1 /******************************************************************************
2 *
3 * Copyright (C) 2018 The Android Open Source Project
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 *
17 *****************************************************************************
18 * Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore
19 */
20 /**
21 *******************************************************************************
22 * @file
23 * ihevc_resi_trans.c
24 *
25 * @brief
26 * Contains function definitions for residual and forward transform
27 *
28 * @author
29 * 100470
30 *
31 * @par List of Functions:
32 * - ihevc_resi_trans_4x4_ttype1()
33 * - ihevc_resi_trans_4x4()
34 * - ihevc_resi_trans_8x8()
35 * - ihevc_resi_trans_16x16()
36 * - ihevc_resi_trans_32x32()
37 *
38 * @remarks
39 * None
40 *
41 *******************************************************************************
42 */
43 #include <stdio.h>
44 #include <string.h>
45 #include <stdlib.h>
46 #include "ihevc_typedefs.h"
47 #include "ihevc_macros.h"
48 #include "ihevc_platform_macros.h"
49 #include "ihevc_defs.h"
50 #include "ihevc_trans_tables.h"
51 #include "ihevc_resi_trans.h"
52 #include "ihevc_func_selector.h"
53 #include "ihevc_trans_macros.h"
54
55 /**
56 *******************************************************************************
57 *
58 * @brief
59 * This function performs residue calculation and forward transform type 1
60 * on input pixels
61 *
62 * @par Description:
63 * Performs residue calculation by subtracting source and prediction and
64 * followed by forward transform
65 *
66 * @param[in] pu1_src
67 * Input 4x4 pixels
68 *
69 * @param[in] pu1_pred
70 * Prediction data
71 *
72 * @param[in] pi2_tmp
73 * Temporary buffer of size 4x4
74 *
75 * @param[out] pi2_dst
76 * Output 4x4 coefficients
77 *
78 * @param[in] src_strd
79 * Input stride
80 *
81 * @param[in] pred_strd
82 * Prediction Stride
83 *
84 * @param[in] dst_strd
85 * Output Stride
86 *
87 * @param[in] e_chroma_plane
88 * Enum singalling chroma plane
89 *
90 *
91 * @returns Void
92 *
93 * @remarks
94 * None
95 *
96 *******************************************************************************
97 */
98
ihevc_resi_trans_4x4_ttype1(UWORD8 * pu1_src,UWORD8 * pu1_pred,WORD32 * pi4_temp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd,CHROMA_PLANE_ID_T e_chroma_plane)99 UWORD32 ihevc_resi_trans_4x4_ttype1(UWORD8 *pu1_src,
100 UWORD8 *pu1_pred,
101 WORD32 *pi4_temp,
102 WORD16 *pi2_dst,
103 WORD32 src_strd,
104 WORD32 pred_strd,
105 WORD32 dst_strd,
106 CHROMA_PLANE_ID_T e_chroma_plane)
107 {
108 WORD32 i, c[4];
109 WORD32 add, shift;
110 WORD32 trans_size;
111 WORD32 *pi4_tmp_orig;
112 WORD16 *pi2_dst_orig;
113 UWORD32 u4_blk_sad = 0;
114 UNUSED(e_chroma_plane);
115
116 pi2_dst_orig = pi2_dst;
117 pi4_tmp_orig = pi4_temp;
118 trans_size = TRANS_SIZE_4;
119
120 /* Residue + Forward Transform 1st stage */
121 shift = 1; // log2(iWidth) - 1 + g_uiBitIncrement
122 add = 1 << (shift - 1);
123
124 for(i = 0; i < trans_size; i++)
125 {
126 WORD32 resi_tmp_1, resi_tmp_2, resi_tmp_3;
127
128 // Intermediate Variables
129 resi_tmp_1 = pu1_src[0] - pu1_pred[0];
130 resi_tmp_2 = pu1_src[3] - pu1_pred[3];
131 c[0] = resi_tmp_1 + resi_tmp_2;
132 u4_blk_sad += abs(resi_tmp_1) + abs(resi_tmp_2);
133
134 resi_tmp_1 = pu1_src[1] - pu1_pred[1];
135 resi_tmp_2 = pu1_src[3] - pu1_pred[3];
136 c[1] = resi_tmp_1 + resi_tmp_2;
137 u4_blk_sad += abs(resi_tmp_1);
138
139 resi_tmp_1 = pu1_src[0] - pu1_pred[0];
140 resi_tmp_2 = pu1_src[1] - pu1_pred[1];
141 c[2] = resi_tmp_1 - resi_tmp_2;
142
143 resi_tmp_1 = pu1_src[2] - pu1_pred[2];
144 c[3] = 74 * resi_tmp_1;
145 u4_blk_sad += abs(resi_tmp_1);
146
147 pi4_temp[0] = (29 * c[0] + 55 * c[1] + c[3] + add) >> shift;
148
149 resi_tmp_1 = pu1_src[0] - pu1_pred[0];
150 resi_tmp_2 = pu1_src[1] - pu1_pred[1];
151 resi_tmp_3 = pu1_src[3] - pu1_pred[3];
152 pi4_temp[trans_size] =
153 (74 * (resi_tmp_1 + resi_tmp_2 - resi_tmp_3) + add)
154 >> shift;
155 pi4_temp[2 * trans_size] = (29 * c[2] + 55 * c[0] - c[3] + add) >> shift;
156 pi4_temp[3 * trans_size] = (55 * c[2] - 29 * c[1] + c[3] + add) >> shift;
157
158 pu1_src += src_strd;
159 pu1_pred += pred_strd;
160 pi4_temp++;
161 }
162
163 pi4_temp = pi4_tmp_orig;
164
165 /* Forward transform 2nd stage */
166 shift = 8; // log2(iHeight) + 6
167 add = 1 << (shift - 1);
168
169 for(i = 0; i < TRANS_SIZE_4; i++)
170 {
171 // Intermediate Variables
172 c[0] = pi4_temp[0] + pi4_temp[3];
173 c[1] = pi4_temp[1] + pi4_temp[3];
174 c[2] = pi4_temp[0] - pi4_temp[1];
175 c[3] = 74 * pi4_temp[2];
176
177 pi2_dst[0] = (29 * c[0] + 55 * c[1] + c[3] + add) >> shift;
178 pi2_dst[dst_strd] = (74 * (pi4_temp[0] + pi4_temp[1] - pi4_temp[3]) + add)
179 >> shift;
180 pi2_dst[2 * dst_strd] = (29 * c[2] + 55 * c[0] - c[3] + add) >> shift;
181 pi2_dst[3 * dst_strd] = (55 * c[2] - 29 * c[1] + c[3] + add) >> shift;
182
183 pi4_temp += trans_size;
184 pi2_dst++;
185 }
186
187 return u4_blk_sad;
188 }
189
190 /**
191 *******************************************************************************
192 *
193 * @brief
194 * This function performs residue calculation and forward transform on
195 * input pixels
196 *
197 * @par Description:
198 * Performs residue calculation by subtracting source and prediction and
199 * followed by forward transform
200 *
201 * @param[in] pu1_src
202 * Input 4x4 pixels
203 *
204 * @param[in] pu1_pred
205 * Prediction data
206 *
207 * @param[in] pi2_tmp
208 * Temporary buffer of size 4x4
209 *
210 * @param[out] pi2_dst
211 * Output 4x4 coefficients
212 *
213 * @param[in] src_strd
214 * Input stride
215 *
216 * @param[in] pred_strd
217 * Prediction Stride
218 *
219 * @param[in] dst_strd
220 * Output Stride
221 *
222 * @param[in] e_chroma_plane
