1 /*
2 * Copyright (c) 2015 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 <stdlib.h>
12
13 #include "vpx_dsp_rtcd.h"
14 #include "vpx_dsp_common.h"
15 #include "mem.h"
16
signed_char_clamp(int t)17 static INLINE int8_t signed_char_clamp(int t) {
18 return (int8_t)clamp(t, -128, 127);
19 }
20
21 #if CONFIG_VP9_HIGHBITDEPTH
signed_char_clamp_high(int t,int bd)22 static INLINE int16_t signed_char_clamp_high(int t, int bd) {
23 switch (bd) {
24 case 10: return (int16_t)clamp(t, -128 * 4, 128 * 4 - 1);
25 case 12: return (int16_t)clamp(t, -128 * 16, 128 * 16 - 1);
26 case 8:
27 default: return (int16_t)clamp(t, -128, 128 - 1);
28 }
29 }
30 #endif
31
32 // Should we apply any filter at all: 11111111 yes, 00000000 no
filter_mask(uint8_t limit,uint8_t blimit,uint8_t p3,uint8_t p2,uint8_t p1,uint8_t p0,uint8_t q0,uint8_t q1,uint8_t q2,uint8_t q3)33 static INLINE int8_t filter_mask(uint8_t limit, uint8_t blimit, uint8_t p3,
34 uint8_t p2, uint8_t p1, uint8_t p0, uint8_t q0,
35 uint8_t q1, uint8_t q2, uint8_t q3) {
36 int8_t mask = 0;
37 mask |= (abs(p3 - p2) > limit) * -1;
38 mask |= (abs(p2 - p1) > limit) * -1;
39 mask |= (abs(p1 - p0) > limit) * -1;
40 mask |= (abs(q1 - q0) > limit) * -1;
41 mask |= (abs(q2 - q1) > limit) * -1;
42 mask |= (abs(q3 - q2) > limit) * -1;
43 mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1;
44 return ~mask;
45 }
46
flat_mask4(uint8_t thresh,uint8_t p3,uint8_t p2,uint8_t p1,uint8_t p0,uint8_t q0,uint8_t q1,uint8_t q2,uint8_t q3)47 static INLINE int8_t flat_mask4(uint8_t thresh, uint8_t p3, uint8_t p2,
48 uint8_t p1, uint8_t p0, uint8_t q0, uint8_t q1,
49 uint8_t q2, uint8_t q3) {
50 int8_t mask = 0;
51 mask |= (abs(p1 - p0) > thresh) * -1;
52 mask |= (abs(q1 - q0) > thresh) * -1;
53 mask |= (abs(p2 - p0) > thresh) * -1;
54 mask |= (abs(q2 - q0) > thresh) * -1;
55 mask |= (abs(p3 - p0) > thresh) * -1;
56 mask |= (abs(q3 - q0) > thresh) * -1;
57 return ~mask;
58 }
59
flat_mask5(uint8_t thresh,uint8_t p4,uint8_t p3,uint8_t p2,uint8_t p1,uint8_t p0,uint8_t q0,uint8_t q1,uint8_t q2,uint8_t q3,uint8_t q4)60 static INLINE int8_t flat_mask5(uint8_t thresh, uint8_t p4, uint8_t p3,
61 uint8_t p2, uint8_t p1, uint8_t p0, uint8_t q0,
62 uint8_t q1, uint8_t q2, uint8_t q3,
63 uint8_t q4) {
64 int8_t mask = ~flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3);
65 mask |= (abs(p4 - p0) > thresh) * -1;
66 mask |= (abs(q4 - q0) > thresh) * -1;
67 return ~mask;
68 }
69
70 // Is there high edge variance internal edge: 11111111 yes, 00000000 no
hev_mask(uint8_t thresh,uint8_t p1,uint8_t p0,uint8_t q0,uint8_t q1)71 static INLINE int8_t hev_mask(uint8_t thresh, uint8_t p1, uint8_t p0,
72 uint8_t q0, uint8_t q1) {
73 int8_t hev = 0;
74 hev |= (abs(p1 - p0) > thresh) * -1;
75 hev |= (abs(q1 - q0) > thresh) * -1;
76 return hev;
77 }
78
filter4(int8_t mask,uint8_t thresh,uint8_t * op1,uint8_t * op0,uint8_t * oq0,uint8_t * oq1)79 static INLINE void filter4(int8_t mask, uint8_t thresh, uint8_t *op1,
80 uint8_t *op0, uint8_t *oq0, uint8_t *oq1) {
81 int8_t filter1, filter2;
82
83 const int8_t ps1 = (int8_t)*op1 ^ 0x80;
84 const int8_t ps0 = (int8_t)*op0 ^ 0x80;
85 const int8_t qs0 = (int8_t)*oq0 ^ 0x80;
86 const int8_t qs1 = (int8_t)*oq1 ^ 0x80;
87 const uint8_t hev = hev_mask(thresh, *op1, *op0, *oq0, *oq1);
88
89 // add outer taps if we have high edge variance
90 int8_t filter = signed_char_clamp(ps1 - qs1) & hev;
91
92 // inner taps
93 filter = signed_char_clamp(filter + 3 * (qs0 - ps0)) & mask;
94
95 // save bottom 3 bits so that we round one side +4 and the other +3
96 // if it equals 4 we'll set it to adjust by -1 to account for the fact
97 // we'd round it by 3 the other way
98 filter1 = signed_char_clamp(filter + 4) >> 3;
99 filter2 = signed_char_clamp(filter + 3) >> 3;
100
101 *oq0 = signed_char_clamp(qs0 - filter1) ^ 0x80;
102 *op0 = signed_char_clamp(ps0 + filter2) ^ 0x80;
103
104 // outer tap adjustments
105 filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev;
