4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "libavutil/imgutils.h"
24 #include "libavutil/intreadwrite.h"
32 /* HQX has four modes - 422, 444, 422alpha and 444alpha - all 12-bit */
40 #define HQX_HEADER_SIZE 59
42 typedef int (*mb_decode_func)(HQXContext *ctx, AVFrame *pic,
43 GetBitContext *gb, int x, int y);
45 /* macroblock selects a group of 4 possible quants and
46 * a block can use any of those four quantisers
47 * one column is powers of 2, the other one is powers of 2 * 3,
48 * then there is the special one, powers of 2 * 5 */
49 static const int hqx_quants[16][4] = {
50 { 0x1, 0x2, 0x4, 0x8 }, { 0x1, 0x3, 0x6, 0xC },
51 { 0x2, 0x4, 0x8, 0x10 }, { 0x3, 0x6, 0xC, 0x18 },
52 { 0x4, 0x8, 0x10, 0x20 }, { 0x6, 0xC, 0x18, 0x30 },
53 { 0x8, 0x10, 0x20, 0x40 },
54 { 0xA, 0x14, 0x28, 0x50 },
55 { 0xC, 0x18, 0x30, 0x60 },
56 { 0x10, 0x20, 0x40, 0x80 }, { 0x18, 0x30, 0x60, 0xC0 },
57 { 0x20, 0x40, 0x80, 0x100 }, { 0x30, 0x60, 0xC0, 0x180 },
58 { 0x40, 0x80, 0x100, 0x200 }, { 0x60, 0xC0, 0x180, 0x300 },
59 { 0x80, 0x100, 0x200, 0x400 }
62 static const uint8_t hqx_quant_luma[64] = {
63 16, 16, 16, 19, 19, 19, 42, 44,
64 16, 16, 19, 19, 19, 38, 43, 45,
65 16, 19, 19, 19, 40, 41, 45, 48,
66 19, 19, 19, 40, 41, 42, 46, 49,
67 19, 19, 40, 41, 42, 43, 48, 101,
68 19, 38, 41, 42, 43, 44, 98, 104,
69 42, 43, 45, 46, 48, 98, 109, 116,
70 44, 45, 48, 49, 101, 104, 116, 123,
73 static const uint8_t hqx_quant_chroma[64] = {
74 16, 16, 19, 25, 26, 26, 42, 44,
75 16, 19, 25, 25, 26, 38, 43, 91,
76 19, 25, 26, 27, 40, 41, 91, 96,
77 25, 25, 27, 40, 41, 84, 93, 197,
78 26, 26, 40, 41, 84, 86, 191, 203,
79 26, 38, 41, 84, 86, 177, 197, 209,
80 42, 43, 91, 93, 191, 197, 219, 232,
81 44, 91, 96, 197, 203, 209, 232, 246,
84 static inline void idct_col(int16_t *blk, const uint8_t *quant)
86 int t0, t1, t2, t3, t4, t5, t6, t7, t8, t9, tA, tB, tC, tD, tE, tF;
87 int t10, t11, t12, t13;
88 int s0, s1, s2, s3, s4, s5, s6, s7;
90 s0 = (int) blk[0 * 8] * quant[0 * 8];
91 s1 = (int) blk[1 * 8] * quant[1 * 8];
92 s2 = (int) blk[2 * 8] * quant[2 * 8];
93 s3 = (int) blk[3 * 8] * quant[3 * 8];
94 s4 = (int) blk[4 * 8] * quant[4 * 8];
95 s5 = (int) blk[5 * 8] * quant[5 * 8];
96 s6 = (int) blk[6 * 8] * quant[6 * 8];
97 s7 = (int) blk[7 * 8] * quant[7 * 8];
99 t0 = (s3 * 19266 + s5 * 12873) >> 15;
100 t1 = (s5 * 19266 - s3 * 12873) >> 15;
101 t2 = ((s7 * 4520 + s1 * 22725) >> 15) - t0;
102 t3 = ((s1 * 4520 - s7 * 22725) >> 15) - t1;
107 t8 = (t6 * 11585) >> 14;
108 t9 = (t7 * 11585) >> 14;
109 tA = (s2 * 8867 - s6 * 21407) >> 14;
110 tB = (s6 * 8867 + s2 * 21407) >> 14;
111 tC = (s0 >> 1) - (s4 >> 1);
112 tD = (s4 >> 1) * 2 + tC;
117 t12 = tE + (tA >> 1) * 2 - t9;
118 t13 = tF + (tB >> 1) * 2 - t4;
120 blk[0 * 8] = t13 + t4 * 2;
121 blk[1 * 8] = t12 + t9 * 2;
122 blk[2 * 8] = t11 + t8 * 2;
123 blk[3 * 8] = t10 + t5 * 2;
130 static inline void idct_row(int16_t *blk)
132 int t0, t1, t2, t3, t4, t5, t6, t7, t8, t9, tA, tB, tC, tD, tE, tF;
133 int t10, t11, t12, t13;
