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 vshuf versions for SSE

Browse source
This commit is contained in:
Ahmet Inan 2024-02-13 13:27:55 +01:00
commit 5721da8bf7
3 changed files with 72 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

22
sse4_1_double.hh
View file

@ -1272,3 +1272,25 @@ inline SIMD<int32_t, 8> vclamp(SIMD<int32_t, 8> x, int32_t a, int32_t b)
return tmp; return tmp;
} }
template <>
inline SIMD<uint8_t, 32> vshuf(SIMD<uint8_t, 32> a, SIMD<uint8_t, 32> b)
{
SIMD<uint8_t, 32> tmp;
for (int i = 0; i < 2; ++i)
tmp.m[i] = _mm_or_si128(
_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
_mm_shuffle_epi8(a.m[1], _mm_sub_epi8(b.m[i], _mm_set1_epi8(16))));
return tmp;
}
template <>
inline SIMD<int8_t, 32> vshuf(SIMD<int8_t, 32> a, SIMD<uint8_t, 32> b)
{
SIMD<int8_t, 32> tmp;
for (int i = 0; i < 2; ++i)
tmp.m[i] = _mm_or_si128(
_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
_mm_shuffle_epi8(a.m[1], _mm_sub_epi8(b.m[i], _mm_set1_epi8(16))));
return tmp;
}

26
sse4_1_quadruple.hh
View file

@ -1272,3 +1272,29 @@ inline SIMD<int32_t, 16> vclamp(SIMD<int32_t, 16> x, int32_t a, int32_t b)
return tmp; return tmp;
} }
template <>
inline SIMD<uint8_t, 64> vshuf(SIMD<uint8_t, 64> a, SIMD<uint8_t, 64> b)
{
SIMD<uint8_t, 64> tmp;
for (int i = 0; i < 4; ++i)
tmp.m[i] = _mm_or_si128(_mm_or_si128(_mm_or_si128(
_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
_mm_shuffle_epi8(a.m[1], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(16)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(31))))),
_mm_shuffle_epi8(a.m[2], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(32)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(47))))),
_mm_shuffle_epi8(a.m[3], _mm_sub_epi8(b.m[i], _mm_set1_epi8(48))));
return tmp;
}
template <>
inline SIMD<int8_t, 64> vshuf(SIMD<int8_t, 64> a, SIMD<uint8_t, 64> b)
{
SIMD<int8_t, 64> tmp;
for (int i = 0; i < 4; ++i)
tmp.m[i] = _mm_or_si128(_mm_or_si128(_mm_or_si128(
_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
_mm_shuffle_epi8(a.m[1], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(16)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(31))))),
_mm_shuffle_epi8(a.m[2], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(32)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(47))))),
_mm_shuffle_epi8(a.m[3], _mm_sub_epi8(b.m[i], _mm_set1_epi8(48))));
return tmp;
}

24
sse4_1_triple.hh
View file

@ -1272,3 +1272,27 @@ inline SIMD<int32_t, 12> vclamp(SIMD<int32_t, 12> x, int32_t a, int32_t b)
return tmp; return tmp;
} }
template <>
inline SIMD<uint8_t, 48> vshuf(SIMD<uint8_t, 48> a, SIMD<uint8_t, 48> b)
{
SIMD<uint8_t, 48> tmp;
for (int i = 0; i < 3; ++i)
tmp.m[i] = _mm_or_si128(_mm_or_si128(
_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
_mm_shuffle_epi8(a.m[1], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(16)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(31))))),
_mm_shuffle_epi8(a.m[2], _mm_sub_epi8(b.m[i], _mm_set1_epi8(32))));
return tmp;
}
template <>
inline SIMD<int8_t, 48> vshuf(SIMD<int8_t, 48> a, SIMD<uint8_t, 48> b)
{
SIMD<int8_t, 48> tmp;
for (int i = 0; i < 3; ++i)
tmp.m[i] = _mm_or_si128(_mm_or_si128(
_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
_mm_shuffle_epi8(a.m[1], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(16)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(31))))),
_mm_shuffle_epi8(a.m[2], _mm_sub_epi8(b.m[i], _mm_set1_epi8(32))));
return tmp;
}