2 * Indeo Video v3 compatible decoder
3 * Copyright (c) 2009 - 2011 Maxim Poliakovski
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * This is a decoder for Intel Indeo Video v3.
25 * It is based on vector quantization, run-length coding and motion compensation.
26 * Known container formats: .avi and .mov
27 * Known FOURCCs: 'IV31', 'IV32'
29 * @see http://wiki.multimedia.cx/index.php?title=Indeo_3
32 #include "libavutil/imgutils.h"
33 #include "libavutil/intreadwrite.h"
35 #include "copy_block.h"
37 #include "bytestream.h"
41 #include "indeo3data.h"
45 RLE_ESC_F9 = 249, ///< same as RLE_ESC_FA + do the same with next block
46 RLE_ESC_FA = 250, ///< INTRA: skip block, INTER: copy data from reference
47 RLE_ESC_FB = 251, ///< apply null delta to N blocks / skip N blocks
48 RLE_ESC_FC = 252, ///< same as RLE_ESC_FD + do the same with next block
49 RLE_ESC_FD = 253, ///< apply null delta to all remaining lines of this block
50 RLE_ESC_FE = 254, ///< apply null delta to all lines up to the 3rd line
51 RLE_ESC_FF = 255 ///< apply null delta to all lines up to the 2nd line
55 /* Some constants for parsing frame bitstream flags. */
56 #define BS_8BIT_PEL (1 << 1) ///< 8bit pixel bitdepth indicator
57 #define BS_KEYFRAME (1 << 2) ///< intra frame indicator
58 #define BS_MV_Y_HALF (1 << 4) ///< vertical mv halfpel resolution indicator
59 #define BS_MV_X_HALF (1 << 5) ///< horizontal mv halfpel resolution indicator
60 #define BS_NONREF (1 << 8) ///< nonref (discardable) frame indicator
61 #define BS_BUFFER 9 ///< indicates which of two frame buffers should be used
64 typedef struct Plane {
66 uint8_t *pixels[2]; ///< pointer to the actual pixel data of the buffers above
72 #define CELL_STACK_MAX 20
75 int16_t xpos; ///< cell coordinates in 4x4 blocks
77 int16_t width; ///< cell width in 4x4 blocks
78 int16_t height; ///< cell height in 4x4 blocks
79 uint8_t tree; ///< tree id: 0- MC tree, 1 - VQ tree
80 const int8_t *mv_ptr; ///< ptr to the motion vector if any
83 typedef struct Indeo3DecodeContext {
84 AVCodecContext *avctx;
91 const uint8_t *next_cell_data;
92 const uint8_t *last_byte;
93 const int8_t *mc_vectors;
94 unsigned num_vectors; ///< number of motion vectors in mc_vectors
96 int16_t width, height;
97 uint32_t frame_num; ///< current frame number (zero-based)
98 uint32_t data_size; ///< size of the frame data in bytes
99 uint16_t frame_flags; ///< frame properties
100 uint8_t cb_offset; ///< needed for selecting VQ tables
101 uint8_t buf_sel; ///< active frame buffer: 0 - primary, 1 -secondary
102 const uint8_t *y_data_ptr;
103 const uint8_t *v_data_ptr;
104 const uint8_t *u_data_ptr;
108 const uint8_t *alt_quant; ///< secondary VQ table set for the modes 1 and 4
110 } Indeo3DecodeContext;
113 static uint8_t requant_tab[8][128];
116 * Build the static requantization table.
117 * This table is used to remap pixel values according to a specific
118 * quant index and thus avoid overflows while adding deltas.
