mbloopfilter_neon.c revision 7bc9febe8749e98a3812a0dc4380ceae75c29450 
1/*
2 *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
3 *
4 *  Use of this source code is governed by a BSD-style license
5 *  that can be found in the LICENSE file in the root of the source
6 *  tree. An additional intellectual property rights grant can be found
7 *  in the file PATENTS.  All contributing project authors may
8 *  be found in the AUTHORS file in the root of the source tree.
9 */
10
11#include <arm_neon.h>
12#include "./vpx_config.h"
13
14static INLINE void vp8_mbloop_filter_neon(uint8x16_t qblimit,  // mblimit
15                                          uint8x16_t qlimit,   // limit
16                                          uint8x16_t qthresh,  // thresh
17                                          uint8x16_t q3,       // p2
18                                          uint8x16_t q4,       // p2
19                                          uint8x16_t q5,       // p1
20                                          uint8x16_t q6,       // p0
21                                          uint8x16_t q7,       // q0
22                                          uint8x16_t q8,       // q1
23                                          uint8x16_t q9,       // q2
24                                          uint8x16_t q10,      // q3
25                                          uint8x16_t *q4r,     // p1
26                                          uint8x16_t *q5r,     // p1
27                                          uint8x16_t *q6r,     // p0
28                                          uint8x16_t *q7r,     // q0
29                                          uint8x16_t *q8r,     // q1
30                                          uint8x16_t *q9r) {   // q1
31  uint8x16_t q0u8, q1u8, q11u8, q12u8, q13u8, q14u8, q15u8;
32  int16x8_t q0s16, q2s16, q11s16, q12s16, q13s16, q14s16, q15s16;
33  int8x16_t q1s8, q6s8, q7s8, q2s8, q11s8, q13s8;
34  uint16x8_t q0u16, q11u16, q12u16, q13u16, q14u16, q15u16;
35  int8x16_t q0s8, q12s8, q14s8, q15s8;
36  int8x8_t d0, d1, d2, d3, d4, d5, d24, d25, d28, d29;
37
38  q11u8 = vabdq_u8(q3, q4);
39  q12u8 = vabdq_u8(q4, q5);
40  q13u8 = vabdq_u8(q5, q6);
41  q14u8 = vabdq_u8(q8, q7);
42  q1u8 = vabdq_u8(q9, q8);
43  q0u8 = vabdq_u8(q10, q9);
44
45  q11u8 = vmaxq_u8(q11u8, q12u8);
46  q12u8 = vmaxq_u8(q13u8, q14u8);
47  q1u8 = vmaxq_u8(q1u8, q0u8);
48  q15u8 = vmaxq_u8(q11u8, q12u8);
49
50  q12u8 = vabdq_u8(q6, q7);
51
52  // vp8_hevmask
53  q13u8 = vcgtq_u8(q13u8, qthresh);
54  q14u8 = vcgtq_u8(q14u8, qthresh);
55  q15u8 = vmaxq_u8(q15u8, q1u8);
56
57  q15u8 = vcgeq_u8(qlimit, q15u8);
58
59  q1u8 = vabdq_u8(q5, q8);
60  q12u8 = vqaddq_u8(q12u8, q12u8);
61
62  // vp8_filter() function
63  // convert to signed
64  q0u8 = vdupq_n_u8(0x80);
65  q9 = veorq_u8(q9, q0u8);
66  q8 = veorq_u8(q8, q0u8);
67  q7 = veorq_u8(q7, q0u8);
68  q6 = veorq_u8(q6, q0u8);
69  q5 = veorq_u8(q5, q0u8);
70  q4 = veorq_u8(q4, q0u8);
71
72  q1u8 = vshrq_n_u8(q1u8, 1);
73  q12u8 = vqaddq_u8(q12u8, q1u8);
74
75  q14u8 = vorrq_u8(q13u8, q14u8);
76  q12u8 = vcgeq_u8(qblimit, q12u8);
77
78  q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)),
79                   vget_low_s8(vreinterpretq_s8_u8(q6)));
80  q13s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)),
81                    vget_high_s8(vreinterpretq_s8_u8(q6)));
82
83  q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5), vreinterpretq_s8_u8(q8));
84
85  q11s16 = vdupq_n_s16(3);
86  q2s16 = vmulq_s16(q2s16, q11s16);
87  q13s16 = vmulq_s16(q13s16, q11s16);
88