223 * Enum singalling chroma plane
224 *
225 * @returns Void
226 *
227 * @remarks
228 * None
229 *
230 *******************************************************************************
231 */
232
ihevc_resi_trans_4x4(UWORD8 * pu1_src,UWORD8 * pu1_pred,WORD32 * pi4_temp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd,CHROMA_PLANE_ID_T e_chroma_plane)233 UWORD32 ihevc_resi_trans_4x4(UWORD8 *pu1_src,
234 UWORD8 *pu1_pred,
235 WORD32 *pi4_temp,
236 WORD16 *pi2_dst,
237 WORD32 src_strd,
238 WORD32 pred_strd,
239 WORD32 dst_strd,
240 CHROMA_PLANE_ID_T e_chroma_plane)
241 {
242 WORD32 i;
243 WORD32 e[2], o[2];
244 WORD32 add, shift;
245 WORD32 trans_size;
246 WORD32 *pi4_tmp_orig;
247 WORD16 *pi2_dst_orig;
248 UWORD32 u4_blk_sad=0;
249 WORD32 chroma_flag = 0;
250
251 if (e_chroma_plane != NULL_PLANE)
252 {
253 chroma_flag = 1;
254 pu1_src += e_chroma_plane;
255 pu1_pred += e_chroma_plane;
256 }
257
258 pi2_dst_orig = pi2_dst;
259 pi4_tmp_orig = pi4_temp;
260 trans_size = TRANS_SIZE_4;
261
262 /* Residue + Forward Transform 1st stage */
263 shift = 1; // log2(iWidth) - 1 + g_uiBitIncrement
264 add = 1 << (shift - 1);
265
266 for(i = 0; i < trans_size; i++)
267 {
268 WORD32 resi_tmp_1, resi_tmp_2;
269
270 /* e and o */
271 resi_tmp_1 = pu1_src[0 + 0*chroma_flag] - pu1_pred[0 + 0*chroma_flag];
272 resi_tmp_2 = pu1_src[3 + 3*chroma_flag] - pu1_pred[3 + 3*chroma_flag];
273 e[0] = resi_tmp_1 + resi_tmp_2;
274 o[0] = resi_tmp_1 - resi_tmp_2;
275 u4_blk_sad += abs(resi_tmp_1);
276 u4_blk_sad += abs(resi_tmp_2);
277
278 resi_tmp_1 = pu1_src[1 + 1*chroma_flag] - pu1_pred[1 + 1*chroma_flag];
279 resi_tmp_2 = pu1_src[2 + 2*chroma_flag] - pu1_pred[2 + 2*chroma_flag];
280 e[1] = resi_tmp_1 + resi_tmp_2;
281 o[1] = resi_tmp_1 - resi_tmp_2;
282 u4_blk_sad += abs(resi_tmp_1);
283 u4_blk_sad += abs(resi_tmp_2);
284
285 pi4_temp[0] = (g_ai2_ihevc_trans_4[0][0] * e[0]
286 + g_ai2_ihevc_trans_4[0][1] * e[1]);// + add) >> shift;
287 pi4_temp[2 * trans_size] = (g_ai2_ihevc_trans_4[2][0] * e[0]
288 + g_ai2_ihevc_trans_4[2][1] * e[1]);// + add) >> shift;
289 pi4_temp[trans_size] = (g_ai2_ihevc_trans_4[1][0] * o[0]
290 + g_ai2_ihevc_trans_4[1][1] * o[1]);// + add) >> shift;
291 pi4_temp[3 * trans_size] = (g_ai2_ihevc_trans_4[3][0] * o[0]
292 + g_ai2_ihevc_trans_4[3][1] * o[1]);// + add) >> shift;
293
294 pu1_src += src_strd;
295 pu1_pred += pred_strd;
296 pi4_temp++;
297 }
298
299 pi4_temp = pi4_tmp_orig;
300 /* Forward Transform 2nd stage */
301 shift = 9; // log2(iHeight) + 6
302 add = 1 << (shift - 1);
303
304 for(i = 0; i < trans_size; i++)
305 {
306
307 /* e and o */
308 e[0] = pi4_temp[0] + pi4_temp[3];
309 o[0] = pi4_temp[0] - pi4_temp[3];
310 e[1] = pi4_temp[1] + pi4_temp[2];
311 o[1] = pi4_temp[1] - pi4_temp[2];
312
313 pi2_dst[0] = (g_ai2_ihevc_trans_4[0][0] * e[0]
314 + g_ai2_ihevc_trans_4[0][1] * e[1] + add) >> shift;
315 pi2_dst[2 * dst_strd] = (g_ai2_ihevc_trans_4[2][0] * e[0]
316 + g_ai2_ihevc_trans_4[2][1] * e[1] + add) >> shift;
317 pi2_dst[dst_strd] = (g_ai2_ihevc_trans_4[1][0] * o[0]
318 + g_ai2_ihevc_trans_4[1][1] * o[1] + add) >> shift;
319 pi2_dst[3 * dst_strd] = (g_ai2_ihevc_trans_4[3][0] * o[0]
320 + g_ai2_ihevc_trans_4[3][1] * o[1] + add) >> shift;
321
322 pi4_temp += trans_size;
323 pi2_dst++;
324 }
325
326 return u4_blk_sad;
327 }
328
ihevc_resi_trans_4x4_16bit(WORD16 * pi2_src,UWORD8 * pu1_pred,WORD16 * pi2_tmp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd)329 void ihevc_resi_trans_4x4_16bit(WORD16 *pi2_src,
330 UWORD8 *pu1_pred,
331 WORD16 *pi2_tmp,
332 WORD16 *pi2_dst,
333 WORD32 src_strd,
334 WORD32 pred_strd,
335 WORD32 dst_strd)
336 {
337 WORD32 i;
338 WORD32 e[2], o[2];
339 WORD32 add, shift;
340 WORD32 trans_size;
341 WORD16 *pi2_tmp_orig;
342 WORD16 *pi2_dst_orig;
343
344 pi2_dst_orig = pi2_dst;
345 pi2_tmp_orig = pi2_tmp;
346 trans_size = TRANS_SIZE_4;
347
348 /* Residue + Forward Transform 1st stage */
349 shift = 1; // log2(iWidth) - 1 + g_uiBitIncrement
350 add = 1 << (shift - 1);
351
352 for(i = 0; i < trans_size; i++)
353 {
354 WORD32 resi_tmp_1, resi_tmp_2;
355
356 /* e and o */
357 resi_tmp_1 = pi2_src[0] - pu1_pred[0];
358 resi_tmp_2 = pi2_src[3] - pu1_pred[3];
359 e[0] = resi_tmp_1 + resi_tmp_2;
360 o[0] = resi_tmp_1 - resi_tmp_2;
361
362 resi_tmp_1 = pi2_src[1] - pu1_pred[1];
363 resi_tmp_2 = pi2_src[2] - pu1_pred[2];
364 e[1] = resi_tmp_1 + resi_tmp_2;
365 o[1] = resi_tmp_1 - resi_tmp_2;
366
367 pi2_tmp[0] = (g_ai2_ihevc_trans_4[0][0] * e[0]
368 + g_ai2_ihevc_trans_4[0][1] * e[1] + add) >> shift;
369 pi2_tmp[2 * trans_size] = (g_ai2_ihevc_trans_4[2][0] * e[0]
370 + g_ai2_ihevc_trans_4[2][1] * e[1] + add) >> shift;
371 pi2_tmp[trans_size] = (g_ai2_ihevc_trans_4[1][0] * o[0]
372 + g_ai2_ihevc_trans_4[1][1] * o[1] + add) >> shift;
373 pi2_tmp[3 * trans_size] = (g_ai2_ihevc_trans_4[3][0] * o[0]
374 + g_ai2_ihevc_trans_4[3][1] * o[1] + add) >> shift;
375
376 pi2_src += src_strd;
377 pu1_pred += pred_strd;
378 pi2_tmp++;
379 }
380
381 pi2_tmp = pi2_tmp_orig;
382 /* Forward Transform 2nd stage */
383 shift = 8; // log2(iHeight) + 6
384 add = 1 << (shift - 1);
385
386 for(i = 0; i < trans_size; i++)
387 {
388
389 /* e and o */
390 e[0] = pi2_tmp[0] + pi2_tmp[3];
391 o[0] = pi2_tmp[0] - pi2_tmp[3];
392 e[1] = pi2_tmp[1] + pi2_tmp[2];
393 o[1] = pi2_tmp[1] - pi2_tmp[2];
394
395 pi2_dst[0] = (g_ai2_ihevc_trans_4[0][0] * e[0]
396 + g_ai2_ihevc_trans_4[0][1] * e[1] + add) >> shift;
397 pi2_dst[2 * dst_strd] = (g_ai2_ihevc_trans_4[2][0] * e[0]
398 + g_ai2_ihevc_trans_4[2][1] * e[1] + add) >> shift;
399 pi2_dst[dst_strd] = (g_ai2_ihevc_trans_4[1][0] * o[0]
400 + g_ai2_ihevc_trans_4[1][1] * o[1] + add) >> shift;
401 pi2_dst[3 * dst_strd] = (g_ai2_ihevc_trans_4[3][0] * o[0]
402 + g_ai2_ihevc_trans_4[3][1] * o[1] + add) >> shift;
