#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 int slice;

static void* mat_mul_thread(void *data) {
	//// TODO: parallelize & optimize matrix multiplication
	int pid = * (int *) data;

	int start = pid * slice;
	int end = ( pid == num_threads - 1 ) ? M : (pid + 1) * slice;

	int bs = 32;
	for (int kk=0; kk < K; kk+=bs) {
		int row_A = start * K;
		int row_C = start * N;
		for (int i= start; i < end; ++i) {
			for (int k=kk; k < ((kk+bs < K)? kk+bs : K); ++k){
				for (int j=0; j < N; ++j) {
					C[row_C + j] += A[row_A + k] * B[k * N + j];
				}
			}
			row_A += K; row_C += N;
		}
	}
	
	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
	if (M < num_threads) num_threads = M;

	slice = M / num_threads;

	pthread_t* threads;
	threads = (pthread_t*)malloc((num_threads) * sizeof(pthread_t));
	for (int i=0; i < num_threads; ++i) {
		int * pid = (int *) malloc(sizeof(int));
		*pid = i;
		pthread_create(&threads[i], NULL, mat_mul_thread, (void*)pid);
	}

	for(int i=0; i < num_threads; ++i)
		pthread_join (threads[i], NULL);
}