118 lines
2.6 KiB
C++
118 lines
2.6 KiB
C++
#include <cmath>
|
|
#include <cfloat>
|
|
#include <vector>
|
|
#include <random>
|
|
#include <chrono>
|
|
#include <functional>
|
|
#include <algorithm>
|
|
|
|
#include <cpuinfo.h>
|
|
#include <nnpack/macros.h>
|
|
#include <nnpack/blas.h>
|
|
#include <nnpack/AlignedAllocator.h>
|
|
|
|
#include <benchmark/benchmark.h>
|
|
|
|
|
|
template<uint32_t mr_, uint32_t nr_>
|
|
class SGEMM : public benchmark::Fixture {
|
|
public:
|
|
inline SGEMM() {
|
|
cpuinfo_initialize();
|
|
const size_t l1d_size = cpuinfo_get_l1d_cache(0)->size;
|
|
const size_t l1d_reserve = 512;
|
|
kc_ = ((l1d_size - l1d_reserve) / sizeof(float) - mr() * nr()) / (mr() + nr());
|
|
}
|
|
|
|
virtual void SetUp(const benchmark::State&) override {
|
|
const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
|
|
auto rng = std::bind(std::uniform_real_distribution<float>(), std::mt19937(seed));
|
|
|
|
a_.resize(mr() * kc());
|
|
std::generate(a_.begin(), a_.end(), std::ref(rng));
|
|
b_.resize(nr() * kc());
|
|
std::generate(b_.begin(), b_.end(), std::ref(rng));
|
|
c_.resize(mr() * nr());
|
|
std::fill(c_.begin(), c_.end(), std::nanf(""));
|
|
}
|
|
|
|
virtual void TearDown(benchmark::State& state) override {
|
|
state.SetItemsProcessed(uint64_t(state.iterations()) * 2 * mr() * nr() * kc());
|
|
a_.clear();
|
|
b_.clear();
|
|
c_.clear();
|
|
}
|
|
|
|
inline const float* a() const {
|
|
return a_.data();
|
|
}
|
|
|
|
inline const float* b() const {
|
|
return b_.data();
|
|
}
|
|
|
|
inline float* c() {
|
|
return c_.data();
|
|
}
|
|
|
|
inline uint32_t mr() const {
|
|
return mr_;
|
|
}
|
|
|
|
inline uint32_t nr() const {
|
|
return nr_;
|
|
}
|
|
|
|
inline uint32_t kc() const {
|
|
return kc_;
|
|
}
|
|
|
|
private:
|
|
std::vector<float, AlignedAllocator<float, 32>> a_;
|
|
std::vector<float, AlignedAllocator<float, 32>> b_;
|
|
std::vector<float> c_;
|
|
uint32_t kc_;
|
|
};
|
|
|
|
#if NNP_BACKEND_X86_64
|
|
BENCHMARK_TEMPLATE_F(SGEMM, fma3, 4, 24)(benchmark::State& state) {
|
|
for (auto _ : state) {
|
|
nnp_sgemm_only_4x24__fma3(kc(), 0, a(), b(), c(), nr());
|
|
}
|
|
}
|
|
#endif
|
|
|
|
#if NNP_BACKEND_ARM && CPUINFO_ARCH_ARM
|
|
BENCHMARK_TEMPLATE_F(SGEMM, aarch32_neon, 6, 8)(benchmark::State& state) {
|
|
for (auto _ : state) {
|
|
nnp_sgemm_only_6x8__aarch32_neon(kc(), 0, a(), b(), c(), nr());
|
|
}
|
|
}
|
|
#endif
|
|
|
|
#if NNP_BACKEND_ARM
|
|
BENCHMARK_TEMPLATE_F(SGEMM, neon, 6, 8)(benchmark::State& state) {
|
|
for (auto _ : state) {
|
|
nnp_sgemm_only_6x8__neon(kc(), 0, a(), b(), c(), nr());
|
|
}
|
|
}
|
|
#endif
|
|
|
|
#if NNP_BACKEND_PSIMD
|
|
BENCHMARK_TEMPLATE_F(SGEMM, psimd, 4, 8)(benchmark::State& state) {
|
|
for (auto _ : state) {
|
|
nnp_sgemm_only_4x8__psimd(kc(), 0, a(), b(), c(), nr());
|
|
}
|
|
}
|
|
#endif
|
|
|
|
#if NNP_BACKEND_SCALAR
|
|
BENCHMARK_TEMPLATE_F(SGEMM, scalar, 4, 3)(benchmark::State& state) {
|
|
for (auto _ : state) {
|
|
nnp_sgemm_only_4x3__scalar(kc(), 0, a(), b(), c(), nr());
|
|
}
|
|
}
|
|
#endif
|
|
|
|
BENCHMARK_MAIN();
|