1 /*
2  *  Copyright (c) 2010 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 <assert.h>
12 #include <math.h>
13 
14 #include "./vp9_rtcd.h"
15 #include "./vpx_config.h"
16 #include "./vpx_dsp_rtcd.h"
17 
18 #include "vp9/common/vp9_blockd.h"
19 #include "vp9/common/vp9_idct.h"
20 #include "vpx_dsp/fwd_txfm.h"
21 #include "vpx_ports/mem.h"
22 
fdct4(const tran_low_t * input,tran_low_t * output)23 static void fdct4(const tran_low_t *input, tran_low_t *output) {
24   tran_high_t step[4];
25   tran_high_t temp1, temp2;
26 
27   step[0] = input[0] + input[3];
28   step[1] = input[1] + input[2];
29   step[2] = input[1] - input[2];
30   step[3] = input[0] - input[3];
31 
32   temp1 = (step[0] + step[1]) * cospi_16_64;
33   temp2 = (step[0] - step[1]) * cospi_16_64;
34   output[0] = (tran_low_t)fdct_round_shift(temp1);
35   output[2] = (tran_low_t)fdct_round_shift(temp2);
36   temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
37   temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
38   output[1] = (tran_low_t)fdct_round_shift(temp1);
39   output[3] = (tran_low_t)fdct_round_shift(temp2);
40 }
41 
fdct8(const tran_low_t * input,tran_low_t * output)42 static void fdct8(const tran_low_t *input, tran_low_t *output) {
43   tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;  // canbe16
44   tran_high_t t0, t1, t2, t3;                  // needs32
45   tran_high_t x0, x1, x2, x3;                  // canbe16
46 
47   // stage 1
48   s0 = input[0] + input[7];
49   s1 = input[1] + input[6];
50   s2 = input[2] + input[5];
51   s3 = input[3] + input[4];
52   s4 = input[3] - input[4];
53   s5 = input[2] - input[5];
54   s6 = input[1] - input[6];
55   s7 = input[0] - input[7];
56 
57   // fdct4(step, step);
58   x0 = s0 + s3;
59   x1 = s1 + s2;
60   x2 = s1 - s2;
61   x3 = s0 - s3;
62   t0 = (x0 + x1) * cospi_16_64;
63   t1 = (x0 - x1) * cospi_16_64;
64   t2 =  x2 * cospi_24_64 + x3 *  cospi_8_64;
65   t3 = -x2 * cospi_8_64  + x3 * cospi_24_64;
66   output[0] = (tran_low_t)fdct_round_shift(t0);
67   output[2] = (tran_low_t)fdct_round_shift(t2);
68   output[4] = (tran_low_t)fdct_round_shift(t1);
69   output[6] = (tran_low_t)fdct_round_shift(t3);
70 
71   // Stage 2
72   t0 = (s6 - s5) * cospi_16_64;
73   t1 = (s6 + s5) * cospi_16_64;
74   t2 = (tran_low_t)fdct_round_shift(t0);
75   t3 = (tran_low_t)fdct_round_shift(t1);
76 
77   // Stage 3
78   x0 = s4 + t2;
79   x1 = s4 - t2;
80   x2 = s7 - t3;
81   x3 = s7 + t3;
82 
83   // Stage 4
84   t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
85   t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
86   t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
87   t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
88   output[1] = (tran_low_t)fdct_round_shift(t0);
89   output[3] = (tran_low_t)fdct_round_shift(t2);
90   output[5] = (tran_low_t)fdct_round_shift(t1);
91   output[7] = (tran_low_t)fdct_round_shift(t3);
92 }
93 
fdct16(const tran_low_t in[16],tran_low_t out[16])94 static void fdct16(const tran_low_t in[16], tran_low_t out[16]) {
95   tran_high_t step1[8];      // canbe16
96   tran_high_t step2[8];      // canbe16
97   tran_high_t step3[8];      // canbe16
98   tran_high_t input[8];      // canbe16
99   tran_high_t temp1, temp2;  // needs32
100 
101   // step 1
102   input[0] = in[0] + in[15];
103   input[1] = in[1] + in[14];
104   input[2] = in[2] + in[13];
105   input[3] = in[3] + in[12];
106   input[4] = in[4] + in[11];
107   input[5] = in[5] + in[10];
108   input[6] = in[6] + in[ 9];
