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

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;


#include "util.h"
#define MASTER 0
#define FROM_MASTER 1
#define FROM_WORKER 2

static int Ndiv_rev[2];

static void mat_conv() {
#pragma omp parallel for num_threads(1024) collapse(3) schedule(dynamic)
 for (int n = 0; n < Ndiv_rev[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;
                         //input[n * C * H * W   +   c * H * W   +   h * W   +  w];


        }
      }
    }
  }

}











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) {


  int Ndiv, Nextra, dest;
  int src;
  int numworkers, mtype;
  MPI_Status status;
  MPI_Request request;

  numworkers = mpi_world_size-1;
  Ndiv       = N/mpi_world_size;
  Nextra     = N%mpi_world_size;

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

  int CHW   = C*H*W;
  int KCRS  = K*C*R*S;
  int KOHOW = K*OH*OW;

  Ndiv_rev[1]   =  Ndiv+Nextra ;
  Ndiv_rev[0]   =  Ndiv;
   // WORKER initialize
  if (mpi_rank > 0 && mpi_world_size==2) {
    alloc_tensor(&input , Ndiv_rev[1], C, H, W);
    alloc_tensor(&output, Ndiv_rev[1], K, OH, OW);
    alloc_tensor(&filter, K, C, R, S);
  }



    // 1. MASTER send to WORKER 
if (mpi_rank == 0) {
  input = _input;
  output = _output;
  filter = _filter;

  mtype  = FROM_MASTER;
  for (dest=1; dest<=numworkers; dest++)
  {
    MPI_Isend(&input[Ndiv_rev[0]*CHW] ,Ndiv_rev[1]*CHW, MPI_FLOAT, dest, mtype, MPI_COMM_WORLD, &request);
    MPI_Isend(&filter[0]         ,    KCRS, MPI_FLOAT, dest, mtype, MPI_COMM_WORLD, &request);
  }
}
else  { // 2. WORKER rcv from MASTER 

    mtype = FROM_MASTER;
    MPI_Recv(&input[0]     , Ndiv_rev[1]*CHW, MPI_FLOAT, MASTER , mtype, MPI_COMM_WORLD, &status);
    MPI_Recv(&filter[0]    ,     KCRS, MPI_FLOAT, MASTER , mtype, MPI_COMM_WORLD, &status);



    mat_conv();


    //printf("\nrank div off  %f, %f,  %f  \n", output[0], output[1], output[2]);
    // WORKER send to MASTER
    mtype = FROM_WORKER;
    MPI_Isend(&output[0]  , Ndiv_rev[1]*KOHOW, MPI_FLOAT, MASTER, mtype, MPI_COMM_WORLD, &request);

                        //output[n * K *OH *OW   +   k *OH *OW   +  oh *OW   + ow] = o;
                         //input[n * C * H * W   +   c * H * W   +   h * W   +  w];

}



  


//MASTER EXECUTE
  if (mpi_rank == 0) {

    mat_conv();
      mtype = FROM_WORKER;
  for (src=1; src<=numworkers; src++)
  {
    MPI_Recv(&output[Ndiv_rev[0]*KOHOW], Ndiv_rev[1]*KOHOW, MPI_FLOAT, src , mtype, MPI_COMM_WORLD, &status);
  }

  }
 


}






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) {
}