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

62 lines
1.3 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;
inline int min(const int a, const int b){return a>b ? b : a;};
static void* mat_mul_thread(void *data) {
int pid = * (int *) data;
int slice = M / num_threads;
int start = pid * slice;
int end = (pid == num_threads-1) ? M : (pid+1)*slice;
float Aik;
int nblk = 32;
for (int kk=0; kk<K; kk+=nblk)
{
for(int i=start; i<end; ++i)
{
int iN = i*N;
for(int k=kk; k<min(kk+nblk, K); ++k)
{
int iK = i*K;
int kN = k*N;
Aik = A[iK + k];
for(int j=0; j<N; ++j)
{
C[iN + j] += Aik * B[kN + 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);
}
}