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

2.9.6-1

Browse Source 
Add support for CUDA graphs.
Fuse BCM Gen4 switches to avoid suboptimal performance on some platforms. Issue #439.
Fix bootstrap issue caused by connection reordering.
Fix CPU locking block.
Improve CollNet algorithm.
Improve performance on DGX A100 for communicators with only one GPU per node.
This commit is contained in:
Sylvain Jeaugey
2021-04-12 16:00:11 -07:00
parent 911d61f214
commit a46ea10583
43 changed files with 2687 additions and 1244 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/Makefile
+5 -5
View File
@@ -1,5 +1,5 @@
#
# Copyright (c) 2015-2020, NVIDIA CORPORATION. All rights reserved.
# Copyright (c) 2015-2021, NVIDIA CORPORATION. All rights reserved.
#
# See LICENSE.txt for license information
#
@@ -10,7 +10,7 @@ include ../makefiles/version.mk
##### src files
INCEXPORTS := nccl.h nccl_net.h
LIBSRCFILES := init.cc channel.cc bootstrap.cc transport.cc enqueue.cc group.cc debug.cc proxy.cc \
misc/nvmlwrap.cc misc/ibvwrap.cc misc/utils.cc misc/argcheck.cc \
misc/nvmlwrap.cc misc/ibvwrap.cc misc/gdrwrap.cc misc/utils.cc misc/argcheck.cc \
transport/p2p.cc transport/shm.cc transport/net.cc transport/net_socket.cc transport/net_ib.cc transport/coll_net.cc \
collectives/sendrecv.cc collectives/all_reduce.cc collectives/all_gather.cc collectives/broadcast.cc collectives/reduce.cc collectives/reduce_scatter.cc \
graph/topo.cc graph/paths.cc graph/search.cc graph/connect.cc graph/rings.cc graph/trees.cc graph/tuning.cc graph/xml.cc
@@ -56,8 +56,8 @@ ALWAYS_REBUILD:
$(LIBDIR)/$(LIBTARGET) $(LIBDIR)/$(STATICLIBTARGET) : $(LIBOBJ)
$(INCDIR)/nccl.h : nccl.h.in
# NCCL_VERSION(X,Y,Z) ((X) * 1000 + (Y) * 100 + (Z))
@$(eval NCCL_VERSION := $(shell printf "%d%d%02d" $(NCCL_MAJOR) $(NCCL_MINOR) $(NCCL_PATCH)))
# NCCL_VERSION(X,Y,Z) ((X) * 10000 + (Y) * 100 + (Z))
@$(eval NCCL_VERSION := $(shell printf "%d%02d%02d" $(NCCL_MAJOR) $(NCCL_MINOR) $(NCCL_PATCH)))
mkdir -p $(INCDIR)
@printf "Generating %-35s > %s\n" $< $@
sed -e "s/\$${nccl:Major}/$(NCCL_MAJOR)/g" \
@@ -129,7 +129,7 @@ install : lib
cp -P -v $(BUILDDIR)/lib/pkgconfig/* $(PREFIX)/lib/pkgconfig/
cp -v $(BUILDDIR)/include/* $(PREFIX)/include/
FILESTOFORMAT := $(shell find . -name ".\#*" -prune -o \( -name "*.cc" -o -name "*.h" \) -print | grep -v -E 'ibvwrap.h|nvmlwrap.h|nccl.h')
FILESTOFORMAT := $(shell find . -name ".\#*" -prune -o \( -name "*.cc" -o -name "*.h" \) -print | grep -v -E 'ibvwrap.h|nvmlwrap.h|gdrwrap.h|nccl.h')
# Note that formatting.mk defines a new target so in order to not overwrite the default target,
# it shouldn't be included at the top. Also, it uses the above definition of FILESTOFORMAT as well
# as the BUILDDIR variable.