sfizz/benchmarks/BM_multiplyAdd.cpp
2020-06-19 20:48:08 +02:00

83 lines
2.9 KiB
C++

// SPDX-License-Identifier: BSD-2-Clause
// This code is part of the sfizz library and is licensed under a BSD 2-clause
// license. You should have receive a LICENSE.md file along with the code.
// If not, contact the sfizz maintainers at https://github.com/sfztools/sfizz
#include "SIMDHelpers.h"
#include <benchmark/benchmark.h>
#include <random>
#include <numeric>
#include <vector>
#include <cmath>
#include <iostream>
class MultiplyAdd : public benchmark::Fixture {
public:
void SetUp(const ::benchmark::State& state) {
std::random_device rd { };
std::mt19937 gen { rd() };
std::uniform_real_distribution<float> dist { 0, 1 };
input = std::vector<float>(state.range(0));
output = std::vector<float>(state.range(0));
gain = std::vector<float>(state.range(0));
std::fill(output.begin(), output.end(), 1.0f );
std::generate(gain.begin(), gain.end(), [&]() { return dist(gen); });
std::generate(input.begin(), input.end(), [&]() { return dist(gen); });
}
void TearDown(const ::benchmark::State& /* state */) {
}
std::vector<float> gain;
std::vector<float> input;
std::vector<float> output;
};
BENCHMARK_DEFINE_F(MultiplyAdd, Straight)(benchmark::State& state) {
for (auto _ : state)
{
for (int i = 0; i < state.range(0); ++i)
output[i] += gain[i] * input[i];
}
}
BENCHMARK_DEFINE_F(MultiplyAdd, Scalar)(benchmark::State& state) {
for (auto _ : state)
{
sfz::setSIMDOpStatus(sfz::SIMDOps::multiplyAdd, false);
sfz::multiplyAdd<float>(gain, input, absl::MakeSpan(output));
}
}
BENCHMARK_DEFINE_F(MultiplyAdd, SIMD)(benchmark::State& state) {
for (auto _ : state)
{
sfz::setSIMDOpStatus(sfz::SIMDOps::multiplyAdd, true);
sfz::multiplyAdd<float>(gain, input, absl::MakeSpan(output));
}
}
BENCHMARK_DEFINE_F(MultiplyAdd, Scalar_Unaligned)(benchmark::State& state) {
for (auto _ : state)
{
sfz::setSIMDOpStatus(sfz::SIMDOps::multiplyAdd, false);
sfz::multiplyAdd<float>(absl::MakeSpan(gain).subspan(1), absl::MakeSpan(input).subspan(1), absl::MakeSpan(output).subspan(1));
}
}
BENCHMARK_DEFINE_F(MultiplyAdd, SIMD_Unaligned)(benchmark::State& state) {
for (auto _ : state)
{
sfz::setSIMDOpStatus(sfz::SIMDOps::multiplyAdd, true);
sfz::multiplyAdd<float>(absl::MakeSpan(gain).subspan(1), absl::MakeSpan(input).subspan(1), absl::MakeSpan(output).subspan(1));
}
}
BENCHMARK_REGISTER_F(MultiplyAdd, Straight)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(MultiplyAdd, Scalar)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(MultiplyAdd, SIMD)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(MultiplyAdd, Scalar_Unaligned)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(MultiplyAdd, SIMD_Unaligned)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_MAIN();