Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

CrossSIMD: make the transpose function compatible with ARM32

Browse source
This commit is contained in:
Henrik Rydgård 2024-12-21 13:21:20 +01:00
parent 2051d55c90
commit 399570e411
1 changed files with 10 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
Common/Math/CrossSIMD.h
View file

@ -264,6 +264,8 @@ struct Vec4F32 {
// One of many possible solutions. Sometimes we could also use vld4q_f32 probably..
static void Transpose(Vec4F32 &col0, Vec4F32 &col1, Vec4F32 &col2, Vec4F32 &col3) {
#if PPSSPP_ARCH(ARM64_NEON)
// Only works on ARM64
float32x4_t temp0 = vzip1q_f32(col0.v, col2.v);
float32x4_t temp1 = vzip2q_f32(col0.v, col2.v);
float32x4_t temp2 = vzip1q_f32(col1.v, col3.v);
@ -272,6 +274,14 @@ struct Vec4F32 {
col1.v = vzip2q_f32(temp0, temp2);
col2.v = vzip1q_f32(temp1, temp3);
col3.v = vzip2q_f32(temp1, temp3);
#else
float32x4x2_t col01 = vtrnq_f32(col0.v, col1.v);
float32x4x2_t col23 = vtrnq_f32(col2.v, col3.v);
col0.v = vcombine_f32(vget_low_f32(col01.val[0]), vget_low_f32(col23.val[0]));
col1.v = vcombine_f32(vget_low_f32(col01.val[1]), vget_low_f32(col23.val[1]));
col2.v = vcombine_f32(vget_high_f32(col01.val[0]), vget_high_f32(col23.val[0]));
col3.v = vcombine_f32(vget_high_f32(col01.val[1]), vget_high_f32(col23.val[1]));
#endif
}
inline Vec4F32 AsVec3ByMatrix44(const Mat4F32 &m) {