Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Some speedup with SSE 4.1 #340

Draft
wants to merge 1 commit into
base: develop
Choose a base branch
from
Draft

Some speedup with SSE 4.1 #340

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion cmake/SfizzConfig.cmake
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@ if (CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang")
add_compile_options(-ffast-math)
add_compile_options(-fno-omit-frame-pointer) # For debugging purposes
if (SFIZZ_SYSTEM_PROCESSOR MATCHES "^(i.86|x86_64)$")
add_compile_options(-msse2)
add_compile_options(-msse4.1)
endif()
elseif (CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
set(CMAKE_CXX_STANDARD 17)
Expand Down
12 changes: 2 additions & 10 deletions src/sfizz/Interpolators.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -101,16 +101,8 @@ class Interpolator<kInterpolatorBspline3, float>
{
__m128 x = _mm_sub_ps(_mm_setr_ps(-1, 0, 1, 2), _mm_set1_ps(coeff));
__m128 h = bspline3x4(x);
__m128 y = _mm_mul_ps(h, _mm_loadu_ps(values - 1));
// sum 4 to 1
__m128 xmm0 = y;
__m128 xmm1 = _mm_shuffle_ps(xmm0, xmm0, 0xe5);
__m128 xmm2 = _mm_movehl_ps(xmm0, xmm0);
xmm1 = _mm_add_ss(xmm1, xmm0);
xmm0 = _mm_shuffle_ps(xmm0, xmm0, 0xe7);
xmm2 = _mm_add_ss(xmm2, xmm1);
xmm0 = _mm_add_ss(xmm0, xmm2);
return _mm_cvtss_f32(xmm0);
__m128 y = _mm_dp_ps(h, _mm_loadu_ps(values - 1), 0xf1);
return _mm_cvtss_f32(y);
}
};
#endif
Expand Down
31 changes: 11 additions & 20 deletions src/sfizz/MathHelpers.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@
#include <cfenv>
#if SFIZZ_HAVE_SSE
#include <xmmintrin.h>
#include <immintrin.h>
#endif

template <class T>
Expand Down Expand Up @@ -162,15 +163,13 @@ inline CXX14_CONSTEXPR void incrementAll(T& first, Args&... rest)
template <class R>
R hermite3(R x)
{
x = std::abs(x);
R x2 = x * x;
x = std::abs(x);
R x3 = x2 * x;
R y = 0;
R q = R(5./2.) * x2; // a reoccurring term
R p1 = R(1) - q + R(3./2.) * x3;
R p2 = R(2) - R(4) * x + q - R(1./2.) * x3;
y = (x < R(2)) ? p2 : y;
y = (x < R(1)) ? p1 : y;
R y = (x < R(1)) ? p1 : p2;
return y;
}

Expand All @@ -183,17 +182,14 @@ R hermite3(R x)
*/
inline __m128 hermite3x4(__m128 x)
{
x = _mm_andnot_ps(_mm_set1_ps(-0.0f), x);
__m128 x2 = _mm_mul_ps(x, x);
x = _mm_andnot_ps(_mm_set1_ps(-0.0f), x);
__m128 x3 = _mm_mul_ps(x2, x);
__m128 y = _mm_set1_ps(0.0f);
__m128 q = _mm_mul_ps(_mm_set1_ps(5./2.), x2);
__m128 p1 = _mm_add_ps(_mm_sub_ps(_mm_set1_ps(1), q), _mm_mul_ps(_mm_set1_ps(3./2.), x3));
__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));
__m128 m2 = _mm_cmple_ps(x, _mm_set1_ps(2));
y = _mm_or_ps(_mm_and_ps(m2, p2), _mm_andnot_ps(m2, y));
__m128 m1 = _mm_cmple_ps(x, _mm_set1_ps(1));
y = _mm_or_ps(_mm_and_ps(m1, p1), _mm_andnot_ps(m1, y));
__m128 m1 = _mm_cmplt_ps(x, _mm_set1_ps(1));
__m128 y = _mm_or_ps(_mm_and_ps(m1, p1), _mm_andnot_ps(m1, p2));
return y;
}
#endif
Expand All @@ -208,14 +204,12 @@ inline __m128 hermite3x4(__m128 x)
template <class R>
R bspline3(R x)
{
x = std::abs(x);
R x2 = x * x;
x = std::abs(x);
R x3 = x2 * x;
R y = 0;
R p1 = R(2./3.) - x2 + R(1./2.) * x3;
R p2 = R(4./3.) - R(2) * x + x2 - R(1./6.) * x3;
y = (x < R(2)) ? p2 : y;
y = (x < R(1)) ? p1 : y;
R y = (x < R(1)) ? p1 : p2;
return y;
}

Expand All @@ -228,16 +222,13 @@ R bspline3(R x)
*/
inline __m128 bspline3x4(__m128 x)
{
x = _mm_andnot_ps(_mm_set1_ps(-0.0f), x);
__m128 x2 = _mm_mul_ps(x, x);
x = _mm_andnot_ps(_mm_set1_ps(-0.0f), x);
__m128 x3 = _mm_mul_ps(x2, x);
__m128 y = _mm_set1_ps(0.0f);
__m128 p1 = _mm_add_ps(_mm_sub_ps(_mm_set1_ps(2./3.), x2), _mm_mul_ps(_mm_set1_ps(1./2.), x3));
__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));
__m128 m2 = _mm_cmple_ps(x, _mm_set1_ps(2));
y = _mm_or_ps(_mm_and_ps(m2, p2), _mm_andnot_ps(m2, y));
__m128 m1 = _mm_cmple_ps(x, _mm_set1_ps(1));
y = _mm_or_ps(_mm_and_ps(m1, p1), _mm_andnot_ps(m1, y));
__m128 m1 = _mm_cmplt_ps(x, _mm_set1_ps(1));
__m128 y = _mm_or_ps(_mm_and_ps(m1, p1), _mm_andnot_ps(m1, p2));
return y;
}
#endif
Expand Down