#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
  //printf("thread #: %d\n", *(int*)data); fflush(stdout);
  int pid = *(int*)data;

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

  int ROW_BLOCK = 64;
  int COL_BLOCK = 1;
  //register float A_reg;
  //for (int kk = 0; kk < K; kk += BLOCK){
  for (int kk = 0; kk < K; kk += ROW_BLOCK){
  //for (int jj = 0; jj < N; jj += BLOCK){
  //for (int jj = 0; jj < N; jj += COL_BLOCK){
  for (int i = start; i < end; ++i) {
    for(int k = kk; k < (kk+ROW_BLOCK < K ? kk+ROW_BLOCK : K); ++k) {	  
	    //A_reg = A[i * K + k];
        //for (int j = jj; j < (jj+COL_BLOCK < N ? jj+COL_BLOCK : N); ++j) {
        for (int j = 0; j < N; ++j) {
          C[i * N + j] += A[i * K + k] * B[k * N + j];
          //C[i * N + j] += A_reg * B[k * N + j];

	  //if(i==0 && j==640) {
	//	printf("val c : %f\n",C[i*N+j]); 
	//	fflush(stdout);
	//  }
        }
    }
  }
  //}
  }
  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_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(&thread[i], NULL, mat_mul_thread, pid);
  }
  for (int i = 0; i< num_threads; i++) {
  	pthread_join(thread[i], NULL);
  }
}