135 t0 = (blk[3] * 19266 + blk[5] * 12873) >> 14;
136 t1 = (blk[5] * 19266 - blk[3] * 12873) >> 14;
137 t2 = ((blk[7] * 4520 + blk[1] * 22725) >> 14) - t0;
138 t3 = ((blk[1] * 4520 - blk[7] * 22725) >> 14) - t1;
143 t8 = (t6 * 11585) >> 14;
144 t9 = (t7 * 11585) >> 14;
145 tA = (blk[2] * 8867 - blk[6] * 21407) >> 14;
146 tB = (blk[6] * 8867 + blk[2] * 21407) >> 14;
147 tC = blk[0] - blk[4];
148 tD = blk[4] * 2 + tC;
153 t12 = tE + tA * 2 - t9;
154 t13 = tF + tB * 2 - t4;
156 blk[0] = (t13 + t4 * 2 + 4) >> 3;
157 blk[1] = (t12 + t9 * 2 + 4) >> 3;
158 blk[2] = (t11 + t8 * 2 + 4) >> 3;
159 blk[3] = (t10 + t5 * 2 + 4) >> 3;
160 blk[4] = (t10 + 4) >> 3;
161 blk[5] = (t11 + 4) >> 3;
162 blk[6] = (t12 + 4) >> 3;
163 blk[7] = (t13 + 4) >> 3;
166 static void hqx_idct(int16_t *block, const uint8_t *quant)
170 for (i = 0; i < 8; i++)
171 idct_col(block + i, quant + i);
172 for (i = 0; i < 8; i++)
173 idct_row(block + i * 8);
176 static void hqx_idct_put(uint16_t *dst, ptrdiff_t stride,
177 int16_t *block, const uint8_t *quant)
181 hqx_idct(block, quant);
183 for (i = 0; i < 8; i++) {
184 for (j = 0; j < 8; j++) {
185 int v = av_clip(block[j + i * 8] + 0x800, 0, 0x1000);
186 dst[j] = (v << 4) | (v >> 8);
192 static inline void put_blocks(AVFrame *pic, int plane,
193 int x, int y, int ilace,
194 int16_t *block0, int16_t *block1,
195 const uint8_t *quant)
197 int fields = ilace ? 2 : 1;
198 int lsize = pic->linesize[plane];
199 uint8_t *p = pic->data[plane] + x * 2;
201 hqx_idct_put((uint16_t *)(p + y * lsize), lsize * fields, block0, quant);
202 hqx_idct_put((uint16_t *)(p + (y + (ilace ? 1 : 8)) * lsize),
203 lsize * fields, block1, quant);
206 static inline void hqx_get_ac(GetBitContext *gb, const HQXAC *ac,
211 val = show_bits(gb, ac->lut_bits);
212 if (ac->lut[val].bits == -1) {
213 GetBitContext gb2 = *gb;
214 skip_bits(&gb2, ac->lut_bits);
215 val = ac->lut[val].lev + show_bits(&gb2, ac->extra_bits);
217 *run = ac->lut[val].run;
218 *lev = ac->lut[val].lev;
219 skip_bits(gb, ac->lut[val].bits);
222 static int decode_block(GetBitContext *gb, VLC *vlc,
223 const int *quants, int dcb,
224 int16_t block[64], int *last_dc)
228 int run, lev, pos = 1;
230 memset(block, 0, 64 * sizeof(*block));
231 dc = get_vlc2(gb, vlc->table, HQX_DC_VLC_BITS, 2);
233 return AVERROR_INVALIDDATA;
236 block[0] = sign_extend(*last_dc << (12 - dcb), 12);
238 q = quants[get_bits(gb, 2)];
240 ac_idx = HQX_AC_Q128;
253 hqx_get_ac(gb, &ff_hqx_ac[ac_idx], &run, &lev);
257 block[ff_zigzag_direct[pos++]] = lev * q;
263 static int hqx_decode_422(HQXContext *ctx, AVFrame *pic,
264 GetBitContext *gb, int x, int y)
272 flag = get_bits1(gb);
276 quants = hqx_quants[get_bits(gb, 4)];
278 for (i = 0; i < 8; i++) {
279 int vlc_index = ctx->dcb - 9;
280 if (i == 0 || i == 4 || i == 6)
282 ret = decode_block(gb, &ctx->dc_vlc[vlc_index], quants,
283 ctx->dcb, ctx->block[i], &last_dc);
288 put_blocks(pic, 0, x, y, flag, ctx->block[0], ctx->block[2], hqx_quant_luma);
