mirror/JUCE
1
0
Fork 0
mirror of https://github.com/juce-framework/JUCE.git synced 2026-01-10 23:44:24 +00:00
Code Activity

DSP: Guarded a rogue FMA instruction in the AVX native ops

Browse source
This commit is contained in:
Tom Poole 2019-06-11 21:23:56 +01:00
parent dab2c9be79
commit 01935e3338
1 changed files with 9 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

10
modules/juce_dsp/native/juce_avx_SIMDNativeOps.h
View file

@ -91,7 +91,6 @@ struct SIMDNativeOps<float>
static forcedinline __m256 JUCE_VECTOR_CALLTYPE greaterThan (__m256 a, __m256 b) noexcept { return _mm256_cmp_ps (a, b, _CMP_GT_OQ); }
static forcedinline __m256 JUCE_VECTOR_CALLTYPE greaterThanOrEqual (__m256 a, __m256 b) noexcept { return _mm256_cmp_ps (a, b, _CMP_GE_OQ); }
static forcedinline bool JUCE_VECTOR_CALLTYPE allEqual (__m256 a, __m256 b) noexcept { return (_mm256_movemask_ps (equal (a, b)) == 0xff); }
static forcedinline __m256 JUCE_VECTOR_CALLTYPE multiplyAdd (__m256 a, __m256 b, __m256 c) noexcept { return _mm256_fmadd_ps (b, c, a); }
static forcedinline __m256 JUCE_VECTOR_CALLTYPE dupeven (__m256 a) noexcept { return _mm256_shuffle_ps (a, a, _MM_SHUFFLE (2, 2, 0, 0)); }
static forcedinline __m256 JUCE_VECTOR_CALLTYPE dupodd (__m256 a) noexcept { return _mm256_shuffle_ps (a, a, _MM_SHUFFLE (3, 3, 1, 1)); }
static forcedinline __m256 JUCE_VECTOR_CALLTYPE swapevenodd (__m256 a) noexcept { return _mm256_shuffle_ps (a, a, _MM_SHUFFLE (2, 3, 0, 1)); }
@ -99,6 +98,15 @@ struct SIMDNativeOps<float>
static forcedinline __m256 JUCE_VECTOR_CALLTYPE set (__m256 v, size_t i, float s) noexcept { return SIMDFallbackOps<float, __m256>::set (v, i, s); }
static forcedinline __m256 JUCE_VECTOR_CALLTYPE truncate (__m256 a) noexcept { return _mm256_cvtepi32_ps (_mm256_cvttps_epi32 (a)); }
static forcedinline __m256 JUCE_VECTOR_CALLTYPE multiplyAdd (__m256 a, __m256 b, __m256 c) noexcept
{
#if __FMA__
return _mm256_fmadd_ps (b, c, a);
#else
return add (a, mul (b, c));
#endif
}
static forcedinline __m256 JUCE_VECTOR_CALLTYPE oddevensum (__m256 a) noexcept
{
a = _mm256_add_ps (_mm256_shuffle_ps (a, a, _MM_SHUFFLE (1, 0, 3, 2)), a);