2 * DSP functions for Indeo Video Interactive codecs (Indeo4 and Indeo5)
4 * Copyright (c) 2009-2011 Maxim Poliakovski
6 * This file is part of Libav.
8 * Libav 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 * Libav 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 Libav; 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.
32 #include "ivi_common.h"
35 void ff_ivi_recompose53(const IVIPlaneDesc *plane, uint8_t *dst,
36 const int dst_pitch, const int num_bands)
39 int32_t p0, p1, p2, p3, tmp0, tmp1, tmp2;
40 int32_t b0_1, b0_2, b1_1, b1_2, b1_3, b2_1, b2_2, b2_3, b2_4, b2_5, b2_6;
41 int32_t b3_1, b3_2, b3_3, b3_4, b3_5, b3_6, b3_7, b3_8, b3_9;
42 int32_t pitch, back_pitch;
43 const IDWTELEM *b0_ptr, *b1_ptr, *b2_ptr, *b3_ptr;
45 /* all bands should have the same pitch */
46 pitch = plane->bands[0].pitch;
48 /* pixels at the position "y-1" will be set to pixels at the "y" for the 1st iteration */
51 /* get pointers to the wavelet bands */
52 b0_ptr = plane->bands[0].buf;
53 b1_ptr = plane->bands[1].buf;
54 b2_ptr = plane->bands[2].buf;
55 b3_ptr = plane->bands[3].buf;
57 for (y = 0; y < plane->height; y += 2) {
58 /* load storage variables with values */
65 b1_1 = b1_ptr[back_pitch];
67 b1_3 = b1_1 - b1_2*6 + b1_ptr[pitch];
71 b2_2 = b2_ptr[0]; // b2[x, y ]
72 b2_3 = b2_2; // b2[x+1,y ] = b2[x,y]
73 b2_5 = b2_ptr[pitch]; // b2[x ,y+1]
74 b2_6 = b2_5; // b2[x+1,y+1] = b2[x,y+1]
78 b3_2 = b3_ptr[back_pitch]; // b3[x ,y-1]
79 b3_3 = b3_2; // b3[x+1,y-1] = b3[x ,y-1]
80 b3_5 = b3_ptr[0]; // b3[x ,y ]
81 b3_6 = b3_5; // b3[x+1,y ] = b3[x ,y ]
82 b3_8 = b3_2 - b3_5*6 + b3_ptr[pitch];
86 for (x = 0, indx = 0; x < plane->width; x+=2, indx++) {
87 /* some values calculated in the previous iterations can */
88 /* be reused in the next ones, so do appropriate copying */
89 b2_1 = b2_2; // b2[x-1,y ] = b2[x, y ]
90 b2_2 = b2_3; // b2[x ,y ] = b2[x+1,y ]
91 b2_4 = b2_5; // b2[x-1,y+1] = b2[x ,y+1]
92 b2_5 = b2_6; // b2[x ,y+1] = b2[x+1,y+1]
93 b3_1 = b3_2; // b3[x-1,y-1] = b3[x ,y-1]
94 b3_2 = b3_3; // b3[x ,y-1] = b3[x+1,y-1]
95 b3_4 = b3_5; // b3[x-1,y ] = b3[x ,y ]
96 b3_5 = b3_6; // b3[x ,y ] = b3[x+1,y ]
97 b3_7 = b3_8; // vert_HPF(x-1)
98 b3_8 = b3_9; // vert_HPF(x )
100 p0 = p1 = p2 = p3 = 0;
102 /* process the LL-band by applying LPF both vertically and horizontally */
106 b0_1 = b0_ptr[indx+1];
107 b0_2 = b0_ptr[pitch+indx+1];
112 p2 = (tmp0 + tmp2) << 3;
113 p3 = (tmp1 + tmp2 + b0_2) << 2;
116 /* process the HL-band by applying HPF vertically and LPF horizontally */
120 b1_2 = b1_ptr[indx+1];
121 b1_1 = b1_ptr[back_pitch+indx+1];
123 tmp2 = tmp1 - tmp0*6 + b1_3;
124 b1_3 = b1_1 - b1_2*6 + b1_ptr[pitch+indx+1];
126 p0 += (tmp0 + tmp1) << 3;
127 p1 += (tmp0 + tmp1 + b1_1 + b1_2) << 2;
129 p3 += (tmp2 + b1_3) << 1;
132 /* process the LH-band by applying LPF vertically and HPF horizontally */
