aboutsummaryrefslogtreecommitdiffstats
path: root/src/vppinfra/vector_neon.h
blob: 81d99a64f0509933b8a205f6219f12386aa101c6 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
/*
 * Copyright (c) 2015 Cisco and/or its affiliates.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at:
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#ifndef included_vector_neon_h
#define included_vector_neon_h
#include <arm_neon.h>

/* Arithmetic */
#define u16x8_sub_saturate(a,b) vsubq_u16(a,b)
#define i16x8_sub_saturate(a,b) vsubq_s16(a,b)
/* Dummy. Aid making uniform macros */
#define vreinterpretq_u8_u8(a)  a
/* Implement the missing intrinsics to make uniform macros */
#define vminvq_u64(x)   \
({  \
  u64 x0 = vgetq_lane_u64(x, 0);    \
  u64 x1 = vgetq_lane_u64(x, 1);    \
  x0 < x1 ? x0 : x1;    \
})

/* Converts all ones/zeros compare mask to bitmap. */
always_inline u32
u8x16_compare_byte_mask (u8x16 v)
{
  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));
}

/* *INDENT-OFF* */
#define foreach_neon_vec128i \
  _(i,8,16,s8) _(i,16,8,s16) _(i,32,4,s32)  _(i,64,2,s64)
#define foreach_neon_vec128u \
  _(u,8,16,u8) _(u,16,8,u16) _(u,32,4,u32)  _(u,64,2,u64)
#define foreach_neon_vec128f \
  _(f,32,4,f32) _(f,64,2,f64)

#define _(t, s, c, i) \
static_always_inline t##s##x##c						\
t##s##x##c##_splat (t##s x)						\
{ return (t##s##x##c) vdupq_n_##i (x); }				\
\
static_always_inline t##s##x##c						\
t##s##x##c##_load_unaligned (void *p)					\
{ return (t##s##x##c) vld1q_##i (p); }					\
\
static_always_inline void						\
t##s##x##c##_store_unaligned (t##s##x##c v, void *p)			\
{ vst1q_##i (p, v); }							\
\
static_always_inline int						\
t##s##x##c##_is_all_zero (t##s##x##c x)					\
{ return !!(vminvq_u##s (vceqq_##i (vdupq_n_##i(0), x))); }						\
\
static_always_inline int						\
t##s##x##c##_is_equal (t##s##x##c a, t##s##x##c b)			\
{ return !!(vminvq_u##s (vceqq_##i (a, b))); }				\
\
static_always_inline int						\
t##s##x##c##_is_all_equal (t##s##x##c v, t##s x)			\
{ return t##s##x##c##_is_equal (v, t##s##x##c##_splat (x)); };		\
\
static_always_inline u32						\
t##s##x##c##_zero_byte_mask (t##s##x##c x)			\
{ uint8x16_t v = vreinterpretq_u8_u##s (vceqq_##i (vdupq_n_##i(0), x));  \
  return u8x16_compare_byte_mask (v); } \
\
static_always_inline u##s##x##c						\
t##s##x##c##_is_greater (t##s##x##c a, t##s##x##c b)			\
{ return (u##s##x##c) vcgtq_##i (a, b); }				\
\
static_always_inline t##s##x##c						\
t##s##x##c##_blend (t##s##x##c dst, t##s##x##c src, u##s##x##c mask)	\
{ return (t##s##x##c) vbslq_##i (mask, src, dst); }

foreach_neon_vec128i foreach_neon_vec128u

#undef _
/* *INDENT-ON* */

static_always_inline u16x8
u16x8_byte_swap (u16x8 v)
{
  return (u16x8) vrev16q_u8 ((u8x16) v);
}

static_always_inline u8x16
u8x16_shuffle (u8x16 v, u8x16 m)
{
  return (u8x16) vqtbl1q_u8 (v, m);
}

static_always_inline u32x4
u32x4_hadd (u32x4 v1, u32x4 v2)
{
  return (u32x4) vpaddq_u32 (v1, v2);
}

static_always_inline u64x2
u32x4_extend_to_u64x2 (u32x4 v)
{
  return vmovl_u32 (vget_low_u32 (v));
}

static_always_inline u64x2
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));
}

static_always_inline u64x2
u64x2_gather (void *p0, void *p1)
{
  u64x2 r = vdupq_n_u64 (*(u64 *) p0);
  r = vsetq_lane_u64 (*(u64 *) p1, r, 1);
  return r;
}

static_always_inline u32x4
u32x4_gather (void *p0, void *p1, void *p2, void *p3)
{
  u32x4 r = vdupq_n_u32 (*(u32 *) p0);
  r = vsetq_lane_u32 (*(u32 *) p1, r, 1);
  r = vsetq_lane_u32 (*(u32 *) p2, r, 2);
  r = vsetq_lane_u32 (*(u32 *) p3, r, 3);
  return r;
}

static_always_inline void
u64x2_scatter (u64x2 r, void *p0, void *p1)
{
  *(u64 *) p0 = vgetq_lane_u64 (r, 0);
  *(u64 *) p1 = vgetq_lane_u64 (r, 1);
}

static_always_inline void
u32x4_scatter (u32x4 r, void *p0, void *p1, void *p2, void *p3)
{
  *(u32 *) p0 = vgetq_lane_u32 (r, 0);
  *(u32 *) p1 = vgetq_lane_u32 (r, 1);
  *(u32 *) p2 = vgetq_lane_u32 (r, 2);
  *(u32 *) p3 = vgetq_lane_u32 (r, 3);
}

static_always_inline u32
u32x4_min_scalar (u32x4 v)
{
  return vminvq_u32 (v);
}

static_always_inline u8x16
u8x16_word_shift_left (u8x16 x, const int n)
{
  return vextq_u8 (u8x16_splat (0), x, 16 - n);
}

static_always_inline u8x16
u8x16_word_shift_right (u8x16 x, const int n)
{
  return vextq_u8 (x, u8x16_splat (0), n);
}

static_always_inline u8x16
u8x16_reflect (u8x16 v)
{
  u8x16 mask = {
    15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
  };
  return (u8x16) vqtbl1q_u8 (v, mask);
}

#define CLIB_HAVE_VEC128_MSB_MASK

#define CLIB_HAVE_VEC128_UNALIGNED_LOAD_STORE
#define CLIB_VEC128_SPLAT_DEFINED
#endif /* included_vector_neon_h */

/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */