2 * DSP functions for Indeo Video Interactive codecs (Indeo4 and Indeo5)
4 * Copyright (c) 2009-2011 Maxim Poliakovski
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * DSP functions (inverse transforms, motion compensation, wavelet recompostions)
26 * for Indeo Video Interactive codecs.
30 #include "ivi_common.h"
33 void ff_ivi_recompose53(const IVIPlaneDesc *plane, uint8_t *dst,
37 int32_t p0, p1, p2, p3, tmp0, tmp1, tmp2;
38 int32_t b0_1, b0_2, b1_1, b1_2, b1_3, b2_1, b2_2, b2_3, b2_4, b2_5, b2_6;
39 int32_t b3_1, b3_2, b3_3, b3_4, b3_5, b3_6, b3_7, b3_8, b3_9;
40 int32_t pitch, back_pitch;
41 const short *b0_ptr, *b1_ptr, *b2_ptr, *b3_ptr;
42 const int num_bands = 4;
44 /* all bands should have the same pitch */
45 pitch = plane->bands[0].pitch;
47 /* pixels at the position "y-1" will be set to pixels at the "y" for the 1st iteration */
50 /* get pointers to the wavelet bands */
51 b0_ptr = plane->bands[0].buf;
52 b1_ptr = plane->bands[1].buf;
53 b2_ptr = plane->bands[2].buf;
54 b3_ptr = plane->bands[3].buf;
56 for (y = 0; y < plane->height; y += 2) {
58 if (y+2 >= plane->height)
60 /* load storage variables with values */
67 b1_1 = b1_ptr[back_pitch];
69 b1_3 = b1_1 - b1_2*6 + b1_ptr[pitch];
73 b2_2 = b2_ptr[0]; // b2[x, y ]
74 b2_3 = b2_2; // b2[x+1,y ] = b2[x,y]
75 b2_5 = b2_ptr[pitch]; // b2[x ,y+1]
76 b2_6 = b2_5; // b2[x+1,y+1] = b2[x,y+1]
80 b3_2 = b3_ptr[back_pitch]; // b3[x ,y-1]
81 b3_3 = b3_2; // b3[x+1,y-1] = b3[x ,y-1]
82 b3_5 = b3_ptr[0]; // b3[x ,y ]
83 b3_6 = b3_5; // b3[x+1,y ] = b3[x ,y ]
84 b3_8 = b3_2 - b3_5*6 + b3_ptr[pitch];
88 for (x = 0, indx = 0; x < plane->width; x+=2, indx++) {
89 if (x+2 >= plane->width) {
96 /* some values calculated in the previous iterations can */
97 /* be reused in the next ones, so do appropriate copying */
98 b2_1 = b2_2; // b2[x-1,y ] = b2[x, y ]
99 b2_2 = b2_3; // b2[x ,y ] = b2[x+1,y ]
100 b2_4 = b2_5; // b2[x-1,y+1] = b2[x ,y+1]
101 b2_5 = b2_6; // b2[x ,y+1] = b2[x+1,y+1]
102 b3_1 = b3_2; // b3[x-1,y-1] = b3[x ,y-1]
103 b3_2 = b3_3; // b3[x ,y-1] = b3[x+1,y-1]
104 b3_4 = b3_5; // b3[x-1,y ] = b3[x ,y ]
105 b3_5 = b3_6; // b3[x ,y ] = b3[x+1,y ]
106 b3_7 = b3_8; // vert_HPF(x-1)
107 b3_8 = b3_9; // vert_HPF(x )
109 p0 = p1 = p2 = p3 = 0;
111 /* process the LL-band by applying LPF both vertically and horizontally */
115 b0_1 = b0_ptr[indx+1];
116 b0_2 = b0_ptr[pitch+indx+1];
121 p2 = (tmp0 + tmp2) << 3;
122 p3 = (tmp1 + tmp2 + b0_2) << 2;
125 /* process the HL-band by applying HPF vertically and LPF horizontally */
129 b1_2 = b1_ptr[indx+1];
130 b1_1 = b1_ptr[back_pitch+indx+1];
132 tmp2 = tmp1 - tmp0*6 + b1_3;
133 b1_3 = b1_1 - b1_2*6 + b1_ptr[pitch+indx+1];
135 p0 += (tmp0 + tmp1) << 3;
136 p1 += (tmp0 + tmp1 + b1_1 + b1_2) << 2;
138 p3 += (tmp2 + b1_3) << 1;
141 /* process the LH-band by applying LPF vertically and HPF horizontally */
143 b2_3 = b2_ptr[indx+1];
144 b2_6 = b2_ptr[pitch+indx+1];
147 tmp1 = b2_1 - b2_2*6 + b2_3;
151 p2 += (tmp0 + b2_4 + b2_5) << 2;
152 p3 += (tmp1 + b2_4 - b2_5*6 + b2_6) << 1;
