1 /******************************************************************************
2 *
3 * Copyright (C) 2012 Ittiam Systems Pvt Ltd, Bangalore
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 /**
19 *******************************************************************************
20 * @file
21 * ihevc_itrans.c
22 *
23 * @brief
24 * Contains function definitions for single stage inverse transform
25 *
26 * @author
27 * 100470
28 *
29 * @par List of Functions:
30 * - ihevc_itrans_4x4_ttype1()
31 * - ihevc_itrans_4x4()
32 * - ihevc_itrans_8x8()
33 * - ihevc_itrans_16x16()
34 * - ihevc_itrans_32x32()
35 *
36 * @remarks
37 * None
38 *
39 *******************************************************************************
40 */
41 #include <stdio.h>
42 #include <string.h>
43 #include "ihevc_typedefs.h"
44 #include "ihevc_macros.h"
45 #include "ihevc_platform_macros.h"
46 #include "ihevc_defs.h"
47 #include "ihevc_trans_tables.h"
48 #include "ihevc_func_selector.h"
49 #include "ihevc_trans_macros.h"
50
51 #define NON_OPTIMIZED 1
52
53 /**
54 *******************************************************************************
55 *
56 * @brief
57 * This function performs Single stage Inverse transform type 1 (DST) for
58 * 4x4 input block
59 *
60 * @par Description:
61 * Performs single stage 4x4 inverse transform type 1 by utilizing the
62 * symmetry of transformation matrix and reducing number of multiplications
63 * wherever possible but keeping the number of operations
64 * (addition,multiplication and shift)same
65 *
66 * @param[in] pi2_src
67 * Input 4x4 coefficients
68 *
69 * @param[out] pi2_dst
70 * Output 4x4 block
71 *
72 * @param[in] src_strd
73 * Input stride
74 *
75 * @param[in] dst_strd
76 * Output Stride
77 *
78 * @param[in] i4_shift
79 * Output shift
80 *
81 * @param[in] zero_cols
82 * Zero columns in pi2_src
83 *
84 * @returns Void
85 *
86 * @remarks
87 * None
88 *
89 *******************************************************************************
90 */
91
92
ihevc_itrans_4x4_ttype1(WORD16 * pi2_src,WORD16 * pi2_dst,WORD32 src_strd,WORD32 dst_strd,WORD32 i4_shift,WORD32 zero_cols)93 void ihevc_itrans_4x4_ttype1(WORD16 *pi2_src,
94 WORD16 *pi2_dst,
95 WORD32 src_strd,
96 WORD32 dst_strd,
97 WORD32 i4_shift,
98 WORD32 zero_cols)
99 {
100 WORD32 i, c[4];
101 WORD32 add;
102
103 add = 1 << (i4_shift - 1);
104
105 for(i = 0; i < TRANS_SIZE_4; i++)
106 {
107 /* Checking for Zero Cols */
108 if((zero_cols & 1) == 1)
109 {
110 memset(pi2_dst, 0, TRANS_SIZE_4 * sizeof(WORD16));
111 }
112 else
113 {
114 // Intermediate Variables
115 c[0] = pi2_src[0] + pi2_src[2 * src_strd];
116 c[1] = pi2_src[2 * src_strd] + pi2_src[3 * src_strd];
117 c[2] = pi2_src[0] - pi2_src[3 * src_strd];
118 c[3] = 74 * pi2_src[src_strd];
119
120 pi2_dst[0] =
121 CLIP_S16((29 * c[0] + 55 * c[1] + c[3] + add) >> i4_shift);
122 pi2_dst[1] =
123 CLIP_S16((55 * c[2] - 29 * c[1] + c[3] + add) >> i4_shift);
124 pi2_dst[2] =
125 CLIP_S16((74 * (pi2_src[0] - pi2_src[2 * src_strd] + pi2_src[3 * src_strd]) + add) >> i4_shift);
126 pi2_dst[3] =
127 CLIP_S16((55 * c[0] + 29 * c[2] - c[3] + add) >> i4_shift);
128 }
129 pi2_src++;
130 pi2_dst += dst_strd;
131 zero_cols = zero_cols >> 1;
132 }
133 }
134
135
136 /**
137 *******************************************************************************
138 *
139 * @brief
140 * This function performs Single stage Inverse transform for 4x4 input
141 * block
142 *
143 * @par Description:
144 * Performs single stage 4x4 inverse transform by utilizing the symmetry of
145 * transformation matrix and reducing number of multiplications wherever
