2.3.5-5

Add support for inter-node communication using sockets and InfiniBand/RoCE. Improve latency. Add support for aggregation. Improve LL/regular tuning. Remove tests as those are now at github.com/nvidia/nccl-tests . [ROCm/rccl commit: f93fe9bfd9]
2018-09-24 16:06:59 -07:00
parent 63ab6df5b3
commit 8ffcfac437
132 changed files with 12424 additions and 9415 deletions
@@ -0,0 +1,7 @@
+/*.debhelper.log
+/*.debhelper
+/*.substvars
+/tmp/
+/files
+/libnccl1/
+/libnccl-dev/
@@ -0,0 +1,58 @@
+#
+# Copyright (c) 2015-2018, NVIDIA CORPORATION. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+include ../../makefiles/common.mk
+include ../../makefiles/version.mk
+BUILDDIR ?= $(abspath ../../build)
+DEBPREPDIR := $(BUILDDIR)/debian
+PKGDIR  := $(BUILDDIR)/pkg/deb/
+
+DEBGEN_IN  := $(wildcard *.in)
+DEBGEN     := $(DEBGEN_IN:.in=)
+DEBFILES   := compat copyright libnccl-dev.install rules $(DEBGEN)
+DEBTARGETS := $(patsubst %, $(DEBPREPDIR)/%, $(DEBFILES))
+
+PKG_TIMESTAMP  := $(shell date -R)
+ARCH           := $(shell uname -m)
+PKG_ARCH       ?= $(shell uname -m | sed -e "s/x86_64/amd64/g" | sed -e "s/ppc64le/ppc64el/g")
+PKG_MULTIARCH  ?= $(shell $(CXX) -print-multiarch)
+ifeq ($(PKG_MULTIARCH),)
+# Hardwire the PKG_MULTIARCH directory as the RHEL6 distribution agnostic compiler (gcc 4.8.3) doesn't set it
+PKG_MULTIARCH  := $(ARCH)-linux-gnu
+endif
+
+prep : $(DEBTARGETS)
+	$(MAKE) -C ../.. lic BUILDDIR=$(BUILDDIR)
+
+build : prep
+	$(MAKE) -C ../.. src.build BUILDDIR=$(BUILDDIR)
+	@printf "Building Debian package\n"
+	(cd $(BUILDDIR); debuild -eLD_LIBRARY_PATH -uc -us -d -b)
+	mkdir -p $(PKGDIR)
+	mv $(BUILDDIR)/../libnccl*.deb $(PKGDIR)/
+
+clean:
+	rm -Rf $(DEBPREPDIR) $(PKGDIR)
+
+$(DEBPREPDIR)/% : %.in
+	@printf "Generating %-35s > %s\n" $< $@
+	mkdir -p $(DEBPREPDIR)
+	sed -e "s/\$${nccl:Major}/$(NCCL_MAJOR)/g" \
+	    -e "s/\$${nccl:Minor}/$(NCCL_MINOR)/g" \
+	    -e "s/\$${nccl:Patch}/$(NCCL_PATCH)/g" \
+	    -e "s/\$${nccl:Suffix}/$(NCCL_SUFFIX)/g" \
+	    -e "s/\$${cuda:Major}/$(CUDA_MAJOR)/g" \
+	    -e "s/\$${cuda:Minor}/$(CUDA_MINOR)/g" \
+	    -e "s/\$${pkg:Revision}/$(PKG_REVISION)/g" \
+	    -e "s/\$${pkg:Timestamp}/$(PKG_TIMESTAMP)/g" \
+	    -e "s/\$${pkg:Arch}/$(PKG_ARCH)/g" \
+	    -e "s/\$${pkg:MultiArch}/$(PKG_MULTIARCH)/g" \
+	    $< > $@
+
+$(DEBPREPDIR)/% : %
+	@printf "Grabbing   %-35s > %s\n" $< $@
+	mkdir -p $(DEBPREPDIR)
+	cp -f $< $@
@@ -0,0 +1,5 @@
+nccl (${nccl:Major}.${nccl:Minor}.${nccl:Patch}${nccl:Suffix}-${pkg:Revision}+cuda${cuda:Major}.${cuda:Minor}) trusty; urgency=medium
+
+  * Automatic Debian package from build
+
+ -- cudatools <cudatools@nvidia.com>  ${pkg:Timestamp}
@@ -0,0 +1 @@
+9
@@ -0,0 +1,30 @@
+Source: nccl
+Section: libs
+Maintainer: cudatools <cudatools@nvidia.com>
+Priority: optional
+Build-depends: debhelper(>=9)
+Standards-Version: 3.9.5
+
+Package: libnccl${nccl:Major}
+Section: libs
+Architecture: ${pkg:Arch}
+Depends: ${misc:Depends}, ${shlibs:Depends}
+Description: NVIDIA Collectives Communication Library (NCCL) Runtime
+ NCCL (pronounced "Nickel") is a stand-alone library of standard collective
+ communication routines for GPUs, implementing all-reduce, all-gather, reduce,
+ broadcast, and reduce-scatter.
+ It has been optimized to achieve high bandwidth on any platform using PCIe,
+ NVLink, NVswitch, as well as networking using InfiniBand Verbs or TCP/IP
+ sockets.
+
+Package: libnccl-dev
+Section: libdevel
+Architecture: ${pkg:Arch}
+Depends: ${misc:Depends}, ${shlibs:Depends}, libnccl${nccl:Major} (= ${binary:Version})
+Description: NVIDIA Collectives Communication Library (NCCL) Development Files
+ NCCL (pronounced "Nickel") is a stand-alone library of standard collective
+ communication routines for GPUs, implementing all-reduce, all-gather, reduce,
+ broadcast, and reduce-scatter.
+ It has been optimized to achieve high bandwidth on any platform using PCIe,
+ NVLink, NVswitch, as well as networking using InfiniBand Verbs or TCP/IP
+ sockets.
@@ -0,0 +1 @@
+../../LICENSE.txt
@@ -0,0 +1,9 @@
+[DEFAULT]
+debian-branch   = master
+upstream-branch = master
+
+ignore-new = True
+
+[git-buildpackage]
+
+no-purge = True
@@ -0,0 +1,3 @@
+include/nccl.h /usr/include
+lib/libnccl.so /usr/lib/${pkg:MultiArch}
+lib/libnccl_static.a /usr/lib/${pkg:MultiArch}
@@ -0,0 +1,2 @@
+lib/libnccl.so.${nccl:Major} /usr/lib/${pkg:MultiArch}
+lib/libnccl.so.${nccl:Major}.${nccl:Minor}.${nccl:Patch} /usr/lib/${pkg:MultiArch}
@@ -0,0 +1,13 @@
+#!/usr/bin/make -f
+
+%:
+	dh $@ --parallel
+
+override_dh_auto_install:
+	PREFIX=debian/tmp dh_auto_install
+
+override_dh_auto_test:
+	# Do not make test
+
+override_dh_auto_clean:
+	# Do not make clean
@@ -0,0 +1 @@
+3.0 (native)