2 * Copyright (C) 2007 Marco Gerards <marco@gnu.org>
3 * Copyright (C) 2009 David Conrad
4 * Copyright (C) 2011 Jordi Ortiz
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * @author Marco Gerards <marco@gnu.org>, David Conrad, Jordi Ortiz <nenjordi@gmail.com>
31 #include "bytestream.h"
34 #include "dirac_arith.h"
35 #include "dirac_vlc.h"
36 #include "mpeg12data.h"
37 #include "libavcodec/mpegvideo.h"
38 #include "mpegvideoencdsp.h"
39 #include "dirac_dwt.h"
46 * The spec limits this to 3 for frame coding, but in practice can be as high as 6
48 #define MAX_REFERENCE_FRAMES 8
49 #define MAX_DELAY 5 /* limit for main profile for frame coding (TODO: field coding) */
50 #define MAX_FRAMES (MAX_REFERENCE_FRAMES + MAX_DELAY + 1)
51 #define MAX_QUANT 255 /* max quant for VC-2 */
52 #define MAX_BLOCKSIZE 32 /* maximum xblen/yblen we support */
55 * DiracBlock->ref flags, if set then the block does MC from the given ref
57 #define DIRAC_REF_MASK_REF1 1
58 #define DIRAC_REF_MASK_REF2 2
59 #define DIRAC_REF_MASK_GLOBAL 4
62 * Value of Picture.reference when Picture is not a reference picture, but
63 * is held for delayed output.
65 #define DELAYED_PIC_REF 4
67 #define CALC_PADDING(size, depth) \
68 (((size + (1 << depth) - 1) >> depth) << depth)
70 #define DIVRNDUP(a, b) (((a) + (b) - 1) / (b))
74 int interpolated[3]; /* 1 if hpel[] is valid */
76 uint8_t *hpel_base[3][4];
84 } u; /* anonymous unions aren't in C99 :( */
88 typedef struct SubBand {
91 int stride; /* in bytes */
97 struct SubBand *parent;
101 const uint8_t *coeff_data;
104 typedef struct Plane {
114 /* block separation (block n+1 starts after this many pixels in block n) */
117 /* amount of overspill on each edge (half of the overlap between blocks) */
121 SubBand band[MAX_DWT_LEVELS][4];
124 /* Used by Low Delay and High Quality profiles */
125 typedef struct DiracSlice {
132 typedef struct DiracContext {
133 AVCodecContext *avctx;
134 MpegvideoEncDSPContext mpvencdsp;
135 VideoDSPContext vdsp;
136 DiracDSPContext diracdsp;
137 DiracGolombLUT *reader_ctx;
138 DiracVersionInfo version;
140 AVDiracSeqHeader seq;
141 int seen_sequence_header;
142 int frame_number; /* number of the next frame to display */
147 int bit_depth; /* bit depth */
148 int pshift; /* pixel shift = bit_depth > 8 */
150 int zero_res; /* zero residue flag */
151 int is_arith; /* whether coeffs use arith or golomb coding */
152 int core_syntax; /* use core syntax only */
153 int low_delay; /* use the low delay syntax */
154 int hq_picture; /* high quality picture, enables low_delay */
155 int ld_picture; /* use low delay picture, turns on low_delay */
156 int dc_prediction; /* has dc prediction */
157 int globalmc_flag; /* use global motion compensation */
158 int num_refs; /* number of reference pictures */
160 /* wavelet decoding */
161 unsigned wavelet_depth; /* depth of the IDWT */
162 unsigned wavelet_idx;
165 * schroedinger older than 1.0.8 doesn't store
166 * quant delta if only one codebook exists in a band
168 unsigned old_delta_quant;
169 unsigned codeblock_mode;
171 unsigned num_x; /* number of horizontal slices */
172 unsigned num_y; /* number of vertical slices */
174 uint8_t *thread_buf; /* Per-thread buffer for coefficient storage */
175 int threads_num_buf; /* Current # of buffers allocated */
176 int thread_buf_size; /* Each thread has a buffer this size */
178 DiracSlice *slice_params_buf;
179 int slice_params_num_buf;
184 } codeblock[MAX_DWT_LEVELS+1];
187 AVRational bytes; /* average bytes per slice */
188 uint8_t quant[MAX_DWT_LEVELS][4]; /* [DIRAC_STD] E.1 */
192 unsigned prefix_bytes;
193 uint64_t size_scaler;
197 int pan_tilt[2]; /* pan/tilt vector */
198 int zrs[2][2]; /* zoom/rotate/shear matrix */
199 int perspective[2]; /* perspective vector */
201 unsigned perspective_exp;
204 /* motion compensation */
205 uint8_t mv_precision; /* [DIRAC_STD] REFS_WT_PRECISION */
206 int16_t weight[2]; /* [DIRAC_STD] REF1_WT and REF2_WT */
207 unsigned weight_log2denom; /* [DIRAC_STD] REFS_WT_PRECISION */
209 int blwidth; /* number of blocks (horizontally) */
210 int blheight; /* number of blocks (vertically) */
211 int sbwidth; /* number of superblocks (horizontally) */
212 int sbheight; /* number of superblocks (vertically) */
215 DiracBlock *blmotion;
217 uint8_t *edge_emu_buffer[4];
218 uint8_t *edge_emu_buffer_base;
220 uint16_t *mctmp; /* buffer holding the MC data multiplied by OBMC weights */
224 DECLARE_ALIGNED(16, uint8_t, obmc_weight)[3][MAX_BLOCKSIZE*MAX_BLOCKSIZE];
226 void (*put_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
227 void (*avg_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
228 void (*add_obmc)(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
229 dirac_weight_func weight_func;
230 dirac_biweight_func biweight_func;
232 DiracFrame *current_picture;
233 DiracFrame *ref_pics[2];
235 DiracFrame *ref_frames[MAX_REFERENCE_FRAMES+1];
236 DiracFrame *delay_frames[MAX_DELAY+1];
237 DiracFrame all_frames[MAX_FRAMES];
248 /* magic number division by 3 from schroedinger */
249 static inline int divide3(int x)
251 return ((x+1)*21845 + 10922) >> 16;
254 static DiracFrame *remove_frame(DiracFrame *framelist[], int picnum)
256 DiracFrame *remove_pic = NULL;
257 int i, remove_idx = -1;
259 for (i = 0; framelist[i]; i++)
260 if (framelist[i]->avframe->display_picture_number == picnum) {
261 remove_pic = framelist[i];
266 for (i = remove_idx; framelist[i]; i++)
267 framelist[i] = framelist[i+1];
272 static int add_frame(DiracFrame *framelist[], int maxframes, DiracFrame *frame)
275 for (i = 0; i < maxframes; i++)
277 framelist[i] = frame;
283 static int alloc_sequence_buffers(DiracContext *s)
285 int sbwidth = DIVRNDUP(s->seq.width, 4);
286 int sbheight = DIVRNDUP(s->seq.height, 4);
287 int i, w, h, top_padding;
289 /* todo: think more about this / use or set Plane here */
290 for (i = 0; i < 3; i++) {
291 int max_xblen = MAX_BLOCKSIZE >> (i ? s->chroma_x_shift : 0);
292 int max_yblen = MAX_BLOCKSIZE >> (i ? s->chroma_y_shift : 0);
293 w = s->seq.width >> (i ? s->chroma_x_shift : 0);
294 h = s->seq.height >> (i ? s->chroma_y_shift : 0);
296 /* we allocate the max we support here since num decompositions can
297 * change from frame to frame. Stride is aligned to 16 for SIMD, and
298 * 1<<MAX_DWT_LEVELS top padding to avoid if(y>0) in arith decoding
