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

Convert the math helpers to simde

Browse source
This commit is contained in:
Jean Pierre Cimalando 2021-02-16 02:31:22 +01:00
parent cb9cc622d0
commit 36d99ec1f3
2 changed files with 35 additions and 30 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/CMakeLists.txt
View file

@ -217,7 +217,7 @@ target_sources(sfizz_parser PRIVATE
${SFIZZ_PARSER_HEADERS} ${SFIZZ_PARSER_SOURCES} ${SFIZZ_PARSER_OTHER}) ${SFIZZ_PARSER_HEADERS} ${SFIZZ_PARSER_SOURCES} ${SFIZZ_PARSER_OTHER})
target_include_directories(sfizz_parser PUBLIC sfizz) target_include_directories(sfizz_parser PUBLIC sfizz)
target_link_libraries(sfizz_parser target_link_libraries(sfizz_parser
PUBLIC sfizz::filesystem absl::strings PUBLIC sfizz::filesystem sfizz::simde absl::strings
PRIVATE absl::flat_hash_map) PRIVATE absl::flat_hash_map)
# OSC messaging library # OSC messaging library

63
src/sfizz/MathHelpers.h
View file

@ -21,8 +21,9 @@
#include <type_traits> #include <type_traits>
#include <cmath> #include <cmath>
#include <cfenv> #include <cfenv>
#if SFIZZ_HAVE_SSE #include <simde/simde-features.h>
#include <xmmintrin.h> #if SIMDE_NATURAL_VECTOR_SIZE_GE(128)
#include <simde/x86/sse.h>
#endif #endif
#if __cplusplus >= 201703L #if __cplusplus >= 201703L
@ -195,26 +196,28 @@ R hermite3(R x)
return y; return y;
} }
#if SFIZZ_HAVE_SSE #if SIMDE_NATURAL_VECTOR_SIZE_GE(128)
/** /**
* @brief Compute 4 parallel elements of the 3rd-order Hermite interpolation polynomial. * @brief Compute 4 parallel elements of the 3rd-order Hermite interpolation polynomial.
* *
* @param x * @param x
* @return __m128 * @return simde__m128
*/ */
inline __m128 hermite3x4(__m128 x) inline simde__m128 hermite3x4(simde__m128 x)
{ {
x = _mm_andnot_ps(_mm_set1_ps(-0.0f), x); // Note(jpc) replace with `simde_x_mm_abs_ps` when fixed
__m128 x2 = _mm_mul_ps(x, x); // https://github.com/simd-everywhere/simde/issues/704
__m128 x3 = _mm_mul_ps(x2, x); x = simde_mm_andnot_ps(simde_mm_set1_ps(-0.0f), x);
__m128 y = _mm_set1_ps(0.0f); simde__m128 x2 = simde_mm_mul_ps(x, x);
__m128 q = _mm_mul_ps(_mm_set1_ps(5./2.), x2); simde__m128 x3 = simde_mm_mul_ps(x2, x);
__m128 p1 = _mm_add_ps(_mm_sub_ps(_mm_set1_ps(1), q), _mm_mul_ps(_mm_set1_ps(3./2.), x3)); simde__m128 y = simde_mm_set1_ps(0.0f);
__m128 p2 = _mm_sub_ps(_mm_add_ps(_mm_sub_ps(_mm_set1_ps(2), _mm_mul_ps(_mm_set1_ps(4), x)), q), _mm_mul_ps(_mm_set1_ps(1./2.), x3)); simde__m128 q = simde_mm_mul_ps(simde_mm_set1_ps(5./2.), x2);
__m128 m2 = _mm_cmple_ps(x, _mm_set1_ps(2)); simde__m128 p1 = simde_mm_add_ps(simde_mm_sub_ps(simde_mm_set1_ps(1), q), simde_mm_mul_ps(simde_mm_set1_ps(3./2.), x3));
y = _mm_or_ps(_mm_and_ps(m2, p2), _mm_andnot_ps(m2, y)); simde__m128 p2 = simde_mm_sub_ps(simde_mm_add_ps(simde_mm_sub_ps(simde_mm_set1_ps(2), simde_mm_mul_ps(simde_mm_set1_ps(4), x)), q), simde_mm_mul_ps(simde_mm_set1_ps(1./2.), x3));