155 /* process the HH-band by applying HPF both vertically and horizontally */
157 b3_6 = b3_ptr[indx+1]; // b3[x+1,y ]
158 b3_3 = b3_ptr[back_pitch+indx+1]; // b3[x+1,y-1]
164 b3_9 = b3_3 - b3_6*6 + b3_ptr[pitch+indx+1];
166 p0 += (tmp0 + tmp1) << 2;
167 p1 += (tmp0 - tmp1*6 + tmp2) << 1;
168 p2 += (b3_7 + b3_8) << 1;
169 p3 += b3_7 - b3_8*6 + b3_9;
172 /* output four pixels */
173 dst[x] = av_clip_uint8((p0 >> 6) + 128);
174 dst[x+1] = av_clip_uint8((p1 >> 6) + 128);
175 dst[dst_pitch+x] = av_clip_uint8((p2 >> 6) + 128);
176 dst[dst_pitch+x+1] = av_clip_uint8((p3 >> 6) + 128);
179 dst += dst_pitch << 1;
190 void ff_ivi_recompose_haar(const IVIPlaneDesc *plane, uint8_t *dst,
193 int x, y, indx, b0, b1, b2, b3, p0, p1, p2, p3;
194 const short *b0_ptr, *b1_ptr, *b2_ptr, *b3_ptr;
197 /* all bands should have the same pitch */
198 pitch = plane->bands[0].pitch;
200 /* get pointers to the wavelet bands */
201 b0_ptr = plane->bands[0].buf;
202 b1_ptr = plane->bands[1].buf;
203 b2_ptr = plane->bands[2].buf;
204 b3_ptr = plane->bands[3].buf;
206 for (y = 0; y < plane->height; y += 2) {
207 for (x = 0, indx = 0; x < plane->width; x += 2, indx++) {
208 /* load coefficients */
209 b0 = b0_ptr[indx]; //should be: b0 = (num_bands > 0) ? b0_ptr[indx] : 0;
210 b1 = b1_ptr[indx]; //should be: b1 = (num_bands > 1) ? b1_ptr[indx] : 0;
211 b2 = b2_ptr[indx]; //should be: b2 = (num_bands > 2) ? b2_ptr[indx] : 0;
212 b3 = b3_ptr[indx]; //should be: b3 = (num_bands > 3) ? b3_ptr[indx] : 0;
214 /* haar wavelet recomposition */
215 p0 = (b0 + b1 + b2 + b3 + 2) >> 2;
216 p1 = (b0 + b1 - b2 - b3 + 2) >> 2;
217 p2 = (b0 - b1 + b2 - b3 + 2) >> 2;
218 p3 = (b0 - b1 - b2 + b3 + 2) >> 2;
220 /* bias, convert and output four pixels */
221 dst[x] = av_clip_uint8(p0 + 128);
222 dst[x + 1] = av_clip_uint8(p1 + 128);
223 dst[dst_pitch + x] = av_clip_uint8(p2 + 128);
224 dst[dst_pitch + x + 1] = av_clip_uint8(p3 + 128);
227 dst += dst_pitch << 1;
236 /** butterfly operation for the inverse Haar transform */
237 #define IVI_HAAR_BFLY(s1, s2, o1, o2, t) \
239 o1 = (s1 + s2) >> 1;\
242 /** inverse 8-point Haar transform */
243 #define INV_HAAR8(s1, s5, s3, s7, s2, s4, s6, s8,\
244 d1, d2, d3, d4, d5, d6, d7, d8,\
245 t0, t1, t2, t3, t4, t5, t6, t7, t8) {\
246 t1 = s1 << 1; t5 = s5 << 1;\
247 IVI_HAAR_BFLY(t1, t5, t1, t5, t0); IVI_HAAR_BFLY(t1, s3, t1, t3, t0);\
248 IVI_HAAR_BFLY(t5, s7, t5, t7, t0); IVI_HAAR_BFLY(t1, s2, t1, t2, t0);\
249 IVI_HAAR_BFLY(t3, s4, t3, t4, t0); IVI_HAAR_BFLY(t5, s6, t5, t6, t0);\
250 IVI_HAAR_BFLY(t7, s8, t7, t8, t0);\
251 d1 = COMPENSATE(t1);\
252 d2 = COMPENSATE(t2);\
253 d3 = COMPENSATE(t3);\
254 d4 = COMPENSATE(t4);\
255 d5 = COMPENSATE(t5);\
256 d6 = COMPENSATE(t6);\
257 d7 = COMPENSATE(t7);\
258 d8 = COMPENSATE(t8); }
260 /** inverse 4-point Haar transform */
261 #define INV_HAAR4(s1, s3, s5, s7) {\
262 HAAR_BFLY(s1, s5); HAAR_BFLY(s1, s3); HAAR_BFLY(s5, s7);\
263 s1 = COMPENSATE(s1);\
264 s3 = COMPENSATE(s3);\
265 s5 = COMPENSATE(s5);\
266 s7 = COMPENSATE(s7); }
268 void ff_ivi_inverse_haar_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
