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
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
|
/*
* Copyright (c) 2014 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <arm_neon.h>
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/arm/mem_neon.h"
#include "vpx_dsp/arm/sum_neon.h"
uint32_t vpx_sad4x4_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride) {
const uint8x16_t src_u8 = load_unaligned_u8q(src_ptr, src_stride);
const uint8x16_t ref_u8 = load_unaligned_u8q(ref_ptr, ref_stride);
uint16x8_t abs = vabdl_u8(vget_low_u8(src_u8), vget_low_u8(ref_u8));
abs = vabal_u8(abs, vget_high_u8(src_u8), vget_high_u8(ref_u8));
return horizontal_add_uint16x8(abs);
}
uint32_t vpx_sad4x4_avg_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const uint8_t *second_pred) {
const uint8x16_t src_u8 = load_unaligned_u8q(src_ptr, src_stride);
const uint8x16_t ref_u8 = load_unaligned_u8q(ref_ptr, ref_stride);
const uint8x16_t second_pred_u8 = vld1q_u8(second_pred);
const uint8x16_t avg = vrhaddq_u8(ref_u8, second_pred_u8);
uint16x8_t abs = vabdl_u8(vget_low_u8(src_u8), vget_low_u8(avg));
abs = vabal_u8(abs, vget_high_u8(src_u8), vget_high_u8(avg));
return horizontal_add_uint16x8(abs);
}
uint32_t vpx_sad4x8_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride) {
int i;
uint16x8_t abs = vdupq_n_u16(0);
for (i = 0; i < 8; i += 4) {
const uint8x16_t src_u8 = load_unaligned_u8q(src_ptr, src_stride);
const uint8x16_t ref_u8 = load_unaligned_u8q(ref_ptr, ref_stride);
src_ptr += 4 * src_stride;
ref_ptr += 4 * ref_stride;
abs = vabal_u8(abs, vget_low_u8(src_u8), vget_low_u8(ref_u8));
abs = vabal_u8(abs, vget_high_u8(src_u8), vget_high_u8(ref_u8));
}
return horizontal_add_uint16x8(abs);
}
uint32_t vpx_sad4x8_avg_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const uint8_t *second_pred) {
int i;
uint16x8_t abs = vdupq_n_u16(0);
for (i = 0; i < 8; i += 4) {
const uint8x16_t src_u8 = load_unaligned_u8q(src_ptr, src_stride);
const uint8x16_t ref_u8 = load_unaligned_u8q(ref_ptr, ref_stride);
const uint8x16_t second_pred_u8 = vld1q_u8(second_pred);
const uint8x16_t avg = vrhaddq_u8(ref_u8, second_pred_u8);
src_ptr += 4 * src_stride;
ref_ptr += 4 * ref_stride;
second_pred += 16;
abs = vabal_u8(abs, vget_low_u8(src_u8), vget_low_u8(avg));
abs = vabal_u8(abs, vget_high_u8(src_u8), vget_high_u8(avg));
}
return horizontal_add_uint16x8(abs);
}
static INLINE uint16x8_t sad8x(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const int height) {
int i;
uint16x8_t abs = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
const uint8x8_t a_u8 = vld1_u8(src_ptr);
const uint8x8_t b_u8 = vld1_u8(ref_ptr);
src_ptr += src_stride;
ref_ptr += ref_stride;
abs = vabal_u8(abs, a_u8, b_u8);
}
return abs;
}
static INLINE uint16x8_t sad8x_avg(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const uint8_t *second_pred,
const int height) {
int i;
uint16x8_t abs = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
const uint8x8_t a_u8 = vld1_u8(src_ptr);
const uint8x8_t b_u8 = vld1_u8(ref_ptr);
const uint8x8_t c_u8 = vld1_u8(second_pred);
const uint8x8_t avg = vrhadd_u8(b_u8, c_u8);
src_ptr += src_stride;
ref_ptr += ref_stride;
second_pred += 8;
abs = vabal_u8(abs, a_u8, avg);
}
return abs;
}
#define SAD8XN(n) \
uint32_t vpx_sad8x##n##_neon(const uint8_t *src_ptr, int src_stride, \
const uint8_t *ref_ptr, int ref_stride) { \
const uint16x8_t abs = sad8x(src_ptr, src_stride, ref_ptr, ref_stride, n); \
return horizontal_add_uint16x8(abs); \
} \
\
uint32_t vpx_sad8x##n##_avg_neon(const uint8_t *src_ptr, int src_stride, \
const uint8_t *ref_ptr, int ref_stride, \
const uint8_t *second_pred) { \
const uint16x8_t abs = \