120 static av_cold void build_requant_tab(void)
122 static int8_t offsets[8] = { 1, 1, 2, -3, -3, 3, 4, 4 };
123 static int8_t deltas [8] = { 0, 1, 0, 4, 4, 1, 0, 1 };
127 for (i = 0; i < 8; i++) {
129 for (j = 0; j < 128; j++)
130 requant_tab[i][j] = (j + offsets[i]) / step * step + deltas[i];
133 /* some last elements calculated above will have values >= 128 */
134 /* pixel values shall never exceed 127 so set them to non-overflowing values */
135 /* according with the quantization step of the respective section */
136 requant_tab[0][127] = 126;
137 requant_tab[1][119] = 118;
138 requant_tab[1][120] = 118;
139 requant_tab[2][126] = 124;
140 requant_tab[2][127] = 124;
141 requant_tab[6][124] = 120;
142 requant_tab[6][125] = 120;
143 requant_tab[6][126] = 120;
144 requant_tab[6][127] = 120;
146 /* Patch for compatibility with the Intel's binary decoders */
147 requant_tab[1][7] = 10;
148 requant_tab[4][8] = 10;
152 static av_cold int allocate_frame_buffers(Indeo3DecodeContext *ctx,
153 AVCodecContext *avctx, int luma_width, int luma_height)
155 int p, chroma_width, chroma_height;
156 int luma_pitch, chroma_pitch, luma_size, chroma_size;
158 if (luma_width < 16 || luma_width > 640 ||
159 luma_height < 16 || luma_height > 480 ||
160 luma_width & 3 || luma_height & 3) {
161 av_log(avctx, AV_LOG_ERROR, "Invalid picture dimensions: %d x %d!\n",
162 luma_width, luma_height);
163 return AVERROR_INVALIDDATA;
166 ctx->width = luma_width ;
167 ctx->height = luma_height;
169 chroma_width = FFALIGN(luma_width >> 2, 4);
170 chroma_height = FFALIGN(luma_height >> 2, 4);
172 luma_pitch = FFALIGN(luma_width, 16);
173 chroma_pitch = FFALIGN(chroma_width, 16);
175 /* Calculate size of the luminance plane. */
176 /* Add one line more for INTRA prediction. */
177 luma_size = luma_pitch * (luma_height + 1);
179 /* Calculate size of a chrominance planes. */
180 /* Add one line more for INTRA prediction. */
181 chroma_size = chroma_pitch * (chroma_height + 1);
183 /* allocate frame buffers */
184 for (p = 0; p < 3; p++) {
185 ctx->planes[p].pitch = !p ? luma_pitch : chroma_pitch;
186 ctx->planes[p].width = !p ? luma_width : chroma_width;
187 ctx->planes[p].height = !p ? luma_height : chroma_height;
189 ctx->planes[p].buffers[0] = av_malloc(!p ? luma_size : chroma_size);
190 ctx->planes[p].buffers[1] = av_malloc(!p ? luma_size : chroma_size);
192 /* fill the INTRA prediction lines with the middle pixel value = 64 */
193 memset(ctx->planes[p].buffers[0], 0x40, ctx->planes[p].pitch);
194 memset(ctx->planes[p].buffers[1], 0x40, ctx->planes[p].pitch);
196 /* set buffer pointers = buf_ptr + pitch and thus skip the INTRA prediction line */
197 ctx->planes[p].pixels[0] = ctx->planes[p].buffers[0] + ctx->planes[p].pitch;
198 ctx->planes[p].pixels[1] = ctx->planes[p].buffers[1] + ctx->planes[p].pitch;
199 memset(ctx->planes[p].pixels[0], 0, ctx->planes[p].pitch * ctx->planes[p].height);
200 memset(ctx->planes[p].pixels[1], 0, ctx->planes[p].pitch * ctx->planes[p].height);
207 static av_cold void free_frame_buffers(Indeo3DecodeContext *ctx)
214 for (p = 0; p < 3; p++) {
215 av_freep(&ctx->planes[p].buffers[0]);
216 av_freep(&ctx->planes[p].buffers[1]);
217 ctx->planes[p].pixels[0] = ctx->planes[p].pixels[1] = 0;
223 * Copy pixels of the cell(x + mv_x, y + mv_y) from the previous frame into
224 * the cell(x, y) in the current frame.
226 * @param ctx pointer to the decoder context
227 * @param plane pointer to the plane descriptor
228 * @param cell pointer to the cell descriptor
230 static void copy_cell(Indeo3DecodeContext *ctx, Plane *plane, Cell *cell)
232 int h, w, mv_x, mv_y, offset, offset_dst;
235 /* setup output and reference pointers */
236 offset_dst = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
237 dst = plane->pixels[ctx->buf_sel] + offset_dst;
239 mv_y = cell->mv_ptr[0];
240 mv_x = cell->mv_ptr[1];
243 offset = offset_dst + mv_y * plane->pitch + mv_x;
244 src = plane->pixels[ctx->buf_sel ^ 1] + offset;
246 h = cell->height << 2;
248 for (w = cell->width; w > 0;) {
249 /* copy using 16xH blocks */
