diff options
Diffstat (limited to 'src/vppinfra/vector_avx512.h')
| -rw-r--r-- | src/vppinfra/vector_avx512.h | 19 |
1 files changed, 17 insertions, 2 deletions
diff --git a/src/vppinfra/vector_avx512.h b/src/vppinfra/vector_avx512.h index 3505961dca5..e71a8a70aa9 100644 --- a/src/vppinfra/vector_avx512.h +++ b/src/vppinfra/vector_avx512.h @@ -322,9 +322,16 @@ u8x64_mask_blend (u8x64 a, u8x64 b, u64 mask) } static_always_inline u8x64 -u8x64_permute (u8x64 v, u8x64 idx) +u8x64_permute (u8x64 idx, u8x64 a) { - return (u8x64) _mm512_permutexvar_epi8 ((__m512i) v, (__m512i) idx); + return (u8x64) _mm512_permutexvar_epi8 ((__m512i) idx, (__m512i) a); +} + +static_always_inline u8x64 +u8x64_permute2 (u8x64 idx, u8x64 a, u8x64 b) +{ + return (u8x64) _mm512_permutex2var_epi8 ((__m512i) a, (__m512i) idx, + (__m512i) b); } #define _(t, m, e, p, it) \ @@ -436,6 +443,14 @@ u16x8_compress (u16x8 v, u8 mask) } #endif +static_always_inline u64 +u64x8_hxor (u64x8 v) +{ + v ^= u64x8_align_right (v, v, 4); + v ^= u64x8_align_right (v, v, 2); + return v[0] ^ v[1]; +} + static_always_inline void u32x16_transpose (u32x16 m[16]) { |