289 put_blocks(pic, 0, x + 8, y, flag, ctx->block[1], ctx->block[3], hqx_quant_luma);
290 put_blocks(pic, 2, x >> 1, y, flag, ctx->block[4], ctx->block[5], hqx_quant_chroma);
291 put_blocks(pic, 1, x >> 1, y, flag, ctx->block[6], ctx->block[7], hqx_quant_chroma);
296 static int hqx_decode_422a(HQXContext *ctx, AVFrame *pic,
297 GetBitContext *gb, int x, int y)
305 cbp = get_vlc2(gb, ctx->cbp_vlc.table, ctx->cbp_vlc.bits, 1);
307 for (i = 0; i < 12; i++)
308 memset(ctx->block[i], 0, sizeof(**ctx->block) * 64);
309 for (i = 0; i < 12; i++)
310 ctx->block[i][0] = -0x800;
313 flag = get_bits1(gb);
315 quants = hqx_quants[get_bits(gb, 4)];
317 cbp |= cbp << 4; // alpha CBP
318 if (cbp & 0x3) // chroma CBP - top
320 if (cbp & 0xC) // chroma CBP - bottom
322 for (i = 0; i < 12; i++) {
323 if (i == 0 || i == 4 || i == 8 || i == 10)
325 if (cbp & (1 << i)) {
326 int vlc_index = ctx->dcb - 9;
327 ret = decode_block(gb, &ctx->dc_vlc[vlc_index], quants,
328 ctx->dcb, ctx->block[i], &last_dc);
335 put_blocks(pic, 3, x, y, flag, ctx->block[ 0], ctx->block[ 2], hqx_quant_luma);
336 put_blocks(pic, 3, x + 8, y, flag, ctx->block[ 1], ctx->block[ 3], hqx_quant_luma);
337 put_blocks(pic, 0, x, y, flag, ctx->block[ 4], ctx->block[ 6], hqx_quant_luma);
338 put_blocks(pic, 0, x + 8, y, flag, ctx->block[ 5], ctx->block[ 7], hqx_quant_luma);
339 put_blocks(pic, 2, x >> 1, y, flag, ctx->block[ 8], ctx->block[ 9], hqx_quant_chroma);
340 put_blocks(pic, 1, x >> 1, y, flag, ctx->block[10], ctx->block[11], hqx_quant_chroma);
345 static int hqx_decode_444(HQXContext *ctx, AVFrame *pic,
346 GetBitContext *gb, int x, int y)
354 flag = get_bits1(gb);
358 quants = hqx_quants[get_bits(gb, 4)];
360 for (i = 0; i < 12; i++) {
361 int vlc_index = ctx->dcb - 9;
362 if (i == 0 || i == 4 || i == 8)
364 ret = decode_block(gb, &ctx->dc_vlc[vlc_index], quants,
365 ctx->dcb, ctx->block[i], &last_dc);
370 put_blocks(pic, 0, x, y, flag, ctx->block[0], ctx->block[ 2], hqx_quant_luma);
371 put_blocks(pic, 0, x + 8, y, flag, ctx->block[1], ctx->block[ 3], hqx_quant_luma);
372 put_blocks(pic, 2, x, y, flag, ctx->block[4], ctx->block[ 6], hqx_quant_chroma);
373 put_blocks(pic, 2, x + 8, y, flag, ctx->block[5], ctx->block[ 7], hqx_quant_chroma);
374 put_blocks(pic, 1, x, y, flag, ctx->block[8], ctx->block[10], hqx_quant_chroma);
375 put_blocks(pic, 1, x + 8, y, flag, ctx->block[9], ctx->block[11], hqx_quant_chroma);
380 static int hqx_decode_444a(HQXContext *ctx, AVFrame *pic,
381 GetBitContext *gb, int x, int y)
389 cbp = get_vlc2(gb, ctx->cbp_vlc.table, ctx->cbp_vlc.bits, 1);
391 for (i = 0; i < 16; i++)
392 memset(ctx->block[i], 0, sizeof(**ctx->block) * 64);
393 for (i = 0; i < 16; i++)
394 ctx->block[i][0] = -0x800;
397 flag = get_bits1(gb);
399 quants = hqx_quants[get_bits(gb, 4)];
401 cbp |= cbp << 4; // alpha CBP
402 cbp |= cbp << 8; // chroma CBP
403 for (i = 0; i < 16; i++) {
404 if (i == 0 || i == 4 || i == 8 || i == 12)
406 if (cbp & (1 << i)) {
407 int vlc_index = ctx->dcb - 9;