250 if (!((cell->xpos << 2) & 15) && w >= 4) {
251 for (; w >= 4; src += 16, dst += 16, w -= 4)
252 ctx->dsp.put_no_rnd_pixels_tab[0][0](dst, src, plane->pitch, h);
255 /* copy using 8xH blocks */
256 if (!((cell->xpos << 2) & 7) && w >= 2) {
257 ctx->dsp.put_no_rnd_pixels_tab[1][0](dst, src, plane->pitch, h);
264 copy_block4(dst, src, plane->pitch, plane->pitch, h);
273 /* Average 4/8 pixels at once without rounding using SWAR */
274 #define AVG_32(dst, src, ref) \
275 AV_WN32A(dst, ((AV_RN32A(src) + AV_RN32A(ref)) >> 1) & 0x7F7F7F7FUL)
277 #define AVG_64(dst, src, ref) \
278 AV_WN64A(dst, ((AV_RN64A(src) + AV_RN64A(ref)) >> 1) & 0x7F7F7F7F7F7F7F7FULL)
282 * Replicate each even pixel as follows:
283 * ABCDEFGH -> AACCEEGG
285 static inline uint64_t replicate64(uint64_t a) {
287 a &= 0xFF00FF00FF00FF00ULL;
290 a &= 0x00FF00FF00FF00FFULL;
296 static inline uint32_t replicate32(uint32_t a) {
308 /* Fill n lines with 64bit pixel value pix */
309 static inline void fill_64(uint8_t *dst, const uint64_t pix, int32_t n,
312 for (; n > 0; dst += row_offset, n--)
317 /* Error codes for cell decoding. */
328 #define BUFFER_PRECHECK \
329 if (*data_ptr >= last_ptr) \
330 return IV3_OUT_OF_DATA; \
332 #define RLE_BLOCK_COPY \
333 if (cell->mv_ptr || !skip_flag) \
334 copy_block4(dst, ref, row_offset, row_offset, 4 << v_zoom)
336 #define RLE_BLOCK_COPY_8 \
337 pix64 = AV_RN64A(ref);\
338 if (is_first_row) {/* special prediction case: top line of a cell */\
339 pix64 = replicate64(pix64);\
340 fill_64(dst + row_offset, pix64, 7, row_offset);\
341 AVG_64(dst, ref, dst + row_offset);\
343 fill_64(dst, pix64, 8, row_offset)
345 #define RLE_LINES_COPY \
346 copy_block4(dst, ref, row_offset, row_offset, num_lines << v_zoom)
348 #define RLE_LINES_COPY_M10 \
349 pix64 = AV_RN64A(ref);\
350 if (is_top_of_cell) {\
351 pix64 = replicate64(pix64);\
352 fill_64(dst + row_offset, pix64, (num_lines << 1) - 1, row_offset);\
353 AVG_64(dst, ref, dst + row_offset);\
355 fill_64(dst, pix64, num_lines << 1, row_offset)
357 #define APPLY_DELTA_4 \
358 AV_WN16A(dst + line_offset ,\
359 (AV_RN16A(ref ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
360 AV_WN16A(dst + line_offset + 2,\
361 (AV_RN16A(ref + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
363 if (is_top_of_cell && !cell->ypos) {\
364 AV_COPY32(dst, dst + row_offset);\
366 AVG_32(dst, ref, dst + row_offset);\
370 #define APPLY_DELTA_8 \
371 /* apply two 32-bit VQ deltas to next even line */\
372 if (is_top_of_cell) { \
373 AV_WN32A(dst + row_offset , \
374 (replicate32(AV_RN32A(ref )) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
375 AV_WN32A(dst + row_offset + 4, \
376 (replicate32(AV_RN32A(ref + 4)) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
378 AV_WN32A(dst + row_offset , \
379 (AV_RN32A(ref ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
380 AV_WN32A(dst + row_offset + 4, \
381 (AV_RN32A(ref + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
383 /* odd lines are not coded but rather interpolated/replicated */\
384 /* first line of the cell on the top of image? - replicate */\
385 /* otherwise - interpolate */\
386 if (is_top_of_cell && !cell->ypos) {\
387 AV_COPY64(dst, dst + row_offset);\
389 AVG_64(dst, ref, dst + row_offset);
392 #define APPLY_DELTA_1011_INTER \
395 (AV_RN32A(dst ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
397 (AV_RN32A(dst + 4 ) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
398 AV_WN32A(dst + row_offset , \
399 (AV_RN32A(dst + row_offset ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
400 AV_WN32A(dst + row_offset + 4, \
401 (AV_RN32A(dst + row_offset + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
404 (AV_RN16A(dst ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
406 (AV_RN16A(dst + 2 ) + delta_tab->deltas[dyad2]) & 0x7F7F);\
407 AV_WN16A(dst + row_offset , \
408 (AV_RN16A(dst + row_offset ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
409 AV_WN16A(dst + row_offset + 2, \
410 (AV_RN16A(dst + row_offset + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
