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

int num_threads = 120;

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

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) {

  MPI_Status status;
  MPI_Request request;
  	int MPSize[2];

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

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

	  //printf("here3 H,W %d %d\n", OH,OW);
  OH = (H + 2 * pad - dilation * (R - 1) - 1) / stride + 1;
  OW = (W + 2 * pad - dilation * (S - 1) - 1) / stride + 1;

if (mpi_world_size == 2) {
  if (mpi_rank == 0) {
    MPI_Isend(&input[MPSize[0]*C*H*W], MPSize[1]*C*H*W, MPI_FLOAT, 1, 0, MPI_COMM_WORLD, &request);
    MPI_Isend(filter, K*C*R*S, MPI_FLOAT, 1, 0, MPI_COMM_WORLD, &request);
  } else {
	  //printf("here1 %d,,%d\n", MPSize[0],MPSize[1]); 
    alloc_tensor(&input, MPSize[1], C, H, W);
    alloc_tensor(&output, MPSize[1], K, OH, OW);
    alloc_tensor(&filter, K, C, R, S);
    
    MPI_Recv(input, MPSize[1]*C*H*W, MPI_FLOAT, 0, 0, MPI_COMM_WORLD, &status);
    MPI_Recv(filter, K*C*R*S, MPI_FLOAT, 0, 0, MPI_COMM_WORLD, &status);
  }
}

#pragma omp parallel for num_threads(num_threads) collapse(3) schedule(dynamic)
    for (int n = 0; n < MPSize[mpi_rank]; ++n) {
      for (int k = 0; k < K; ++k) {
        for (int oh = 0; oh < OH; ++oh) {
          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;
          }
        }
      }
    }

	  //printf("here2 %d\n", mpi_world_size);
if (mpi_world_size == 2) {
  if (mpi_rank == 0) 
    MPI_Recv(&output[MPSize[0]*K*OH*OW], MPSize[1]*K*OH*OW, MPI_FLOAT, 1, 0, MPI_COMM_WORLD, &status);
  else
    MPI_Isend(output, MPSize[1]*K*OH*OW, MPI_FLOAT, 0, 0, MPI_COMM_WORLD, &request);
}
}

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_size(MPI_COMM_WORLD, &mpi_world_size);
  MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
}

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