chundoong-lab-ta/SHPC2022/hw4_answer/matmul/matmul.c

#define _GNU_SOURCE
#include "util.h"
#include <immintrin.h>
#include <mpi.h>
#include <omp.h>
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>

void matmul(const float *A, const float *B, float *C, int M, int N, int K,
            int threads_per_process, int mpi_rank, int mpi_world_size) {
  int start = ((M) / mpi_world_size) * mpi_rank;
  int end = ((M) / mpi_world_size) * (mpi_rank + 1);

  int nrow = end - start;

  if (nrow > 0) {
    MPI_Scatter(A, (nrow)*K, MPI_FLOAT, A, (nrow)*K, MPI_FLOAT, 0,
                MPI_COMM_WORLD);
    MPI_Bcast(B, K * N, MPI_FLOAT, 0, MPI_COMM_WORLD);

#pragma omp parallel for
    for (int i = 0; i < nrow; ++i) {
      for (int k = 0; k < K; ++k) {
        for (int j = 0; j < N; ++j) {
          C[i * N + j] += A[i * K + k] * B[k * N + j];
        }
      }
    }
  }

  if (mpi_rank == 0) {
#pragma omp parallel for
    for (int i = nrow * mpi_world_size; i < M; ++i) {
      for (int k = 0; k < K; ++k) {
        for (int j = 0; j < N; ++j) {
          C[i * N + j] += A[i * K + k] * B[k * N + j];
        }
      }
    }
  }

  if (nrow > 0) {
    MPI_Gather(C, (nrow)*N, MPI_FLOAT, C, (nrow)*N, MPI_FLOAT, 0,
               MPI_COMM_WORLD);
  }
}