134 b2_3 = b2_ptr[indx+1];
135 b2_6 = b2_ptr[pitch+indx+1];
138 tmp1 = b2_1 - b2_2*6 + b2_3;
142 p2 += (tmp0 + b2_4 + b2_5) << 2;
143 p3 += (tmp1 + b2_4 - b2_5*6 + b2_6) << 1;
146 /* process the HH-band by applying HPF both vertically and horizontally */
148 b3_6 = b3_ptr[indx+1]; // b3[x+1,y ]
149 b3_3 = b3_ptr[back_pitch+indx+1]; // b3[x+1,y-1]
155 b3_9 = b3_3 - b3_6*6 + b3_ptr[pitch+indx+1];
157 p0 += (tmp0 + tmp1) << 2;
158 p1 += (tmp0 - tmp1*6 + tmp2) << 1;
159 p2 += (b3_7 + b3_8) << 1;
160 p3 += b3_7 - b3_8*6 + b3_9;
163 /* output four pixels */
164 dst[x] = av_clip_uint8((p0 >> 6) + 128);
165 dst[x+1] = av_clip_uint8((p1 >> 6) + 128);
166 dst[dst_pitch+x] = av_clip_uint8((p2 >> 6) + 128);
167 dst[dst_pitch+x+1] = av_clip_uint8((p3 >> 6) + 128);
170 dst += dst_pitch << 1;
181 void ff_ivi_recompose_haar(const IVIPlaneDesc *plane, uint8_t *dst,
182 const int dst_pitch, const int num_bands)
184 int x, y, indx, b0, b1, b2, b3, p0, p1, p2, p3;
185 const IDWTELEM *b0_ptr, *b1_ptr, *b2_ptr, *b3_ptr;
188 /* all bands should have the same pitch */
189 pitch = plane->bands[0].pitch;
191 /* get pointers to the wavelet bands */
192 b0_ptr = plane->bands[0].buf;
193 b1_ptr = plane->bands[1].buf;
194 b2_ptr = plane->bands[2].buf;
195 b3_ptr = plane->bands[3].buf;
197 for (y = 0; y < plane->height; y += 2) {
198 for (x = 0, indx = 0; x < plane->width; x += 2, indx++) {
199 /* load coefficients */
200 b0 = b0_ptr[indx]; //should be: b0 = (num_bands > 0) ? b0_ptr[indx] : 0;
201 b1 = b1_ptr[indx]; //should be: b1 = (num_bands > 1) ? b1_ptr[indx] : 0;
202 b2 = b2_ptr[indx]; //should be: b2 = (num_bands > 2) ? b2_ptr[indx] : 0;
203 b3 = b3_ptr[indx]; //should be: b3 = (num_bands > 3) ? b3_ptr[indx] : 0;
205 /* haar wavelet recomposition */
206 p0 = (b0 + b1 + b2 + b3 + 2) >> 2;
207 p1 = (b0 + b1 - b2 - b3 + 2) >> 2;
208 p2 = (b0 - b1 + b2 - b3 + 2) >> 2;
209 p3 = (b0 - b1 - b2 + b3 + 2) >> 2;
211 /* bias, convert and output four pixels */
212 dst[x] = av_clip_uint8(p0 + 128);
213 dst[x + 1] = av_clip_uint8(p1 + 128);
214 dst[dst_pitch + x] = av_clip_uint8(p2 + 128);
215 dst[dst_pitch + x + 1] = av_clip_uint8(p3 + 128);
218 dst += dst_pitch << 1;
227 /** butterfly operation for the inverse Haar transform */
228 #define IVI_HAAR_BFLY(s1, s2, o1, o2, t) \
230 o1 = (s1 + s2) >> 1;\
233 /** inverse 8-point Haar transform */
234 #define INV_HAAR8(s1, s5, s3, s7, s2, s4, s6, s8,\
235 d1, d2, d3, d4, d5, d6, d7, d8,\
236 t0, t1, t2, t3, t4, t5, t6, t7, t8) {\
237 t1 = s1 << 1; t5 = s5 << 1;\
238 IVI_HAAR_BFLY(t1, t5, t1, t5, t0); IVI_HAAR_BFLY(t1, s3, t1, t3, t0);\
239 IVI_HAAR_BFLY(t5, s7, t5, t7, t0); IVI_HAAR_BFLY(t1, s2, t1, t2, t0);\
240 IVI_HAAR_BFLY(t3, s4, t3, t4, t0); IVI_HAAR_BFLY(t5, s6, t5, t6, t0);\
241 IVI_HAAR_BFLY(t7, s8, t7, t8, t0);\
242 d1 = COMPENSATE(t1);\
