add curve

07/01_bandwidth/01/CMakeLists.txt

@@ -0,0 +1,23 @@
+cmake_minimum_required(VERSION 3.10)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_BUILD_TYPE Release)
+
+project(main LANGUAGES CXX)
+
+add_executable(main main.cpp)
+
+find_package(OpenMP REQUIRED)
+target_link_libraries(main PUBLIC OpenMP::OpenMP_CXX)
+
+#find_package(TBB REQUIRED)
+#target_link_libraries(main PUBLIC TBB::tbb)
+
+find_package(benchmark REQUIRED)
+target_link_libraries(main PUBLIC benchmark::benchmark)
+
+if (MSVC)
+    target_compile_options(main PUBLIC /fp:fast /arch:AVX)
+else()
+    target_compile_options(main PUBLIC -ffast-math -march=native)
+endif()

07/01_bandwidth/01/main.cpp

@@ -0,0 +1,53 @@
+#include <iostream>
+#include <vector>
+#include <cmath>
+#include <cstring>
+#include <cstdlib>
+#include <array>
+#include <benchmark/benchmark.h>
+#include <x86intrin.h>
+#include <omp.h>
+
+constexpr size_t n = 1<<26;
+
+std::vector<float> a(n);  // 256MB
+
+void BM_fill(benchmark::State &bm) {
+    for (auto _: bm) {
+        for (size_t i = 0; i < n; i++) {
+            a[i] = 1;
+        }
+    }
+}
+BENCHMARK(BM_fill);
+
+void BM_parallel_fill(benchmark::State &bm) {
+    for (auto _: bm) {
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = 1;
+        }
+    }
+}
+BENCHMARK(BM_parallel_fill);
+
+void BM_sine(benchmark::State &bm) {
+    for (auto _: bm) {
+        for (size_t i = 0; i < n; i++) {
+            a[i] = std::sin(i);
+        }
+    }
+}
+BENCHMARK(BM_sine);
+
+void BM_parallel_sine(benchmark::State &bm) {
+    for (auto _: bm) {
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = std::sin(i);
+        }
+    }
+}
+BENCHMARK(BM_parallel_sine);
+
+BENCHMARK_MAIN();

07/01_bandwidth/02/CMakeLists.txt

@@ -0,0 +1,23 @@
+cmake_minimum_required(VERSION 3.10)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_BUILD_TYPE Release)
+
+project(main LANGUAGES CXX)
+
+add_executable(main main.cpp)
+
+find_package(OpenMP REQUIRED)
+target_link_libraries(main PUBLIC OpenMP::OpenMP_CXX)
+
+#find_package(TBB REQUIRED)
+#target_link_libraries(main PUBLIC TBB::tbb)
+
+find_package(benchmark REQUIRED)
+target_link_libraries(main PUBLIC benchmark::benchmark)
+
+if (MSVC)
+    target_compile_options(main PUBLIC /fp:fast /arch:AVX)
+else()
+    target_compile_options(main PUBLIC -ffast-math -march=native)
+endif()

07/01_bandwidth/02/main.cpp

@@ -0,0 +1,36 @@
+#include <iostream>
+#include <vector>
+#include <cmath>
+#include <cstring>
+#include <cstdlib>
+#include <array>
+#include <benchmark/benchmark.h>
+#include <x86intrin.h>
+#include <omp.h>
+
+constexpr size_t n = 1<<28;
+
+std::vector<float> a(n);  // 1GB
+
+void BM_serial_add(benchmark::State &bm) {
+    for (auto _: bm) {
+        for (size_t i = 0; i < n; i++) {
+            a[i] = a[i] + 1;
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_serial_add);
+
+void BM_parallel_add(benchmark::State &bm) {
+    for (auto _: bm) {
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = a[i] + 1;
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_parallel_add);
+
+BENCHMARK_MAIN();

07/01_bandwidth/03/CMakeLists.txt

@@ -0,0 +1,23 @@
+cmake_minimum_required(VERSION 3.10)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_BUILD_TYPE Release)
+
+project(main LANGUAGES CXX)
+
+add_executable(main main.cpp)
+
+find_package(OpenMP REQUIRED)
+target_link_libraries(main PUBLIC OpenMP::OpenMP_CXX)
+
+#find_package(TBB REQUIRED)
+#target_link_libraries(main PUBLIC TBB::tbb)
+
+find_package(benchmark REQUIRED)
+target_link_libraries(main PUBLIC benchmark::benchmark)
+
+if (MSVC)
+    target_compile_options(main PUBLIC /fp:fast /arch:AVX)
+else()
+    target_compile_options(main PUBLIC -ffast-math -march=native)
+endif()