106
107 *oq1 = signed_char_clamp(qs1 - filter) ^ 0x80;
108 *op1 = signed_char_clamp(ps1 + filter) ^ 0x80;
109 }
110
eb_vp9_lpf_horizontal_4_c(uint8_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh)111 void eb_vp9_lpf_horizontal_4_c(uint8_t *s, int p /* pitch */,
112 const uint8_t *blimit, const uint8_t *limit,
113 const uint8_t *thresh) {
114 int i;
115
116 // loop filter designed to work using chars so that we can make maximum use
117 // of 8 bit simd instructions.
118 for (i = 0; i < 8; ++i) {
119 const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
120 const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
121 const int8_t mask =
122 filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
123 filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p);
124 ++s;
125 }
126 }
127
eb_vp9_lpf_horizontal_4_dual_c(uint8_t * s,int p,const uint8_t * blimit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * blimit1,const uint8_t * limit1,const uint8_t * thresh1)128 void eb_vp9_lpf_horizontal_4_dual_c(uint8_t *s, int p, const uint8_t *blimit0,
129 const uint8_t *limit0, const uint8_t *thresh0,
130 const uint8_t *blimit1, const uint8_t *limit1,
131 const uint8_t *thresh1) {
132 eb_vp9_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0);
133 eb_vp9_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1);
134 }
135
eb_vp9_lpf_vertical_4_c(uint8_t * s,int pitch,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh)136 void eb_vp9_lpf_vertical_4_c(uint8_t *s, int pitch, const uint8_t *blimit,
137 const uint8_t *limit, const uint8_t *thresh) {
138 int i;
139
140 // loop filter designed to work using chars so that we can make maximum use
141 // of 8 bit simd instructions.
142 for (i = 0; i < 8; ++i) {
143 const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
144 const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
145 const int8_t mask =
146 filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
147 filter4(mask, *thresh, s - 2, s - 1, s, s + 1);
148 s += pitch;
149 }
150 }
151
eb_vp9_lpf_vertical_4_dual_c(uint8_t * s,int pitch,const uint8_t * blimit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * blimit1,const uint8_t * limit1,const uint8_t * thresh1)152 void eb_vp9_lpf_vertical_4_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0,
153 const uint8_t *limit0, const uint8_t *thresh0,
154 const uint8_t *blimit1, const uint8_t *limit1,
155 const uint8_t *thresh1) {
156 eb_vp9_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0);
157 eb_vp9_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1);
158 }
159
filter8(int8_t mask,uint8_t thresh,uint8_t flat,uint8_t * op3,uint8_t * op2,uint8_t * op1,uint8_t * op0,uint8_t * oq0,uint8_t * oq1,uint8_t * oq2,uint8_t * oq3)160 static INLINE void filter8(int8_t mask, uint8_t thresh, uint8_t flat,
161 uint8_t *op3, uint8_t *op2, uint8_t *op1,
162 uint8_t *op0, uint8_t *oq0, uint8_t *oq1,
163 uint8_t *oq2, uint8_t *oq3) {
164 if (flat && mask) {
165 const uint8_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0;
166 const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3;
167
168 // 7-tap filter [1, 1, 1, 2, 1, 1, 1]
169 *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3);
170 *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3);
171 *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3);
172 *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3);
173 *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3);
174 *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3);
175 } else {
176 filter4(mask, thresh, op1, op0, oq0, oq1);
177 }
178 }
179
eb_vp9_lpf_horizontal_8_c(uint8_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh)180 void eb_vp9_lpf_horizontal_8_c(uint8_t *s, int p, const uint8_t *blimit,
181 const uint8_t *limit, const uint8_t *thresh) {
182 int i;
183
184 // loop filter designed to work using chars so that we can make maximum use
185 // of 8 bit simd instructions.
