chundoong-lab-ta/SamsungDS22/submissions/HW2/jh8846.kim/mat_mul.cpp

114 lines
2.1 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 M_div = M / num_threads;
int start = pid * M_div;
int end;
if(pid==num_threads - 1) end = M;
else end = (pid+1)*M_div;
float temp;
int bs = 50;
int k_min =0;
for (int kk = 0; kk < K; kk += bs)
{
for (int i = start; i < end; i++)
{
if(K>=(kk+bs)) k_min = kk+bs;
else k_min = K;
for (int k = kk; k < k_min; k++)
{
temp = A[i * K + k];
for (int j = 0; j < N; j++)
{
C[i*N+(j+0)] += temp*B[k*N+(j+0)];
}
}
}
}
/*
for (int i = 0; i < M; i++)
{
for (int k = 0; k < N; k++)
{
float temp = A[i*K + k];
for (int j = start; j < end; j=j+8)
{
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)];
}
}
}
*/
/*
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];
}
}
}
*/
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);
pthread_join(thread, NULL);
*/
}