github.com/johnnyeven/libtools@v0.0.0-20191126065708-61829c1adf46/third_party/eigen3/gpu_packet_math.patch (about) 1 --- a/Eigen/src/Geometry/arch/Geometry_SSE.h 2 +++ b/Eigen/src/Geometry/arch/Geometry_SSE.h 3 @@ -33,13 +33,14 @@ 4 Packet4f b = be.template packet<BAlignment,Packet4f>(0); 5 Packet4f s1 = pmul(vec4f_swizzle1(a,1,2,0,2),vec4f_swizzle1(b,2,0,1,2)); 6 Packet4f s2 = pmul(vec4f_swizzle1(a,3,3,3,1),vec4f_swizzle1(b,0,1,2,1)); 7 - pstoret<float,Packet4f,ResAlignment>( 8 - &res.x(), 9 - padd(psub(pmul(a,vec4f_swizzle1(b,3,3,3,3)), 10 - pmul(vec4f_swizzle1(a,2,0,1,0), 11 - vec4f_swizzle1(b,1,2,0,0))), 12 - pxor(mask,padd(s1,s2)))); 13 - 14 + pstoret<float, Packet4f, ResAlignment>( 15 + &res.x(), 16 + padd<Packet4f>( 17 + psub<Packet4f>(pmul<Packet4f>(a, vec4f_swizzle1(b, 3, 3, 3, 3)), 18 + pmul<Packet4f>(vec4f_swizzle1(a, 2, 0, 1, 0), 19 + vec4f_swizzle1(b, 1, 2, 0, 0))), 20 + pxor<Packet4f>(mask, padd(s1, s2)))); 21 + 22 return res; 23 } 24 };