146 * possible but keeping the number of operations(addition,multiplication and
147 * shift) same
148 *
149 * @param[in] pi2_src
150 * Input 4x4 coefficients
151 *
152 * @param[out] pi2_dst
153 * Output 4x4 block
154 *
155 * @param[in] src_strd
156 * Input stride
157 *
158 * @param[in] dst_strd
159 * Output Stride
160 *
161 * @param[in] i4_shift
162 * Output shift
163 *
164 * @param[in] zero_cols
165 * Zero columns in pi2_src
166 *
167 * @returns Void
168 *
169 * @remarks
170 * None
171 *
172 *******************************************************************************
173 */
174
175 #if NON_OPTIMIZED
ihevc_itrans_4x4(WORD16 * pi2_src,WORD16 * pi2_dst,WORD32 src_strd,WORD32 dst_strd,WORD32 i4_shift,WORD32 zero_cols)176 void ihevc_itrans_4x4(WORD16 *pi2_src,
177 WORD16 *pi2_dst,
178 WORD32 src_strd,
179 WORD32 dst_strd,
180 WORD32 i4_shift,
181 WORD32 zero_cols)
182 {
183 WORD32 j;
184 WORD32 e[2], o[2];
185 WORD32 add;
186
187 add = 1 << (i4_shift - 1);
188
189 for(j = 0; j < TRANS_SIZE_4; j++)
190 {
191 /* Checking for Zero Cols */
192 if((zero_cols & 1) == 1)
193 {
194 memset(pi2_dst, 0, TRANS_SIZE_4 * sizeof(WORD16));
195 }
196 else
197 {
198
199 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
200 o[0] = g_ai2_ihevc_trans_4[1][0] * pi2_src[src_strd]
201 + g_ai2_ihevc_trans_4[3][0] * pi2_src[3 * src_strd];
202 o[1] = g_ai2_ihevc_trans_4[1][1] * pi2_src[src_strd]
203 + g_ai2_ihevc_trans_4[3][1] * pi2_src[3 * src_strd];
204 e[0] = g_ai2_ihevc_trans_4[0][0] * pi2_src[0]
205 + g_ai2_ihevc_trans_4[2][0] * pi2_src[2 * src_strd];
206 e[1] = g_ai2_ihevc_trans_4[0][1] * pi2_src[0]
207 + g_ai2_ihevc_trans_4[2][1] * pi2_src[2 * src_strd];
208
209 pi2_dst[0] =
210 CLIP_S16(((e[0] + o[0] + add) >> i4_shift));
211 pi2_dst[1] =
212 CLIP_S16(((e[1] + o[1] + add) >> i4_shift));
213 pi2_dst[2] =
214 CLIP_S16(((e[1] - o[1] + add) >> i4_shift));
215 pi2_dst[3] =
216 CLIP_S16(((e[0] - o[0] + add) >> i4_shift));
217
218 }
219 pi2_src++;
220 pi2_dst += dst_strd;
221 zero_cols = zero_cols >> 1;
222 }
223 }
224 #else
ihevc_itrans_4x4(WORD16 * pi2_src,WORD16 * pi2_dst,WORD32 src_strd,WORD32 dst_strd,WORD32 i4_shift,WORD32 zero_cols)225 void ihevc_itrans_4x4(WORD16 *pi2_src,
226 WORD16 *pi2_dst,
227 WORD32 src_strd,
228 WORD32 dst_strd,
229 WORD32 i4_shift,
230 WORD32 zero_cols)
231 {
232 WORD32 j;
233 WORD32 e[2], o[2];
234 WORD32 add;
235
236 add = 1 << (i4_shift - 1);
237
238 /***************************************************************************/
239 /* Transform Matrix 4x4 */
240 /* 0 1 2 3 */
241 /* 0 { 64, 64, 64, 64}, */
242 /* 1 { 83, 36,-36,-83}, */
243 /* 2 { 64,-64,-64, 64}, */
244 /* 3 { 36,-83, 83,-36} */
245 /***************************************************************************/
246
247 for(j = 0; j < TRANS_SIZE_4; j++)
248 {
249 WORD32 temp;
250
251 /* Checking for Zero Cols */
252 if((zero_cols & 1) == 1)
253 {
254 memset(pi2_dst, 0, TRANS_SIZE_4 * sizeof(WORD16));
255 }
256 else
257 {
258 /* Common operation in o[0] and o[1] */
259 temp = (pi2_src[src_strd] + pi2_src[3 * src_strd]) * 36;
260
261 o[0] = temp + 47 * pi2_src[src_strd];
262 o[1] = temp - 119 * pi2_src[3 * src_strd];
263 e[0] = (pi2_src[0] + pi2_src[2 * src_strd]) << 6;
264 e[1] = (pi2_src[0] - pi2_src[2 * src_strd]) << 6;
265
266 pi2_dst[0] =
267 CLIP_S16(((e[0] + o[0] + add) >> i4_shift));
268 pi2_dst[1] =
269 CLIP_S16(((e[1] + o[1] + add) >> i4_shift));
270 pi2_dst[2] =
271 CLIP_S16(((e[1] - o[1] + add) >> i4_shift));
272 pi2_dst[3] =
273 CLIP_S16(((e[0] - o[0] + add) >> i4_shift));
274 }
275 pi2_src++;
276 pi2_dst += dst_strd;
277 zero_cols = zero_cols >> 1;
