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

#define NODE 2

static float *input, *output, *filter;
static int N, C, H, W;
static int K, R, S;
static int OH, OW;
static int pad;
static int dilation;
static int stride;
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) {
  input = _input;
  output = _output;
  filter = _filter;
  
  MPI_Status status;
  MPI_Request request1,request2;

  OH = (H + 2 * pad - dilation * (R - 1) - 1) / stride + 1;
  OW = (W + 2 * pad - dilation * (S - 1) - 1) / stride + 1;
  
  int rows;
  int offset;
  int half = N/mpi_world_size;
  if(mpi_world_size==2){
    rows = N-half;
    offset = N/mpi_world_size;
  }else{
    rows = N;
    offset =0;
  }
  
    
  

  if (mpi_rank == 0) {
    if(mpi_world_size==2){
      MPI_Isend(&input[(offset)*C*H*W], (rows)*C*H*W, MPI_FLOAT, 1 , 1, MPI_COMM_WORLD,&request1);
      MPI_Isend(filter, K*C*R*S, MPI_FLOAT, 1, 1, MPI_COMM_WORLD,&request2);
    }
      
#pragma omp parallel for collapse(4) num_threads(100)
    for (int n = 0; n < N/mpi_world_size; ++n) {
      //printf("n=%d\n",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;
          }
        }
      }
    }    

    if(mpi_world_size==2){
      MPI_Recv(&output[(offset)*K*OH*OW], (rows)*K*OH*OW, MPI_FLOAT, 1, 2, MPI_COMM_WORLD, &status);
    }
    






  }else if(mpi_rank >0){
    alloc_tensor(&input, rows, C,H,W);
    alloc_tensor(&filter, K,C,R,S);
    alloc_tensor(&output, rows, K,OH,OW);

    MPI_Recv(input, (rows)*C*H*W, MPI_FLOAT, 0, 1, MPI_COMM_WORLD, &status);
    MPI_Recv(filter,  K*C*R*S, MPI_FLOAT, 0, 1, MPI_COMM_WORLD, &status);

#pragma omp parallel for collapse(4) num_threads(100)
    for (int n = 0; n < rows; ++n) {
      //printf("n=%d\n",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;
          }
        }
      }
    }      
    
   
    MPI_Send(output, (rows)*K*OH*OW, MPI_FLOAT, 0, 2, MPI_COMM_WORLD);

  }

}


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);
}

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