403
404 pi2_tmp += trans_size;
405 pi2_dst++;
406 }
407 }
408 /**
409 *******************************************************************************
410 *
411 * @brief
412 * This function performs residue calculation and forward transform on
413 * input pixels
414 *
415 * @par Description:
416 * Performs residue calculation by subtracting source and prediction and
417 * followed by forward transform
418 *
419 * @param[in] pu1_src
420 * Input 8x8 pixels
421 *
422 * @param[in] pu1_pred
423 * Prediction data
424 *
425 * @param[in] pi2_tmp
426 * Temporary buffer of size 8x8
427 *
428 * @param[out] pi2_dst
429 * Output 8x8 coefficients
430 *
431 * @param[in] src_strd
432 * Input stride
433 *
434 * @param[in] pred_strd
435 * Prediction Stride
436 *
437 * @param[in] dst_strd
438 * Output Stride
439 *
440 * @param[in] e_chroma_plane
441 * Enum singalling chroma plane
442 *
443 * @returns Void
444 *
445 * @remarks
446 * None
447 *
448 *******************************************************************************
449 */
450
ihevc_resi_trans_8x8(UWORD8 * pu1_src,UWORD8 * pu1_pred,WORD32 * pi4_temp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd,CHROMA_PLANE_ID_T e_chroma_plane)451 UWORD32 ihevc_resi_trans_8x8(UWORD8 *pu1_src,
452 UWORD8 *pu1_pred,
453 WORD32 *pi4_temp,
454 WORD16 *pi2_dst,
455 WORD32 src_strd,
456 WORD32 pred_strd,
457 WORD32 dst_strd,
458 CHROMA_PLANE_ID_T e_chroma_plane)
459 {
460 WORD32 i, k;
461 WORD32 e[4], o[4];
462 WORD32 ee[2], eo[2];
463 WORD32 add, shift;
464 WORD32 trans_size;
465 WORD32 *pi4_tmp_orig;
466 // WORD16 *pi2_tmp;
467 WORD16 *pi2_dst_orig;
468 UWORD32 u4_blk_sad=0;
469 WORD32 chroma_flag = 0;
470
471 if (e_chroma_plane != NULL_PLANE)
472 {
473 chroma_flag = 1;
474 pu1_src += e_chroma_plane;
475 pu1_pred += e_chroma_plane;
476 }
477
478 pi2_dst_orig = pi2_dst;
479 pi4_tmp_orig = pi4_temp;
480 trans_size = TRANS_SIZE_8;
481 /* Residue + Forward Transform 1st stage */
482 shift = 2; // log2(iWidth) - 1 + g_uiBitIncrement
483 add = 1 << (shift - 1);
484
485 for(i = 0; i < trans_size; i++)
486 {
487 WORD32 resi_tmp_1, resi_tmp_2;
488
489 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
490 /* e and o*/
491 for(k = 0; k < 4; k++)
492 {
493 resi_tmp_1 = pu1_src[k*(1+chroma_flag)] - pu1_pred[k*(1+chroma_flag)];
494 resi_tmp_2 = pu1_src[(7-k)*(1+chroma_flag)] - pu1_pred[(7-k)*(1+chroma_flag)];
495 e[k] = resi_tmp_1 + resi_tmp_2;
496 o[k] = resi_tmp_1 - resi_tmp_2;
497 u4_blk_sad += abs(resi_tmp_1) + abs(resi_tmp_2);
498 }
499 /* ee and eo */
500 ee[0] = e[0] + e[3];
501 eo[0] = e[0] - e[3];
502 ee[1] = e[1] + e[2];
503 eo[1] = e[1] - e[2];
504
505 pi4_temp[0] = (g_ai2_ihevc_trans_8[0][0] * ee[0]
506 + g_ai2_ihevc_trans_8[0][1] * ee[1]);// + add) >> shift;
507 pi4_temp[4 * trans_size] = (g_ai2_ihevc_trans_8[4][0] * ee[0]
508 + g_ai2_ihevc_trans_8[4][1] * ee[1]);// + add) >> shift;
509 pi4_temp[2 * trans_size] = (g_ai2_ihevc_trans_8[2][0] * eo[0]
510 + g_ai2_ihevc_trans_8[2][1] * eo[1]);// + add) >> shift;
511 pi4_temp[6 * trans_size] = (g_ai2_ihevc_trans_8[6][0] * eo[0]
512 + g_ai2_ihevc_trans_8[6][1] * eo[1]);// + add) >> shift;
513
514 pi4_temp[trans_size] = (g_ai2_ihevc_trans_8[1][0] * o[0]
515 + g_ai2_ihevc_trans_8[1][1] * o[1]
516 + g_ai2_ihevc_trans_8[1][2] * o[2]
517 + g_ai2_ihevc_trans_8[1][3] * o[3]);// + add) >> shift;
518 pi4_temp[3 * trans_size] = (g_ai2_ihevc_trans_8[3][0] * o[0]
519 + g_ai2_ihevc_trans_8[3][1] * o[1]
520 + g_ai2_ihevc_trans_8[3][2] * o[2]
521 + g_ai2_ihevc_trans_8[3][3] * o[3]);// + add) >> shift;
522 pi4_temp[5 * trans_size] = (g_ai2_ihevc_trans_8[5][0] * o[0]
523 + g_ai2_ihevc_trans_8[5][1] * o[1]
524 + g_ai2_ihevc_trans_8[5][2] * o[2]
525 + g_ai2_ihevc_trans_8[5][3] * o[3]);// + add) >> shift;
526 pi4_temp[7 * trans_size] = (g_ai2_ihevc_trans_8[7][0] * o[0]
527 + g_ai2_ihevc_trans_8[7][1] * o[1]
528 + g_ai2_ihevc_trans_8[7][2] * o[2]
529 + g_ai2_ihevc_trans_8[7][3] * o[3]);// + add) >> shift;
530
531 pu1_src += src_strd;
532 pu1_pred += pred_strd;
533 pi4_temp++;
534 }
535
536 pi4_temp = pi4_tmp_orig;
537 /* Forward Transform 2nd stage */
538 shift = 11; // log2(iHeight) + 6
539 add = 1 << (shift - 1);
540
541 for(i = 0; i < trans_size; i++)
542 {
543 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
544 /* e and o*/
545 for(k = 0; k < 4; k++)
546 {
547 e[k] = pi4_temp[k] + pi4_temp[7 - k];
548 o[k] = pi4_temp[k] - pi4_temp[7 - k];
549 }
550 /* ee and eo */
551 ee[0] = e[0] + e[3];
552 eo[0] = e[0] - e[3];
553 ee[1] = e[1] + e[2];
554 eo[1] = e[1] - e[2];
555
556 pi2_dst[0] = (g_ai2_ihevc_trans_8[0][0] * ee[0]
557 + g_ai2_ihevc_trans_8[0][1] * ee[1] + add) >> shift;
558 pi2_dst[4 * dst_strd] = (g_ai2_ihevc_trans_8[4][0] * ee[0]
559 + g_ai2_ihevc_trans_8[4][1] * ee[1] + add) >> shift;
560 pi2_dst[2 * dst_strd] = (g_ai2_ihevc_trans_8[2][0] * eo[0]
561 + g_ai2_ihevc_trans_8[2][1] * eo[1] + add) >> shift;
562 pi2_dst[6 * dst_strd] = (g_ai2_ihevc_trans_8[6][0] * eo[0]
563 + g_ai2_ihevc_trans_8[6][1] * eo[1] + add) >> shift;
564
565 pi2_dst[dst_strd] = (g_ai2_ihevc_trans_8[1][0] * o[0]
566 + g_ai2_ihevc_trans_8[1][1] * o[1]
567 + g_ai2_ihevc_trans_8[1][2] * o[2]
568 + g_ai2_ihevc_trans_8[1][3] * o[3] + add) >> shift;
569 pi2_dst[3 * dst_strd] = (g_ai2_ihevc_trans_8[3][0] * o[0]
570 + g_ai2_ihevc_trans_8[3][1] * o[1]
571 + g_ai2_ihevc_trans_8[3][2] * o[2]
572 + g_ai2_ihevc_trans_8[3][3] * o[3] + add) >> shift;
573 pi2_dst[5 * dst_strd] = (g_ai2_ihevc_trans_8[5][0] * o[0]
574 + g_ai2_ihevc_trans_8[5][1] * o[1]
575 + g_ai2_ihevc_trans_8[5][2] * o[2]
576 + g_ai2_ihevc_trans_8[5][3] * o[3] + add) >> shift;
577 pi2_dst[7 * dst_strd] = (g_ai2_ihevc_trans_8[7][0] * o[0]
578 + g_ai2_ihevc_trans_8[7][1] * o[1]
579 + g_ai2_ihevc_trans_8[7][2] * o[2]
