Merge pull request #86 from lanl/benchmark

Add google benchmark

.gitignore

@@ -1 +1,4 @@
 install-*
+benchmark/benchmark*
+build-matar-*
+install/*

benchmark/CMakeLists.txt

@@ -0,0 +1,54 @@
+cmake_minimum_required(VERSION 3.1.3)
+
+project (matarbenchmark)
+
+set(benchmark_DIR "benchmark/build")
+
+find_package(benchmark REQUIRED)
+find_package(Matar REQUIRED)
+
+
+
+
+
+if (NOT KOKKOS)
+  add_executable(BM_Carray src/CArray_benchmark.cpp)
+  target_link_libraries(BM_Carray matar benchmark::benchmark)
+endif()
+
+if (KOKKOS)
+  find_package(Kokkos REQUIRED) #new
+
+  add_definitions(-DHAVE_KOKKOS=1)
+
+  add_executable(BM_CArray src/CArray_benchmark.cpp)
+  target_link_libraries(BM_CArray matar Kokkos::kokkos benchmark::benchmark)
+
+  add_executable(BM_CArrayDevice src/CArrayDevice_benchmark.cpp)
+  target_link_libraries(BM_CArrayDevice matar Kokkos::kokkos benchmark::benchmark)
+
+  if (CUDA)
+    add_definitions(-DHAVE_CUDA=1)
+  elseif (HIP)
+    add_definitions(-DHAVE_HIP=1)
+  elseif (OPENMP)
+    add_definitions(-DHAVE_OPENMP=1)
+  elseif (THREADS)
+    add_definitions(-DHAVE_THREADS=1)
+  endif()
+endif()
+
+# find_package(Kokkos REQUIRED) #new
+
+# set(This matar_benchmark)
+
+# set(Sources
+#     src/serial_types_benchmark.cpp
+# )
+
+
+
+
+# if (KOKKOKS)
+#   target_link_libraries(${This} matar Kokkos::kokkos benchmark::benchmark)
+# endif ()

benchmark/src/CArrayDevice_benchmark.cpp

@@ -0,0 +1,112 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string>
+#include <assert.h>
+#include <memory> // for shared_ptr
+#include <benchmark/benchmark.h>
+#include "matar.h"
+
+using namespace mtr; // matar namespace
+
+// ------- vector vector multiply ------------- //
+static void BM_CArrayDevice_1d_multiply(benchmark::State& state) 
+{
+
+    int size = state.range(0);
+
+    CArrayDevice<double> A(size);
+    CArrayDevice<double> B(size);
+    CArrayDevice<double> C(size);
+
+    FOR_ALL(i, 0, size, {
+        A(i) = (double)i+1.0;
+        B(i) = (double)i+2.0;
+    });
+    // Begin benchmarked section
+    for (auto _ : state){
+        FOR_ALL(i, 0, size, {
+            C(i) = A(i)*B(i);
+        });
+    } // end benchmarked section
+
+    // Kokkos::finalize();
+
+}
+BENCHMARK(BM_CArrayDevice_1d_multiply)
+->Unit(benchmark::kMillisecond)
+->Name("Benchmark Multiplying 2 1D CArrayDevice of size ")
+->RangeMultiplier(2)->Range(1<<12, 1<<20);
+
+// ------- vector vector dot product ------------- //
+static void BM_CArrayDevice_vec_vec_dot(benchmark::State& state) 
+{
+    int size = state.range(0);
+
+    CArrayDevice<double> A(size);
+    CArrayDevice<double> B(size);
+    double C = 0.0;
+
+    FOR_ALL(i, 0, size, {
+        A(i) = (double)i+1.0;
+        B(i) = (double)i+2.0;
+    });
+
+
+    // Begin benchmarked section
+    for (auto _ : state){
+
+        double loc_sum = 0;
+        double C  = 0;
+        REDUCE_SUM(i, 0, size,
+            loc_sum, {
+            loc_sum += A(i)*B(i);
+        }, C);
+    } // end benchmarked section
+}
+BENCHMARK(BM_CArrayDevice_vec_vec_dot)
+->Unit(benchmark::kMillisecond)
+->Name("Benchmark dot product of 2 1D CArrayDevice of size ")
+->RangeMultiplier(2)->Range(1<<12, 1<<20);
+
+
+// ------- matrix matrix multiply ------------- //
+static void BM_CArrayDevice_mat_mat_multiply(benchmark::State& state) 
+{
+    int size = state.range(0);
+
+    CArrayDevice<double> A(size, size);
+    CArrayDevice<double> B(size, size);
+    CArrayDevice<double> C(size, size);
+
+    FOR_ALL(i, 0, size,
+            j, 0, size, {
+        A(i,j) = (double)i+(double)j+1.0;
+        B(i,j) = (double)i+(double)j+2.0;
+        C(i,j) = 0.0;
+    });
+
+    // Begin benchmarked section
+    for (auto _ : state){
+
+        FOR_ALL(i, 0, size,
+                j, 0, size, {
+            for(int k = 0; k < size; k++){
+                C(i,k) += A(i,j)*B(j,k);
+            }
+        });
+    } // end benchmarked section
+}
+BENCHMARK(BM_CArrayDevice_mat_mat_multiply)
+->Unit(benchmark::kMillisecond)
+->Name("Benchmark matrix-matrix multiply of CArrayDevice of size ")
+->RangeMultiplier(2)->Range(1<<3, 1<<10);
+
+
+// Run Benchmarks
+int main(int argc, char** argv)
+{
+   Kokkos::initialize();
+   ::benchmark::Initialize(&argc, argv);
+   ::benchmark::RunSpecifiedBenchmarks();
+   Kokkos::finalize();
+}