414 static int decode_cell_data(Indeo3DecodeContext *ctx, Cell *cell,
415 uint8_t *block, uint8_t *ref_block,
416 int pitch, int h_zoom, int v_zoom, int mode,
417 const vqEntry *delta[2], int swap_quads[2],
418 const uint8_t **data_ptr, const uint8_t *last_ptr)
420 int x, y, line, num_lines;
422 uint8_t code, *dst, *ref;
423 const vqEntry *delta_tab;
424 unsigned int dyad1, dyad2;
426 int skip_flag = 0, is_top_of_cell, is_first_row = 1;
427 int row_offset, blk_row_offset, line_offset;
430 blk_row_offset = (row_offset << (2 + v_zoom)) - (cell->width << 2);
431 line_offset = v_zoom ? row_offset : 0;
433 if (cell->height & v_zoom || cell->width & h_zoom)
436 for (y = 0; y < cell->height; is_first_row = 0, y += 1 + v_zoom) {
437 for (x = 0; x < cell->width; x += 1 + h_zoom) {
441 if (rle_blocks > 0) {
444 } else if (mode == 10 && !cell->mv_ptr) {
449 for (line = 0; line < 4;) {
451 is_top_of_cell = is_first_row && !line;
453 /* select primary VQ table for odd, secondary for even lines */
455 delta_tab = delta[line & 1];
457 delta_tab = delta[1];
459 code = bytestream_get_byte(data_ptr);
461 if (code < delta_tab->num_dyads) {
463 dyad1 = bytestream_get_byte(data_ptr);
465 if (dyad1 >= delta_tab->num_dyads || dyad1 >= 248)
469 code -= delta_tab->num_dyads;
470 dyad1 = code / delta_tab->quad_exp;
471 dyad2 = code % delta_tab->quad_exp;
472 if (swap_quads[line & 1])
473 FFSWAP(unsigned int, dyad1, dyad2);
477 } else if (mode == 10 && !cell->mv_ptr) {
480 APPLY_DELTA_1011_INTER;
483 /* process RLE codes */
493 num_lines = 257 - code - line;
498 } else if (mode == 10 && !cell->mv_ptr) {
504 code = bytestream_get_byte(data_ptr);
505 rle_blocks = (code & 0x1F) - 1; /* set block counter */
506 if (code >= 64 || rle_blocks < 0)
507 return IV3_BAD_COUNTER;
508 skip_flag = code & 0x20;
509 num_lines = 4 - line; /* enforce next block processing */
510 if (mode >= 10 || (cell->mv_ptr || !skip_flag)) {
513 } else if (mode == 10 && !cell->mv_ptr) {
525 num_lines = 4; /* enforce next block processing */
529 } else if (mode == 10 && !cell->mv_ptr) {
535 return IV3_UNSUPPORTED;
540 ref += row_offset * (num_lines << v_zoom);
541 dst += row_offset * (num_lines << v_zoom);
545 /* move to next horizontal block */
546 block += 4 << h_zoom;
547 ref_block += 4 << h_zoom;
550 /* move to next line of blocks */
551 ref_block += blk_row_offset;
552 block += blk_row_offset;
559 * Decode a vector-quantized cell.
560 * It consists of several routines, each of which handles one or more "modes"
561 * with which a cell can be encoded.
563 * @param ctx pointer to the decoder context
564 * @param avctx ptr to the AVCodecContext
565 * @param plane pointer to the plane descriptor
566 * @param cell pointer to the cell descriptor
567 * @param data_ptr pointer to the compressed data
568 * @param last_ptr pointer to the last byte to catch reads past end of buffer
569 * @return number of consumed bytes or negative number in case of error
571 static int decode_cell(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
572 Plane *plane, Cell *cell, const uint8_t *data_ptr,
573 const uint8_t *last_ptr)
575 int x, mv_x, mv_y, mode, vq_index, prim_indx, second_indx;
577 int offset, error = 0, swap_quads[2];
578 uint8_t code, *block, *ref_block = 0;
579 const vqEntry *delta[2];
580 const uint8_t *data_start = data_ptr;
582 /* get coding mode and VQ table index from the VQ descriptor byte */
585 vq_index = code & 0xF;
587 /* setup output and reference pointers */
588 offset = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
589 block = plane->pixels[ctx->buf_sel] + offset;
592 mv_y = cell->mv_ptr[0];
593 mv_x = cell->mv_ptr[1];
594 if ( mv_x + 4*cell->xpos < 0
595 || mv_y + 4*cell->ypos < 0
596 || mv_x + 4*cell->xpos + 4*cell->width > plane->width
597 || mv_y + 4*cell->ypos + 4*cell->height > plane->height) {
598 av_log(avctx, AV_LOG_ERROR, "motion vector %d %d outside reference\n", mv_x + 4*cell->xpos, mv_y + 4*cell->ypos);
599 return AVERROR_INVALIDDATA;
604 /* use previous line as reference for INTRA cells */
605 ref_block = block - plane->pitch;