299 * MAX_BLOCKSIZE padding for MC: blocks can spill up to half of that
301 top_padding = FFMAX(1<<MAX_DWT_LEVELS, max_yblen/2);
302 w = FFALIGN(CALC_PADDING(w, MAX_DWT_LEVELS), 8); /* FIXME: Should this be 16 for SSE??? */
303 h = top_padding + CALC_PADDING(h, MAX_DWT_LEVELS) + max_yblen/2;
305 s->plane[i].idwt.buf_base = av_mallocz_array((w+max_xblen), h * (2 << s->pshift));
306 s->plane[i].idwt.tmp = av_malloc_array((w+16), 2 << s->pshift);
307 s->plane[i].idwt.buf = s->plane[i].idwt.buf_base + (top_padding*w)*(2 << s->pshift);
308 if (!s->plane[i].idwt.buf_base || !s->plane[i].idwt.tmp)
309 return AVERROR(ENOMEM);
312 /* fixme: allocate using real stride here */
313 s->sbsplit = av_malloc_array(sbwidth, sbheight);
314 s->blmotion = av_malloc_array(sbwidth, sbheight * 16 * sizeof(*s->blmotion));
316 if (!s->sbsplit || !s->blmotion)
317 return AVERROR(ENOMEM);
321 static int alloc_buffers(DiracContext *s, int stride)
323 int w = s->seq.width;
324 int h = s->seq.height;
326 av_assert0(stride >= w);
329 if (s->buffer_stride >= stride)
331 s->buffer_stride = 0;
333 av_freep(&s->edge_emu_buffer_base);
334 memset(s->edge_emu_buffer, 0, sizeof(s->edge_emu_buffer));
336 av_freep(&s->mcscratch);
338 s->edge_emu_buffer_base = av_malloc_array(stride, MAX_BLOCKSIZE);
340 s->mctmp = av_malloc_array((stride+MAX_BLOCKSIZE), (h+MAX_BLOCKSIZE) * sizeof(*s->mctmp));
341 s->mcscratch = av_malloc_array(stride, MAX_BLOCKSIZE);
343 if (!s->edge_emu_buffer_base || !s->mctmp || !s->mcscratch)
344 return AVERROR(ENOMEM);
346 s->buffer_stride = stride;
350 static void free_sequence_buffers(DiracContext *s)
354 for (i = 0; i < MAX_FRAMES; i++) {
355 if (s->all_frames[i].avframe->data[0]) {
356 av_frame_unref(s->all_frames[i].avframe);
357 memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));
360 for (j = 0; j < 3; j++)
361 for (k = 1; k < 4; k++)
362 av_freep(&s->all_frames[i].hpel_base[j][k]);
365 memset(s->ref_frames, 0, sizeof(s->ref_frames));
366 memset(s->delay_frames, 0, sizeof(s->delay_frames));
368 for (i = 0; i < 3; i++) {
369 av_freep(&s->plane[i].idwt.buf_base);
370 av_freep(&s->plane[i].idwt.tmp);
373 s->buffer_stride = 0;
374 av_freep(&s->sbsplit);
375 av_freep(&s->blmotion);
376 av_freep(&s->edge_emu_buffer_base);
379 av_freep(&s->mcscratch);
382 static av_cold int dirac_decode_init(AVCodecContext *avctx)
384 DiracContext *s = avctx->priv_data;
388 s->frame_number = -1;
390 s->thread_buf = NULL;
391 s->threads_num_buf = -1;
392 s->thread_buf_size = -1;
394 ff_dirac_golomb_reader_init(&s->reader_ctx);
395 ff_diracdsp_init(&s->diracdsp);
396 ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
397 ff_videodsp_init(&s->vdsp, 8);
399 for (i = 0; i < MAX_FRAMES; i++) {
400 s->all_frames[i].avframe = av_frame_alloc();
401 if (!s->all_frames[i].avframe) {
403 av_frame_free(&s->all_frames[--i].avframe);
404 return AVERROR(ENOMEM);
411 static void dirac_decode_flush(AVCodecContext *avctx)
413 DiracContext *s = avctx->priv_data;
414 free_sequence_buffers(s);
415 s->seen_sequence_header = 0;
416 s->frame_number = -1;
419 static av_cold int dirac_decode_end(AVCodecContext *avctx)
421 DiracContext *s = avctx->priv_data;
424 ff_dirac_golomb_reader_end(&s->reader_ctx);
426 dirac_decode_flush(avctx);
427 for (i = 0; i < MAX_FRAMES; i++)
428 av_frame_free(&s->all_frames[i].avframe);
430 av_freep(&s->thread_buf);
431 av_freep(&s->slice_params_buf);
436 static inline int coeff_unpack_golomb(GetBitContext *gb, int qfactor, int qoffset)
438 int coeff = dirac_get_se_golomb(gb);
439 const int sign = FFSIGN(coeff);
441 coeff = sign*((sign * coeff * qfactor + qoffset) >> 2);
445 #define SIGN_CTX(x) (CTX_SIGN_ZERO + ((x) > 0) - ((x) < 0))
447 #define UNPACK_ARITH(n, type) \
448 static inline void coeff_unpack_arith_##n(DiracArith *c, int qfactor, int qoffset, \
449 SubBand *b, type *buf, int x, int y) \
451 int coeff, sign, sign_pred = 0, pred_ctx = CTX_ZPZN_F1; \
452 const int mstride = -(b->stride >> (1+b->pshift)); \
454 const type *pbuf = (type *)b->parent->ibuf; \
455 const int stride = b->parent->stride >> (1+b->parent->pshift); \
456 pred_ctx += !!pbuf[stride * (y>>1) + (x>>1)] << 1; \
458 if (b->orientation == subband_hl) \
459 sign_pred = buf[mstride]; \
461 pred_ctx += !(buf[-1] | buf[mstride] | buf[-1 + mstride]); \
462 if (b->orientation == subband_lh) \
463 sign_pred = buf[-1]; \
465 pred_ctx += !buf[mstride]; \
467 coeff = dirac_get_arith_uint(c, pred_ctx, CTX_COEFF_DATA); \
469 coeff = (coeff * qfactor + qoffset) >> 2; \
470 sign = dirac_get_arith_bit(c, SIGN_CTX(sign_pred)); \
471 coeff = (coeff ^ -sign) + sign; \
476 UNPACK_ARITH(8, int16_t)
477 UNPACK_ARITH(10, int32_t)
480 * Decode the coeffs in the rectangle defined by left, right, top, bottom
481 * [DIRAC_STD] 13.4.3.2 Codeblock unpacking loop. codeblock()
483 static inline void codeblock(DiracContext *s, SubBand *b,
484 GetBitContext *gb, DiracArith *c,
485 int left, int right, int top, int bottom,
486 int blockcnt_one, int is_arith)
488 int x, y, zero_block;
489 int qoffset, qfactor;
492 /* check for any coded coefficients in this codeblock */
495 zero_block = dirac_get_arith_bit(c, CTX_ZERO_BLOCK);
497 zero_block = get_bits1(gb);
503 if (s->codeblock_mode && !(s->old_delta_quant && blockcnt_one)) {
504 int quant = b->quant;
506 quant += dirac_get_arith_int(c, CTX_DELTA_Q_F, CTX_DELTA_Q_DATA);
508 quant += dirac_get_se_golomb(gb);
510 av_log(s->avctx, AV_LOG_ERROR, "Invalid quant\n");
516 if (b->quant > (DIRAC_MAX_QUANT_INDEX - 1)) {
517 av_log(s->avctx, AV_LOG_ERROR, "Unsupported quant %d\n", b->quant);
522 qfactor = ff_dirac_qscale_tab[b->quant];
523 /* TODO: context pointer? */
525 qoffset = ff_dirac_qoffset_intra_tab[b->quant] + 2;
527 qoffset = ff_dirac_qoffset_inter_tab[b->quant] + 2;
529 buf = b->ibuf + top * b->stride;
531 for (y = top; y < bottom; y++) {
532 for (x = left; x < right; x++) {
534 coeff_unpack_arith_10(c, qfactor, qoffset, b, (int32_t*)(buf)+x, x, y);
536 coeff_unpack_arith_8(c, qfactor, qoffset, b, (int16_t*)(buf)+x, x, y);
542 for (y = top; y < bottom; y++) {
543 for (x = left; x < right; x++) {
544 int val = coeff_unpack_golomb(gb, qfactor, qoffset);
546 AV_WN32(&buf[4*x], val);
548 AV_WN16(&buf[2*x], val);
557 * Dirac Specification ->
558 * 13.3 intra_dc_prediction(band)
560 #define INTRA_DC_PRED(n, type) \
561 static inline void intra_dc_prediction_##n(SubBand *b) \
563 type *buf = (type*)b->ibuf; \
