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"
36 #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;
90 const uint8_t *next_cell_data;
91 const uint8_t *last_byte;
92 const int8_t *mc_vectors;
93 unsigned num_vectors; ///< number of motion vectors in mc_vectors
95 int16_t width, height;
96 uint32_t frame_num; ///< current frame number (zero-based)
97 int data_size; ///< size of the frame data in bytes
98 uint16_t frame_flags; ///< frame properties
99 uint8_t cb_offset; ///< needed for selecting VQ tables
100 uint8_t buf_sel; ///< active frame buffer: 0 - primary, 1 -secondary
101 const uint8_t *y_data_ptr;
102 const uint8_t *v_data_ptr;
103 const uint8_t *u_data_ptr;
107 const uint8_t *alt_quant; ///< secondary VQ table set for the modes 1 and 4
109 } Indeo3DecodeContext;
112 static uint8_t requant_tab[8][128];
115 * Build the static requantization table.
116 * This table is used to remap pixel values according to a specific
117 * quant index and thus avoid overflows while adding deltas.
119 static av_cold void build_requant_tab(void)
121 static const int8_t offsets[8] = { 1, 1, 2, -3, -3, 3, 4, 4 };
122 static const int8_t deltas [8] = { 0, 1, 0, 4, 4, 1, 0, 1 };
126 for (i = 0; i < 8; i++) {
128 for (j = 0; j < 128; j++)
129 requant_tab[i][j] = (j + offsets[i]) / step * step + deltas[i];
132 /* some last elements calculated above will have values >= 128 */
133 /* pixel values shall never exceed 127 so set them to non-overflowing values */
134 /* according with the quantization step of the respective section */
135 requant_tab[0][127] = 126;
136 requant_tab[1][119] = 118;
137 requant_tab[1][120] = 118;
138 requant_tab[2][126] = 124;
139 requant_tab[2][127] = 124;
140 requant_tab[6][124] = 120;
141 requant_tab[6][125] = 120;
142 requant_tab[6][126] = 120;
143 requant_tab[6][127] = 120;
145 /* Patch for compatibility with the Intel's binary decoders */
146 requant_tab[1][7] = 10;
147 requant_tab[4][8] = 10;
151 static av_cold void free_frame_buffers(Indeo3DecodeContext *ctx)
155 ctx->width = ctx->height = 0;
157 for (p = 0; p < 3; p++) {
158 av_freep(&ctx->planes[p].buffers[0]);
159 av_freep(&ctx->planes[p].buffers[1]);
160 ctx->planes[p].pixels[0] = ctx->planes[p].pixels[1] = 0;
165 static av_cold int allocate_frame_buffers(Indeo3DecodeContext *ctx,
166 AVCodecContext *avctx, int luma_width, int luma_height)
168 int p, chroma_width, chroma_height;
169 int luma_pitch, chroma_pitch, luma_size, chroma_size;
171 if (luma_width < 16 || luma_width > 640 ||
172 luma_height < 16 || luma_height > 480 ||
173 luma_width & 3 || luma_height & 3) {
174 av_log(avctx, AV_LOG_ERROR, "Invalid picture dimensions: %d x %d!\n",
175 luma_width, luma_height);
176 return AVERROR_INVALIDDATA;
179 ctx->width = luma_width ;
180 ctx->height = luma_height;
182 chroma_width = FFALIGN(luma_width >> 2, 4);
183 chroma_height = FFALIGN(luma_height >> 2, 4);
185 luma_pitch = FFALIGN(luma_width, 16);
186 chroma_pitch = FFALIGN(chroma_width, 16);
188 /* Calculate size of the luminance plane. */
189 /* Add one line more for INTRA prediction. */
190 luma_size = luma_pitch * (luma_height + 1);
192 /* Calculate size of a chrominance planes. */
193 /* Add one line more for INTRA prediction. */
194 chroma_size = chroma_pitch * (chroma_height + 1);
196 /* allocate frame buffers */
197 for (p = 0; p < 3; p++) {
198 ctx->planes[p].pitch = !p ? luma_pitch : chroma_pitch;
199 ctx->planes[p].width = !p ? luma_width : chroma_width;
200 ctx->planes[p].height = !p ? luma_height : chroma_height;
202 ctx->planes[p].buffers[0] = av_malloc(!p ? luma_size : chroma_size);
203 ctx->planes[p].buffers[1] = av_malloc(!p ? luma_size : chroma_size);
205 if (!ctx->planes[p].buffers[0] || !ctx->planes[p].buffers[1]) {
206 free_frame_buffers(ctx);
207 return AVERROR(ENOMEM);
210 /* fill the INTRA prediction lines with the middle pixel value = 64 */
211 memset(ctx->planes[p].buffers[0], 0x40, ctx->planes[p].pitch);
212 memset(ctx->planes[p].buffers[1], 0x40, ctx->planes[p].pitch);
214 /* set buffer pointers = buf_ptr + pitch and thus skip the INTRA prediction line */
215 ctx->planes[p].pixels[0] = ctx->planes[p].buffers[0] + ctx->planes[p].pitch;
216 ctx->planes[p].pixels[1] = ctx->planes[p].buffers[1] + ctx->planes[p].pitch;
217 memset(ctx->planes[p].pixels[0], 0, ctx->planes[p].pitch * ctx->planes[p].height);
218 memset(ctx->planes[p].pixels[1], 0, ctx->planes[p].pitch * ctx->planes[p].height);
225 * Copy pixels of the cell(x + mv_x, y + mv_y) from the previous frame into
226 * the cell(x, y) in the current frame.