243 d2 = COMPENSATE(t2);\
244 d3 = COMPENSATE(t3);\
245 d4 = COMPENSATE(t4);\
246 d5 = COMPENSATE(t5);\
247 d6 = COMPENSATE(t6);\
248 d7 = COMPENSATE(t7);\
249 d8 = COMPENSATE(t8); }
251 /** inverse 4-point Haar transform */
252 #define INV_HAAR4(s1, s3, s5, s7) {\
253 HAAR_BFLY(s1, s5); HAAR_BFLY(s1, s3); HAAR_BFLY(s5, s7);\
254 s1 = COMPENSATE(s1);\
255 s3 = COMPENSATE(s3);\
256 s5 = COMPENSATE(s5);\
257 s7 = COMPENSATE(s7); }
259 void ff_ivi_inverse_haar_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
260 const uint8_t *flags)
262 int i, shift, sp1, sp2, sp3, sp4;
266 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
268 /* apply the InvHaar8 to all columns */
269 #define COMPENSATE(x) (x)
272 for (i = 0; i < 8; i++) {
276 sp1 = src[ 0] << shift;
277 sp2 = src[ 8] << shift;
278 sp3 = src[16] << shift;
279 sp4 = src[24] << shift;
280 INV_HAAR8( sp1, sp2, sp3, sp4,
281 src[32], src[40], src[48], src[56],
282 dst[ 0], dst[ 8], dst[16], dst[24],
283 dst[32], dst[40], dst[48], dst[56],
284 t0, t1, t2, t3, t4, t5, t6, t7, t8);
286 dst[ 0] = dst[ 8] = dst[16] = dst[24] =
287 dst[32] = dst[40] = dst[48] = dst[56] = 0;
294 /* apply the InvHaar8 to all rows */
295 #define COMPENSATE(x) (x)
297 for (i = 0; i < 8; i++) {
298 if ( !src[0] && !src[1] && !src[2] && !src[3]
299 && !src[4] && !src[5] && !src[6] && !src[7]) {
300 memset(out, 0, 8 * sizeof(out[0]));
302 INV_HAAR8(src[0], src[1], src[2], src[3],
303 src[4], src[5], src[6], src[7],
304 out[0], out[1], out[2], out[3],
305 out[4], out[5], out[6], out[7],
306 t0, t1, t2, t3, t4, t5, t6, t7, t8);
314 void ff_ivi_dc_haar_2d(const int32_t *in, int16_t *out, uint32_t pitch,
320 dc_coeff = (*in + 0) >> 3;
322 for (y = 0; y < blk_size; out += pitch, y++) {
323 for (x = 0; x < blk_size; x++)
328 /** butterfly operation for the inverse slant transform */
329 #define IVI_SLANT_BFLY(s1, s2, o1, o2, t) \
334 /** This is a reflection a,b = 1/2, 5/4 for the inverse slant transform */
335 #define IVI_IREFLECT(s1, s2, o1, o2, t) \
336 t = ((s1 + s2*2 + 2) >> 2) + s1;\
337 o2 = ((s1*2 - s2 + 2) >> 2) - s2;\
340 /** This is a reflection a,b = 1/2, 7/8 for the inverse slant transform */
341 #define IVI_SLANT_PART4(s1, s2, o1, o2, t) \
342 t = s2 + ((s1*4 - s2 + 4) >> 3);\
343 o2 = s1 + ((-s1 - s2*4 + 4) >> 3);\
346 /** inverse slant8 transform */
347 #define IVI_INV_SLANT8(s1, s4, s8, s5, s2, s6, s3, s7,\
348 d1, d2, d3, d4, d5, d6, d7, d8,\
349 t0, t1, t2, t3, t4, t5, t6, t7, t8) {\
350 IVI_SLANT_PART4(s4, s5, t4, t5, t0);\
352 IVI_SLANT_BFLY(s1, t5, t1, t5, t0); IVI_SLANT_BFLY(s2, s6, t2, t6, t0);\
353 IVI_SLANT_BFLY(s7, s3, t7, t3, t0); IVI_SLANT_BFLY(t4, s8, t4, t8, t0);\
355 IVI_SLANT_BFLY(t1, t2, t1, t2, t0); IVI_IREFLECT (t4, t3, t4, t3, t0);\
356 IVI_SLANT_BFLY(t5, t6, t5, t6, t0); IVI_IREFLECT (t8, t7, t8, t7, t0);\
