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

Reduce RCCL kernel count as we don't pass first coll in argument

Browse Source
This commit is contained in:
Wenkai Du
2020-06-12 21:02:27 +00:00
parent a6d621176c
commit b257676f30
2 changed files with 2 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/enqueue.cc
+2 -10
View File
@@ -58,15 +58,7 @@
typedef void(*ncclKern_t)(struct ncclDevComm*);
// Must be consistent with the ncclFuncSet enum
static ncclKern_t const ncclKerns[4+NCCL_NUM_FUNCTIONS*ncclNumOps*ncclNumTypes*NCCL_NUM_ALGORITHMS*NCCL_NUM_PROTOCOLS] = {
NCCL_FUNCS2B(ncclBroadcast),
NCCL_FUNCS2A(ncclReduce),
NCCL_FUNCS2B(ncclAllGather),
NCCL_FUNCS2A(ncclReduceScatter),
NCCL_FUNCS2A(ncclAllReduce),
NCCL_KERN_NAME(ncclGather, copy, i8),
NCCL_KERN_NAME(ncclScatter, copy, i8),
NCCL_KERN_NAME(ncclAllToAll, copy, i8),
static ncclKern_t const ncclKerns[1] = {
NCCL_KERN_NAME(ncclSendRecv, copy, i8)
};
@@ -128,7 +120,7 @@ ncclResult_t setupLaunch(struct ncclComm* comm, hipLaunchParams* params) {
struct ncclColl* coll = comm->channels[0].collectives+comm->channels[0].collStart;
comm->args = comm->devComm;
params->func = (void *)ncclKerns[coll->funcIndex];
params->func = (void *)ncclKerns[0];
return ncclSuccess;
}