89  q15u8 = vandq_u8(q15u8, q12u8);
90
91  q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8));
92  q13s16 = vaddw_s8(q13s16, vget_high_s8(q1s8));
93
94  q12u8 = vdupq_n_u8(3);
95  q11u8 = vdupq_n_u8(4);
96  // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
97  d2 = vqmovn_s16(q2s16);
98  d3 = vqmovn_s16(q13s16);
99  q1s8 = vcombine_s8(d2, d3);
100  q1s8 = vandq_s8(q1s8, vreinterpretq_s8_u8(q15u8));
101  q13s8 = vandq_s8(q1s8, vreinterpretq_s8_u8(q14u8));
102
103  q2s8 = vqaddq_s8(q13s8, vreinterpretq_s8_u8(q11u8));
104  q13s8 = vqaddq_s8(q13s8, vreinterpretq_s8_u8(q12u8));
105  q2s8 = vshrq_n_s8(q2s8, 3);
106  q13s8 = vshrq_n_s8(q13s8, 3);
107
108  q7s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q2s8);
109  q6s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q13s8);
110
111  q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8));
112
113  q0u16 = q11u16 = q12u16 = q13u16 = q14u16 = q15u16 = vdupq_n_u16(63);
114  d5 = vdup_n_s8(9);
115  d4 = vdup_n_s8(18);
116
117  q0s16 = vmlal_s8(vreinterpretq_s16_u16(q0u16), vget_low_s8(q1s8), d5);
118  q11s16 = vmlal_s8(vreinterpretq_s16_u16(q11u16), vget_high_s8(q1s8), d5);
119  d5 = vdup_n_s8(27);
120  q12s16 = vmlal_s8(vreinterpretq_s16_u16(q12u16), vget_low_s8(q1s8), d4);
121  q13s16 = vmlal_s8(vreinterpretq_s16_u16(q13u16), vget_high_s8(q1s8), d4);
122  q14s16 = vmlal_s8(vreinterpretq_s16_u16(q14u16), vget_low_s8(q1s8), d5);
123  q15s16 = vmlal_s8(vreinterpretq_s16_u16(q15u16), vget_high_s8(q1s8), d5);
124
125  d0 = vqshrn_n_s16(q0s16, 7);
126  d1 = vqshrn_n_s16(q11s16, 7);
127  d24 = vqshrn_n_s16(q12s16, 7);
128  d25 = vqshrn_n_s16(q13s16, 7);
129  d28 = vqshrn_n_s16(q14s16, 7);
130  d29 = vqshrn_n_s16(q15s16, 7);
131
132  q0s8 = vcombine_s8(d0, d1);
133  q12s8 = vcombine_s8(d24, d25);
134  q14s8 = vcombine_s8(d28, d29);
135
136  q11s8 = vqsubq_s8(vreinterpretq_s8_u8(q9), q0s8);
137  q0s8 = vqaddq_s8(vreinterpretq_s8_u8(q4), q0s8);
138  q13s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q12s8);
139  q12s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q12s8);
140  q15s8 = vqsubq_s8((q7s8), q14s8);
141  q14s8 = vqaddq_s8((q6s8), q14s8);
142
143  q1u8 = vdupq_n_u8(0x80);
144  *q9r = veorq_u8(vreinterpretq_u8_s8(q11s8), q1u8);
145  *q8r = veorq_u8(vreinterpretq_u8_s8(q13s8), q1u8);
146  *q7r = veorq_u8(vreinterpretq_u8_s8(q15s8), q1u8);
147  *q6r = veorq_u8(vreinterpretq_u8_s8(q14s8), q1u8);
148  *q5r = veorq_u8(vreinterpretq_u8_s8(q12s8), q1u8);
149  *q4r = veorq_u8(vreinterpretq_u8_s8(q0s8), q1u8);
150  return;
151}
152
153void vp8_mbloop_filter_horizontal_edge_y_neon(unsigned char *src, int pitch,
154                                              unsigned char blimit,
155                                              unsigned char limit,
156                                              unsigned char thresh) {
157  uint8x16_t qblimit, qlimit, qthresh, q3, q4;
158  uint8x16_t q5, q6, q7, q8, q9, q10;
159
160  qblimit = vdupq_n_u8(blimit);
161  qlimit = vdupq_n_u8(limit);
162  qthresh = vdupq_n_u8(thresh);
163
164  src -= (pitch << 2);
165
166  q3 = vld1q_u8(src);
167  src += pitch;
168  q4 = vld1q_u8(src);
169  src += pitch;
170  q5 = vld1q_u8(src);
171  src += pitch;
172  q6 = vld1q_u8(src);
173  src += pitch;
174  q7 = vld1q_u8(src);
175  src += pitch;
176  q8 = vld1q_u8(src);
177  src += pitch;
178  q9 = vld1q_u8(src);
179  src += pitch;
180  q10 = vld1q_u8(src);
181
182  vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, q5, q6, q7, q8, q9,
183                         q10, &q4, &q5, &q6, &q7, &q8, &q9);
184
185  src -= (pitch * 6);
