#include "mat_mul.h"
#include "util.h"

#include <stdio.h>
#include <stdlib.h>

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

#define max(a,b)  (((a) > (b)) ? (a) : (b))
#define min(a,b)  (((a) < (b)) ? (a) : (b))

static float *A, *B, *C;  //, *B_T;
static int M, N, K;
static int num_threads;

#if 0
static void* mat_mul_thread(void *data) {
  // TODO: parallelize & optimize matrix multiplication
  for (int i = 0; i < M; ++i) {
      for (int j = 0; j < N; ++j) {
        for (int k = 0; k < K; ++k) {
          C[i * N + j] += A[i * K + k] * 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
  pthread_t thread;
  pthread_create(&thread, NULL, mat_mul_thread, NULL);
  pthread_join(thread, NULL);
}

#else

static void* mat_mul_thread(void *data) {
  const int block_size = 64;
  int tid, blk_size;
  int row_start, row_end;
  
  tid = *(int *)(data);
  blk_size = M / num_threads;

  row_start = tid * blk_size;
  row_end = (tid == num_threads - 1) ? M : ((tid + 1) * blk_size);
  
  if ((K % 4) == 0) {
    float c0, c1, c2, c3;
    for (int kk = 0; kk < K; kk += block_size) {
      for (int jj = 0; jj < N; jj += block_size) {
        for (int i = row_start; i < row_end; i++) {
          int iK = i * K;
          for (int k = kk; k < min(kk + block_size, K); k += 4) {
            float Aik0 = A[iK + k];
            float Aik1 = A[iK + k + 1];
            float Aik2 = A[iK + k + 2];
            float Aik3 = A[iK + k + 3];
            int iN = i * N;
            int k0N = k * N;
            int k1N = (k + 1)* N;
            int k2N = (k + 2)* N;
            int k3N = (k + 3)* N;
            for (int j = jj; j < min(jj + block_size, N); j++) {
              c0 = Aik0 * B[k0N + j];
              c1 = Aik1 * B[k1N + j];
              c2 = Aik2 * B[k2N + j];
              c3 = Aik3 * B[k3N + j];
              C[iN + j] += (c0 + c1 + c2 + c3);
            }
          }
        }
      }
    }
  }
  else {
    for (int kk = 0; kk < K; kk += block_size) {
      for (int jj = 0; jj < N; jj += block_size) {
        for (int i = row_start; i < row_end; i++) {
          int iK = i * K;
          for (int k = kk; k < min(kk + block_size, K); k++) {
            float Aik = A[iK + k];
            int iN = i * N;
            int kN = k * N;
            for (int j = jj; j < min(jj + block_size, 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;

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

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

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

#endif