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

use struct versions of vqtbl

Browse source
This commit is contained in:
Ahmet Inan 2026-01-20 11:24:13 +01:00
commit d94b8e8ce9
3 changed files with 6 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

4
neon_double.hh
View file

@ -1058,7 +1058,7 @@ 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.val[i] = vorrq_u8(vqtbl1q_u8(a.m.val[0], b.m.val[i]), vqtbl1q_u8(a.m.val[1], vsubq_u8(b.m.val[i], vdupq_n_u8(16))));
tmp.m.val[i] = vqtbl2q_u8(a.m, b.m.val[i]);
return tmp;
}
@ -1067,7 +1067,7 @@ 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.val[i] = vorrq_s8(vqtbl1q_s8(a.m.val[0], b.m.val[i]), vqtbl1q_s8(a.m.val[1], vsubq_u8(b.m.val[i], vdupq_n_u8(16))));
tmp.m.val[i] = vqtbl2q_s8(a.m, b.m.val[i]);
return tmp;
}
#endif

10
neon_quadruple.hh
View file

@ -1058,10 +1058,7 @@ 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.val[i] = vorrq_u8(vorrq_u8(vorrq_u8(vqtbl1q_u8(a.m.val[0], b.m.val[i]),
vqtbl1q_u8(a.m.val[1], vsubq_u8(b.m.val[i], vdupq_n_u8(16)))),
vqtbl1q_u8(a.m.val[2], vsubq_u8(b.m.val[i], vdupq_n_u8(32)))),
vqtbl1q_u8(a.m.val[3], vsubq_u8(b.m.val[i], vdupq_n_u8(48))));
tmp.m.val[i] = vqtbl4q_u8(a.m, b.m.val[i]);
return tmp;
}
@ -1070,10 +1067,7 @@ 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.val[i] = vorrq_s8(vorrq_s8(vorrq_s8(vqtbl1q_s8(a.m.val[0], b.m.val[i]),
vqtbl1q_s8(a.m.val[1], vsubq_u8(b.m.val[i], vdupq_n_u8(16)))),
vqtbl1q_s8(a.m.val[2], vsubq_u8(b.m.val[i], vdupq_n_u8(32)))),
vqtbl1q_s8(a.m.val[3], vsubq_u8(b.m.val[i], vdupq_n_u8(48))));
tmp.m.val[i] = vqtbl4q_s8(a.m, b.m.val[i]);
return tmp;
}
#endif

8
neon_triple.hh
View file

@ -1058,9 +1058,7 @@ 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.val[i] = vorrq_u8(vorrq_u8(vqtbl1q_u8(a.m.val[0], b.m.val[i]),
vqtbl1q_u8(a.m.val[1], vsubq_u8(b.m.val[i], vdupq_n_u8(16)))),
vqtbl1q_u8(a.m.val[2], vsubq_u8(b.m.val[i], vdupq_n_u8(32))));
tmp.m.val[i] = vqtbl3q_u8(a.m, b.m.val[i]);
return tmp;
}
@ -1069,9 +1067,7 @@ 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.val[i] = vorrq_s8(vorrq_s8(vqtbl1q_s8(a.m.val[0], b.m.val[i]),
vqtbl1q_s8(a.m.val[1], vsubq_u8(b.m.val[i], vdupq_n_u8(16)))),
vqtbl1q_s8(a.m.val[2], vsubq_u8(b.m.val[i], vdupq_n_u8(32))));
tmp.m.val[i] = vqtbl3q_s8(a.m, b.m.val[i]);
return tmp;
}
#endif