mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-01-24 14:45:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix ptranspose overload prototypes for NEON

Browse Source
This commit is contained in:
Gael Guennebaud 2014-04-25 11:15:13 +02:00
parent c79bd4b64b
commit 4def7b1fa5
2 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Eigen/src/Core/arch/NEON/Complex.h
View File

@ -263,7 +263,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, con
return Packet2cf(pdiv(res.v, vaddq_f32(s,rev_s)));
}
template<> EIGEN_DEVICE_FUNC inline void
EIGEN_DEVICE_FUNC inline void
ptranspose(PacketBlock<Packet2cf,2>& kernel) {
float32x4_t tmp = vcombine_f32(vget_high_f32(kernel.packet[0].v), vget_high_f32(kernel.packet[1].v));
kernel.packet[0].v = vcombine_f32(vget_low_f32(kernel.packet[0].v), vget_low_f32(kernel.packet[1].v));

4
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@ -451,7 +451,7 @@ PALIGN_NEON(3,Packet4i,vextq_s32)
#undef PALIGN_NEON
template<> EIGEN_DEVICE_FUNC inline void
EIGEN_DEVICE_FUNC inline void
ptranspose(PacketBlock<Packet4f,4>& kernel) {
float32x4x2_t tmp1 = vzipq_f32(kernel.packet[0], kernel.packet[1]);
float32x4x2_t tmp2 = vzipq_f32(kernel.packet[2], kernel.packet[3]);
@ -462,7 +462,7 @@ ptranspose(PacketBlock<Packet4f,4>& kernel) {
kernel.packet[3] = vcombine_f32(vget_high_f32(tmp1.val[1]), vget_high_f32(tmp2.val[1]));
}
template<> EIGEN_DEVICE_FUNC inline void
EIGEN_DEVICE_FUNC inline void
ptranspose(PacketBlock<Packet4i,4>& kernel) {
int32x4x2_t tmp1 = vzipq_s32(kernel.packet[0], kernel.packet[1]);
int32x4x2_t tmp2 = vzipq_s32(kernel.packet[2], kernel.packet[3]);