diff options
| author |  Gael Guennebaud <g.gael@free.fr> | 2013-02-14 23:34:05 +0100 |
|---|---|---|
| committer | Gael Guennebaud <g.gael@free.fr> | 2013-02-14 23:34:05 +0100 |
| commit | 8745da14d8609425764c399c6ac1f44235277ef3 (patch) | |
| tree | aa26f6fb5b7890e973f8f810d9847d98ed3c1be6 /Eigen/src/Core/arch/SSE/MathFunctions.h | |
| parent | 912ba10efe420ed17dabb6085f258c2ec4bd6789 (diff) | |
Fix SSE plog<float> to return -INF on 0
Diffstat (limited to 'Eigen/src/Core/arch/SSE/MathFunctions.h')
| -rw-r--r-- | Eigen/src/Core/arch/SSE/MathFunctions.h | 10 |
1 files changed, 7 insertions, 3 deletions
diff --git a/Eigen/src/Core/arch/SSE/MathFunctions.h b/Eigen/src/Core/arch/SSE/MathFunctions.h index 557af8455..5ede55fba 100644 --- a/Eigen/src/Core/arch/SSE/MathFunctions.h +++ b/Eigen/src/Core/arch/SSE/MathFunctions.h @@ -31,7 +31,8 @@ Packet4f plog<Packet4f>(const Packet4f& _x) /* the smallest non denormalized float number */ _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(min_norm_pos, 0x00800000); - + _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_inf, 0xff800000);//-1.f/0.f); + /* natural logarithm computed for 4 simultaneous float return NaN for x <= 0 */ @@ -51,7 +52,8 @@ Packet4f plog<Packet4f>(const Packet4f& _x) Packet4i emm0; - Packet4f invalid_mask = _mm_cmple_ps(x, _mm_setzero_ps()); + Packet4f invalid_mask = _mm_cmplt_ps(x, _mm_setzero_ps()); + Packet4f iszero_mask = _mm_cmpeq_ps(x, _mm_setzero_ps()); x = pmax(x, p4f_min_norm_pos); /* cut off denormalized stuff */ emm0 = _mm_srli_epi32(_mm_castps_si128(x), 23); @@ -96,7 +98,9 @@ Packet4f plog<Packet4f>(const Packet4f& _x) y2 = pmul(e, p4f_cephes_log_q2); x = padd(x, y); x = padd(x, y2); - return _mm_or_ps(x, invalid_mask); // negative arg will be NAN + // negative arg will be NAN, 0 will be -INF + return _mm_or_ps(_mm_andnot_ps(iszero_mask, _mm_or_ps(x, invalid_mask)), + _mm_and_ps(iszero_mask, p4f_minus_inf)); } template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED |