FD.io VPP  v18.04-17-g3a0d853
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 always_inline int
26 {
27  return !(vaddvq_u8 (x));
28 }
29 
30 always_inline int
32 {
33  return !(vaddvq_u16 (x));
34 }
35 
36 always_inline int
38 {
39  return !(vaddvq_u32 (x));
40 }
41 
42 always_inline int
44 {
45  return !(vaddvq_u64 (x));
46 }
47 
48 /* Converts all ones/zeros compare mask to bitmap. */
51 {
52  uint8x16_t mask_shift =
53  { -7, -6, -5, -4, -3, -2, -1, 0, -7, -6, -5, -4, -3, -2, -1, 0 };
54  uint8x16_t mask_and = vdupq_n_u8 (0x80);
55  x = vandq_u8 (x, mask_and);
56  x = vshlq_u8 (x, vreinterpretq_s8_u8 (mask_shift));
57  x = vpaddq_u8 (x, x);
58  x = vpaddq_u8 (x, x);
59  x = vpaddq_u8 (x, x);
60  return vgetq_lane_u8 (x, 0) | (vgetq_lane_u8 (x, 1) << 8);
61 }
62 
64 u16x8_zero_byte_mask (u16x8 input)
65 {
66  u8x16 vall_one = vdupq_n_u8 (0x0);
67  u8x16 res_values = { 0x01, 0x02, 0x04, 0x08,
68  0x10, 0x20, 0x40, 0x80,
69  0x01, 0x02, 0x04, 0x08,
70  0x10, 0x20, 0x40, 0x80
71  };
72 
73  /* input --> [0x80, 0x40, 0x01, 0xf0, ... ] */
74  u8x16 test_result =
75  vreinterpretq_u8_u16 (vceqq_u16 (input, vreinterpretq_u16_u8 (vall_one)));
76  u8x16 before_merge = vminq_u8 (test_result, res_values);
77  /*before_merge--> [0x80, 0x00, 0x00, 0x10, ... ] */
78  /* u8x16 --> [a,b,c,d, e,f,g,h, i,j,k,l, m,n,o,p] */
79  /* pair add until we have 2 uint64_t */
80  u16x8 merge1 = vpaddlq_u8 (before_merge);
81  /* u16x8--> [a+b,c+d, e+f,g+h, i+j,k+l, m+n,o+p] */
82  u32x4 merge2 = vpaddlq_u16 (merge1);
83  /* u32x4--> [a+b+c+d, e+f+g+h, i+j+k+l, m+n+o+p] */
84  u64x2 merge3 = vpaddlq_u32 (merge2);
85  /* u64x2--> [a+b+c+d+e+f+g+h, i+j+k+l+m+n+o+p] */
86  return (u32) (vgetq_lane_u64 (merge3, 1) << 8) + vgetq_lane_u64 (merge3, 0);
87 }
88 
90 u8x16_zero_byte_mask (u8x16 input)
91 {
92  return u16x8_zero_byte_mask ((u16x8) input);
93 }
94 
97 {
98  return u16x8_zero_byte_mask ((u16x8) input);
99 }
100 
102 u64x2_zero_byte_mask (u64x2 input)
103 {
104  return u16x8_zero_byte_mask ((u16x8) input);
105 }
106 
107 
108 
109 #endif /* included_vector_neon_h */
110 
111 /*
112  * fd.io coding-style-patch-verification: ON
113  *
114  * Local Variables:
115  * eval: (c-set-style "gnu")
116  * End:
117  */
static int u16x8_is_all_zero(u16x8 x)
Definition: vector_neon.h:31
static int u8x16_is_all_zero(u8x16 x)
Definition: vector_neon.h:25
#define always_inline
Definition: clib.h:92
unsigned long long u32x4
Definition: ixge.c:28
static int u64x2_is_all_zero(u64x2 x)
Definition: vector_neon.h:43
static u32 u32x4_zero_byte_mask(u32x4 input)
Definition: vector_neon.h:96
static u32 u16x8_zero_byte_mask(u16x8 input)
Definition: vector_neon.h:64
static u32 u8x16_zero_byte_mask(u8x16 input)
Definition: vector_neon.h:90
unsigned int u32
Definition: types.h:88
static int u32x4_is_all_zero(u32x4 x)
Definition: vector_neon.h:37
static u32 u64x2_zero_byte_mask(u64x2 input)
Definition: vector_neon.h:102
static u32 u8x16_compare_byte_mask(u8x16 x)
Definition: vector_neon.h:50