186 for (i = 0; i < 8; ++i) {
187 const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
188 const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
189
190 const int8_t mask =
191 filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
192 const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
193 filter8(mask, *thresh, flat, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s,
194 s + 1 * p, s + 2 * p, s + 3 * p);
195 ++s;
196 }
197 }
198
eb_vp9_lpf_horizontal_8_dual_c(uint8_t * s,int p,const uint8_t * blimit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * blimit1,const uint8_t * limit1,const uint8_t * thresh1)199 void eb_vp9_lpf_horizontal_8_dual_c(uint8_t *s, int p, const uint8_t *blimit0,
200 const uint8_t *limit0, const uint8_t *thresh0,
201 const uint8_t *blimit1, const uint8_t *limit1,
202 const uint8_t *thresh1) {
203 eb_vp9_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0);
204 eb_vp9_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1);
205 }
206
eb_vp9_lpf_vertical_8_c(uint8_t * s,int pitch,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh)207 void eb_vp9_lpf_vertical_8_c(uint8_t *s, int pitch, const uint8_t *blimit,
208 const uint8_t *limit, const uint8_t *thresh) {
209 int i;
210
211 for (i = 0; i < 8; ++i) {
212 const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
213 const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
214 const int8_t mask =
215 filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
216 const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
217 filter8(mask, *thresh, flat, s - 4, s - 3, s - 2, s - 1, s, s + 1, s + 2,
218 s + 3);
219 s += pitch;
220 }
221 }
222
eb_vp9_lpf_vertical_8_dual_c(uint8_t * s,int pitch,const uint8_t * blimit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * blimit1,const uint8_t * limit1,const uint8_t * thresh1)223 void eb_vp9_lpf_vertical_8_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0,
224 const uint8_t *limit0, const uint8_t *thresh0,
225 const uint8_t *blimit1, const uint8_t *limit1,
226 const uint8_t *thresh1) {
227 eb_vp9_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0);
228 eb_vp9_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1);
229 }
230
filter16(int8_t mask,uint8_t thresh,uint8_t flat,uint8_t flat2,uint8_t * op7,uint8_t * op6,uint8_t * op5,uint8_t * op4,uint8_t * op3,uint8_t * op2,uint8_t * op1,uint8_t * op0,uint8_t * oq0,uint8_t * oq1,uint8_t * oq2,uint8_t * oq3,uint8_t * oq4,uint8_t * oq5,uint8_t * oq6,uint8_t * oq7)231 static INLINE void filter16(int8_t mask, uint8_t thresh, uint8_t flat,
232 uint8_t flat2, uint8_t *op7, uint8_t *op6,
233 uint8_t *op5, uint8_t *op4, uint8_t *op3,
234 uint8_t *op2, uint8_t *op1, uint8_t *op0,
235 uint8_t *oq0, uint8_t *oq1, uint8_t *oq2,
236 uint8_t *oq3, uint8_t *oq4, uint8_t *oq5,
237 uint8_t *oq6, uint8_t *oq7) {
238 if (flat2 && flat && mask) {
239 const uint8_t p7 = *op7, p6 = *op6, p5 = *op5, p4 = *op4, p3 = *op3,
240 p2 = *op2, p1 = *op1, p0 = *op0;
241
242 const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3, q4 = *oq4,
243 q5 = *oq5, q6 = *oq6, q7 = *oq7;
244
245 // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1]
246 *op6 = ROUND_POWER_OF_TWO(
247 p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 + q0, 4);
248 *op5 = ROUND_POWER_OF_TWO(
249 p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 + q0 + q1, 4);
250 *op4 = ROUND_POWER_OF_TWO(
251 p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 + q0 + q1 + q2, 4);
252 *op3 = ROUND_POWER_OF_TWO(
253 p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 + q0 + q1 + q2 + q3, 4);
254 *op2 = ROUND_POWER_OF_TWO(
255 p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 + q0 + q1 + q2 + q3 + q4,
256 4);
257 *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 +
258 q0 + q1 + q2 + q3 + q4 + q5,
259 4);
260 *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 + q0 +
261 q1 + q2 + q3 + q4 + q5 + q6,
262 4);
263 *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 + q0 * 2 + q1 +
264 q2 + q3 + q4 + q5 + q6 + q7,
265 4);
266 *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 + q0 + q1 * 2 + q2 +
267 q3 + q4 + q5 + q6 + q7 * 2,
268 4);
269 *oq2 = ROUND_POWER_OF_TWO(
270 p4 + p3 + p2 + p1 + p0 + q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3,
271 4);
272 *oq3 = ROUND_POWER_OF_TWO(
273 p3 + p2 + p1 + p0 + q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4);
274 *oq4 = ROUND_POWER_OF_TWO(
275 p2 + p1 + p0 + q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4);
276 *oq5 = ROUND_POWER_OF_TWO(
277 p1 + p0 + q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4);
278 *oq6 = ROUND_POWER_OF_TWO(
279 p0 + q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4);
280 } else {
281 filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3);
282 }
283 }
284
mb_lpf_horizontal_edge_w(uint8_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int count)285 static void mb_lpf_horizontal_edge_w(uint8_t *s, int p, const uint8_t *blimit,
286 const uint8_t *limit,
287 const uint8_t *thresh, int count) {
288 int i;
289
290 // loop filter designed to work using chars so that we can make maximum use
291 // of 8 bit simd instructions.
