89 lines
1.6 KiB
C++
89 lines
1.6 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
|
|
|
|
//for (int i = row_start; i < row_end; ++i) {
|
|
//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 x = * (int *) data;
|
|
int start = x * (M / num_threads);
|
|
int end = start + (M / num_threads);
|
|
|
|
if (x == num_threads - 1)
|
|
{
|
|
end = M;
|
|
}
|
|
|
|
float Aik;
|
|
int bs = 50;
|
|
int k_max;
|
|
|
|
for (int kk = 0; kk < K; kk += bs)
|
|
{
|
|
if ((kk + bs) < K)
|
|
{
|
|
k_max = kk + bs;
|
|
}
|
|
else
|
|
{
|
|
k_max = K;
|
|
}
|
|
|
|
for (int i = start; i < end; ++i)
|
|
{
|
|
for (int k = kk; k < k_max; ++k)
|
|
{
|
|
Aik = A[i * K + k];
|
|
|
|
for (int j = 0; j < N; ++j)
|
|
{
|
|
C[i * N + j] += Aik * 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
|
|
int i;
|
|
int* p = (int*)malloc(sizeof(int) * num_threads);
|
|
|
|
pthread_t thread[num_threads];
|
|
|
|
for (i = 0; i < num_threads; i++)
|
|
{
|
|
p[i] = i;
|
|
|
|
pthread_create(&thread[i], NULL, mat_mul_thread, &p[i]);
|
|
}
|
|
|
|
for (i = 0; i <num_threads; i++)
|
|
{
|
|
pthread_join(thread[i], NULL);
|
|
}
|
|
|
|
free(p);
|
|
|
|
}
|