408 ret = decode_block(gb, &ctx->dc_vlc[vlc_index], quants,
409 ctx->dcb, ctx->block[i], &last_dc);
416 put_blocks(pic, 3, x, y, flag, ctx->block[ 0], ctx->block[ 2], hqx_quant_luma);
417 put_blocks(pic, 3, x + 8, y, flag, ctx->block[ 1], ctx->block[ 3], hqx_quant_luma);
418 put_blocks(pic, 0, x, y, flag, ctx->block[ 4], ctx->block[ 6], hqx_quant_luma);
419 put_blocks(pic, 0, x + 8, y, flag, ctx->block[ 5], ctx->block[ 7], hqx_quant_luma);
420 put_blocks(pic, 2, x, y, flag, ctx->block[ 8], ctx->block[10], hqx_quant_chroma);
421 put_blocks(pic, 2, x + 8, y, flag, ctx->block[ 9], ctx->block[11], hqx_quant_chroma);
422 put_blocks(pic, 1, x, y, flag, ctx->block[12], ctx->block[14], hqx_quant_chroma);
423 put_blocks(pic, 1, x + 8, y, flag, ctx->block[13], ctx->block[15], hqx_quant_chroma);
428 static const int shuffle_16[16] = {
429 0, 5, 11, 14, 2, 7, 9, 13, 1, 4, 10, 15, 3, 6, 8, 12
432 static int decode_slice(HQXContext *ctx, AVFrame *pic, GetBitContext *gb,
433 int slice_no, mb_decode_func decode_func)
435 int mb_w = (ctx->width + 15) >> 4;
436 int mb_h = (ctx->height + 15) >> 4;
437 int grp_w = (mb_w + 4) / 5;
438 int grp_h = (mb_h + 4) / 5;
439 int grp_h_edge = grp_w * (mb_w / grp_w);
440 int grp_v_edge = grp_h * (mb_h / grp_h);
441 int grp_v_rest = mb_w - grp_h_edge;
442 int grp_h_rest = mb_h - grp_v_edge;
443 int num_mbs = mb_w * mb_h;
444 int num_tiles = (num_mbs + 479) / 480;
445 int std_tile_blocks = num_mbs / (16 * num_tiles);
446 int g_tile = slice_no * num_tiles;
447 int blk_addr, loc_addr, mb_x, mb_y, pos, loc_row, i;
448 int tile_blocks, tile_limit, tile_no;
450 for (tile_no = 0; tile_no < num_tiles; tile_no++, g_tile++) {
451 tile_blocks = std_tile_blocks;
453 if (g_tile < num_mbs - std_tile_blocks * 16 * num_tiles) {
454 tile_limit = num_mbs / (16 * num_tiles);
457 for (i = 0; i < tile_blocks; i++) {
459 blk_addr = g_tile + 16 * num_tiles * i;
461 blk_addr = tile_no + 16 * num_tiles * i +
462 num_tiles * shuffle_16[(i + slice_no) & 0xF];
463 loc_row = grp_h * (blk_addr / (grp_h * mb_w));
464 loc_addr = blk_addr % (grp_h * mb_w);
465 if (loc_row >= grp_v_edge) {
466 mb_x = grp_w * (loc_addr / (grp_h_rest * grp_w));
467 pos = loc_addr % (grp_h_rest * grp_w);
469 mb_x = grp_w * (loc_addr / (grp_h * grp_w));
470 pos = loc_addr % (grp_h * grp_w);
472 if (mb_x >= grp_h_edge) {
473 mb_x += pos % grp_v_rest;
474 mb_y = loc_row + (pos / grp_v_rest);
477 mb_y = loc_row + (pos / grp_w);
479 decode_func(ctx, pic, gb, mb_x * 16, mb_y * 16);
486 static int hqx_decode_frame(AVCodecContext *avctx, void *data,
487 int *got_picture_ptr, AVPacket *avpkt)
489 HQXContext *ctx = avctx->priv_data;
491 uint8_t *src = avpkt->data;
492 uint32_t info_tag, info_offset;
498 uint32_t slice_off[17];
499 mb_decode_func decode_func = 0;
502 return AVERROR_INVALIDDATA;
504 /* Skip the INFO header if present */
506 info_tag = AV_RL32(src);
507 if (info_tag == MKTAG('I', 'N', 'F', 'O')) {
508 info_offset = AV_RL32(src + 4);
509 if (info_offset > UINT32_MAX - 8 || info_offset + 8 > avpkt->size) {
