#include "convolution.h"
#include "util.h"
#include <mpi.h>
#include <stdio.h>

static int mpi_rank, mpi_world_size;
static float *input, *output, *filter;
static int N, C, H, W, K, R, S;
static int OH, OW;
static int pad;
static int dilation;
static int stride;
static int size[2];

int num_threads = 40;

void convolution(
    float *_input, float *_output, float *_filter, 
    int _N, int _C, int _H, int _W,
    int _K, int _R, int _S,
    int _pad, int _dilation, int _stride) {

  input  = _input;
  output = _output;
  filter = _filter;

  if ((mpi_rank == 0) && (mpi_world_size == 2))
  {
      MPI_Send(&input[size[0]*C*H*W],size[1]*C*H*W, MPI_FLOAT,1,0,MPI_COMM_WORLD);
      MPI_Send(filter,K*C*R*S,MPI_FLOAT,1,0,MPI_COMM_WORLD);
  }
  else if (mpi_world_size == 2)
  {
      alloc_tensor(&input,size[1],C,H,W);
      alloc_tensor(&output,size[1],K,OH,OW);
      alloc_tensor(&filter,K,C,R,S);
      MPI_Recv(input,size[1]*C*H*W,MPI_FLOAT,0,0,MPI_COMM_WORLD,nullptr);
      MPI_Recv(filter,K*C*R*S,MPI_FLOAT,0,0,MPI_COMM_WORLD,nullptr);
  }

  #pragma omp parallel for num_threads(num_threads)  collapse(3) schedule(dynamic)
  for (int n = 0; n < size[mpi_rank]; ++n) {
    for (int k = 0; k < K; ++k) {
      for (int oh = 0; oh < OH; ++oh) {
        #pragma omp parallel for num_threads(num_threads) schedule(dynamic)
        for (int ow = 0; ow < OW; ++ow) {
          float o = 0.f;
          for (int c = 0; c < C; ++c) {
            for (int r = 0; r < R; ++r) {
              for (int s = 0; s < S; ++s) {
                int h = oh * stride - pad + r * dilation;
                int w = ow * stride - pad + s * dilation;
                if (h < 0 || h >= H || w < 0 || w >= W) continue;
                float i = input[n * C * H * W + c * H * W + h * W + w];
                float f = filter[k * C * R * S + c * R * S + r * S + s];
                o += i * f;
              }
            }
          }
          output[n * K * OH * OW + k * OH * OW + oh * OW + ow] = o;
        }
      }
    }
  }

}

void convolution_init(
    int _N, int _C, int _H, int _W,
    int _K, int _R, int _S,
    int _pad, int _dilation, int _stride) {
  N = _N; C = _C; H = _H; W = _W;
  K = _K; R = _R; S = _S;
  pad = _pad;
  dilation = _dilation;
  stride = _stride;

  MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
  MPI_Comm_size(MPI_COMM_WORLD, &mpi_world_size);

  if(mpi_world_size==2)
  {
      size[1] = _N/2;
  }
  else
  {
      size[1] = 0;
  }
  size[0] = N - size[1];

  OH = (H + 2 * pad - dilation * (R - 1) - 1) / stride + 1;
  OW = (W + 2 * pad - dilation * (S - 1) - 1) / stride + 1;
}

void convolution_final(
    int _N, int _C, int _H, int _W,
    int _K, int _R, int _S,
    int _pad, int _dilation, int _stride) {

  if((mpi_rank == 0) && (mpi_world_size == 2)) {
      MPI_Recv(&output[size[0]*K*OH*OW],size[1]*K*OH*OW,MPI_FLOAT,1,0,MPI_COMM_WORLD,nullptr);
  }
  else if (mpi_world_size == 2) {
      MPI_Send(output,size[1]*K*OH*OW,MPI_FLOAT,0,0,MPI_COMM_WORLD);
  }
}