292 for (i = 0; i < 8 * count; ++i) {
293 const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
294 const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
295 const int8_t mask =
296 filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
297 const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
298 const int8_t flat2 =
299 flat_mask5(1, s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, q0,
300 s[4 * p], s[5 * p], s[6 * p], s[7 * p]);
301
302 filter16(mask, *thresh, flat, flat2, s - 8 * p, s - 7 * p, s - 6 * p,
303 s - 5 * p, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s,
304 s + 1 * p, s + 2 * p, s + 3 * p, s + 4 * p, s + 5 * p, s + 6 * p,
305 s + 7 * p);
306 ++s;
307 }
308 }
309
eb_vp9_lpf_horizontal_16_c(uint8_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh)310 void eb_vp9_lpf_horizontal_16_c(uint8_t *s, int p, const uint8_t *blimit,
311 const uint8_t *limit, const uint8_t *thresh) {
312 mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 1);
313 }
314
eb_vp9_lpf_horizontal_16_dual_c(uint8_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh)315 void eb_vp9_lpf_horizontal_16_dual_c(uint8_t *s, int p, const uint8_t *blimit,
316 const uint8_t *limit, const uint8_t *thresh) {
317 mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 2);
318 }
319
mb_lpf_vertical_edge_w(uint8_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int count)320 static void mb_lpf_vertical_edge_w(uint8_t *s, int p, const uint8_t *blimit,
321 const uint8_t *limit, const uint8_t *thresh,
322 int count) {
323 int i;
324
325 for (i = 0; i < count; ++i) {
326 const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
327 const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
328 const int8_t mask =
329 filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
330 const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
331 const int8_t flat2 = flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0, q0, s[4],
332 s[5], s[6], s[7]);
333
334 filter16(mask, *thresh, flat, flat2, s - 8, s - 7, s - 6, s - 5, s - 4,
335 s - 3, s - 2, s - 1, s, s + 1, s + 2, s + 3, s + 4, s + 5, s + 6,
336 s + 7);
337 s += p;
338 }
339 }
340
eb_vp9_lpf_vertical_16_c(uint8_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh)341 void eb_vp9_lpf_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit,
342 const uint8_t *limit, const uint8_t *thresh) {
343 mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8);
344 }
345
eb_vp9_lpf_vertical_16_dual_c(uint8_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh)346 void eb_vp9_lpf_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit,
347 const uint8_t *limit, const uint8_t *thresh) {
348 mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16);
349 }
350
351 #if CONFIG_VP9_HIGHBITDEPTH
352 // Should we apply any filter at all: 11111111 yes, 00000000 no ?
