[build] Add cub library for cuda projects (#2819)

not needed now but will be in future.

.gitignore

@@ -144,3 +144,6 @@ GSYMS
 /tools/mmseg-1.3.0.tar.gz
 /tools/mmseg-1.3.0/
 /kaldiwin_vs*
+/tools/cub-1.8.0.zip
+/tools/cub-1.8.0/
+/tools/cub

src/configure

@@ -863,6 +863,7 @@ android=false
 MATHLIB='ATLAS'
 ATLASROOT=`rel2abs ../tools/ATLAS_headers/`
 FSTROOT=`rel2abs ../tools/openfst`
+CUBROOT=`rel2abs ../tools/cub`
 
 # Save the command line to include in kaldi.mk
 cmd_line="$0 $@"
@@ -955,6 +956,9 @@ do
   --fst-root=*)
     FSTROOT=`read_dirname $1`;
     shift ;;
+  --cub-root=*)
+    CUBROOT=`read_dirname $1`;
+    shift ;;
   --clapack-root=*)
     CLAPACKROOT=`read_dirname $1`;
     shift ;;
@@ -1092,6 +1096,13 @@ if $double_precision; then
 else
   echo "DOUBLE_PRECISION = 0" >> kaldi.mk
 fi
+$use_cuda && echo "Checking cub library in $CUBROOT ..."
+if [[ $use_cuda && ! -f $CUBROOT/cub/cub.cuh ]]; then
+  failure "Could not find file $CUBROOT/cub/cub.cuh:
+  you may not have installed cub. See ../tools/INSTALL"
+else
+  echo "CUBROOT = $CUBROOT" >> kaldi.mk
+fi
 echo "Checking OpenFst library in $FSTROOT ..."
 if [ ! -f $FSTROOT/include/fst/fst.h  ]; then
   failure "Could not find file $FSTROOT/include/fst/fst.h:

src/makefiles/cuda_32bit.mk

@@ -5,10 +5,11 @@ ifndef CUDATKDIR
 $(error CUDATKDIR not defined.)
 endif
 
-CUDA_INCLUDE= -I$(CUDATKDIR)/include
-CUDA_FLAGS = -g -Xcompiler -fPIC --verbose --machine 32 -DHAVE_CUDA \
+CUDA_INCLUDE= -I$(CUDATKDIR)/include -I$(CUBROOT)
+CUDA_FLAGS = -Xcompiler "-fPIC -pthread -isystem $(OPENFSTINC)" --verbose --machine 32 -DHAVE_CUDA \
              -ccbin $(CXX) -DKALDI_DOUBLEPRECISION=$(DOUBLE_PRECISION) \
-             -DCUDA_API_PER_THREAD_DEFAULT_STREAM
+             -std=c++11 -DCUDA_API_PER_THREAD_DEFAULT_STREAM
+
 CXXFLAGS += -DHAVE_CUDA -I$(CUDATKDIR)/include
-LDFLAGS += -L$(CUDATKDIR)/lib -Wl,-rpath=$(CUDATKDIR)/lib
-LDLIBS += -lcublas -lcusparse -lcudart -lcurand #LDLIBS : The libs are loaded later than static libs in implicit rule
+CUDA_LDFLAGS += -L$(CUDATKDIR)/lib -Wl,-rpath,$(CUDATKDIR)/lib
+CUDA_LDLIBS += -lcuda -lcublas -lcusparse -lcudart -lcurand -lnvToolsExt #LDLIBS : The libs are loaded later than static libs in implicit rule