580 + g_ai2_ihevc_trans_8[7][3] * o[3] + add) >> shift;
581
582 pi4_temp += trans_size;
583 pi2_dst++;
584 }
585
586 return u4_blk_sad;
587 }
588
ihevc_resi_trans_8x8_16bit(WORD16 * pi2_src,UWORD8 * pu1_pred,WORD16 * pi2_tmp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd)589 void ihevc_resi_trans_8x8_16bit(WORD16 *pi2_src,
590 UWORD8 *pu1_pred,
591 WORD16 *pi2_tmp,
592 WORD16 *pi2_dst,
593 WORD32 src_strd,
594 WORD32 pred_strd,
595 WORD32 dst_strd)
596 {
597 WORD32 i, k;
598 WORD32 e[4], o[4];
599 WORD32 ee[2], eo[2];
600 WORD32 add, shift;
601 WORD32 trans_size;
602 WORD16 *pi2_tmp_orig;
603 WORD16 *pi2_dst_orig;
604
605 pi2_dst_orig = pi2_dst;
606 pi2_tmp_orig = pi2_tmp;
607 trans_size = TRANS_SIZE_8;
608 /* Residue + Forward Transform 1st stage */
609 shift = 2; // log2(iWidth) - 1 + g_uiBitIncrement
610 add = 1 << (shift - 1);
611
612 for(i = 0; i < trans_size; i++)
613 {
614 WORD32 resi_tmp_1, resi_tmp_2;
615
616 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
617 /* e and o*/
618 for(k = 0; k < 4; k++)
619 {
620 resi_tmp_1 = pi2_src[k] - pu1_pred[k];
621 resi_tmp_2 = pi2_src[7 - k] - pu1_pred[7 - k];
622 e[k] = resi_tmp_1 + resi_tmp_2;
623 o[k] = resi_tmp_1 - resi_tmp_2;
624 }
625 /* ee and eo */
626 ee[0] = e[0] + e[3];
627 eo[0] = e[0] - e[3];
628 ee[1] = e[1] + e[2];
629 eo[1] = e[1] - e[2];
630
631 pi2_tmp[0] = (g_ai2_ihevc_trans_8[0][0] * ee[0]
632 + g_ai2_ihevc_trans_8[0][1] * ee[1] + add) >> shift;
633 pi2_tmp[4 * trans_size] = (g_ai2_ihevc_trans_8[4][0] * ee[0]
634 + g_ai2_ihevc_trans_8[4][1] * ee[1] + add) >> shift;
635 pi2_tmp[2 * trans_size] = (g_ai2_ihevc_trans_8[2][0] * eo[0]
636 + g_ai2_ihevc_trans_8[2][1] * eo[1] + add) >> shift;
637 pi2_tmp[6 * trans_size] = (g_ai2_ihevc_trans_8[6][0] * eo[0]
638 + g_ai2_ihevc_trans_8[6][1] * eo[1] + add) >> shift;
639
640 pi2_tmp[trans_size] = (g_ai2_ihevc_trans_8[1][0] * o[0]
641 + g_ai2_ihevc_trans_8[1][1] * o[1]
642 + g_ai2_ihevc_trans_8[1][2] * o[2]
643 + g_ai2_ihevc_trans_8[1][3] * o[3] + add) >> shift;
644 pi2_tmp[3 * trans_size] = (g_ai2_ihevc_trans_8[3][0] * o[0]
645 + g_ai2_ihevc_trans_8[3][1] * o[1]
646 + g_ai2_ihevc_trans_8[3][2] * o[2]
647 + g_ai2_ihevc_trans_8[3][3] * o[3] + add) >> shift;
648 pi2_tmp[5 * trans_size] = (g_ai2_ihevc_trans_8[5][0] * o[0]
649 + g_ai2_ihevc_trans_8[5][1] * o[1]
650 + g_ai2_ihevc_trans_8[5][2] * o[2]
651 + g_ai2_ihevc_trans_8[5][3] * o[3] + add) >> shift;
652 pi2_tmp[7 * trans_size] = (g_ai2_ihevc_trans_8[7][0] * o[0]
653 + g_ai2_ihevc_trans_8[7][1] * o[1]
654 + g_ai2_ihevc_trans_8[7][2] * o[2]
655 + g_ai2_ihevc_trans_8[7][3] * o[3] + add) >> shift;
656
657 pi2_src += src_strd;
658 pu1_pred += pred_strd;
659 pi2_tmp++;
660 }
661
662 pi2_tmp = pi2_tmp_orig;
663 /* Forward Transform 2nd stage */
664 shift = 9; // log2(iHeight) + 6
665 add = 1 << (shift - 1);
666
667 for(i = 0; i < trans_size; i++)
668 {
669 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
670 /* e and o*/
671 for(k = 0; k < 4; k++)
672 {
673 e[k] = pi2_tmp[k] + pi2_tmp[7 - k];
674 o[k] = pi2_tmp[k] - pi2_tmp[7 - k];
675 }
676 /* ee and eo */
677 ee[0] = e[0] + e[3];
678 eo[0] = e[0] - e[3];
679 ee[1] = e[1] + e[2];
680 eo[1] = e[1] - e[2];
681
682 pi2_dst[0] = (g_ai2_ihevc_trans_8[0][0] * ee[0]
683 + g_ai2_ihevc_trans_8[0][1] * ee[1] + add) >> shift;
684 pi2_dst[4 * dst_strd] = (g_ai2_ihevc_trans_8[4][0] * ee[0]
685 + g_ai2_ihevc_trans_8[4][1] * ee[1] + add) >> shift;
686 pi2_dst[2 * dst_strd] = (g_ai2_ihevc_trans_8[2][0] * eo[0]
687 + g_ai2_ihevc_trans_8[2][1] * eo[1] + add) >> shift;
688 pi2_dst[6 * dst_strd] = (g_ai2_ihevc_trans_8[6][0] * eo[0]
689 + g_ai2_ihevc_trans_8[6][1] * eo[1] + add) >> shift;
690
691 pi2_dst[dst_strd] = (g_ai2_ihevc_trans_8[1][0] * o[0]
692 + g_ai2_ihevc_trans_8[1][1] * o[1]
693 + g_ai2_ihevc_trans_8[1][2] * o[2]
694 + g_ai2_ihevc_trans_8[1][3] * o[3] + add) >> shift;
695 pi2_dst[3 * dst_strd] = (g_ai2_ihevc_trans_8[3][0] * o[0]
696 + g_ai2_ihevc_trans_8[3][1] * o[1]
697 + g_ai2_ihevc_trans_8[3][2] * o[2]
698 + g_ai2_ihevc_trans_8[3][3] * o[3] + add) >> shift;
699 pi2_dst[5 * dst_strd] = (g_ai2_ihevc_trans_8[5][0] * o[0]
700 + g_ai2_ihevc_trans_8[5][1] * o[1]
701 + g_ai2_ihevc_trans_8[5][2] * o[2]
702 + g_ai2_ihevc_trans_8[5][3] * o[3] + add) >> shift;
703 pi2_dst[7 * dst_strd] = (g_ai2_ihevc_trans_8[7][0] * o[0]
704 + g_ai2_ihevc_trans_8[7][1] * o[1]
705 + g_ai2_ihevc_trans_8[7][2] * o[2]
706 + g_ai2_ihevc_trans_8[7][3] * o[3] + add) >> shift;
707
708 pi2_tmp += trans_size;
709 pi2_dst++;
710 }
711 }
712 /**
713 *******************************************************************************
714 *
715 * @brief
716 * This function performs residue calculation and forward transform on
717 * input pixels
718 *
719 * @par Description:
720 * Performs residue calculation by subtracting source and prediction and
721 * followed by forward transform
722 *
723 * @param[in] pu1_src
724 * Input 16x16 pixels
725 *
726 * @param[in] pu1_pred
727 * Prediction data
728 *
729 * @param[in] pi2_tmp
730 * Temporary buffer of size 16x16
731 *
732 * @param[out] pi2_dst
733 * Output 16x16 coefficients
734 *
735 * @param[in] src_strd
736 * Input stride
737 *
738 * @param[in] pred_strd
739 * Prediction Stride
740 *
741 * @param[in] dst_strd
742 * Output Stride
743 *
744 * @param[in] e_chroma_plane
745 * Enum singalling chroma plane
746 *
747 * @returns Void
748 *
749 * @remarks
750 * None
751 *
752 *******************************************************************************
753 */
754
ihevc_resi_trans_16x16(UWORD8 * pu1_src,UWORD8 * pu1_pred,WORD32 * pi4_temp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd,CHROMA_PLANE_ID_T e_chroma_plane)755 UWORD32 ihevc_resi_trans_16x16(UWORD8 *pu1_src,
756 UWORD8 *pu1_pred,
757 WORD32 *pi4_temp,
758 WORD16 *pi2_dst,