109   input[7] = in[7] + in[ 8];
110 
111   step1[0] = in[7] - in[ 8];
112   step1[1] = in[6] - in[ 9];
113   step1[2] = in[5] - in[10];
114   step1[3] = in[4] - in[11];
115   step1[4] = in[3] - in[12];
116   step1[5] = in[2] - in[13];
117   step1[6] = in[1] - in[14];
118   step1[7] = in[0] - in[15];
119 
120   // fdct8(step, step);
121   {
122     tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;  // canbe16
123     tran_high_t t0, t1, t2, t3;                  // needs32
124     tran_high_t x0, x1, x2, x3;                  // canbe16
125 
126     // stage 1
127     s0 = input[0] + input[7];
128     s1 = input[1] + input[6];
129     s2 = input[2] + input[5];
130     s3 = input[3] + input[4];
131     s4 = input[3] - input[4];
132     s5 = input[2] - input[5];
133     s6 = input[1] - input[6];
134     s7 = input[0] - input[7];
135 
136     // fdct4(step, step);
137     x0 = s0 + s3;
138     x1 = s1 + s2;
139     x2 = s1 - s2;
140     x3 = s0 - s3;
141     t0 = (x0 + x1) * cospi_16_64;
142     t1 = (x0 - x1) * cospi_16_64;
143     t2 = x3 * cospi_8_64  + x2 * cospi_24_64;
144     t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
145     out[0] = (tran_low_t)fdct_round_shift(t0);
146     out[4] = (tran_low_t)fdct_round_shift(t2);
147     out[8] = (tran_low_t)fdct_round_shift(t1);
148     out[12] = (tran_low_t)fdct_round_shift(t3);
149 
150     // Stage 2
151     t0 = (s6 - s5) * cospi_16_64;
152     t1 = (s6 + s5) * cospi_16_64;
153     t2 = fdct_round_shift(t0);
154     t3 = fdct_round_shift(t1);
155 
156     // Stage 3
157     x0 = s4 + t2;
158     x1 = s4 - t2;
159     x2 = s7 - t3;
160     x3 = s7 + t3;
161 
162     // Stage 4
163     t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
164     t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
165     t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
166     t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
167     out[2] = (tran_low_t)fdct_round_shift(t0);
168     out[6] = (tran_low_t)fdct_round_shift(t2);
169     out[10] = (tran_low_t)fdct_round_shift(t1);
170     out[14] = (tran_low_t)fdct_round_shift(t3);
171   }
172 
173   // step 2
174   temp1 = (step1[5] - step1[2]) * cospi_16_64;
175   temp2 = (step1[4] - step1[3]) * cospi_16_64;
176   step2[2] = fdct_round_shift(temp1);
177   step2[3] = fdct_round_shift(temp2);
178   temp1 = (step1[4] + step1[3]) * cospi_16_64;
179   temp2 = (step1[5] + step1[2]) * cospi_16_64;
180   step2[4] = fdct_round_shift(temp1);
181   step2[5] = fdct_round_shift(temp2);
182 
183   // step 3
184   step3[0] = step1[0] + step2[3];
185   step3[1] = step1[1] + step2[2];
186   step3[2] = step1[1] - step2[2];
187   step3[3] = step1[0] - step2[3];
188   step3[4] = step1[7] - step2[4];
189   step3[5] = step1[6] - step2[5];
190   step3[6] = step1[6] + step2[5];
191   step3[7] = step1[7] + step2[4];
192 
193   // step 4
194   temp1 = step3[1] *  -cospi_8_64 + step3[6] * cospi_24_64;
195   temp2 = step3[2] * cospi_24_64 + step3[5] *  cospi_8_64;
196   step2[1] = fdct_round_shift(temp1);
197   step2[2] = fdct_round_shift(temp2);
198   temp1 = step3[2] * cospi_8_64 - step3[5] * cospi_24_64;
199   temp2 = step3[1] * cospi_24_64 + step3[6] *  cospi_8_64;
200   step2[5] = fdct_round_shift(temp1);
201   step2[6] = fdct_round_shift(temp2);
202 
203   // step 5
204   step1[0] = step3[0] + step2[1];
205   step1[1] = step3[0] - step2[1];
206   step1[2] = step3[3] + step2[2];
207   step1[3] = step3[3] - step2[2];
208   step1[4] = step3[4] - step2[5];
209   step1[5] = step3[4] + step2[5];
