xref: /aosp_15_r20/external/libdav1d/src/itx_1d.c (revision c09093415860a1c2373dacd84c4fde00c507cdfd)
1 /*
2  * Copyright © 2018-2019, VideoLAN and dav1d authors
3  * Copyright © 2018-2019, Two Orioles, LLC
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions are met:
8  *
9  * 1. Redistributions of source code must retain the above copyright notice, this
10  *    list of conditions and the following disclaimer.
11  *
12  * 2. Redistributions in binary form must reproduce the above copyright notice,
13  *    this list of conditions and the following disclaimer in the documentation
14  *    and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
19  * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
20  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
22  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
23  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include "config.h"
29 
30 #include <stddef.h>
31 #include <stdint.h>
32 
33 #include "common/intops.h"
34 
35 #include "src/itx_1d.h"
36 
37 #define CLIP(a) iclip(a, min, max)
38 
39 /*
40  * In some places, we use the pattern like this:
41  * t2 = ((in1 *  1567         - in3 * (3784 - 4096) + 2048) >> 12) - in3;
42  * even though the reference code might use something like:
43  * t2 =  (in1 *  1567         - in3 *  3784         + 2048) >> 12;
44  *
45  * The reason for this is that for 12 bits/component bitstreams (corrupt/
46  * invalid ones, but they are codable nonetheless), each coefficient or
47  * input can be 19(+sign) bits, and therefore if the combination of the
48  * two multipliers (each 12 bits) is >= 4096, the result of the add/sub
49  * after the pair of multiplies will exceed the 31+sign bit range. Signed
50  * integer overflows are UB in C, and we'd like to prevent that.
51  *
52  * To workaround this, we invert one of the two coefficients (or, if both are
53  * multiples of 2, we reduce their magnitude by one bit). It should be noted
54  * that SIMD implementations do not have to follow this exact behaviour. The
55  * AV1 spec clearly states that the result of the multiply/add pairs should
56  * fit in 31+sign bit intermediates, and that streams violating this convention
57  * are not AV1-compliant. So, as long as we don't trigger UB (which some people
58  * would consider a security vulnerability), we're fine. So, SIMD can simply
59  * use the faster implementation, even if that might in some cases result in
60  * integer overflows, since these are not considered valid AV1 anyway, and in
61  * e.g. x86 assembly, integer overflows are not considered UB, but they merely
62  * wrap around.
63  */
64 
65 static NOINLINE void
inv_dct4_1d_internal_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max,const int tx64)66 inv_dct4_1d_internal_c(int32_t *const c, const ptrdiff_t stride,
67                        const int min, const int max, const int tx64)
68 {
69     assert(stride > 0);
70     const int in0 = c[0 * stride], in1 = c[1 * stride];
71 
72     int t0, t1, t2, t3;
73     if (tx64) {
74         t0 = t1 = (in0 * 181 + 128) >> 8;
75         t2 = (in1 * 1567 + 2048) >> 12;
76         t3 = (in1 * 3784 + 2048) >> 12;
77     } else {
78         const int in2 = c[2 * stride], in3 = c[3 * stride];
79 
80         t0 = ((in0 + in2) * 181 + 128) >> 8;
81         t1 = ((in0 - in2) * 181 + 128) >> 8;
82         t2 = ((in1 *  1567         - in3 * (3784 - 4096) + 2048) >> 12) - in3;
83         t3 = ((in1 * (3784 - 4096) + in3 *  1567         + 2048) >> 12) + in1;
84     }
85 
86     c[0 * stride] = CLIP(t0 + t3);
87     c[1 * stride] = CLIP(t1 + t2);
88     c[2 * stride] = CLIP(t1 - t2);
89     c[3 * stride] = CLIP(t0 - t3);
90 }
91 
inv_dct4_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)92 static void inv_dct4_1d_c(int32_t *const c, const ptrdiff_t stride,
93                           const int min, const int max)
94 {
95     inv_dct4_1d_internal_c(c, stride, min, max, 0);
96 }
97 
98 static NOINLINE void
inv_dct8_1d_internal_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max,const int tx64)99 inv_dct8_1d_internal_c(int32_t *const c, const ptrdiff_t stride,
100                        const int min, const int max, const int tx64)
101 {
102     assert(stride > 0);
103     inv_dct4_1d_internal_c(c, stride << 1, min, max, tx64);
104 
105     const int in1 = c[1 * stride], in3 = c[3 * stride];
106 
107     int t4a, t5a, t6a, t7a;
108     if (tx64) {
109         t4a = (in1 *   799 + 2048) >> 12;
110         t5a = (in3 * -2276 + 2048) >> 12;
111         t6a = (in3 *  3406 + 2048) >> 12;
112         t7a = (in1 *  4017 + 2048) >> 12;
113     } else {
114         const int in5 = c[5 * stride], in7 = c[7 * stride];
115 
116         t4a = ((in1 *   799         - in7 * (4017 - 4096) + 2048) >> 12) - in7;
117         t5a =  (in5 *  1703         - in3 *  1138         + 1024) >> 11;
118         t6a =  (in5 *  1138         + in3 *  1703         + 1024) >> 11;
119         t7a = ((in1 * (4017 - 4096) + in7 *  799          + 2048) >> 12) + in1;
120     }
121 
122     const int t4  = CLIP(t4a + t5a);
123               t5a = CLIP(t4a - t5a);
124     const int t7  = CLIP(t7a + t6a);
125               t6a = CLIP(t7a - t6a);
126 
127     const int t5  = ((t6a - t5a) * 181 + 128) >> 8;
128     const int t6  = ((t6a + t5a) * 181 + 128) >> 8;
129 
130     const int t0 = c[0 * stride];
131     const int t1 = c[2 * stride];
132     const int t2 = c[4 * stride];
133     const int t3 = c[6 * stride];
134 
135     c[0 * stride] = CLIP(t0 + t7);
136     c[1 * stride] = CLIP(t1 + t6);
137     c[2 * stride] = CLIP(t2 + t5);
138     c[3 * stride] = CLIP(t3 + t4);
139     c[4 * stride] = CLIP(t3 - t4);
140     c[5 * stride] = CLIP(t2 - t5);
141     c[6 * stride] = CLIP(t1 - t6);
142     c[7 * stride] = CLIP(t0 - t7);
143 }
144 
inv_dct8_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)145 static void inv_dct8_1d_c(int32_t *const c, const ptrdiff_t stride,
146                           const int min, const int max)
147 {
148     inv_dct8_1d_internal_c(c, stride, min, max, 0);
149 }
150 
151 static NOINLINE void