228 * @param ctx pointer to the decoder context
229 * @param plane pointer to the plane descriptor
230 * @param cell pointer to the cell descriptor
232 static int copy_cell(Indeo3DecodeContext *ctx, Plane *plane, Cell *cell)
234 int h, w, mv_x, mv_y, offset, offset_dst;
237 /* setup output and reference pointers */
238 offset_dst = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
239 dst = plane->pixels[ctx->buf_sel] + offset_dst;
241 mv_y = cell->mv_ptr[0];
242 mv_x = cell->mv_ptr[1];
246 /* -1 because there is an extra line on top for prediction */
247 if ((cell->ypos << 2) + mv_y < -1 || (cell->xpos << 2) + mv_x < 0 ||
248 ((cell->ypos + cell->height) << 2) + mv_y > plane->height ||
249 ((cell->xpos + cell->width) << 2) + mv_x > plane->width) {
250 av_log(ctx->avctx, AV_LOG_ERROR,
251 "Motion vectors point out of the frame.\n");
252 return AVERROR_INVALIDDATA;
255 offset = offset_dst + mv_y * plane->pitch + mv_x;
256 src = plane->pixels[ctx->buf_sel ^ 1] + offset;
258 h = cell->height << 2;
260 for (w = cell->width; w > 0;) {
261 /* copy using 16xH blocks */
262 if (!((cell->xpos << 2) & 15) && w >= 4) {
263 for (; w >= 4; src += 16, dst += 16, w -= 4)
264 ctx->hdsp.put_pixels_tab[0][0](dst, src, plane->pitch, h);
267 /* copy using 8xH blocks */
268 if (!((cell->xpos << 2) & 7) && w >= 2) {
269 ctx->hdsp.put_pixels_tab[1][0](dst, src, plane->pitch, h);
274 ctx->hdsp.put_pixels_tab[2][0](dst, src, plane->pitch, h);
285 /* Average 4/8 pixels at once without rounding using SWAR */
286 #define AVG_32(dst, src, ref) \
287 AV_WN32A(dst, ((AV_RN32(src) + AV_RN32(ref)) >> 1) & 0x7F7F7F7FUL)
289 #define AVG_64(dst, src, ref) \
290 AV_WN64A(dst, ((AV_RN64(src) + AV_RN64(ref)) >> 1) & 0x7F7F7F7F7F7F7F7FULL)
294 * Replicate each even pixel as follows:
295 * ABCDEFGH -> AACCEEGG
297 static inline uint64_t replicate64(uint64_t a) {
299 a &= 0xFF00FF00FF00FF00ULL;
302 a &= 0x00FF00FF00FF00FFULL;
308 static inline uint32_t replicate32(uint32_t a) {
320 /* Fill n lines with 64bit pixel value pix */
321 static inline void fill_64(uint8_t *dst, const uint64_t pix, int32_t n,
324 for (; n > 0; dst += row_offset, n--)
329 /* Error codes for cell decoding. */
340 #define BUFFER_PRECHECK \
341 if (*data_ptr >= last_ptr) \
342 return IV3_OUT_OF_DATA; \
344 #define RLE_BLOCK_COPY \
345 if (cell->mv_ptr || !skip_flag) \
346 copy_block4(dst, ref, row_offset, row_offset, 4 << v_zoom)
348 #define RLE_BLOCK_COPY_8 \
349 pix64 = AV_RN64(ref);\
350 if (is_first_row) {/* special prediction case: top line of a cell */\
351 pix64 = replicate64(pix64);\
352 fill_64(dst + row_offset, pix64, 7, row_offset);\
353 AVG_64(dst, ref, dst + row_offset);\
355 fill_64(dst, pix64, 8, row_offset)
357 #define RLE_LINES_COPY \
358 copy_block4(dst, ref, row_offset, row_offset, num_lines << v_zoom)
360 #define RLE_LINES_COPY_M10 \
361 pix64 = AV_RN64(ref);\
362 if (is_top_of_cell) {\
363 pix64 = replicate64(pix64);\
364 fill_64(dst + row_offset, pix64, (num_lines << 1) - 1, row_offset);\
365 AVG_64(dst, ref, dst + row_offset);\
367 fill_64(dst, pix64, num_lines << 1, row_offset)
369 #define APPLY_DELTA_4 \
370 AV_WN16A(dst + line_offset ,\
371 (AV_RN16(ref ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
372 AV_WN16A(dst + line_offset + 2,\
373 (AV_RN16(ref + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
375 if (is_top_of_cell && !cell->ypos) {\
376 AV_COPY32U(dst, dst + row_offset);\
378 AVG_32(dst, ref, dst + row_offset);\
382 #define APPLY_DELTA_8 \
