Benchmark tools

benchmark/benchmark.sh

+#!/bin/bash
+
+export REPEAT=20
+export PROG="a.out"
+export FILE="gemm"
+export TIME=/usr/bin/time
+
+rm -f ${FILE}_simple.res
+rm -f ${FILE}_eigen.res
+rm -f ${FILE}_custom.res
+
+for size in $(seq 3 8); do
+    echo "Size=$size"
+
+    echo -e "\t Simple"
+    g++ gemm_simple.cpp -DSIZE=$size -O3 -march=native
+    for i in $(seq 1 ${REPEAT}); do
+        ${TIME} -f "%U, " ./a.out 2> >(tr -d '\n' | tee -a ${FILE}_simple.res) && echo
+    done
+    echo >> ${FILE}_simple.res
+
+    echo -e "\t Eigen"
+    g++ gemm_eigen.cpp -DSIZE=$size -O3 -march=native
+    for i in $(seq 1 ${REPEAT}); do
+        ${TIME} -f "%U, " ./a.out 2> >(tr -d '\n' | tee -a ${FILE}_eigen.res) && echo
+    done
+    echo >> ${FILE}_eigen.res
+
+    echo -e "\t Custom"
+    g++ gemm_custom.cpp -I ${BOOST_SIMD_ROOT} -DSIZE=$size -O3 -march=native
+    for i in $(seq 1 ${REPEAT}); do
+        ${TIME} -f "%U, " ./a.out 2> >(tr -d '\n' | tee -a ${FILE}_custom.res) && echo
+    done
+    echo >> ${FILE}_custom.res
+
+done

benchmark/gemm_custom.cpp

+#include<iostream>
+#include "gemm_header.h"
+#include "../fast_matrix.hpp"
+
+int main(int argc, char **argv) {
+    alignas(32) float a[SIZE*SIZE];
+    alignas(32) float b[SIZE*SIZE];
+
+    for (int i=0; i<SIZE; i++) {
+        for (int j=0; j<SIZE; j++) {
+            a[i*SIZE+j] = i+j;
+        }
+    }
+
+    for (int i=0; i<SIZE; i++) {
+        for (int j=0; j<SIZE; j++) {
+            float val;
+            if (i == 0 && j == 1) val = -1;
+            else if (i == 1 && j == 0) val = 1;
+            else if (i > 1 && i == j && i % 2) val = -1;
+            else if (i > 1 && i == j && !(i % 2)) val = 1;
+            else val = 0;
+            b[i*SIZE+j] = val;
+        }
+    }
+
+    using M = BaseMatrix<float, SIZE, VECSIZE>;
+    M pa(a), pb(b), pc;
+
+    for (unsigned int i=0; i < REPEAT; i++) {
+        matrix_mul_m_m<M>(pa, pb, pc);
+        matrix_mul_m_m<M>(pb, pc, pa);
+    }
+
+    std::cout << pa << std::endl;
+    return 0;
+}

benchmark/gemm_eigen.cpp

+#include<eigen3/Eigen/Dense>
+#include<iostream>
+
+#include "gemm_header.h"
+
+
+int main(int argc, char **argv) {
+    Eigen::Matrix<float, SIZE, SIZE> pa, pb, pc;
+
+    for (int i=0; i<SIZE; i++) {
+        for (int j=0; j<SIZE; j++) {
+            pa(i,j) = i+j;
+        }
+    }
+
+    for (int i=0; i<SIZE; i++) {
+        for (int j=0; j<SIZE; j++) {
+            float val;
+            if (i == 0 && j == 1) val = -1;
+            else if (i == 1 && j == 0) val = 1;
+            else if (i > 1 && i == j && i % 2) val = -1;
+            else if (i > 1 && i == j && !(i % 2)) val = 1;
+            else val = 0;
+            pb(i,j) = val;
+        }
+    }
+
+    for (unsigned int i=0; i < REPEAT; i++) {
+        pc = pa * pb;
+        pa = pb * pc;
+    }
+
+    std::cout << pa << std::endl;
+    return 0;
+}

benchmark/gemm_header.h

+#define REPEAT 200000000
+
+#ifndef SIZE
+#define SIZE 4
+#endif
+
+#ifndef VECSIZE
+#define VECSIZE 8
+#endif
+

benchmark/gemm_simple.cpp

+#include <iostream>
+
+#include "gemm_header.h"
+
+template <typename T>
+void print_matrix(T* matrix, int n, int m) {
+    for (int i=0; i<n; i++) {
+        for (int j=0; j<m; j++) {
+            std::cout << matrix[i*m+j] << (j == m-1 ? "" : ", ");
+        }
+        std::cout << std::endl;
+    }
+}
+
+template <typename T, int N>
+void gemm(T *a, T *b, T*c) {
+    for (int i=0; i < N; i++) {
+        for (int j=0; j < N; j++) {
+            T res = 0;
+            for (int k=0; k < N; k++) res += a[i*N+k] * b[k*N+j];
+            c[i*N+j] = res;
+        }
+    }
+}
+
+int main(int argc, char **argv) {
+    alignas(32) float a[SIZE*SIZE];
+    alignas(32) float b[SIZE*SIZE];
+    alignas(32) float c[SIZE*SIZE];
+
+    for (int i=0; i<SIZE; i++) {
+        for (int j=0; j<SIZE; j++) {
+            a[i*SIZE+j] = i+j;
+        }
+    }
+
+    for (int i=0; i<SIZE; i++) {
+        for (int j=0; j<SIZE; j++) {
+            float val;
+            if (i == 0 && j == 1) val = -1;
+            else if (i == 1 && j == 0) val = 1;
+            else if (i > 1 && i == j && i % 2) val = -1;
+            else if (i > 1 && i == j && !(i % 2)) val = 1;
+            else val = 0;
+            b[i*SIZE+j] = val;
+        }
+    }
+
+    for (unsigned int i=0; i<REPEAT; i++) { // Computation Loop
+        gemm<float, SIZE>(a, b, c);
+        gemm<float, SIZE>(b, c, a);
+    }
+
+    print_matrix<float>(a, SIZE, SIZE);
+    return 0;
+}