269 const uint8_t *flags)
271 int i, shift, sp1, sp2, sp3, sp4;
275 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
277 /* apply the InvHaar8 to all columns */
278 #define COMPENSATE(x) (x)
281 for (i = 0; i < 8; i++) {
285 sp1 = src[ 0] << shift;
286 sp2 = src[ 8] << shift;
287 sp3 = src[16] << shift;
288 sp4 = src[24] << shift;
289 INV_HAAR8( sp1, sp2, sp3, sp4,
290 src[32], src[40], src[48], src[56],
291 dst[ 0], dst[ 8], dst[16], dst[24],
292 dst[32], dst[40], dst[48], dst[56],
293 t0, t1, t2, t3, t4, t5, t6, t7, t8);
295 dst[ 0] = dst[ 8] = dst[16] = dst[24] =
296 dst[32] = dst[40] = dst[48] = dst[56] = 0;
303 /* apply the InvHaar8 to all rows */
304 #define COMPENSATE(x) (x)
306 for (i = 0; i < 8; i++) {
307 if ( !src[0] && !src[1] && !src[2] && !src[3]
308 && !src[4] && !src[5] && !src[6] && !src[7]) {
309 memset(out, 0, 8 * sizeof(out[0]));
311 INV_HAAR8(src[0], src[1], src[2], src[3],
312 src[4], src[5], src[6], src[7],
313 out[0], out[1], out[2], out[3],
314 out[4], out[5], out[6], out[7],
315 t0, t1, t2, t3, t4, t5, t6, t7, t8);
323 void ff_ivi_inverse_haar_1x8(const int32_t *in, int16_t *out, uint32_t pitch,
324 const uint8_t *flags)
328 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
330 /* apply the InvHaar8 to all columns */
331 #define COMPENSATE(x) (x)
333 for (i = 0; i < 8; i++) {
335 INV_HAAR8(src[ 0], src[ 8], src[16], src[24],
336 src[32], src[40], src[48], src[56],
337 out[ 0], out[pitch], out[2*pitch], out[3*pitch],
338 out[4*pitch], out[5*pitch], out[6*pitch], out[7*pitch],
339 t0, t1, t2, t3, t4, t5, t6, t7, t8);
341 out[ 0]= out[ pitch]= out[2*pitch]= out[3*pitch]=
342 out[4*pitch]= out[5*pitch]= out[6*pitch]= out[7*pitch]= 0;
350 void ff_ivi_inverse_haar_8x1(const int32_t *in, int16_t *out, uint32_t pitch,
351 const uint8_t *flags)
355 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
357 /* apply the InvHaar8 to all rows */
358 #define COMPENSATE(x) (x)
360 for (i = 0; i < 8; i++) {
361 if ( !src[0] && !src[1] && !src[2] && !src[3]
362 && !src[4] && !src[5] && !src[6] && !src[7]) {
363 memset(out, 0, 8 * sizeof(out[0]));
365 INV_HAAR8(src[0], src[1], src[2], src[3],
366 src[4], src[5], src[6], src[7],
367 out[0], out[1], out[2], out[3],
368 out[4], out[5], out[6], out[7],
369 t0, t1, t2, t3, t4, t5, t6, t7, t8);
377 void ff_ivi_dc_haar_2d(const int32_t *in, int16_t *out, uint32_t pitch,
383 dc_coeff = (*in + 0) >> 3;
385 for (y = 0; y < blk_size; out += pitch, y++) {
386 for (x = 0; x < blk_size; x++)
391 /** butterfly operation for the inverse slant transform */
392 #define IVI_SLANT_BFLY(s1, s2, o1, o2, t) \
397 /** This is a reflection a,b = 1/2, 5/4 for the inverse slant transform */
398 #define IVI_IREFLECT(s1, s2, o1, o2, t) \
399 t = ((s1 + s2*2 + 2) >> 2) + s1;\
400 o2 = ((s1*2 - s2 + 2) >> 2) - s2;\
403 /** This is a reflection a,b = 1/2, 7/8 for the inverse slant transform */
404 #define IVI_SLANT_PART4(s1, s2, o1, o2, t) \
405 t = s2 + ((s1*4 - s2 + 4) >> 3);\
406 o2 = s1 + ((-s1 - s2*4 + 4) >> 3);\
409 /** inverse slant8 transform */
