Snapshot of kokkos-kernels.git from commit 6e8e97a977564673bdadc15085…

…d27a6a56b98114 From repository at [email protected]:kokkos/kokkos-kernels.git At commit: commit 6e8e97a977564673bdadc15085d27a6a56b98114 Merge: 00b1648 f81778c Author: Nathan Ellingwood <[email protected]> Date: Wed Mar 7 16:29:28 2018 -0700 Merge branch 'develop' for 2.6.00 Part of Kokkos C++ Performance Portability Programming EcoSystem 2.6
trilinos · Mar 7, 2018 · 35f8c36 · 35f8c36
1 parent dee20c5
commit 35f8c36
Show file tree

Hide file tree

Showing 185 changed files with 17,198 additions and 11,498 deletions.
diff --git a/packages/kokkos-kernels/.gitignore b/packages/kokkos-kernels/.gitignore
diff --git a/packages/kokkos-kernels/CHANGELOG.md b/packages/kokkos-kernels/CHANGELOG.md
@@ -1,5 +1,19 @@
 # Change Log
 
+## [2.6.00](https:/kokkos/kokkos-kernels/tree/2.6.00) (2018-03-07)
+[Full Changelog](https:/kokkos/kokkos-kernels/compare/2.5.00...2.6.00)
+
+**Implemented enhancements:**
+
+- Spgemm hash promotion [\#171](https:/kokkos/kokkos-kernels/issues/171)
+- Batched BLAS enhancement [\#170](https:/kokkos/kokkos-kernels/issues/170)
+
+**Fixed bugs:**
+
+- d2\_graph\_color doesn't have a default coloring mechanism [\#168](https:/kokkos/kokkos-kernels/issues/168)
+- Build error when MKL TPL is enabled [\#135](https:/kokkos/kokkos-kernels/issues/135)
+
+
 ## [2.5.00](https:/kokkos/kokkos-kernels/tree/2.5.00) (2017-12-15)
 [Full Changelog](https:/kokkos/kokkos-kernels/compare/0.10.03...2.5.00)
 

diff --git a/packages/kokkos-kernels/Makefile.kokkos-kernels b/packages/kokkos-kernels/Makefile.kokkos-kernels
@@ -389,8 +389,10 @@ endif
 KOKKOSKERNELS_INTERNAL_HEADERS = $(wildcard ${KOKKOSKERNELS_PATH}/src/impl/*.hpp)
 KOKKOSKERNELS_INTERNAL_HEADERS += $(wildcard ${KOKKOSKERNELS_PATH}/src/impl/generated_specializations_hpp/*/*eti_spec*.hpp)
 
+ifdef KOKKOSKERNELS_INTERNAL_SRC_SPARSE
 vpath %.cpp $(sort $(dir $(KOKKOSKERNELS_INTERNAL_SRC_SPARSE)))
 vpath %.cpp $(sort $(dir $(KOKKOSKERNELS_INTERNAL_SRC_BLAS)))
+endif
 
 DEPFLAGS = -M
 

diff --git a/packages/kokkos-kernels/README b/packages/kokkos-kernels/README
@@ -70,6 +70,36 @@ Building Kokkoskernels
  1. Modify example/buildlib/compileKokkosKernelsSimple.sh or
  example/buildlib/compileKokkosKernels.sh for your environment
  and run it to generate the required makefiles. 
+ - KOKKOS_DEVICES can be as below. You can remove any backend 
+ that you don't need. If cuda backend is used, CXX compiler should point to ${KOKKOS_PATH}/config/nvcc_wrapper.
+ If you enable Cuda, a host space, either OpenMP or Serial should be enabled.
+ KOKKOS_DEVICES=OpenMP,Serial,Cuda
+
+ - For the best performance give the architecture flag to proper architecture.
+ e.g. KNLs: KOKKOS_ARCHS=KNL, KOKKOS_ARCHS=HSW. 
+ If you compile for P100 GPUs with Power8 Processor, give both architectures.
+ KOKKOS_ARCHS=Pascal60,Power8
 
+ For the architecture flags, run below command.
+ %: scripts/generate_makefile.bash --help 
  2. Run "make build-test" to compile the tests.
 