278 }
279 }
280 #endif
281
282 /**
283 *******************************************************************************
284 *
285 * @brief
286 * This function performs Single stage Inverse transform for 8x8 input
287 * block
288 *
289 * @par Description:
290 * Performs single stage 8x8 inverse transform by utilizing the symmetry of
291 * transformation matrix and reducing number of multiplications wherever
292 * possible but keeping the number of operations(addition,multiplication and
293 * shift) same
294 *
295 * @param[in] pi2_src
296 * Input 8x8 coefficients
297 *
298 * @param[out] pi2_dst
299 * Output 8x8 block
300 *
301 * @param[in] src_strd
302 * Input stride
303 *
304 * @param[in] dst_strd
305 * Output Stride
306 *
307 * @param[in] i4_shift
308 * Output shift
309 *
310 * @param[in] zero_cols
311 * Zero columns in pi2_src
312 *
313 * @returns Void
314 *
315 * @remarks
316 * None
317 *
318 *******************************************************************************
319 */
320
321 #if NON_OPTIMIZED
ihevc_itrans_8x8(WORD16 * pi2_src,WORD16 * pi2_dst,WORD32 src_strd,WORD32 dst_strd,WORD32 i4_shift,WORD32 zero_cols)322 void ihevc_itrans_8x8(WORD16 *pi2_src,
323 WORD16 *pi2_dst,
324 WORD32 src_strd,
325 WORD32 dst_strd,
326 WORD32 i4_shift,
327 WORD32 zero_cols)
328 {
329 WORD32 j, k;
330 WORD32 e[4], o[4];
331 WORD32 ee[2], eo[2];
332 WORD32 add;
333
334 add = 1 << (i4_shift - 1);
335
336 for(j = 0; j < TRANS_SIZE_8; j++)
337 {
338 /* Checking for Zero Cols */
339 if((zero_cols & 1) == 1)
340 {
341 memset(pi2_dst, 0, TRANS_SIZE_8 * sizeof(WORD16));
342 }
343 else
344 {
345 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
346 for(k = 0; k < 4; k++)
347 {
348 o[k] = g_ai2_ihevc_trans_8[1][k] * pi2_src[src_strd]
349 + g_ai2_ihevc_trans_8[3][k]
350 * pi2_src[3 * src_strd]
351 + g_ai2_ihevc_trans_8[5][k]
352 * pi2_src[5 * src_strd]
353 + g_ai2_ihevc_trans_8[7][k]
354 * pi2_src[7 * src_strd];
355 }
356
357 eo[0] = g_ai2_ihevc_trans_8[2][0] * pi2_src[2 * src_strd]
358 + g_ai2_ihevc_trans_8[6][0] * pi2_src[6 * src_strd];
359 eo[1] = g_ai2_ihevc_trans_8[2][1] * pi2_src[2 * src_strd]
360 + g_ai2_ihevc_trans_8[6][1] * pi2_src[6 * src_strd];
361 ee[0] = g_ai2_ihevc_trans_8[0][0] * pi2_src[0]
362 + g_ai2_ihevc_trans_8[4][0] * pi2_src[4 * src_strd];
363 ee[1] = g_ai2_ihevc_trans_8[0][1] * pi2_src[0]
364 + g_ai2_ihevc_trans_8[4][1] * pi2_src[4 * src_strd];
365
366 /* Combining e and o terms at each hierarchy levels to calculate the final spatial domain vector */
367 e[0] = ee[0] + eo[0];
368 e[3] = ee[0] - eo[0];
369 e[1] = ee[1] + eo[1];
370 e[2] = ee[1] - eo[1];
371 for(k = 0; k < 4; k++)
372 {
373 pi2_dst[k] =
374 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
375 pi2_dst[k + 4] =
376 CLIP_S16(((e[3 - k] - o[3 - k] + add) >> i4_shift));
377 }
378 }
379 pi2_src++;
380 pi2_dst += dst_strd;
381 zero_cols = zero_cols >> 1;
382 }
383 }
384
385 #else
ihevc_itrans_8x8(WORD16 * pi2_src,WORD16 * pi2_dst,WORD32 src_strd,WORD32 dst_strd,WORD32 i4_shift,WORD32 zero_cols)386 void ihevc_itrans_8x8(WORD16 *pi2_src,
387 WORD16 *pi2_dst,
388 WORD32 src_strd,
389 WORD32 dst_strd,
390 WORD32 i4_shift,
391 WORD32 zero_cols)
392 {
393 /* Transform Matrix 8x8 */
394 /* 0 1 2 3 4 5 6 7 */
395 /* 0 - 64 64 64 64 64 64 64 64 */
396 /* 1 - 89 75 50 18 -18 -50 -75 -89 */
397 /* 2 - 83 36 -36 -83 -83 -36 36 83 */
398 /* 3 - 75 -18 -89 -50 50 89 18 -75 */
399 /* 4 - 64 -64 -64 64 64 -64 -64 64 */
400 /* 5 - 50 -89 18 75 -75 -18 89 -50 */
401 /* 6 - 36 -83 83 -36 -36 83 -83 36 */
402 /* 7 - 18 -50 75 -89 89 -75 50 -18 */
403