inv_dct16_1d_internal_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max,int tx64)152 inv_dct16_1d_internal_c(int32_t *const c, const ptrdiff_t stride,
153                         const int min, const int max, int tx64)
154 {
155     assert(stride > 0);
156     inv_dct8_1d_internal_c(c, stride << 1, min, max, tx64);
157 
158     const int in1 = c[1 * stride], in3 = c[3 * stride];
159     const int in5 = c[5 * stride], in7 = c[7 * stride];
160 
161     int t8a, t9a, t10a, t11a, t12a, t13a, t14a, t15a;
162     if (tx64) {
163         t8a  = (in1 *   401 + 2048) >> 12;
164         t9a  = (in7 * -2598 + 2048) >> 12;
165         t10a = (in5 *  1931 + 2048) >> 12;
166         t11a = (in3 * -1189 + 2048) >> 12;
167         t12a = (in3 *  3920 + 2048) >> 12;
168         t13a = (in5 *  3612 + 2048) >> 12;
169         t14a = (in7 *  3166 + 2048) >> 12;
170         t15a = (in1 *  4076 + 2048) >> 12;
171     } else {
172         const int in9  = c[ 9 * stride], in11 = c[11 * stride];
173         const int in13 = c[13 * stride], in15 = c[15 * stride];
174 
175         t8a  = ((in1  *   401         - in15 * (4076 - 4096) + 2048) >> 12) - in15;
176         t9a  =  (in9  *  1583         - in7  *  1299         + 1024) >> 11;
177         t10a = ((in5  *  1931         - in11 * (3612 - 4096) + 2048) >> 12) - in11;
178         t11a = ((in13 * (3920 - 4096) - in3  *  1189         + 2048) >> 12) + in13;
179         t12a = ((in13 *  1189         + in3  * (3920 - 4096) + 2048) >> 12) + in3;
180         t13a = ((in5  * (3612 - 4096) + in11 *  1931         + 2048) >> 12) + in5;
181         t14a =  (in9  *  1299         + in7  *  1583         + 1024) >> 11;
182         t15a = ((in1  * (4076 - 4096) + in15 *   401         + 2048) >> 12) + in1;
183     }
184 
185     int t8  = CLIP(t8a  + t9a);
186     int t9  = CLIP(t8a  - t9a);
187     int t10 = CLIP(t11a - t10a);
188     int t11 = CLIP(t11a + t10a);
189     int t12 = CLIP(t12a + t13a);
190     int t13 = CLIP(t12a - t13a);
191     int t14 = CLIP(t15a - t14a);
192     int t15 = CLIP(t15a + t14a);
193 
194     t9a  = ((  t14 *  1567         - t9  * (3784 - 4096)  + 2048) >> 12) - t9;
195     t14a = ((  t14 * (3784 - 4096) + t9  *  1567          + 2048) >> 12) + t14;
196     t10a = ((-(t13 * (3784 - 4096) + t10 *  1567)         + 2048) >> 12) - t13;
197     t13a = ((  t13 *  1567         - t10 * (3784 - 4096)  + 2048) >> 12) - t10;
198 
199     t8a  = CLIP(t8   + t11);
200     t9   = CLIP(t9a  + t10a);
201     t10  = CLIP(t9a  - t10a);
202     t11a = CLIP(t8   - t11);
203     t12a = CLIP(t15  - t12);
204     t13  = CLIP(t14a - t13a);
205     t14  = CLIP(t14a + t13a);
206     t15a = CLIP(t15  + t12);
207 
208     t10a = ((t13  - t10)  * 181 + 128) >> 8;
209     t13a = ((t13  + t10)  * 181 + 128) >> 8;
210     t11  = ((t12a - t11a) * 181 + 128) >> 8;
211     t12  = ((t12a + t11a) * 181 + 128) >> 8;
212 
213     const int t0 = c[ 0 * stride];
214     const int t1 = c[ 2 * stride];
215     const int t2 = c[ 4 * stride];
216     const int t3 = c[ 6 * stride];
217     const int t4 = c[ 8 * stride];
218     const int t5 = c[10 * stride];
219     const int t6 = c[12 * stride];
220     const int t7 = c[14 * stride];
221 
222     c[ 0 * stride] = CLIP(t0 + t15a);
223     c[ 1 * stride] = CLIP(t1 + t14);
224     c[ 2 * stride] = CLIP(t2 + t13a);
225     c[ 3 * stride] = CLIP(t3 + t12);
226     c[ 4 * stride] = CLIP(t4 + t11);
227     c[ 5 * stride] = CLIP(t5 + t10a);
228     c[ 6 * stride] = CLIP(t6 + t9);
229     c[ 7 * stride] = CLIP(t7 + t8a);
230     c[ 8 * stride] = CLIP(t7 - t8a);
231     c[ 9 * stride] = CLIP(t6 - t9);
232     c[10 * stride] = CLIP(t5 - t10a);
233     c[11 * stride] = CLIP(t4 - t11);
234     c[12 * stride] = CLIP(t3 - t12);
235     c[13 * stride] = CLIP(t2 - t13a);
236     c[14 * stride] = CLIP(t1 - t14);
237     c[15 * stride] = CLIP(t0 - t15a);
238 }
239 
inv_dct16_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)240 static void inv_dct16_1d_c(int32_t *const c, const ptrdiff_t stride,
241                            const int min, const int max)
242 {
243     inv_dct16_1d_internal_c(c, stride, min, max, 0);
244 }
245 
246 static NOINLINE void
inv_dct32_1d_internal_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max,const int tx64)247 inv_dct32_1d_internal_c(int32_t *const c, const ptrdiff_t stride,
248                         const int min, const int max, const int tx64)
249 {
250     assert(stride > 0);
251     inv_dct16_1d_internal_c(c, stride << 1, min, max, tx64);
252 
253     const int in1  = c[ 1 * stride], in3  = c[ 3 * stride];
254     const int in5  = c[ 5 * stride], in7  = c[ 7 * stride];
255     const int in9  = c[ 9 * stride], in11 = c[11 * stride];
256     const int in13 = c[13 * stride], in15 = c[15 * stride];
257 
258     int t16a, t17a, t18a, t19a, t20a, t21a, t22a, t23a;
259     int t24a, t25a, t26a, t27a, t28a, t29a, t30a, t31a;
260     if (tx64) {
261         t16a = (in1  *   201 + 2048) >> 12;
262         t17a = (in15 * -2751 + 2048) >> 12;
263         t18a = (in9  *  1751 + 2048) >> 12;
264         t19a = (in7  * -1380 + 2048) >> 12;
265         t20a = (in5  *   995 + 2048) >> 12;
266         t21a = (in11 * -2106 + 2048) >> 12;
267         t22a = (in13 *  2440 + 2048) >> 12;
268         t23a = (in3  *  -601 + 2048) >> 12;
269         t24a = (in3  *  4052 + 2048) >> 12;
270         t25a = (in13 *  3290 + 2048) >> 12;
271         t26a = (in11 *  3513 + 2048) >> 12;
272         t27a = (in5  *  3973 + 2048) >> 12;
273         t28a = (in7  *  3857 + 2048) >> 12;
274         t29a = (in9  *  3703 + 2048) >> 12;
275         t30a = (in15 *  3035 + 2048) >> 12;
276         t31a = (in1  *  4091 + 2048) >> 12;
277     } else {
278         const int in17 = c[17 * stride], in19 = c[19 * stride];
279         const int in21 = c[21 * stride], in23 = c[23 * stride];
280         const int in25 = c[25 * stride], in27 = c[27 * stride];
281         const int in29 = c[29 * stride], in31 = c[31 * stride];
282 
283         t16a = ((in1  *   201         - in31 * (4091 - 4096) + 2048) >> 12) - in31;
284         t17a = ((in17 * (3035 - 4096) - in15 *  2751         + 2048) >> 12) + in17;
285         t18a = ((in9  *  1751         - in23 * (3703 - 4096) + 2048) >> 12) - in23;
