diff options
| -rw-r--r-- | src/vlib/buffer_funcs.h | 8 | ||||
| -rw-r--r-- | src/vppinfra/vector_sse42.h | 36 |
2 files changed, 44 insertions, 0 deletions
diff --git a/src/vlib/buffer_funcs.h b/src/vlib/buffer_funcs.h index 205eaa37eaf..667063cd693 100644 --- a/src/vlib/buffer_funcs.h +++ b/src/vlib/buffer_funcs.h @@ -97,6 +97,14 @@ vlib_get_buffers_with_offset (vlib_main_t * vm, u32 * bi, void **b, int count, u64x4 b0 = u32x4_extend_to_u64x4 (u32x4_load_unaligned (bi)); /* shift and add to get vlib_buffer_t pointer */ u64x4_store_unaligned ((b0 << CLIB_LOG2_CACHE_LINE_BYTES) + off, b); +#elif defined (CLIB_HAVE_VEC128) && defined (__x86_64__) + u64x2 off = u64x2_splat (buffer_main.buffer_mem_start + offset); + u32x4 bi4 = u32x4_load_unaligned (bi); + u64x2 b0 = u32x4_extend_to_u64x2 ((u32x4) bi4); + bi4 = u32x4_shuffle (bi4, 2, 3, 0, 1); + u64x2 b1 = u32x4_extend_to_u64x2 ((u32x4) bi4); + u64x2_store_unaligned ((b0 << CLIB_LOG2_CACHE_LINE_BYTES) + off, b); + u64x2_store_unaligned ((b1 << CLIB_LOG2_CACHE_LINE_BYTES) + off, b + 2); #else b[0] = ((u8 *) vlib_get_buffer (vm, bi[0])) + offset; b[1] = ((u8 *) vlib_get_buffer (vm, bi[1])) + offset; diff --git a/src/vppinfra/vector_sse42.h b/src/vppinfra/vector_sse42.h index 5134855d2ae..76737940230 100644 --- a/src/vppinfra/vector_sse42.h +++ b/src/vppinfra/vector_sse42.h @@ -613,6 +613,42 @@ u8x16_shuffle (u8x16 v, u8x16 m) return (u8x16) _mm_shuffle_epi8 ((__m128i) v, (__m128i) m); } +static_always_inline u32x4 +u32x4_shuffle (u32x4 v, const int a, const int b, const int c, const int d) +{ +#ifdef __clang__ + u32x4 r = { v[a], v[b], v[c], v[d] }; + return r; +#else + return (u32x4) _mm_shuffle_epi32 ((__m128i) v, + a | b << 2 | c << 4 | d << 6); +#endif +} + +/* _extend_to_ */ +/* *INDENT-OFF* */ +#define _(f,t,i) \ +static_always_inline t \ +f##_extend_to_##t (f x) \ +{ return (t) _mm_cvt##i ((__m128i) x); } + +_(u8x16, u16x8, epu8_epi16) +_(u8x16, u32x4, epu8_epi32) +_(u8x16, u64x2, epu8_epi64) +_(u16x8, u32x4, epu16_epi32) +_(u16x8, u64x2, epu16_epi64) +_(u32x4, u64x2, epu32_epi64) + +_(i8x16, i16x8, epi8_epi16) +_(i8x16, i32x4, epi8_epi32) +_(i8x16, i64x2, epi8_epi64) +_(i16x8, i32x4, epi16_epi32) +_(i16x8, i64x2, epi16_epi64) +_(i32x4, i64x2, epi32_epi64) +#undef _ +/* *INDENT-ON* */ + + #endif /* included_vector_sse2_h */ /* |