2
0
Add tree algorithms for allreduce to improve performance at scale.
Add ncclCommAbort() and ncclCommGetAsyncError() to properly handle
network errors and be permit recover.
Detect initial CPU affinity and no longer escape it.
Este cometimento está contido em:
Sylvain Jeaugey
2018-12-13 15:56:12 -08:00
ascendente 4861e197fd
cometimento 1450d42675
66 ficheiros modificados com 3746 adições e 3251 eliminações
+1
Ver ficheiro
@@ -36,4 +36,5 @@ $(TXZPREPDIR)/% : %.in
-e "s/\$${nccl:Minor}/$(NCCL_MINOR)/g" \
-e "s/\$${nccl:Patch}/$(NCCL_PATCH)/g" \
-e "s/\$${nccl:Suffix}/$(NCCL_SUFFIX)/g" \
-e "s/\$${pkg:Revision}/$(PKG_REVISION)/g" \
$< > $@
+2 -1
Ver ficheiro
@@ -25,8 +25,9 @@ NCCL_MAJOR=${nccl:Major}
NCCL_MINOR=${nccl:Minor}
NCCL_PATCH=${nccl:Patch}
NCCL_SUFFIX=${nccl:Suffix}
NCCL_BUILD=${pkg:Revision}
NCCLNAME="nccl-src_${NCCL_MAJOR}.${NCCL_MINOR}.${NCCL_PATCH}${NCCL_SUFFIX}"
NCCLNAME="nccl-src_${NCCL_MAJOR}.${NCCL_MINOR}.${NCCL_PATCH}${NCCL_SUFFIX}-${NCCL_BUILD}"
tar --exclude build \
--exclude ".git*" \