286         t19a = ((in25 * (3857 - 4096) - in7  *  1380         + 2048) >> 12) + in25;
287         t20a = ((in5  *   995         - in27 * (3973 - 4096) + 2048) >> 12) - in27;
288         t21a = ((in21 * (3513 - 4096) - in11 *  2106         + 2048) >> 12) + in21;
289         t22a =  (in13 *  1220         - in19 *  1645         + 1024) >> 11;
290         t23a = ((in29 * (4052 - 4096) - in3  *   601         + 2048) >> 12) + in29;
291         t24a = ((in29 *   601         + in3  * (4052 - 4096) + 2048) >> 12) + in3;
292         t25a =  (in13 *  1645         + in19 *  1220         + 1024) >> 11;
293         t26a = ((in21 *  2106         + in11 * (3513 - 4096) + 2048) >> 12) + in11;
294         t27a = ((in5  * (3973 - 4096) + in27 *   995         + 2048) >> 12) + in5;
295         t28a = ((in25 *  1380         + in7  * (3857 - 4096) + 2048) >> 12) + in7;
296         t29a = ((in9  * (3703 - 4096) + in23 *  1751         + 2048) >> 12) + in9;
297         t30a = ((in17 *  2751         + in15 * (3035 - 4096) + 2048) >> 12) + in15;
298         t31a = ((in1  * (4091 - 4096) + in31 *   201         + 2048) >> 12) + in1;
299     }
300 
301     int t16 = CLIP(t16a + t17a);
302     int t17 = CLIP(t16a - t17a);
303     int t18 = CLIP(t19a - t18a);
304     int t19 = CLIP(t19a + t18a);
305     int t20 = CLIP(t20a + t21a);
306     int t21 = CLIP(t20a - t21a);
307     int t22 = CLIP(t23a - t22a);
308     int t23 = CLIP(t23a + t22a);
309     int t24 = CLIP(t24a + t25a);
310     int t25 = CLIP(t24a - t25a);
311     int t26 = CLIP(t27a - t26a);
312     int t27 = CLIP(t27a + t26a);
313     int t28 = CLIP(t28a + t29a);
314     int t29 = CLIP(t28a - t29a);
315     int t30 = CLIP(t31a - t30a);
316     int t31 = CLIP(t31a + t30a);
317 
318     t17a = ((  t30 *   799         - t17 * (4017 - 4096)  + 2048) >> 12) - t17;
319     t30a = ((  t30 * (4017 - 4096) + t17 *   799          + 2048) >> 12) + t30;
320     t18a = ((-(t29 * (4017 - 4096) + t18 *   799)         + 2048) >> 12) - t29;
321     t29a = ((  t29 *   799         - t18 * (4017 - 4096)  + 2048) >> 12) - t18;
322     t21a =  (  t26 *  1703         - t21 *  1138          + 1024) >> 11;
323     t26a =  (  t26 *  1138         + t21 *  1703          + 1024) >> 11;
324     t22a =  (-(t25 *  1138         + t22 *  1703        ) + 1024) >> 11;
325     t25a =  (  t25 *  1703         - t22 *  1138          + 1024) >> 11;
326 
327     t16a = CLIP(t16  + t19);
328     t17  = CLIP(t17a + t18a);
329     t18  = CLIP(t17a - t18a);
330     t19a = CLIP(t16  - t19);
331     t20a = CLIP(t23  - t20);
332     t21  = CLIP(t22a - t21a);
333     t22  = CLIP(t22a + t21a);
334     t23a = CLIP(t23  + t20);
335     t24a = CLIP(t24  + t27);
336     t25  = CLIP(t25a + t26a);
337     t26  = CLIP(t25a - t26a);
338     t27a = CLIP(t24  - t27);
339     t28a = CLIP(t31  - t28);
340     t29  = CLIP(t30a - t29a);
341     t30  = CLIP(t30a + t29a);
342     t31a = CLIP(t31  + t28);
343 
344     t18a = ((  t29  *  1567         - t18  * (3784 - 4096)  + 2048) >> 12) - t18;
345     t29a = ((  t29  * (3784 - 4096) + t18  *  1567          + 2048) >> 12) + t29;
346     t19  = ((  t28a *  1567         - t19a * (3784 - 4096)  + 2048) >> 12) - t19a;
347     t28  = ((  t28a * (3784 - 4096) + t19a *  1567          + 2048) >> 12) + t28a;
348     t20  = ((-(t27a * (3784 - 4096) + t20a *  1567)         + 2048) >> 12) - t27a;
349     t27  = ((  t27a *  1567         - t20a * (3784 - 4096)  + 2048) >> 12) - t20a;
350     t21a = ((-(t26  * (3784 - 4096) + t21  *  1567)         + 2048) >> 12) - t26;
351     t26a = ((  t26  *  1567         - t21  * (3784 - 4096)  + 2048) >> 12) - t21;
352 
353     t16  = CLIP(t16a + t23a);
354     t17a = CLIP(t17  + t22);
355     t18  = CLIP(t18a + t21a);
356     t19a = CLIP(t19  + t20);
357     t20a = CLIP(t19  - t20);
358     t21  = CLIP(t18a - t21a);
359     t22a = CLIP(t17  - t22);
360     t23  = CLIP(t16a - t23a);
361     t24  = CLIP(t31a - t24a);
362     t25a = CLIP(t30  - t25);
363     t26  = CLIP(t29a - t26a);
364     t27a = CLIP(t28  - t27);
365     t28a = CLIP(t28  + t27);
366     t29  = CLIP(t29a + t26a);
367     t30a = CLIP(t30  + t25);
368     t31  = CLIP(t31a + t24a);
369 
370     t20  = ((t27a - t20a) * 181 + 128) >> 8;
371     t27  = ((t27a + t20a) * 181 + 128) >> 8;
372     t21a = ((t26  - t21 ) * 181 + 128) >> 8;
373     t26a = ((t26  + t21 ) * 181 + 128) >> 8;
374     t22  = ((t25a - t22a) * 181 + 128) >> 8;
375     t25  = ((t25a + t22a) * 181 + 128) >> 8;
376     t23a = ((t24  - t23 ) * 181 + 128) >> 8;
377     t24a = ((t24  + t23 ) * 181 + 128) >> 8;
378 
379     const int t0  = c[ 0 * stride];
380     const int t1  = c[ 2 * stride];
381     const int t2  = c[ 4 * stride];
382     const int t3  = c[ 6 * stride];
383     const int t4  = c[ 8 * stride];
384     const int t5  = c[10 * stride];
385     const int t6  = c[12 * stride];
386     const int t7  = c[14 * stride];
387     const int t8  = c[16 * stride];
388     const int t9  = c[18 * stride];
389     const int t10 = c[20 * stride];
390     const int t11 = c[22 * stride];
391     const int t12 = c[24 * stride];
392     const int t13 = c[26 * stride];
393     const int t14 = c[28 * stride];
394     const int t15 = c[30 * stride];
395 
396     c[ 0 * stride] = CLIP(t0  + t31);
397     c[ 1 * stride] = CLIP(t1  + t30a);
398     c[ 2 * stride] = CLIP(t2  + t29);
399     c[ 3 * stride] = CLIP(t3  + t28a);
400     c[ 4 * stride] = CLIP(t4  + t27);
401     c[ 5 * stride] = CLIP(t5  + t26a);
402     c[ 6 * stride] = CLIP(t6  + t25);
403     c[ 7 * stride] = CLIP(t7  + t24a);
404     c[ 8 * stride] = CLIP(t8  + t23a);
405     c[ 9 * stride] = CLIP(t9  + t22);
406     c[10 * stride] = CLIP(t10 + t21a);
407     c[11 * stride] = CLIP(t11 + t20);
408     c[12 * stride] = CLIP(t12 + t19a);
409     c[13 * stride] = CLIP(t13 + t18);
410     c[14 * stride] = CLIP(t14 + t17a);
411     c[15 * stride] = CLIP(t15 + t16);
412     c[16 * stride] = CLIP(t15 - t16);
413     c[17 * stride] = CLIP(t14 - t17a);
414     c[18 * stride] = CLIP(t13 - t18);
415     c[19 * stride] = CLIP(t12 - t19a);
416     c[20 * stride] = CLIP(t11 - t20);