566 for (x = 1; x < b->width; x++) \
567 buf[x] += buf[x-1]; \
568 buf += (b->stride >> (1+b->pshift)); \
570 for (y = 1; y < b->height; y++) { \
571 buf[0] += buf[-(b->stride >> (1+b->pshift))]; \
573 for (x = 1; x < b->width; x++) { \
574 int pred = buf[x - 1] + buf[x - (b->stride >> (1+b->pshift))] + buf[x - (b->stride >> (1+b->pshift))-1]; \
575 buf[x] += divide3(pred); \
577 buf += (b->stride >> (1+b->pshift)); \
581 INTRA_DC_PRED(8, int16_t)
582 INTRA_DC_PRED(10, int32_t)
585 * Dirac Specification ->
586 * 13.4.2 Non-skipped subbands. subband_coeffs()
588 static av_always_inline void decode_subband_internal(DiracContext *s, SubBand *b, int is_arith)
590 int cb_x, cb_y, left, right, top, bottom;
593 int cb_width = s->codeblock[b->level + (b->orientation != subband_ll)].width;
594 int cb_height = s->codeblock[b->level + (b->orientation != subband_ll)].height;
595 int blockcnt_one = (cb_width + cb_height) == 2;
600 init_get_bits8(&gb, b->coeff_data, b->length);
603 ff_dirac_init_arith_decoder(&c, &gb, b->length);
606 for (cb_y = 0; cb_y < cb_height; cb_y++) {
607 bottom = (b->height * (cb_y+1LL)) / cb_height;
609 for (cb_x = 0; cb_x < cb_width; cb_x++) {
610 right = (b->width * (cb_x+1LL)) / cb_width;
611 codeblock(s, b, &gb, &c, left, right, top, bottom, blockcnt_one, is_arith);
617 if (b->orientation == subband_ll && s->num_refs == 0) {
619 intra_dc_prediction_10(b);
621 intra_dc_prediction_8(b);
626 static int decode_subband_arith(AVCodecContext *avctx, void *b)
628 DiracContext *s = avctx->priv_data;
629 decode_subband_internal(s, b, 1);
633 static int decode_subband_golomb(AVCodecContext *avctx, void *arg)
635 DiracContext *s = avctx->priv_data;
637 decode_subband_internal(s, *b, 0);
642 * Dirac Specification ->
643 * [DIRAC_STD] 13.4.1 core_transform_data()
645 static void decode_component(DiracContext *s, int comp)
647 AVCodecContext *avctx = s->avctx;
648 SubBand *bands[3*MAX_DWT_LEVELS+1];
649 enum dirac_subband orientation;
650 int level, num_bands = 0;
652 /* Unpack all subbands at all levels. */
653 for (level = 0; level < s->wavelet_depth; level++) {
654 for (orientation = !!level; orientation < 4; orientation++) {
655 SubBand *b = &s->plane[comp].band[level][orientation];
656 bands[num_bands++] = b;
658 align_get_bits(&s->gb);
659 /* [DIRAC_STD] 13.4.2 subband() */
660 b->length = get_interleaved_ue_golomb(&s->gb);
662 b->quant = get_interleaved_ue_golomb(&s->gb);
663 align_get_bits(&s->gb);
664 b->coeff_data = s->gb.buffer + get_bits_count(&s->gb)/8;
665 b->length = FFMIN(b->length, FFMAX(get_bits_left(&s->gb)/8, 0));
666 skip_bits_long(&s->gb, b->length*8);
669 /* arithmetic coding has inter-level dependencies, so we can only execute one level at a time */
671 avctx->execute(avctx, decode_subband_arith, &s->plane[comp].band[level][!!level],
672 NULL, 4-!!level, sizeof(SubBand));
674 /* golomb coding has no inter-level dependencies, so we can execute all subbands in parallel */
676 avctx->execute(avctx, decode_subband_golomb, bands, NULL, num_bands, sizeof(SubBand*));
679 #define PARSE_VALUES(type, x, gb, ebits, buf1, buf2) \
680 type *buf = (type *)buf1; \
681 buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); \
682 if (get_bits_count(gb) >= ebits) \
685 buf = (type *)buf2; \
686 buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); \
687 if (get_bits_count(gb) >= ebits) \
691 static void decode_subband(DiracContext *s, GetBitContext *gb, int quant,
692 int slice_x, int slice_y, int bits_end,
693 SubBand *b1, SubBand *b2)
695 int left = b1->width * slice_x / s->num_x;
696 int right = b1->width *(slice_x+1) / s->num_x;
697 int top = b1->height * slice_y / s->num_y;
698 int bottom = b1->height *(slice_y+1) / s->num_y;
700 int qfactor, qoffset;
702 uint8_t *buf1 = b1->ibuf + top * b1->stride;
703 uint8_t *buf2 = b2 ? b2->ibuf + top * b2->stride: NULL;
706 if (quant > (DIRAC_MAX_QUANT_INDEX - 1)) {
707 av_log(s->avctx, AV_LOG_ERROR, "Unsupported quant %d\n", quant);
710 qfactor = ff_dirac_qscale_tab[quant];
711 qoffset = ff_dirac_qoffset_intra_tab[quant] + 2;
712 /* we have to constantly check for overread since the spec explicitly
713 requires this, with the meaning that all remaining coeffs are set to 0 */
714 if (get_bits_count(gb) >= bits_end)
718 for (y = top; y < bottom; y++) {
719 for (x = left; x < right; x++) {
720 PARSE_VALUES(int32_t, x, gb, bits_end, buf1, buf2);
728 for (y = top; y < bottom; y++) {
729 for (x = left; x < right; x++) {
730 PARSE_VALUES(int16_t, x, gb, bits_end, buf1, buf2);
740 * Dirac Specification ->
741 * 13.5.2 Slices. slice(sx,sy)
743 static int decode_lowdelay_slice(AVCodecContext *avctx, void *arg)
745 DiracContext *s = avctx->priv_data;
746 DiracSlice *slice = arg;
747 GetBitContext *gb = &slice->gb;
748 enum dirac_subband orientation;
749 int level, quant, chroma_bits, chroma_end;
751 int quant_base = get_bits(gb, 7); /*[DIRAC_STD] qindex */
752 int length_bits = av_log2(8 * slice->bytes)+1;
753 int luma_bits = get_bits_long(gb, length_bits);
754 int luma_end = get_bits_count(gb) + FFMIN(luma_bits, get_bits_left(gb));
756 /* [DIRAC_STD] 13.5.5.2 luma_slice_band */
757 for (level = 0; level < s->wavelet_depth; level++)
758 for (orientation = !!level; orientation < 4; orientation++) {
759 quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);
760 decode_subband(s, gb, quant, slice->slice_x, slice->slice_y, luma_end,
761 &s->plane[0].band[level][orientation], NULL);
764 /* consume any unused bits from luma */
765 skip_bits_long(gb, get_bits_count(gb) - luma_end);
767 chroma_bits = 8*slice->bytes - 7 - length_bits - luma_bits;
768 chroma_end = get_bits_count(gb) + FFMIN(chroma_bits, get_bits_left(gb));
769 /* [DIRAC_STD] 13.5.5.3 chroma_slice_band */
770 for (level = 0; level < s->wavelet_depth; level++)
771 for (orientation = !!level; orientation < 4; orientation++) {
772 quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);
773 decode_subband(s, gb, quant, slice->slice_x, slice->slice_y, chroma_end,
774 &s->plane[1].band[level][orientation],
775 &s->plane[2].band[level][orientation]);
781 typedef struct SliceCoeffs {
789 static int subband_coeffs(DiracContext *s, int x, int y, int p,
790 SliceCoeffs c[MAX_DWT_LEVELS])
793 for (level = 0; level < s->wavelet_depth; level++) {
794 SliceCoeffs *o = &c[level];
795 SubBand *b = &s->plane[p].band[level][3]; /* orientation doens't matter */