759 WORD32 src_strd,
760 WORD32 pred_strd,
761 WORD32 dst_strd,
762 CHROMA_PLANE_ID_T e_chroma_plane)
763 {
764 WORD32 i, k;
765 WORD32 e[8], o[8];
766 WORD32 ee[4], eo[4];
767 WORD32 eee[2], eeo[2];
768 WORD32 add, shift;
769 WORD32 trans_size;
770 WORD32 *pi4_tmp_orig;
771 WORD16 *pi2_dst_orig;
772 UWORD32 u4_blk_sad = 0;
773 WORD32 chroma_flag = 0;
774
775 if (e_chroma_plane != NULL_PLANE)
776 {
777 chroma_flag = 1;
778 pu1_src += e_chroma_plane;
779 pu1_pred += e_chroma_plane;
780 }
781
782 pi2_dst_orig = pi2_dst;
783 pi4_tmp_orig = pi4_temp;
784 trans_size = TRANS_SIZE_16;
785 /* Residue + Forward Transform 1st stage */
786 shift = 3; // log2(iWidth) - 1 + g_uiBitIncrement
787 add = 1 << (shift - 1);
788
789 for(i = 0; i < trans_size; i++)
790 {
791 WORD32 resi_tmp_1, resi_tmp_2;
792 /* e and o*/
793 for(k = 0; k < 8; k++)
794 {
795 resi_tmp_1 = pu1_src[k*(1+chroma_flag)] - pu1_pred[k*(1+chroma_flag)];
796 resi_tmp_2 = pu1_src[(15-k)*(1+chroma_flag)] - pu1_pred[(15-k)*(1+chroma_flag)];
797 e[k] = resi_tmp_1 + resi_tmp_2;
798 o[k] = resi_tmp_1 - resi_tmp_2;
799 u4_blk_sad += abs(resi_tmp_1) + abs(resi_tmp_2);
800 }
801 /* ee and eo */
802 for(k = 0; k < 4; k++)
803 {
804 ee[k] = e[k] + e[7 - k];
805 eo[k] = e[k] - e[7 - k];
806 }
807 /* eee and eeo */
808 eee[0] = ee[0] + ee[3];
809 eeo[0] = ee[0] - ee[3];
810 eee[1] = ee[1] + ee[2];
811 eeo[1] = ee[1] - ee[2];
812
813 pi4_temp[0] = (g_ai2_ihevc_trans_16[0][0] * eee[0]
814 + g_ai2_ihevc_trans_16[0][1] * eee[1]);// + add) >> shift;
815 pi4_temp[8 * trans_size] = (g_ai2_ihevc_trans_16[8][0] * eee[0]
816 + g_ai2_ihevc_trans_16[8][1] * eee[1]);// + add) >> shift;
817 pi4_temp[4 * trans_size] = (g_ai2_ihevc_trans_16[4][0] * eeo[0]
818 + g_ai2_ihevc_trans_16[4][1] * eeo[1]);// + add) >> shift;
819 pi4_temp[12 * trans_size] = (g_ai2_ihevc_trans_16[12][0] * eeo[0]
820 + g_ai2_ihevc_trans_16[12][1] * eeo[1]);// + add) >> shift;
821
822 for(k = 2; k < 16; k += 4)
823 {
824 pi4_temp[k * trans_size] = (g_ai2_ihevc_trans_16[k][0] * eo[0]
825 + g_ai2_ihevc_trans_16[k][1] * eo[1]
826 + g_ai2_ihevc_trans_16[k][2] * eo[2]
827 + g_ai2_ihevc_trans_16[k][3] * eo[3]);// + add)>> shift;
828
829 }
830
831 for(k = 1; k < 16; k += 2)
832 {
833 pi4_temp[k * trans_size] = (g_ai2_ihevc_trans_16[k][0] * o[0]
834 + g_ai2_ihevc_trans_16[k][1] * o[1]
835 + g_ai2_ihevc_trans_16[k][2] * o[2]
836 + g_ai2_ihevc_trans_16[k][3] * o[3]
837 + g_ai2_ihevc_trans_16[k][4] * o[4]
838 + g_ai2_ihevc_trans_16[k][5] * o[5]
839 + g_ai2_ihevc_trans_16[k][6] * o[6]
840 + g_ai2_ihevc_trans_16[k][7] * o[7]);// + add) >> shift;
841 }
842 pu1_src += src_strd;
843 pu1_pred += pred_strd;
844 pi4_temp++;
845 }
846
847 pi4_temp = pi4_tmp_orig;
848 /* Forward Transform 2nd stage */
849 shift = 13; // log2(iHeight) + 6
850 add = 1 << (shift - 1);
851
852 for(i = 0; i < TRANS_SIZE_16; i++)
853 {
854 /* e and o*/
855 for(k = 0; k < 8; k++)
856 {
857 e[k] = pi4_temp[k] + pi4_temp[15 - k];
858 o[k] = pi4_temp[k] - pi4_temp[15 - k];
859 }
860 /* ee and eo */
861 for(k = 0; k < 4; k++)
862 {
863 ee[k] = e[k] + e[7 - k];
864 eo[k] = e[k] - e[7 - k];
865 }
866 /* eee and eeo */
867 eee[0] = ee[0] + ee[3];
868 eeo[0] = ee[0] - ee[3];
869 eee[1] = ee[1] + ee[2];
870 eeo[1] = ee[1] - ee[2];
871
872 pi2_dst[0] = (g_ai2_ihevc_trans_16[0][0] * eee[0]
873 + g_ai2_ihevc_trans_16[0][1] * eee[1] + add) >> shift;
874 pi2_dst[8 * dst_strd] = (g_ai2_ihevc_trans_16[8][0] * eee[0]
875 + g_ai2_ihevc_trans_16[8][1] * eee[1] + add) >> shift;
876 pi2_dst[4 * dst_strd] = (g_ai2_ihevc_trans_16[4][0] * eeo[0]
877 + g_ai2_ihevc_trans_16[4][1] * eeo[1] + add) >> shift;
878 pi2_dst[12 * dst_strd] = (g_ai2_ihevc_trans_16[12][0] * eeo[0]
879 + g_ai2_ihevc_trans_16[12][1] * eeo[1] + add) >> shift;
880
881 for(k = 2; k < 16; k += 4)
882 {
883 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_16[k][0] * eo[0]
884 + g_ai2_ihevc_trans_16[k][1] * eo[1]
885 + g_ai2_ihevc_trans_16[k][2] * eo[2]
886 + g_ai2_ihevc_trans_16[k][3] * eo[3] + add)
887 >> shift;
888 }
889
890 for(k = 1; k < 16; k += 2)
891 {
892 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_16[k][0] * o[0]
893 + g_ai2_ihevc_trans_16[k][1] * o[1]
894 + g_ai2_ihevc_trans_16[k][2] * o[2]
895 + g_ai2_ihevc_trans_16[k][3] * o[3]
896 + g_ai2_ihevc_trans_16[k][4] * o[4]
897 + g_ai2_ihevc_trans_16[k][5] * o[5]
898 + g_ai2_ihevc_trans_16[k][6] * o[6]
899 + g_ai2_ihevc_trans_16[k][7] * o[7] + add) >> shift;
900 }
901
902 pi4_temp += trans_size;
903 pi2_dst++;
904 }
905
906 return u4_blk_sad;
907 }
908
909
ihevc_resi_trans_16x16_16bit(WORD16 * pi2_src,UWORD8 * pu1_pred,WORD16 * pi2_tmp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd)910 void ihevc_resi_trans_16x16_16bit(WORD16 *pi2_src,
911 UWORD8 *pu1_pred,
912 WORD16 *pi2_tmp,
913 WORD16 *pi2_dst,
914 WORD32 src_strd,
915 WORD32 pred_strd,
916 WORD32 dst_strd)
917 {
918 WORD32 i, k;
919 WORD32 e[8], o[8];
920 WORD32 ee[4], eo[4];
921 WORD32 eee[2], eeo[2];
922 WORD32 add, shift;
923 WORD32 trans_size;
924 WORD16 *pi2_tmp_orig;
925 WORD16 *pi2_dst_orig;
926
927 pi2_dst_orig = pi2_dst;
928 pi2_tmp_orig = pi2_tmp;
929 trans_size = TRANS_SIZE_16;
930 /* Residue + Forward Transform 1st stage */
931 shift = 3; // log2(iWidth) - 1 + g_uiBitIncrement
932 add = 1 << (shift - 1);
933
934 for(i = 0; i < trans_size; i++)
935 {
936 WORD32 resi_tmp_1, resi_tmp_2;
937 /* e and o*/
938 for(k = 0; k < 8; k++)
939 {
940 resi_tmp_1 = pi2_src[k] - pu1_pred[k];
941 resi_tmp_2 = pi2_src[15 - k] - pu1_pred[15 - k];
942 e[k] = resi_tmp_1 + resi_tmp_2;
943 o[k] = resi_tmp_1 - resi_tmp_2;
944 }
945 /* ee and eo */
946 for(k = 0; k < 4; k++)
947 {
948 ee[k] = e[k] + e[7 - k];
949 eo[k] = e[k] - e[7 - k];
950 }
951 /* eee and eeo */
952 eee[0] = ee[0] + ee[3];
953 eeo[0] = ee[0] - ee[3];