210   step1[6] = step3[7] - step2[6];
211   step1[7] = step3[7] + step2[6];
212 
213   // step 6
214   temp1 = step1[0] * cospi_30_64 + step1[7] *  cospi_2_64;
215   temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
216   out[1] = (tran_low_t)fdct_round_shift(temp1);
217   out[9] = (tran_low_t)fdct_round_shift(temp2);
218 
219   temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
220   temp2 = step1[3] *  cospi_6_64 + step1[4] * cospi_26_64;
221   out[5] = (tran_low_t)fdct_round_shift(temp1);
222   out[13] = (tran_low_t)fdct_round_shift(temp2);
223 
224   temp1 = step1[3] * -cospi_26_64 + step1[4] *  cospi_6_64;
225   temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
226   out[3] = (tran_low_t)fdct_round_shift(temp1);
227   out[11] = (tran_low_t)fdct_round_shift(temp2);
228 
229   temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
230   temp2 = step1[0] *  -cospi_2_64 + step1[7] * cospi_30_64;
231   out[7] = (tran_low_t)fdct_round_shift(temp1);
232   out[15] = (tran_low_t)fdct_round_shift(temp2);
233 }
234 
fadst4(const tran_low_t * input,tran_low_t * output)235 static void fadst4(const tran_low_t *input, tran_low_t *output) {
236   tran_high_t x0, x1, x2, x3;
237   tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
238 
239   x0 = input[0];
240   x1 = input[1];
241   x2 = input[2];
242   x3 = input[3];
243 
244   if (!(x0 | x1 | x2 | x3)) {
245     output[0] = output[1] = output[2] = output[3] = 0;
246     return;
247   }
248 
249   s0 = sinpi_1_9 * x0;
250   s1 = sinpi_4_9 * x0;
251   s2 = sinpi_2_9 * x1;
252   s3 = sinpi_1_9 * x1;
253   s4 = sinpi_3_9 * x2;
254   s5 = sinpi_4_9 * x3;
255   s6 = sinpi_2_9 * x3;
256   s7 = x0 + x1 - x3;
257 
258   x0 = s0 + s2 + s5;
259   x1 = sinpi_3_9 * s7;
260   x2 = s1 - s3 + s6;
261   x3 = s4;
262 
263   s0 = x0 + x3;
264   s1 = x1;
265   s2 = x2 - x3;
266   s3 = x2 - x0 + x3;
267 
268   // 1-D transform scaling factor is sqrt(2).
269   output[0] = (tran_low_t)fdct_round_shift(s0);
270   output[1] = (tran_low_t)fdct_round_shift(s1);
271   output[2] = (tran_low_t)fdct_round_shift(s2);
272   output[3] = (tran_low_t)fdct_round_shift(s3);
273 }
274 
fadst8(const tran_low_t * input,tran_low_t * output)275 static void fadst8(const tran_low_t *input, tran_low_t *output) {
276   tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
277 
278   tran_high_t x0 = input[7];
279   tran_high_t x1 = input[0];
280   tran_high_t x2 = input[5];
281   tran_high_t x3 = input[2];
282   tran_high_t x4 = input[3];
283   tran_high_t x5 = input[4];
284   tran_high_t x6 = input[1];
285   tran_high_t x7 = input[6];
286 
287   // stage 1
288   s0 = cospi_2_64  * x0 + cospi_30_64 * x1;
289   s1 = cospi_30_64 * x0 - cospi_2_64  * x1;
290   s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
291   s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
292   s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
293   s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
294   s6 = cospi_26_64 * x6 + cospi_6_64  * x7;
295   s7 = cospi_6_64  * x6 - cospi_26_64 * x7;
296 
297   x0 = fdct_round_shift(s0 + s4);
298   x1 = fdct_round_shift(s1 + s5);
299   x2 = fdct_round_shift(s2 + s6);
300   x3 = fdct_round_shift(s3 + s7);
301   x4 = fdct_round_shift(s0 - s4);
302   x5 = fdct_round_shift(s1 - s5);
303   x6 = fdct_round_shift(s2 - s6);
304   x7 = fdct_round_shift(s3 - s7);
305 
306   // stage 2
307   s0 = x0;
308   s1 = x1;
309   s2 = x2;
310   s3 = x3;
311   s4 = cospi_8_64  * x4 + cospi_24_64 * x5;