606 } else if (mode >= 10) {
607 /* for mode 10 and 11 INTER first copy the predicted cell into the current one */
608 /* so we don't need to do data copying for each RLE code later */
609 copy_cell(ctx, plane, cell);
611 /* set the pointer to the reference pixels for modes 0-4 INTER */
612 mv_y = cell->mv_ptr[0];
613 mv_x = cell->mv_ptr[1];
614 offset += mv_y * plane->pitch + mv_x;
615 ref_block = plane->pixels[ctx->buf_sel ^ 1] + offset;
618 /* select VQ tables as follows: */
619 /* modes 0 and 3 use only the primary table for all lines in a block */
620 /* while modes 1 and 4 switch between primary and secondary tables on alternate lines */
621 if (mode == 1 || mode == 4) {
622 code = ctx->alt_quant[vq_index];
623 prim_indx = (code >> 4) + ctx->cb_offset;
624 second_indx = (code & 0xF) + ctx->cb_offset;
626 vq_index += ctx->cb_offset;
627 prim_indx = second_indx = vq_index;
630 if (prim_indx >= 24 || second_indx >= 24) {
631 av_log(avctx, AV_LOG_ERROR, "Invalid VQ table indexes! Primary: %d, secondary: %d!\n",
632 prim_indx, second_indx);
633 return AVERROR_INVALIDDATA;
636 delta[0] = &vq_tab[second_indx];
637 delta[1] = &vq_tab[prim_indx];
638 swap_quads[0] = second_indx >= 16;
639 swap_quads[1] = prim_indx >= 16;
641 /* requantize the prediction if VQ index of this cell differs from VQ index */
642 /* of the predicted cell in order to avoid overflows. */
643 if (vq_index >= 8 && ref_block) {
644 for (x = 0; x < cell->width << 2; x++)
645 ref_block[x] = requant_tab[vq_index & 7][ref_block[x] & 127];
651 case 0: /*------------------ MODES 0 & 1 (4x4 block processing) --------------------*/
653 case 3: /*------------------ MODES 3 & 4 (4x8 block processing) --------------------*/
655 if (mode >= 3 && cell->mv_ptr) {
656 av_log(avctx, AV_LOG_ERROR, "Attempt to apply Mode 3/4 to an INTER cell!\n");
657 return AVERROR_INVALIDDATA;
660 zoom_fac = mode >= 3;
661 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
662 0, zoom_fac, mode, delta, swap_quads,
663 &data_ptr, last_ptr);
665 case 10: /*-------------------- MODE 10 (8x8 block processing) ---------------------*/
666 case 11: /*----------------- MODE 11 (4x8 INTER block processing) ------------------*/
667 if (mode == 10 && !cell->mv_ptr) { /* MODE 10 INTRA processing */
668 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
669 1, 1, mode, delta, swap_quads,
670 &data_ptr, last_ptr);
671 } else { /* mode 10 and 11 INTER processing */
672 if (mode == 11 && !cell->mv_ptr) {
673 av_log(avctx, AV_LOG_ERROR, "Attempt to use Mode 11 for an INTRA cell!\n");
674 return AVERROR_INVALIDDATA;
677 zoom_fac = mode == 10;
678 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
679 zoom_fac, 1, mode, delta, swap_quads,
680 &data_ptr, last_ptr);
684 av_log(avctx, AV_LOG_ERROR, "Unsupported coding mode: %d\n", mode);
685 return AVERROR_INVALIDDATA;
690 av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE code %X is not allowed at the current line\n",
692 return AVERROR_INVALIDDATA;
694 av_log(avctx, AV_LOG_ERROR, "Mode %d: invalid VQ data\n", mode);
695 return AVERROR_INVALIDDATA;
696 case IV3_BAD_COUNTER:
697 av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE-FB invalid counter: %d\n", mode, code);
698 return AVERROR_INVALIDDATA;
699 case IV3_UNSUPPORTED:
700 av_log(avctx, AV_LOG_ERROR, "Mode %d: unsupported RLE code: %X\n", mode, data_ptr[-1]);
701 return AVERROR_INVALIDDATA;
702 case IV3_OUT_OF_DATA:
703 av_log(avctx, AV_LOG_ERROR, "Mode %d: attempt to read past end of buffer\n", mode);
704 return AVERROR_INVALIDDATA;
707 return data_ptr - data_start; /* report number of bytes consumed from the input buffer */
711 /* Binary tree codes. */
720 #define SPLIT_CELL(size, new_size) (new_size) = ((size) > 2) ? ((((size) + 2) >> 2) << 1) : 1
722 #define UPDATE_BITPOS(n) \
723 ctx->skip_bits += (n); \
726 #define RESYNC_BITSTREAM \
727 if (ctx->need_resync && !(get_bits_count(&ctx->gb) & 7)) { \
728 skip_bits_long(&ctx->gb, ctx->skip_bits); \
729 ctx->skip_bits = 0; \
730 ctx->need_resync = 0; \
734 if (curr_cell.xpos + curr_cell.width > (plane->width >> 2) || \