+
+Using Kokkoskernels Test Drivers 
+--------------------------
+
+In perf_test there are test drivers.
+
+-- KokkosGraph_triangle.exe : Triangle counting driver. 
+-- KokkosSparse_spgemm.exe : Sparse Matrix Sparse Matrix Multiply: 
+ *****NOTE: KKMEM is outdated. Use default algorithm: KKSPGEMM = KKDEFAULT = DEFAULT****
+ Or within the code:
+ kh.create_spgemm_handle(KokkosSparse::SPGEMM_KK);
+-- KokkosSparse_spmv.exe : Sparse matvec.
+-- KokkosSparse_pcg.exe: CG method with Gauss Seidel as preconditioner.
+-- KokkosGraph_color.exe: Distance-1 Graph coloring 
+-- KokkosKernels_MatrixConverter.exe: given a matrix market format, converts it ".bin"
+ binary format for fast input output readings, which can be read by other test drivers.
+
+Please report bugs or performance issues to: https:/kokkos/kokkos-kernels/issues
+
diff --git a/packages/kokkos-kernels/example/wiki/Makefile b/packages/kokkos-kernels/example/wiki/Makefile
@@ -0,0 +1,54 @@
+KOKKOS_PATH ?= ${HOME}/Kokkos/kokkos
+KOKKOSKERNELS_PATH ?= ${HOME}/Kokkos/kokkos-kernels
+
+# Turn of ETI
+KOKKOSKERNELS_SCALARS = 
+
+SRC ?= $(wildcard *.cpp)
+HDR ?= $(wildcard *.hpp)
+
+KOKKOS_DEVICES=OpenMP
+KOKKOS_ARCH = "SNB,Kepler35"
+KOKKOS_CUDA_OPTIONS=enable_lambda
+
+default: build
+ echo "Start Build"
+
+ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
+CXX = ${KOKKOS_PATH}/config/nvcc_wrapper
+else
+CXX = g++
+endif
+
+LINK = ${CXX}
+
+CXXFLAGS = -O3 -g 
+override CXXFLAGS += -I./
+LINKFLAGS = 
+
+EXE = test.x
+DEPFLAGS = -M
+
+vpath %.cpp $(sort $(dir $(SRC)))
+
+OBJ = $(notdir $(SRC:.cpp=.o))
+LIB =
+
+include $(KOKKOS_PATH)/Makefile.kokkos
+include ${KOKKOSKERNELS_PATH}/Makefile.kokkos-kernels
+
+$(warning $(OBJ) $(EXE) $(sort $(dir $(SRC))))
+
+build: $(EXE)
+
+$(EXE): $(OBJ) $(KOKKOS_LINK_DEPENDS) $(KOKKOSKERNELS_LINK_DEPENDS)
+ $(LINK) $(KOKKOS_LDFLAGS) $(KOKKOSKERNELS_LDFLAGS) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(KOKKOS_LIBS) $(KOKKOSKERNELS_LIBS) $(LIB) -o $(EXE)
+
+clean: kokkos-clean 
+ rm -f *.o *.cuda *.host
+
+# Compilation rules
+
+%.o:%.cpp $(KOKKOS_CPP_DEPENDS) $(HDR)
+ $(CXX) $(KOKKOS_CPPFLAGS) $(KOKKOS_CXXFLAGS) $(KOKKOSKERNELS_CPPFLAGS) $(CXXFLAGS) $(EXTRA_INC) -c $< -o $(notdir $@)
+
diff --git a/packages/kokkos-kernels/example/wiki/blas/abs/Makefile b/packages/kokkos-kernels/example/wiki/blas/abs/Makefile
@@ -0,0 +1,14 @@
+MAKEFILE_PATH := $(subst Makefile,,$(abspath $(lastword $(MAKEFILE_LIST))))
+
+ifndef KOKKOSKERNELS_PATH
+ KOKKOSKERNELS_PATH = $(MAKEFILE_PATH)../../../..
+endif
+ifndef KOKKOS_PATH
+ KOKKOS_PATH = $(KOKKOSKERNELS_PATH)/../kokkos
+endif
+
+SRC = $(wildcard $(MAKEFILE_PATH)*.cpp)
+HDR = $(wildcard $(MAKEFILE_PATH)*.hpp)
+
+include $(KOKKOSKERNELS_PATH)/example/wiki/Makefile
+
diff --git a/packages/kokkos-kernels/example/wiki/blas/abs/abs.cpp b/packages/kokkos-kernels/example/wiki/blas/abs/abs.cpp
@@ -0,0 +1,24 @@
+#include<Kokkos_Core.hpp>
+#include<KokkosBlas1_abs.hpp>
+
+int main(int argc, char* argv[]) {
+ Kokkos::initialize();
+
+ int N = atoi(argv[1]);
+
+ Kokkos::View<double*> x("X",N);
+ Kokkos::View<double*> y("Y",N);
+ Kokkos::deep_copy(x,-1.0);
+
+ KokkosBlas::abs(y,x);
+
+ double sum = 0.0;
+ Kokkos::parallel_reduce("CheckValue", N, KOKKOS_LAMBDA (const int& i, double& lsum) {
+ lsum += y(i);
+ },sum);
+
+ printf("Sum: %lf Expected: %lf Diff: %e\n",sum,1.0*N,sum-1.0*N);
+
+ Kokkos::finalize();
+}
+
diff --git a/packages/kokkos-kernels/master_history.txt b/packages/kokkos-kernels/master_history.txt
@@ -1,4 +1,5 @@
 tag: 0.10.00 date: 08/22/1017 master: xxxxxxxx develop: e323bcb0
 tag: 0.10.03 date: 09/11/2017 master: e323bcb0 develop: 4cb87390
 tag: 2.5.00 date: 12/15/2017 master: e4c645e9 develop: 04d58766
+tag: 2.6.00 date: 03/07/2018 master: 00b16484 develop: f81778ce
 
diff --git a/packages/kokkos-kernels/perf_test/Makefile b/packages/kokkos-kernels/perf_test/Makefile
@@ -77,11 +77,31 @@ EXECUTABLES += $(wildcard ${KOKKOSKERNELS_SRC_PATH}/perf_test/sparse/*cpp)
 EXECUTABLES += $(wildcard ${KOKKOSKERNELS_SRC_PATH}/perf_test/graph/*cpp) 
 EXECUTABLES += $(wildcard ${KOKKOSKERNELS_SRC_PATH}/perf_test/common/*cpp) 
 
+
+
+#=======================================================================
+#===================== BATCHED BLAS/LAPACK PERF ========================
+#=======================================================================
+
+BATCHEDINC = -I$(KOKKOSKERNELS_SRC_PATH)/perf_test/batched
+INC += $(BATCHEDINC)
+
+BATCHED_HEADERS = $(wildcard $(KOKKOSKERNELS_SRC_PATH)/perf_test/batched/*.hpp)
+TEST_HEADERS += $(BATCHED_HEADERS)
+
+BATCHED_EXECUTABLES = $(wildcard ${KOKKOSKERNELS_SRC_PATH}/perf_test/batched/*cpp)
+EXECUTABLES += $(BATCHED_EXECUTABLES)
+
+#=======================================================================
+#===================== TEST TARGETS ====================================
+#=======================================================================
+
 vpath %.cpp $(sort $(dir $(EXECUTABLES)))
 OBJS = $(notdir $(EXECUTABLES:.cpp=.o))
 TEST_TARGETS = $(notdir $(EXECUTABLES:.cpp=.exe))
-
 #TEST_TARGETS = $(patsubst %.cpp, %.exe, $(EXECUTABLES))
+
+
 #=======================================================================
 #=================== Make Rules ========================================
 #=======================================================================
@@ -93,7 +113,7 @@ clean: kokkos-clean kokkoskernels-clean
 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 $@
 