404 /* 0th and 4th row will have no multiplications */
405 /* 2nd and 6th row has only two coefff multiplies */
406 /* 1st, 3rd, 5th and 7th rows have o mirror symmetry */
407 WORD32 j, k;
408 WORD32 temp1, temp2;
409 WORD32 e[4], o[4];
410 WORD32 ee[2], eo[2];
411 WORD32 add;
412
413 add = 1 << (i4_shift - 1);
414
415 for(j = 0; j < TRANS_SIZE_8; j++)
416 {
417 /* Checking for Zero Cols */
418 if((zero_cols & 1) == 1)
419 {
420 memset(pi2_dst, 0, TRANS_SIZE_8 * sizeof(WORD16));
421 }
422 else
423 {
424
425 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
426 /*
427 o[0] = 89 *pi2_src[8] + 75 *pi2_src[3*8] + 50 *pi2_src[5*8] + 18 *pi2_src[7*8];
428 o[1] = 75 *pi2_src[8] + -18 *pi2_src[3*8] + -89 *pi2_src[5*8] + -50 *pi2_src[7*8];
429 o[2] = 50 *pi2_src[8] + -89 *pi2_src[3*8] + 18 *pi2_src[5*8] + 75 *pi2_src[7*8];
430 o[3] = 18 *pi2_src[8] + -50 *pi2_src[3*8] + 75 *pi2_src[5*8] + -89 *pi2_src[7*8];
431 */
432
433 /* Optimization: 4 mul + 2 add ---> 3 mul + 3 add */
434 /*
435 temp1 = (pi2_src[8 ] + pi2_src[3*8]) * 75;
436 temp2 = (pi2_src[5*8] + pi2_src[7*8]) * 50;
437
438 o[0] = temp1 + 14 * pi2_src[8 ] + temp2 - 32 * pi2_src[7*8];
439 o[1] = temp1 - 93 * pi2_src[3*8] - temp2 - 39 * pi2_src[5*8];
440 */
441
442 temp1 = (pi2_src[src_strd] + pi2_src[3 * src_strd]) * 75;
443 temp2 = (pi2_src[5 * src_strd] + pi2_src[7 * src_strd]) * 50;
444
445 o[0] = temp1 + 14 * pi2_src[src_strd] + temp2
446 - (pi2_src[7 * src_strd] << 5);
447 o[1] = temp1 - 93 * pi2_src[3 * src_strd] - temp2
448 - 39 * pi2_src[5 * src_strd];
449
450 /* Optimization: 4 mul + 2 add ---> 3 mul + 3 add */
451 /*
452 temp1 = (pi2_src[8 ] - pi2_src[3*8]) * 50;
453 temp2 = (pi2_src[5*8] + pi2_src[7*8]) * 75;
454
455 o[2] = temp1 - 39 * pi2_src[3*8] + temp2 - 57 * pi2_src[5*8];
456 o[3] = temp1 - 32 * pi2_src[8 ] + temp2 - 164 * pi2_src[7*8];
457 */
458
459 temp1 = (pi2_src[src_strd] - pi2_src[3 * src_strd]) * 50;
460 temp2 = (pi2_src[5 * src_strd] + pi2_src[7 * src_strd]) * 75;
461
462 o[2] = temp1 - 39 * pi2_src[3 * src_strd] + temp2
463 - 57 * pi2_src[5 * src_strd];
464 o[3] = temp1 - (pi2_src[src_strd] << 5) + temp2
465 - 164 * pi2_src[7 * src_strd];
466
467 /*
468 eo[0] = 83 *pi2_src[ 2*8 ] + 36 *pi2_src[ 6*8 ];
469 eo[1] = 36 *pi2_src[ 2*8 ] + -83 *pi2_src[ 6*8 ];
470 ee[0] = 64 *pi2_src[ 0 ] + 64 *pi2_src[ 4*8 ];
471 ee[1] = 64 *pi2_src[ 0 ] + -64 *pi2_src[ 4*8 ];
472 */
473
474 /* Optimization: 4 mul + 2 add ---> 3 mul + 3 add */
475 temp1 = (pi2_src[2 * src_strd] + pi2_src[6 * src_strd]) * 36;
476 eo[0] = temp1 + 47 * pi2_src[2 * src_strd];
477 eo[1] = temp1 - 119 * pi2_src[6 * src_strd];
478
479 /* Optimization: 4 mul + 2 add ---> 2 i4_shift + 2 add */
480 ee[0] = (pi2_src[0] + pi2_src[4 * src_strd]) << 6;
481 ee[1] = (pi2_src[0] - pi2_src[4 * src_strd]) << 6;
482
483 e[0] = ee[0] + eo[0];
484 e[3] = ee[0] - eo[0];
485 e[1] = ee[1] + eo[1];
486 e[2] = ee[1] - eo[1];
487
488 for(k = 0; k < 4; k++)
489 {
490 pi2_dst[k] =
491 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
492 pi2_dst[k + 4] =
493 CLIP_S16(((e[3 - k] - o[3 - k] + add) >> i4_shift));
494 }
495 }
496 pi2_src++;
497 pi2_dst += dst_strd;
498 zero_cols = zero_cols >> 1;
499 }
500
501 }
502 #endif
503
504
505 /**
506 *******************************************************************************
507 *
508 * @brief
509 * This function performs Single stage Inverse transform for 16x16 input
510 * block
511 *
512 * @par Description:
513 * Performs single stage 16x16 inverse transform by utilizing the symmetry
514 * of transformation matrix and reducing number of multiplications wherever
