AI/rocm-systems
2
0 Fork 0
Código Issues Pull Requests Ações Pacotes Projetos Versões Wiki Atividade

Fix ncclDevFuncId for AllReduceWithBias (#1980)

Ver Código-Fonte 
[ROCm/rccl commit: c35bc721ad]
Esse commit está contido em:
Nilesh M Negi authored and GitHub committed 2025-10-17 09:28:57 -05:00
1 pai fca120343f
commit 0aa56fb0a5
4 files changed
+49 -46
Baixar Arquivo de Patch Baixar Arquivo de Diff Expandir todos os arquivos Colapsar todos os arquivos

No files matched your search

projects/rccl/src/enqueue.cc
+2 -6
Ver Arquivo
@@ -501,13 +501,9 @@ ncclResult_t ncclPrepareTasks(struct ncclComm* comm, bool* algoNeedConnect, bool
NCCLCHECK(getAlgoInfo(comm, &agg, collNetSupport, nvlsSupport, nTasksPerChannel, simInfo));
if(agg.func==ncclFuncAllReduce && agg.acc != nullptr)
{
agg.devFuncId = ncclDevFuncId(ncclFuncAllReduceWithBias, agg.opDev.op, agg.datatype, agg.algorithm, agg.protocol, agg.pipeline);
}
agg.devFuncId = ncclDevFuncId(agg.func, agg.opDev.op, agg.datatype, agg.algorithm, agg.protocol, 1, agg.pipeline);
else
{
agg.devFuncId = ncclDevFuncId(agg.func, agg.opDev.op, agg.datatype, agg.algorithm, agg.protocol, agg.pipeline);
}
agg.devFuncId = ncclDevFuncId(agg.func, agg.opDev.op, agg.datatype, agg.algorithm, agg.protocol, 0, agg.pipeline);
if (agg.devFuncId < 0) {
WARN("%s: unsupported collective. Please ensure the collective has been enabled in build.", __func__);
return ncclInvalidUsage;