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.


[ROCm/rccl commit: 1450d42675]
This commit is contained in:
Sylvain Jeaugey
2018-12-13 15:56:12 -08:00
parent 3a7974d4c4
commit ffc55560a1
66 changed files with 3746 additions and 3251 deletions
+2 -2
View File
@@ -1,6 +1,6 @@
Name: libnccl
Version: ${nccl:Major}.${nccl:Minor}.${nccl:Patch}
Release: ${pkg:Revision}
Version: ${nccl:Major}.${nccl:Minor}.${nccl:Patch}${nccl:Suffix}
Release: ${pkg:Revision}+cuda${cuda:Major}.${cuda:Minor}
Summary: NVIDIA Collectives Communication Library (NCCL) Runtime
Group: Development/Libraries
+1
View File
@@ -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" \
$< > $@
+2 -1
View File
@@ -25,8 +25,9 @@ NCCL_MAJOR=${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*" \