796 o->top = b->height * y / s->num_y;
797 o->left = b->width * x / s->num_x;
798 o->tot_h = ((b->width * (x + 1)) / s->num_x) - o->left;
799 o->tot_v = ((b->height * (y + 1)) / s->num_y) - o->top;
800 o->tot = o->tot_h*o->tot_v;
801 coef += o->tot * (4 - !!level);
807 * VC-2 Specification ->
808 * 13.5.3 hq_slice(sx,sy)
810 static int decode_hq_slice(DiracContext *s, DiracSlice *slice, uint8_t *tmp_buf)
812 int i, level, orientation, quant_idx;
813 int qfactor[MAX_DWT_LEVELS][4], qoffset[MAX_DWT_LEVELS][4];
814 GetBitContext *gb = &slice->gb;
815 SliceCoeffs coeffs_num[MAX_DWT_LEVELS];
817 skip_bits_long(gb, 8*s->highquality.prefix_bytes);
818 quant_idx = get_bits(gb, 8);
820 if (quant_idx > DIRAC_MAX_QUANT_INDEX) {
821 av_log(s->avctx, AV_LOG_ERROR, "Invalid quantization index - %i\n", quant_idx);
822 return AVERROR_INVALIDDATA;
825 /* Slice quantization (slice_quantizers() in the specs) */
826 for (level = 0; level < s->wavelet_depth; level++) {
827 for (orientation = !!level; orientation < 4; orientation++) {
828 const int quant = FFMAX(quant_idx - s->lowdelay.quant[level][orientation], 0);
829 qfactor[level][orientation] = ff_dirac_qscale_tab[quant];
830 qoffset[level][orientation] = ff_dirac_qoffset_intra_tab[quant] + 2;
834 /* Luma + 2 Chroma planes */
835 for (i = 0; i < 3; i++) {
836 int coef_num, coef_par, off = 0;
837 int64_t length = s->highquality.size_scaler*get_bits(gb, 8);
838 int64_t bits_end = get_bits_count(gb) + 8*length;
839 const uint8_t *addr = align_get_bits(gb);
841 if (length*8 > get_bits_left(gb)) {
842 av_log(s->avctx, AV_LOG_ERROR, "end too far away\n");
843 return AVERROR_INVALIDDATA;
846 coef_num = subband_coeffs(s, slice->slice_x, slice->slice_y, i, coeffs_num);
849 coef_par = ff_dirac_golomb_read_32bit(s->reader_ctx, addr,
850 length, tmp_buf, coef_num);
852 coef_par = ff_dirac_golomb_read_16bit(s->reader_ctx, addr,
853 length, tmp_buf, coef_num);
855 if (coef_num > coef_par) {
856 const int start_b = coef_par * (1 << (s->pshift + 1));
857 const int end_b = coef_num * (1 << (s->pshift + 1));
858 memset(&tmp_buf[start_b], 0, end_b - start_b);
861 for (level = 0; level < s->wavelet_depth; level++) {
862 const SliceCoeffs *c = &coeffs_num[level];
863 for (orientation = !!level; orientation < 4; orientation++) {
864 const SubBand *b1 = &s->plane[i].band[level][orientation];
865 uint8_t *buf = b1->ibuf + c->top * b1->stride + (c->left << (s->pshift + 1));
867 /* Change to c->tot_h <= 4 for AVX2 dequantization */
868 const int qfunc = s->pshift + 2*(c->tot_h <= 2);
869 s->diracdsp.dequant_subband[qfunc](&tmp_buf[off], buf, b1->stride,
870 qfactor[level][orientation],
871 qoffset[level][orientation],
874 off += c->tot << (s->pshift + 1);
878 skip_bits_long(gb, bits_end - get_bits_count(gb));
884 static int decode_hq_slice_row(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
887 DiracContext *s = avctx->priv_data;
888 DiracSlice *slices = ((DiracSlice *)arg) + s->num_x*jobnr;
889 uint8_t *thread_buf = &s->thread_buf[s->thread_buf_size*threadnr];
890 for (i = 0; i < s->num_x; i++)
891 decode_hq_slice(s, &slices[i], thread_buf);
896 * Dirac Specification ->
897 * 13.5.1 low_delay_transform_data()
899 static int decode_lowdelay(DiracContext *s)
901 AVCodecContext *avctx = s->avctx;
902 int slice_x, slice_y, bufsize;
903 int64_t coef_buf_size, bytes = 0;
906 SliceCoeffs tmp[MAX_DWT_LEVELS];
909 if (s->slice_params_num_buf != (s->num_x * s->num_y)) {
910 s->slice_params_buf = av_realloc_f(s->slice_params_buf, s->num_x * s->num_y, sizeof(DiracSlice));
911 if (!s->slice_params_buf) {
912 av_log(s->avctx, AV_LOG_ERROR, "slice params buffer allocation failure\n");
913 s->slice_params_num_buf = 0;
914 return AVERROR(ENOMEM);
916 s->slice_params_num_buf = s->num_x * s->num_y;
918 slices = s->slice_params_buf;
920 /* 8 becacuse that's how much the golomb reader could overread junk data
921 * from another plane/slice at most, and 512 because SIMD */
922 coef_buf_size = subband_coeffs(s, s->num_x - 1, s->num_y - 1, 0, tmp) + 8;
923 coef_buf_size = (coef_buf_size << (1 + s->pshift)) + 512;
925 if (s->threads_num_buf != avctx->thread_count ||
926 s->thread_buf_size != coef_buf_size) {
927 s->threads_num_buf = avctx->thread_count;
928 s->thread_buf_size = coef_buf_size;
929 s->thread_buf = av_realloc_f(s->thread_buf, avctx->thread_count, s->thread_buf_size);
930 if (!s->thread_buf) {
931 av_log(s->avctx, AV_LOG_ERROR, "thread buffer allocation failure\n");
932 return AVERROR(ENOMEM);
936 align_get_bits(&s->gb);
937 /*[DIRAC_STD] 13.5.2 Slices. slice(sx,sy) */
938 buf = s->gb.buffer + get_bits_count(&s->gb)/8;
939 bufsize = get_bits_left(&s->gb);
944 for (slice_y = 0; bufsize > 0 && slice_y < s->num_y; slice_y++) {
945 for (slice_x = 0; bufsize > 0 && slice_x < s->num_x; slice_x++) {
946 bytes = s->highquality.prefix_bytes + 1;
947 for (i = 0; i < 3; i++) {
948 if (bytes <= bufsize/8)
949 bytes += buf[bytes] * s->highquality.size_scaler + 1;
951 if (bytes >= INT_MAX || bytes*8 > bufsize) {
952 av_log(s->avctx, AV_LOG_ERROR, "too many bytes\n");
953 return AVERROR_INVALIDDATA;
956 slices[slice_num].bytes = bytes;
957 slices[slice_num].slice_x = slice_x;
958 slices[slice_num].slice_y = slice_y;
959 init_get_bits(&slices[slice_num].gb, buf, bufsize);
963 if (bufsize/8 >= bytes)
970 if (s->num_x*s->num_y != slice_num) {
971 av_log(s->avctx, AV_LOG_ERROR, "too few slices\n");
972 return AVERROR_INVALIDDATA;
975 avctx->execute2(avctx, decode_hq_slice_row, slices, NULL, s->num_y);
977 for (slice_y = 0; bufsize > 0 && slice_y < s->num_y; slice_y++) {
978 for (slice_x = 0; bufsize > 0 && slice_x < s->num_x; slice_x++) {
979 bytes = (slice_num+1) * (int64_t)s->lowdelay.bytes.num / s->lowdelay.bytes.den
980 - slice_num * (int64_t)s->lowdelay.bytes.num / s->lowdelay.bytes.den;
981 slices[slice_num].bytes = bytes;
982 slices[slice_num].slice_x = slice_x;
983 slices[slice_num].slice_y = slice_y;
984 init_get_bits(&slices[slice_num].gb, buf, bufsize);
988 if (bufsize/8 >= bytes)
994 avctx->execute(avctx, decode_lowdelay_slice, slices, NULL, slice_num,
995 sizeof(DiracSlice)); /* [DIRAC_STD] 13.5.2 Slices */
998 if (s->dc_prediction) {
1000 intra_dc_prediction_10(&s->plane[0].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1001 intra_dc_prediction_10(&s->plane[1].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1002 intra_dc_prediction_10(&s->plane[2].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1004 intra_dc_prediction_8(&s->plane[0].band[0][0]);