410 #define IVI_INV_SLANT8(s1, s4, s8, s5, s2, s6, s3, s7,\
411 d1, d2, d3, d4, d5, d6, d7, d8,\
412 t0, t1, t2, t3, t4, t5, t6, t7, t8) {\
413 IVI_SLANT_PART4(s4, s5, t4, t5, t0);\
415 IVI_SLANT_BFLY(s1, t5, t1, t5, t0); IVI_SLANT_BFLY(s2, s6, t2, t6, t0);\
416 IVI_SLANT_BFLY(s7, s3, t7, t3, t0); IVI_SLANT_BFLY(t4, s8, t4, t8, t0);\
418 IVI_SLANT_BFLY(t1, t2, t1, t2, t0); IVI_IREFLECT (t4, t3, t4, t3, t0);\
419 IVI_SLANT_BFLY(t5, t6, t5, t6, t0); IVI_IREFLECT (t8, t7, t8, t7, t0);\
420 IVI_SLANT_BFLY(t1, t4, t1, t4, t0); IVI_SLANT_BFLY(t2, t3, t2, t3, t0);\
421 IVI_SLANT_BFLY(t5, t8, t5, t8, t0); IVI_SLANT_BFLY(t6, t7, t6, t7, t0);\
422 d1 = COMPENSATE(t1);\
423 d2 = COMPENSATE(t2);\
424 d3 = COMPENSATE(t3);\
425 d4 = COMPENSATE(t4);\
426 d5 = COMPENSATE(t5);\
427 d6 = COMPENSATE(t6);\
428 d7 = COMPENSATE(t7);\
429 d8 = COMPENSATE(t8);}
431 /** inverse slant4 transform */
432 #define IVI_INV_SLANT4(s1, s4, s2, s3, d1, d2, d3, d4, t0, t1, t2, t3, t4) {\
433 IVI_SLANT_BFLY(s1, s2, t1, t2, t0); IVI_IREFLECT (s4, s3, t4, t3, t0);\
435 IVI_SLANT_BFLY(t1, t4, t1, t4, t0); IVI_SLANT_BFLY(t2, t3, t2, t3, t0);\
436 d1 = COMPENSATE(t1);\
437 d2 = COMPENSATE(t2);\
438 d3 = COMPENSATE(t3);\
439 d4 = COMPENSATE(t4);}
441 void ff_ivi_inverse_slant_8x8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
447 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
449 #define COMPENSATE(x) (x)
452 for (i = 0; i < 8; i++) {
454 IVI_INV_SLANT8(src[0], src[8], src[16], src[24], src[32], src[40], src[48], src[56],
455 dst[0], dst[8], dst[16], dst[24], dst[32], dst[40], dst[48], dst[56],
456 t0, t1, t2, t3, t4, t5, t6, t7, t8);
458 dst[0] = dst[8] = dst[16] = dst[24] = dst[32] = dst[40] = dst[48] = dst[56] = 0;
465 #define COMPENSATE(x) ((x + 1)>>1)
467 for (i = 0; i < 8; i++) {
468 if (!src[0] && !src[1] && !src[2] && !src[3] && !src[4] && !src[5] && !src[6] && !src[7]) {
469 memset(out, 0, 8*sizeof(out[0]));
471 IVI_INV_SLANT8(src[0], src[1], src[2], src[3], src[4], src[5], src[6], src[7],
472 out[0], out[1], out[2], out[3], out[4], out[5], out[6], out[7],
473 t0, t1, t2, t3, t4, t5, t6, t7, t8);
481 void ff_ivi_inverse_slant_4x4(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
487 int t0, t1, t2, t3, t4;
489 #define COMPENSATE(x) (x)
492 for (i = 0; i < 4; i++) {
494 IVI_INV_SLANT4(src[0], src[4], src[8], src[12],
495 dst[0], dst[4], dst[8], dst[12],
498 dst[0] = dst[4] = dst[8] = dst[12] = 0;
505 #define COMPENSATE(x) ((x + 1)>>1)
507 for (i = 0; i < 4; i++) {
508 if (!src[0] && !src[1] && !src[2] && !src[3]) {
509 out[0] = out[1] = out[2] = out[3] = 0;
511 IVI_INV_SLANT4(src[0], src[1], src[2], src[3],
512 out[0], out[1], out[2], out[3],
521 void ff_ivi_dc_slant_2d(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
526 dc_coeff = (*in + 1) >> 1;
528 for (y = 0; y < blk_size; out += pitch, y++) {
529 for (x = 0; x < blk_size; x++)
534 void ff_ivi_row_slant8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
537 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
539 #define COMPENSATE(x) ((x + 1)>>1)
540 for (i = 0; i < 8; i++) {
