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

Increase minimal channels for gfx908 (#259)

Browse Source
This commit is contained in:
Wenkai Du
2020-08-26 11:40:11 -07:00
committed by GitHub
parent b0919dc46c
commit c5cbece6d0
8 changed files with 107 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/connect.cc
+2 -1
View File
@@ -255,7 +255,7 @@ int ncclMaxNchannels() {
return maxNchannels;
}
ncclResult_t ncclTopoPostset(struct ncclComm* comm, int* firstRanks, struct ncclTopoRanks** allTopoRanks, int* rings) {
ncclResult_t ncclTopoPostset(struct ncclComm* comm, int* firstRanks, struct ncclTopoRanks** allTopoRanks, int* rings, int gcn) {
// Gather data from all ranks
int *ringRecv, *ringSend, *ringPrev, *ringNext, *treeUpRecv, *treeUpSend, *treeDnRecv,*treeDnSend;
int nranks = comm->nRanks;
@@ -290,6 +290,7 @@ ncclResult_t ncclTopoPostset(struct ncclComm* comm, int* firstRanks, struct nccl
memcpy(ringNext+nChannels*nranks, ringNext, nChannels*nranks*sizeof(int));
int nc = nChannels*2;
if (gcn == 908) nc = std::max(nc, 4);
if (comm->topo->nodes[NET].count == 0 && comm->topo->type == RCCL_TOPO_CR8G) nc = nChannels*3;
if (comm->topo->nodes[NET].count && comm->topo->type == RCCL_TOPO_4P2H_ROME) nc = 4*comm->topo->nodes[NET].count;
int end = std::min((int)ncclMaxNchannels(), std::max(nc, ncclMinNchannels()));