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

Added a copy helper

Browse source
This commit is contained in:
paulfd 2019-09-06 22:15:19 +02:00
parent fad715559f
commit a5212fd2d6
8 changed files with 86 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
benchmarks/CMakeLists.txt
View file

@ -57,6 +57,9 @@ target_link_libraries(bm_ADSR benchmark absl::span absl::algorithm)
add_executable(bm_add BM_add.cpp ${SFIZZ_SIMD_SOURCES})
target_link_libraries(bm_add benchmark absl::span absl::algorithm)
add_executable(bm_copy BM_copy.cpp ${SFIZZ_SIMD_SOURCES})
target_link_libraries(bm_copy benchmark absl::span absl::algorithm)
add_custom_target(sfizz_benchmarks)
add_dependencies(sfizz_benchmarks
bm_opf_high_vs_low

2
external/Catch2 vendored

@ -1 +1 @@
Subproject commit ce42deb72fab2be85a862f559984580c24cb76c4
Subproject commit b77ab74b723cbd55ee12883562e22c7e467dbcb1

2
external/abseil-cpp vendored

@ -1 +1 @@
Subproject commit 83c1d65c90a92aa49632b9ac5a793214bb0768bc
Subproject commit 325fd7b042ff4ec34f7dd32e602cd81ad0e24b22

1
sfizz/Config.h
View file

@ -54,4 +54,5 @@ namespace SIMDConfig {
constexpr bool linearRamp { false };
constexpr bool multiplicativeRamp { true };
constexpr bool add { false };
constexpr bool copy { false };
}

6
sfizz/SIMDDummy.cpp
View file

@ -106,4 +106,10 @@ template <>
void add<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept
{
add<float, false>(input, output);
}
template <>
void copy<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept
{
copy<float, false>(input, output);
}

23
sfizz/SIMDHelpers.h
View file

@ -325,4 +325,25 @@ 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;
void add<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept;
template <class T>
void snippetCopy(const T*& input, T*& output)
{
*output++ = *input++;
}
template <class T, bool SIMD = SIMDConfig::copy>
void copy(absl::Span<const T> input, absl::Span<T> output) noexcept
{
ASSERT(output.size() >= input.size());
auto* in = input.begin();
auto* out = output.begin();
auto* sentinel = out + min(input.size(), output.size());
while (out < sentinel)
snippetCopy(in, out);
}
template <>
void copy<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept;

22
sfizz/SIMDSSE.cpp
View file

@ -494,4 +494,26 @@ void add<float, true>(absl::Span<const float> input, absl::Span<float> output) n
while (out < sentinel)
snippetAdd<float>(in, out);
}
template <>
void copy<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept
{
ASSERT(output.size() >= input.size());
auto* in = input.begin();
auto* out = output.begin();
auto* sentinel = out + min(input.size(), output.size());
const auto* lastAligned = prevAligned(sentinel);
while (unaligned(in, out) && out < lastAligned)
snippetCopy<float>(in, out);
while (out < lastAligned) {
_mm_store_ps(out, _mm_load_ps(in));
out += TypeAlignment;
in += TypeAlignment;
}
while (out < sentinel)
snippetCopy<float>(in, out);
}

30
tests/SIMDHelpersT.cpp
View file

@ -622,3 +622,33 @@ TEST_CASE("[Helpers] Add (SIMD vs scalar)")
add<float, true>(input, absl::MakeSpan(outputSIMD));
REQUIRE(approxEqual<float>(outputScalar, outputSIMD));
}
TEST_CASE("[Helpers] copy")
{
std::array<float, 5> input { 1.0f, 2.0f, 3.0f, 4.0f, 5.0f };
std::array<float, 5> output { 1.0f, 1.0f, 1.0f, 1.0f, 1.0f };
copy<float, false>(input, absl::MakeSpan(output));
REQUIRE(output == input);
}
TEST_CASE("[Helpers] copy (SIMD)")
{
std::array<float, 5> input { 1.0f, 2.0f, 3.0f, 4.0f, 5.0f };
std::array<float, 5> output { 1.0f, 1.0f, 1.0f, 1.0f, 1.0f };
copy<float, true>(input, absl::MakeSpan(output));
REQUIRE(output == input);
}
TEST_CASE("[Helpers] copy (SIMD vs scalar)")
{
std::vector<float> input(bigBufferSize);
std::vector<float> outputScalar(bigBufferSize);
std::vector<float> outputSIMD(bigBufferSize);
absl::c_iota(input, 0.0);
absl::c_fill(outputScalar, 0.0);
absl::c_fill(outputSIMD, 0.0);
add<float, false>(input, absl::MakeSpan(outputScalar));
add<float, true>(input, absl::MakeSpan(outputSIMD));
REQUIRE(approxEqual<float>(outputScalar, outputSIMD));
}