357 IVI_SLANT_BFLY(t1, t4, t1, t4, t0); IVI_SLANT_BFLY(t2, t3, t2, t3, t0);\
358 IVI_SLANT_BFLY(t5, t8, t5, t8, t0); IVI_SLANT_BFLY(t6, t7, t6, t7, t0);\
359 d1 = COMPENSATE(t1);\
360 d2 = COMPENSATE(t2);\
361 d3 = COMPENSATE(t3);\
362 d4 = COMPENSATE(t4);\
363 d5 = COMPENSATE(t5);\
364 d6 = COMPENSATE(t6);\
365 d7 = COMPENSATE(t7);\
366 d8 = COMPENSATE(t8);}
368 /** inverse slant4 transform */
369 #define IVI_INV_SLANT4(s1, s4, s2, s3, d1, d2, d3, d4, t0, t1, t2, t3, t4) {\
370 IVI_SLANT_BFLY(s1, s2, t1, t2, t0); IVI_IREFLECT (s4, s3, t4, t3, t0);\
372 IVI_SLANT_BFLY(t1, t4, t1, t4, t0); IVI_SLANT_BFLY(t2, t3, t2, t3, t0);\
373 d1 = COMPENSATE(t1);\
374 d2 = COMPENSATE(t2);\
375 d3 = COMPENSATE(t3);\
376 d4 = COMPENSATE(t4);}
378 void ff_ivi_inverse_slant_8x8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
384 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
386 #define COMPENSATE(x) (x)
389 for (i = 0; i < 8; i++) {
391 IVI_INV_SLANT8(src[0], src[8], src[16], src[24], src[32], src[40], src[48], src[56],
392 dst[0], dst[8], dst[16], dst[24], dst[32], dst[40], dst[48], dst[56],
393 t0, t1, t2, t3, t4, t5, t6, t7, t8);
395 dst[0] = dst[8] = dst[16] = dst[24] = dst[32] = dst[40] = dst[48] = dst[56] = 0;
402 #define COMPENSATE(x) ((x + 1)>>1)
404 for (i = 0; i < 8; i++) {
405 if (!src[0] && !src[1] && !src[2] && !src[3] && !src[4] && !src[5] && !src[6] && !src[7]) {
406 memset(out, 0, 8*sizeof(out[0]));
408 IVI_INV_SLANT8(src[0], src[1], src[2], src[3], src[4], src[5], src[6], src[7],
409 out[0], out[1], out[2], out[3], out[4], out[5], out[6], out[7],
410 t0, t1, t2, t3, t4, t5, t6, t7, t8);
418 void ff_ivi_inverse_slant_4x4(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
424 int t0, t1, t2, t3, t4;
426 #define COMPENSATE(x) (x)
429 for (i = 0; i < 4; i++) {
431 IVI_INV_SLANT4(src[0], src[4], src[8], src[12],
432 dst[0], dst[4], dst[8], dst[12],
435 dst[0] = dst[4] = dst[8] = dst[12] = 0;
442 #define COMPENSATE(x) ((x + 1)>>1)
444 for (i = 0; i < 4; i++) {
445 if (!src[0] && !src[1] && !src[2] && !src[3]) {
446 out[0] = out[1] = out[2] = out[3] = 0;
448 IVI_INV_SLANT4(src[0], src[1], src[2], src[3],
449 out[0], out[1], out[2], out[3],
458 void ff_ivi_dc_slant_2d(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
463 dc_coeff = (*in + 1) >> 1;
465 for (y = 0; y < blk_size; out += pitch, y++) {
466 for (x = 0; x < blk_size; x++)
471 void ff_ivi_row_slant8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
474 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
476 #define COMPENSATE(x) ((x + 1)>>1)
477 for (i = 0; i < 8; i++) {
478 if (!in[0] && !in[1] && !in[2] && !in[3] && !in[4] && !in[5] && !in[6] && !in[7]) {
479 memset(out, 0, 8*sizeof(out[0]));
481 IVI_INV_SLANT8( in[0], in[1], in[2], in[3], in[4], in[5], in[6], in[7],
482 out[0], out[1], out[2], out[3], out[4], out[5], out[6], out[7],