312   s5 = cospi_24_64 * x4 - cospi_8_64  * x5;
313   s6 = - cospi_24_64 * x6 + cospi_8_64  * x7;
314   s7 =   cospi_8_64  * x6 + cospi_24_64 * x7;
315 
316   x0 = s0 + s2;
317   x1 = s1 + s3;
318   x2 = s0 - s2;
319   x3 = s1 - s3;
320   x4 = fdct_round_shift(s4 + s6);
321   x5 = fdct_round_shift(s5 + s7);
322   x6 = fdct_round_shift(s4 - s6);
323   x7 = fdct_round_shift(s5 - s7);
324 
325   // stage 3
326   s2 = cospi_16_64 * (x2 + x3);
327   s3 = cospi_16_64 * (x2 - x3);
328   s6 = cospi_16_64 * (x6 + x7);
329   s7 = cospi_16_64 * (x6 - x7);
330 
331   x2 = fdct_round_shift(s2);
332   x3 = fdct_round_shift(s3);
333   x6 = fdct_round_shift(s6);
334   x7 = fdct_round_shift(s7);
335 
336   output[0] = (tran_low_t)x0;
337   output[1] = (tran_low_t)-x4;
338   output[2] = (tran_low_t)x6;
339   output[3] = (tran_low_t)-x2;
340   output[4] = (tran_low_t)x3;
341   output[5] = (tran_low_t)-x7;
342   output[6] = (tran_low_t)x5;
343   output[7] = (tran_low_t)-x1;
344 }
345 
fadst16(const tran_low_t * input,tran_low_t * output)346 static void fadst16(const tran_low_t *input, tran_low_t *output) {
347   tran_high_t s0, s1, s2, s3, s4, s5, s6, s7, s8;
348   tran_high_t s9, s10, s11, s12, s13, s14, s15;
349 
350   tran_high_t x0 = input[15];
351   tran_high_t x1 = input[0];
352   tran_high_t x2 = input[13];
353   tran_high_t x3 = input[2];
354   tran_high_t x4 = input[11];
355   tran_high_t x5 = input[4];
356   tran_high_t x6 = input[9];
357   tran_high_t x7 = input[6];
358   tran_high_t x8 = input[7];
359   tran_high_t x9 = input[8];
360   tran_high_t x10 = input[5];
361   tran_high_t x11 = input[10];
362   tran_high_t x12 = input[3];
363   tran_high_t x13 = input[12];
364   tran_high_t x14 = input[1];
365   tran_high_t x15 = input[14];
366 
367   // stage 1
368   s0 = x0 * cospi_1_64  + x1 * cospi_31_64;
369   s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
370   s2 = x2 * cospi_5_64  + x3 * cospi_27_64;
371   s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
372   s4 = x4 * cospi_9_64  + x5 * cospi_23_64;
373   s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
374   s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
375   s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
376   s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
377   s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
378   s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
379   s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
380   s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
381   s13 = x12 * cospi_7_64  - x13 * cospi_25_64;
382   s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
383   s15 = x14 * cospi_3_64  - x15 * cospi_29_64;
384 
385   x0 = fdct_round_shift(s0 + s8);
386   x1 = fdct_round_shift(s1 + s9);
387   x2 = fdct_round_shift(s2 + s10);
388   x3 = fdct_round_shift(s3 + s11);
389   x4 = fdct_round_shift(s4 + s12);
390   x5 = fdct_round_shift(s5 + s13);
391   x6 = fdct_round_shift(s6 + s14);
392   x7 = fdct_round_shift(s7 + s15);
393   x8  = fdct_round_shift(s0 - s8);
394   x9  = fdct_round_shift(s1 - s9);
395   x10 = fdct_round_shift(s2 - s10);
396   x11 = fdct_round_shift(s3 - s11);
397   x12 = fdct_round_shift(s4 - s12);
398   x13 = fdct_round_shift(s5 - s13);
399   x14 = fdct_round_shift(s6 - s14);
400   x15 = fdct_round_shift(s7 - s15);
401 
402   // stage 2
403   s0 = x0;
404   s1 = x1;
405   s2 = x2;
406   s3 = x3;
407   s4 = x4;
408   s5 = x5;
409   s6 = x6;
410   s7 = x7;
411   s8 =    x8 * cospi_4_64   + x9 * cospi_28_64;