src/makefiles/cuda_64bit.mk

@@ -5,10 +5,11 @@ ifndef CUDATKDIR
 $(error CUDATKDIR not defined.)
 endif
 
-CUDA_INCLUDE= -I$(CUDATKDIR)/include
-CUDA_FLAGS = -g -Xcompiler -fPIC --verbose --machine 64 -DHAVE_CUDA \
+CUDA_INCLUDE= -I$(CUDATKDIR)/include -I$(CUBROOT)
+CUDA_FLAGS = -Xcompiler "-fPIC -pthread -isystem $(OPENFSTINC)" --verbose --machine 64 -DHAVE_CUDA \
              -ccbin $(CXX) -DKALDI_DOUBLEPRECISION=$(DOUBLE_PRECISION) \
-             -DCUDA_API_PER_THREAD_DEFAULT_STREAM
+             -std=c++11 -DCUDA_API_PER_THREAD_DEFAULT_STREAM
+
 CXXFLAGS += -DHAVE_CUDA -I$(CUDATKDIR)/include
 CUDA_LDFLAGS += -L$(CUDATKDIR)/lib64 -Wl,-rpath,$(CUDATKDIR)/lib64
-CUDA_LDLIBS += -lcublas -lcusparse -lcudart -lcurand #LDLIBS : The libs are loaded later than static libs in implicit rule
+CUDA_LDLIBS += -lcuda -lcublas -lcusparse -lcudart -lcurand -lnvToolsExt #LDLIBS : The libs are loaded later than static libs in implicit rule

tools/Makefile

@@ -8,6 +8,7 @@ CC = gcc         # used for sph2pipe
 # Note: OpenFst requires a relatively recent C++ compiler with C++11 support,
 # e.g. g++ >= 4.7, Apple clang >= 5.0 or LLVM clang >= 3.3.
 OPENFST_VERSION ?= 1.6.7
+CUB_VERSION ?= 1.8.0
 
 # Default features configured for OpenFST; can be overridden in the make command line.
 OPENFST_CONFIGURE ?= --enable-static --enable-shared --enable-far --enable-ngram-fsts
@@ -18,7 +19,7 @@ ifeq ("$(shell expr $(OPENFST_VER_NUM) \< 10600)","1")
             Supported versions: >= 1.6.0)
 endif
 
-all: check_required_programs sph2pipe sclite openfst
+all: check_required_programs sph2pipe sclite openfst cub
 	@echo -e "\n\n"
 	@echo "Warning: IRSTLM is not installed by default anymore. If you need IRSTLM"
 	@echo "Warning: use the script extras/install_irstlm.sh"
@@ -49,6 +50,9 @@ distclean:
 	rm -rf openfst-$(OPENFST_VERSION).tar.gz
 	rm -f openfst
 	rm -rf libsndfile-1.0.25{,.tar.gz} BeamformIt-3.51{,.tgz}
+	rm -f cub-$(CUB_VERSION).zip
+	rm -rf cub-$(CUB_VERSION)
+	rm -f cub
 
 
 .PHONY: openfst # so target will be made even though "openfst" exists.
@@ -119,7 +123,7 @@ sph2pipe_v2.5/sph2pipe: | sph2pipe_v2.5
 	$(CC) -o sph2pipe  *.c -lm
 
 sph2pipe_v2.5: sph2pipe_v2.5.tar.gz
-	tar xzf sph2pipe_v2.5.tar.gz
+	tar --no-same-owner -xzf sph2pipe_v2.5.tar.gz
 
 sph2pipe_v2.5.tar.gz:
 	wget -T 10 -t 3 http://www.openslr.org/resources/3/sph2pipe_v2.5.tar.gz || \
@@ -149,3 +153,11 @@ openblas_compiled:
 	cd OpenBLAS; sed 's:# FCOMMON_OPT = -frecursive:FCOMMON_OPT = -frecursive:' < Makefile.rule >tmp && mv tmp Makefile.rule
 	# $(MAKE) PREFIX=`pwd`/OpenBLAS/install FC=gfortran $(fortran_opt) DEBUG=1 USE_THREAD=1 NUM_THREADS=64 -C OpenBLAS all install
 	$(MAKE) PREFIX=`pwd`/OpenBLAS/install FC=gfortran $(fortran_opt) DEBUG=1 USE_THREAD=0 -C OpenBLAS all install
+
+
+.PHONY: cub
+cub:
+	wget -T 10 -t 3 -O cub-$(CUB_VERSION).zip https://github.com/NVlabs/cub/archive/$(CUB_VERSION).zip
+	unzip -oq cub-$(CUB_VERSION).zip
+	rm -f cub
+	ln -s cub-$(CUB_VERSION) cub

tools/extras/travis_script.sh

@@ -69,6 +69,7 @@ echo "Building tools..." [Time: $(date)]
 runvx cd tools
 runvx make -j$MAXPAR openfst "$CCC" CXXFLAGS="$CF" \
       OPENFST_CONFIGURE="--disable-static --enable-shared --disable-bin --disable-dependency-tracking"
+runvx make -j$MAXPAR cub "$CCC" CXXFLAGS="$CF"
 cd ..
 
 runvx cd src