417     c[21 * stride] = CLIP(t10 - t21a);
418     c[22 * stride] = CLIP(t9  - t22);
419     c[23 * stride] = CLIP(t8  - t23a);
420     c[24 * stride] = CLIP(t7  - t24a);
421     c[25 * stride] = CLIP(t6  - t25);
422     c[26 * stride] = CLIP(t5  - t26a);
423     c[27 * stride] = CLIP(t4  - t27);
424     c[28 * stride] = CLIP(t3  - t28a);
425     c[29 * stride] = CLIP(t2  - t29);
426     c[30 * stride] = CLIP(t1  - t30a);
427     c[31 * stride] = CLIP(t0  - t31);
428 }
429 
inv_dct32_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)430 static void inv_dct32_1d_c(int32_t *const c, const ptrdiff_t stride,
431                            const int min, const int max)
432 {
433     inv_dct32_1d_internal_c(c, stride, min, max, 0);
434 }
435 
inv_dct64_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)436 static void inv_dct64_1d_c(int32_t *const c, const ptrdiff_t stride,
437                            const int min, const int max)
438 {
439     assert(stride > 0);
440     inv_dct32_1d_internal_c(c, stride << 1, min, max, 1);
441 
442     const int in1  = c[ 1 * stride], in3  = c[ 3 * stride];
443     const int in5  = c[ 5 * stride], in7  = c[ 7 * stride];
444     const int in9  = c[ 9 * stride], in11 = c[11 * stride];
445     const int in13 = c[13 * stride], in15 = c[15 * stride];
446     const int in17 = c[17 * stride], in19 = c[19 * stride];
447     const int in21 = c[21 * stride], in23 = c[23 * stride];
448     const int in25 = c[25 * stride], in27 = c[27 * stride];
449     const int in29 = c[29 * stride], in31 = c[31 * stride];
450 
451     int t32a = (in1  *   101 + 2048) >> 12;
452     int t33a = (in31 * -2824 + 2048) >> 12;
453     int t34a = (in17 *  1660 + 2048) >> 12;
454     int t35a = (in15 * -1474 + 2048) >> 12;
455     int t36a = (in9  *   897 + 2048) >> 12;
456     int t37a = (in23 * -2191 + 2048) >> 12;
457     int t38a = (in25 *  2359 + 2048) >> 12;
458     int t39a = (in7  *  -700 + 2048) >> 12;
459     int t40a = (in5  *   501 + 2048) >> 12;
460     int t41a = (in27 * -2520 + 2048) >> 12;
461     int t42a = (in21 *  2019 + 2048) >> 12;
462     int t43a = (in11 * -1092 + 2048) >> 12;
463     int t44a = (in13 *  1285 + 2048) >> 12;
464     int t45a = (in19 * -1842 + 2048) >> 12;
465     int t46a = (in29 *  2675 + 2048) >> 12;
466     int t47a = (in3  *  -301 + 2048) >> 12;
467     int t48a = (in3  *  4085 + 2048) >> 12;
468     int t49a = (in29 *  3102 + 2048) >> 12;
469     int t50a = (in19 *  3659 + 2048) >> 12;
470     int t51a = (in13 *  3889 + 2048) >> 12;
471     int t52a = (in11 *  3948 + 2048) >> 12;
472     int t53a = (in21 *  3564 + 2048) >> 12;
473     int t54a = (in27 *  3229 + 2048) >> 12;
474     int t55a = (in5  *  4065 + 2048) >> 12;
475     int t56a = (in7  *  4036 + 2048) >> 12;
476     int t57a = (in25 *  3349 + 2048) >> 12;
477     int t58a = (in23 *  3461 + 2048) >> 12;
478     int t59a = (in9  *  3996 + 2048) >> 12;
479     int t60a = (in15 *  3822 + 2048) >> 12;
480     int t61a = (in17 *  3745 + 2048) >> 12;
481     int t62a = (in31 *  2967 + 2048) >> 12;
482     int t63a = (in1  *  4095 + 2048) >> 12;
483 
484     int t32 = CLIP(t32a + t33a);
485     int t33 = CLIP(t32a - t33a);
486     int t34 = CLIP(t35a - t34a);
487     int t35 = CLIP(t35a + t34a);
488     int t36 = CLIP(t36a + t37a);
489     int t37 = CLIP(t36a - t37a);
490     int t38 = CLIP(t39a - t38a);
491     int t39 = CLIP(t39a + t38a);
492     int t40 = CLIP(t40a + t41a);
493     int t41 = CLIP(t40a - t41a);
494     int t42 = CLIP(t43a - t42a);
495     int t43 = CLIP(t43a + t42a);
496     int t44 = CLIP(t44a + t45a);
497     int t45 = CLIP(t44a - t45a);
498     int t46 = CLIP(t47a - t46a);
499     int t47 = CLIP(t47a + t46a);
500     int t48 = CLIP(t48a + t49a);
501     int t49 = CLIP(t48a - t49a);
502     int t50 = CLIP(t51a - t50a);
503     int t51 = CLIP(t51a + t50a);
504     int t52 = CLIP(t52a + t53a);
505     int t53 = CLIP(t52a - t53a);
506     int t54 = CLIP(t55a - t54a);
507     int t55 = CLIP(t55a + t54a);
508     int t56 = CLIP(t56a + t57a);
509     int t57 = CLIP(t56a - t57a);
510     int t58 = CLIP(t59a - t58a);
511     int t59 = CLIP(t59a + t58a);
512     int t60 = CLIP(t60a + t61a);
513     int t61 = CLIP(t60a - t61a);
514     int t62 = CLIP(t63a - t62a);
515     int t63 = CLIP(t63a + t62a);
516 
517     t33a = ((t33 * (4096 - 4076) + t62 *   401         + 2048) >> 12) - t33;
518     t34a = ((t34 *  -401         + t61 * (4096 - 4076) + 2048) >> 12) - t61;
519     t37a =  (t37 * -1299         + t58 *  1583         + 1024) >> 11;
520     t38a =  (t38 * -1583         + t57 * -1299         + 1024) >> 11;
521     t41a = ((t41 * (4096 - 3612) + t54 *  1931         + 2048) >> 12) - t41;
522     t42a = ((t42 * -1931         + t53 * (4096 - 3612) + 2048) >> 12) - t53;
523     t45a = ((t45 * -1189         + t50 * (3920 - 4096) + 2048) >> 12) + t50;
524     t46a = ((t46 * (4096 - 3920) + t49 * -1189         + 2048) >> 12) - t46;
525     t49a = ((t46 * -1189         + t49 * (3920 - 4096) + 2048) >> 12) + t49;
526     t50a = ((t45 * (3920 - 4096) + t50 *  1189         + 2048) >> 12) + t45;
527     t53a = ((t42 * (4096 - 3612) + t53 *  1931         + 2048) >> 12) - t42;
528     t54a = ((t41 *  1931         + t54 * (3612 - 4096) + 2048) >> 12) + t54;
529     t57a =  (t38 * -1299         + t57 *  1583         + 1024) >> 11;
530     t58a =  (t37 *  1583         + t58 *  1299         + 1024) >> 11;
531     t61a = ((t34 * (4096 - 4076) + t61 *   401         + 2048) >> 12) - t34;
532     t62a = ((t33 *   401         + t62 * (4076 - 4096) + 2048) >> 12) + t62;
533 
534     t32a = CLIP(t32  + t35);
535     t33  = CLIP(t33a + t34a);
536     t34  = CLIP(t33a - t34a);
537     t35a = CLIP(t32  - t35);
538     t36a = CLIP(t39  - t36);
539     t37  = CLIP(t38a - t37a);
540     t38  = CLIP(t38a + t37a);
541     t39a = CLIP(t39  + t36);
542     t40a = CLIP(t40  + t43);
543     t41  = CLIP(t41a + t42a);
544     t42  = CLIP(t41a - t42a);
545     t43a = CLIP(t40  - t43);
546     t44a = CLIP(t47  - t44);
547     t45  = CLIP(t46a - t45a);
548     t46  = CLIP(t46a + t45a);
549     t47a = CLIP(t47  + t44);
550     t48a = CLIP(t48  + t51);