412   s9 =    x8 * cospi_28_64  - x9 * cospi_4_64;
413   s10 =   x10 * cospi_20_64 + x11 * cospi_12_64;
414   s11 =   x10 * cospi_12_64 - x11 * cospi_20_64;
415   s12 = - x12 * cospi_28_64 + x13 * cospi_4_64;
416   s13 =   x12 * cospi_4_64  + x13 * cospi_28_64;
417   s14 = - x14 * cospi_12_64 + x15 * cospi_20_64;
418   s15 =   x14 * cospi_20_64 + x15 * cospi_12_64;
419 
420   x0 = s0 + s4;
421   x1 = s1 + s5;
422   x2 = s2 + s6;
423   x3 = s3 + s7;
424   x4 = s0 - s4;
425   x5 = s1 - s5;
426   x6 = s2 - s6;
427   x7 = s3 - s7;
428   x8 = fdct_round_shift(s8 + s12);
429   x9 = fdct_round_shift(s9 + s13);
430   x10 = fdct_round_shift(s10 + s14);
431   x11 = fdct_round_shift(s11 + s15);
432   x12 = fdct_round_shift(s8 - s12);
433   x13 = fdct_round_shift(s9 - s13);
434   x14 = fdct_round_shift(s10 - s14);
435   x15 = fdct_round_shift(s11 - s15);
436 
437   // stage 3
438   s0 = x0;
439   s1 = x1;
440   s2 = x2;
441   s3 = x3;
442   s4 = x4 * cospi_8_64  + x5 * cospi_24_64;
443   s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
444   s6 = - x6 * cospi_24_64 + x7 * cospi_8_64;
445   s7 =   x6 * cospi_8_64  + x7 * cospi_24_64;
446   s8 = x8;
447   s9 = x9;
448   s10 = x10;
449   s11 = x11;
450   s12 = x12 * cospi_8_64  + x13 * cospi_24_64;
451   s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
452   s14 = - x14 * cospi_24_64 + x15 * cospi_8_64;
453   s15 =   x14 * cospi_8_64  + x15 * cospi_24_64;
454 
455   x0 = s0 + s2;
456   x1 = s1 + s3;
457   x2 = s0 - s2;
458   x3 = s1 - s3;
459   x4 = fdct_round_shift(s4 + s6);
460   x5 = fdct_round_shift(s5 + s7);
461   x6 = fdct_round_shift(s4 - s6);
462   x7 = fdct_round_shift(s5 - s7);
463   x8 = s8 + s10;
464   x9 = s9 + s11;
465   x10 = s8 - s10;
466   x11 = s9 - s11;
467   x12 = fdct_round_shift(s12 + s14);
468   x13 = fdct_round_shift(s13 + s15);
469   x14 = fdct_round_shift(s12 - s14);
470   x15 = fdct_round_shift(s13 - s15);
471 
472   // stage 4
473   s2 = (- cospi_16_64) * (x2 + x3);
474   s3 = cospi_16_64 * (x2 - x3);
475   s6 = cospi_16_64 * (x6 + x7);
476   s7 = cospi_16_64 * (- x6 + x7);
477   s10 = cospi_16_64 * (x10 + x11);
478   s11 = cospi_16_64 * (- x10 + x11);
479   s14 = (- cospi_16_64) * (x14 + x15);
480   s15 = cospi_16_64 * (x14 - x15);
481 
482   x2 = fdct_round_shift(s2);
483   x3 = fdct_round_shift(s3);
484   x6 = fdct_round_shift(s6);
485   x7 = fdct_round_shift(s7);
486   x10 = fdct_round_shift(s10);
487   x11 = fdct_round_shift(s11);
488   x14 = fdct_round_shift(s14);
489   x15 = fdct_round_shift(s15);
490 
491   output[0] = (tran_low_t)x0;
492   output[1] = (tran_low_t)-x8;
493   output[2] = (tran_low_t)x12;
494   output[3] = (tran_low_t)-x4;
495   output[4] = (tran_low_t)x6;
496   output[5] = (tran_low_t)x14;
497   output[6] = (tran_low_t)x10;
498   output[7] = (tran_low_t)x2;
499   output[8] = (tran_low_t)x3;
500   output[9] = (tran_low_t)x11;
501   output[10] = (tran_low_t)x15;
502   output[11] = (tran_low_t)x7;
503   output[12] = (tran_low_t)x5;
504   output[13] = (tran_low_t)-x13;
505   output[14] = (tran_low_t)x9;
506   output[15] = (tran_low_t)-x1;
507 }
508 
509 static const transform_2d FHT_4[] = {
510   { fdct4,  fdct4  },  // DCT_DCT  = 0
511   { fadst4, fdct4  },  // ADST_DCT = 1
512   { fdct4,  fadst4 },  // DCT_ADST = 2
513   { fadst4, fadst4 }   // ADST_ADST = 3
514 };
515 
516 static const transform_2d FHT_8[] = {
517   { fdct8,  fdct8  },  // DCT_DCT  = 0
518   { fadst8, fdct8  },  // ADST_DCT = 1
519   { fdct8,  fadst8 },  // DCT_ADST = 2
520   { fadst8, fadst8 }   // ADST_ADST = 3