1005 intra_dc_prediction_8(&s->plane[1].band[0][0]);
1006 intra_dc_prediction_8(&s->plane[2].band[0][0]);
1013 static void init_planes(DiracContext *s)
1015 int i, w, h, level, orientation;
1017 for (i = 0; i < 3; i++) {
1018 Plane *p = &s->plane[i];
1020 p->width = s->seq.width >> (i ? s->chroma_x_shift : 0);
1021 p->height = s->seq.height >> (i ? s->chroma_y_shift : 0);
1022 p->idwt.width = w = CALC_PADDING(p->width , s->wavelet_depth);
1023 p->idwt.height = h = CALC_PADDING(p->height, s->wavelet_depth);
1024 p->idwt.stride = FFALIGN(p->idwt.width, 8) << (1 + s->pshift);
1026 for (level = s->wavelet_depth-1; level >= 0; level--) {
1029 for (orientation = !!level; orientation < 4; orientation++) {
1030 SubBand *b = &p->band[level][orientation];
1032 b->pshift = s->pshift;
1033 b->ibuf = p->idwt.buf;
1035 b->stride = p->idwt.stride << (s->wavelet_depth - level);
1038 b->orientation = orientation;
1040 if (orientation & 1)
1041 b->ibuf += w << (1+b->pshift);
1042 if (orientation > 1)
1043 b->ibuf += (b->stride>>1);
1046 b->parent = &p->band[level-1][orientation];
1051 p->xblen = s->plane[0].xblen >> s->chroma_x_shift;
1052 p->yblen = s->plane[0].yblen >> s->chroma_y_shift;
1053 p->xbsep = s->plane[0].xbsep >> s->chroma_x_shift;
1054 p->ybsep = s->plane[0].ybsep >> s->chroma_y_shift;
1057 p->xoffset = (p->xblen - p->xbsep)/2;
1058 p->yoffset = (p->yblen - p->ybsep)/2;
1063 * Unpack the motion compensation parameters
1064 * Dirac Specification ->
1065 * 11.2 Picture prediction data. picture_prediction()
1067 static int dirac_unpack_prediction_parameters(DiracContext *s)
1069 static const uint8_t default_blen[] = { 4, 12, 16, 24 };
1071 GetBitContext *gb = &s->gb;
1075 /* [DIRAC_STD] 11.2.2 Block parameters. block_parameters() */
1076 /* Luma and Chroma are equal. 11.2.3 */
1077 idx = get_interleaved_ue_golomb(gb); /* [DIRAC_STD] index */
1080 av_log(s->avctx, AV_LOG_ERROR, "Block prediction index too high\n");
1081 return AVERROR_INVALIDDATA;
1085 s->plane[0].xblen = get_interleaved_ue_golomb(gb);
1086 s->plane[0].yblen = get_interleaved_ue_golomb(gb);
1087 s->plane[0].xbsep = get_interleaved_ue_golomb(gb);
1088 s->plane[0].ybsep = get_interleaved_ue_golomb(gb);
1090 /*[DIRAC_STD] preset_block_params(index). Table 11.1 */
1091 s->plane[0].xblen = default_blen[idx-1];
1092 s->plane[0].yblen = default_blen[idx-1];
1093 s->plane[0].xbsep = 4 * idx;
1094 s->plane[0].ybsep = 4 * idx;
1096 /*[DIRAC_STD] 11.2.4 motion_data_dimensions()
1097 Calculated in function dirac_unpack_block_motion_data */
1099 if (s->plane[0].xblen % (1 << s->chroma_x_shift) != 0 ||
1100 s->plane[0].yblen % (1 << s->chroma_y_shift) != 0 ||
1101 !s->plane[0].xblen || !s->plane[0].yblen) {
1102 av_log(s->avctx, AV_LOG_ERROR,
1103 "invalid x/y block length (%d/%d) for x/y chroma shift (%d/%d)\n",
1104 s->plane[0].xblen, s->plane[0].yblen, s->chroma_x_shift, s->chroma_y_shift);
1105 return AVERROR_INVALIDDATA;
1107 if (!s->plane[0].xbsep || !s->plane[0].ybsep || s->plane[0].xbsep < s->plane[0].xblen/2 || s->plane[0].ybsep < s->plane[0].yblen/2) {
1108 av_log(s->avctx, AV_LOG_ERROR, "Block separation too small\n");
1109 return AVERROR_INVALIDDATA;
1111 if (s->plane[0].xbsep > s->plane[0].xblen || s->plane[0].ybsep > s->plane[0].yblen) {
1112 av_log(s->avctx, AV_LOG_ERROR, "Block separation greater than size\n");
1113 return AVERROR_INVALIDDATA;
1115 if (FFMAX(s->plane[0].xblen, s->plane[0].yblen) > MAX_BLOCKSIZE) {
1116 av_log(s->avctx, AV_LOG_ERROR, "Unsupported large block size\n");
1117 return AVERROR_PATCHWELCOME;
1120 /*[DIRAC_STD] 11.2.5 Motion vector precision. motion_vector_precision()
1121 Read motion vector precision */
1122 s->mv_precision = get_interleaved_ue_golomb(gb);
1123 if (s->mv_precision > 3) {
1124 av_log(s->avctx, AV_LOG_ERROR, "MV precision finer than eighth-pel\n");
1125 return AVERROR_INVALIDDATA;
1128 /*[DIRAC_STD] 11.2.6 Global motion. global_motion()
1129 Read the global motion compensation parameters */
1130 s->globalmc_flag = get_bits1(gb);
1131 if (s->globalmc_flag) {
1132 memset(s->globalmc, 0, sizeof(s->globalmc));
1133 /* [DIRAC_STD] pan_tilt(gparams) */
1134 for (ref = 0; ref < s->num_refs; ref++) {
1135 if (get_bits1(gb)) {
1136 s->globalmc[ref].pan_tilt[0] = dirac_get_se_golomb(gb);
1137 s->globalmc[ref].pan_tilt[1] = dirac_get_se_golomb(gb);
1139 /* [DIRAC_STD] zoom_rotate_shear(gparams)
1140 zoom/rotation/shear parameters */
1141 if (get_bits1(gb)) {
1142 s->globalmc[ref].zrs_exp = get_interleaved_ue_golomb(gb);
1143 s->globalmc[ref].zrs[0][0] = dirac_get_se_golomb(gb);
1144 s->globalmc[ref].zrs[0][1] = dirac_get_se_golomb(gb);
1145 s->globalmc[ref].zrs[1][0] = dirac_get_se_golomb(gb);
1146 s->globalmc[ref].zrs[1][1] = dirac_get_se_golomb(gb);
1148 s->globalmc[ref].zrs[0][0] = 1;
1149 s->globalmc[ref].zrs[1][1] = 1;
1151 /* [DIRAC_STD] perspective(gparams) */
1152 if (get_bits1(gb)) {
1153 s->globalmc[ref].perspective_exp = get_interleaved_ue_golomb(gb);
1154 s->globalmc[ref].perspective[0] = dirac_get_se_golomb(gb);
1155 s->globalmc[ref].perspective[1] = dirac_get_se_golomb(gb);
1160 /*[DIRAC_STD] 11.2.7 Picture prediction mode. prediction_mode()
1161 Picture prediction mode, not currently used. */
1162 if (get_interleaved_ue_golomb(gb)) {
1163 av_log(s->avctx, AV_LOG_ERROR, "Unknown picture prediction mode\n");
1164 return AVERROR_INVALIDDATA;
1167 /* [DIRAC_STD] 11.2.8 Reference picture weight. reference_picture_weights()
1168 just data read, weight calculation will be done later on. */
1169 s->weight_log2denom = 1;
1173 if (get_bits1(gb)) {
1174 s->weight_log2denom = get_interleaved_ue_golomb(gb);
1175 s->weight[0] = dirac_get_se_golomb(gb);
1176 if (s->num_refs == 2)
1177 s->weight[1] = dirac_get_se_golomb(gb);
1183 * Dirac Specification ->
1184 * 11.3 Wavelet transform data. wavelet_transform()
1186 static int dirac_unpack_idwt_params(DiracContext *s)
1188 GetBitContext *gb = &s->gb;
1192 #define CHECKEDREAD(dst, cond, errmsg) \
1193 tmp = get_interleaved_ue_golomb(gb); \
1195 av_log(s->avctx, AV_LOG_ERROR, errmsg); \
1196 return AVERROR_INVALIDDATA; \
1202 s->zero_res = s->num_refs ? get_bits1(gb) : 0;
1206 /*[DIRAC_STD] 11.3.1 Transform parameters. transform_parameters() */
1207 CHECKEDREAD(s->wavelet_idx, tmp > 6, "wavelet_idx is too big\n")