551     t49  = CLIP(t49a + t50a);
552     t50  = CLIP(t49a - t50a);
553     t51a = CLIP(t48  - t51);
554     t52a = CLIP(t55  - t52);
555     t53  = CLIP(t54a - t53a);
556     t54  = CLIP(t54a + t53a);
557     t55a = CLIP(t55  + t52);
558     t56a = CLIP(t56  + t59);
559     t57  = CLIP(t57a + t58a);
560     t58  = CLIP(t57a - t58a);
561     t59a = CLIP(t56  - t59);
562     t60a = CLIP(t63  - t60);
563     t61  = CLIP(t62a - t61a);
564     t62  = CLIP(t62a + t61a);
565     t63a = CLIP(t63  + t60);
566 
567     t34a = ((t34  * (4096 - 4017) + t61  *   799         + 2048) >> 12) - t34;
568     t35  = ((t35a * (4096 - 4017) + t60a *   799         + 2048) >> 12) - t35a;
569     t36  = ((t36a *  -799         + t59a * (4096 - 4017) + 2048) >> 12) - t59a;
570     t37a = ((t37  *  -799         + t58  * (4096 - 4017) + 2048) >> 12) - t58;
571     t42a =  (t42  * -1138         + t53  *  1703         + 1024) >> 11;
572     t43  =  (t43a * -1138         + t52a *  1703         + 1024) >> 11;
573     t44  =  (t44a * -1703         + t51a * -1138         + 1024) >> 11;
574     t45a =  (t45  * -1703         + t50  * -1138         + 1024) >> 11;
575     t50a =  (t45  * -1138         + t50  *  1703         + 1024) >> 11;
576     t51  =  (t44a * -1138         + t51a *  1703         + 1024) >> 11;
577     t52  =  (t43a *  1703         + t52a *  1138         + 1024) >> 11;
578     t53a =  (t42  *  1703         + t53  *  1138         + 1024) >> 11;
579     t58a = ((t37  * (4096 - 4017) + t58  *   799         + 2048) >> 12) - t37;
580     t59  = ((t36a * (4096 - 4017) + t59a *   799         + 2048) >> 12) - t36a;
581     t60  = ((t35a *   799         + t60a * (4017 - 4096) + 2048) >> 12) + t60a;
582     t61a = ((t34  *   799         + t61  * (4017 - 4096) + 2048) >> 12) + t61;
583 
584     t32  = CLIP(t32a + t39a);
585     t33a = CLIP(t33  + t38);
586     t34  = CLIP(t34a + t37a);
587     t35a = CLIP(t35  + t36);
588     t36a = CLIP(t35  - t36);
589     t37  = CLIP(t34a - t37a);
590     t38a = CLIP(t33  - t38);
591     t39  = CLIP(t32a - t39a);
592     t40  = CLIP(t47a - t40a);
593     t41a = CLIP(t46  - t41);
594     t42  = CLIP(t45a - t42a);
595     t43a = CLIP(t44  - t43);
596     t44a = CLIP(t44  + t43);
597     t45  = CLIP(t45a + t42a);
598     t46a = CLIP(t46  + t41);
599     t47  = CLIP(t47a + t40a);
600     t48  = CLIP(t48a + t55a);
601     t49a = CLIP(t49  + t54);
602     t50  = CLIP(t50a + t53a);
603     t51a = CLIP(t51  + t52);
604     t52a = CLIP(t51  - t52);
605     t53  = CLIP(t50a - t53a);
606     t54a = CLIP(t49  - t54);
607     t55  = CLIP(t48a - t55a);
608     t56  = CLIP(t63a - t56a);
609     t57a = CLIP(t62  - t57);
610     t58  = CLIP(t61a - t58a);
611     t59a = CLIP(t60  - t59);
612     t60a = CLIP(t60  + t59);
613     t61  = CLIP(t61a + t58a);
614     t62a = CLIP(t62  + t57);
615     t63  = CLIP(t63a + t56a);
616 
617     t36  = ((t36a * (4096 - 3784) + t59a *  1567         + 2048) >> 12) - t36a;
618     t37a = ((t37  * (4096 - 3784) + t58  *  1567         + 2048) >> 12) - t37;
619     t38  = ((t38a * (4096 - 3784) + t57a *  1567         + 2048) >> 12) - t38a;
620     t39a = ((t39  * (4096 - 3784) + t56  *  1567         + 2048) >> 12) - t39;
621     t40a = ((t40  * -1567         + t55  * (4096 - 3784) + 2048) >> 12) - t55;
622     t41  = ((t41a * -1567         + t54a * (4096 - 3784) + 2048) >> 12) - t54a;
623     t42a = ((t42  * -1567         + t53  * (4096 - 3784) + 2048) >> 12) - t53;
624     t43  = ((t43a * -1567         + t52a * (4096 - 3784) + 2048) >> 12) - t52a;
625     t52  = ((t43a * (4096 - 3784) + t52a *  1567         + 2048) >> 12) - t43a;
626     t53a = ((t42  * (4096 - 3784) + t53  *  1567         + 2048) >> 12) - t42;
627     t54  = ((t41a * (4096 - 3784) + t54a *  1567         + 2048) >> 12) - t41a;
628     t55a = ((t40  * (4096 - 3784) + t55  *  1567         + 2048) >> 12) - t40;
629     t56a = ((t39  *  1567         + t56  * (3784 - 4096) + 2048) >> 12) + t56;
630     t57  = ((t38a *  1567         + t57a * (3784 - 4096) + 2048) >> 12) + t57a;
631     t58a = ((t37  *  1567         + t58  * (3784 - 4096) + 2048) >> 12) + t58;
632     t59  = ((t36a *  1567         + t59a * (3784 - 4096) + 2048) >> 12) + t59a;
633 
634     t32a = CLIP(t32  + t47);
635     t33  = CLIP(t33a + t46a);
636     t34a = CLIP(t34  + t45);
637     t35  = CLIP(t35a + t44a);
638     t36a = CLIP(t36  + t43);
639     t37  = CLIP(t37a + t42a);
640     t38a = CLIP(t38  + t41);
641     t39  = CLIP(t39a + t40a);
642     t40  = CLIP(t39a - t40a);
643     t41a = CLIP(t38  - t41);
644     t42  = CLIP(t37a - t42a);
645     t43a = CLIP(t36  - t43);
646     t44  = CLIP(t35a - t44a);
647     t45a = CLIP(t34  - t45);
648     t46  = CLIP(t33a - t46a);
649     t47a = CLIP(t32  - t47);
650     t48a = CLIP(t63  - t48);
651     t49  = CLIP(t62a - t49a);
652     t50a = CLIP(t61  - t50);
653     t51  = CLIP(t60a - t51a);
654     t52a = CLIP(t59  - t52);
655     t53  = CLIP(t58a - t53a);
656     t54a = CLIP(t57  - t54);
657     t55  = CLIP(t56a - t55a);
658     t56  = CLIP(t56a + t55a);
659     t57a = CLIP(t57  + t54);
660     t58  = CLIP(t58a + t53a);
661     t59a = CLIP(t59  + t52);
662     t60  = CLIP(t60a + t51a);
663     t61a = CLIP(t61  + t50);
664     t62  = CLIP(t62a + t49a);
665     t63a = CLIP(t63  + t48);
666 
667     t40a = ((t55  - t40 ) * 181 + 128) >> 8;
668     t41  = ((t54a - t41a) * 181 + 128) >> 8;
669     t42a = ((t53  - t42 ) * 181 + 128) >> 8;
670     t43  = ((t52a - t43a) * 181 + 128) >> 8;
671     t44a = ((t51  - t44 ) * 181 + 128) >> 8;
672     t45  = ((t50a - t45a) * 181 + 128) >> 8;
673     t46a = ((t49  - t46 ) * 181 + 128) >> 8;
674     t47  = ((t48a - t47a) * 181 + 128) >> 8;
675     t48  = ((t47a + t48a) * 181 + 128) >> 8;
676     t49a = ((t46  + t49 ) * 181 + 128) >> 8;
677     t50  = ((t45a + t50a) * 181 + 128) >> 8;
678     t51a = ((t44  + t51 ) * 181 + 128) >> 8;
679     t52  = ((t43a + t52a) * 181 + 128) >> 8;
680     t53a = ((t42  + t53 ) * 181 + 128) >> 8;
681     t54  = ((t41a + t54a) * 181 + 128) >> 8;
682     t55a = ((t40  + t55 ) * 181 + 128) >> 8;
683 
684     const int t0  = c[ 0 * stride];
685     const int t1  = c[ 2 * stride];
