dgemm.cc
No OneTemporary
Actions

Subscribers

None

File Metadata

	#include <chrono>
	#include <cmath>
	#include <cstdio>
	#include <omp.h>
	#include <vector>

	using clk = std::chrono::high_resolution_clock;
	using second = std::chrono::duration<double>;
	using time_point = std::chrono::time_point<clk>;

	int main() {
	int N = 2000;
	std::vector<double> A(N * N), B(N * N), C(N * N);

	for (auto &a : A)
	a = 1.0;

	for (auto &b : B)
	b = 2.0;

	for (auto &c : C)
	c = 0.0;

	auto verif = [&](auto &&A) {
	double sum = 0.;
	for (auto &a : A)
	sum += a;
	return sum;
	};

	auto nthreads = omp_get_max_threads();

	auto t1 = clk::now();
	#pragma omp parallel for collapse(1) schedule(static, N / nthreads)
	for (int i = 0; i < N; ++i)
	for (int j = 0; j < N; ++j)
	for (int k = 0; k < N; ++k)
	C[i * N + j] += A[i * N + k] * B[k * N + j];

	second elapsed = clk::now() - t1;

	std::printf("DGEMM with %i threads, collapse(1): %g GFLOP/s (verif %g)\n",
	nthreads, ((2. * N * N * N) / elapsed.count() / 1e8),
	verif(C) / (1. * N * N * N));

	return 0;
	}