735 curr_cell.ypos + curr_cell.height > (plane->height >> 2)) { \
736 av_log(avctx, AV_LOG_ERROR, "Invalid cell: x=%d, y=%d, w=%d, h=%d\n", \
737 curr_cell.xpos, curr_cell.ypos, curr_cell.width, curr_cell.height); \
738 return AVERROR_INVALIDDATA; \
742 static int parse_bintree(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
743 Plane *plane, int code, Cell *ref_cell,
744 const int depth, const int strip_width)
751 av_log(avctx, AV_LOG_ERROR, "Stack overflow (corrupted binary tree)!\n");
752 return AVERROR_INVALIDDATA; // unwind recursion
755 curr_cell = *ref_cell; // clone parent cell
756 if (code == H_SPLIT) {
757 SPLIT_CELL(ref_cell->height, curr_cell.height);
758 ref_cell->ypos += curr_cell.height;
759 ref_cell->height -= curr_cell.height;
760 if (ref_cell->height <= 0 || curr_cell.height <= 0)
761 return AVERROR_INVALIDDATA;
762 } else if (code == V_SPLIT) {
763 if (curr_cell.width > strip_width) {
765 curr_cell.width = (curr_cell.width <= (strip_width << 1) ? 1 : 2) * strip_width;
767 SPLIT_CELL(ref_cell->width, curr_cell.width);
768 ref_cell->xpos += curr_cell.width;
769 ref_cell->width -= curr_cell.width;
770 if (ref_cell->width <= 0 || curr_cell.width <= 0)
771 return AVERROR_INVALIDDATA;
774 while (get_bits_left(&ctx->gb) >= 2) { /* loop until return */
776 switch (code = get_bits(&ctx->gb, 2)) {
779 if (parse_bintree(ctx, avctx, plane, code, &curr_cell, depth - 1, strip_width))
780 return AVERROR_INVALIDDATA;
783 if (!curr_cell.tree) { /* MC tree INTRA code */
784 curr_cell.mv_ptr = 0; /* mark the current strip as INTRA */
785 curr_cell.tree = 1; /* enter the VQ tree */
786 } else { /* VQ tree NULL code */
788 code = get_bits(&ctx->gb, 2);
790 av_log(avctx, AV_LOG_ERROR, "Invalid VQ_NULL code: %d\n", code);
791 return AVERROR_INVALIDDATA;
794 av_log(avctx, AV_LOG_ERROR, "SkipCell procedure not implemented yet!\n");
797 if (!curr_cell.mv_ptr)
798 return AVERROR_INVALIDDATA;
800 mv_y = curr_cell.mv_ptr[0];
801 mv_x = curr_cell.mv_ptr[1];
802 if ( mv_x + 4*curr_cell.xpos < 0
803 || mv_y + 4*curr_cell.ypos < 0
804 || mv_x + 4*curr_cell.xpos + 4*curr_cell.width > plane->width
805 || mv_y + 4*curr_cell.ypos + 4*curr_cell.height > plane->height) {
806 av_log(avctx, AV_LOG_ERROR, "motion vector %d %d outside reference\n", mv_x + 4*curr_cell.xpos, mv_y + 4*curr_cell.ypos);
807 return AVERROR_INVALIDDATA;
810 copy_cell(ctx, plane, &curr_cell);
815 if (!curr_cell.tree) { /* MC tree INTER code */
817 /* get motion vector index and setup the pointer to the mv set */
818 if (!ctx->need_resync)
819 ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
820 if (ctx->next_cell_data >= ctx->last_byte) {
821 av_log(avctx, AV_LOG_ERROR, "motion vector out of array\n");
822 return AVERROR_INVALIDDATA;
824 mv_idx = *(ctx->next_cell_data++);
825 if (mv_idx >= ctx->num_vectors) {
826 av_log(avctx, AV_LOG_ERROR, "motion vector index out of range\n");
827 return AVERROR_INVALIDDATA;
829 curr_cell.mv_ptr = &ctx->mc_vectors[mv_idx << 1];
830 curr_cell.tree = 1; /* enter the VQ tree */
832 } else { /* VQ tree DATA code */
833 if (!ctx->need_resync)
834 ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
837 bytes_used = decode_cell(ctx, avctx, plane, &curr_cell,
838 ctx->next_cell_data, ctx->last_byte);
840 return AVERROR_INVALIDDATA;
842 UPDATE_BITPOS(bytes_used << 3);
843 ctx->next_cell_data += bytes_used;
850 return AVERROR_INVALIDDATA;
854 static int decode_plane(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
855 Plane *plane, const uint8_t *data, int32_t data_size,
859 unsigned num_vectors;
861 /* each plane data starts with mc_vector_count field, */
862 /* an optional array of motion vectors followed by the vq data */
863 num_vectors = bytestream_get_le32(&data); data_size -= 4;
864 if (num_vectors > 256) {
865 av_log(ctx->avctx, AV_LOG_ERROR,
866 "Read invalid number of motion vectors %d\n", num_vectors);
867 return AVERROR_INVALIDDATA;
869 if (num_vectors * 2 > data_size)
870 return AVERROR_INVALIDDATA;
872 ctx->num_vectors = num_vectors;
873 ctx->mc_vectors = num_vectors ? data : 0;