383 /* apply two 32-bit VQ deltas to next even line */\
384 if (is_top_of_cell) { \
385 AV_WN32A(dst + row_offset , \
386 (replicate32(AV_RN32(ref )) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
387 AV_WN32A(dst + row_offset + 4, \
388 (replicate32(AV_RN32(ref + 4)) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
390 AV_WN32A(dst + row_offset , \
391 (AV_RN32(ref ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
392 AV_WN32A(dst + row_offset + 4, \
393 (AV_RN32(ref + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
395 /* odd lines are not coded but rather interpolated/replicated */\
396 /* first line of the cell on the top of image? - replicate */\
397 /* otherwise - interpolate */\
398 if (is_top_of_cell && !cell->ypos) {\
399 AV_COPY64U(dst, dst + row_offset);\
401 AVG_64(dst, ref, dst + row_offset);
404 #define APPLY_DELTA_1011_INTER \
407 (AV_RN32(dst ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
409 (AV_RN32(dst + 4 ) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
410 AV_WN32A(dst + row_offset , \
411 (AV_RN32(dst + row_offset ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
412 AV_WN32A(dst + row_offset + 4, \
413 (AV_RN32(dst + row_offset + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
416 (AV_RN16(dst ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
418 (AV_RN16(dst + 2 ) + delta_tab->deltas[dyad2]) & 0x7F7F);\
419 AV_WN16A(dst + row_offset , \
420 (AV_RN16(dst + row_offset ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
421 AV_WN16A(dst + row_offset + 2, \
422 (AV_RN16(dst + row_offset + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
426 static int decode_cell_data(Indeo3DecodeContext *ctx, Cell *cell,
427 uint8_t *block, uint8_t *ref_block,
428 int pitch, int h_zoom, int v_zoom, int mode,
429 const vqEntry *delta[2], int swap_quads[2],
430 const uint8_t **data_ptr, const uint8_t *last_ptr)
432 int x, y, line, num_lines;
434 uint8_t code, *dst, *ref;
435 const vqEntry *delta_tab;
436 unsigned int dyad1, dyad2;
438 int skip_flag = 0, is_top_of_cell, is_first_row = 1;
439 int row_offset, blk_row_offset, line_offset;
442 blk_row_offset = (row_offset << (2 + v_zoom)) - (cell->width << 2);
443 line_offset = v_zoom ? row_offset : 0;
445 if (cell->height & v_zoom || cell->width & h_zoom)
448 for (y = 0; y < cell->height; is_first_row = 0, y += 1 + v_zoom) {
449 for (x = 0; x < cell->width; x += 1 + h_zoom) {
453 if (rle_blocks > 0) {
456 } else if (mode == 10 && !cell->mv_ptr) {
461 for (line = 0; line < 4;) {
463 is_top_of_cell = is_first_row && !line;
465 /* select primary VQ table for odd, secondary for even lines */
467 delta_tab = delta[line & 1];
469 delta_tab = delta[1];
471 code = bytestream_get_byte(data_ptr);
473 if (code < delta_tab->num_dyads) {
475 dyad1 = bytestream_get_byte(data_ptr);
477 if (dyad1 >= delta_tab->num_dyads || dyad1 >= 248)
481 code -= delta_tab->num_dyads;
482 dyad1 = code / delta_tab->quad_exp;
483 dyad2 = code % delta_tab->quad_exp;
484 if (swap_quads[line & 1])
485 FFSWAP(unsigned int, dyad1, dyad2);
489 } else if (mode == 10 && !cell->mv_ptr) {
492 APPLY_DELTA_1011_INTER;
495 /* process RLE codes */
505 num_lines = 257 - code - line;
510 } else if (mode == 10 && !cell->mv_ptr) {
516 code = bytestream_get_byte(data_ptr);
517 rle_blocks = (code & 0x1F) - 1; /* set block counter */
518 if (code >= 64 || rle_blocks < 0)
519 return IV3_BAD_COUNTER;
520 skip_flag = code & 0x20;
521 num_lines = 4 - line; /* enforce next block processing */
