diff options
| -rw-r--r-- | src/vppinfra/vector_avx2.h | 16 | ||||
| -rw-r--r-- | src/vppinfra/vector_sse42.h | 15 |
2 files changed, 31 insertions, 0 deletions
diff --git a/src/vppinfra/vector_avx2.h b/src/vppinfra/vector_avx2.h index 6dd110c52e3..6b298be8bfd 100644 --- a/src/vppinfra/vector_avx2.h +++ b/src/vppinfra/vector_avx2.h @@ -121,6 +121,22 @@ _(i8x16, i64x4, epi8_epi64) #undef _ /* *INDENT-ON* */ +static_always_inline u16x16 +u16x16_byte_swap (u16x16 v) +{ + u8x32 swap = { + 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, + 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14 + }; + return (u16x16) _mm256_shuffle_epi8 ((__m256i) v, (__m256i) swap); +} + +static_always_inline u32x8 +u32x8_hadd (u32x8 v1, u32x8 v2) +{ + return (u32x8) _mm256_hadd_epi32 ((__m256i) v1, (__m256i) v2); +} + #endif /* included_vector_avx2_h */ /* diff --git a/src/vppinfra/vector_sse42.h b/src/vppinfra/vector_sse42.h index fa377cc983b..deb98776f7f 100644 --- a/src/vppinfra/vector_sse42.h +++ b/src/vppinfra/vector_sse42.h @@ -592,6 +592,21 @@ u8x16_msb_mask (u8x16 v) #undef _signed_binop +static_always_inline u16x8 +u16x8_byte_swap (u16x8 v) +{ + u8x16 swap = { + 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, + }; + return (u16x8) _mm_shuffle_epi8 ((__m128i) v, (__m128i) swap); +} + +static_always_inline u32x4 +u32x4_hadd (u32x4 v1, u32x4 v2) +{ + return (u32x4) _mm_hadd_epi32 ((__m128i) v1, (__m128i) v2); +} + #endif /* included_vector_sse2_h */ /* |