aboutsummaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
authorGravatar Rasmus Munk Larsen <rmlarsen@google.com>2019-01-09 16:29:18 -0800
committerGravatar Rasmus Munk Larsen <rmlarsen@google.com>2019-01-09 16:29:18 -0800
commitf2767112c88762ddc62e8c066dd3377a3d89da31 (patch)
tree51595b6651df9fc5dd6c532e2d0a2578f9726b80
parentcb955df9a6fd5cb2673a7a15172609ce2dafdde8 (diff)
Simplify a bit.
-rw-r--r--Eigen/src/Core/arch/AVX/Complex.h6
-rw-r--r--Eigen/src/Core/arch/AVX512/Complex.h10
-rw-r--r--Eigen/src/Core/arch/SSE/Complex.h6
3 files changed, 6 insertions, 16 deletions
diff --git a/Eigen/src/Core/arch/AVX/Complex.h b/Eigen/src/Core/arch/AVX/Complex.h
index d880ef593..9f1bb969e 100644
--- a/Eigen/src/Core/arch/AVX/Complex.h
+++ b/Eigen/src/Core/arch/AVX/Complex.h
@@ -72,8 +72,7 @@ template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, con
template <>
EIGEN_STRONG_INLINE Packet4cf pcmp_eq(const Packet4cf& a, const Packet4cf& b) {
__m256 eq = _mm256_cmp_ps(a.v, b.v, _CMP_EQ_OQ);
- __m256 real_and_imag_equal = _mm256_and_ps(eq, _mm256_permute_ps(eq, 0xb1));
- return Packet4cf(real_and_imag_equal);
+ return Packet4cf(_mm256_and_ps(eq, _mm256_permute_ps(eq, 0xb1)));
}
template<> EIGEN_STRONG_INLINE Packet4cf pones<Packet4cf>(const Packet4cf& a) { return Packet4cf(pones(a.v)); }
@@ -288,8 +287,7 @@ template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, con
template <>
EIGEN_STRONG_INLINE Packet2cd pcmp_eq(const Packet2cd& a, const Packet2cd& b) {
__m256d eq = _mm256_cmp_pd(a.v, b.v, _CMP_EQ_OQ);
- __m256d real_and_imag_equal = _mm256_and_pd(eq, _mm256_permute_pd(eq, 0x5));
- return Packet2cd(real_and_imag_equal);
+ return Packet2cd(pand(eq, _mm256_permute_pd(eq, 0x5)));
}
template<> EIGEN_STRONG_INLINE Packet2cd pones<Packet2cd>(const Packet2cd& a) { return Packet2cd(pones(a.v)); }
diff --git a/Eigen/src/Core/arch/AVX512/Complex.h b/Eigen/src/Core/arch/AVX512/Complex.h
index 2c613f870..154fedc25 100644
--- a/Eigen/src/Core/arch/AVX512/Complex.h
+++ b/Eigen/src/Core/arch/AVX512/Complex.h
@@ -83,10 +83,7 @@ template<> EIGEN_STRONG_INLINE Packet8cf pandnot<Packet8cf>(const Packet8cf& a,
template <>
EIGEN_STRONG_INLINE Packet8cf pcmp_eq(const Packet8cf& a, const Packet8cf& b) {
__m512 eq = pcmp_eq<Packet16f>(a.v, b.v);
- __m512 eq_swap_real_imag = _mm512_permute_ps(eq, 0xB1);
- __m512i real_and_imag_equal = _mm512_and_si512(
- _mm512_castps_si512(eq), _mm512_castps_si512(eq_swap_real_imag));
- return Packet8cf(_mm512_castsi512_ps(real_and_imag_equal));
+ return Packet8cf(pand(eq, _mm512_permute_ps(eq, 0xB1)));
}
template<> EIGEN_STRONG_INLINE Packet8cf pload <Packet8cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet8cf(pload<Packet16f>(&numext::real_ref(*from))); }
@@ -279,10 +276,7 @@ template<> EIGEN_STRONG_INLINE Packet4cd pandnot<Packet4cd>(const Packet4cd& a,
template <>
EIGEN_STRONG_INLINE Packet4cd pcmp_eq(const Packet4cd& a, const Packet4cd& b) {
__m512d eq = pcmp_eq<Packet8d>(a.v, b.v);
- __m512d eq_swap_real_imag = _mm512_permute_pd(eq, 0x55);
- __m512i real_and_imag_equal = _mm512_and_si512(
- _mm512_castpd_si512(eq), _mm512_castpd_si512(eq_swap_real_imag));
- return Packet4cd(_mm512_castsi512_pd(real_and_imag_equal));
+ return Packet4cd(pand(eq, _mm512_permute_pd(eq, 0x55)));
}
template<> EIGEN_STRONG_INLINE Packet4cd pload <Packet4cd>(const std::complex<double>* from)
diff --git a/Eigen/src/Core/arch/SSE/Complex.h b/Eigen/src/Core/arch/SSE/Complex.h
index 8372cedfb..875cb09e0 100644
--- a/Eigen/src/Core/arch/SSE/Complex.h
+++ b/Eigen/src/Core/arch/SSE/Complex.h
@@ -447,15 +447,13 @@ ptranspose(PacketBlock<Packet2cf,2>& kernel) {
template<> EIGEN_STRONG_INLINE Packet2cf pcmp_eq(const Packet2cf& a, const Packet2cf& b)
{
__m128 eq = _mm_cmpeq_ps(a.v, b.v);
- __m128 real_and_imag_equal = _mm_and_ps(eq, vec4f_swizzle1(eq, 1, 0, 3, 2));
- return Packet2cf(real_and_imag_equal);
+ return Packet2cf(pand(eq, vec4f_swizzle1(eq, 1, 0, 3, 2)));
}
template<> EIGEN_STRONG_INLINE Packet1cd pcmp_eq(const Packet1cd& a, const Packet1cd& b)
{
__m128d eq = _mm_cmpeq_pd(a.v, b.v);
- __m128d real_and_imag_equal = _mm_and_pd(eq, vec2d_swizzle1(eq, 1, 0));
- return Packet1cd(real_and_imag_equal);
+ return Packet1cd(pand(eq, vec2d_swizzle1(eq, 1, 0)));
}
template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {