#include "mat_mul.h"

#include <cstdlib>
#include <cstdio>
#include <pthread.h>
#include <algorithm>

#define BLOCK_SIZE 32 

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 i,j,k;
  int kk;
  int Bsize = BLOCK_SIZE;
  int id = *(int*)data;
  int slice = M / num_threads;
  int start_ptr = slice * id;
  int end_ptr  = (id == (num_threads-1)) ? M : (id+1) * slice;

  float A_mac;

  for (kk = 0; kk < K; kk += Bsize)
  {
    for (i = start_ptr; i < end_ptr; ++i)
    {
      for (k = kk; k < std::min(kk + Bsize, K); ++k)
      {
        A_mac = A[i*K+k];

        for (j = 0; j < N; ++j)
	{
          C[i*N+j] += A_mac * 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; 

  pthread_t *pthreads;
  pthreads = (pthread_t*) malloc(sizeof(pthread_t) * num_threads);

  for (i=0; i<num_threads; i++)
  {
      int * thread_id = (int *) malloc(sizeof(int));
      // tID = i;
      // thread_id = &tID;
      *thread_id = i; //init

      pthread_create(&pthreads[i], NULL, mat_mul_thread, thread_id);
  }

  for (i = 0; i < num_threads; i++)
  {
      pthread_join(pthreads[i], NULL);
  }
  free(pthreads);  
}