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

Disable Colltrace for --fast option (#778)

Browse Source 
* Disable Colltrace for --fast option

* Limit nprocs for CI
This commit is contained in:
Bertan Dogancay
2023-06-21 16:16:09 -04:00
committed by GitHub
parent 399d31ed40
commit 0c77c66221
8 changed files with 85 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/enqueue.cc
+7
View File
@@ -28,11 +28,18 @@ struct ncclKernelMatch {
};
typedef void(*ncclKern_t)(struct ncclDevComm* comm, uint64_t channelMask, struct ncclWork* workHead);
// Must be consistent with the ncclFuncSet enum
#ifdef ENABLE_COLLTRACE
static ncclKernelMatch const ncclKerns[2] = {
{(void *)NCCL_KERN_NAME(SendRecv, RING, SIMPLE, Sum, int8_t), true},
{(void *)NCCL_KERN_NAME_DEBUG(SendRecv, RING, SIMPLE, Sum, int8_t), true},
};
#else
static ncclKernelMatch const ncclKerns[1] = {
{(void*)NCCL_KERN_NAME(SendRecv, RING, SIMPLE, Sum, int8_t), true}
};
#endif
static ncclResult_t computeColl(struct ncclInfo* info /* input */, int* workFuncIndex, struct ncclWorkElem* work, struct ncclProxyOp* proxyOp /* output */);