103 lines
2.3 KiB
C++
103 lines
2.3 KiB
C++
#include "mat_mul.h"
|
|
|
|
#include <cstdlib>
|
|
#include <cstdio>
|
|
#include <pthread.h>
|
|
#include <algorithm>
|
|
|
|
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
|
|
//for (int i = 0; i < M; ++i) {
|
|
// for (int j = 0; j < N; ++j) {
|
|
// for (int k = 0; k < K; ++k) {
|
|
// C[i * N + j] += A[i * K + k] * B[k * N + j];
|
|
// }
|
|
// }
|
|
//}
|
|
|
|
int pid = * (int *) data;
|
|
int slice = M / num_threads;
|
|
//int slice = 210;
|
|
int start = pid * slice;
|
|
int end = pid == num_threads - 1 ? M : (pid + 1) * slice;
|
|
|
|
//float Aik;
|
|
int bs = 40;
|
|
short k_min;
|
|
|
|
//float* _B;
|
|
|
|
|
|
// _B = (float*) malloc(sizeof(float) * 4096);
|
|
// for (int k =0; k<K; k++)
|
|
// for (int j=0; j<N; j++)
|
|
// _B[k*N+j]= B[j*N+k];
|
|
//
|
|
|
|
for (int kk = 0; kk < K; kk += bs) {
|
|
k_min = std::min(kk+bs,K);
|
|
//for (int jj = 0; jj < N; jj += bs) {
|
|
//j_min = std::min(jj + bs, N);
|
|
//j_min = j_min>>2;
|
|
for (int i = start; i < end; ++i) {
|
|
//for (int k = kk; k < std::min(kk + bs, K); ++k) {
|
|
for (int k = kk; k < k_min; ++k) {
|
|
//for (int j = jj; j < j_min; j=j+1) {
|
|
//Aik = A[i * K + k];
|
|
for (int j = 0; j < N; j=j+1) {
|
|
//for (int j = jj; j < std::min(jj + bs, N); ++j) {
|
|
//C[i * N + j] += Aik * B[k * N + j];
|
|
C[i * N + j] += A[i*K +k] * B[k * N + j];
|
|
//C[i * N + j] += A[i*K +k] * _B[j * N + k];
|
|
}
|
|
}
|
|
//}
|
|
}
|
|
}
|
|
|
|
//free(_B);
|
|
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 thread;
|
|
//pthread_create(&thread, NULL, mat_mul_thread, NULL);
|
|
//pthread_join(thread, NULL);
|
|
|
|
|
|
|
|
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);
|
|
}
|
|
|
|
|
|
|
|
|
|
}
|