686     const int t2  = c[ 4 * stride];
687     const int t3  = c[ 6 * stride];
688     const int t4  = c[ 8 * stride];
689     const int t5  = c[10 * stride];
690     const int t6  = c[12 * stride];
691     const int t7  = c[14 * stride];
692     const int t8  = c[16 * stride];
693     const int t9  = c[18 * stride];
694     const int t10 = c[20 * stride];
695     const int t11 = c[22 * stride];
696     const int t12 = c[24 * stride];
697     const int t13 = c[26 * stride];
698     const int t14 = c[28 * stride];
699     const int t15 = c[30 * stride];
700     const int t16 = c[32 * stride];
701     const int t17 = c[34 * stride];
702     const int t18 = c[36 * stride];
703     const int t19 = c[38 * stride];
704     const int t20 = c[40 * stride];
705     const int t21 = c[42 * stride];
706     const int t22 = c[44 * stride];
707     const int t23 = c[46 * stride];
708     const int t24 = c[48 * stride];
709     const int t25 = c[50 * stride];
710     const int t26 = c[52 * stride];
711     const int t27 = c[54 * stride];
712     const int t28 = c[56 * stride];
713     const int t29 = c[58 * stride];
714     const int t30 = c[60 * stride];
715     const int t31 = c[62 * stride];
716 
717     c[ 0 * stride] = CLIP(t0  + t63a);
718     c[ 1 * stride] = CLIP(t1  + t62);
719     c[ 2 * stride] = CLIP(t2  + t61a);
720     c[ 3 * stride] = CLIP(t3  + t60);
721     c[ 4 * stride] = CLIP(t4  + t59a);
722     c[ 5 * stride] = CLIP(t5  + t58);
723     c[ 6 * stride] = CLIP(t6  + t57a);
724     c[ 7 * stride] = CLIP(t7  + t56);
725     c[ 8 * stride] = CLIP(t8  + t55a);
726     c[ 9 * stride] = CLIP(t9  + t54);
727     c[10 * stride] = CLIP(t10 + t53a);
728     c[11 * stride] = CLIP(t11 + t52);
729     c[12 * stride] = CLIP(t12 + t51a);
730     c[13 * stride] = CLIP(t13 + t50);
731     c[14 * stride] = CLIP(t14 + t49a);
732     c[15 * stride] = CLIP(t15 + t48);
733     c[16 * stride] = CLIP(t16 + t47);
734     c[17 * stride] = CLIP(t17 + t46a);
735     c[18 * stride] = CLIP(t18 + t45);
736     c[19 * stride] = CLIP(t19 + t44a);
737     c[20 * stride] = CLIP(t20 + t43);
738     c[21 * stride] = CLIP(t21 + t42a);
739     c[22 * stride] = CLIP(t22 + t41);
740     c[23 * stride] = CLIP(t23 + t40a);
741     c[24 * stride] = CLIP(t24 + t39);
742     c[25 * stride] = CLIP(t25 + t38a);
743     c[26 * stride] = CLIP(t26 + t37);
744     c[27 * stride] = CLIP(t27 + t36a);
745     c[28 * stride] = CLIP(t28 + t35);
746     c[29 * stride] = CLIP(t29 + t34a);
747     c[30 * stride] = CLIP(t30 + t33);
748     c[31 * stride] = CLIP(t31 + t32a);
749     c[32 * stride] = CLIP(t31 - t32a);
750     c[33 * stride] = CLIP(t30 - t33);
751     c[34 * stride] = CLIP(t29 - t34a);
752     c[35 * stride] = CLIP(t28 - t35);
753     c[36 * stride] = CLIP(t27 - t36a);
754     c[37 * stride] = CLIP(t26 - t37);
755     c[38 * stride] = CLIP(t25 - t38a);
756     c[39 * stride] = CLIP(t24 - t39);
757     c[40 * stride] = CLIP(t23 - t40a);
758     c[41 * stride] = CLIP(t22 - t41);
759     c[42 * stride] = CLIP(t21 - t42a);
760     c[43 * stride] = CLIP(t20 - t43);
761     c[44 * stride] = CLIP(t19 - t44a);
762     c[45 * stride] = CLIP(t18 - t45);
763     c[46 * stride] = CLIP(t17 - t46a);
764     c[47 * stride] = CLIP(t16 - t47);
765     c[48 * stride] = CLIP(t15 - t48);
766     c[49 * stride] = CLIP(t14 - t49a);
767     c[50 * stride] = CLIP(t13 - t50);
768     c[51 * stride] = CLIP(t12 - t51a);
769     c[52 * stride] = CLIP(t11 - t52);
770     c[53 * stride] = CLIP(t10 - t53a);
771     c[54 * stride] = CLIP(t9  - t54);
772     c[55 * stride] = CLIP(t8  - t55a);
773     c[56 * stride] = CLIP(t7  - t56);
774     c[57 * stride] = CLIP(t6  - t57a);
775     c[58 * stride] = CLIP(t5  - t58);
776     c[59 * stride] = CLIP(t4  - t59a);
777     c[60 * stride] = CLIP(t3  - t60);
778     c[61 * stride] = CLIP(t2  - t61a);
779     c[62 * stride] = CLIP(t1  - t62);
780     c[63 * stride] = CLIP(t0  - t63a);
781 }
782 
783 static NOINLINE void
inv_adst4_1d_internal_c(const int32_t * const in,const ptrdiff_t in_s,const int min,const int max,int32_t * const out,const ptrdiff_t out_s)784 inv_adst4_1d_internal_c(const int32_t *const in, const ptrdiff_t in_s,
785                         const int min, const int max,
786                         int32_t *const out, const ptrdiff_t out_s)
787 {
788     assert(in_s > 0 && out_s != 0);
789     const int in0 = in[0 * in_s], in1 = in[1 * in_s];
790     const int in2 = in[2 * in_s], in3 = in[3 * in_s];
791 
792     out[0 * out_s] = (( 1321         * in0 + (3803 - 4096) * in2 +
793                        (2482 - 4096) * in3 + (3344 - 4096) * in1 + 2048) >> 12) +
794                      in2 + in3 + in1;
795     out[1 * out_s] = (((2482 - 4096) * in0 -  1321         * in2 -
796                        (3803 - 4096) * in3 + (3344 - 4096) * in1 + 2048) >> 12) +
797                      in0 - in3 + in1;
798     out[2 * out_s] = (209 * (in0 - in2 + in3) + 128) >> 8;
799     out[3 * out_s] = (((3803 - 4096) * in0 + (2482 - 4096) * in2 -
800                         1321         * in3 - (3344 - 4096) * in1 + 2048) >> 12) +
801                      in0 + in2 - in1;
802 }
803 
804 static NOINLINE void
inv_adst8_1d_internal_c(const int32_t * const in,const ptrdiff_t in_s,const int min,const int max,int32_t * const out,const ptrdiff_t out_s)805 inv_adst8_1d_internal_c(const int32_t *const in, const ptrdiff_t in_s,
806                         const int min, const int max,
807                         int32_t *const out, const ptrdiff_t out_s)
808 {
809     assert(in_s > 0 && out_s != 0);
810     const int in0 = in[0 * in_s], in1 = in[1 * in_s];
811     const int in2 = in[2 * in_s], in3 = in[3 * in_s];
812     const int in4 = in[4 * in_s], in5 = in[5 * in_s];
813     const int in6 = in[6 * in_s], in7 = in[7 * in_s];
814 
815     const int t0a = (((4076 - 4096) * in7 +   401         * in0 + 2048) >> 12) + in7;
816     const int t1a = ((  401         * in7 - (4076 - 4096) * in0 + 2048) >> 12) - in0;
817     const int t2a = (((3612 - 4096) * in5 +  1931         * in2 + 2048) >> 12) + in5;
818     const int t3a = (( 1931         * in5 - (3612 - 4096) * in2 + 2048) >> 12) - in2;