186  vst1q_u8(src, q4);
187  src += pitch;
188  vst1q_u8(src, q5);
189  src += pitch;
190  vst1q_u8(src, q6);
191  src += pitch;
192  vst1q_u8(src, q7);
193  src += pitch;
194  vst1q_u8(src, q8);
195  src += pitch;
196  vst1q_u8(src, q9);
197  return;
198}
199
200void vp8_mbloop_filter_horizontal_edge_uv_neon(unsigned char *u, int pitch,
201                                               unsigned char blimit,
202                                               unsigned char limit,
203                                               unsigned char thresh,
204                                               unsigned char *v) {
205  uint8x16_t qblimit, qlimit, qthresh, q3, q4;
206  uint8x16_t q5, q6, q7, q8, q9, q10;
207  uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
208  uint8x8_t d15, d16, d17, d18, d19, d20, d21;
209
210  qblimit = vdupq_n_u8(blimit);
211  qlimit = vdupq_n_u8(limit);
212  qthresh = vdupq_n_u8(thresh);
213
214  u -= (pitch << 2);
215  v -= (pitch << 2);
216
217  d6 = vld1_u8(u);
218  u += pitch;
219  d7 = vld1_u8(v);
220  v += pitch;
221  d8 = vld1_u8(u);
222  u += pitch;
223  d9 = vld1_u8(v);
224  v += pitch;
225  d10 = vld1_u8(u);
226  u += pitch;
227  d11 = vld1_u8(v);
228  v += pitch;
229  d12 = vld1_u8(u);
230  u += pitch;
231  d13 = vld1_u8(v);
232  v += pitch;
233  d14 = vld1_u8(u);
234  u += pitch;
235  d15 = vld1_u8(v);
236  v += pitch;
237  d16 = vld1_u8(u);
238  u += pitch;
239  d17 = vld1_u8(v);
240  v += pitch;
241  d18 = vld1_u8(u);
242  u += pitch;
243  d19 = vld1_u8(v);
244  v += pitch;
245  d20 = vld1_u8(u);
246  d21 = vld1_u8(v);
247
248  q3 = vcombine_u8(d6, d7);
249  q4 = vcombine_u8(d8, d9);
250  q5 = vcombine_u8(d10, d11);
251  q6 = vcombine_u8(d12, d13);
252  q7 = vcombine_u8(d14, d15);
253  q8 = vcombine_u8(d16, d17);
254  q9 = vcombine_u8(d18, d19);
255  q10 = vcombine_u8(d20, d21);
256
257  vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, q5, q6, q7, q8, q9,
258                         q10, &q4, &q5, &q6, &q7, &q8, &q9);
259
260  u -= (pitch * 6);
261  v -= (pitch * 6);
262  vst1_u8(u, vget_low_u8(q4));
263  u += pitch;
264  vst1_u8(v, vget_high_u8(q4));
265  v += pitch;
266  vst1_u8(u, vget_low_u8(q5));
267  u += pitch;
268  vst1_u8(v, vget_high_u8(q5));
269  v += pitch;
270  vst1_u8(u, vget_low_u8(q6));
271  u += pitch;
272  vst1_u8(v, vget_high_u8(q6));
273  v += pitch;
274  vst1_u8(u, vget_low_u8(q7));
275  u += pitch;
276  vst1_u8(v, vget_high_u8(q7));
277  v += pitch;
278  vst1_u8(u, vget_low_u8(q8));
279  u += pitch;
280  vst1_u8(v, vget_high_u8(q8));
281  v += pitch;
282  vst1_u8(u, vget_low_u8(q9));
283  vst1_u8(v, vget_high_u8(q9));
284  return;
285}
286
287void vp8_mbloop_filter_vertical_edge_y_neon(unsigned char *src, int pitch,
288                                            unsigned char blimit,
289                                            unsigned char limit,
290                                            unsigned char thresh) {
291  unsigned char *s1, *s2;
292  uint8x16_t qblimit, qlimit, qthresh, q3, q4;
293  uint8x16_t q5, q6, q7, q8, q9, q10;
294  uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
295  uint8x8_t d15, d16, d17, d18, d19, d20, d21;
296  uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3;
297  uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7;
298  uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11;
299
300  qblimit = vdupq_n_u8(blimit);
301  qlimit = vdupq_n_u8(limit);
302  qthresh = vdupq_n_u8(thresh);
303
304  s1 = src - 4;
305  s2 = s1 + 8 * pitch;
306  d6 = vld1_u8(s1);
307  s1 += pitch;