highbd_filter_mask(uint8_t limit,uint8_t blimit,uint16_t p3,uint16_t p2,uint16_t p1,uint16_t p0,uint16_t q0,uint16_t q1,uint16_t q2,uint16_t q3,int bd)353 static INLINE int8_t highbd_filter_mask(uint8_t limit, uint8_t blimit,
354 uint16_t p3, uint16_t p2, uint16_t p1,
355 uint16_t p0, uint16_t q0, uint16_t q1,
356 uint16_t q2, uint16_t q3, int bd) {
357 int8_t mask = 0;
358 int16_t limit16 = (uint16_t)limit << (bd - 8);
359 int16_t blimit16 = (uint16_t)blimit << (bd - 8);
360 mask |= (abs(p3 - p2) > limit16) * -1;
361 mask |= (abs(p2 - p1) > limit16) * -1;
362 mask |= (abs(p1 - p0) > limit16) * -1;
363 mask |= (abs(q1 - q0) > limit16) * -1;
364 mask |= (abs(q2 - q1) > limit16) * -1;
365 mask |= (abs(q3 - q2) > limit16) * -1;
366 mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit16) * -1;
367 return ~mask;
368 }
369
highbd_flat_mask4(uint8_t thresh,uint16_t p3,uint16_t p2,uint16_t p1,uint16_t p0,uint16_t q0,uint16_t q1,uint16_t q2,uint16_t q3,int bd)370 static INLINE int8_t highbd_flat_mask4(uint8_t thresh, uint16_t p3, uint16_t p2,
371 uint16_t p1, uint16_t p0, uint16_t q0,
372 uint16_t q1, uint16_t q2, uint16_t q3,
373 int bd) {
374 int8_t mask = 0;
375 int16_t thresh16 = (uint16_t)thresh << (bd - 8);
376 mask |= (abs(p1 - p0) > thresh16) * -1;
377 mask |= (abs(q1 - q0) > thresh16) * -1;
378 mask |= (abs(p2 - p0) > thresh16) * -1;
379 mask |= (abs(q2 - q0) > thresh16) * -1;
380 mask |= (abs(p3 - p0) > thresh16) * -1;
381 mask |= (abs(q3 - q0) > thresh16) * -1;
382 return ~mask;
383 }
384
highbd_flat_mask5(uint8_t thresh,uint16_t p4,uint16_t p3,uint16_t p2,uint16_t p1,uint16_t p0,uint16_t q0,uint16_t q1,uint16_t q2,uint16_t q3,uint16_t q4,int bd)385 static INLINE int8_t highbd_flat_mask5(uint8_t thresh, uint16_t p4, uint16_t p3,
386 uint16_t p2, uint16_t p1, uint16_t p0,
387 uint16_t q0, uint16_t q1, uint16_t q2,
388 uint16_t q3, uint16_t q4, int bd) {
389 int8_t mask = ~highbd_flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3, bd);
390 int16_t thresh16 = (uint16_t)thresh << (bd - 8);
391 mask |= (abs(p4 - p0) > thresh16) * -1;
392 mask |= (abs(q4 - q0) > thresh16) * -1;
393 return ~mask;
394 }
395
396 // Is there high edge variance internal edge:
397 // 11111111_11111111 yes, 00000000_00000000 no ?
highbd_hev_mask(uint8_t thresh,uint16_t p1,uint16_t p0,uint16_t q0,uint16_t q1,int bd)398 static INLINE int16_t highbd_hev_mask(uint8_t thresh, uint16_t p1, uint16_t p0,
399 uint16_t q0, uint16_t q1, int bd) {
400 int16_t hev = 0;
401 int16_t thresh16 = (uint16_t)thresh << (bd - 8);
402 hev |= (abs(p1 - p0) > thresh16) * -1;
403 hev |= (abs(q1 - q0) > thresh16) * -1;
404 return hev;
405 }
406
highbd_filter4(int8_t mask,uint8_t thresh,uint16_t * op1,uint16_t * op0,uint16_t * oq0,uint16_t * oq1,int bd)407 static INLINE void highbd_filter4(int8_t mask, uint8_t thresh, uint16_t *op1,
408 uint16_t *op0, uint16_t *oq0, uint16_t *oq1,
409 int bd) {
410 int16_t filter1, filter2;
411 // ^0x80 equivalent to subtracting 0x80 from the values to turn them
412 // into -128 to +127 instead of 0 to 255.
413 int shift = bd - 8;
414 const int16_t ps1 = (int16_t)*op1 - (0x80 << shift);
415 const int16_t ps0 = (int16_t)*op0 - (0x80 << shift);
416 const int16_t qs0 = (int16_t)*oq0 - (0x80 << shift);
417 const int16_t qs1 = (int16_t)*oq1 - (0x80 << shift);
418 const uint16_t hev = highbd_hev_mask(thresh, *op1, *op0, *oq0, *oq1, bd);
419
420 // Add outer taps if we have high edge variance.
421 int16_t filter = signed_char_clamp_high(ps1 - qs1, bd) & hev;
422
423 // Inner taps.
424 filter = signed_char_clamp_high(filter + 3 * (qs0 - ps0), bd) & mask;
425
426 // Save bottom 3 bits so that we round one side +4 and the other +3
427 // if it equals 4 we'll set it to adjust by -1 to account for the fact
428 // we'd round it by 3 the other way.
429 filter1 = signed_char_clamp_high(filter + 4, bd) >> 3;
430 filter2 = signed_char_clamp_high(filter + 3, bd) >> 3;
431
432 *oq0 = signed_char_clamp_high(qs0 - filter1, bd) + (0x80 << shift);
433 *op0 = signed_char_clamp_high(ps0 + filter2, bd) + (0x80 << shift);
434
435 // Outer tap adjustments.