521 };
522 
523 static const transform_2d FHT_16[] = {
524   { fdct16,  fdct16  },  // DCT_DCT  = 0
525   { fadst16, fdct16  },  // ADST_DCT = 1
526   { fdct16,  fadst16 },  // DCT_ADST = 2
527   { fadst16, fadst16 }   // ADST_ADST = 3
528 };
529 
vp9_fht4x4_c(const int16_t * input,tran_low_t * output,int stride,int tx_type)530 void vp9_fht4x4_c(const int16_t *input, tran_low_t *output,
531                   int stride, int tx_type) {
532   if (tx_type == DCT_DCT) {
533     vpx_fdct4x4_c(input, output, stride);
534   } else {
535     tran_low_t out[4 * 4];
536     int i, j;
537     tran_low_t temp_in[4], temp_out[4];
538     const transform_2d ht = FHT_4[tx_type];
539 
540     // Columns
541     for (i = 0; i < 4; ++i) {
542       for (j = 0; j < 4; ++j)
543         temp_in[j] = input[j * stride + i] * 16;
544       if (i == 0 && temp_in[0])
545         temp_in[0] += 1;
546       ht.cols(temp_in, temp_out);
547       for (j = 0; j < 4; ++j)
548         out[j * 4 + i] = temp_out[j];
549     }
550 
551     // Rows
552     for (i = 0; i < 4; ++i) {
553       for (j = 0; j < 4; ++j)
554         temp_in[j] = out[j + i * 4];
555       ht.rows(temp_in, temp_out);
556       for (j = 0; j < 4; ++j)
557         output[j + i * 4] = (temp_out[j] + 1) >> 2;
558     }
559   }
560 }
561 
vp9_fdct8x8_quant_c(const int16_t * input,int stride,tran_low_t * coeff_ptr,intptr_t n_coeffs,int skip_block,const int16_t * zbin_ptr,const int16_t * round_ptr,const int16_t * quant_ptr,const int16_t * quant_shift_ptr,tran_low_t * qcoeff_ptr,tran_low_t * dqcoeff_ptr,const int16_t * dequant_ptr,uint16_t * eob_ptr,const int16_t * scan,const int16_t * iscan)562 void vp9_fdct8x8_quant_c(const int16_t *input, int stride,
563                          tran_low_t *coeff_ptr, intptr_t n_coeffs,
564                          int skip_block,
565                          const int16_t *zbin_ptr, const int16_t *round_ptr,
566                          const int16_t *quant_ptr,
567                          const int16_t *quant_shift_ptr,
568                          tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
569                          const int16_t *dequant_ptr,
570                          uint16_t *eob_ptr,
571                          const int16_t *scan, const int16_t *iscan) {
572   int eob = -1;
573 
574   int i, j;
575   tran_low_t intermediate[64];
576 
577   // Transform columns
578   {
579     tran_low_t *output = intermediate;
580     tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;  // canbe16
581     tran_high_t t0, t1, t2, t3;                  // needs32
582     tran_high_t x0, x1, x2, x3;                  // canbe16
583 
584     int i;
585     for (i = 0; i < 8; i++) {
586       // stage 1
587       s0 = (input[0 * stride] + input[7 * stride]) * 4;
588       s1 = (input[1 * stride] + input[6 * stride]) * 4;
589       s2 = (input[2 * stride] + input[5 * stride]) * 4;
590       s3 = (input[3 * stride] + input[4 * stride]) * 4;
591       s4 = (input[3 * stride] - input[4 * stride]) * 4;
592       s5 = (input[2 * stride] - input[5 * stride]) * 4;
593       s6 = (input[1 * stride] - input[6 * stride]) * 4;
594       s7 = (input[0 * stride] - input[7 * stride]) * 4;
595 
596       // fdct4(step, step);
597       x0 = s0 + s3;
598       x1 = s1 + s2;
599       x2 = s1 - s2;
600       x3 = s0 - s3;
601       t0 = (x0 + x1) * cospi_16_64;
602       t1 = (x0 - x1) * cospi_16_64;
603       t2 =  x2 * cospi_24_64 + x3 *  cospi_8_64;
604       t3 = -x2 * cospi_8_64  + x3 * cospi_24_64;
