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

Add another Rome model (#249)

Browse Source 
* Add another Rome model

* Add gfx908 4P3L models and support

* Revert "Use cached value for detecting GDR support only once"

This reverts commit 67c8e72ce3.

* Skip using ibverb for GPU direct RDMA detection

* Fine tune one Rome model
This commit is contained in:
Wenkai Du
2020-08-17 10:51:02 -07:00
committed by GitHub
parent c985478133
commit a51e4071e3
7 changed files with 257 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/topo.cc
-5
View File
@@ -743,11 +743,6 @@ ncclResult_t ncclTopoSetAffinity(struct ncclTopoSystem* system, int rank) {
NCCLCHECK(ncclCpusetToStr(&finalMask, affinityStr));
INFO(NCCL_INIT, "Setting affinity for GPU %d to %s", gpu->gpu.dev, affinityStr);
SYSCHECK(sched_setaffinity(0, sizeof(cpu_set_t), &finalMask), "sched_setaffinity");
int ret = numa_run_on_node(cpu->id);
if (ret != 0)
INFO(NCCL_INIT, "Failed to run on numa node %ld", cpu->id);
else
INFO(NCCL_INIT, "Thread is set to run on numa node %ld", cpu->id);
}
return ncclSuccess;
}