yezhengmao/nccl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove error logging from a normal path

Browse Source 
When initNet fails, we should not print the backtrace as it is
supposed to be normal operation (falling back to sockets)
This commit is contained in:
Sylvain Jeaugey 2018-12-04 14:47:41 -08:00
parent 4b39a4cf91
commit 57368189e1
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/init.cu
View File

@ -72,10 +72,11 @@ int ncclCudaFullCompCap() {
return ccMajor*10+ccMinor; return ccMajor*10+ccMinor;
} }
// Returns ncclInternalError if anything fails, causing that network to be ignored.
ncclResult_t initNet(ncclNet_t* net) { ncclResult_t initNet(ncclNet_t* net) {
int ndev; int ndev;
NCCLCHECK(net->init(ncclDebugLog)); if (net->init(ncclDebugLog) != ncclSuccess) return ncclInternalError;
NCCLCHECK(net->devices(&ndev)); if (net->devices(&ndev) != ncclSuccess) return ncclInternalError;
if (ndev <= 0) { if (ndev <= 0) {
INFO(NCCL_INIT|NCCL_NET, "Net/%s: call to devices() returned 0 devices.", net->name); INFO(NCCL_INIT|NCCL_NET, "Net/%s: call to devices() returned 0 devices.", net->name);
return ncclSystemError; return ncclSystemError;