522 if (mode >= 10 || (cell->mv_ptr || !skip_flag)) {
525 } else if (mode == 10 && !cell->mv_ptr) {
537 num_lines = 4; /* enforce next block processing */
541 } else if (mode == 10 && !cell->mv_ptr) {
547 return IV3_UNSUPPORTED;
552 ref += row_offset * (num_lines << v_zoom);
553 dst += row_offset * (num_lines << v_zoom);
557 /* move to next horizontal block */
558 block += 4 << h_zoom;
559 ref_block += 4 << h_zoom;
562 /* move to next line of blocks */
563 ref_block += blk_row_offset;
564 block += blk_row_offset;
571 * Decode a vector-quantized cell.
572 * It consists of several routines, each of which handles one or more "modes"
573 * with which a cell can be encoded.
575 * @param ctx pointer to the decoder context
576 * @param avctx ptr to the AVCodecContext
577 * @param plane pointer to the plane descriptor
578 * @param cell pointer to the cell descriptor
579 * @param data_ptr pointer to the compressed data
580 * @param last_ptr pointer to the last byte to catch reads past end of buffer
581 * @return number of consumed bytes or negative number in case of error
583 static int decode_cell(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
584 Plane *plane, Cell *cell, const uint8_t *data_ptr,
585 const uint8_t *last_ptr)
587 int x, mv_x, mv_y, mode, vq_index, prim_indx, second_indx;
589 int offset, error = 0, swap_quads[2];
590 uint8_t code, *block, *ref_block = 0;
591 const vqEntry *delta[2];
592 const uint8_t *data_start = data_ptr;
594 /* get coding mode and VQ table index from the VQ descriptor byte */
597 vq_index = code & 0xF;
599 /* setup output and reference pointers */
600 offset = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
601 block = plane->pixels[ctx->buf_sel] + offset;
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 int ret = copy_cell(ctx, plane, cell);
613 /* set the pointer to the reference pixels for modes 0-4 INTER */
614 mv_y = cell->mv_ptr[0];
615 mv_x = cell->mv_ptr[1];
617 /* -1 because there is an extra line on top for prediction */
618 if ((cell->ypos << 2) + mv_y < -1 || (cell->xpos << 2) + mv_x < 0 ||
619 ((cell->ypos + cell->height) << 2) + mv_y > plane->height ||
620 ((cell->xpos + cell->width) << 2) + mv_x > plane->width) {
621 av_log(ctx->avctx, AV_LOG_ERROR,
622 "Motion vectors point out of the frame.\n");
623 return AVERROR_INVALIDDATA;
626 offset += mv_y * plane->pitch + mv_x;
627 ref_block = plane->pixels[ctx->buf_sel ^ 1] + offset;
630 /* select VQ tables as follows: */
631 /* modes 0 and 3 use only the primary table for all lines in a block */
632 /* while modes 1 and 4 switch between primary and secondary tables on alternate lines */
633 if (mode == 1 || mode == 4) {
634 code = ctx->alt_quant[vq_index];
635 prim_indx = (code >> 4) + ctx->cb_offset;
636 second_indx = (code & 0xF) + ctx->cb_offset;
638 vq_index += ctx->cb_offset;
639 prim_indx = second_indx = vq_index;
642 if (prim_indx >= 24 || second_indx >= 24) {
643 av_log(avctx, AV_LOG_ERROR, "Invalid VQ table indexes! Primary: %d, secondary: %d!\n",
644 prim_indx, second_indx);
645 return AVERROR_INVALIDDATA;
648 delta[0] = &vq_tab[second_indx];
649 delta[1] = &vq_tab[prim_indx];
650 swap_quads[0] = second_indx >= 16;
651 swap_quads[1] = prim_indx >= 16;
653 /* requantize the prediction if VQ index of this cell differs from VQ index */
654 /* of the predicted cell in order to avoid overflows. */
655 if (vq_index >= 8 && ref_block) {
656 for (x = 0; x < cell->width << 2; x++)