308  d7 = vld1_u8(s2);
309  s2 += pitch;
310  d8 = vld1_u8(s1);
311  s1 += pitch;
312  d9 = vld1_u8(s2);
313  s2 += pitch;
314  d10 = vld1_u8(s1);
315  s1 += pitch;
316  d11 = vld1_u8(s2);
317  s2 += pitch;
318  d12 = vld1_u8(s1);
319  s1 += pitch;
320  d13 = vld1_u8(s2);
321  s2 += pitch;
322  d14 = vld1_u8(s1);
323  s1 += pitch;
324  d15 = vld1_u8(s2);
325  s2 += pitch;
326  d16 = vld1_u8(s1);
327  s1 += pitch;
328  d17 = vld1_u8(s2);
329  s2 += pitch;
330  d18 = vld1_u8(s1);
331  s1 += pitch;
332  d19 = vld1_u8(s2);
333  s2 += pitch;
334  d20 = vld1_u8(s1);
335  d21 = vld1_u8(s2);
336
337  q3 = vcombine_u8(d6, d7);
338  q4 = vcombine_u8(d8, d9);
339  q5 = vcombine_u8(d10, d11);
340  q6 = vcombine_u8(d12, d13);
341  q7 = vcombine_u8(d14, d15);
342  q8 = vcombine_u8(d16, d17);
343  q9 = vcombine_u8(d18, d19);
344  q10 = vcombine_u8(d20, d21);
345
346  q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7));
347  q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8));
348  q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9));
349  q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10));
350
351  q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]),
352                     vreinterpretq_u16_u32(q2tmp2.val[0]));
353  q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]),
354                     vreinterpretq_u16_u32(q2tmp3.val[0]));
355  q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]),
356                     vreinterpretq_u16_u32(q2tmp2.val[1]));
357  q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]),
358                     vreinterpretq_u16_u32(q2tmp3.val[1]));
359
360  q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]),
361                    vreinterpretq_u8_u16(q2tmp5.val[0]));
362  q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]),
363                    vreinterpretq_u8_u16(q2tmp5.val[1]));
364  q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]),
365                     vreinterpretq_u8_u16(q2tmp7.val[0]));
366  q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]),
367                     vreinterpretq_u8_u16(q2tmp7.val[1]));
368
369  q3 = q2tmp8.val[0];
370  q4 = q2tmp8.val[1];
371  q5 = q2tmp9.val[0];
372  q6 = q2tmp9.val[1];
373  q7 = q2tmp10.val[0];
374  q8 = q2tmp10.val[1];
375  q9 = q2tmp11.val[0];
376  q10 = q2tmp11.val[1];
377
378  vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, q5, q6, q7, q8, q9,
379                         q10, &q4, &q5, &q6, &q7, &q8, &q9);
380
381  q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7));
382  q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8));
383  q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9));
384  q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10));
385
386  q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]),
387                     vreinterpretq_u16_u32(q2tmp2.val[0]));
388  q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]),
389                     vreinterpretq_u16_u32(q2tmp3.val[0]));
390  q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]),
391                     vreinterpretq_u16_u32(q2tmp2.val[1]));
392  q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]),
393                     vreinterpretq_u16_u32(q2tmp3.val[1]));
394
395  q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]),
396                    vreinterpretq_u8_u16(q2tmp5.val[0]));
397  q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]),
398                    vreinterpretq_u8_u16(q2tmp5.val[1]));
399  q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]),