605       output[0 * 8] = (tran_low_t)fdct_round_shift(t0);
606       output[2 * 8] = (tran_low_t)fdct_round_shift(t2);
607       output[4 * 8] = (tran_low_t)fdct_round_shift(t1);
608       output[6 * 8] = (tran_low_t)fdct_round_shift(t3);
609 
610       // Stage 2
611       t0 = (s6 - s5) * cospi_16_64;
612       t1 = (s6 + s5) * cospi_16_64;
613       t2 = fdct_round_shift(t0);
614       t3 = fdct_round_shift(t1);
615 
616       // Stage 3
617       x0 = s4 + t2;
618       x1 = s4 - t2;
619       x2 = s7 - t3;
620       x3 = s7 + t3;
621 
622       // Stage 4
623       t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
624       t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
625       t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
626       t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
627       output[1 * 8] = (tran_low_t)fdct_round_shift(t0);
628       output[3 * 8] = (tran_low_t)fdct_round_shift(t2);
629       output[5 * 8] = (tran_low_t)fdct_round_shift(t1);
630       output[7 * 8] = (tran_low_t)fdct_round_shift(t3);
631       input++;
632       output++;
633     }
634   }
635 
636   // Rows
637   for (i = 0; i < 8; ++i) {
638     fdct8(&intermediate[i * 8], &coeff_ptr[i * 8]);
639     for (j = 0; j < 8; ++j)
640       coeff_ptr[j + i * 8] /= 2;
641   }
642 
643   // TODO(jingning) Decide the need of these arguments after the
644   // quantization process is completed.
645   (void)zbin_ptr;
646   (void)quant_shift_ptr;
647   (void)iscan;
648 
649   memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
650   memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
651 
652   if (!skip_block) {
653     // Quantization pass: All coefficients with index >= zero_flag are
654     // skippable. Note: zero_flag can be zero.
655     for (i = 0; i < n_coeffs; i++) {
656       const int rc = scan[i];
657       const int coeff = coeff_ptr[rc];
658       const int coeff_sign = (coeff >> 31);
659       const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
660 
661       int tmp = clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
662       tmp = (tmp * quant_ptr[rc != 0]) >> 16;
663 
664       qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
665       dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
666 
667       if (tmp)
668         eob = i;
669     }
670   }
671   *eob_ptr = eob + 1;
672 }
673 
vp9_fht8x8_c(const int16_t * input,tran_low_t * output,int stride,int tx_type)674 void vp9_fht8x8_c(const int16_t *input, tran_low_t *output,
675                   int stride, int tx_type) {
676   if (tx_type == DCT_DCT) {
677     vpx_fdct8x8_c(input, output, stride);
678   } else {
679     tran_low_t out[64];
680     int i, j;
681     tran_low_t temp_in[8], temp_out[8];
682     const transform_2d ht = FHT_8[tx_type];
683 
684     // Columns
685     for (i = 0; i < 8; ++i) {
686       for (j = 0; j < 8; ++j)
687         temp_in[j] = input[j * stride + i] * 4;
688       ht.cols(temp_in, temp_out);
689       for (j = 0; j < 8; ++j)
690         out[j * 8 + i] = temp_out[j];
691     }
692 
693     // Rows
694     for (i = 0; i < 8; ++i) {
695       for (j = 0; j < 8; ++j)
696         temp_in[j] = out[j + i * 8];
697       ht.rows(temp_in, temp_out);
698       for (j = 0; j < 8; ++j)
699         output[j + i * 8] = (temp_out[j] + (temp_out[j] < 0)) >> 1;
700     }
701   }
702 }
703 
704 /* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
705    pixel. */