benchmark/src/CArray_benchmark.cpp

@@ -0,0 +1,106 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string>
+#include <assert.h>
+#include <memory> // for shared_ptr
+#include <benchmark/benchmark.h>
+#include "matar.h"
+
+using namespace mtr; // matar namespace
+
+// ------- vector vector multiply ------------- //
+static void BM_CArray_1d_multiply(benchmark::State& state) 
+{
+    // const int size = 4000;
+
+    int size = state.range(0);
+
+    CArray<double> A(size);
+    CArray<double> B(size);
+    CArray<double> C(size);
+
+    for(int i=0; i<size; i++){
+        A(i) = (double)i+1.0;
+        B(i) = (double)i+2.0;
+    }
+
+    // Begin benchmarked section
+    for (auto _ : state){
+        for(int i=0; i<size; i++){
+            C(i) = A(i)*B(i);
+        }
+    } // end benchmarked section
+}
+BENCHMARK(BM_CArray_1d_multiply)
+->Unit(benchmark::kMillisecond)
+->Name("Benchmark Multiplying 2 1D CArrays of size ")
+->RangeMultiplier(2)->Range(1<<12, 1<<20);
+
+
+// ------- vector vector dot product ------------- //
+static void BM_Carray_vec_vec_dot(benchmark::State& state) 
+{
+    int size = state.range(0);
+
+    // Begin benchmarked section
+    for (auto _ : state){
+        CArray<double> A(size);
+        CArray<double> B(size);
+        double C = 0.0;
+
+        for(int i = 0; i < size; i++){
+            A(i) = (double)i+1.0;
+            B(i) = (double)i+2.0;
+        }
+
+        for(int i = 0; i < size; i++){
+            C += A(i)*B(i);
+        }
+    } // end benchmarked section
+}
+BENCHMARK(BM_CArray_1d_multiply)
+->Unit(benchmark::kMillisecond)
+->Name("Benchmark dot product of 2 1D CArrays of size ")
+->RangeMultiplier(2)->Range(1<<12, 1<<20);
+
+// ------- matrix matrix multiply ------------- //
+static void BM_CArray_mat_mat_multiply(benchmark::State& state) 
+{
+    int size = state.range(0);
+
+    // Begin benchmarked section
+    for (auto _ : state){
+        CArray<double> A(size, size);
+        CArray<double> B(size, size);
+        CArray<double> C(size, size);
+
+        for(int i = 0; i < size; i++){
+            for(int j = 0; j < size; j++){
+                A(i,j) = (double)i+(double)j+1.0;
+                B(i,j) = (double)i+(double)j+2.0;
+                C(i,j) = 0.0;
+            }
+        }
+
+        for(int i = 0; i < size; i++){
+            for(int j = 0; j < size; j++){
+                for(int k = 0; k < size; k++){
+                    C(i,k) += A(i,j)*B(j,k);
+                }
+            }
+        }
+    } // end benchmarked section
+}
+BENCHMARK(BM_CArray_mat_mat_multiply)
+->Unit(benchmark::kMillisecond)
+->Name("Benchmark matrix-matrix multiply of CArrays of size ")
+->RangeMultiplier(2)->Range(1<<3, 1<<10);
+
+// Run benchmarks
+int main(int argc, char** argv)
+{
+   Kokkos::initialize();
+   ::benchmark::Initialize(&argc, argv);
+   ::benchmark::RunSpecifiedBenchmarks();
+   Kokkos::finalize();
+}

scripts/build-matar.sh

@@ -23,6 +23,7 @@ show_help() {
     echo " "
     echo "          examples                    builds examples"
     echo "          test                        builds tests"
+    echo "          benchmark                   builds benchmarks for MATAR"
     echo " "
     echo "      --kokkos_build_type             The desired kokkos parallel backend to use. The default is 'serial'"
     echo " "
@@ -52,7 +53,7 @@ build_cores="1"
 