657 ref_block[x] = requant_tab[vq_index & 7][ref_block[x] & 127];
663 case 0: /*------------------ MODES 0 & 1 (4x4 block processing) --------------------*/
665 case 3: /*------------------ MODES 3 & 4 (4x8 block processing) --------------------*/
667 if (mode >= 3 && cell->mv_ptr) {
668 av_log(avctx, AV_LOG_ERROR, "Attempt to apply Mode 3/4 to an INTER cell!\n");
669 return AVERROR_INVALIDDATA;
672 zoom_fac = mode >= 3;
673 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
674 0, zoom_fac, mode, delta, swap_quads,
675 &data_ptr, last_ptr);
677 case 10: /*-------------------- MODE 10 (8x8 block processing) ---------------------*/
678 case 11: /*----------------- MODE 11 (4x8 INTER block processing) ------------------*/
679 if (mode == 10 && !cell->mv_ptr) { /* MODE 10 INTRA processing */
680 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
681 1, 1, mode, delta, swap_quads,
682 &data_ptr, last_ptr);
683 } else { /* mode 10 and 11 INTER processing */
684 if (mode == 11 && !cell->mv_ptr) {
685 av_log(avctx, AV_LOG_ERROR, "Attempt to use Mode 11 for an INTRA cell!\n");
686 return AVERROR_INVALIDDATA;
689 zoom_fac = mode == 10;
690 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
691 zoom_fac, 1, mode, delta, swap_quads,
692 &data_ptr, last_ptr);
696 av_log(avctx, AV_LOG_ERROR, "Unsupported coding mode: %d\n", mode);
697 return AVERROR_INVALIDDATA;
702 av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE code %X is not allowed at the current line\n",
704 return AVERROR_INVALIDDATA;
706 av_log(avctx, AV_LOG_ERROR, "Mode %d: invalid VQ data\n", mode);
707 return AVERROR_INVALIDDATA;
708 case IV3_BAD_COUNTER:
709 av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE-FB invalid counter: %d\n", mode, code);
710 return AVERROR_INVALIDDATA;
711 case IV3_UNSUPPORTED:
712 av_log(avctx, AV_LOG_ERROR, "Mode %d: unsupported RLE code: %X\n", mode, data_ptr[-1]);
713 return AVERROR_INVALIDDATA;
714 case IV3_OUT_OF_DATA:
715 av_log(avctx, AV_LOG_ERROR, "Mode %d: attempt to read past end of buffer\n", mode);
716 return AVERROR_INVALIDDATA;
719 return data_ptr - data_start; /* report number of bytes consumed from the input buffer */
723 /* Binary tree codes. */
732 #define SPLIT_CELL(size, new_size) (new_size) = ((size) > 2) ? ((((size) + 2) >> 2) << 1) : 1
734 #define UPDATE_BITPOS(n) \
735 ctx->skip_bits += (n); \
738 #define RESYNC_BITSTREAM \
739 if (ctx->need_resync && !(get_bits_count(&ctx->gb) & 7)) { \
740 skip_bits_long(&ctx->gb, ctx->skip_bits); \
741 ctx->skip_bits = 0; \
742 ctx->need_resync = 0; \
746 if (curr_cell.xpos + curr_cell.width > (plane->width >> 2) || \
747 curr_cell.ypos + curr_cell.height > (plane->height >> 2)) { \
748 av_log(avctx, AV_LOG_ERROR, "Invalid cell: x=%d, y=%d, w=%d, h=%d\n", \
749 curr_cell.xpos, curr_cell.ypos, curr_cell.width, curr_cell.height); \
750 return AVERROR_INVALIDDATA; \
754 static int parse_bintree(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
755 Plane *plane, int code, Cell *ref_cell,
756 const int depth, const int strip_width)
762 av_log(avctx, AV_LOG_ERROR, "Stack overflow (corrupted binary tree)!\n");
763 return AVERROR_INVALIDDATA; // unwind recursion
766 curr_cell = *ref_cell; // clone parent cell
767 if (code == H_SPLIT) {
768 SPLIT_CELL(ref_cell->height, curr_cell.height);
769 ref_cell->ypos += curr_cell.height;
770 ref_cell->height -= curr_cell.height;
771 if (ref_cell->height <= 0 || curr_cell.height <= 0)
772 return AVERROR_INVALIDDATA;
773 } else if (code == V_SPLIT) {