vp9_fwht4x4_c(const int16_t * input,tran_low_t * output,int stride)706 void vp9_fwht4x4_c(const int16_t *input, tran_low_t *output, int stride) {
707   int i;
708   tran_high_t a1, b1, c1, d1, e1;
709   const int16_t *ip_pass0 = input;
710   const tran_low_t *ip = NULL;
711   tran_low_t *op = output;
712 
713   for (i = 0; i < 4; i++) {
714     a1 = ip_pass0[0 * stride];
715     b1 = ip_pass0[1 * stride];
716     c1 = ip_pass0[2 * stride];
717     d1 = ip_pass0[3 * stride];
718 
719     a1 += b1;
720     d1 = d1 - c1;
721     e1 = (a1 - d1) >> 1;
722     b1 = e1 - b1;
723     c1 = e1 - c1;
724     a1 -= c1;
725     d1 += b1;
726     op[0] = (tran_low_t)a1;
727     op[4] = (tran_low_t)c1;
728     op[8] = (tran_low_t)d1;
729     op[12] = (tran_low_t)b1;
730 
731     ip_pass0++;
732     op++;
733   }
734   ip = output;
735   op = output;
736 
737   for (i = 0; i < 4; i++) {
738     a1 = ip[0];
739     b1 = ip[1];
740     c1 = ip[2];
741     d1 = ip[3];
742 
743     a1 += b1;
744     d1 -= c1;
745     e1 = (a1 - d1) >> 1;
746     b1 = e1 - b1;
747     c1 = e1 - c1;
748     a1 -= c1;
749     d1 += b1;
750     op[0] = (tran_low_t)(a1 * UNIT_QUANT_FACTOR);
751     op[1] = (tran_low_t)(c1 * UNIT_QUANT_FACTOR);
752     op[2] = (tran_low_t)(d1 * UNIT_QUANT_FACTOR);
753     op[3] = (tran_low_t)(b1 * UNIT_QUANT_FACTOR);
754 
755     ip += 4;
756     op += 4;
757   }
758 }
759 
vp9_fht16x16_c(const int16_t * input,tran_low_t * output,int stride,int tx_type)760 void vp9_fht16x16_c(const int16_t *input, tran_low_t *output,
761                     int stride, int tx_type) {
762   if (tx_type == DCT_DCT) {
763     vpx_fdct16x16_c(input, output, stride);
764   } else {
765     tran_low_t out[256];
766     int i, j;
767     tran_low_t temp_in[16], temp_out[16];
768     const transform_2d ht = FHT_16[tx_type];
769 
770     // Columns
771     for (i = 0; i < 16; ++i) {
772       for (j = 0; j < 16; ++j)
773         temp_in[j] = input[j * stride + i] * 4;
774       ht.cols(temp_in, temp_out);
775       for (j = 0; j < 16; ++j)
776         out[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] < 0)) >> 2;
777     }
778 
779     // Rows
780     for (i = 0; i < 16; ++i) {
781       for (j = 0; j < 16; ++j)
782         temp_in[j] = out[j + i * 16];
783       ht.rows(temp_in, temp_out);
784       for (j = 0; j < 16; ++j)
785         output[j + i * 16] = temp_out[j];
786     }
787   }
788 }
789 
790 #if CONFIG_VP9_HIGHBITDEPTH
vp9_highbd_fht4x4_c(const int16_t * input,tran_low_t * output,int stride,int tx_type)791 void vp9_highbd_fht4x4_c(const int16_t *input, tran_low_t *output,
792                          int stride, int tx_type) {
793   vp9_fht4x4_c(input, output, stride, tx_type);
794 }
795 
vp9_highbd_fht8x8_c(const int16_t * input,tran_low_t * output,int stride,int tx_type)796 void vp9_highbd_fht8x8_c(const int16_t *input, tran_low_t *output,
797                          int stride, int tx_type) {
798   vp9_fht8x8_c(input, output, stride, tx_type);
799 }
800 
vp9_highbd_fwht4x4_c(const int16_t * input,tran_low_t * output,int stride)801 void vp9_highbd_fwht4x4_c(const int16_t *input, tran_low_t *output,
802                           int stride) {
803   vp9_fwht4x4_c(input, output, stride);
804 }
805 
vp9_highbd_fht16x16_c(const int16_t * input,tran_low_t * output,int stride,int tx_type)806 void vp9_highbd_fht16x16_c(const int16_t *input, tran_low_t *output,
807                            int stride, int tx_type) {
808   vp9_fht16x16_c(input, output, stride, tx_type);
809 }
810 #endif  // CONFIG_VP9_HIGHBITDEPTH
811