875 /* init the bitreader */
876 init_get_bits(&ctx->gb, &data[num_vectors * 2], (data_size - num_vectors * 2) << 3);
878 ctx->need_resync = 0;
880 ctx->last_byte = data + data_size;
882 /* initialize the 1st cell and set its dimensions to whole plane */
883 curr_cell.xpos = curr_cell.ypos = 0;
884 curr_cell.width = plane->width >> 2;
885 curr_cell.height = plane->height >> 2;
886 curr_cell.tree = 0; // we are in the MC tree now
887 curr_cell.mv_ptr = 0; // no motion vector = INTRA cell
889 return parse_bintree(ctx, avctx, plane, INTRA_NULL, &curr_cell, CELL_STACK_MAX, strip_width);
893 #define OS_HDR_ID MKBETAG('F', 'R', 'M', 'H')
895 static int decode_frame_headers(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
896 const uint8_t *buf, int buf_size)
898 const uint8_t *buf_ptr = buf, *bs_hdr;
899 uint32_t frame_num, word2, check_sum, data_size;
900 uint32_t y_offset, u_offset, v_offset, starts[3], ends[3];
901 uint16_t height, width;
904 /* parse and check the OS header */
905 frame_num = bytestream_get_le32(&buf_ptr);
906 word2 = bytestream_get_le32(&buf_ptr);
907 check_sum = bytestream_get_le32(&buf_ptr);
908 data_size = bytestream_get_le32(&buf_ptr);
910 if ((frame_num ^ word2 ^ data_size ^ OS_HDR_ID) != check_sum) {
911 av_log(avctx, AV_LOG_ERROR, "OS header checksum mismatch!\n");
912 return AVERROR_INVALIDDATA;
915 /* parse the bitstream header */
919 if (bytestream_get_le16(&buf_ptr) != 32) {
920 av_log(avctx, AV_LOG_ERROR, "Unsupported codec version!\n");
921 return AVERROR_INVALIDDATA;
924 ctx->frame_num = frame_num;
925 ctx->frame_flags = bytestream_get_le16(&buf_ptr);
926 ctx->data_size = (bytestream_get_le32(&buf_ptr) + 7) >> 3;
927 ctx->cb_offset = *buf_ptr++;
929 if (ctx->data_size == 16)
931 if (ctx->data_size > buf_size)
932 ctx->data_size = buf_size;
934 buf_ptr += 3; // skip reserved byte and checksum
936 /* check frame dimensions */
937 height = bytestream_get_le16(&buf_ptr);
938 width = bytestream_get_le16(&buf_ptr);
939 if (av_image_check_size(width, height, 0, avctx))
940 return AVERROR_INVALIDDATA;
942 if (width != ctx->width || height != ctx->height) {
945 av_dlog(avctx, "Frame dimensions changed!\n");
947 if (width < 16 || width > 640 ||
948 height < 16 || height > 480 ||
949 width & 3 || height & 3) {
950 av_log(avctx, AV_LOG_ERROR,
951 "Invalid picture dimensions: %d x %d!\n", width, height);
952 return AVERROR_INVALIDDATA;
954 free_frame_buffers(ctx);
955 if ((res = allocate_frame_buffers(ctx, avctx, width, height)) < 0)
957 avcodec_set_dimensions(avctx, width, height);
960 y_offset = bytestream_get_le32(&buf_ptr);
961 v_offset = bytestream_get_le32(&buf_ptr);
962 u_offset = bytestream_get_le32(&buf_ptr);
964 /* unfortunately there is no common order of planes in the buffer */
965 /* so we use that sorting algo for determining planes data sizes */
966 starts[0] = y_offset;
967 starts[1] = v_offset;
968 starts[2] = u_offset;
970 for (j = 0; j < 3; j++) {
971 ends[j] = ctx->data_size;
972 for (i = 2; i >= 0; i--)
973 if (starts[i] < ends[j] && starts[i] > starts[j])
977 ctx->y_data_size = ends[0] - starts[0];
978 ctx->v_data_size = ends[1] - starts[1];
979 ctx->u_data_size = ends[2] - starts[2];
980 if (FFMAX3(y_offset, v_offset, u_offset) >= ctx->data_size - 16 ||
981 FFMIN3(ctx->y_data_size, ctx->v_data_size, ctx->u_data_size) <= 0) {
982 av_log(avctx, AV_LOG_ERROR, "One of the y/u/v offsets is invalid\n");
983 return AVERROR_INVALIDDATA;
986 ctx->y_data_ptr = bs_hdr + y_offset;
987 ctx->v_data_ptr = bs_hdr + v_offset;
988 ctx->u_data_ptr = bs_hdr + u_offset;
989 ctx->alt_quant = buf_ptr + sizeof(uint32_t);
991 if (ctx->data_size == 16) {
992 av_log(avctx, AV_LOG_DEBUG, "Sync frame encountered!\n");
996 if (ctx->frame_flags & BS_8BIT_PEL) {
997 av_log_ask_for_sample(avctx, "8-bit pixel format\n");
998 return AVERROR_PATCHWELCOME;
1001 if (ctx->frame_flags & BS_MV_X_HALF || ctx->frame_flags & BS_MV_Y_HALF) {
1002 av_log_ask_for_sample(avctx, "halfpel motion vectors\n");
1003 return AVERROR_PATCHWELCOME;
1011 * Convert and output the current plane.