510 av_log(avctx, AV_LOG_ERROR,
511 "Invalid INFO header offset: 0x%08"PRIX32" is too large.\n",
513 return AVERROR_INVALIDDATA;
519 av_log(avctx, AV_LOG_DEBUG, "Skipping INFO chunk.\n");
522 data_start = src - avpkt->data;
523 data_size = avpkt->size - data_start;
525 if (data_size < HQX_HEADER_SIZE) {
526 av_log(avctx, AV_LOG_ERROR, "Frame too small.\n");
527 return AVERROR_INVALIDDATA;
530 if (src[0] != 'H' || src[1] != 'Q') {
531 av_log(avctx, AV_LOG_ERROR, "Not an HQX frame.\n");
532 return AVERROR_INVALIDDATA;
534 ctx->interlaced = !(src[2] & 0x80);
535 ctx->format = src[2] & 7;
536 ctx->dcb = (src[3] & 3) + 8;
537 ctx->width = AV_RB16(src + 4);
538 ctx->height = AV_RB16(src + 6);
539 for (i = 0; i < 17; i++)
540 slice_off[i] = AV_RB24(src + 8 + i * 3);
543 av_log(avctx, AV_LOG_ERROR, "Invalid DC precision %d.\n", ctx->dcb);
544 return AVERROR_INVALIDDATA;
546 ret = av_image_check_size(ctx->width, ctx->height, 0, avctx);
548 av_log(avctx, AV_LOG_ERROR, "Invalid stored dimensions %dx%d.\n",
549 ctx->width, ctx->height);
550 return AVERROR_INVALIDDATA;
553 avctx->coded_width = FFALIGN(ctx->width, 16);
554 avctx->coded_height = FFALIGN(ctx->height, 16);
555 avctx->width = ctx->width;
556 avctx->height = ctx->height;
557 avctx->bits_per_raw_sample = 10;
559 switch (ctx->format) {
561 avctx->pix_fmt = AV_PIX_FMT_YUV422P16;
562 decode_func = hqx_decode_422;
565 avctx->pix_fmt = AV_PIX_FMT_YUV444P16;
566 decode_func = hqx_decode_444;
569 avctx->pix_fmt = AV_PIX_FMT_YUVA422P16;
570 decode_func = hqx_decode_422a;
573 avctx->pix_fmt = AV_PIX_FMT_YUVA444P16;
574 decode_func = hqx_decode_444a;
578 av_log(avctx, AV_LOG_ERROR, "Invalid format: %d.\n", ctx->format);
579 return AVERROR_INVALIDDATA;
582 ret = ff_get_buffer(avctx, pic, 0);
586 for (slice = 0; slice < 16; slice++) {
587 if (slice_off[slice] < HQX_HEADER_SIZE ||
588 slice_off[slice] >= slice_off[slice + 1] ||
589 slice_off[slice + 1] > data_size) {
590 av_log(avctx, AV_LOG_ERROR, "Invalid slice size.\n");
593 ret = init_get_bits8(&gb, src + slice_off[slice],
594 slice_off[slice + 1] - slice_off[slice]);
597 ret = decode_slice(ctx, pic, &gb, slice, decode_func);
599 av_log(avctx, AV_LOG_ERROR, "Error decoding slice %d.\n", slice);
604 pic->pict_type = AV_PICTURE_TYPE_I;
606 *got_picture_ptr = 1;
611 static av_cold int hqx_decode_close(AVCodecContext *avctx)
614 HQXContext *ctx = avctx->priv_data;
616 ff_free_vlc(&ctx->cbp_vlc);
617 for (i = 0; i < 3; i++) {
618 ff_free_vlc(&ctx->dc_vlc[i]);
624 static av_cold int hqx_decode_init(AVCodecContext *avctx)
626 HQXContext *ctx = avctx->priv_data;
627 int ret = ff_hqx_init_vlcs(ctx);
629 hqx_decode_close(avctx);
633 AVCodec ff_hqx_decoder = {
635 .long_name = NULL_IF_CONFIG_SMALL("Canopus HQX"),
636 .type = AVMEDIA_TYPE_VIDEO,
637 .id = AV_CODEC_ID_HQX,
638 .priv_data_size = sizeof(HQXContext),
639 .init = hqx_decode_init,
640 .decode = hqx_decode_frame,
641 .close = hqx_decode_close,
642 .capabilities = CODEC_CAP_DR1,