sad8x_avg(src_ptr, src_stride, ref_ptr, ref_stride, second_pred, n); \
return horizontal_add_uint16x8(abs); \
}
SAD8XN(4)
SAD8XN(8)
SAD8XN(16)
static INLINE uint16x8_t sad16x(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const int height) {
int i;
uint16x8_t abs = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
const uint8x16_t a_u8 = vld1q_u8(src_ptr);
const uint8x16_t b_u8 = vld1q_u8(ref_ptr);
src_ptr += src_stride;
ref_ptr += ref_stride;
abs = vabal_u8(abs, vget_low_u8(a_u8), vget_low_u8(b_u8));
abs = vabal_u8(abs, vget_high_u8(a_u8), vget_high_u8(b_u8));
}
return abs;
}
static INLINE uint16x8_t sad16x_avg(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const uint8_t *second_pred,
const int height) {
int i;
uint16x8_t abs = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
const uint8x16_t a_u8 = vld1q_u8(src_ptr);
const uint8x16_t b_u8 = vld1q_u8(ref_ptr);
const uint8x16_t c_u8 = vld1q_u8(second_pred);
const uint8x16_t avg = vrhaddq_u8(b_u8, c_u8);
src_ptr += src_stride;
ref_ptr += ref_stride;
second_pred += 16;
abs = vabal_u8(abs, vget_low_u8(a_u8), vget_low_u8(avg));
abs = vabal_u8(abs, vget_high_u8(a_u8), vget_high_u8(avg));
}
return abs;
}
#define SAD16XN(n) \
uint32_t vpx_sad16x##n##_neon(const uint8_t *src_ptr, int src_stride, \
const uint8_t *ref_ptr, int ref_stride) { \
const uint16x8_t abs = \
sad16x(src_ptr, src_stride, ref_ptr, ref_stride, n); \
return horizontal_add_uint16x8(abs); \
} \
\
uint32_t vpx_sad16x##n##_avg_neon(const uint8_t *src_ptr, int src_stride, \
const uint8_t *ref_ptr, int ref_stride, \
const uint8_t *second_pred) { \
const uint16x8_t abs = \
sad16x_avg(src_ptr, src_stride, ref_ptr, ref_stride, second_pred, n); \
return horizontal_add_uint16x8(abs); \
}
SAD16XN(8)
SAD16XN(16)
SAD16XN(32)
static INLINE uint16x8_t sad32x(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const int height) {
int i;
uint16x8_t abs = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
const uint8x16_t a_lo = vld1q_u8(src_ptr);
const uint8x16_t a_hi = vld1q_u8(src_ptr + 16);
const uint8x16_t b_lo = vld1q_u8(ref_ptr);
const uint8x16_t b_hi = vld1q_u8(ref_ptr + 16);
src_ptr += src_stride;
ref_ptr += ref_stride;
abs = vabal_u8(abs, vget_low_u8(a_lo), vget_low_u8(b_lo));
abs = vabal_u8(abs, vget_high_u8(a_lo), vget_high_u8(b_lo));
abs = vabal_u8(abs, vget_low_u8(a_hi), vget_low_u8(b_hi));
abs = vabal_u8(abs, vget_high_u8(a_hi), vget_high_u8(b_hi));
}
return abs;
}
static INLINE uint16x8_t sad32x_avg(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const uint8_t *second_pred,
const int height) {
int i;
uint16x8_t abs = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
const uint8x16_t a_lo = vld1q_u8(src_ptr);
const uint8x16_t a_hi = vld1q_u8(src_ptr + 16);
const uint8x16_t b_lo = vld1q_u8(ref_ptr);
const uint8x16_t b_hi = vld1q_u8(ref_ptr + 16);
const uint8x16_t c_lo = vld1q_u8(second_pred);
const uint8x16_t c_hi = vld1q_u8(second_pred + 16);
const uint8x16_t avg_lo = vrhaddq_u8(b_lo, c_lo);
const uint8x16_t avg_hi = vrhaddq_u8(b_hi, c_hi);
src_ptr += src_stride;
ref_ptr += ref_stride;
second_pred += 32;
abs = vabal_u8(abs, vget_low_u8(a_lo), vget_low_u8(avg_lo));
abs = vabal_u8(abs, vget_high_u8(a_lo), vget_high_u8(avg_lo));
abs = vabal_u8(abs, vget_low_u8(a_hi), vget_low_u8(avg_hi));
abs = vabal_u8(abs, vget_high_u8(a_hi), vget_high_u8(avg_hi));
}
return abs;
}
#define SAD32XN(n) \
uint32_t vpx_sad32x##n##_neon(const uint8_t *src_ptr, int src_stride, \
const uint8_t *ref_ptr, int ref_stride) { \
const uint16x8_t abs = \
sad32x(src_ptr, src_stride, ref_ptr, ref_stride, n); \
return horizontal_add_uint16x8(abs); \
} \
\
uint32_t vpx_sad32x##n##_avg_neon(const uint8_t *src_ptr, int src_stride, \
const uint8_t *ref_ptr, int ref_stride, \
