sam/sfizz
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Added a "scalar add" SIMD helper

Browse source
This commit is contained in:
paulfd 2019-09-15 20:16:42 +02:00
parent 2aa086d971
commit 9324d137bc
3 changed files with 56 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
benchmarks/BM_add.cpp
View file

@ -49,6 +49,33 @@ public:
std::vector<float> output;
};
BENCHMARK_DEFINE_F(AddArray, Value_Scalar)(benchmark::State& state) {
for (auto _ : state)
{
add<float, false>(1.1f, absl::MakeSpan(output));
}
}
BENCHMARK_DEFINE_F(AddArray, Value_SIMD)(benchmark::State& state) {
for (auto _ : state)
{
add<float, true>(1.1f, absl::MakeSpan(output));
}
}
BENCHMARK_DEFINE_F(AddArray, Value_Scalar_Unaligned)(benchmark::State& state) {
for (auto _ : state)
{
add<float, false>(1.1f, absl::MakeSpan(output).subspan(1));
}
}
BENCHMARK_DEFINE_F(AddArray, Value_SIMD_Unaligned)(benchmark::State& state) {
for (auto _ : state)
{
add<float, true>(1.1f, absl::MakeSpan(output).subspan(1));
}
}
BENCHMARK_DEFINE_F(AddArray, Scalar)(benchmark::State& state) {
for (auto _ : state)
@ -78,6 +105,10 @@ BENCHMARK_DEFINE_F(AddArray, SIMD_Unaligned)(benchmark::State& state) {
}
}
BENCHMARK_REGISTER_F(AddArray, Value_Scalar)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(AddArray, Value_SIMD)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(AddArray, Value_Scalar_Unaligned)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(AddArray, Value_SIMD_Unaligned)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(AddArray, Scalar)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(AddArray, SIMD)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
BENCHMARK_REGISTER_F(AddArray, Scalar_Unaligned)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);

6
sfizz/SIMDDummy.cpp
View file

@ -114,6 +114,12 @@ void add<float, true>(absl::Span<const float> input, absl::Span<float> output) n
add<float, false>(input, output);
}
template <>
void add<float, true>(float value, absl::Span<float> output) noexcept
{
add<float, false>(value, output);
}
template <>
void subtract<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept
{

19
sfizz/SIMDHelpers.h
View file

@ -349,6 +349,25 @@ void add(absl::Span<const T> input, absl::Span<T> output) noexcept
template <>
void add<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept;
template <class T>
inline void snippetAdd(const T value, T*& output)
{
*output++ += value;
}
template <class T, bool SIMD = SIMDConfig::add>
void add(T value, absl::Span<T> output) noexcept
{
auto* out = output.begin();
auto* sentinel = output.end();
while (out < sentinel)
snippetAdd(value, out);
}
template <>
void add<float, true>(float value, absl::Span<float> output) noexcept;
template <class T>
inline void snippetSubtract(const T*& input, T*& output)
{