400                     vreinterpretq_u8_u16(q2tmp7.val[0]));
401  q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]),
402                     vreinterpretq_u8_u16(q2tmp7.val[1]));
403
404  q3 = q2tmp8.val[0];
405  q4 = q2tmp8.val[1];
406  q5 = q2tmp9.val[0];
407  q6 = q2tmp9.val[1];
408  q7 = q2tmp10.val[0];
409  q8 = q2tmp10.val[1];
410  q9 = q2tmp11.val[0];
411  q10 = q2tmp11.val[1];
412
413  s1 -= 7 * pitch;
414  s2 -= 7 * pitch;
415
416  vst1_u8(s1, vget_low_u8(q3));
417  s1 += pitch;
418  vst1_u8(s2, vget_high_u8(q3));
419  s2 += pitch;
420  vst1_u8(s1, vget_low_u8(q4));
421  s1 += pitch;
422  vst1_u8(s2, vget_high_u8(q4));
423  s2 += pitch;
424  vst1_u8(s1, vget_low_u8(q5));
425  s1 += pitch;
426  vst1_u8(s2, vget_high_u8(q5));
427  s2 += pitch;
428  vst1_u8(s1, vget_low_u8(q6));
429  s1 += pitch;
430  vst1_u8(s2, vget_high_u8(q6));
431  s2 += pitch;
432  vst1_u8(s1, vget_low_u8(q7));
433  s1 += pitch;
434  vst1_u8(s2, vget_high_u8(q7));
435  s2 += pitch;
436  vst1_u8(s1, vget_low_u8(q8));
437  s1 += pitch;
438  vst1_u8(s2, vget_high_u8(q8));
439  s2 += pitch;
440  vst1_u8(s1, vget_low_u8(q9));
441  s1 += pitch;
442  vst1_u8(s2, vget_high_u8(q9));
443  s2 += pitch;
444  vst1_u8(s1, vget_low_u8(q10));
445  vst1_u8(s2, vget_high_u8(q10));
446  return;
447}
448
449void vp8_mbloop_filter_vertical_edge_uv_neon(unsigned char *u, int pitch,
450                                             unsigned char blimit,
451                                             unsigned char limit,
452                                             unsigned char thresh,
453                                             unsigned char *v) {
454  unsigned char *us, *ud;
455  unsigned char *vs, *vd;
456  uint8x16_t qblimit, qlimit, qthresh, q3, q4;
457  uint8x16_t q5, q6, q7, q8, q9, q10;
458  uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
459  uint8x8_t d15, d16, d17, d18, d19, d20, d21;
460  uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3;
461  uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7;
462  uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11;
463
464  qblimit = vdupq_n_u8(blimit);
465  qlimit = vdupq_n_u8(limit);
466  qthresh = vdupq_n_u8(thresh);
467
468  us = u - 4;
469  vs = v - 4;
470  d6 = vld1_u8(us);
471  us += pitch;
472  d7 = vld1_u8(vs);
473  vs += pitch;
474  d8 = vld1_u8(us);
475  us += pitch;
476  d9 = vld1_u8(vs);
477  vs += pitch;
478  d10 = vld1_u8(us);
479  us += pitch;
480  d11 = vld1_u8(vs);
481  vs += pitch;
482  d12 = vld1_u8(us);
483  us += pitch;
484  d13 = vld1_u8(vs);
485  vs += pitch;
486  d14 = vld1_u8(us);
487  us += pitch;
488  d15 = vld1_u8(vs);
489  vs += pitch;
490  d16 = vld1_u8(us);
491  us += pitch;
492  d17 = vld1_u8(vs);
493  vs += pitch;
494  d18 = vld1_u8(us);
495  us += pitch;
496  d19 = vld1_u8(vs);
497  vs += pitch;
498  d20 = vld1_u8(us);
499  d21 = vld1_u8(vs);
500
501  q3 = vcombine_u8(d6, d7);
502  q4 = vcombine_u8(d8, d9);
503  q5 = vcombine_u8(d10, d11);
504  q6 = vcombine_u8(d12, d13);
505  q7 = vcombine_u8(d14, d15);
506  q8 = vcombine_u8(d16, d17);
507  q9 = vcombine_u8(d18, d19);
508  q10 = vcombine_u8(d20, d21);
509
510  q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7));
511  q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8));
512  q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9));
513  q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10));
514
515  q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]),
516                     vreinterpretq_u16_u32(q2tmp2.val[0]));
517  q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]),
