92 lines
2.4 KiB
C++
92 lines
2.4 KiB
C++
#include "mat_mul.h"
|
|
|
|
#include <cstdlib>
|
|
#include <cstdio>
|
|
#include <pthread.h>
|
|
|
|
static float *A, *B, *C;
|
|
static int M, N, K;
|
|
static int num_threads;
|
|
|
|
static void* mat_mul_thread(void *data) {
|
|
// TODO: parallelize & optimize matrix multiplication
|
|
|
|
int pid = * (int *) data;
|
|
|
|
int slice = M / num_threads;
|
|
int start = pid * slice;
|
|
int end = pid == num_threads - 1 ? M : (pid + 1) * slice;
|
|
|
|
float temp;
|
|
int bs = 32;
|
|
int j;
|
|
|
|
for (int kk = 0; kk < K; kk += bs) {
|
|
for (int i = start; i < end; ++i) {
|
|
for (int k = kk; k < (kk + bs > K ? K : kk + bs); ++k) {
|
|
|
|
temp = A[i * K + k];
|
|
|
|
for (j = 0; j < ((N / 8) * 8); j += 8)
|
|
{
|
|
//C[i * N + j] += temp * B[k * N + j];
|
|
C[i * N + j + 0] += temp * B[k * N + j + 0];
|
|
C[i * N + j + 1] += temp * B[k * N + j + 1];
|
|
C[i * N + j + 2] += temp * B[k * N + j + 2];
|
|
C[i * N + j + 3] += temp * B[k * N + j + 3];
|
|
|
|
C[i * N + j + 4] += temp * B[k * N + j + 4];
|
|
C[i * N + j + 5] += temp * B[k * N + j + 5];
|
|
C[i * N + j + 6] += temp * B[k * N + j + 6];
|
|
C[i * N + j + 7] += temp * B[k * N + j + 7];
|
|
/*
|
|
C[i * N + j + 8] += temp * B[k * N + j + 8];
|
|
C[i * N + j + 9] += temp * B[k * N + j + 9];
|
|
C[i * N + j + 10] += temp * B[k * N + j + 10];
|
|
C[i * N + j + 11] += temp * B[k * N + j + 11];
|
|
|
|
C[i * N + j + 12] += temp * B[k * N + j + 12];
|
|
C[i * N + j + 13] += temp * B[k * N + j + 13];
|
|
C[i * N + j + 14] += temp * B[k * N + j + 14];
|
|
C[i * N + j + 15] += temp * B[k * N + j + 15];
|
|
*/
|
|
}
|
|
|
|
for (; j < N; ++j) {
|
|
C[i * N + j] += temp * B[k * N + j];
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
void mat_mul(float *_A, float *_B, float *_C, int _M, int _N, int _K, int _num_threads) {
|
|
A = _A, B = _B, C = _C;
|
|
M = _M, N = _N, K = _K;
|
|
num_threads = _num_threads;
|
|
|
|
// TODO: create '_num_threads' pthreads
|
|
|
|
pthread_t * threads;
|
|
threads = (pthread_t *) malloc(sizeof(pthread_t) * num_threads);
|
|
|
|
for (int i = 0; i < num_threads; i++) {
|
|
int * pid = (int *) malloc(sizeof(int));
|
|
*pid = i;
|
|
pthread_create(&threads[i], NULL, mat_mul_thread, pid);
|
|
}
|
|
|
|
for (int i = 0; i < num_threads; i++) {
|
|
pthread_join(threads[i], NULL);
|
|
}
|
|
|
|
|
|
/*
|
|
pthread_t thread;
|
|
pthread_create(&thread, NULL, mat_mul_thread, NULL);
|
|
1pthread_join(thread, NULL);
|
|
*/
|
|
}
|