774 if (curr_cell.width > strip_width) {
776 curr_cell.width = (curr_cell.width <= (strip_width << 1) ? 1 : 2) * strip_width;
778 SPLIT_CELL(ref_cell->width, curr_cell.width);
779 ref_cell->xpos += curr_cell.width;
780 ref_cell->width -= curr_cell.width;
781 if (ref_cell->width <= 0 || curr_cell.width <= 0)
782 return AVERROR_INVALIDDATA;
785 while (get_bits_left(&ctx->gb) >= 2) { /* loop until return */
787 switch (code = get_bits(&ctx->gb, 2)) {
790 if (parse_bintree(ctx, avctx, plane, code, &curr_cell, depth - 1, strip_width))
791 return AVERROR_INVALIDDATA;
794 if (!curr_cell.tree) { /* MC tree INTRA code */
795 curr_cell.mv_ptr = 0; /* mark the current strip as INTRA */
796 curr_cell.tree = 1; /* enter the VQ tree */
797 } else { /* VQ tree NULL code */
799 code = get_bits(&ctx->gb, 2);
801 av_log(avctx, AV_LOG_ERROR, "Invalid VQ_NULL code: %d\n", code);
802 return AVERROR_INVALIDDATA;
805 av_log(avctx, AV_LOG_ERROR, "SkipCell procedure not implemented yet!\n");
808 if (!curr_cell.mv_ptr)
809 return AVERROR_INVALIDDATA;
811 ret = copy_cell(ctx, plane, &curr_cell);
816 if (!curr_cell.tree) { /* MC tree INTER code */
818 /* get motion vector index and setup the pointer to the mv set */
819 if (!ctx->need_resync)
820 ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
821 if (ctx->next_cell_data >= ctx->last_byte) {
822 av_log(avctx, AV_LOG_ERROR, "motion vector out of array\n");
823 return AVERROR_INVALIDDATA;
825 mv_idx = *(ctx->next_cell_data++);
826 if (mv_idx >= ctx->num_vectors) {
827 av_log(avctx, AV_LOG_ERROR, "motion vector index out of range\n");
828 return AVERROR_INVALIDDATA;
830 curr_cell.mv_ptr = &ctx->mc_vectors[mv_idx << 1];
831 curr_cell.tree = 1; /* enter the VQ tree */
833 } else { /* VQ tree DATA code */
834 if (!ctx->need_resync)
835 ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
838 bytes_used = decode_cell(ctx, avctx, plane, &curr_cell,
839 ctx->next_cell_data, ctx->last_byte);
841 return AVERROR_INVALIDDATA;
843 UPDATE_BITPOS(bytes_used << 3);
844 ctx->next_cell_data += bytes_used;
851 return AVERROR_INVALIDDATA;
855 static int decode_plane(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
856 Plane *plane, const uint8_t *data, int32_t data_size,
860 unsigned num_vectors;
862 /* each plane data starts with mc_vector_count field, */
863 /* an optional array of motion vectors followed by the vq data */
864 num_vectors = bytestream_get_le32(&data); data_size -= 4;
865 if (num_vectors > 256) {
866 av_log(ctx->avctx, AV_LOG_ERROR,
867 "Read invalid number of motion vectors %d\n", num_vectors);
868 return AVERROR_INVALIDDATA;
870 if (num_vectors * 2 > data_size)
871 return AVERROR_INVALIDDATA;
873 ctx->num_vectors = num_vectors;
874 ctx->mc_vectors = num_vectors ? data : 0;
876 /* init the bitreader */
877 init_get_bits(&ctx->gb, &data[num_vectors * 2], (data_size - num_vectors * 2) << 3);
879 ctx->need_resync = 0;
881 ctx->last_byte = data + data_size;
883 /* initialize the 1st cell and set its dimensions to whole plane */
884 curr_cell.xpos = curr_cell.ypos = 0;
885 curr_cell.width = plane->width >> 2;
886 curr_cell.height = plane->height >> 2;
887 curr_cell.tree = 0; // we are in the MC tree now
888 curr_cell.mv_ptr = 0; // no motion vector = INTRA cell
890 return parse_bintree(ctx, avctx, plane, INTRA_NULL, &curr_cell, CELL_STACK_MAX, strip_width);
894 #define OS_HDR_ID MKBETAG('F', 'R', 'M', 'H')
896 static int decode_frame_headers(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
