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

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 slice = M / num_threads;
  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) {
    for(int i = start; i < end; ++i) {
      for(int k = kk; k < std::min(kk+bs, K); ++k) {
	float Aik = A[i*K + k];
	int j = 0;
	for(; j < N-7; j += 8) {
	  C[i*N + j] += Aik * B[k*N + j]; 
	  C[i*N + j + 1] += Aik * B[k*N + j + 1]; 
	  C[i*N + j + 2] += Aik * B[k*N + j + 2]; 
	  C[i*N + j + 3] += Aik * B[k*N + j + 3]; 
	  C[i*N + j + 4] += Aik * B[k*N + j + 4]; 
	  C[i*N + j + 5] += Aik * B[k*N + j + 5]; 
	  C[i*N + j + 6] += Aik * B[k*N + j + 6]; 
	  C[i*N + j + 7] += Aik * B[k*N + j + 7]; 
	}
        for(; j < N; ++j)
	  C[i*N + j] += Aik * B[k*N + j];	
      }
    }
  }
  /*
  float Aik;
  int bs = 32;
  for (int kk = 0; kk < K; kk += bs) {
    for (int jj = 0; jj < N; jj += bs) {
      for (int i = start; i < end; ++i) {
        for (int k = kk; k < std::min(kk + bs, K); ++k) {
          Aik = A[i * K + k];
          for (int j = jj; j < std::min(jj + bs, N); ++j) {
            C[i * N + j] += Aik * B[k * N + j];
          }
        }
      }
    }
  }
  */  
  free(data);

  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[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);
  }
}