FD.io VPP  v19.01.3-6-g70449b9b9
Vector Packet Processing
vector_neon.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2015 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 
16 #ifndef included_vector_neon_h
17 #define included_vector_neon_h
18 #include <arm_neon.h>
19 
20 /* Arithmetic */
21 #define u16x8_sub_saturate(a,b) vsubq_u16(a,b)
22 #define i16x8_sub_saturate(a,b) vsubq_s16(a,b)
23 /* Dummy. Aid making uniform macros */
24 #define vreinterpretq_u8_u8(a) a
25 
26 /* Converts all ones/zeros compare mask to bitmap. */
29 {
30  uint8x16_t mask = { 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
31  0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80
32  };
33  /* v --> [0xFF, 0x00, 0xFF, 0xFF, 0xFF, 0x00, 0xFF, 0x00, ... ] */
34  uint8x16_t x = vandq_u8 (v, mask);
35  /* after v & mask,
36  * x --> [0x01, 0x00, 0x04, 0x08, 0x10, 0x00, 0x40, 0x00, ... ] */
37  uint64x2_t x64 = vpaddlq_u32 (vpaddlq_u16 (vpaddlq_u8 (x)));
38  /* after merge, x64 --> [0x5D, 0x.. ] */
39  return (u32) (vgetq_lane_u64 (x64, 0) + (vgetq_lane_u64 (x64, 1) << 8));
40 }
41 
42 /* *INDENT-OFF* */
43 #define foreach_neon_vec128i \
44  _(i,8,16,s8) _(i,16,8,s16) _(i,32,4,s32) _(i,64,2,s64)
45 #define foreach_neon_vec128u \
46  _(u,8,16,u8) _(u,16,8,u16) _(u,32,4,u32) _(u,64,2,u64)
47 #define foreach_neon_vec128f \
48  _(f,32,4,f32) _(f,64,2,f64)
49 
50 #define _(t, s, c, i) \
51 static_always_inline t##s##x##c \
52 t##s##x##c##_splat (t##s x) \
53 { return (t##s##x##c) vdupq_n_##i (x); } \
54 \
55 static_always_inline t##s##x##c \
56 t##s##x##c##_load_unaligned (void *p) \
57 { return (t##s##x##c) vld1q_##i (p); } \
58 \
59 static_always_inline void \
60 t##s##x##c##_store_unaligned (t##s##x##c v, void *p) \
61 { vst1q_##i (p, v); } \
62 \
63 static_always_inline int \
64 t##s##x##c##_is_all_zero (t##s##x##c x) \
65 { return !(vaddvq_##i (x)); } \
66 \
67 static_always_inline int \
68 t##s##x##c##_is_equal (t##s##x##c a, t##s##x##c b) \
69 { return t##s##x##c##_is_all_zero (a ^ b); } \
70 \
71 static_always_inline int \
72 t##s##x##c##_is_all_equal (t##s##x##c v, t##s x) \
73 { return t##s##x##c##_is_equal (v, t##s##x##c##_splat (x)); }; \
74 \
75 static_always_inline u32 \
76 t##s##x##c##_zero_byte_mask (t##s##x##c x) \
77 { uint8x16_t v = vreinterpretq_u8_u##s (vceqq_##i (vdupq_n_##i(0), x)); \
78  return u8x16_compare_byte_mask (v); } \
79 
81 
82 #undef _
83 /* *INDENT-ON* */
84 
86 u16x8_byte_swap (u16x8 v)
87 {
88  return (u16x8) vrev16q_u8 ((u8x16) v);
89 }
90 
92 u8x16_shuffle (u8x16 v, u8x16 m)
93 {
94  return (u8x16) vqtbl1q_u8 (v, m);
95 }
96 
99 {
100  return (u32x4) vpaddq_u32 (v1, v2);
101 }
102 
105 {
106  return vmovl_u32 (vget_low_u32 (v));
107 }
108 
111 {
112  return vmovl_high_u32 (v);
113 }
114 
115 /* Creates a mask made up of the MSB of each byte of the source vector */
117 u8x16_msb_mask (u8x16 v)
118 {
119  int8x16_t shift =
120  { -7, -6, -5, -4, -3, -2, -1, 0, -7, -6, -5, -4, -3, -2, -1, 0 };
121  /* v --> [0x80, 0x7F, 0xF0, 0xAF, 0xF0, 0x00, 0xF2, 0x00, ... ] */
122  uint8x16_t x = vshlq_u8 (vandq_u8 (v, vdupq_n_u8 (0x80)), shift);
123  /* after (v & 0x80) >> shift,
124  * x --> [0x01, 0x00, 0x04, 0x08, 0x10, 0x00, 0x40, 0x00, ... ] */
125  uint64x2_t x64 = vpaddlq_u32 (vpaddlq_u16 (vpaddlq_u8 (x)));
126  /* after merge, x64 --> [0x5D, 0x.. ] */
127  return (u16) (vgetq_lane_u64 (x64, 0) + (vgetq_lane_u64 (x64, 1) << 8));
128 }
129 
130 #define CLIB_HAVE_VEC128_MSB_MASK
131 
132 #define CLIB_HAVE_VEC128_UNALIGNED_LOAD_STORE
133 #define CLIB_VEC128_SPLAT_DEFINED
134 #endif /* included_vector_neon_h */
135 
136 /*
137  * fd.io coding-style-patch-verification: ON
138  *
139  * Local Variables:
140  * eval: (c-set-style "gnu")
141  * End:
142  */
#define foreach_neon_vec128i
Definition: vector_neon.h:43
static_always_inline u64x2 u32x4_extend_to_u64x2_high(u32x4 v)
Definition: vector_neon.h:110
static_always_inline u32x4 u32x4_hadd(u32x4 v1, u32x4 v2)
Definition: vector_neon.h:98
#define static_always_inline
Definition: clib.h:99
#define always_inline
Definition: clib.h:98
unsigned int u32
Definition: types.h:88
epu8_epi32 epu16_epi32 u64x2
Definition: vector_sse42.h:640
unsigned short u16
Definition: types.h:57
static_always_inline u16 u8x16_msb_mask(u8x16 v)
Definition: vector_neon.h:117
static u32 u8x16_compare_byte_mask(u8x16 v)
Definition: vector_neon.h:28
static_always_inline u64x2 u32x4_extend_to_u64x2(u32x4 v)
Definition: vector_neon.h:104
#define foreach_neon_vec128u
Definition: vector_neon.h:45
static_always_inline u8x16 u8x16_shuffle(u8x16 v, u8x16 m)
Definition: vector_neon.h:92
unsigned long long u32x4
Definition: ixge.c:28
foreach_neon_vec128i foreach_neon_vec128u static_always_inline u16x8 u16x8_byte_swap(u16x8 v)
Definition: vector_neon.h:86