mirror of https://github.com/QMCPACK/qmcpack.git
126 lines
3.1 KiB
C++
126 lines
3.1 KiB
C++
// -*-indent-tabs-mode:t;c-basic-offset:4;tab-width:4-*-
|
|
// © Alfredo A. Correa 2019-2021
|
|
|
|
#include <benchmark/benchmark.h>
|
|
|
|
#include "../array.hpp"
|
|
#include "../algorithms/gemm.hpp"
|
|
#include "../adaptors/blas/gemm.hpp"
|
|
|
|
namespace multi = boost::multi;
|
|
|
|
template<class MatrixA, class MatrixB, class MatrixC>
|
|
auto naive_product(MatrixA const& A, MatrixB const& B, double beta, MatrixC&& C) -> MatrixC&& {
|
|
assert( C .size() == A. size() );
|
|
assert( (~C).size() == (~B).size() );
|
|
for(auto i : extension(C)) {
|
|
for(auto j : extension(~C)) {
|
|
C[i][j] = std::inner_product(begin(A[i]), end(A[i]), begin((~B)[j]), beta*C[i][j]);
|
|
}
|
|
}
|
|
return std::forward<MatrixC>(C);
|
|
}
|
|
|
|
auto const N = 1024;
|
|
|
|
static void Bnaive_product(benchmark::State& _) {
|
|
auto const A = []{
|
|
multi::array<double, 2> A({ N, N} );
|
|
std::iota(begin(A.elements()), end(A.elements()), 0.1);
|
|
return A;
|
|
}();
|
|
|
|
auto const B = [&]{
|
|
multi::array<double, 2> B({(~A).size(), N} );
|
|
std::iota(begin(B.elements()), end(B.elements()), 0.2);
|
|
return B;
|
|
}();
|
|
|
|
multi::array<double, 2> C({ A.size() , (~B).size()}, 0.);
|
|
|
|
benchmark::ClobberMemory();
|
|
while(_.KeepRunning()) {
|
|
naive_product(A, B, 0., C);
|
|
benchmark::DoNotOptimize(C.data_elements());
|
|
benchmark::ClobberMemory();
|
|
}
|
|
}
|
|
|
|
static void multi_detail_naive_gemm(benchmark::State& state) {
|
|
auto const A = []{
|
|
multi::array<double, 2> A({ N, N} );
|
|
std::iota(begin(A.elements()), end(A.elements()), 0.1);
|
|
return A;
|
|
}();
|
|
|
|
auto const B = [&]{
|
|
multi::array<double, 2> B({(~A).size(), N} );
|
|
std::iota(begin(B.elements()), end(B.elements()), 0.2);
|
|
return B;
|
|
}();
|
|
|
|
multi::array<double, 2> C({ A.size() , (~B).size()}, 0.);
|
|
|
|
benchmark::ClobberMemory();
|
|
for(auto _ : state) {
|
|
multi::detail::naive_gemm(1., A, B, 0., C);
|
|
benchmark::DoNotOptimize(C.data_elements());
|
|
benchmark::ClobberMemory();
|
|
}
|
|
}
|
|
|
|
static void multi_gemm(benchmark::State& state) {
|
|
auto const A = []{
|
|
multi::array<double, 2> A({ N, N} );
|
|
std::iota(begin(A.elements()), end(A.elements()), 0.1);
|
|
return A;
|
|
}();
|
|
|
|
auto const B = [&]{
|
|
multi::array<double, 2> B({(~A).size(), N} );
|
|
std::iota(begin(B.elements()), end(B.elements()), 0.2);
|
|
return B;
|
|
}();
|
|
|
|
multi::array<double, 2> C({ A.size() , (~B).size()}, 0.);
|
|
|
|
benchmark::ClobberMemory();
|
|
for(auto _ : state) {
|
|
multi::gemm(1., A, B, 0., C);
|
|
benchmark::DoNotOptimize(C);
|
|
benchmark::ClobberMemory();
|
|
}
|
|
}
|
|
|
|
static void multi_blas_gemm(benchmark::State& state) {
|
|
auto const A = [] {
|
|
multi::array<double, 2> A({ N, N} );
|
|
std::iota(begin(A.elements()), end(A.elements()), 0.1);
|
|
return A;
|
|
}();
|
|
|
|
auto const B = [&] {
|
|
multi::array<double, 2> B({(~A).size(), N} );
|
|
std::iota(begin(B.elements()), end(B.elements()), 0.2);
|
|
return B;
|
|
}();
|
|
|
|
multi::array<double, 2> C({A.size() , (~B).size()}, 0.);
|
|
|
|
benchmark::ClobberMemory();
|
|
for(auto _ : state) {
|
|
multi::blas::gemm(1., A, B, 0., C);
|
|
benchmark::DoNotOptimize(C);
|
|
benchmark::ClobberMemory();
|
|
}
|
|
}
|
|
|
|
BENCHMARK(Bnaive_product);
|
|
|
|
BENCHMARK(multi_detail_naive_gemm);
|
|
BENCHMARK(multi_gemm);
|
|
|
|
BENCHMARK(multi_blas_gemm);
|
|
|
|
BENCHMARK_MAIN();
|