpu.cu == 38)
? comm->topo->nodes[GPU].nodes[0].gpu.cu : MAXCHANNELS) : 2*CHANNEL_LIMIT;