483 t0, t1, t2, t3, t4, t5, t6, t7, t8);
491 void ff_ivi_dc_row_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
496 dc_coeff = (*in + 1) >> 1;
498 for (x = 0; x < blk_size; x++)
503 for (y = 1; y < blk_size; out += pitch, y++) {
504 for (x = 0; x < blk_size; x++)
509 void ff_ivi_col_slant8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
511 int i, row2, row4, row8;
512 int t0, t1, t2, t3, t4, t5, t6, t7, t8;
518 #define COMPENSATE(x) ((x + 1)>>1)
519 for (i = 0; i < 8; i++) {
521 IVI_INV_SLANT8(in[0], in[8], in[16], in[24], in[32], in[40], in[48], in[56],
522 out[0], out[pitch], out[row2], out[row2 + pitch], out[row4],
523 out[row4 + pitch], out[row4 + row2], out[row8 - pitch],
524 t0, t1, t2, t3, t4, t5, t6, t7, t8);
526 out[0] = out[pitch] = out[row2] = out[row2 + pitch] = out[row4] =
527 out[row4 + pitch] = out[row4 + row2] = out[row8 - pitch] = 0;
536 void ff_ivi_dc_col_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
541 dc_coeff = (*in + 1) >> 1;
543 for (y = 0; y < blk_size; out += pitch, y++) {
545 for (x = 1; x < blk_size; x++)
550 void ff_ivi_put_pixels_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
551 const uint8_t *flags)
555 for (y = 0; y < 8; out += pitch, in += 8, y++)
556 for (x = 0; x < 8; x++)
560 void ff_ivi_put_dc_pixel_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
566 memset(out + 1, 0, 7*sizeof(out[0]));
569 for (y = 1; y < 8; out += pitch, y++)
570 memset(out, 0, 8*sizeof(out[0]));
573 #define IVI_MC_TEMPLATE(size, suffix, OP) \
574 void ff_ivi_mc_ ## size ##x## size ## suffix (int16_t *buf, const int16_t *ref_buf, \
575 uint32_t pitch, int mc_type) \
578 const int16_t *wptr; \
581 case 0: /* fullpel (no interpolation) */ \
582 for (i = 0; i < size; i++, buf += pitch, ref_buf += pitch) { \
583 for (j = 0; j < size; j++) {\
584 OP(buf[j], ref_buf[j]); \
588 case 1: /* horizontal halfpel interpolation */ \
589 for (i = 0; i < size; i++, buf += pitch, ref_buf += pitch) \
590 for (j = 0; j < size; j++) \
591 OP(buf[j], (ref_buf[j] + ref_buf[j+1]) >> 1); \
593 case 2: /* vertical halfpel interpolation */ \
594 wptr = ref_buf + pitch; \
595 for (i = 0; i < size; i++, buf += pitch, wptr += pitch, ref_buf += pitch) \
596 for (j = 0; j < size; j++) \
597 OP(buf[j], (ref_buf[j] + wptr[j]) >> 1); \
599 case 3: /* vertical and horizontal halfpel interpolation */ \
600 wptr = ref_buf + pitch; \
601 for (i = 0; i < size; i++, buf += pitch, wptr += pitch, ref_buf += pitch) \
602 for (j = 0; j < size; j++) \
603 OP(buf[j], (ref_buf[j] + ref_buf[j+1] + wptr[j] + wptr[j+1]) >> 2); \
608 #define OP_PUT(a, b) (a) = (b)
609 #define OP_ADD(a, b) (a) += (b)
611 IVI_MC_TEMPLATE(8, _no_delta, OP_PUT)
612 IVI_MC_TEMPLATE(8, _delta, OP_ADD)
613 IVI_MC_TEMPLATE(4, _no_delta, OP_PUT)
614 IVI_MC_TEMPLATE(4, _delta, OP_ADD)