515 * possible but keeping the number of operations (addition,multiplication
516 * and shift) same
517 *
518 * @param[in] pi2_src
519 * Input 16x16 coefficients
520 *
521 * @param[out] pi2_dst
522 * Output 16x16 block
523 *
524 * @param[in] src_strd
525 * Input stride
526 *
527 * @param[in] dst_strd
528 * Output Stride
529 *
530 * @param[in] i4_shift
531 * Output shift
532 *
533 * @param[in] zero_cols
534 * Zero columns in pi2_src
535 *
536 * @returns Void
537 *
538 * @remarks
539 * None
540 *
541 *******************************************************************************
542 */
543
544 #if NON_OPTIMIZED
ihevc_itrans_16x16(WORD16 * pi2_src,WORD16 * pi2_dst,WORD32 src_strd,WORD32 dst_strd,WORD32 i4_shift,WORD32 zero_cols)545 void ihevc_itrans_16x16(WORD16 *pi2_src,
546 WORD16 *pi2_dst,
547 WORD32 src_strd,
548 WORD32 dst_strd,
549 WORD32 i4_shift,
550 WORD32 zero_cols)
551 {
552 WORD32 j, k;
553 WORD32 e[8], o[8];
554 WORD32 ee[4], eo[4];
555 WORD32 eee[2], eeo[2];
556 WORD32 add;
557
558 add = 1 << (i4_shift - 1);
559
560 for(j = 0; j < TRANS_SIZE_16; j++)
561 {
562 /* Checking for Zero Cols */
563 if((zero_cols & 1) == 1)
564 {
565 memset(pi2_dst, 0, TRANS_SIZE_16 * sizeof(WORD16));
566 }
567 else
568 {
569 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
570 for(k = 0; k < 8; k++)
571 {
572 o[k] = g_ai2_ihevc_trans_16[1][k] * pi2_src[src_strd]
573 + g_ai2_ihevc_trans_16[3][k]
574 * pi2_src[3 * src_strd]
575 + g_ai2_ihevc_trans_16[5][k]
576 * pi2_src[5 * src_strd]
577 + g_ai2_ihevc_trans_16[7][k]
578 * pi2_src[7 * src_strd]
579 + g_ai2_ihevc_trans_16[9][k]
580 * pi2_src[9 * src_strd]
581 + g_ai2_ihevc_trans_16[11][k]
582 * pi2_src[11 * src_strd]
583 + g_ai2_ihevc_trans_16[13][k]
584 * pi2_src[13 * src_strd]
585 + g_ai2_ihevc_trans_16[15][k]
586 * pi2_src[15 * src_strd];
587 }
588 for(k = 0; k < 4; k++)
589 {
590 eo[k] = g_ai2_ihevc_trans_16[2][k] * pi2_src[2 * src_strd]
591 + g_ai2_ihevc_trans_16[6][k]
592 * pi2_src[6 * src_strd]
593 + g_ai2_ihevc_trans_16[10][k]
594 * pi2_src[10 * src_strd]
595 + g_ai2_ihevc_trans_16[14][k]
596 * pi2_src[14 * src_strd];
597 }
598 eeo[0] = g_ai2_ihevc_trans_16[4][0] * pi2_src[4 * src_strd]
599 + g_ai2_ihevc_trans_16[12][0]
600 * pi2_src[12 * src_strd];
601 eee[0] =
602 g_ai2_ihevc_trans_16[0][0] * pi2_src[0]
603 + g_ai2_ihevc_trans_16[8][0]
604 * pi2_src[8
605 * src_strd];
606 eeo[1] = g_ai2_ihevc_trans_16[4][1] * pi2_src[4 * src_strd]
607 + g_ai2_ihevc_trans_16[12][1]
608 * pi2_src[12 * src_strd];
609 eee[1] =
610 g_ai2_ihevc_trans_16[0][1] * pi2_src[0]
611 + g_ai2_ihevc_trans_16[8][1]
612 * pi2_src[8
613 * src_strd];
614
615 /* Combining e and o terms at each hierarchy levels to calculate the final spatial domain vector */
616 for(k = 0; k < 2; k++)
617 {
618 ee[k] = eee[k] + eeo[k];
619 ee[k + 2] = eee[1 - k] - eeo[1 - k];
620 }
621 for(k = 0; k < 4; k++)
622 {
623 e[k] = ee[k] + eo[k];
624 e[k + 4] = ee[3 - k] - eo[3 - k];
625 }
626 for(k = 0; k < 8; k++)
627 {
628 pi2_dst[k] =
629 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
630 pi2_dst[k + 8] =
631 CLIP_S16(((e[7 - k] - o[7 - k] + add) >> i4_shift));
632 }
633 }
634 pi2_src++;
635 pi2_dst += dst_strd;
636 zero_cols = zero_cols >> 1;
637 }
638 }
639 #else
ihevc_itrans_16x16(WORD16 * pi2_src,WORD16 * pi2_dst,WORD32 src_strd,WORD32 dst_strd,WORD32 i4_shift,WORD32 zero_cols)640 void ihevc_itrans_16x16(WORD16 *pi2_src,
641 WORD16 *pi2_dst,
642 WORD32 src_strd,
643 WORD32 dst_strd,
644 WORD32 i4_shift,
645 WORD32 zero_cols)
646 {
647 WORD32 j, k;
648 WORD32 e[8], o[8];
649 WORD32 ee[4], eo[4];
650 WORD32 eee[2], eeo[2];
651 WORD32 add;
652 WORD32 temp1, temp2;