436 filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev;
437
438 *oq1 = signed_char_clamp_high(qs1 - filter, bd) + (0x80 << shift);
439 *op1 = signed_char_clamp_high(ps1 + filter, bd) + (0x80 << shift);
440 }
441
vpx_highbd_lpf_horizontal_4_c(uint16_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int bd)442 void vpx_highbd_lpf_horizontal_4_c(uint16_t *s, int p /* pitch */,
443 const uint8_t *blimit, const uint8_t *limit,
444 const uint8_t *thresh, int bd) {
445 int i;
446
447 // loop filter designed to work using chars so that we can make maximum use
448 // of 8 bit simd instructions.
449 for (i = 0; i < 8; ++i) {
450 const uint16_t p3 = s[-4 * p];
451 const uint16_t p2 = s[-3 * p];
452 const uint16_t p1 = s[-2 * p];
453 const uint16_t p0 = s[-p];
454 const uint16_t q0 = s[0 * p];
455 const uint16_t q1 = s[1 * p];
456 const uint16_t q2 = s[2 * p];
457 const uint16_t q3 = s[3 * p];
458 const int8_t mask =
459 highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
460 highbd_filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p, bd);
461 ++s;
462 }
463 }
464
vpx_highbd_lpf_horizontal_4_dual_c(uint16_t * s,int p,const uint8_t * blimit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * blimit1,const uint8_t * limit1,const uint8_t * thresh1,int bd)465 void vpx_highbd_lpf_horizontal_4_dual_c(
466 uint16_t *s, int p, const uint8_t *blimit0, const uint8_t *limit0,
467 const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
468 const uint8_t *thresh1, int bd) {
469 vpx_highbd_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, bd);
470 vpx_highbd_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, bd);
471 }
472
vpx_highbd_lpf_vertical_4_c(uint16_t * s,int pitch,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int bd)473 void vpx_highbd_lpf_vertical_4_c(uint16_t *s, int pitch, const uint8_t *blimit,
474 const uint8_t *limit, const uint8_t *thresh,
475 int bd) {
476 int i;
477
478 // loop filter designed to work using chars so that we can make maximum use
479 // of 8 bit simd instructions.
480 for (i = 0; i < 8; ++i) {
481 const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
482 const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
483 const int8_t mask =
484 highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
485 highbd_filter4(mask, *thresh, s - 2, s - 1, s, s + 1, bd);
486 s += pitch;
487 }
488 }
489
vpx_highbd_lpf_vertical_4_dual_c(uint16_t * s,int pitch,const uint8_t * blimit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * blimit1,const uint8_t * limit1,const uint8_t * thresh1,int bd)490 void vpx_highbd_lpf_vertical_4_dual_c(
491 uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0,
492 const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
493 const uint8_t *thresh1, int bd) {
494 vpx_highbd_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, bd);
495 vpx_highbd_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1,
496 bd);
497 }
498
highbd_filter8(int8_t mask,uint8_t thresh,uint8_t flat,uint16_t * op3,uint16_t * op2,uint16_t * op1,uint16_t * op0,uint16_t * oq0,uint16_t * oq1,uint16_t * oq2,uint16_t * oq3,int bd)499 static INLINE void highbd_filter8(int8_t mask, uint8_t thresh, uint8_t flat,
500 uint16_t *op3, uint16_t *op2, uint16_t *op1,
501 uint16_t *op0, uint16_t *oq0, uint16_t *oq1,
502 uint16_t *oq2, uint16_t *oq3, int bd) {
503 if (flat && mask) {
504 const uint16_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0;
505 const uint16_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3;
506
507 // 7-tap filter [1, 1, 1, 2, 1, 1, 1]
508 *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3);
509 *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3);
510 *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3);
511 *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3);
512 *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3);
513 *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3);
514 } else {
515 highbd_filter4(mask, thresh, op1, op0, oq0, oq1, bd);
516 }
517 }
518
vpx_highbd_lpf_horizontal_8_c(uint16_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int bd)519 void vpx_highbd_lpf_horizontal_8_c(uint16_t *s, int p, const uint8_t *blimit,
520 const uint8_t *limit, const uint8_t *thresh,
521 int bd) {
522 int i;
523
524 // loop filter designed to work using chars so that we can make maximum use
525 // of 8 bit simd instructions.