1209 CHECKEDREAD(s->wavelet_depth, tmp > MAX_DWT_LEVELS || tmp < 1, "invalid number of DWT decompositions\n")
1211 if (!s->low_delay) {
1212 /* Codeblock parameters (core syntax only) */
1213 if (get_bits1(gb)) {
1214 for (i = 0; i <= s->wavelet_depth; i++) {
1215 CHECKEDREAD(s->codeblock[i].width , tmp < 1 || tmp > (s->avctx->width >>s->wavelet_depth-i), "codeblock width invalid\n")
1216 CHECKEDREAD(s->codeblock[i].height, tmp < 1 || tmp > (s->avctx->height>>s->wavelet_depth-i), "codeblock height invalid\n")
1219 CHECKEDREAD(s->codeblock_mode, tmp > 1, "unknown codeblock mode\n")
1222 for (i = 0; i <= s->wavelet_depth; i++)
1223 s->codeblock[i].width = s->codeblock[i].height = 1;
1227 s->num_x = get_interleaved_ue_golomb(gb);
1228 s->num_y = get_interleaved_ue_golomb(gb);
1229 if (s->num_x * s->num_y == 0 || s->num_x * (uint64_t)s->num_y > INT_MAX) {
1230 av_log(s->avctx,AV_LOG_ERROR,"Invalid numx/y\n");
1231 s->num_x = s->num_y = 0;
1232 return AVERROR_INVALIDDATA;
1234 if (s->ld_picture) {
1235 s->lowdelay.bytes.num = get_interleaved_ue_golomb(gb);
1236 s->lowdelay.bytes.den = get_interleaved_ue_golomb(gb);
1237 if (s->lowdelay.bytes.den <= 0) {
1238 av_log(s->avctx,AV_LOG_ERROR,"Invalid lowdelay.bytes.den\n");
1239 return AVERROR_INVALIDDATA;
1241 } else if (s->hq_picture) {
1242 s->highquality.prefix_bytes = get_interleaved_ue_golomb(gb);
1243 s->highquality.size_scaler = get_interleaved_ue_golomb(gb);
1244 if (s->highquality.prefix_bytes >= INT_MAX / 8) {
1245 av_log(s->avctx,AV_LOG_ERROR,"too many prefix bytes\n");
1246 return AVERROR_INVALIDDATA;
1250 /* [DIRAC_STD] 11.3.5 Quantisation matrices (low-delay syntax). quant_matrix() */
1251 if (get_bits1(gb)) {
1252 av_log(s->avctx,AV_LOG_DEBUG,"Low Delay: Has Custom Quantization Matrix!\n");
1253 /* custom quantization matrix */
1254 s->lowdelay.quant[0][0] = get_interleaved_ue_golomb(gb);
1255 for (level = 0; level < s->wavelet_depth; level++) {
1256 s->lowdelay.quant[level][1] = get_interleaved_ue_golomb(gb);
1257 s->lowdelay.quant[level][2] = get_interleaved_ue_golomb(gb);
1258 s->lowdelay.quant[level][3] = get_interleaved_ue_golomb(gb);
1261 if (s->wavelet_depth > 4) {
1262 av_log(s->avctx,AV_LOG_ERROR,"Mandatory custom low delay matrix missing for depth %d\n", s->wavelet_depth);
1263 return AVERROR_INVALIDDATA;
1265 /* default quantization matrix */
1266 for (level = 0; level < s->wavelet_depth; level++)
1267 for (i = 0; i < 4; i++) {
1268 s->lowdelay.quant[level][i] = ff_dirac_default_qmat[s->wavelet_idx][level][i];
1269 /* haar with no shift differs for different depths */
1270 if (s->wavelet_idx == 3)
1271 s->lowdelay.quant[level][i] += 4*(s->wavelet_depth-1 - level);
1278 static inline int pred_sbsplit(uint8_t *sbsplit, int stride, int x, int y)
1280 static const uint8_t avgsplit[7] = { 0, 0, 1, 1, 1, 2, 2 };
1287 return sbsplit[-stride];
1289 return avgsplit[sbsplit[-1] + sbsplit[-stride] + sbsplit[-stride-1]];
1292 static inline int pred_block_mode(DiracBlock *block, int stride, int x, int y, int refmask)
1299 return block[-1].ref & refmask;
1301 return block[-stride].ref & refmask;
1303 /* return the majority */
1304 pred = (block[-1].ref & refmask) + (block[-stride].ref & refmask) + (block[-stride-1].ref & refmask);
1305 return (pred >> 1) & refmask;
1308 static inline void pred_block_dc(DiracBlock *block, int stride, int x, int y)
1312 memset(block->u.dc, 0, sizeof(block->u.dc));
1314 if (x && !(block[-1].ref & 3)) {
1315 for (i = 0; i < 3; i++)
1316 block->u.dc[i] += block[-1].u.dc[i];
1320 if (y && !(block[-stride].ref & 3)) {
1321 for (i = 0; i < 3; i++)
1322 block->u.dc[i] += block[-stride].u.dc[i];
1326 if (x && y && !(block[-1-stride].ref & 3)) {
1327 for (i = 0; i < 3; i++)
1328 block->u.dc[i] += block[-1-stride].u.dc[i];
1333 for (i = 0; i < 3; i++)
1334 block->u.dc[i] = (block->u.dc[i]+1)>>1;
1335 } else if (n == 3) {
1336 for (i = 0; i < 3; i++)
1337 block->u.dc[i] = divide3(block->u.dc[i]);
1341 static inline void pred_mv(DiracBlock *block, int stride, int x, int y, int ref)
1344 int refmask = ref+1;
1345 int mask = refmask | DIRAC_REF_MASK_GLOBAL; /* exclude gmc blocks */
1348 if (x && (block[-1].ref & mask) == refmask)
1349 pred[n++] = block[-1].u.mv[ref];
1351 if (y && (block[-stride].ref & mask) == refmask)
1352 pred[n++] = block[-stride].u.mv[ref];
1354 if (x && y && (block[-stride-1].ref & mask) == refmask)
1355 pred[n++] = block[-stride-1].u.mv[ref];
1359 block->u.mv[ref][0] = 0;
1360 block->u.mv[ref][1] = 0;
1363 block->u.mv[ref][0] = pred[0][0];
1364 block->u.mv[ref][1] = pred[0][1];
1367 block->u.mv[ref][0] = (pred[0][0] + pred[1][0] + 1) >> 1;
1368 block->u.mv[ref][1] = (pred[0][1] + pred[1][1] + 1) >> 1;
1371 block->u.mv[ref][0] = mid_pred(pred[0][0], pred[1][0], pred[2][0]);
1372 block->u.mv[ref][1] = mid_pred(pred[0][1], pred[1][1], pred[2][1]);
1377 static void global_mv(DiracContext *s, DiracBlock *block, int x, int y, int ref)
1379 int ez = s->globalmc[ref].zrs_exp;
1380 int ep = s->globalmc[ref].perspective_exp;
1381 int (*A)[2] = s->globalmc[ref].zrs;
1382 int *b = s->globalmc[ref].pan_tilt;
1383 int *c = s->globalmc[ref].perspective;
1385 int m = (1<<ep) - (c[0]*x + c[1]*y);
1386 int mx = m * ((A[0][0] * x + A[0][1]*y) + (1<<ez) * b[0]);
1387 int my = m * ((A[1][0] * x + A[1][1]*y) + (1<<ez) * b[1]);
1389 block->u.mv[ref][0] = (mx + (1<<(ez+ep))) >> (ez+ep);
1390 block->u.mv[ref][1] = (my + (1<<(ez+ep))) >> (ez+ep);
1393 static void decode_block_params(DiracContext *s, DiracArith arith[8], DiracBlock *block,
1394 int stride, int x, int y)
1398 block->ref = pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF1);
1399 block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF1);
1401 if (s->num_refs == 2) {
1402 block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF2);
1403 block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF2) << 1;
1407 pred_block_dc(block, stride, x, y);
1408 for (i = 0; i < 3; i++)
1409 block->u.dc[i] += dirac_get_arith_int(arith+1+i, CTX_DC_F1, CTX_DC_DATA);
1413 if (s->globalmc_flag) {
1414 block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_GLOBAL);
1415 block->ref ^= dirac_get_arith_bit(arith, CTX_GLOBAL_BLOCK) << 2;
1418 for (i = 0; i < s->num_refs; i++)
1419 if (block->ref & (i+1)) {
1420 if (block->ref & DIRAC_REF_MASK_GLOBAL) {
1421 global_mv(s, block, x, y, i);
1423 pred_mv(block, stride, x, y, i);