653
654 add = 1 << (i4_shift - 1);
655 /***************************************************************************/
656 /* Transform Matrix 16x16 */
657 /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 */
658 /* 0 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64}, */
659 /* 1 { 90, 87, 80, 70, 57, 43, 25, 9, -9,-25,-43,-57,-70,-80,-87,-90}, */
660 /* 2 { 89, 75, 50, 18,-18,-50,-75,-89,-89,-75,-50,-18, 18, 50, 75, 89}, */
661 /* 3 { 87, 57, 9,-43,-80,-90,-70,-25, 25, 70, 90, 80, 43, -9,-57,-87}, */
662 /* 4 { 83, 36,-36,-83,-83,-36, 36, 83, 83, 36,-36,-83,-83,-36, 36, 83}, */
663 /* 5 { 80, 9,-70,-87,-25, 57, 90, 43,-43,-90,-57, 25, 87, 70, -9,-80}, */
664 /* 6 { 75,-18,-89,-50, 50, 89, 18,-75,-75, 18, 89, 50,-50,-89,-18, 75}, */
665 /* 7 { 70,-43,-87, 9, 90, 25,-80,-57, 57, 80,-25,-90, -9, 87, 43,-70}, */
666 /* 8 { 64,-64,-64, 64, 64,-64,-64, 64, 64,-64,-64, 64, 64,-64,-64, 64}, */
667 /* 9 { 57,-80,-25, 90, -9,-87, 43, 70,-70,-43, 87, 9,-90, 25, 80,-57}, */
668 /* 10 { 50,-89, 18, 75,-75,-18, 89,-50,-50, 89,-18,-75, 75, 18,-89, 50}, */
669 /* 11 { 43,-90, 57, 25,-87, 70, 9,-80, 80, -9,-70, 87,-25,-57, 90,-43}, */
670 /* 12 { 36,-83, 83,-36,-36, 83,-83, 36, 36,-83, 83,-36,-36, 83,-83, 36}, */
671 /* 13 { 25,-70, 90,-80, 43, 9,-57, 87,-87, 57, -9,-43, 80,-90, 70,-25}, */
672 /* 14 { 18,-50, 75,-89, 89,-75, 50,-18,-18, 50,-75, 89,-89, 75,-50, 18}, */
673 /* 15 { 9,-25, 43,-57, 70,-80, 87,-90, 90,-87, 80,-70, 57,-43, 25, -9} */
674 /***************************************************************************/
675
676 for(j = 0; j < TRANS_SIZE_16; j++)
677 {
678 /* Checking for Zero Cols */
679 if((zero_cols & 1) == 1)
680 {
681 memset(pi2_dst, 0, TRANS_SIZE_16 * sizeof(WORD16));
682 }
683 else
684 {
685 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
686 {
687 /*
688 o[k] = g_ai2_ihevc_trans_16[ 1][k]*pi2_src[ src_strd ] + g_ai2_ihevc_trans_16[ 3][k]*pi2_src[ 3*src_strd ] + g_ai2_ihevc_trans_16[ 5][k]*pi2_src[ 5*src_strd ] + g_ai2_ihevc_trans_16[ 7][k]*pi2_src[ 7*src_strd ] +
689 g_ai2_ihevc_trans_16[ 9][k]*pi2_src[ 9*src_strd ] + g_ai2_ihevc_trans_16[11][k]*pi2_src[11*src_strd ] + g_ai2_ihevc_trans_16[13][k]*pi2_src[13*src_strd ] + g_ai2_ihevc_trans_16[15][k]*pi2_src[15*src_strd ];
690 */
691
692 o[0] = 90 * pi2_src[src_strd] + 87 * pi2_src[3 * src_strd]
693 + 80 * pi2_src[5 * src_strd]
694 + 70 * pi2_src[7 * src_strd]
695 + 57 * pi2_src[9 * src_strd]
696 + 43 * pi2_src[11 * src_strd]
697 + 25 * pi2_src[13 * src_strd]
698 + 9 * pi2_src[15 * src_strd];
699
700 o[1] = 87 * pi2_src[src_strd] + 57 * pi2_src[3 * src_strd]
701 + 9 * pi2_src[5 * src_strd]
702 + -43 * pi2_src[7 * src_strd]
703 + -80 * pi2_src[9 * src_strd]
704 + -90 * pi2_src[11 * src_strd]
705 + -70 * pi2_src[13 * src_strd]
706 + -25 * pi2_src[15 * src_strd];
707
708 o[2] = 80 * pi2_src[src_strd] + 9 * pi2_src[3 * src_strd]
709 + -70 * pi2_src[5 * src_strd]
710 + -87 * pi2_src[7 * src_strd]
711 + -25 * pi2_src[9 * src_strd]
712 + 57 * pi2_src[11 * src_strd]
713 + 90 * pi2_src[13 * src_strd]
714 + 43 * pi2_src[15 * src_strd];
715
716 o[3] = 70 * pi2_src[src_strd] + -43 * pi2_src[3 * src_strd]
717 + -87 * pi2_src[5 * src_strd]
718 + 9 * pi2_src[7 * src_strd]
719 + 90 * pi2_src[9 * src_strd]
720 + 25 * pi2_src[11 * src_strd]
721 + -80 * pi2_src[13 * src_strd]
722 + -57 * pi2_src[15 * src_strd];
723
724 o[4] = 57 * pi2_src[src_strd] + -80 * pi2_src[3 * src_strd]
725 + -25 * pi2_src[5 * src_strd]
726 + 90 * pi2_src[7 * src_strd]
727 + -9 * pi2_src[9 * src_strd]
728 + -87 * pi2_src[11 * src_strd]
729 + 43 * pi2_src[13 * src_strd]
730 + 70 * pi2_src[15 * src_strd];
731
732 o[5] = 43 * pi2_src[src_strd] + -90 * pi2_src[3 * src_strd]
733 + 57 * pi2_src[5 * src_strd]
734 + 25 * pi2_src[7 * src_strd]
735 + -87 * pi2_src[9 * src_strd]
736 + 70 * pi2_src[11 * src_strd]
737 + 9 * pi2_src[13 * src_strd]
738 + -80 * pi2_src[15 * src_strd];
739
740 o[6] = 25 * pi2_src[src_strd] + -70 * pi2_src[3 * src_strd]
741 + 90 * pi2_src[5 * src_strd]
742 + -80 * pi2_src[7 * src_strd]
743 + 43 * pi2_src[9 * src_strd]
744 + 9 * pi2_src[11 * src_strd]
745 + -57 * pi2_src[13 * src_strd]
746 + 87 * pi2_src[15 * src_strd];
747
748 o[7] = 9 * pi2_src[src_strd] + -25 * pi2_src[3 * src_strd]
749 + 43 * pi2_src[5 * src_strd]
750 + -57 * pi2_src[7 * src_strd]
751 + 70 * pi2_src[9 * src_strd]
752 + -80 * pi2_src[11 * src_strd]
753 + 87 * pi2_src[13 * src_strd]
754 + -90 * pi2_src[15 * src_strd];
755 }
756 {
757 temp1 = (pi2_src[2 * src_strd] + pi2_src[6 * src_strd]) * 75;
758 temp2 = (pi2_src[10 * src_strd] + pi2_src[14 * src_strd]) * 50;
759 eo[0] = temp1 + 14 * pi2_src[2 * src_strd] + temp2
760 - (pi2_src[14 * src_strd] << 5);
761 eo[1] = temp1 - 93 * pi2_src[6 * src_strd] - temp2
762 - 39 * pi2_src[10 * src_strd];
763
764 temp1 = (pi2_src[2 * src_strd] - pi2_src[6 * src_strd]) * 50;
765 temp2 = (pi2_src[10 * src_strd] + pi2_src[14 * src_strd]) * 75;
766 eo[2] = temp1 - 39 * pi2_src[6 * src_strd] + temp2
767 - 57 * pi2_src[10 * src_strd];
768 eo[3] = temp1 - (pi2_src[2 * src_strd] << 5) + temp2
769 - 164 * pi2_src[14 * src_strd];
770 }
771
772 temp1 = (pi2_src[4 * src_strd] + pi2_src[12 * src_strd]) * 36;
773 eeo[0] = temp1 + 47 * pi2_src[4 * src_strd];
774 eeo[1] = temp1 - 119 * pi2_src[12 * src_strd];
775
776 eee[0] = (pi2_src[0] + pi2_src[8 * src_strd]) << 6;
777 eee[1] = (pi2_src[0] - pi2_src[8 * src_strd]) << 6;
778
779 /* Combining e and o terms at each hierarchy levels to calculate the final spatial domain vector */
780 for(k = 0; k < 2; k++)
781 {
782 ee[k] = eee[k] + eeo[k];
783 ee[k + 2] = eee[1 - k] - eeo[1 - k];
784 }
785 for(k = 0; k < 4; k++)
786 {
787 e[k] = ee[k] + eo[k];
788 e[k + 4] = ee[3 - k] - eo[3 - k];
789 }
790 for(k = 0; k < 8; k++)
791 {
792 pi2_dst[k] =
793 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
794 pi2_dst[k + 8] =
795 CLIP_S16(((e[7 - k] - o[7 - k] + add) >> i4_shift));
796 }
797 }
798 pi2_src++;
799 pi2_dst += dst_strd;
800 zero_cols = zero_cols >> 1;
801 }
802 }
803 #endif
804
805 /**
806 *******************************************************************************
807 *
808 * @brief
809 * This function performs Single stage Inverse transform for 32x32 input
810 * block
811 *
812 * @par Description:
813 * Performs single stage 32x32 inverse transform by utilizing the symmetry
814 * of transformation matrix and reducing number of multiplications wherever
815 * possible but keeping the number of operations (addition,multiplication
816 * and shift) same
817 *
818 * @param[in] pi2_src
819 * Input 32x32 coefficients
820 *
821 * @param[out] pi2_dst
822 * Output 32x32 block
823 *
824 * @param[in] src_strd
825 * Input stride
826 *
827 * @param[in] dst_strd
828 * Output Stride
829 *
830 * @param[in] i4_shift
831 * Output shift
832 *
833 * @param[in] zero_cols
834 * Zero columns in pi2_src
835 *
836 * @returns Void
837 *
838 * @remarks
839 * None
840 *
841 *******************************************************************************
842 */
843
844
ihevc_itrans_32x32(WORD16 * pi2_src,WORD16 * pi2_dst,WORD32 src_strd,WORD32 dst_strd,WORD32 i4_shift,WORD32 zero_cols)845 void ihevc_itrans_32x32(WORD16 *pi2_src,
846 WORD16 *pi2_dst,
847 WORD32 src_strd,