819           int t4a =  ( 1299         * in3 +  1583         * in4 + 1024) >> 11;
820           int t5a =  ( 1583         * in3 -  1299         * in4 + 1024) >> 11;
821           int t6a = (( 1189         * in1 + (3920 - 4096) * in6 + 2048) >> 12) + in6;
822           int t7a = (((3920 - 4096) * in1 -  1189         * in6 + 2048) >> 12) + in1;
823 
824     const int t0 = CLIP(t0a + t4a);
825     const int t1 = CLIP(t1a + t5a);
826           int t2 = CLIP(t2a + t6a);
827           int t3 = CLIP(t3a + t7a);
828     const int t4 = CLIP(t0a - t4a);
829     const int t5 = CLIP(t1a - t5a);
830           int t6 = CLIP(t2a - t6a);
831           int t7 = CLIP(t3a - t7a);
832 
833     t4a = (((3784 - 4096) * t4 +  1567         * t5 + 2048) >> 12) + t4;
834     t5a = (( 1567         * t4 - (3784 - 4096) * t5 + 2048) >> 12) - t5;
835     t6a = (((3784 - 4096) * t7 -  1567         * t6 + 2048) >> 12) + t7;
836     t7a = (( 1567         * t7 + (3784 - 4096) * t6 + 2048) >> 12) + t6;
837 
838     out[0 * out_s] =  CLIP(t0  + t2 );
839     out[7 * out_s] = -CLIP(t1  + t3 );
840     t2             =  CLIP(t0  - t2 );
841     t3             =  CLIP(t1  - t3 );
842     out[1 * out_s] = -CLIP(t4a + t6a);
843     out[6 * out_s] =  CLIP(t5a + t7a);
844     t6             =  CLIP(t4a - t6a);
845     t7             =  CLIP(t5a - t7a);
846 
847     out[3 * out_s] = -(((t2 + t3) * 181 + 128) >> 8);
848     out[4 * out_s] =   ((t2 - t3) * 181 + 128) >> 8;
849     out[2 * out_s] =   ((t6 + t7) * 181 + 128) >> 8;
850     out[5 * out_s] = -(((t6 - t7) * 181 + 128) >> 8);
851 }
852 
853 static NOINLINE void
inv_adst16_1d_internal_c(const int32_t * const in,const ptrdiff_t in_s,const int min,const int max,int32_t * const out,const ptrdiff_t out_s)854 inv_adst16_1d_internal_c(const int32_t *const in, const ptrdiff_t in_s,
855                          const int min, const int max,
856                          int32_t *const out, const ptrdiff_t out_s)
857 {
858     assert(in_s > 0 && out_s != 0);
859     const int in0  = in[ 0 * in_s], in1  = in[ 1 * in_s];
860     const int in2  = in[ 2 * in_s], in3  = in[ 3 * in_s];
861     const int in4  = in[ 4 * in_s], in5  = in[ 5 * in_s];
862     const int in6  = in[ 6 * in_s], in7  = in[ 7 * in_s];
863     const int in8  = in[ 8 * in_s], in9  = in[ 9 * in_s];
864     const int in10 = in[10 * in_s], in11 = in[11 * in_s];
865     const int in12 = in[12 * in_s], in13 = in[13 * in_s];
866     const int in14 = in[14 * in_s], in15 = in[15 * in_s];
867 
868     int t0  = ((in15 * (4091 - 4096) + in0  *   201         + 2048) >> 12) + in15;
869     int t1  = ((in15 *   201         - in0  * (4091 - 4096) + 2048) >> 12) - in0;
870     int t2  = ((in13 * (3973 - 4096) + in2  *   995         + 2048) >> 12) + in13;
871     int t3  = ((in13 *   995         - in2  * (3973 - 4096) + 2048) >> 12) - in2;
872     int t4  = ((in11 * (3703 - 4096) + in4  *  1751         + 2048) >> 12) + in11;
873     int t5  = ((in11 *  1751         - in4  * (3703 - 4096) + 2048) >> 12) - in4;
874     int t6  =  (in9  *  1645         + in6  *  1220         + 1024) >> 11;
875     int t7  =  (in9  *  1220         - in6  *  1645         + 1024) >> 11;
876     int t8  = ((in7  *  2751         + in8  * (3035 - 4096) + 2048) >> 12) + in8;
877     int t9  = ((in7  * (3035 - 4096) - in8  *  2751         + 2048) >> 12) + in7;
878     int t10 = ((in5  *  2106         + in10 * (3513 - 4096) + 2048) >> 12) + in10;
879     int t11 = ((in5  * (3513 - 4096) - in10 *  2106         + 2048) >> 12) + in5;
880     int t12 = ((in3  *  1380         + in12 * (3857 - 4096) + 2048) >> 12) + in12;
881     int t13 = ((in3  * (3857 - 4096) - in12 *  1380         + 2048) >> 12) + in3;
882     int t14 = ((in1  *   601         + in14 * (4052 - 4096) + 2048) >> 12) + in14;
883     int t15 = ((in1  * (4052 - 4096) - in14 *   601         + 2048) >> 12) + in1;
884 
885     int t0a  = CLIP(t0 + t8 );
886     int t1a  = CLIP(t1 + t9 );
887     int t2a  = CLIP(t2 + t10);
888     int t3a  = CLIP(t3 + t11);
889     int t4a  = CLIP(t4 + t12);
890     int t5a  = CLIP(t5 + t13);
891     int t6a  = CLIP(t6 + t14);
892     int t7a  = CLIP(t7 + t15);
893     int t8a  = CLIP(t0 - t8 );
894     int t9a  = CLIP(t1 - t9 );
895     int t10a = CLIP(t2 - t10);
896     int t11a = CLIP(t3 - t11);
897     int t12a = CLIP(t4 - t12);
898     int t13a = CLIP(t5 - t13);
899     int t14a = CLIP(t6 - t14);
900     int t15a = CLIP(t7 - t15);
901 
902     t8   = ((t8a  * (4017 - 4096) + t9a  *   799         + 2048) >> 12) + t8a;
903     t9   = ((t8a  *   799         - t9a  * (4017 - 4096) + 2048) >> 12) - t9a;
904     t10  = ((t10a *  2276         + t11a * (3406 - 4096) + 2048) >> 12) + t11a;
905     t11  = ((t10a * (3406 - 4096) - t11a *  2276         + 2048) >> 12) + t10a;
906     t12  = ((t13a * (4017 - 4096) - t12a *   799         + 2048) >> 12) + t13a;
907     t13  = ((t13a *   799         + t12a * (4017 - 4096) + 2048) >> 12) + t12a;
908     t14  = ((t15a *  2276         - t14a * (3406 - 4096) + 2048) >> 12) - t14a;
909     t15  = ((t15a * (3406 - 4096) + t14a *  2276         + 2048) >> 12) + t15a;
910 
911     t0   = CLIP(t0a + t4a);
912     t1   = CLIP(t1a + t5a);
913     t2   = CLIP(t2a + t6a);
914     t3   = CLIP(t3a + t7a);
915     t4   = CLIP(t0a - t4a);
916     t5   = CLIP(t1a - t5a);
917     t6   = CLIP(t2a - t6a);
918     t7   = CLIP(t3a - t7a);
919     t8a  = CLIP(t8  + t12);
920     t9a  = CLIP(t9  + t13);
921     t10a = CLIP(t10 + t14);
922     t11a = CLIP(t11 + t15);
923     t12a = CLIP(t8  - t12);
924     t13a = CLIP(t9  - t13);
925     t14a = CLIP(t10 - t14);
926     t15a = CLIP(t11 - t15);
927 
928     t4a  = ((t4   * (3784 - 4096) + t5   *  1567         + 2048) >> 12) + t4;
929     t5a  = ((t4   *  1567         - t5   * (3784 - 4096) + 2048) >> 12) - t5;
930     t6a  = ((t7   * (3784 - 4096) - t6   *  1567         + 2048) >> 12) + t7;
931     t7a  = ((t7   *  1567         + t6   * (3784 - 4096) + 2048) >> 12) + t6;
932     t12  = ((t12a * (3784 - 4096) + t13a *  1567         + 2048) >> 12) + t12a;
933     t13  = ((t12a *  1567         - t13a * (3784 - 4096) + 2048) >> 12) - t13a;