954 eee[1] = ee[1] + ee[2];
955 eeo[1] = ee[1] - ee[2];
956
957 pi2_tmp[0] = (g_ai2_ihevc_trans_16[0][0] * eee[0]
958 + g_ai2_ihevc_trans_16[0][1] * eee[1] + add) >> shift;
959 pi2_tmp[8 * trans_size] = (g_ai2_ihevc_trans_16[8][0] * eee[0]
960 + g_ai2_ihevc_trans_16[8][1] * eee[1] + add) >> shift;
961 pi2_tmp[4 * trans_size] = (g_ai2_ihevc_trans_16[4][0] * eeo[0]
962 + g_ai2_ihevc_trans_16[4][1] * eeo[1] + add) >> shift;
963 pi2_tmp[12 * trans_size] = (g_ai2_ihevc_trans_16[12][0] * eeo[0]
964 + g_ai2_ihevc_trans_16[12][1] * eeo[1] + add) >> shift;
965
966 for(k = 2; k < 16; k += 4)
967 {
968 pi2_tmp[k * trans_size] = (g_ai2_ihevc_trans_16[k][0] * eo[0]
969 + g_ai2_ihevc_trans_16[k][1] * eo[1]
970 + g_ai2_ihevc_trans_16[k][2] * eo[2]
971 + g_ai2_ihevc_trans_16[k][3] * eo[3] + add)
972 >> shift;
973 }
974
975 for(k = 1; k < 16; k += 2)
976 {
977 pi2_tmp[k * trans_size] = (g_ai2_ihevc_trans_16[k][0] * o[0]
978 + g_ai2_ihevc_trans_16[k][1] * o[1]
979 + g_ai2_ihevc_trans_16[k][2] * o[2]
980 + g_ai2_ihevc_trans_16[k][3] * o[3]
981 + g_ai2_ihevc_trans_16[k][4] * o[4]
982 + g_ai2_ihevc_trans_16[k][5] * o[5]
983 + g_ai2_ihevc_trans_16[k][6] * o[6]
984 + g_ai2_ihevc_trans_16[k][7] * o[7] + add) >> shift;
985 }
986 pi2_src += src_strd;
987 pu1_pred += pred_strd;
988 pi2_tmp++;
989 }
990
991 pi2_tmp = pi2_tmp_orig;
992 /* Forward Transform 2nd stage */
993 shift = 10; // log2(iHeight) + 6
994 add = 1 << (shift - 1);
995
996 for(i = 0; i < TRANS_SIZE_16; i++)
997 {
998 /* e and o*/
999 for(k = 0; k < 8; k++)
1000 {
1001 e[k] = pi2_tmp[k] + pi2_tmp[15 - k];
1002 o[k] = pi2_tmp[k] - pi2_tmp[15 - k];
1003 }
1004 /* ee and eo */
1005 for(k = 0; k < 4; k++)
1006 {
1007 ee[k] = e[k] + e[7 - k];
1008 eo[k] = e[k] - e[7 - k];
1009 }
1010 /* eee and eeo */
1011 eee[0] = ee[0] + ee[3];
1012 eeo[0] = ee[0] - ee[3];
1013 eee[1] = ee[1] + ee[2];
1014 eeo[1] = ee[1] - ee[2];
1015
1016 pi2_dst[0] = (g_ai2_ihevc_trans_16[0][0] * eee[0]
1017 + g_ai2_ihevc_trans_16[0][1] * eee[1] + add) >> shift;
1018 pi2_dst[8 * dst_strd] = (g_ai2_ihevc_trans_16[8][0] * eee[0]
1019 + g_ai2_ihevc_trans_16[8][1] * eee[1] + add) >> shift;
1020 pi2_dst[4 * dst_strd] = (g_ai2_ihevc_trans_16[4][0] * eeo[0]
1021 + g_ai2_ihevc_trans_16[4][1] * eeo[1] + add) >> shift;
1022 pi2_dst[12 * dst_strd] = (g_ai2_ihevc_trans_16[12][0] * eeo[0]
1023 + g_ai2_ihevc_trans_16[12][1] * eeo[1] + add) >> shift;
1024
1025 for(k = 2; k < 16; k += 4)
1026 {
1027 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_16[k][0] * eo[0]
1028 + g_ai2_ihevc_trans_16[k][1] * eo[1]
1029 + g_ai2_ihevc_trans_16[k][2] * eo[2]
1030 + g_ai2_ihevc_trans_16[k][3] * eo[3] + add)
1031 >> shift;
1032 }
1033
1034 for(k = 1; k < 16; k += 2)
1035 {
1036 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_16[k][0] * o[0]
1037 + g_ai2_ihevc_trans_16[k][1] * o[1]
1038 + g_ai2_ihevc_trans_16[k][2] * o[2]
1039 + g_ai2_ihevc_trans_16[k][3] * o[3]
1040 + g_ai2_ihevc_trans_16[k][4] * o[4]
1041 + g_ai2_ihevc_trans_16[k][5] * o[5]
1042 + g_ai2_ihevc_trans_16[k][6] * o[6]
1043 + g_ai2_ihevc_trans_16[k][7] * o[7] + add) >> shift;
1044 }
1045
1046 pi2_tmp += trans_size;
1047 pi2_dst++;
1048 }
1049 }
1050
1051 /**
1052 *******************************************************************************
1053 *
1054 * @brief
1055 * This function performs residue calculation and forward transform on
1056 * input pixels
1057 *
1058 * @par Description:
1059 * Performs residue calculation by subtracting source and prediction and
1060 * followed by forward transform
1061 *
1062 * @param[in] pu1_src
1063 * Input 32x32 pixels
1064 *
1065 * @param[in] pu1_pred
1066 * Prediction data
1067 *
1068 * @param[in] pi2_tmp
1069 * Temporary buffer of size 32x32
1070 *
1071 * @param[out] pi2_dst
1072 * Output 32x32 coefficients
1073 *
1074 * @param[in] src_strd
1075 * Input stride
1076 *
1077 * @param[in] pred_strd
1078 * Prediction Stride
1079 *
1080 * @param[in] dst_strd
1081 * Output Stride
1082 *
1083 * @param[in] e_chroma_plane
1084 * Enum singalling chroma plane
1085 *
1086 * @returns Void
1087 *
1088 * @remarks
1089 * None
1090 *
1091 *******************************************************************************
1092 */
1093
ihevc_resi_trans_32x32(UWORD8 * pu1_src,UWORD8 * pu1_pred,WORD32 * pi4_temp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd,CHROMA_PLANE_ID_T e_chroma_plane)1094 UWORD32 ihevc_resi_trans_32x32(UWORD8 *pu1_src,
1095 UWORD8 *pu1_pred,
1096 WORD32 *pi4_temp,
1097 WORD16 *pi2_dst,
1098 WORD32 src_strd,
1099 WORD32 pred_strd,
1100 WORD32 dst_strd,
1101 CHROMA_PLANE_ID_T e_chroma_plane)
1102 {
1103 WORD32 i, k;
1104 WORD32 e[16], o[16];
1105 WORD32 ee[8], eo[8];
1106 WORD32 eee[4], eeo[4];
1107 WORD32 eeee[2], eeeo[2];
1108 WORD32 add, shift;
1109 WORD32 trans_size;
1110 WORD32 *pi4_tmp_orig;
1111 WORD16 *pi2_dst_orig;
1112 UWORD32 u4_blk_sad = 0 ;
1113 UNUSED(e_chroma_plane);
1114
1115 pi2_dst_orig = pi2_dst;
1116 pi4_tmp_orig = pi4_temp;
1117 trans_size = TRANS_SIZE_32;
1118 /* Residue + Forward Transform 1st stage */
1119 /* Made to zero to match with intrinsics */
1120 shift = 0; // 4 : log2(iWidth) - 1 + g_uiBitIncrement
1121 add = 0 ; //1 << (shift - 1);
1122
1123 for(i = 0; i < trans_size; i++)
1124 {
1125 WORD32 resi_tmp_1, resi_tmp_2;
1126 /* e and o*/
1127 for(k = 0; k < 16; k++)
1128 {
1129 resi_tmp_1 = pu1_src[k] - pu1_pred[k];
1130 resi_tmp_2 = pu1_src[31 - k] - pu1_pred[31 - k];
1131 e[k] = resi_tmp_1 + resi_tmp_2;
1132 o[k] = resi_tmp_1 - resi_tmp_2;
1133 u4_blk_sad += abs(resi_tmp_1) + abs(resi_tmp_2);
1134 }
1135 /* ee and eo */
1136 for(k = 0; k < 8; k++)
1137 {
1138 ee[k] = e[k] + e[15 - k];
1139 eo[k] = e[k] - e[15 - k];
1140 }
1141 /* eee and eeo */
1142 for(k = 0; k < 4; k++)
1143 {
1144 eee[k] = ee[k] + ee[7 - k];