518                     vreinterpretq_u16_u32(q2tmp3.val[0]));
519  q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]),
520                     vreinterpretq_u16_u32(q2tmp2.val[1]));
521  q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]),
522                     vreinterpretq_u16_u32(q2tmp3.val[1]));
523
524  q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]),
525                    vreinterpretq_u8_u16(q2tmp5.val[0]));
526  q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]),
527                    vreinterpretq_u8_u16(q2tmp5.val[1]));
528  q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]),
529                     vreinterpretq_u8_u16(q2tmp7.val[0]));
530  q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]),
531                     vreinterpretq_u8_u16(q2tmp7.val[1]));
532
533  q3 = q2tmp8.val[0];
534  q4 = q2tmp8.val[1];
535  q5 = q2tmp9.val[0];
536  q6 = q2tmp9.val[1];
537  q7 = q2tmp10.val[0];
538  q8 = q2tmp10.val[1];
539  q9 = q2tmp11.val[0];
540  q10 = q2tmp11.val[1];
541
542  vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, q5, q6, q7, q8, q9,
543                         q10, &q4, &q5, &q6, &q7, &q8, &q9);
544
545  q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7));
546  q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8));
547  q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9));
548  q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10));
549
550  q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]),
551                     vreinterpretq_u16_u32(q2tmp2.val[0]));
552  q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]),
553                     vreinterpretq_u16_u32(q2tmp3.val[0]));
554  q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]),
555                     vreinterpretq_u16_u32(q2tmp2.val[1]));
556  q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]),
557                     vreinterpretq_u16_u32(q2tmp3.val[1]));
558
559  q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]),
560                    vreinterpretq_u8_u16(q2tmp5.val[0]));
561  q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]),
562                    vreinterpretq_u8_u16(q2tmp5.val[1]));
563  q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]),
564                     vreinterpretq_u8_u16(q2tmp7.val[0]));
565  q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]),
566                     vreinterpretq_u8_u16(q2tmp7.val[1]));
567
568  q3 = q2tmp8.val[0];
569  q4 = q2tmp8.val[1];
570  q5 = q2tmp9.val[0];
571  q6 = q2tmp9.val[1];
572  q7 = q2tmp10.val[0];
573  q8 = q2tmp10.val[1];
574  q9 = q2tmp11.val[0];
575  q10 = q2tmp11.val[1];
576
577  ud = u - 4;
578  vst1_u8(ud, vget_low_u8(q3));
579  ud += pitch;
580  vst1_u8(ud, vget_low_u8(q4));
581  ud += pitch;
582  vst1_u8(ud, vget_low_u8(q5));
583  ud += pitch;
584  vst1_u8(ud, vget_low_u8(q6));
585  ud += pitch;
586  vst1_u8(ud, vget_low_u8(q7));
587  ud += pitch;
588  vst1_u8(ud, vget_low_u8(q8));
589  ud += pitch;
590  vst1_u8(ud, vget_low_u8(q9));
591  ud += pitch;
592  vst1_u8(ud, vget_low_u8(q10));
593
594  vd = v - 4;
595  vst1_u8(vd, vget_high_u8(q3));
596  vd += pitch;
597  vst1_u8(vd, vget_high_u8(q4));
598  vd += pitch;
599  vst1_u8(vd, vget_high_u8(q5));
600  vd += pitch;
601  vst1_u8(vd, vget_high_u8(q6));
602  vd += pitch;
603  vst1_u8(vd, vget_high_u8(q7));
604  vd += pitch;
605  vst1_u8(vd, vget_high_u8(q8));
606  vd += pitch;
607  vst1_u8(vd, vget_high_u8(q9));
608  vd += pitch;
609  vst1_u8(vd, vget_high_u8(q10));
610  return;
611}
612