526 for (i = 0; i < 8; ++i) {
527 const uint16_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
528 const uint16_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
529
530 const int8_t mask =
531 highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
532 const int8_t flat =
533 highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
534 highbd_filter8(mask, *thresh, flat, s - 4 * p, s - 3 * p, s - 2 * p,
535 s - 1 * p, s, s + 1 * p, s + 2 * p, s + 3 * p, bd);
536 ++s;
537 }
538 }
539
vpx_highbd_lpf_horizontal_8_dual_c(uint16_t * s,int p,const uint8_t * blimit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * blimit1,const uint8_t * limit1,const uint8_t * thresh1,int bd)540 void vpx_highbd_lpf_horizontal_8_dual_c(
541 uint16_t *s, int p, const uint8_t *blimit0, const uint8_t *limit0,
542 const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
543 const uint8_t *thresh1, int bd) {
544 vpx_highbd_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, bd);
545 vpx_highbd_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, bd);
546 }
547
vpx_highbd_lpf_vertical_8_c(uint16_t * s,int pitch,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int bd)548 void vpx_highbd_lpf_vertical_8_c(uint16_t *s, int pitch, const uint8_t *blimit,
549 const uint8_t *limit, const uint8_t *thresh,
550 int bd) {
551 int i;
552
553 for (i = 0; i < 8; ++i) {
554 const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
555 const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
556 const int8_t mask =
557 highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
558 const int8_t flat =
559 highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
560 highbd_filter8(mask, *thresh, flat, s - 4, s - 3, s - 2, s - 1, s, s + 1,
561 s + 2, s + 3, bd);
562 s += pitch;
563 }
564 }
565
vpx_highbd_lpf_vertical_8_dual_c(uint16_t * s,int pitch,const uint8_t * blimit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * blimit1,const uint8_t * limit1,const uint8_t * thresh1,int bd)566 void vpx_highbd_lpf_vertical_8_dual_c(
567 uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0,
568 const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
569 const uint8_t *thresh1, int bd) {
570 vpx_highbd_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, bd);
571 vpx_highbd_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1,
572 bd);
573 }
574
highbd_filter16(int8_t mask,uint8_t thresh,uint8_t flat,uint8_t flat2,uint16_t * op7,uint16_t * op6,uint16_t * op5,uint16_t * op4,uint16_t * op3,uint16_t * op2,uint16_t * op1,uint16_t * op0,uint16_t * oq0,uint16_t * oq1,uint16_t * oq2,uint16_t * oq3,uint16_t * oq4,uint16_t * oq5,uint16_t * oq6,uint16_t * oq7,int bd)575 static INLINE void highbd_filter16(int8_t mask, uint8_t thresh, uint8_t flat,
576 uint8_t flat2, uint16_t *op7, uint16_t *op6,
577 uint16_t *op5, uint16_t *op4, uint16_t *op3,
578 uint16_t *op2, uint16_t *op1, uint16_t *op0,
579 uint16_t *oq0, uint16_t *oq1, uint16_t *oq2,
580 uint16_t *oq3, uint16_t *oq4, uint16_t *oq5,
581 uint16_t *oq6, uint16_t *oq7, int bd) {
582 if (flat2 && flat && mask) {
583 const uint16_t p7 = *op7;
584 const uint16_t p6 = *op6;
585 const uint16_t p5 = *op5;
586 const uint16_t p4 = *op4;
587 const uint16_t p3 = *op3;
588 const uint16_t p2 = *op2;
589 const uint16_t p1 = *op1;
590 const uint16_t p0 = *op0;
591 const uint16_t q0 = *oq0;
592 const uint16_t q1 = *oq1;
593 const uint16_t q2 = *oq2;
594 const uint16_t q3 = *oq3;
595 const uint16_t q4 = *oq4;
596 const uint16_t q5 = *oq5;
597 const uint16_t q6 = *oq6;
598 const uint16_t q7 = *oq7;
599
600 // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1]
601 *op6 = ROUND_POWER_OF_TWO(
602 p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 + q0, 4);
603 *op5 = ROUND_POWER_OF_TWO(
604 p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 + q0 + q1, 4);
605 *op4 = ROUND_POWER_OF_TWO(
606 p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 + q0 + q1 + q2, 4);
607 *op3 = ROUND_POWER_OF_TWO(
608 p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 + q0 + q1 + q2 + q3, 4);
609 *op2 = ROUND_POWER_OF_TWO(
610 p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 + q0 + q1 + q2 + q3 + q4,
611 4);
612 *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 +
613 q0 + q1 + q2 + q3 + q4 + q5,
614 4);
615 *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 + q0 +
616 q1 + q2 + q3 + q4 + q5 + q6,
617 4);
618 *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 + q0 * 2 + q1 +
619 q2 + q3 + q4 + q5 + q6 + q7,
620 4);
621 *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 + q0 + q1 * 2 + q2 +
622 q3 + q4 + q5 + q6 + q7 * 2,
623 4);
624 *oq2 = ROUND_POWER_OF_TWO(
625 p4 + p3 + p2 + p1 + p0 + q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3,
626 4);
627 *oq3 = ROUND_POWER_OF_TWO(
628 p3 + p2 + p1 + p0 + q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4);
629 *oq4 = ROUND_POWER_OF_TWO(
630 p2 + p1 + p0 + q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4);
631 *oq5 = ROUND_POWER_OF_TWO(
632 p1 + p0 + q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4);
633 *oq6 = ROUND_POWER_OF_TWO(
634 p0 + q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4);
635 } else {
636 highbd_filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3,
637 bd);
638 }
639 }
640
highbd_mb_lpf_horizontal_edge_w(uint16_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int count,int bd)641 static void highbd_mb_lpf_horizontal_edge_w(uint16_t *s, int p,
642 const uint8_t *blimit,
643 const uint8_t *limit,
644 const uint8_t *thresh, int count,
645 int bd) {
646 int i;
647
648 // loop filter designed to work using chars so that we can make maximum use
649 // of 8 bit simd instructions.