897 const uint8_t *buf, int buf_size)
900 const uint8_t *bs_hdr;
901 uint32_t frame_num, word2, check_sum, data_size;
902 int y_offset, u_offset, v_offset;
903 uint32_t starts[3], ends[3];
904 uint16_t height, width;
907 bytestream2_init(&gb, buf, buf_size);
909 /* parse and check the OS header */
910 frame_num = bytestream2_get_le32(&gb);
911 word2 = bytestream2_get_le32(&gb);
912 check_sum = bytestream2_get_le32(&gb);
913 data_size = bytestream2_get_le32(&gb);
915 if ((frame_num ^ word2 ^ data_size ^ OS_HDR_ID) != check_sum) {
916 av_log(avctx, AV_LOG_ERROR, "OS header checksum mismatch!\n");
917 return AVERROR_INVALIDDATA;
920 /* parse the bitstream header */
923 if (bytestream2_get_le16(&gb) != 32) {
924 av_log(avctx, AV_LOG_ERROR, "Unsupported codec version!\n");
925 return AVERROR_INVALIDDATA;
928 ctx->frame_num = frame_num;
929 ctx->frame_flags = bytestream2_get_le16(&gb);
930 ctx->data_size = (bytestream2_get_le32(&gb) + 7) >> 3;
931 ctx->cb_offset = bytestream2_get_byte(&gb);
933 if (ctx->data_size == 16)
935 ctx->data_size = FFMIN(ctx->data_size, buf_size - 16);
937 bytestream2_skip(&gb, 3); // skip reserved byte and checksum
939 /* check frame dimensions */
940 height = bytestream2_get_le16(&gb);
941 width = bytestream2_get_le16(&gb);
942 if (av_image_check_size(width, height, 0, avctx))
943 return AVERROR_INVALIDDATA;
945 if (width != ctx->width || height != ctx->height) {
948 ff_dlog(avctx, "Frame dimensions changed!\n");
950 if (width < 16 || width > 640 ||
951 height < 16 || height > 480 ||
952 width & 3 || height & 3) {
953 av_log(avctx, AV_LOG_ERROR,
954 "Invalid picture dimensions: %d x %d!\n", width, height);
955 return AVERROR_INVALIDDATA;
957 free_frame_buffers(ctx);
958 if ((res = allocate_frame_buffers(ctx, avctx, width, height)) < 0)
960 if ((res = ff_set_dimensions(avctx, width, height)) < 0)
964 y_offset = bytestream2_get_le32(&gb);
965 v_offset = bytestream2_get_le32(&gb);
966 u_offset = bytestream2_get_le32(&gb);
967 bytestream2_skip(&gb, 4);
969 /* unfortunately there is no common order of planes in the buffer */
970 /* so we use that sorting algo for determining planes data sizes */
971 starts[0] = y_offset;
972 starts[1] = v_offset;
973 starts[2] = u_offset;
975 for (j = 0; j < 3; j++) {
976 ends[j] = ctx->data_size;
977 for (i = 2; i >= 0; i--)
978 if (starts[i] < ends[j] && starts[i] > starts[j])
982 ctx->y_data_size = ends[0] - starts[0];
983 ctx->v_data_size = ends[1] - starts[1];
984 ctx->u_data_size = ends[2] - starts[2];
985 if (FFMIN3(y_offset, v_offset, u_offset) < 0 ||
986 FFMAX3(y_offset, v_offset, u_offset) >= ctx->data_size - 16 ||
987 FFMIN3(y_offset, v_offset, u_offset) < gb.buffer - bs_hdr + 16 ||
988 FFMIN3(ctx->y_data_size, ctx->v_data_size, ctx->u_data_size) <= 0) {
989 av_log(avctx, AV_LOG_ERROR, "One of the y/u/v offsets is invalid\n");
990 return AVERROR_INVALIDDATA;
993 ctx->y_data_ptr = bs_hdr + y_offset;
994 ctx->v_data_ptr = bs_hdr + v_offset;
995 ctx->u_data_ptr = bs_hdr + u_offset;
996 ctx->alt_quant = gb.buffer;
998 if (ctx->data_size == 16) {
999 av_log(avctx, AV_LOG_DEBUG, "Sync frame encountered!\n");
1003 if (ctx->frame_flags & BS_8BIT_PEL) {
1004 avpriv_request_sample(avctx, "8-bit pixel format");
1005 return AVERROR_PATCHWELCOME;
1008 if (ctx->frame_flags & BS_MV_X_HALF || ctx->frame_flags & BS_MV_Y_HALF) {
1009 avpriv_request_sample(avctx, "Halfpel motion vectors");
1010 return AVERROR_PATCHWELCOME;
1018 * Convert and output the current plane.