__m128 m1 = _mm_cmple_ps(x, _mm_set1_ps(1)); simde__m128 m2 = simde_mm_cmple_ps(x, simde_mm_set1_ps(2));
y = _mm_or_ps(_mm_and_ps(m1, p1), _mm_andnot_ps(m1, y)); y = simde_mm_or_ps(simde_mm_and_ps(m2, p2), simde_mm_andnot_ps(m2, y));
simde__m128 m1 = simde_mm_cmple_ps(x, simde_mm_set1_ps(1));
y = simde_mm_or_ps(simde_mm_and_ps(m1, p1), simde_mm_andnot_ps(m1, y));
return y; return y;
} }
#endif #endif
@ -240,25 +243,27 @@ R bspline3(R x)
return y; return y;
} }
#if SFIZZ_HAVE_SSE #if SIMDE_NATURAL_VECTOR_SIZE_GE(128)
/** /**
* @brief Compute 4 parallel elements of the 3rd-order B-spline interpolation polynomial. * @brief Compute 4 parallel elements of the 3rd-order B-spline interpolation polynomial.
* *
* @param x * @param x
* @return __m128 * @return simde__m128
*/ */
inline __m128 bspline3x4(__m128 x) inline simde__m128 bspline3x4(simde__m128 x)
{ {
x = _mm_andnot_ps(_mm_set1_ps(-0.0f), x); // Note(jpc) replace with `simde_x_mm_abs_ps` when fixed
__m128 x2 = _mm_mul_ps(x, x); // https://github.com/simd-everywhere/simde/issues/704
__m128 x3 = _mm_mul_ps(x2, x); x = simde_mm_andnot_ps(simde_mm_set1_ps(-0.0f), x);
__m128 y = _mm_set1_ps(0.0f); simde__m128 x2 = simde_mm_mul_ps(x, x);
__m128 p1 = _mm_add_ps(_mm_sub_ps(_mm_set1_ps(2./3.), x2), _mm_mul_ps(_mm_set1_ps(1./2.), x3)); simde__m128 x3 = simde_mm_mul_ps(x2, x);
__m128 p2 = _mm_sub_ps(_mm_add_ps(_mm_sub_ps(_mm_set1_ps(4./3.), _mm_mul_ps(_mm_set1_ps(2), x)), x2), _mm_mul_ps(_mm_set1_ps(1./6.), x3)); simde__m128 y = simde_mm_set1_ps(0.0f);
__m128 m2 = _mm_cmple_ps(x, _mm_set1_ps(2)); simde__m128 p1 = simde_mm_add_ps(simde_mm_sub_ps(simde_mm_set1_ps(2./3.), x2), simde_mm_mul_ps(simde_mm_set1_ps(1./2.), x3));
y = _mm_or_ps(_mm_and_ps(m2, p2), _mm_andnot_ps(m2, y)); simde__m128 p2 = simde_mm_sub_ps(simde_mm_add_ps(simde_mm_sub_ps(simde_mm_set1_ps(4./3.), simde_mm_mul_ps(simde_mm_set1_ps(2), x)), x2), simde_mm_mul_ps(simde_mm_set1_ps(1./6.), x3));
__m128 m1 = _mm_cmple_ps(x, _mm_set1_ps(1)); simde__m128 m2 = simde_mm_cmple_ps(x, simde_mm_set1_ps(2));
y = _mm_or_ps(_mm_and_ps(m1, p1), _mm_andnot_ps(m1, y)); y = simde_mm_or_ps(simde_mm_and_ps(m2, p2), simde_mm_andnot_ps(m2, y));
simde__m128 m1 = simde_mm_cmple_ps(x, simde_mm_set1_ps(1));
y = simde_mm_or_ps(simde_mm_and_ps(m1, p1), simde_mm_andnot_ps(m1, y));
return y; return y;
} }
#endif #endif