void cpu_benchmark(int n, int samples, double &duration, double &GFLOPS)
{
type_precision* A = new type_precision[n * n];
type_precision* B = new type_precision[n * n];
type_precision* C = new type_precision[n * n];
cputime_type start_tick, end_tick;
duration = 9999999999.0;
int b = 0;
for (int i = 0; i < samples; i++)
{
re_random_vec(A, n*n);
re_random_vec(B, n*n);
re_random_vec(C, n*n);
get_ticks(start_tick);
cblas_sgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, n, n, n,
1.0, A, n, B, n, 1.0, C, n);
get_ticks(end_tick);
duration = min(duration, static_cast<double>(ticks2sec(end_tick, start_tick)));
int a = 0;
for (int j = 0; j < n * n ; j++)
{
a += A[j] + B[j] + C[j];
}
b += a;
}
//!2nnn - nn + 2nn (from+c)
GFLOPS = gemm_flops(n, n, n, 0);
cout << b;
delete []A;
delete []B;
delete []C;
}
static void dgemm_driver(blas_idx_t m_global, blas_idx_t n_global, blas_idx_t k_global)
{
auto grid = std::make_shared<blacs_grid_t>();
auto a = block_cyclic_mat_t::random(grid, m_global, k_global);
// auto b = block_cyclic_mat_t::random(grid, k_global, n_global);
auto c = block_cyclic_mat_t::random(grid, m_global, n_global);
//for test TODO
double *dd = new double[m_global*k_global];
for (int i = 0; i < m_global*k_global; i++)
{
dd[i] = i;
}
auto b = block_cyclic_mat_t::createWithArray(grid, m_global, k_global, dd);
//从这里开始矩阵的运算
MPI_Barrier(MPI_COMM_WORLD);
double alpha = 1.0, beta = 0.0;
double t0 = MPI_Wtime();
char NEIN = 'N'; //表示不进行转置
blas_idx_t ia = 1, ja = 1, ib = 1, jb = 1, ic = 1, jc = 1;
// sub(C) = alpha*op(sub(A))*op(sub(B)) + beta*sub(C)
pdgemm_ (NEIN, NEIN, m_global, n_global, k_global,
alpha,
a->local_data(), ia, ja, a->descriptor(),
b->local_data(), ib, jb, b->descriptor(),
beta,
c->local_data(), ic, jc, c->descriptor()
);
double t1 = MPI_Wtime() - t0;
double t_glob;
//获取所有进程所用时间中最长的时间
MPI_Reduce(&t1, &t_glob, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
if (grid->iam() == 0) // 进程号为0的进程
{
double gflops = gemm_flops(m_global, n_global, k_global)/t_glob/grid->nprocs();
printf("\n"
"MATRIX MULTIPLY BENCHMARK SUMMARY\n"
"=================================\n"
"M = %d\tN = %d\tK = %d\tNP = %d\tNP_ROW = %d\tNP_COL = %d\n"
"Time for PxGEMM = %10.7f seconds\tGFlops/Proc = %10.7f\n",
m_global, n_global, k_global, grid->nprocs(), grid->nprows(), grid->npcols(),
t_glob, gflops); fflush(stdout);
for (int i = 0; i < 10; i++)
{
for (int j = 0; j < 10; j++)
{
printf("%f ", c->local_data()[i*k_global + j]);
}
printf("\n");
}
fflush(stdout);
}
}