650 for (i = 0; i < 8 * count; ++i) {
651 const uint16_t p3 = s[-4 * p];
652 const uint16_t p2 = s[-3 * p];
653 const uint16_t p1 = s[-2 * p];
654 const uint16_t p0 = s[-p];
655 const uint16_t q0 = s[0 * p];
656 const uint16_t q1 = s[1 * p];
657 const uint16_t q2 = s[2 * p];
658 const uint16_t q3 = s[3 * p];
659 const int8_t mask =
660 highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
661 const int8_t flat =
662 highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
663 const int8_t flat2 =
664 highbd_flat_mask5(1, s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, q0,
665 s[4 * p], s[5 * p], s[6 * p], s[7 * p], bd);
666
667 highbd_filter16(mask, *thresh, flat, flat2, s - 8 * p, s - 7 * p, s - 6 * p,
668 s - 5 * p, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s,
669 s + 1 * p, s + 2 * p, s + 3 * p, s + 4 * p, s + 5 * p,
670 s + 6 * p, s + 7 * p, bd);
671 ++s;
672 }
673 }
674
vpx_highbd_lpf_horizontal_16_c(uint16_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int bd)675 void vpx_highbd_lpf_horizontal_16_c(uint16_t *s, int p, const uint8_t *blimit,
676 const uint8_t *limit, const uint8_t *thresh,
677 int bd) {
678 highbd_mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 1, bd);
679 }
680
vpx_highbd_lpf_horizontal_16_dual_c(uint16_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int bd)681 void vpx_highbd_lpf_horizontal_16_dual_c(uint16_t *s, int p,
682 const uint8_t *blimit,
683 const uint8_t *limit,
684 const uint8_t *thresh, int bd) {
685 highbd_mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 2, bd);
686 }
687
highbd_mb_lpf_vertical_edge_w(uint16_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int count,int bd)688 static void highbd_mb_lpf_vertical_edge_w(uint16_t *s, int p,
689 const uint8_t *blimit,
690 const uint8_t *limit,
691 const uint8_t *thresh, int count,
692 int bd) {
693 int i;
694
695 for (i = 0; i < count; ++i) {
696 const uint16_t p3 = s[-4];
697 const uint16_t p2 = s[-3];
698 const uint16_t p1 = s[-2];
699 const uint16_t p0 = s[-1];
700 const uint16_t q0 = s[0];
701 const uint16_t q1 = s[1];
702 const uint16_t q2 = s[2];
703 const uint16_t q3 = s[3];
704 const int8_t mask =
705 highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
706 const int8_t flat =
707 highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
708 const int8_t flat2 = highbd_flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0,
709 q0, s[4], s[5], s[6], s[7], bd);
710
711 highbd_filter16(mask, *thresh, flat, flat2, s - 8, s - 7, s - 6, s - 5,
712 s - 4, s - 3, s - 2, s - 1, s, s + 1, s + 2, s + 3, s + 4,
713 s + 5, s + 6, s + 7, bd);
714 s += p;
715 }
716 }
717
vpx_highbd_lpf_vertical_16_c(uint16_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int bd)718 void vpx_highbd_lpf_vertical_16_c(uint16_t *s, int p, const uint8_t *blimit,
719 const uint8_t *limit, const uint8_t *thresh,
720 int bd) {
721 highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8, bd);
722 }
723
vpx_highbd_lpf_vertical_16_dual_c(uint16_t * s,int p,const uint8_t * blimit,const uint8_t * limit,const uint8_t * thresh,int bd)724 void vpx_highbd_lpf_vertical_16_dual_c(uint16_t *s, int p,
725 const uint8_t *blimit,
726 const uint8_t *limit,
727 const uint8_t *thresh, int bd) {
728 highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16, bd);
729 }
730 #endif // CONFIG_VP9_HIGHBITDEPTH
731