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 more shuffle operations for AVX2

Browse source
This commit is contained in:
Ahmet Inan 2024-02-14 10:48:45 +01:00
commit 41b8abaf83
2 changed files with 30 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

8
avx2.hh
View file

@ -1178,6 +1178,14 @@ inline SIMD<int8_t, 32> vshuf(SIMD<int8_t, 32> a, SIMD<uint8_t, 32> b)
return tmp; return tmp;
} }
template <>
inline SIMD<uint32_t, 8> vshuf(SIMD<uint32_t, 8> a, SIMD<uint32_t, 8> b)
{
SIMD<uint32_t, 8> tmp;
tmp.m = _mm256_permutevar8x32_epi32(a.m, b.m);
return tmp;
}
template <> template <>
inline SIMD<float, 8> vshuf(SIMD<float, 8> a, SIMD<uint32_t, 8> b) inline SIMD<float, 8> vshuf(SIMD<float, 8> a, SIMD<uint32_t, 8> b)
{ {

22
avx2_double.hh
View file

@ -1321,3 +1321,25 @@ inline SIMD<int8_t, 64> vshuf(SIMD<int8_t, 64> a, SIMD<uint8_t, 64> b)
return tmp; return tmp;
} }
template <>
inline SIMD<uint32_t, 16> vshuf(SIMD<uint32_t, 16> a, SIMD<uint32_t, 16> b)
{
SIMD<uint32_t, 16> tmp;
for (int i = 0; i < 2; ++i)
tmp.m[i] = _mm256_or_si256(
_mm256_and_si256(_mm256_permutevar8x32_epi32(a.m[0], b.m[i]), _mm256_cmpgt_epi32(_mm256_set1_epi32(8), b.m[i])),
_mm256_and_si256(_mm256_permutevar8x32_epi32(a.m[1], _mm256_sub_epi32(b.m[i], _mm256_set1_epi32(8))), _mm256_cmpgt_epi32(b.m[i], _mm256_set1_epi32(7))));
return tmp;
}
template <>
inline SIMD<float, 16> vshuf(SIMD<float, 16> a, SIMD<uint32_t, 16> b)
{
SIMD<float, 16> tmp;
for (int i = 0; i < 2; ++i)
tmp.m[i] = _mm256_or_ps(
_mm256_and_ps(_mm256_permutevar8x32_ps(a.m[0], b.m[i]), _mm256_cmpgt_epi32(_mm256_set1_epi32(8), b.m[i])),
_mm256_and_ps(_mm256_permutevar8x32_ps(a.m[1], _mm256_sub_epi32(b.m[i], _mm256_set1_epi32(8))), _mm256_cmpgt_epi32(b.m[i], _mm256_set1_epi32(7))));
return tmp;
}