diff options
Diffstat (limited to 'Eigen/src/Core/arch/SSE/MathFunctions.h')
| -rw-r--r-- | Eigen/src/Core/arch/SSE/MathFunctions.h | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/Eigen/src/Core/arch/SSE/MathFunctions.h b/Eigen/src/Core/arch/SSE/MathFunctions.h index 8f78b3a6c..9ffba5b41 100644 --- a/Eigen/src/Core/arch/SSE/MathFunctions.h +++ b/Eigen/src/Core/arch/SSE/MathFunctions.h @@ -52,7 +52,7 @@ Packet4f plog<Packet4f>(const Packet4f& _x) Packet4i emm0; - Packet4f invalid_mask = _mm_cmplt_ps(x, _mm_setzero_ps()); + Packet4f invalid_mask = _mm_cmpnge_ps(x, _mm_setzero_ps()); // not greater equal is true if x is NaN Packet4f iszero_mask = _mm_cmpeq_ps(x, _mm_setzero_ps()); x = pmax(x, p4f_min_norm_pos); /* cut off denormalized stuff */ @@ -167,7 +167,7 @@ Packet4f pexp<Packet4f>(const Packet4f& _x) emm0 = _mm_cvttps_epi32(fx); emm0 = _mm_add_epi32(emm0, p4i_0x7f); emm0 = _mm_slli_epi32(emm0, 23); - return pmul(y, Packet4f(_mm_castsi128_ps(emm0))); + return pmax(pmul(y, Packet4f(_mm_castsi128_ps(emm0))), _x); } template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet2d pexp<Packet2d>(const Packet2d& _x) @@ -241,7 +241,7 @@ Packet2d pexp<Packet2d>(const Packet2d& _x) emm0 = _mm_add_epi32(emm0, p4i_1023_0); emm0 = _mm_slli_epi32(emm0, 20); emm0 = _mm_shuffle_epi32(emm0, _MM_SHUFFLE(1,2,0,3)); - return pmul(x, Packet2d(_mm_castsi128_pd(emm0))); + return pmax(pmul(x, Packet2d(_mm_castsi128_pd(emm0))), _x); } /* evaluation of 4 sines at onces, using SSE2 intrinsics. |