848 WORD32 dst_strd,
849 WORD32 i4_shift,
850 WORD32 zero_cols)
851 {
852 WORD32 j, k;
853 WORD32 e[16], o[16];
854 WORD32 ee[8], eo[8];
855 WORD32 eee[4], eeo[4];
856 WORD32 eeee[2], eeeo[2];
857 WORD32 add;
858
859 add = 1 << (i4_shift - 1);
860
861 for(j = 0; j < TRANS_SIZE_32; j++)
862 {
863 /* Checking for Zero Cols */
864 if((zero_cols & 1) == 1)
865 {
866 memset(pi2_dst, 0, TRANS_SIZE_32 * sizeof(WORD16));
867 }
868 else
869 {
870 /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
871 for(k = 0; k < 16; k++)
872 {
873 o[k] = g_ai2_ihevc_trans_32[1][k] * pi2_src[src_strd]
874 + g_ai2_ihevc_trans_32[3][k]
875 * pi2_src[3 * src_strd]
876 + g_ai2_ihevc_trans_32[5][k]
877 * pi2_src[5 * src_strd]
878 + g_ai2_ihevc_trans_32[7][k]
879 * pi2_src[7 * src_strd]
880 + g_ai2_ihevc_trans_32[9][k]
881 * pi2_src[9 * src_strd]
882 + g_ai2_ihevc_trans_32[11][k]
883 * pi2_src[11 * src_strd]
884 + g_ai2_ihevc_trans_32[13][k]
885 * pi2_src[13 * src_strd]
886 + g_ai2_ihevc_trans_32[15][k]
887 * pi2_src[15 * src_strd]
888 + g_ai2_ihevc_trans_32[17][k]
889 * pi2_src[17 * src_strd]
890 + g_ai2_ihevc_trans_32[19][k]
891 * pi2_src[19 * src_strd]
892 + g_ai2_ihevc_trans_32[21][k]
893 * pi2_src[21 * src_strd]
894 + g_ai2_ihevc_trans_32[23][k]
895 * pi2_src[23 * src_strd]
896 + g_ai2_ihevc_trans_32[25][k]
897 * pi2_src[25 * src_strd]
898 + g_ai2_ihevc_trans_32[27][k]
899 * pi2_src[27 * src_strd]
900 + g_ai2_ihevc_trans_32[29][k]
901 * pi2_src[29 * src_strd]
902 + g_ai2_ihevc_trans_32[31][k]
903 * pi2_src[31 * src_strd];
904 }
905 for(k = 0; k < 8; k++)
906 {
907 eo[k] = g_ai2_ihevc_trans_32[2][k] * pi2_src[2 * src_strd]
908 + g_ai2_ihevc_trans_32[6][k]
909 * pi2_src[6 * src_strd]
910 + g_ai2_ihevc_trans_32[10][k]
911 * pi2_src[10 * src_strd]
912 + g_ai2_ihevc_trans_32[14][k]
913 * pi2_src[14 * src_strd]
914 + g_ai2_ihevc_trans_32[18][k]
915 * pi2_src[18 * src_strd]
916 + g_ai2_ihevc_trans_32[22][k]
917 * pi2_src[22 * src_strd]
918 + g_ai2_ihevc_trans_32[26][k]
919 * pi2_src[26 * src_strd]
920 + g_ai2_ihevc_trans_32[30][k]
921 * pi2_src[30 * src_strd];
922 }
923 for(k = 0; k < 4; k++)
924 {
925 eeo[k] = g_ai2_ihevc_trans_32[4][k] * pi2_src[4 * src_strd]
926 + g_ai2_ihevc_trans_32[12][k]
927 * pi2_src[12 * src_strd]
928 + g_ai2_ihevc_trans_32[20][k]
929 * pi2_src[20 * src_strd]
930 + g_ai2_ihevc_trans_32[28][k]
931 * pi2_src[28 * src_strd];
932 }
933 eeeo[0] = g_ai2_ihevc_trans_32[8][0] * pi2_src[8 * src_strd]
934 + g_ai2_ihevc_trans_32[24][0]
935 * pi2_src[24 * src_strd];
936 eeeo[1] = g_ai2_ihevc_trans_32[8][1] * pi2_src[8 * src_strd]
937 + g_ai2_ihevc_trans_32[24][1]
938 * pi2_src[24 * src_strd];
939 eeee[0] = g_ai2_ihevc_trans_32[0][0] * pi2_src[0]
940 + g_ai2_ihevc_trans_32[16][0]
941 * pi2_src[16 * src_strd];
942 eeee[1] = g_ai2_ihevc_trans_32[0][1] * pi2_src[0]
943 + g_ai2_ihevc_trans_32[16][1]
944 * pi2_src[16 * src_strd];
945
946 /* Combining e and o terms at each hierarchy levels to calculate the final spatial domain vector */
947 eee[0] = eeee[0] + eeeo[0];
948 eee[3] = eeee[0] - eeeo[0];
949 eee[1] = eeee[1] + eeeo[1];
950 eee[2] = eeee[1] - eeeo[1];
951 for(k = 0; k < 4; k++)
952 {
953 ee[k] = eee[k] + eeo[k];
954 ee[k + 4] = eee[3 - k] - eeo[3 - k];
955 }
956 for(k = 0; k < 8; k++)
957 {
958 e[k] = ee[k] + eo[k];
959 e[k + 8] = ee[7 - k] - eo[7 - k];
960 }
961 for(k = 0; k < 16; k++)
962 {
963 pi2_dst[k] =
964 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
965 pi2_dst[k + 16] =
966 CLIP_S16(((e[15 - k] - o[15 - k] + add) >> i4_shift));
967 }
968 }
969 pi2_src++;
970 pi2_dst += dst_strd;
971 zero_cols = zero_cols >> 1;
972 }
973 }
974
975