1145 eeo[k] = ee[k] - ee[7 - k];
1146 }
1147 /* eeee and eeeo */
1148 eeee[0] = eee[0] + eee[3];
1149 eeeo[0] = eee[0] - eee[3];
1150 eeee[1] = eee[1] + eee[2];
1151 eeeo[1] = eee[1] - eee[2];
1152
1153 pi4_temp[0] = (g_ai2_ihevc_trans_32[0][0] * eeee[0]
1154 + g_ai2_ihevc_trans_32[0][1] * eeee[1]);// + add) >> shift;
1155 pi4_temp[16 * trans_size] = (g_ai2_ihevc_trans_32[16][0] * eeee[0]
1156 + g_ai2_ihevc_trans_32[16][1] * eeee[1]);// + add) >> shift;
1157 pi4_temp[8 * trans_size] = (g_ai2_ihevc_trans_32[8][0] * eeeo[0]
1158 + g_ai2_ihevc_trans_32[8][1] * eeeo[1]);// + add) >> shift;
1159 pi4_temp[24 * trans_size] = (g_ai2_ihevc_trans_32[24][0] * eeeo[0]
1160 + g_ai2_ihevc_trans_32[24][1] * eeeo[1]);// + add) >> shift;
1161 for(k = 4; k < 32; k += 8)
1162 {
1163 pi4_temp[k * trans_size] = (g_ai2_ihevc_trans_32[k][0] * eeo[0]
1164 + g_ai2_ihevc_trans_32[k][1] * eeo[1]
1165 + g_ai2_ihevc_trans_32[k][2] * eeo[2]
1166 + g_ai2_ihevc_trans_32[k][3] * eeo[3]);// + add)>> shift;
1167 }
1168 for(k = 2; k < 32; k += 4)
1169 {
1170 pi4_temp[k * trans_size] = (g_ai2_ihevc_trans_32[k][0] * eo[0]
1171 + g_ai2_ihevc_trans_32[k][1] * eo[1]
1172 + g_ai2_ihevc_trans_32[k][2] * eo[2]
1173 + g_ai2_ihevc_trans_32[k][3] * eo[3]
1174 + g_ai2_ihevc_trans_32[k][4] * eo[4]
1175 + g_ai2_ihevc_trans_32[k][5] * eo[5]
1176 + g_ai2_ihevc_trans_32[k][6] * eo[6]
1177 + g_ai2_ihevc_trans_32[k][7] * eo[7]);// + add)>> shift;
1178 }
1179 for(k = 1; k < 32; k += 2)
1180 {
1181 pi4_temp[k * trans_size] = (g_ai2_ihevc_trans_32[k][0] * o[0]
1182 + g_ai2_ihevc_trans_32[k][1] * o[1]
1183 + g_ai2_ihevc_trans_32[k][2] * o[2]
1184 + g_ai2_ihevc_trans_32[k][3] * o[3]
1185 + g_ai2_ihevc_trans_32[k][4] * o[4]
1186 + g_ai2_ihevc_trans_32[k][5] * o[5]
1187 + g_ai2_ihevc_trans_32[k][6] * o[6]
1188 + g_ai2_ihevc_trans_32[k][7] * o[7]
1189 + g_ai2_ihevc_trans_32[k][8] * o[8]
1190 + g_ai2_ihevc_trans_32[k][9] * o[9]
1191 + g_ai2_ihevc_trans_32[k][10] * o[10]
1192 + g_ai2_ihevc_trans_32[k][11] * o[11]
1193 + g_ai2_ihevc_trans_32[k][12] * o[12]
1194 + g_ai2_ihevc_trans_32[k][13] * o[13]
1195 + g_ai2_ihevc_trans_32[k][14] * o[14]
1196 + g_ai2_ihevc_trans_32[k][15] * o[15]);// + add) >> shift;
1197 }
1198 pu1_src += src_strd;
1199 pu1_pred += pred_strd;
1200 pi4_temp++;
1201 }
1202
1203 pi4_temp = pi4_tmp_orig;
1204 /* Forward Transform 2nd stage */
1205 shift = 15; // log2(iHeight) + 6
1206 add = 1 << (shift - 1);
1207
1208 for(i = 0; i < TRANS_SIZE_32; i++)
1209 {
1210 /* e and o*/
1211 for(k = 0; k < 16; k++)
1212 {
1213 e[k] = pi4_temp[k] + pi4_temp[31 - k];
1214 o[k] = pi4_temp[k] - pi4_temp[31 - k];
1215 }
1216 /* ee and eo */
1217 for(k = 0; k < 8; k++)
1218 {
1219 ee[k] = e[k] + e[15 - k];
1220 eo[k] = e[k] - e[15 - k];
1221 }
1222 /* eee and eeo */
1223 for(k = 0; k < 4; k++)
1224 {
1225 eee[k] = ee[k] + ee[7 - k];
1226 eeo[k] = ee[k] - ee[7 - k];
1227 }
1228 /* eeee and eeeo */
1229 eeee[0] = eee[0] + eee[3];
1230 eeeo[0] = eee[0] - eee[3];
1231 eeee[1] = eee[1] + eee[2];
1232 eeeo[1] = eee[1] - eee[2];
1233
1234 pi2_dst[0] = (g_ai2_ihevc_trans_32[0][0] * eeee[0]
1235 + g_ai2_ihevc_trans_32[0][1] * eeee[1] + add) >> shift;
1236 pi2_dst[16 * dst_strd] = (g_ai2_ihevc_trans_32[16][0] * eeee[0]
1237 + g_ai2_ihevc_trans_32[16][1] * eeee[1] + add) >> shift;
1238 pi2_dst[8 * dst_strd] = (g_ai2_ihevc_trans_32[8][0] * eeeo[0]
1239 + g_ai2_ihevc_trans_32[8][1] * eeeo[1] + add) >> shift;
1240 pi2_dst[24 * dst_strd] = (g_ai2_ihevc_trans_32[24][0] * eeeo[0]
1241 + g_ai2_ihevc_trans_32[24][1] * eeeo[1] + add) >> shift;
1242 for(k = 4; k < 32; k += 8)
1243 {
1244 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_32[k][0] * eeo[0]
1245 + g_ai2_ihevc_trans_32[k][1] * eeo[1]
1246 + g_ai2_ihevc_trans_32[k][2] * eeo[2]
1247 + g_ai2_ihevc_trans_32[k][3] * eeo[3] + add)
1248 >> shift;
1249 }
1250 for(k = 2; k < 32; k += 4)
1251 {
1252 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_32[k][0] * eo[0]
1253 + g_ai2_ihevc_trans_32[k][1] * eo[1]
1254 + g_ai2_ihevc_trans_32[k][2] * eo[2]
1255 + g_ai2_ihevc_trans_32[k][3] * eo[3]
1256 + g_ai2_ihevc_trans_32[k][4] * eo[4]
1257 + g_ai2_ihevc_trans_32[k][5] * eo[5]
1258 + g_ai2_ihevc_trans_32[k][6] * eo[6]
1259 + g_ai2_ihevc_trans_32[k][7] * eo[7] + add)
1260 >> shift;
1261 }
1262 for(k = 1; k < 32; k += 2)
1263 {
1264 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_32[k][0] * o[0]
1265 + g_ai2_ihevc_trans_32[k][1] * o[1]
1266 + g_ai2_ihevc_trans_32[k][2] * o[2]
1267 + g_ai2_ihevc_trans_32[k][3] * o[3]
1268 + g_ai2_ihevc_trans_32[k][4] * o[4]
1269 + g_ai2_ihevc_trans_32[k][5] * o[5]
1270 + g_ai2_ihevc_trans_32[k][6] * o[6]
1271 + g_ai2_ihevc_trans_32[k][7] * o[7]
1272 + g_ai2_ihevc_trans_32[k][8] * o[8]
1273 + g_ai2_ihevc_trans_32[k][9] * o[9]
1274 + g_ai2_ihevc_trans_32[k][10] * o[10]
1275 + g_ai2_ihevc_trans_32[k][11] * o[11]
1276 + g_ai2_ihevc_trans_32[k][12] * o[12]
1277 + g_ai2_ihevc_trans_32[k][13] * o[13]
1278 + g_ai2_ihevc_trans_32[k][14] * o[14]
1279 + g_ai2_ihevc_trans_32[k][15] * o[15] + add)
1280 >> shift;
1281 }
1282
1283 pi4_temp += trans_size;
1284 pi2_dst++;
1285 }
1286
1287 return u4_blk_sad;
1288 }
1289
1290
1291
ihevc_resi_trans_32x32_16bit(WORD16 * pi2_src,UWORD8 * pu1_pred,WORD16 * pi2_tmp,WORD16 * pi2_dst,WORD32 src_strd,WORD32 pred_strd,WORD32 dst_strd)1292 void ihevc_resi_trans_32x32_16bit(WORD16 *pi2_src,
1293 UWORD8 *pu1_pred,
1294 WORD16 *pi2_tmp,
1295 WORD16 *pi2_dst,
1296 WORD32 src_strd,
1297 WORD32 pred_strd,
1298 WORD32 dst_strd)
1299 {
1300 WORD32 i, k;
1301 WORD32 e[16], o[16];
1302 WORD32 ee[8], eo[8];
1303 WORD32 eee[4], eeo[4];
1304 WORD32 eeee[2], eeeo[2];
1305 WORD32 add, shift;
1306 WORD32 trans_size;
1307 WORD16 *pi2_tmp_orig;
1308 WORD16 *pi2_dst_orig;
1309
1310 pi2_dst_orig = pi2_dst;
1311 pi2_tmp_orig = pi2_tmp;
