diff options
| author |   Lijian Zhang <Lijian.Zhang@arm.com> | 2018-10-25 18:50:33 +0800 |
|---|---|---|
| committer |   Damjan Marion <dmarion@me.com> | 2018-10-31 10:24:22 +0000 |
| commit | a22ba8859a108569b457d18873a52a91f1387a38 (patch) | |
| tree | b4d6d0641b970e6ad6a4a2600b00dea4f5144d65 | |
| parent | 11d47af5af3274b7f5dd1ac2b3a1f9fa025b2414 (diff) | |
Add and enable msb mask vector intrinsic for aarch64.
This patch enables the use of this function for enqueuing frames to the next graph node.
Change-Id: I4003110db59870f7106e0d13942d6ff7bc54b46d
Signed-off-by: Lijian Zhang <Lijian.Zhang@arm.com>
Reviewed-by: Sirshak Das <Sirshak.Das@arm.com>
Reviewed-by: Honnappa Nagarahalli <Honnappa.Nagarahalli@arm.com>
Reviewed-by: Steve Capper <Steve.Capper@arm.com>
| -rw-r--r-- | src/vppinfra/vector_neon.h | 38 |
1 files changed, 28 insertions, 10 deletions
diff --git a/src/vppinfra/vector_neon.h b/src/vppinfra/vector_neon.h index f6d953cab81..3ac41bbb72c 100644 --- a/src/vppinfra/vector_neon.h +++ b/src/vppinfra/vector_neon.h @@ -23,17 +23,18 @@ /* Converts all ones/zeros compare mask to bitmap. */ always_inline u32 -u8x16_compare_byte_mask (u8x16 x) +u8x16_compare_byte_mask (u8x16 v) { - uint8x16_t mask_shift = - { -7, -6, -5, -4, -3, -2, -1, 0, -7, -6, -5, -4, -3, -2, -1, 0 }; - uint8x16_t mask_and = vdupq_n_u8 (0x80); - x = vandq_u8 (x, mask_and); - x = vshlq_u8 (x, vreinterpretq_s8_u8 (mask_shift)); - x = vpaddq_u8 (x, x); - x = vpaddq_u8 (x, x); - x = vpaddq_u8 (x, x); - return vgetq_lane_u8 (x, 0) | (vgetq_lane_u8 (x, 1) << 8); + uint8x16_t mask = { 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, + 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80 + }; + /* v --> [0xFF, 0x00, 0xFF, 0xFF, 0xFF, 0x00, 0xFF, 0x00, ... ] */ + uint8x16_t x = vandq_u8 (v, mask); + /* after v & mask, + * x --> [0x01, 0x00, 0x04, 0x08, 0x10, 0x00, 0x40, 0x00, ... ] */ + uint64x2_t x64 = vpaddlq_u32 (vpaddlq_u16 (vpaddlq_u8 (x))); + /* after merge, x64 --> [0x5D, 0x.. ] */ + return (u32) (vgetq_lane_u64 (x64, 0) + (vgetq_lane_u64 (x64, 1) << 8)); } always_inline u32 @@ -148,6 +149,23 @@ u32x4_extend_to_u64x2_high (u32x4 v) return vmovl_high_u32 (v); } +/* Creates a mask made up of the MSB of each byte of the source vector */ +static_always_inline u16 +u8x16_msb_mask (u8x16 v) +{ + int8x16_t shift = + { -7, -6, -5, -4, -3, -2, -1, 0, -7, -6, -5, -4, -3, -2, -1, 0 }; + /* v --> [0x80, 0x7F, 0xF0, 0xAF, 0xF0, 0x00, 0xF2, 0x00, ... ] */ + uint8x16_t x = vshlq_u8 (vandq_u8 (v, vdupq_n_u8 (0x80)), shift); + /* after (v & 0x80) >> shift, + * x --> [0x01, 0x00, 0x04, 0x08, 0x10, 0x00, 0x40, 0x00, ... ] */ + uint64x2_t x64 = vpaddlq_u32 (vpaddlq_u16 (vpaddlq_u8 (x))); + /* after merge, x64 --> [0x5D, 0x.. ] */ + return (u16) (vgetq_lane_u64 (x64, 0) + (vgetq_lane_u64 (x64, 1) << 8)); +} + +#define CLIB_HAVE_VEC128_MSB_MASK + #define CLIB_HAVE_VEC128_UNALIGNED_LOAD_STORE #define CLIB_VEC128_SPLAT_DEFINED #endif /* included_vector_neon_h */ |