#include "mat_mul.h"
#include "omp.h"
#include "util.h"

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

#define MIN(a,b) ((a>b) ? b : a)

#define ITILESIZE (20)
#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;
static int rows,offset; 

MPI_Status status;
MPI_Request request;

static void mat_mul_omp() {

#pragma omp parallel for
	for (int ii = 0; ii < rows; ii += ITILESIZE) {
		for (int jj = 0; jj < N; jj += JTILESIZE) {
			for (int kk = 0; kk < K; kk += KTILESIZE) {
				for (int k = kk; k < MIN(kk + KTILESIZE,K); k++) {
					for (int i = ii; i < MIN(ii + ITILESIZE, rows); i++) {
						float ar = A[i * K + k];
						for (int j = jj; j < MIN(jj + JTILESIZE, N); j++) {
							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) {

	int source,dest,remain;


	M = _M, N = _N, K = _K;
	num_threads = _num_threads, mpi_rank = _mpi_rank,
		    mpi_world_size = _mpi_world_size;

	omp_set_num_threads(num_threads);


	if (mpi_rank == 0) {
		/* send matrix data to the worker tasks */
		A = _A, B = _B, C = _C;

		rows = M/(mpi_world_size);
		remain = M%(mpi_world_size);
		offset = remain;

		for (dest=1; dest < mpi_world_size; dest++)
		{
			offset = offset + rows;
			MPI_Isend(&offset, 1, MPI_INT, dest, 1, MPI_COMM_WORLD, &request);
			MPI_Isend(&rows, 1, MPI_INT, dest, 1, MPI_COMM_WORLD, &request);
			MPI_Isend(&A[offset*K], rows*K, MPI_FLOAT,dest,1, MPI_COMM_WORLD, &request);
			MPI_Isend(B, K*N, MPI_FLOAT, dest, 1, MPI_COMM_WORLD, &request);

		}
		offset = 0;
		rows = offset+rows+remain;
		mat_mul_omp();

		/* wait for results from all worker tasks */
		for (source = 1; source < mpi_world_size; source++)
		{
			MPI_Recv(&offset, 1, MPI_INT, source, 2, MPI_COMM_WORLD, &status);
			MPI_Recv(&rows, 1, MPI_INT, source, 2, MPI_COMM_WORLD, &status);
			MPI_Recv(&C[offset*N], rows*N, MPI_FLOAT, source, 2, MPI_COMM_WORLD, &status);
		}

	}
	/*---------------------------- worker----------------------------*/
	if (mpi_rank > 0) {
		/* Matrix multiplication */
		MPI_Recv(&offset, 1, MPI_INT, 0, 1, MPI_COMM_WORLD, &status);
		MPI_Recv(&rows, 1, MPI_INT, 0, 1, MPI_COMM_WORLD, &status);
		alloc_mat(&A, rows, K);
		alloc_mat(&B, K, N);
		alloc_mat(&C, rows, N);
		zero_mat(C, rows, N); 
		MPI_Recv(A, rows*K, MPI_FLOAT, 0, 1, MPI_COMM_WORLD, &status);
		MPI_Recv(B, K*N, MPI_FLOAT, 0, 1, MPI_COMM_WORLD, &status);

		mat_mul_omp();

		MPI_Isend(&offset, 1, MPI_INT, 0, 2, MPI_COMM_WORLD, &request);
		MPI_Isend(&rows, 1, MPI_INT, 0, 2, MPI_COMM_WORLD, &request);
		MPI_Isend(C, rows*N, MPI_FLOAT, 0, 2, MPI_COMM_WORLD, &request);


	}
}