diff options
| author |   Adrian Oanca <adrian.oanca@enea.com> | 2018-02-23 16:27:41 +0100 |
|---|---|---|
| committer |   Damjan Marion <dmarion.lists@gmail.com> | 2018-02-26 22:28:50 +0000 |
| commit | 22ac59b59157af057d5e1cade2f888341b92d6c0 (patch) | |
| tree | ffd5cc6ab195833b166c8578f4d0bee341f025fa | |
| parent | 200fa32213fce6824bcb75c907989ef8daba4a29 (diff) | |
Added u8x16,u32x4,u64x2 variants of _zero_byte_mask(x) for ARM/NEON platform. VPP-1129
Change-Id: I954acb56d901e42976e71534317f38d7c4359bcf
Signed-off-by: Adrian Oanca <adrian.oanca@enea.com>
| -rw-r--r-- | src/vnet/classify/vnet_classify.h | 12 | ||||
| -rw-r--r-- | src/vppinfra/vector_neon.h | 20 |
2 files changed, 24 insertions, 8 deletions
diff --git a/src/vnet/classify/vnet_classify.h b/src/vnet/classify/vnet_classify.h index 6cbbf10aa2e..791b8fd18b7 100644 --- a/src/vnet/classify/vnet_classify.h +++ b/src/vnet/classify/vnet_classify.h @@ -41,10 +41,6 @@ extern vlib_node_registration_t ip6_classify_node; #define CLASSIFY_TRACE 0 -#ifdef CLIB_HAVE_VEC128 -#define CLASSIFY_USE_SSE //Allow usage of SSE operations -#endif - #define U32X4_ALIGNED(p) PREDICT_TRUE((((intptr_t)p) & 0xf) == 0) /* @@ -233,7 +229,7 @@ vnet_classify_hash_packet_inline (vnet_classify_table_t * t, u8 * h) ASSERT (t); mask = t->mask; -#ifdef CLASSIFY_USE_SSE +#ifdef CLIB_HAVE_VEC128 if (U32X4_ALIGNED (h)) { //SSE can't handle unaligned data u32x4 *data = (u32x4 *) h; @@ -259,7 +255,7 @@ vnet_classify_hash_packet_inline (vnet_classify_table_t * t, u8 * h) } } else -#endif /* CLASSIFY_USE_SSE */ +#endif /* CLIB_HAVE_VEC128 */ { u32 skip_u64 = t->skip_n_vectors * 2; u64 *data64 = (u64 *) h; @@ -407,7 +403,7 @@ vnet_classify_find_entry_inline (vnet_classify_table_t * t, v = vnet_classify_entry_at_index (t, v, value_index); -#ifdef CLASSIFY_USE_SSE +#ifdef CLIB_HAVE_VEC128 if (U32X4_ALIGNED (h)) { u32x4 *data = (u32x4 *) h; @@ -452,7 +448,7 @@ vnet_classify_find_entry_inline (vnet_classify_table_t * t, } } else -#endif /* CLASSIFY_USE_SSE */ +#endif /* CLIB_HAVE_VEC128 */ { u32 skip_u64 = t->skip_n_vectors * 2; u64 *data64 = (u64 *) h; diff --git a/src/vppinfra/vector_neon.h b/src/vppinfra/vector_neon.h index 331b8ed6f5d..3ed78360296 100644 --- a/src/vppinfra/vector_neon.h +++ b/src/vppinfra/vector_neon.h @@ -86,6 +86,26 @@ u16x8_zero_byte_mask (u16x8 input) return (u32) (vgetq_lane_u64 (merge3, 1) << 8) + vgetq_lane_u64 (merge3, 0); } +always_inline u32 +u8x16_zero_byte_mask (u8x16 input) +{ + return u16x8_zero_byte_mask ((u16x8) input); +} + +always_inline u32 +u32x4_zero_byte_mask (u32x4 input) +{ + return u16x8_zero_byte_mask ((u16x8) input); +} + +always_inline u32 +u64x2_zero_byte_mask (u64x2 input) +{ + return u16x8_zero_byte_mask ((u16x8) input); +} + + + #endif /* included_vector_neon_h */ /* |