1312 trans_size = TRANS_SIZE_32;
1313 /* Residue + Forward Transform 1st stage */
1314 shift = 4; // log2(iWidth) - 1 + g_uiBitIncrement
1315 add = 1 << (shift - 1);
1316
1317 for(i = 0; i < trans_size; i++)
1318 {
1319 WORD32 resi_tmp_1, resi_tmp_2;
1320 /* e and o*/
1321 for(k = 0; k < 16; k++)
1322 {
1323 resi_tmp_1 = pi2_src[k] - pu1_pred[k];
1324 resi_tmp_2 = pi2_src[31 - k] - pu1_pred[31 - k];
1325 e[k] = resi_tmp_1 + resi_tmp_2;
1326 o[k] = resi_tmp_1 - resi_tmp_2;
1327 }
1328 /* ee and eo */
1329 for(k = 0; k < 8; k++)
1330 {
1331 ee[k] = e[k] + e[15 - k];
1332 eo[k] = e[k] - e[15 - k];
1333 }
1334 /* eee and eeo */
1335 for(k = 0; k < 4; k++)
1336 {
1337 eee[k] = ee[k] + ee[7 - k];
1338 eeo[k] = ee[k] - ee[7 - k];
1339 }
1340 /* eeee and eeeo */
1341 eeee[0] = eee[0] + eee[3];
1342 eeeo[0] = eee[0] - eee[3];
1343 eeee[1] = eee[1] + eee[2];
1344 eeeo[1] = eee[1] - eee[2];
1345
1346 pi2_tmp[0] = (g_ai2_ihevc_trans_32[0][0] * eeee[0]
1347 + g_ai2_ihevc_trans_32[0][1] * eeee[1] + add) >> shift;
1348 pi2_tmp[16 * trans_size] = (g_ai2_ihevc_trans_32[16][0] * eeee[0]
1349 + g_ai2_ihevc_trans_32[16][1] * eeee[1] + add) >> shift;
1350 pi2_tmp[8 * trans_size] = (g_ai2_ihevc_trans_32[8][0] * eeeo[0]
1351 + g_ai2_ihevc_trans_32[8][1] * eeeo[1] + add) >> shift;
1352 pi2_tmp[24 * trans_size] = (g_ai2_ihevc_trans_32[24][0] * eeeo[0]
1353 + g_ai2_ihevc_trans_32[24][1] * eeeo[1] + add) >> shift;
1354 for(k = 4; k < 32; k += 8)
1355 {
1356 pi2_tmp[k * trans_size] = (g_ai2_ihevc_trans_32[k][0] * eeo[0]
1357 + g_ai2_ihevc_trans_32[k][1] * eeo[1]
1358 + g_ai2_ihevc_trans_32[k][2] * eeo[2]
1359 + g_ai2_ihevc_trans_32[k][3] * eeo[3] + add)
1360 >> shift;
1361 }
1362 for(k = 2; k < 32; k += 4)
1363 {
1364 pi2_tmp[k * trans_size] = (g_ai2_ihevc_trans_32[k][0] * eo[0]
1365 + g_ai2_ihevc_trans_32[k][1] * eo[1]
1366 + g_ai2_ihevc_trans_32[k][2] * eo[2]
1367 + g_ai2_ihevc_trans_32[k][3] * eo[3]
1368 + g_ai2_ihevc_trans_32[k][4] * eo[4]
1369 + g_ai2_ihevc_trans_32[k][5] * eo[5]
1370 + g_ai2_ihevc_trans_32[k][6] * eo[6]
1371 + g_ai2_ihevc_trans_32[k][7] * eo[7] + add)
1372 >> shift;
1373 }
1374 for(k = 1; k < 32; k += 2)
1375 {
1376 pi2_tmp[k * trans_size] = (g_ai2_ihevc_trans_32[k][0] * o[0]
1377 + g_ai2_ihevc_trans_32[k][1] * o[1]
1378 + g_ai2_ihevc_trans_32[k][2] * o[2]
1379 + g_ai2_ihevc_trans_32[k][3] * o[3]
1380 + g_ai2_ihevc_trans_32[k][4] * o[4]
1381 + g_ai2_ihevc_trans_32[k][5] * o[5]
1382 + g_ai2_ihevc_trans_32[k][6] * o[6]
1383 + g_ai2_ihevc_trans_32[k][7] * o[7]
1384 + g_ai2_ihevc_trans_32[k][8] * o[8]
1385 + g_ai2_ihevc_trans_32[k][9] * o[9]
1386 + g_ai2_ihevc_trans_32[k][10] * o[10]
1387 + g_ai2_ihevc_trans_32[k][11] * o[11]
1388 + g_ai2_ihevc_trans_32[k][12] * o[12]
1389 + g_ai2_ihevc_trans_32[k][13] * o[13]
1390 + g_ai2_ihevc_trans_32[k][14] * o[14]
1391 + g_ai2_ihevc_trans_32[k][15] * o[15] + add)
1392 >> shift;
1393 }
1394 pi2_src += src_strd;
1395 pu1_pred += pred_strd;
1396 pi2_tmp++;
1397 }
1398
1399 pi2_tmp = pi2_tmp_orig;
1400 /* Forward Transform 2nd stage */
1401 shift = 11; // log2(iHeight) + 6
1402 add = 1 << (shift - 1);
1403
1404 for(i = 0; i < TRANS_SIZE_32; i++)
1405 {
1406 /* e and o*/
1407 for(k = 0; k < 16; k++)
1408 {
1409 e[k] = pi2_tmp[k] + pi2_tmp[31 - k];
1410 o[k] = pi2_tmp[k] - pi2_tmp[31 - k];
1411 }
1412 /* ee and eo */
1413 for(k = 0; k < 8; k++)
1414 {
1415 ee[k] = e[k] + e[15 - k];
1416 eo[k] = e[k] - e[15 - k];
1417 }
1418 /* eee and eeo */
1419 for(k = 0; k < 4; k++)
1420 {
1421 eee[k] = ee[k] + ee[7 - k];
1422 eeo[k] = ee[k] - ee[7 - k];
1423 }
1424 /* eeee and eeeo */
1425 eeee[0] = eee[0] + eee[3];
1426 eeeo[0] = eee[0] - eee[3];
1427 eeee[1] = eee[1] + eee[2];
1428 eeeo[1] = eee[1] - eee[2];
1429
1430 pi2_dst[0] = (g_ai2_ihevc_trans_32[0][0] * eeee[0]
1431 + g_ai2_ihevc_trans_32[0][1] * eeee[1] + add) >> shift;
1432 pi2_dst[16 * dst_strd] = (g_ai2_ihevc_trans_32[16][0] * eeee[0]
1433 + g_ai2_ihevc_trans_32[16][1] * eeee[1] + add) >> shift;
1434 pi2_dst[8 * dst_strd] = (g_ai2_ihevc_trans_32[8][0] * eeeo[0]
1435 + g_ai2_ihevc_trans_32[8][1] * eeeo[1] + add) >> shift;
1436 pi2_dst[24 * dst_strd] = (g_ai2_ihevc_trans_32[24][0] * eeeo[0]
1437 + g_ai2_ihevc_trans_32[24][1] * eeeo[1] + add) >> shift;
1438 for(k = 4; k < 32; k += 8)
1439 {
1440 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_32[k][0] * eeo[0]
1441 + g_ai2_ihevc_trans_32[k][1] * eeo[1]
1442 + g_ai2_ihevc_trans_32[k][2] * eeo[2]
1443 + g_ai2_ihevc_trans_32[k][3] * eeo[3] + add)
1444 >> shift;
1445 }
1446 for(k = 2; k < 32; k += 4)
1447 {
1448 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_32[k][0] * eo[0]
1449 + g_ai2_ihevc_trans_32[k][1] * eo[1]
1450 + g_ai2_ihevc_trans_32[k][2] * eo[2]
1451 + g_ai2_ihevc_trans_32[k][3] * eo[3]
1452 + g_ai2_ihevc_trans_32[k][4] * eo[4]
1453 + g_ai2_ihevc_trans_32[k][5] * eo[5]
1454 + g_ai2_ihevc_trans_32[k][6] * eo[6]
1455 + g_ai2_ihevc_trans_32[k][7] * eo[7] + add)
1456 >> shift;
1457 }
1458 for(k = 1; k < 32; k += 2)
1459 {
1460 pi2_dst[k * dst_strd] = (g_ai2_ihevc_trans_32[k][0] * o[0]
1461 + g_ai2_ihevc_trans_32[k][1] * o[1]
1462 + g_ai2_ihevc_trans_32[k][2] * o[2]
1463 + g_ai2_ihevc_trans_32[k][3] * o[3]
1464 + g_ai2_ihevc_trans_32[k][4] * o[4]
1465 + g_ai2_ihevc_trans_32[k][5] * o[5]
1466 + g_ai2_ihevc_trans_32[k][6] * o[6]
1467 + g_ai2_ihevc_trans_32[k][7] * o[7]
1468 + g_ai2_ihevc_trans_32[k][8] * o[8]
1469 + g_ai2_ihevc_trans_32[k][9] * o[9]
1470 + g_ai2_ihevc_trans_32[k][10] * o[10]
1471 + g_ai2_ihevc_trans_32[k][11] * o[11]
1472 + g_ai2_ihevc_trans_32[k][12] * o[12]
1473 + g_ai2_ihevc_trans_32[k][13] * o[13]
1474 + g_ai2_ihevc_trans_32[k][14] * o[14]
1475 + g_ai2_ihevc_trans_32[k][15] * o[15] + add)
1476 >> shift;
1477 }
1478
1479 pi2_tmp += trans_size;
1480 pi2_dst++;
1481 }
1482 }
1483
1484