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 .
2018-09-24 16:06:59 -07:00
parent 286916a1a3
commit f93fe9bfd9
132 changed files with 12424 additions and 9415 deletions
@@ -0,0 +1,62 @@
+#
+# 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)
+RPMPREPDIR := $(BUILDDIR)/redhat
+PKGDIR  := $(BUILDDIR)/pkg/rpm/
+
+RPMGEN_IN  := $(wildcard *.in)
+RPMGEN     := $(RPMGEN_IN:.in=)
+RPMFILES   := $(RPMGEN)
+RPMTARGETS := $(patsubst %, $(RPMPREPDIR)/%, $(RPMFILES))
+
+PKG_TIMESTAMP  := $(shell date -R)
+ARCH           := $(shell uname -m)
+PKG_ARCH       ?= $(shell uname -m)
+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 : $(RPMTARGETS)
+	$(MAKE) -C ../.. lic BUILDDIR=$(BUILDDIR)
+
+build : prep
+	$(MAKE) -C ../.. src.build BUILDDIR=$(BUILDDIR)
+	$(MAKE) -C ../txz build BUILDDIR=$(BUILDDIR)
+	@printf "Building Redhat package\n"
+	mkdir -p $(PKGDIR)
+	rpmbuild --define "_sourcedir $(BUILDDIR)/pkg/txz" \
+                 --define "_rpmdir $(PKGDIR)" \
+                 --define "_builddir $(PKGDIR)/build/" \
+                 --define "_buildrootdir $(PKGDIR)/buildroot/" \
+                 -bb $(BUILDDIR)/redhat/nccl.spec
+
+clean:
+	rm -Rf $(RPMPREPDIR) $(PKGDIR)
+
+$(RPMPREPDIR)/% : %.in
+	@printf "Generating %-35s > %s\n" $< $@
+	mkdir -p $(RPMPREPDIR)
+	sed -e "s/\$${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" \
+	    $< > $@
+
+$(RPMPREPDIR)/% : %
+	@printf "Grabbing   %-35s > %s\n" $< $@
+	mkdir -p $(RPMPREPDIR)
+	cp -f $< $@
@@ -0,0 +1,73 @@
+Name:           libnccl
+Version:        ${nccl:Major}.${nccl:Minor}.${nccl:Patch}
+Release:        ${pkg:Revision}
+Summary:        NVIDIA Collectives Communication Library (NCCL) Runtime
+
+Group:          Development/Libraries
+License:        BSD
+URL:            http://developer.nvidia.com/nccl
+Source0:        nccl_${nccl:Major}.${nccl:Minor}.${nccl:Patch}${nccl:Suffix}-${pkg:Revision}+cuda${cuda:Major}.${cuda:Minor}_${pkg:Arch}.txz
+
+%description
+NCCL (pronounced "Nickel") is a stand-alone library of standard collective
+communication routines for GPUs, implementing all-reduce, all-gather, reduce,
+broadcast, and reduce-scatter.
+It has been optimized to achieve high bandwidth on any platform using PCIe,
+NVLink, NVswitch, as well as networking using InfiniBand Verbs or TCP/IP
+sockets.
+
+%package devel
+Summary:        NVIDIA Collectives Communication Library (NCCL) Runtime
+Group:          Development/Libraries
+%description devel
+NCCL development files
+
+%package static
+Summary:        NVIDIA Collectives Communication Library (NCCL) Runtime
+Group:          Development/Libraries
+%description static
+NCCL static library
+
+%define debug_package %{nil}
+
+%prep
+%setup -n nccl_${nccl:Major}.${nccl:Minor}.${nccl:Patch}${nccl:Suffix}-${pkg:Revision}+cuda${cuda:Major}.${cuda:Minor}_${pkg:Arch} -q
+
+%build
+
+%install
+rm -rf $RPM_BUILD_ROOT
+install -m 755 -d $RPM_BUILD_ROOT
+install -m 755 -d $RPM_BUILD_ROOT/%{_libdir}
+install -m 755 lib/libnccl.so.${nccl:Major}.${nccl:Minor}.${nccl:Patch} $RPM_BUILD_ROOT/%{_libdir}
+ln -s libnccl.so.${nccl:Major}.${nccl:Minor}.${nccl:Patch} $RPM_BUILD_ROOT/%{_libdir}/libnccl.so.${nccl:Major}
+
+# devel
+install -m 755 -d $RPM_BUILD_ROOT/%{_includedir}
+install -m 644 include/nccl.h $RPM_BUILD_ROOT/%{_includedir}
+ln -s libnccl.so.${nccl:Major} $RPM_BUILD_ROOT/%{_libdir}/libnccl.so
+
+# static
+install -m 644 lib/libnccl_static.a $RPM_BUILD_ROOT/%{_libdir}
+
+%clean
+rm -rf $RPM_BUILD_ROOT
+
+%files devel
+%doc LICENSE.txt
+%defattr(-,root,root,-)
+%{_includedir}/nccl.h
+%{_libdir}/libnccl.so
+
+%files static
+%doc LICENSE.txt
+%defattr(-,root,root,-)
+%{_libdir}/libnccl_static.a
+
+%files
+%doc LICENSE.txt
+%defattr(-,root,root,-)
+%{_libdir}/libnccl.so.${nccl:Major}
+%{_libdir}/libnccl.so.${nccl:Major}.${nccl:Minor}.${nccl:Patch}
+
+%changelog