const uint8_t *second_pred) { \
const uint16x8_t abs = \
sad32x_avg(src_ptr, src_stride, ref_ptr, ref_stride, second_pred, n); \
return horizontal_add_uint16x8(abs); \
}
SAD32XN(16)
SAD32XN(32)
SAD32XN(64)
static INLINE uint32x4_t sad64x(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const int height) {
int i;
uint16x8_t abs_0 = vdupq_n_u16(0);
uint16x8_t abs_1 = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
const uint8x16_t a_0 = vld1q_u8(src_ptr);
const uint8x16_t a_1 = vld1q_u8(src_ptr + 16);
const uint8x16_t a_2 = vld1q_u8(src_ptr + 32);
const uint8x16_t a_3 = vld1q_u8(src_ptr + 48);
const uint8x16_t b_0 = vld1q_u8(ref_ptr);
const uint8x16_t b_1 = vld1q_u8(ref_ptr + 16);
const uint8x16_t b_2 = vld1q_u8(ref_ptr + 32);
const uint8x16_t b_3 = vld1q_u8(ref_ptr + 48);
src_ptr += src_stride;
ref_ptr += ref_stride;
abs_0 = vabal_u8(abs_0, vget_low_u8(a_0), vget_low_u8(b_0));
abs_0 = vabal_u8(abs_0, vget_high_u8(a_0), vget_high_u8(b_0));
abs_0 = vabal_u8(abs_0, vget_low_u8(a_1), vget_low_u8(b_1));
abs_0 = vabal_u8(abs_0, vget_high_u8(a_1), vget_high_u8(b_1));
abs_1 = vabal_u8(abs_1, vget_low_u8(a_2), vget_low_u8(b_2));
abs_1 = vabal_u8(abs_1, vget_high_u8(a_2), vget_high_u8(b_2));
abs_1 = vabal_u8(abs_1, vget_low_u8(a_3), vget_low_u8(b_3));
abs_1 = vabal_u8(abs_1, vget_high_u8(a_3), vget_high_u8(b_3));
}
{
const uint32x4_t sum = vpaddlq_u16(abs_0);
return vpadalq_u16(sum, abs_1);
}
}
static INLINE uint32x4_t sad64x_avg(const uint8_t *src_ptr, int src_stride,
const uint8_t *ref_ptr, int ref_stride,
const uint8_t *second_pred,
const int height) {
int i;
uint16x8_t abs_0 = vdupq_n_u16(0);
uint16x8_t abs_1 = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
const uint8x16_t a_0 = vld1q_u8(src_ptr);
const uint8x16_t a_1 = vld1q_u8(src_ptr + 16);
const uint8x16_t a_2 = vld1q_u8(src_ptr + 32);
const uint8x16_t a_3 = vld1q_u8(src_ptr + 48);
const uint8x16_t b_0 = vld1q_u8(ref_ptr);
const uint8x16_t b_1 = vld1q_u8(ref_ptr + 16);
const uint8x16_t b_2 = vld1q_u8(ref_ptr + 32);
const uint8x16_t b_3 = vld1q_u8(ref_ptr + 48);
const uint8x16_t c_0 = vld1q_u8(second_pred);
const uint8x16_t c_1 = vld1q_u8(second_pred + 16);
const uint8x16_t c_2 = vld1q_u8(second_pred + 32);
const uint8x16_t c_3 = vld1q_u8(second_pred + 48);
const uint8x16_t avg_0 = vrhaddq_u8(b_0, c_0);
const uint8x16_t avg_1 = vrhaddq_u8(b_1, c_1);
const uint8x16_t avg_2 = vrhaddq_u8(b_2, c_2);
const uint8x16_t avg_3 = vrhaddq_u8(b_3, c_3);
src_ptr += src_stride;
ref_ptr += ref_stride;
second_pred += 64;
abs_0 = vabal_u8(abs_0, vget_low_u8(a_0), vget_low_u8(avg_0));
abs_0 = vabal_u8(abs_0, vget_high_u8(a_0), vget_high_u8(avg_0));
abs_0 = vabal_u8(abs_0, vget_low_u8(a_1), vget_low_u8(avg_1));
abs_0 = vabal_u8(abs_0, vget_high_u8(a_1), vget_high_u8(avg_1));
abs_1 = vabal_u8(abs_1, vget_low_u8(a_2), vget_low_u8(avg_2));
abs_1 = vabal_u8(abs_1, vget_high_u8(a_2), vget_high_u8(avg_2));
abs_1 = vabal_u8(abs_1, vget_low_u8(a_3), vget_low_u8(avg_3));
abs_1 = vabal_u8(abs_1, vget_high_u8(a_3), vget_high_u8(avg_3));
}
{
const uint32x4_t sum = vpaddlq_u16(abs_0);
return vpadalq_u16(sum, abs_1);
}
}
#define SAD64XN(n) \
uint32_t vpx_sad64x##n##_neon(const uint8_t *src_ptr, int src_stride, \
const uint8_t *ref_ptr, int ref_stride) { \
const uint32x4_t abs = \
sad64x(src_ptr, src_stride, ref_ptr, ref_stride, n); \
return horizontal_add_uint32x4(abs); \
} \
\
uint32_t vpx_sad64x##n##_avg_neon(const uint8_t *src_ptr, int src_stride, \
const uint8_t *ref_ptr, int ref_stride, \
const uint8_t *second_pred) { \
const uint32x4_t abs = \
sad64x_avg(src_ptr, src_stride, ref_ptr, ref_stride, second_pred, n); \
return horizontal_add_uint32x4(abs); \
}
SAD64XN(32)
SAD64XN(64)
|