934     t14  = ((t15a * (3784 - 4096) - t14a *  1567         + 2048) >> 12) + t15a;
935     t15  = ((t15a *  1567         + t14a * (3784 - 4096) + 2048) >> 12) + t14a;
936 
937     out[ 0 * out_s] =  CLIP(t0  + t2  );
938     out[15 * out_s] = -CLIP(t1  + t3  );
939     t2a             =  CLIP(t0  - t2  );
940     t3a             =  CLIP(t1  - t3  );
941     out[ 3 * out_s] = -CLIP(t4a + t6a );
942     out[12 * out_s] =  CLIP(t5a + t7a );
943     t6              =  CLIP(t4a - t6a );
944     t7              =  CLIP(t5a - t7a );
945     out[ 1 * out_s] = -CLIP(t8a + t10a);
946     out[14 * out_s] =  CLIP(t9a + t11a);
947     t10             =  CLIP(t8a - t10a);
948     t11             =  CLIP(t9a - t11a);
949     out[ 2 * out_s] =  CLIP(t12 + t14 );
950     out[13 * out_s] = -CLIP(t13 + t15 );
951     t14a            =  CLIP(t12 - t14 );
952     t15a            =  CLIP(t13 - t15 );
953 
954     out[ 7 * out_s] = -(((t2a  + t3a)  * 181 + 128) >> 8);
955     out[ 8 * out_s] =   ((t2a  - t3a)  * 181 + 128) >> 8;
956     out[ 4 * out_s] =   ((t6   + t7)   * 181 + 128) >> 8;
957     out[11 * out_s] = -(((t6   - t7)   * 181 + 128) >> 8);
958     out[ 6 * out_s] =   ((t10  + t11)  * 181 + 128) >> 8;
959     out[ 9 * out_s] = -(((t10  - t11)  * 181 + 128) >> 8);
960     out[ 5 * out_s] = -(((t14a + t15a) * 181 + 128) >> 8);
961     out[10 * out_s] =   ((t14a - t15a) * 181 + 128) >> 8;
962 }
963 
964 #define inv_adst_1d(sz) \
965 static void inv_adst##sz##_1d_c(int32_t *const c, const ptrdiff_t stride, \
966                                 const int min, const int max) \
967 { \
968     inv_adst##sz##_1d_internal_c(c, stride, min, max, c, stride); \
969 } \
970 static void inv_flipadst##sz##_1d_c(int32_t *const c, const ptrdiff_t stride, \
971                                           const int min, const int max) \
972 { \
973     inv_adst##sz##_1d_internal_c(c, stride, min, max, \
974                                  &c[(sz - 1) * stride], -stride); \
975 }
976 
977 inv_adst_1d( 4)
978 inv_adst_1d( 8)
979 inv_adst_1d(16)
980 
981 #undef inv_adst_1d
982 
inv_identity4_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)983 static void inv_identity4_1d_c(int32_t *const c, const ptrdiff_t stride,
984                                const int min, const int max)
985 {
986     assert(stride > 0);
987     for (int i = 0; i < 4; i++) {
988         const int in = c[stride * i];
989         c[stride * i] = in + ((in * 1697 + 2048) >> 12);
990     }
991 }
992 
inv_identity8_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)993 static void inv_identity8_1d_c(int32_t *const c, const ptrdiff_t stride,
994                                const int min, const int max)
995 {
996     assert(stride > 0);
997     for (int i = 0; i < 8; i++)
998         c[stride * i] *= 2;
999 }
1000 
inv_identity16_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)1001 static void inv_identity16_1d_c(int32_t *const c, const ptrdiff_t stride,
1002                                 const int min, const int max)
1003 {
1004     assert(stride > 0);
1005     for (int i = 0; i < 16; i++) {
1006         const int in = c[stride * i];
1007         c[stride * i] = 2 * in + ((in * 1697 + 1024) >> 11);
1008     }
1009 }
1010 
inv_identity32_1d_c(int32_t * const c,const ptrdiff_t stride,const int min,const int max)1011 static void inv_identity32_1d_c(int32_t *const c, const ptrdiff_t stride,
1012                                 const int min, const int max)
1013 {
1014     assert(stride > 0);
1015     for (int i = 0; i < 32; i++)
1016         c[stride * i] *= 4;
1017 }
1018 
1019 const itx_1d_fn dav1d_tx1d_fns[N_TX_SIZES][N_TX_1D_TYPES] = {
1020     [TX_4X4] = {
1021         [DCT] = inv_dct4_1d_c,
1022         [ADST] = inv_adst4_1d_c,
1023         [FLIPADST] = inv_flipadst4_1d_c,
1024         [IDENTITY] = inv_identity4_1d_c,
1025     }, [TX_8X8] = {
1026         [DCT] = inv_dct8_1d_c,
1027         [ADST] = inv_adst8_1d_c,
1028         [FLIPADST] = inv_flipadst8_1d_c,
1029         [IDENTITY] = inv_identity8_1d_c,
1030     }, [TX_16X16] = {
1031         [DCT] = inv_dct16_1d_c,
1032         [ADST] = inv_adst16_1d_c,
1033         [FLIPADST] = inv_flipadst16_1d_c,
1034         [IDENTITY] = inv_identity16_1d_c,
1035     }, [TX_32X32] = {
1036         [DCT] = inv_dct32_1d_c,
1037         [IDENTITY] = inv_identity32_1d_c,
1038     }, [TX_64X64] = {
1039         [DCT] = inv_dct64_1d_c,
1040     },
1041 };
1042 
1043 const uint8_t /* enum Tx1dType */ dav1d_tx1d_types[N_TX_TYPES][2] = {
1044     [DCT_DCT]           = { DCT, DCT },
1045     [ADST_DCT]          = { ADST, DCT },
1046     [DCT_ADST]          = { DCT, ADST },
1047     [ADST_ADST]         = { ADST, ADST },
1048     [FLIPADST_DCT]      = { FLIPADST, DCT },
1049     [DCT_FLIPADST]      = { DCT, FLIPADST },
1050     [FLIPADST_FLIPADST] = { FLIPADST, FLIPADST },
1051     [ADST_FLIPADST]     = { ADST, FLIPADST },
1052     [FLIPADST_ADST]     = { FLIPADST, ADST },
1053     [IDTX]              = { IDENTITY, IDENTITY },
1054     [V_DCT]             = { DCT, IDENTITY },
1055     [H_DCT]             = { IDENTITY, DCT },
1056     [V_ADST]            = { ADST, IDENTITY },
1057     [H_ADST]            = { IDENTITY, ADST },
1058     [V_FLIPADST]        = { FLIPADST, IDENTITY },
1059     [H_FLIPADST]        = { IDENTITY, FLIPADST },
1060 };
1061 
1062 #if !(HAVE_ASM && TRIM_DSP_FUNCTIONS && ( \
1063   ARCH_AARCH64 || \
1064   (ARCH_ARM && (defined(__ARM_NEON) || defined(__APPLE__) || defined(_WIN32))) \
1065 ))
dav1d_inv_wht4_1d_c(int32_t * const c,const ptrdiff_t stride)1066 void dav1d_inv_wht4_1d_c(int32_t *const c, const ptrdiff_t stride) {
1067     assert(stride > 0);
1068     const int in0 = c[0 * stride], in1 = c[1 * stride];
1069     const int in2 = c[2 * stride], in3 = c[3 * stride];
1070 
1071     const int t0 = in0 + in1;
1072     const int t2 = in2 - in3;
1073     const int t4 = (t0 - t2) >> 1;
1074     const int t3 = t4 - in3;
1075     const int t1 = t4 - in1;
1076 
1077     c[0 * stride] = t0 - t3;
1078     c[1 * stride] = t3;
1079     c[2 * stride] = t1;
1080     c[3 * stride] = t2 + t1;
1081 }
1082 #endif
1083