1012 * All pixel values will be upsampled by shifting right by one bit.
1014 * @param[in] plane pointer to the descriptor of the plane being processed
1015 * @param[in] buf_sel indicates which frame buffer the input data stored in
1016 * @param[out] dst pointer to the buffer receiving converted pixels
1017 * @param[in] dst_pitch pitch for moving to the next y line
1018 * @param[in] dst_height output plane height
1020 static void output_plane(const Plane *plane, int buf_sel, uint8_t *dst,
1021 int dst_pitch, int dst_height)
1024 const uint8_t *src = plane->pixels[buf_sel];
1025 uint32_t pitch = plane->pitch;
1027 dst_height = FFMIN(dst_height, plane->height);
1028 for (y = 0; y < dst_height; y++) {
1029 /* convert four pixels at once using SWAR */
1030 for (x = 0; x < plane->width >> 2; x++) {
1031 AV_WN32A(dst, (AV_RN32A(src) & 0x7F7F7F7F) << 1);
1036 for (x <<= 2; x < plane->width; x++)
1037 *dst++ = *src++ << 1;
1039 src += pitch - plane->width;
1040 dst += dst_pitch - plane->width;
1045 static av_cold int decode_init(AVCodecContext *avctx)
1047 Indeo3DecodeContext *ctx = avctx->priv_data;
1050 avctx->pix_fmt = AV_PIX_FMT_YUV410P;
1051 avcodec_get_frame_defaults(&ctx->frame);
1053 build_requant_tab();
1055 ff_dsputil_init(&ctx->dsp, avctx);
1057 return allocate_frame_buffers(ctx, avctx, avctx->width, avctx->height);
1061 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
1064 Indeo3DecodeContext *ctx = avctx->priv_data;
1065 const uint8_t *buf = avpkt->data;
1066 int buf_size = avpkt->size;
1069 res = decode_frame_headers(ctx, avctx, buf, buf_size);
1073 /* skip sync(null) frames */
1075 // we have processed 16 bytes but no data was decoded
1080 /* skip droppable INTER frames if requested */
1081 if (ctx->frame_flags & BS_NONREF &&
1082 (avctx->skip_frame >= AVDISCARD_NONREF))
1085 /* skip INTER frames if requested */
1086 if (!(ctx->frame_flags & BS_KEYFRAME) && avctx->skip_frame >= AVDISCARD_NONKEY)
1089 /* use BS_BUFFER flag for buffer switching */
1090 ctx->buf_sel = (ctx->frame_flags >> BS_BUFFER) & 1;
1092 if (ctx->frame.data[0])
1093 avctx->release_buffer(avctx, &ctx->frame);
1095 ctx->frame.reference = 0;
1096 if ((res = ff_get_buffer(avctx, &ctx->frame)) < 0) {
1097 av_log(ctx->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1101 /* decode luma plane */
1102 if ((res = decode_plane(ctx, avctx, ctx->planes, ctx->y_data_ptr, ctx->y_data_size, 40)))
1105 /* decode chroma planes */
1106 if ((res = decode_plane(ctx, avctx, &ctx->planes[1], ctx->u_data_ptr, ctx->u_data_size, 10)))
1109 if ((res = decode_plane(ctx, avctx, &ctx->planes[2], ctx->v_data_ptr, ctx->v_data_size, 10)))
1112 output_plane(&ctx->planes[0], ctx->buf_sel,
1113 ctx->frame.data[0], ctx->frame.linesize[0],
1115 output_plane(&ctx->planes[1], ctx->buf_sel,
1116 ctx->frame.data[1], ctx->frame.linesize[1],
1117 (avctx->height + 3) >> 2);
1118 output_plane(&ctx->planes[2], ctx->buf_sel,
1119 ctx->frame.data[2], ctx->frame.linesize[2],
1120 (avctx->height + 3) >> 2);
1123 *(AVFrame*)data = ctx->frame;
1129 static av_cold int decode_close(AVCodecContext *avctx)
1131 Indeo3DecodeContext *ctx = avctx->priv_data;
1133 free_frame_buffers(avctx->priv_data);
1135 if (ctx->frame.data[0])
1136 avctx->release_buffer(avctx, &ctx->frame);
1141 AVCodec ff_indeo3_decoder = {
1143 .type = AVMEDIA_TYPE_VIDEO,
1144 .id = AV_CODEC_ID_INDEO3,
1145 .priv_data_size = sizeof(Indeo3DecodeContext),
1146 .init = decode_init,
1147 .close = decode_close,
1148 .decode = decode_frame,
1149 .long_name = NULL_IF_CONFIG_SMALL("Intel Indeo 3"),
1150 .capabilities = CODEC_CAP_DR1,