Mirror/aicodix___code
1
0
Fork 0
mirror of https://github.com/aicodix/code.git synced 2026-04-27 14:30:36 +00:00
Code Issues Projects Releases Packages Wiki Activity

added vmul() and vsignum()

Browse source
This commit is contained in:
Ahmet Inan 2020-07-12 22:44:46 +02:00
commit 48303e7f58
4 changed files with 263 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

35
neon.hh
View file

@ -444,6 +444,22 @@ inline SIMD<uint16_t, 8> vqsub(SIMD<uint16_t, 8> a, SIMD<uint16_t, 8> b)
return tmp;
}
template <>
inline SIMD<float, 4> vmul(SIMD<float, 4> a, SIMD<float, 4> b)
{
SIMD<float, 4> tmp;
tmp.m = vmulq_f32(a.m, b.m);
return tmp;
}
template <>
inline SIMD<int8_t, 16> vmul(SIMD<int8_t, 16> a, SIMD<int8_t, 16> b)
{
SIMD<int8_t, 16> tmp;
tmp.m = vmulq_s8(a.m, b.m);
return tmp;
}
template <>
inline SIMD<float, 4> vabs(SIMD<float, 4> a)
{
@ -468,6 +484,25 @@ inline SIMD<int16_t, 8> vqabs(SIMD<int16_t, 8> a)
return tmp;
}
template <>
inline SIMD<float, 4> vsignum(SIMD<float, 4> a)
{
SIMD<float, 4> tmp;
tmp.m = (float32x4_t)vbicq_u32(
veorq_u32((uint32x4_t)vdupq_n_f32(1.f), vandq_u32((uint32x4_t)vdupq_n_f32(-0.f), (uint32x4_t)a.m)),
vceqq_f32(a.m, vdupq_n_f32(0.f)));
return tmp;
}
template <>
inline SIMD<int8_t, 16> vsignum(SIMD<int8_t, 16> a)
{
SIMD<int8_t, 16> tmp;
tmp.m = (int8x16_t)vorrq_u8(vcgtq_s8(vdupq_n_s8(0), a.m),
vandq_u8(vcgtq_s8(a.m, vdupq_n_s8(0)), (uint8x16_t)vdupq_n_s8(1)));
return tmp;
}
template <>
inline SIMD<float, 4> vsign(SIMD<float, 4> a, SIMD<float, 4> b)
{