541 if (!in[0] && !in[1] && !in[2] && !in[3] && !in[4] && !in[5] && !in[6] && !in[7]) {
542 memset(out, 0, 8*sizeof(out[0]));
544 IVI_INV_SLANT8( in[0], in[1], in[2], in[3], in[4], in[5], in[6], in[7],
545 out[0], out[1], out[2], out[3], out[4], out[5], out[6], out[7],
546 t0, t1, t2, t3, t4, t5, t6, t7, t8);
554 void ff_ivi_dc_row_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
559 dc_coeff = (*in + 1) >> 1;
561 for (x = 0; x < blk_size; x++)
566 for (y = 1; y < blk_size; out += pitch, y++) {
567 for (x = 0; x < blk_size; x++)
572 void ff_ivi_col_slant8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
574 int i, row2, row4, row8;
575 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
581 #define COMPENSATE(x) ((x + 1)>>1)
582 for (i = 0; i < 8; i++) {
584 IVI_INV_SLANT8(in[0], in[8], in[16], in[24], in[32], in[40], in[48], in[56],
585 out[0], out[pitch], out[row2], out[row2 + pitch], out[row4],
586 out[row4 + pitch], out[row4 + row2], out[row8 - pitch],
587 t0, t1, t2, t3, t4, t5, t6, t7, t8);
589 out[0] = out[pitch] = out[row2] = out[row2 + pitch] = out[row4] =
590 out[row4 + pitch] = out[row4 + row2] = out[row8 - pitch] = 0;
599 void ff_ivi_dc_col_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
604 dc_coeff = (*in + 1) >> 1;
606 for (y = 0; y < blk_size; out += pitch, y++) {
608 for (x = 1; x < blk_size; x++)
613 void ff_ivi_put_pixels_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
614 const uint8_t *flags)
618 for (y = 0; y < 8; out += pitch, in += 8, y++)
619 for (x = 0; x < 8; x++)
623 void ff_ivi_put_dc_pixel_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
629 memset(out + 1, 0, 7*sizeof(out[0]));
632 for (y = 1; y < 8; out += pitch, y++)
633 memset(out, 0, 8*sizeof(out[0]));
636 #define IVI_MC_TEMPLATE(size, suffix, OP) \
637 void ff_ivi_mc_ ## size ##x## size ## suffix (int16_t *buf, const int16_t *ref_buf, \
638 uint32_t pitch, int mc_type) \
641 const int16_t *wptr; \
644 case 0: /* fullpel (no interpolation) */ \
645 for (i = 0; i < size; i++, buf += pitch, ref_buf += pitch) { \
646 for (j = 0; j < size; j++) {\
647 OP(buf[j], ref_buf[j]); \
651 case 1: /* horizontal halfpel interpolation */ \
652 for (i = 0; i < size; i++, buf += pitch, ref_buf += pitch) \
653 for (j = 0; j < size; j++) \
654 OP(buf[j], (ref_buf[j] + ref_buf[j+1]) >> 1); \
656 case 2: /* vertical halfpel interpolation */ \
657 wptr = ref_buf + pitch; \
658 for (i = 0; i < size; i++, buf += pitch, wptr += pitch, ref_buf += pitch) \
659 for (j = 0; j < size; j++) \
660 OP(buf[j], (ref_buf[j] + wptr[j]) >> 1); \
662 case 3: /* vertical and horizontal halfpel interpolation */ \
663 wptr = ref_buf + pitch; \
664 for (i = 0; i < size; i++, buf += pitch, wptr += pitch, ref_buf += pitch) \
665 for (j = 0; j < size; j++) \
666 OP(buf[j], (ref_buf[j] + ref_buf[j+1] + wptr[j] + wptr[j+1]) >> 2); \
671 #define OP_PUT(a, b) (a) = (b)
672 #define OP_ADD(a, b) (a) += (b)
674 IVI_MC_TEMPLATE(8, _no_delta, OP_PUT)
675 IVI_MC_TEMPLATE(8, _delta, OP_ADD)
676 IVI_MC_TEMPLATE(4, _no_delta, OP_PUT)
677 IVI_MC_TEMPLATE(4, _delta, OP_ADD)