add jaocib

.gitignore

@@ -3,3 +3,4 @@ a.out
 build
 .~*
 GNUmakefile
+r000*/

07/01_bandwidth/07/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/07/main.cpp

@@ -0,0 +1,26 @@
+#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_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/07/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();
+}

07/01_bandwidth/07/pod.h

@@ -0,0 +1,66 @@
+#pragma once
+
+#include <new>
+#include <utility>
+
+template <class T>
+struct pod {
+private:
+    T m_t;
+public:
+    pod() {}
+
+    pod(pod &&p) : m_t(std::move(p.m_t)) {}
+
+    pod(pod const &p) : m_t(p.m_t) {}
+
+    pod &operator=(pod &&p) {
+        m_t = std::move(p.m_t);
+        return *this;
+    }
+
+    pod &operator=(pod const &p) {
+        m_t = p.m_t;
+        return *this;
+    }
+
+    pod(T &&t) : m_t(std::move(t)) {}
+
+    pod(T const &t) : m_t(t) {}
+
+    pod &operator=(T &&t) {
+        m_t = std::move(t);
+        return *this;
+    }
+
+    pod &operator=(T const &t) {
+        m_t = t;
+        return *this;
+    }
+
+    operator T const &() const {
+        return m_t;
+    }
+
+    operator T &() {
+        return m_t;
+    }
+
+    T const &get() const {
+        return m_t;
+    }
+
+    T &get() {
+        return m_t;
+    }
+
+    template <class ...Ts>
+    pod &emplace(Ts &&...ts) {
+        ::new (&m_t) T(std::forward<Ts>(ts)...);
+        return *this;
+    }
+
+    void destroy() {
+        m_t.~T();
+    }
+};

07/01_bandwidth/07/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;

07/02_cache/01/main.cpp

@@ -8,7 +8,9 @@
 #include <x86intrin.h>
 #include <omp.h>
 
-// 32KB, 256KB, 12MB
+// L1: 32KB
+// L2: 256KB
+// L3: 12MB
 
 constexpr size_t n = 1<<28;
 
@@ -23,14 +25,23 @@ void BM_fill1GB(benchmark::State &bm) {
 }
 BENCHMARK(BM_fill1GB);
 
-void BM_fill32MB(benchmark::State &bm) {
+void BM_fill128MB(benchmark::State &bm) {
     for (auto _: bm) {
-        for (size_t i = 0; i < (1<<23); i++) {
+        for (size_t i = 0; i < (1<<25); i++) {
             a[i] = 1;
         }
     }
 }
-BENCHMARK(BM_fill32MB);
+BENCHMARK(BM_fill128MB);
+
+void BM_fill16MB(benchmark::State &bm) {
+    for (auto _: bm) {
+        for (size_t i = 0; i < (1<<22); i++) {
+            a[i] = 1;
+        }
+    }
+}
+BENCHMARK(BM_fill16MB);
 
 void BM_fill1MB(benchmark::State &bm) {
     for (auto _: bm) {
@@ -50,13 +61,13 @@ void BM_fill128KB(benchmark::State &bm) {
 }
 BENCHMARK(BM_fill128KB);
 
-void BM_fill4KB(benchmark::State &bm) {
+void BM_fill16KB(benchmark::State &bm) {
     for (auto _: bm) {
-        for (size_t i = 0; i < (1<<10); i++) {
+        for (size_t i = 0; i < (1<<12); i++) {
             a[i] = 1;
         }
     }
 }
-BENCHMARK(BM_fill4KB);
+BENCHMARK(BM_fill16KB);
 
 BENCHMARK_MAIN();

07/04_fusion/00/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()

07/04_fusion/00/main.cpp

@@ -0,0 +1,46 @@
+#include <iostream>
+#include <vector>
+#include <cmath>
+#include <cstring>
+#include <cstdlib>
+#include <array>
+#include <benchmark/benchmark.h>
+#include <x86intrin.h>
+#include <omp.h>
+
+// L1: 32KB
+// L2: 256KB
+// L3: 12MB
+
+constexpr size_t n = 1<<28;
+
+std::vector<float> a(n);
+
+void BM_original(benchmark::State &bm) {
+    for (auto _: bm) {
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = a[i] * 2;
+        }
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = a[i] + 1;
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_original);
+
+void BM_optimized(benchmark::State &bm) {
+    for (auto _: bm) {
+#pragma omp parallel for
+        for (size_t i = 0; i < n; i++) {
+            a[i] = a[i] * 2;
+            a[i] = a[i] + 1;
+        }
+        benchmark::DoNotOptimize(a);
+    }
+}
+BENCHMARK(BM_optimized);
+
+BENCHMARK_MAIN();

07/04_fusion/00/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();
+}

07/04_fusion/00/pod.h

@@ -0,0 +1,66 @@
+#pragma once
+
+#include <new>
+#include <utility>
+
+template <class T>
+struct pod {
+private:
+    T m_t;
+public:
+    pod() {}
+
+    pod(pod &&p) : m_t(std::move(p.m_t)) {}
+
+    pod(pod const &p) : m_t(p.m_t) {}
+
+    pod &operator=(pod &&p) {
+        m_t = std::move(p.m_t);
+        return *this;
+    }
+
+    pod &operator=(pod const &p) {
+        m_t = p.m_t;
+        return *this;
+    }
+
+    pod(T &&t) : m_t(std::move(t)) {}
+
+    pod(T const &t) : m_t(t) {}
+
+    pod &operator=(T &&t) {
+        m_t = std::move(t);
+        return *this;
+    }
+
+    pod &operator=(T const &t) {
+        m_t = t;
+        return *this;
+    }
+
+    operator T const &() const {
+        return m_t;
+    }
+
+    operator T &() {
+        return m_t;
+    }
+
+    T const &get() const {
+        return m_t;
+    }
+
+    T &get() {
+        return m_t;
+    }
+
+    template <class ...Ts>
+    pod &emplace(Ts &&...ts) {
+        ::new (&m_t) T(std::forward<Ts>(ts)...);
+        return *this;
+    }
+
+    void destroy() {
+        m_t.~T();
+    }
+};

07/04_fusion/00/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;

07/04_fusion/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()