Merge pull request #555 from ndellingwood/sptrsv-pr-2

sptrsv: Add tplchain and cusparse
kokkos · Dec 20, 2019 · 6684d19 · 6684d19
2 parents d29beba + bfac0c0
commit 6684d19
Show file tree

Hide file tree

Showing 15 changed files with 3,402 additions and 353 deletions.
diff --git a/example/buildlib/compileKokkosKernelsSimple.sh b/example/buildlib/compileKokkosKernelsSimple.sh
@@ -4,10 +4,10 @@ KOKKOSKERNELS_LAYOUTS=LayoutLeft #the layout types to instantiate.
 KOKKOSKERNELS_ORDINALS=int #ordinal types to instantiate
 KOKKOSKERNELS_OFFSETS=int #offset types to instantiate
 KOKKOSKERNELS_PATH=../.. #path to kokkos-kernels top directory.
-CXX=icpc #${KOKKOS_PATH}/config/nvcc_wrapper #icpc #
+CXX=${KOKKOS_PATH}/bin/nvcc_wrapper #icpc #
 KOKKOSKERNELS_OPTIONS=eti-only #options for kokkoskernels  
-KOKKOS_DEVICES=OpenMP # other devices Cuda,Serial ..
-KOKKOS_ARCHS=KNL
+KOKKOS_DEVICES=Cuda # other devices Cuda,Serial ..
+KOKKOS_ARCHS=SKX,Volta70
 CXXFLAGS="-Wall -pedantic -Werror -O3 -g -Wshadow -Wsign-compare -Wtype-limits -Wuninitialized"
 
 ../../scripts/generate_makefile.bash --kokkoskernels-path=${KOKKOSKERNELS_PATH} --with-scalars=${KOKKOSKERNELS_SCALARS} --with-ordinals=${KOKKOSKERNELS_ORDINALS} --with-offsets=${KOKKOSKERNELS_OFFSETS} --kokkos-path=${KOKKOS_PATH} --with-devices=${KOKKOS_DEVICES} --arch=${KOKKOS_ARCHS} --compiler=${CXX} --with-options=${KOKKOSKERNELS_OPTIONS}  --cxxflags="${CXXFLAGS}"
diff --git a/perf_test/Makefile b/perf_test/Makefile
@@ -119,7 +119,7 @@ default: $(TEST_TARGETS)
 build: $(TEST_TARGETS)
 
 %.exe:%.o $(KOKKOS_LINK_DEPENDS) $(KOKKOSKERNELS_LINK_DEPENDS) $(TEST_HEADERS)
-	$(LINK) $(EXTRA_PATH) $< $(KOKKOSKERNELS_LDFLAGS) $(KOKKOSKERNELS_LIBS) $(KOKKOS_LIBS) $(LIB) $(KOKKOS_LDFLAGS) $(LDFLAGS) -o $@
+	$(LINK) $(EXTRA_PATH) $< $(KOKKOSKERNELS_LDFLAGS) $(KOKKOSKERNELS_LIBS) $(KOKKOS_LIBS) $(LIB) $(KOKKOS_LDFLAGS) $(LDFLAGS) $(LDLIB) -o $@
 
 %.o:%.cpp $(KOKKOS_CPP_DEPENDS) $(KOKKOSKERNELS_CPP_DEPENDS) $(TEST_HEADERS)
 	$(CXX) $(KOKKOS_CPPFLAGS) $(KOKKOSKERNELS_CPPFLAGS) $(KOKKOS_CXXFLAGS) $(INC) $(CXXFLAGS) $(EXTRA_INC) -I. -c $< -o $(notdir $@)

diff --git a/perf_test/sparse/KokkosSparse_spiluk.cpp b/perf_test/sparse/KokkosSparse_spiluk.cpp
@@ -90,7 +90,7 @@ using namespace KokkosKernels::Experimental;
 enum {DEFAULT, CUSPARSE, LVLSCHED_RP, LVLSCHED_TP1/*, LVLSCHED_TP2*/};
 
 template<typename Scalar>
-int test_spiluk_perf(std::vector<int> tests, std::string afilename, int k, int team_size, int vector_length, /*int idx_offset,*/ int loop) {
+int test_spiluk_perf(std::vector<int> tests, std::string afilename, int K, int team_size, int vector_length, /*int idx_offset,*/ int loop) {
 
   typedef Scalar scalar_t;
   typedef default_lno_t lno_t;
@@ -119,12 +119,12 @@ int test_spiluk_perf(std::vector<int> tests, std::string afilename, int k, int t
 
   std::cout << "\n\n" << std::endl;
   if (!afilename.empty()) {
-    std::cout << "ILU(k) Begin: Read matrix filename " << afilename << std::endl;
+    std::cout << "ILU(K) Begin: Read matrix filename " << afilename << std::endl;
     crsmat_t A    = KokkosKernels::Impl::read_kokkos_crst_matrix<crsmat_t>(afilename.c_str()); //in_matrix
     graph_t graph = A.graph; // in_graph
     const size_type nrows = graph.numRows();
     const int       nnz   = A.nnz();
-    const typename KernelHandle::const_nnz_lno_t fill_lev = lno_t(k) ;
+    const typename KernelHandle::const_nnz_lno_t fill_lev = lno_t(K) ;
 
 #ifdef KOKKOSKERNELS_ENABLE_TPL_CUSPARSE
     //cuSPARSE requires lno_t = size_type = int. For both, int is always used (if enabled)