1424 block->u.mv[i][0] += dirac_get_arith_int(arith + 4 + 2 * i, CTX_MV_F1, CTX_MV_DATA);
1425 block->u.mv[i][1] += dirac_get_arith_int(arith + 5 + 2 * i, CTX_MV_F1, CTX_MV_DATA);
1431 * Copies the current block to the other blocks covered by the current superblock split mode
1433 static void propagate_block_data(DiracBlock *block, int stride, int size)
1436 DiracBlock *dst = block;
1438 for (x = 1; x < size; x++)
1441 for (y = 1; y < size; y++) {
1443 for (x = 0; x < size; x++)
1449 * Dirac Specification ->
1450 * 12. Block motion data syntax
1452 static int dirac_unpack_block_motion_data(DiracContext *s)
1454 GetBitContext *gb = &s->gb;
1455 uint8_t *sbsplit = s->sbsplit;
1457 DiracArith arith[8];
1461 /* [DIRAC_STD] 11.2.4 and 12.2.1 Number of blocks and superblocks */
1462 s->sbwidth = DIVRNDUP(s->seq.width, 4*s->plane[0].xbsep);
1463 s->sbheight = DIVRNDUP(s->seq.height, 4*s->plane[0].ybsep);
1464 s->blwidth = 4 * s->sbwidth;
1465 s->blheight = 4 * s->sbheight;
1467 /* [DIRAC_STD] 12.3.1 Superblock splitting modes. superblock_split_modes()
1468 decode superblock split modes */
1469 ff_dirac_init_arith_decoder(arith, gb, get_interleaved_ue_golomb(gb)); /* get_interleaved_ue_golomb(gb) is the length */
1470 for (y = 0; y < s->sbheight; y++) {
1471 for (x = 0; x < s->sbwidth; x++) {
1472 unsigned int split = dirac_get_arith_uint(arith, CTX_SB_F1, CTX_SB_DATA);
1474 return AVERROR_INVALIDDATA;
1475 sbsplit[x] = (split + pred_sbsplit(sbsplit+x, s->sbwidth, x, y)) % 3;
1477 sbsplit += s->sbwidth;
1480 /* setup arith decoding */
1481 ff_dirac_init_arith_decoder(arith, gb, get_interleaved_ue_golomb(gb));
1482 for (i = 0; i < s->num_refs; i++) {
1483 ff_dirac_init_arith_decoder(arith + 4 + 2 * i, gb, get_interleaved_ue_golomb(gb));
1484 ff_dirac_init_arith_decoder(arith + 5 + 2 * i, gb, get_interleaved_ue_golomb(gb));
1486 for (i = 0; i < 3; i++)
1487 ff_dirac_init_arith_decoder(arith+1+i, gb, get_interleaved_ue_golomb(gb));
1489 for (y = 0; y < s->sbheight; y++)
1490 for (x = 0; x < s->sbwidth; x++) {
1491 int blkcnt = 1 << s->sbsplit[y * s->sbwidth + x];
1492 int step = 4 >> s->sbsplit[y * s->sbwidth + x];
1494 for (q = 0; q < blkcnt; q++)
1495 for (p = 0; p < blkcnt; p++) {
1496 int bx = 4 * x + p*step;
1497 int by = 4 * y + q*step;
1498 DiracBlock *block = &s->blmotion[by*s->blwidth + bx];
1499 decode_block_params(s, arith, block, s->blwidth, bx, by);
1500 propagate_block_data(block, s->blwidth, step);
1507 static int weight(int i, int blen, int offset)
1509 #define ROLLOFF(i) offset == 1 ? ((i) ? 5 : 3) : \
1510 (1 + (6*(i) + offset - 1) / (2*offset - 1))
1514 else if (i > blen-1 - 2*offset)
1515 return ROLLOFF(blen-1 - i);
1519 static void init_obmc_weight_row(Plane *p, uint8_t *obmc_weight, int stride,
1520 int left, int right, int wy)
1523 for (x = 0; left && x < p->xblen >> 1; x++)
1524 obmc_weight[x] = wy*8;
1525 for (; x < p->xblen >> right; x++)
1526 obmc_weight[x] = wy*weight(x, p->xblen, p->xoffset);
1527 for (; x < p->xblen; x++)
1528 obmc_weight[x] = wy*8;
1529 for (; x < stride; x++)
1533 static void init_obmc_weight(Plane *p, uint8_t *obmc_weight, int stride,
1534 int left, int right, int top, int bottom)
1537 for (y = 0; top && y < p->yblen >> 1; y++) {
1538 init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);
1539 obmc_weight += stride;
1541 for (; y < p->yblen >> bottom; y++) {
1542 int wy = weight(y, p->yblen, p->yoffset);
1543 init_obmc_weight_row(p, obmc_weight, stride, left, right, wy);
1544 obmc_weight += stride;
1546 for (; y < p->yblen; y++) {
1547 init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);
1548 obmc_weight += stride;
1552 static void init_obmc_weights(DiracContext *s, Plane *p, int by)
1555 int bottom = by == s->blheight-1;
1557 /* don't bother re-initing for rows 2 to blheight-2, the weights don't change */
1558 if (top || bottom || by == 1) {
1559 init_obmc_weight(p, s->obmc_weight[0], MAX_BLOCKSIZE, 1, 0, top, bottom);
1560 init_obmc_weight(p, s->obmc_weight[1], MAX_BLOCKSIZE, 0, 0, top, bottom);
1561 init_obmc_weight(p, s->obmc_weight[2], MAX_BLOCKSIZE, 0, 1, top, bottom);
1565 static const uint8_t epel_weights[4][4][4] = {
1585 * For block x,y, determine which of the hpel planes to do bilinear
1586 * interpolation from and set src[] to the location in each hpel plane
1589 * @return the index of the put_dirac_pixels_tab function to use
1590 * 0 for 1 plane (fpel,hpel), 1 for 2 planes (qpel), 2 for 4 planes (qpel), and 3 for epel
1592 static int mc_subpel(DiracContext *s, DiracBlock *block, const uint8_t *src[5],
1593 int x, int y, int ref, int plane)
1595 Plane *p = &s->plane[plane];
1596 uint8_t **ref_hpel = s->ref_pics[ref]->hpel[plane];
1597 int motion_x = block->u.mv[ref][0];
1598 int motion_y = block->u.mv[ref][1];
1599 int mx, my, i, epel, nplanes = 0;
1602 motion_x >>= s->chroma_x_shift;
1603 motion_y >>= s->chroma_y_shift;
1606 mx = motion_x & ~(-1U << s->mv_precision);
1607 my = motion_y & ~(-1U << s->mv_precision);
1608 motion_x >>= s->mv_precision;
1609 motion_y >>= s->mv_precision;
1610 /* normalize subpel coordinates to epel */
1611 /* TODO: template this function? */
1612 mx <<= 3 - s->mv_precision;
1613 my <<= 3 - s->mv_precision;
1622 src[0] = ref_hpel[(my>>1)+(mx>>2)] + y*p->stride + x;
1626 for (i = 0; i < 4; i++)
1627 src[i] = ref_hpel[i] + y*p->stride + x;
1629 /* if we're interpolating in the right/bottom halves, adjust the planes as needed
1630 we increment x/y because the edge changes for half of the pixels */
1637 src[0] += p->stride;
1638 src[1] += p->stride;
1646 /* check if we really only need 2 planes since either mx or my is
1647 a hpel position. (epel weights of 0 handle this there) */
1649 /* mx == 0: average [0] and [2]
1650 mx == 4: average [1] and [3] */
1651 src[!mx] = src[2 + !!mx];
1653 } else if (!(my&3)) {
1654 src[0] = src[(my>>1) ];
1655 src[1] = src[(my>>1)+1];
1659 /* adjust the ordering if needed so the weights work */
1661 FFSWAP(const uint8_t *, src[0], src[1]);
1662 FFSWAP(const uint8_t *, src[2], src[3]);
1665 FFSWAP(const uint8_t *, src[0], src[2]);
1666 FFSWAP(const uint8_t *, src[1], src[3]);
1668 src[4] = epel_weights[my&3][mx&3];
1672 /* fixme: v/h _edge_pos */
1673 if (x + p->xblen > p->width +EDGE_WIDTH/2 ||
1674 y + p->yblen > p->height+EDGE_WIDTH/2 ||
1676 for (i = 0; i < nplanes; i++) {
1677 s->vdsp.emulated_edge_mc(s->edge_emu_buffer[i], src[i],
