#include "mat_mul.h"

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

#include "util.h"
#include <immintrin.h>

#define MASTER 0
#define FROM_MASTER 1
#define FROM_WORKER 2
#define ITILESIZE (32)
#define JTILESIZE (1024)
#define KTILESIZE (1024)

static float *A, *B, *C;
static int M, N, K;
static int num_threads;
static int mpi_rank, mpi_world_size;

MPI_Status status;
MPI_Request request;

int numworkers, source, dest, mtype, rows, averow, extra, offset, i, j, k;

static int min(int x, int y) {
  return (x<y)? x: y;
}

static void mat_mul_omp() {
  // TODO: parallelize & optimize matrix multiplication
  // Use num_threads per node
  
int is=0;
int ie=rows;


#pragma omp parallel for num_threads(num_threads) schedule(dynamic)

  for (int ii = is; ii < ie; ii += ITILESIZE) {
    for (int jj = 0; jj < N; jj += JTILESIZE) {
      for (int kk = 0; kk < K; kk += KTILESIZE) {
        
        for (k = kk; k < min(K, kk + KTILESIZE); k++) {
          for (i = ii; i < min(ie, ii + ITILESIZE); i++) {
            float ar = A[i * K + k];
            for (j = jj; j < min(N, jj+ JTILESIZE); j +=1) {
              C[i * N + j] += ar * B[k * N + j];
            }
          }
        }
      }
    }
  }

      
}

void mat_mul(float *_A, float *_B, float *_C, int _M, int _N, int _K,
             int _num_threads, int _mpi_rank, int _mpi_world_size) {
  A = _A, B = _B, C = _C;
  M = _M, N = _N, K = _K;
  num_threads = _num_threads, mpi_rank = _mpi_rank,
  mpi_world_size = _mpi_world_size;

  // TODO: parallelize & optimize matrix multiplication on multi-node
  // You must allocate & initialize A, B, C for non-root processes

	numworkers = mpi_world_size;
if(numworkers ==0) {
	rows=M;
    mat_mul_omp();
} else {


	if(mpi_rank == MASTER) {


	averow = M/numworkers;
	int s, e;
	extra = M%numworkers;
	offset =0;
	mtype=FROM_MASTER;
		for(dest=1;dest<numworkers;dest++){
		//rows=(dest+1<=extra) ? averow+1:averow;
		//offset = offset+rows;
		s=offset=dest*averow;
		e=dest==numworkers-1 ? M : (dest+1)*averow;
		rows=e-s;
		printf("sending %d rows to task %d offset =%d\n", rows, dest, offset);
		MPI_Isend(&offset, 1, MPI_INT, dest, mtype,MPI_COMM_WORLD,&request);
		MPI_Isend(&rows, 1, MPI_INT, dest, mtype,MPI_COMM_WORLD,&request);
		MPI_Isend(&A[offset*K], rows*K, MPI_FLOAT, dest, mtype,MPI_COMM_WORLD,&request);

		MPI_Isend(B, K*N, MPI_FLOAT, dest, mtype,MPI_COMM_WORLD,&request);
		}
		//rows= (extra > 0) ? averow+1:averow;
		rows=averow;
    mat_mul_omp();
		mtype=FROM_WORKER;
		for(i=1;i<numworkers;i++){
		source=i;
		MPI_Recv(&offset, 1, MPI_INT, source,mtype,MPI_COMM_WORLD,&status);
		MPI_Recv(&rows, 1, MPI_INT, source,mtype,MPI_COMM_WORLD,&status);
		MPI_Recv(&C[offset*N], rows*N, MPI_FLOAT, source, mtype,MPI_COMM_WORLD,&status);
		printf("Received result from task %d\n", source);
		}

	} //MASTER ==0;
	
	if(mpi_rank > MASTER) {
		alloc_mat(&A, M, K);
		alloc_mat(&B, K, N);
		alloc_mat(&C, M, N);
		zero_mat(C,M,N);
		mtype=FROM_MASTER;
		MPI_Recv(&offset, 1, MPI_INT, MASTER, mtype, MPI_COMM_WORLD,&status);
		MPI_Recv(&rows, 1, MPI_INT, MASTER, mtype, MPI_COMM_WORLD,&status);
		MPI_Recv(A, rows*K, MPI_FLOAT, MASTER, mtype, MPI_COMM_WORLD,&status);
		MPI_Recv(B, K*N, MPI_FLOAT, MASTER, mtype, MPI_COMM_WORLD,&status);

    mat_mul_omp();
		mtype = FROM_WORKER;
		MPI_Isend(&offset, 1, MPI_INT, MASTER, mtype, MPI_COMM_WORLD,&request);
		MPI_Isend(&rows, 1, MPI_INT, MASTER, mtype, MPI_COMM_WORLD,&request);
		MPI_Isend(C, rows*N, MPI_FLOAT, MASTER, mtype, MPI_COMM_WORLD,&request);
	}
}




}