07/01_bandwidth/03/main.cpp

@@ -0,0 +1,40 @@
+#include <iostream>
+#include <vector>
+#include <cmath>
+#include <cstring>
+#include <cstdlib>
+#include <array>
+#include <benchmark/benchmark.h>
+#include <x86intrin.h>
+#include <omp.h>
+
+constexpr size_t n = 1<<28;
+
+std::vector<float> a(n);  // 1GB
+
+static float func(float x) {
+    return x * (x * x + x * 3.14f - 1 / (x + 1)) + 42 / (2.718f - x);
+}
+
+void BM_serial_func(benchmark::State &bm) {
+    for (auto _: bm) {
+        for (size_t i = 0; i < n; i++) {
+            a[i] = func(a[i]);
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_serial_func);
+
+void BM_parallel_func(benchmark::State &bm) {
+    for (auto _: bm) {
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = func(a[i]);
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_parallel_func);
+
+BENCHMARK_MAIN();

07/01_bandwidth/03/ticktock.h

@@ -0,0 +1,9 @@
+#pragma once
+
+//#include <chrono>
+//#define TICK(x) auto bench_##x = std::chrono::steady_clock::now();
+//#define TOCK(x) std::cout << #x ": " << std::chrono::duration_cast<std::chrono::duration<double>>(std::chrono::steady_clock::now() - bench_##x).count() << "s" << std::endl;
+
+#include <tbb/tick_count.h>
+#define TICK(x) auto bench_##x = tbb::tick_count::now();
+#define TOCK(x) std::cout << #x ": " << (tbb::tick_count::now() - bench_##x).seconds() << "s" << std::endl;

07/01_bandwidth/04/CMakeLists.txt

@@ -0,0 +1,23 @@
+cmake_minimum_required(VERSION 3.10)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_BUILD_TYPE Release)
+
+project(main LANGUAGES CXX)
+
+add_executable(main main.cpp)
+
+find_package(OpenMP REQUIRED)
+target_link_libraries(main PUBLIC OpenMP::OpenMP_CXX)
+
+#find_package(TBB REQUIRED)
+#target_link_libraries(main PUBLIC TBB::tbb)
+
+find_package(benchmark REQUIRED)
+target_link_libraries(main PUBLIC benchmark::benchmark)
+
+if (MSVC)
+    target_compile_options(main PUBLIC /fp:fast /arch:AVX)
+else()
+    target_compile_options(main PUBLIC -ffast-math -march=native)
+endif()

07/01_bandwidth/04/main.cpp

@@ -0,0 +1,91 @@
+#include <iostream>
+#include <vector>
+#include <cmath>
+#include <cstring>
+#include <cstdlib>
+#include <array>
+#include <benchmark/benchmark.h>
+#include <x86intrin.h>
+#include <omp.h>
+
+constexpr size_t n = 1<<28;
+
+std::vector<float> a(n);  // 1GB
+
+static float funcA(float x) {
+    return sqrtf(x) * x;
+}
+
+void BM_1funcA(benchmark::State &bm) {
+    for (auto _: bm) {
+        omp_set_num_threads(1);
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = funcA(a[i]);
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_1funcA);
+
+void BM_2funcA(benchmark::State &bm) {
+    for (auto _: bm) {
+        omp_set_num_threads(2);
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = funcA(a[i]);
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_2funcA);
+
+void BM_4funcA(benchmark::State &bm) {
+    for (auto _: bm) {
+        omp_set_num_threads(4);
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = funcA(a[i]);
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_4funcA);
+
+void BM_6funcA(benchmark::State &bm) {
+    for (auto _: bm) {
+        omp_set_num_threads(6);
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = funcA(a[i]);
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_6funcA);
+
+void BM_8funcA(benchmark::State &bm) {
+    for (auto _: bm) {
+        omp_set_num_threads(8);
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = funcA(a[i]);
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_8funcA);
+
+void BM_10funcA(benchmark::State &bm) {
+    for (auto _: bm) {
+        omp_set_num_threads(10);
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = funcA(a[i]);
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_10funcA);
+
+BENCHMARK_MAIN();

07/01_bandwidth/04/mtprint.h

@@ -0,0 +1,13 @@
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+template <class T, class ...Ts>
+static void mtprint(T &&t, Ts &&...ts) {
+    std::stringstream ss;
+    ss << std::forward<T>(t);
+    ((ss << ' ' << std::forward<Ts>(ts)), ...);
+    ss << std::endl;
+    std::cout << ss.str();
+}