FD.io VPP  v18.10-34-gcce845e
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 
24 /* Converts all ones/zeros compare mask to bitmap. */
27 {
28  uint8x16_t mask_shift =
29  { -7, -6, -5, -4, -3, -2, -1, 0, -7, -6, -5, -4, -3, -2, -1, 0 };
30  uint8x16_t mask_and = vdupq_n_u8 (0x80);
31  x = vandq_u8 (x, mask_and);
32  x = vshlq_u8 (x, vreinterpretq_s8_u8 (mask_shift));
33  x = vpaddq_u8 (x, x);
34  x = vpaddq_u8 (x, x);
35  x = vpaddq_u8 (x, x);
36  return vgetq_lane_u8 (x, 0) | (vgetq_lane_u8 (x, 1) << 8);
37 }
38 
40 u16x8_zero_byte_mask (u16x8 input)
41 {
42  u8x16 vall_one = vdupq_n_u8 (0x0);
43  u8x16 res_values = { 0x01, 0x02, 0x04, 0x08,
44  0x10, 0x20, 0x40, 0x80,
45  0x01, 0x02, 0x04, 0x08,
46  0x10, 0x20, 0x40, 0x80
47  };
48 
49  /* input --> [0x80, 0x40, 0x01, 0xf0, ... ] */
50  u8x16 test_result =
51  vreinterpretq_u8_u16 (vceqq_u16 (input, vreinterpretq_u16_u8 (vall_one)));
52  u8x16 before_merge = vminq_u8 (test_result, res_values);
53  /*before_merge--> [0x80, 0x00, 0x00, 0x10, ... ] */
54  /* u8x16 --> [a,b,c,d, e,f,g,h, i,j,k,l, m,n,o,p] */
55  /* pair add until we have 2 uint64_t */
56  u16x8 merge1 = vpaddlq_u8 (before_merge);
57  /* u16x8--> [a+b,c+d, e+f,g+h, i+j,k+l, m+n,o+p] */
58  u32x4 merge2 = vpaddlq_u16 (merge1);
59  /* u32x4--> [a+b+c+d, e+f+g+h, i+j+k+l, m+n+o+p] */
60  u64x2 merge3 = vpaddlq_u32 (merge2);
61  /* u64x2--> [a+b+c+d+e+f+g+h, i+j+k+l+m+n+o+p] */
62  return (u32) (vgetq_lane_u64 (merge3, 1) << 8) + vgetq_lane_u64 (merge3, 0);
63 }
64 
66 u8x16_zero_byte_mask (u8x16 input)
67 {
68  return u16x8_zero_byte_mask ((u16x8) input);
69 }
70 
73 {
74  return u16x8_zero_byte_mask ((u16x8) input);
75 }
76 
79 {
80  return u16x8_zero_byte_mask ((u16x8) input);
81 }
82 
83 /* *INDENT-OFF* */
84 #define foreach_neon_vec128i \
85  _(i,8,16,s8) _(i,16,8,s16) _(i,32,4,s32) _(i,64,2,s64)
86 #define foreach_neon_vec128u \
87  _(u,8,16,u8) _(u,16,8,u16) _(u,32,4,u32) _(u,64,2,u64)
88 #define foreach_neon_vec128f \
89  _(f,32,4,f32) _(f,64,2,f64)
90 
91 #define _(t, s, c, i) \
92 static_always_inline t##s##x##c \
93 t##s##x##c##_splat (t##s x) \
94 { return (t##s##x##c) vdupq_n_##i (x); } \
95 \
96 static_always_inline t##s##x##c \
97 t##s##x##c##_load_unaligned (void *p) \
98 { return (t##s##x##c) vld1q_##i (p); } \
99 \
100 static_always_inline void \
101 t##s##x##c##_store_unaligned (t##s##x##c v, void *p) \
102 { vst1q_##i (p, v); } \
103 \
104 static_always_inline int \
105 t##s##x##c##_is_all_zero (t##s##x##c x) \
106 { return !(vaddvq_##i (x)); } \
107 \
108 static_always_inline int \
109 t##s##x##c##_is_equal (t##s##x##c a, t##s##x##c b) \
110 { return t##s##x##c##_is_all_zero (a ^ b); } \
111 \
112 static_always_inline int \
113 t##s##x##c##_is_all_equal (t##s##x##c v, t##s x) \
114 { return t##s##x##c##_is_equal (v, t##s##x##c##_splat (x)); }; \
115 
117 
118 #undef _
119 /* *INDENT-ON* */
120 
123 {
124  return (u16x8) vrev16q_u8 ((u8x16) v);
125 }
126 
128 u8x16_shuffle (u8x16 v, u8x16 m)
129 {
130  return (u8x16) vqtbl1q_u8 (v, m);
131 }
132 
135 {
136  return (u32x4) vpaddq_u32 (v1, v2);
137 }
138 
141 {
142  return vmovl_u32 (vget_low_u32 (v));
143 }
144 
147 {
148  return vmovl_high_u32 (v);
149 }
150 
151 #define CLIB_HAVE_VEC128_UNALIGNED_LOAD_STORE
152 #define CLIB_VEC128_SPLAT_DEFINED
153 #endif /* included_vector_neon_h */
154 
155 /*
156  * fd.io coding-style-patch-verification: ON
157  *
158  * Local Variables:
159  * eval: (c-set-style "gnu")
160  * End:
161  */
#define foreach_neon_vec128i
Definition: vector_neon.h:84
static_always_inline u64x2 u32x4_extend_to_u64x2_high(u32x4 v)
Definition: vector_neon.h:146
static_always_inline u32x4 u32x4_hadd(u32x4 v1, u32x4 v2)
Definition: vector_neon.h:134
#define static_always_inline
Definition: clib.h:95
#define always_inline
Definition: clib.h:94
unsigned int u32
Definition: types.h:88
static u32 u32x4_zero_byte_mask(u32x4 input)
Definition: vector_neon.h:72
epu8_epi32 epu16_epi32 u64x2
Definition: vector_sse42.h:640
static u32 u16x8_zero_byte_mask(u16x8 input)
Definition: vector_neon.h:40
static u32 u8x16_zero_byte_mask(u8x16 input)
Definition: vector_neon.h:66
#define v
Definition: acl.c:496
static_always_inline u64x2 u32x4_extend_to_u64x2(u32x4 v)
Definition: vector_neon.h:140
#define foreach_neon_vec128u
Definition: vector_neon.h:86
static_always_inline u8x16 u8x16_shuffle(u8x16 v, u8x16 m)
Definition: vector_neon.h:128
foreach_neon_vec128i foreach_neon_vec128u static_always_inline u16x8 u16x8_byte_swap(u16x8 v)
Definition: vector_neon.h:122
unsigned long long u32x4
Definition: ixge.c:28
static u32 u64x2_zero_byte_mask(u64x2 input)
Definition: vector_neon.h:78
static u32 u8x16_compare_byte_mask(u8x16 x)
Definition: vector_neon.h:26