GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-10 18:59:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix pandnot order in AVX512

Browse Source
This commit is contained in:
Gael Guennebaud 2018-11-30 14:32:06 +01:00
parent b477d60bc6
commit f91500d303
2 changed files with 7 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Eigen/Core
View File

@ -154,8 +154,10 @@ using std::ptrdiff_t;
#if defined EIGEN_VECTORIZE_AVX512 #if defined EIGEN_VECTORIZE_AVX512
#include "src/Core/arch/SSE/PacketMath.h" #include "src/Core/arch/SSE/PacketMath.h"
#include "src/Core/arch/SSE/TypeCasting.h" #include "src/Core/arch/SSE/TypeCasting.h"
#include "src/Core/arch/SSE/Complex.h"
#include "src/Core/arch/AVX/PacketMath.h" #include "src/Core/arch/AVX/PacketMath.h"
#include "src/Core/arch/AVX/TypeCasting.h" #include "src/Core/arch/AVX/TypeCasting.h"
#include "src/Core/arch/AVX/Complex.h"
#include "src/Core/arch/AVX512/PacketMath.h" #include "src/Core/arch/AVX512/PacketMath.h"
#include "src/Core/arch/SSE/MathFunctions.h" #include "src/Core/arch/SSE/MathFunctions.h"
#include "src/Core/arch/AVX/MathFunctions.h" #include "src/Core/arch/AVX/MathFunctions.h"

10
Eigen/src/Core/arch/AVX512/PacketMath.h
View File

@ -393,24 +393,24 @@ template <>
EIGEN_STRONG_INLINE Packet16f pandnot<Packet16f>(const Packet16f& a, EIGEN_STRONG_INLINE Packet16f pandnot<Packet16f>(const Packet16f& a,
const Packet16f& b) { const Packet16f& b) {
#ifdef EIGEN_VECTORIZE_AVX512DQ #ifdef EIGEN_VECTORIZE_AVX512DQ
return _mm512_andnot_ps(a, b); return _mm512_andnot_ps(b, a);
#else #else
Packet16f res = _mm512_undefined_ps(); Packet16f res = _mm512_undefined_ps();
Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0); Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0); Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
res = _mm512_insertf32x4(res, _mm_andnot_ps(lane0_a, lane0_b), 0); res = _mm512_insertf32x4(res, pandnot(lane0_a, lane0_b), 0);
Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1); Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1); Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
res = _mm512_insertf32x4(res, _mm_andnot_ps(lane1_a, lane1_b), 1); res = _mm512_insertf32x4(res, pandnot(lane1_a, lane1_b), 1);
Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2); Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2); Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
res = _mm512_insertf32x4(res, _mm_andnot_ps(lane2_a, lane2_b), 2); res = _mm512_insertf32x4(res, pandnot(lane2_a, lane2_b), 2);
Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3); Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3); Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
res = _mm512_insertf32x4(res, _mm_andnot_ps(lane3_a, lane3_b), 3); res = _mm512_insertf32x4(res, pandnot(lane3_a, lane3_b), 3);
return res; return res;
#endif #endif