diff --git a/packages/kokkos-kernels/perf_test/batched/KokkosBatched_Test_BlockCrs_Cuda.cpp b/packages/kokkos-kernels/perf_test/batched/KokkosBatched_Test_BlockCrs_Cuda.cpp
@@ -0,0 +1,121 @@
+/* Implementation for testing KokkosKernels on BCRS operations
+ - block-tridiagonal factorization
+ - block-tridiagonal solve
+ - bcrs matvec
+
+ StructuredBlock represents a 3D mesh having ni, nj, nk cells in each
+ dimension. Variable ordering is such that the k index is the fastest and the
+ i index is slowest. Smoothing lines are built in the k direction.
+ BlockCrsMatrix is a simple block CRS data structure.
+ BlockTridiagMatrices holds the block tridiagonal matrices.
+
+ An example run is
+ ./driver -ni 32 -nj 32 -nk 128 -bs 5 -c
+
+ This runs a sequence of unit tests, then runs a problem having a 32x32x128
+ structured block with the lines oriented along the third dimension (line
+ length = 128). The block size is 5. -c adds a somewhat expensive check of the
+ answer. It's good to run with -c once in a while, but the cheap unit tests
+ that always run before the big problem already provide good coverage.
+*/
+
+#include "Kokkos_Core.hpp"
+#include "impl/Kokkos_Timer.hpp"
+
+#if defined(KOKKOS_ENABLE_CUDA) 
+#define __KOKKOSBATCHED_TEST_ENABLE_CUDA__
+
+#include "KokkosBatched_Util.hpp"
+
+#define KOKKOSBATCHED_USE_UNBLOCKED_ALGO 1
+//#define KOKKOSBATCHED_USE_BLOCKED_ALGO 1
+
+#if defined (KOKKOSBATCHED_USE_UNBLOCKED_ALGO)
+typedef KokkosBatched::Experimental::Algo::LU::Unblocked AlgoLU;
+typedef KokkosBatched::Experimental::Algo::Trsm::Unblocked AlgoTrsm;
+typedef KokkosBatched::Experimental::Algo::Gemm::Unblocked AlgoGemm;
+
+typedef KokkosBatched::Experimental::Algo::Trsv::Unblocked AlgoTrsv;
+typedef KokkosBatched::Experimental::Algo::Gemv::Unblocked AlgoGemv;
+#endif
+#if defined (KOKKOSBATCHED_USE_BLOCKED_ALGO)
+typedef KokkosBatched::Experimental::Algo::LU::Blocked AlgoLU;
+typedef KokkosBatched::Experimental::Algo::Trsm::Blocked AlgoTrsm;
+typedef KokkosBatched::Experimental::Algo::Gemm::Blocked AlgoGemm;
+
+typedef KokkosBatched::Experimental::Algo::Trsv::Blocked AlgoTrsv;
+typedef KokkosBatched::Experimental::Algo::Gemv::Blocked AlgoGemv;
+#endif
+
+#include "KokkosBatched_Test_BlockCrs.hpp"
+
+using namespace KokkosBatched;
+
+int main (int argc, char *argv[]) {
+ Kokkos::initialize(argc, argv); 
+
+ typedef Kokkos::DefaultHostExecutionSpace HostSpaceType;
+ typedef Kokkos::DefaultExecutionSpace DeviceSpaceType;
+
+ const bool detail = false;
+
+ Kokkos::print_configuration(std::cout, detail);
+
+ enum : int { VectorLength = DefaultVectorLength<Test::scalar_type,typename DeviceSpaceType::memory_space>::value,
+ RangeTagOper = 0,
+ TeamTagOper = 1 };
+
+ // Unit tests
+ bool profile = false;
+ for (int i=1;i<argc;++i) {
+ const std::string& token = argv[i];
+ if (strncmp(token.c_str(), "-profile", 8) == 0) profile = true;
+ }
+
+
+ if (!profile) {
+ // std::cout << " Unit Test::Range :: Begin\n";
+ // {
+ // Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,RangeTagOper>( 3, 4, 2, 25, 2);
+ // Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,RangeTagOper>(44, 63, 15, 4, 1);
+ // Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,RangeTagOper>( 2, 2, 15, 3, 3);
+ // Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,RangeTagOper>( 1, 1, 2, 63, 8);
+
+ // for (int nrhs=1;nrhs<=33;++nrhs)
+ // Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,RangeTagOper>(2, 2, 15, 3, nrhs);
+ // }
+ // std::cout << " Unit Test::Range :: End\n";
+
+ std::cout << " Unit Test::Team :: Begin\n";
+ {
+ Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,TeamTagOper>( 3, 4, 2, 25, 2);
+ Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,TeamTagOper>(44, 63, 15, 4, 1);
+ Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,TeamTagOper>( 2, 2, 15, 3, 3);
+ Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,TeamTagOper>( 1, 1, 2, 63, 8);
+
+ for (int nrhs=1;nrhs<=33;++nrhs)
+ Test::run<DeviceSpaceType,Test::scalar_type,VectorLength,TeamTagOper>(2, 2, 15, 3, nrhs);
+ }
+ std::cout << " Unit Test::Team :: End\n";
+ }
+
+ // Performance tests
+ std::cout << " Perf Test:: Begin\n";
+ {
+ const Test::Input<DeviceSpaceType> input(argc, argv); 
+ Test::run<DeviceSpaceType,Test::scalar_type,VectorLength>(input);
+ } 
+ std::cout << " Perf Test:: End\n";
+
+ Kokkos::finalize();
+
+ return 0;
+}
+#else
+
+int main(int argc, char *argv[]) {
+ std::cout << "Kokkos::Cuda is not enabled\n";
+ return -1;
+}
+
+#endif