 # Define arrays of valid options
 valid_build_action=("full-app" "set-env" "install-matar" "install-kokkos" "matar")
-valid_execution=("examples" "test")
+valid_execution=("examples" "test" "benchmark")
 valid_kokkos_build_types=("none" "serial" "openmp" "pthreads" "cuda" "hip")
 valid_machines=("darwin" "chicoma" "linux" "mac")
 

scripts/cmake_build_benchmark.sh

@@ -0,0 +1,42 @@
+#!/bin/bash -e
+
+kokkos_build_type="${1}"
+
+if [ ! -d "${BENCHMARK_SOURCE_DIR}/benchmark" ]
+then
+  echo "Missing googlebenchmark for benchmarking, downloading and installing...."
+  git clone https://github.com/google/benchmark.git ${BENCHMARK_SOURCE_DIR}/benchmark
+  cd ${BENCHMARK_SOURCE_DIR}/benchmark
+  cmake -E make_directory "build"
+  cmake -E chdir "build" cmake -DBENCHMARK_DOWNLOAD_DEPENDENCIES=on -DCMAKE_BUILD_TYPE=Release ../
+  cmake --build  "build" --config Release -j${MATAR_BUILD_CORES}
+  # Test install
+  cmake -E chdir "build" ctest --build-config Release
+fi
+
+cmake_options=(
+    -D CMAKE_PREFIX_PATH="${MATAR_INSTALL_DIR};${KOKKOS_INSTALL_DIR};${BENCHMARK_INSTALL_DIR}"
+    -D BENCHMARK_DOWNLOAD_DEPENDENCIES=on
+    -DCMAKE_BUILD_TYPE=Release
+)
+
+
+
+if [ "$kokkos_build_type" = "none" ]; then
+    cmake_options+=(
+        -D KOKKOS=OFF
+    )
+else
+    cmake_options+=(
+        -D KOKKOS=ON
+    )
+fi
+# Print CMake options for reference
+echo "CMake Options: ${cmake_options[@]}"
+
+cmake "${cmake_options[@]}" -B "${BENCHMARK_BUILD_DIR}" -S "${BENCHMARK_SOURCE_DIR}"
+
+# Build benchmark
+make -C "${BENCHMARK_BUILD_DIR}" -j${MATAR_BUILD_CORES}
+
+cd $basedir

scripts/setup-env.sh

@@ -23,6 +23,10 @@ export EXAMPLE_BUILD_DIR=${builddir}
 export TEST_SOURCE_DIR=${basedir}/test
 export TEST_BUILD_DIR=${builddir}
 
+export BENCHMARK_SOURCE_DIR=${basedir}/benchmark
+export BENCHMARK_INSTALL_DIR=${basedir}/benchmark/build
+export BENCHMARK_BUILD_DIR=${builddir}
+
 export KOKKOS_SOURCE_DIR=${basedir}/src/Kokkos/kokkos
 export KOKKOS_BUILD_DIR=${builddir}/kokkos
 export KOKKOS_INSTALL_DIR=${installdir}/kokkos

src/kokkos_types.h

@@ -7523,7 +7523,7 @@ DynamicRaggedRightArrayKokkos<T,Layout,ExecSpace,MemoryTraits>::DynamicRaggedRig
     stride_ = SArray1D(strides_tag_string, dim1_);
     #ifdef HAVE_CLASS_LAMBDA
     Kokkos::parallel_for("StridesInit", dim1_, KOKKOS_CLASS_LAMBDA(const int i) {
-      strides_((i) = 0;
+      strides_(i) = 0;
     });
     #else
     set_strides_functor execution_functor(0, stride_);
@@ -7700,7 +7700,7 @@ DynamicRaggedDownArrayKokkos<T,Layout,ExecSpace,MemoryTraits>::DynamicRaggedDown
     stride_ = SArray1D(strides_tag_string, dim2_);
     #ifdef HAVE_CLASS_LAMBDA
     Kokkos::parallel_for("StridesInit", dim2_, KOKKOS_CLASS_LAMBDA(const int i) {
-      strides_((i) = 0;
+      strides_(i) = 0;
     });
     #else
     set_strides_functor execution_functor(0, stride_);