1019 * All pixel values will be upsampled by shifting right by one bit.
1021 * @param[in] plane pointer to the descriptor of the plane being processed
1022 * @param[in] buf_sel indicates which frame buffer the input data stored in
1023 * @param[out] dst pointer to the buffer receiving converted pixels
1024 * @param[in] dst_pitch pitch for moving to the next y line
1025 * @param[in] dst_height output plane height
1027 static void output_plane(const Plane *plane, int buf_sel, uint8_t *dst,
1028 int dst_pitch, int dst_height)
1031 const uint8_t *src = plane->pixels[buf_sel];
1032 uint32_t pitch = plane->pitch;
1034 dst_height = FFMIN(dst_height, plane->height);
1035 for (y = 0; y < dst_height; y++) {
1036 /* convert four pixels at once using SWAR */
1037 for (x = 0; x < plane->width >> 2; x++) {
1038 AV_WN32A(dst, (AV_RN32A(src) & 0x7F7F7F7F) << 1);
1043 for (x <<= 2; x < plane->width; x++)
1044 *dst++ = *src++ << 1;
1046 src += pitch - plane->width;
1047 dst += dst_pitch - plane->width;
1052 static av_cold int decode_init(AVCodecContext *avctx)
1054 Indeo3DecodeContext *ctx = avctx->priv_data;
1057 avctx->pix_fmt = AV_PIX_FMT_YUV410P;
1059 build_requant_tab();
1061 ff_hpeldsp_init(&ctx->hdsp, avctx->flags);
1063 return allocate_frame_buffers(ctx, avctx, avctx->width, avctx->height);
1067 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
1070 Indeo3DecodeContext *ctx = avctx->priv_data;
1071 const uint8_t *buf = avpkt->data;
1072 int buf_size = avpkt->size;
1073 AVFrame *frame = data;
1076 res = decode_frame_headers(ctx, avctx, buf, buf_size);
1080 /* skip sync(null) frames */
1082 // we have processed 16 bytes but no data was decoded
1087 /* skip droppable INTER frames if requested */
1088 if (ctx->frame_flags & BS_NONREF &&
1089 (avctx->skip_frame >= AVDISCARD_NONREF))
1092 /* skip INTER frames if requested */
1093 if (!(ctx->frame_flags & BS_KEYFRAME) && avctx->skip_frame >= AVDISCARD_NONKEY)
1096 /* use BS_BUFFER flag for buffer switching */
1097 ctx->buf_sel = (ctx->frame_flags >> BS_BUFFER) & 1;
1099 if ((res = ff_get_buffer(avctx, frame, 0)) < 0)
1102 /* decode luma plane */
1103 if ((res = decode_plane(ctx, avctx, ctx->planes, ctx->y_data_ptr, ctx->y_data_size, 40)))
1106 /* decode chroma planes */
1107 if ((res = decode_plane(ctx, avctx, &ctx->planes[1], ctx->u_data_ptr, ctx->u_data_size, 10)))
1110 if ((res = decode_plane(ctx, avctx, &ctx->planes[2], ctx->v_data_ptr, ctx->v_data_size, 10)))
1113 output_plane(&ctx->planes[0], ctx->buf_sel,
1114 frame->data[0], frame->linesize[0],
1116 output_plane(&ctx->planes[1], ctx->buf_sel,
1117 frame->data[1], frame->linesize[1],
1118 (avctx->height + 3) >> 2);
1119 output_plane(&ctx->planes[2], ctx->buf_sel,
1120 frame->data[2], frame->linesize[2],
1121 (avctx->height + 3) >> 2);
1129 static av_cold int decode_close(AVCodecContext *avctx)
1131 free_frame_buffers(avctx->priv_data);
1136 AVCodec ff_indeo3_decoder = {
1138 .long_name = NULL_IF_CONFIG_SMALL("Intel Indeo 3"),
1139 .type = AVMEDIA_TYPE_VIDEO,
1140 .id = AV_CODEC_ID_INDEO3,
1141 .priv_data_size = sizeof(Indeo3DecodeContext),
1142 .init = decode_init,
1143 .close = decode_close,
1144 .decode = decode_frame,
1145 .capabilities = AV_CODEC_CAP_DR1,