1678 p->stride, p->stride,
1679 p->xblen, p->yblen, x, y,
1680 p->width+EDGE_WIDTH/2, p->height+EDGE_WIDTH/2);
1681 src[i] = s->edge_emu_buffer[i];
1684 return (nplanes>>1) + epel;
1687 static void add_dc(uint16_t *dst, int dc, int stride,
1688 uint8_t *obmc_weight, int xblen, int yblen)
1693 for (y = 0; y < yblen; y++) {
1694 for (x = 0; x < xblen; x += 2) {
1695 dst[x ] += dc * obmc_weight[x ];
1696 dst[x+1] += dc * obmc_weight[x+1];
1699 obmc_weight += MAX_BLOCKSIZE;
1703 static void block_mc(DiracContext *s, DiracBlock *block,
1704 uint16_t *mctmp, uint8_t *obmc_weight,
1705 int plane, int dstx, int dsty)
1707 Plane *p = &s->plane[plane];
1708 const uint8_t *src[5];
1711 switch (block->ref&3) {
1713 add_dc(mctmp, block->u.dc[plane], p->stride, obmc_weight, p->xblen, p->yblen);
1717 idx = mc_subpel(s, block, src, dstx, dsty, (block->ref&3)-1, plane);
1718 s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1720 s->weight_func(s->mcscratch, p->stride, s->weight_log2denom,
1721 s->weight[0] + s->weight[1], p->yblen);
1724 idx = mc_subpel(s, block, src, dstx, dsty, 0, plane);
1725 s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1726 idx = mc_subpel(s, block, src, dstx, dsty, 1, plane);
1727 if (s->biweight_func) {
1728 /* fixme: +32 is a quick hack */
1729 s->put_pixels_tab[idx](s->mcscratch + 32, src, p->stride, p->yblen);
1730 s->biweight_func(s->mcscratch, s->mcscratch+32, p->stride, s->weight_log2denom,
1731 s->weight[0], s->weight[1], p->yblen);
1733 s->avg_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1736 s->add_obmc(mctmp, s->mcscratch, p->stride, obmc_weight, p->yblen);
1739 static void mc_row(DiracContext *s, DiracBlock *block, uint16_t *mctmp, int plane, int dsty)
1741 Plane *p = &s->plane[plane];
1742 int x, dstx = p->xbsep - p->xoffset;
1744 block_mc(s, block, mctmp, s->obmc_weight[0], plane, -p->xoffset, dsty);
1747 for (x = 1; x < s->blwidth-1; x++) {
1748 block_mc(s, block+x, mctmp, s->obmc_weight[1], plane, dstx, dsty);
1752 block_mc(s, block+x, mctmp, s->obmc_weight[2], plane, dstx, dsty);
1755 static void select_dsp_funcs(DiracContext *s, int width, int height, int xblen, int yblen)
1763 memcpy(s->put_pixels_tab, s->diracdsp.put_dirac_pixels_tab[idx], sizeof(s->put_pixels_tab));
1764 memcpy(s->avg_pixels_tab, s->diracdsp.avg_dirac_pixels_tab[idx], sizeof(s->avg_pixels_tab));
1765 s->add_obmc = s->diracdsp.add_dirac_obmc[idx];
1766 if (s->weight_log2denom > 1 || s->weight[0] != 1 || s->weight[1] != 1) {
1767 s->weight_func = s->diracdsp.weight_dirac_pixels_tab[idx];
1768 s->biweight_func = s->diracdsp.biweight_dirac_pixels_tab[idx];
1770 s->weight_func = NULL;
1771 s->biweight_func = NULL;
1775 static int interpolate_refplane(DiracContext *s, DiracFrame *ref, int plane, int width, int height)
1777 /* chroma allocates an edge of 8 when subsampled
1778 which for 4:2:2 means an h edge of 16 and v edge of 8
1779 just use 8 for everything for the moment */
1780 int i, edge = EDGE_WIDTH/2;
1782 ref->hpel[plane][0] = ref->avframe->data[plane];
1783 s->mpvencdsp.draw_edges(ref->hpel[plane][0], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM); /* EDGE_TOP | EDGE_BOTTOM values just copied to make it build, this needs to be ensured */
1785 /* no need for hpel if we only have fpel vectors */
1786 if (!s->mv_precision)
1789 for (i = 1; i < 4; i++) {
1790 if (!ref->hpel_base[plane][i])
1791 ref->hpel_base[plane][i] = av_malloc((height+2*edge) * ref->avframe->linesize[plane] + 32);
1792 if (!ref->hpel_base[plane][i]) {
1793 return AVERROR(ENOMEM);
1795 /* we need to be 16-byte aligned even for chroma */
1796 ref->hpel[plane][i] = ref->hpel_base[plane][i] + edge*ref->avframe->linesize[plane] + 16;
1799 if (!ref->interpolated[plane]) {
1800 s->diracdsp.dirac_hpel_filter(ref->hpel[plane][1], ref->hpel[plane][2],
1801 ref->hpel[plane][3], ref->hpel[plane][0],
1802 ref->avframe->linesize[plane], width, height);
1803 s->mpvencdsp.draw_edges(ref->hpel[plane][1], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1804 s->mpvencdsp.draw_edges(ref->hpel[plane][2], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1805 s->mpvencdsp.draw_edges(ref->hpel[plane][3], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1807 ref->interpolated[plane] = 1;
1813 * Dirac Specification ->
1814 * 13.0 Transform data syntax. transform_data()
1816 static int dirac_decode_frame_internal(DiracContext *s)
1819 int y, i, comp, dsty;
1823 /* [DIRAC_STD] 13.5.1 low_delay_transform_data() */
1824 if (!s->hq_picture) {
1825 for (comp = 0; comp < 3; comp++) {
1826 Plane *p = &s->plane[comp];
1827 memset(p->idwt.buf, 0, p->idwt.stride * p->idwt.height);
1831 if ((ret = decode_lowdelay(s)) < 0)
1836 for (comp = 0; comp < 3; comp++) {
1837 Plane *p = &s->plane[comp];
1838 uint8_t *frame = s->current_picture->avframe->data[comp];
1840 /* FIXME: small resolutions */
1841 for (i = 0; i < 4; i++)
1842 s->edge_emu_buffer[i] = s->edge_emu_buffer_base + i*FFALIGN(p->width, 16);
1844 if (!s->zero_res && !s->low_delay)
1846 memset(p->idwt.buf, 0, p->idwt.stride * p->idwt.height);
1847 decode_component(s, comp); /* [DIRAC_STD] 13.4.1 core_transform_data() */
1849 ret = ff_spatial_idwt_init(&d, &p->idwt, s->wavelet_idx+2,
1850 s->wavelet_depth, s->bit_depth);
1854 if (!s->num_refs) { /* intra */
1855 for (y = 0; y < p->height; y += 16) {
1856 int idx = (s->bit_depth - 8) >> 1;
1857 ff_spatial_idwt_slice2(&d, y+16); /* decode */
1858 s->diracdsp.put_signed_rect_clamped[idx](frame + y*p->stride,
1860 p->idwt.buf + y*p->idwt.stride,
1861 p->idwt.stride, p->width, 16);
1863 } else { /* inter */
1864 int rowheight = p->ybsep*p->stride;
1866 select_dsp_funcs(s, p->width, p->height, p->xblen, p->yblen);
1868 for (i = 0; i < s->num_refs; i++) {
1869 int ret = interpolate_refplane(s, s->ref_pics[i], comp, p->width, p->height);
1874 memset(s->mctmp, 0, 4*p->yoffset*p->stride);
1877 for (y = 0; y < s->blheight; y++) {
1879 start = FFMAX(dsty, 0);
1880 uint16_t *mctmp = s->mctmp + y*rowheight;
1881 DiracBlock *blocks = s->blmotion + y*s->blwidth;
1883 init_obmc_weights(s, p, y);
1885 if (y == s->blheight-1 || start+p->ybsep > p->height)
1886 h = p->height - start;
1888 h = p->ybsep - (start - dsty);
1892 memset(mctmp+2*p->yoffset*p->stride, 0, 2*rowheight);
1893 mc_row(s, blocks, mctmp, comp, dsty);
1895 mctmp += (start - dsty)*p->stride + p->xoffset;
1896 ff_spatial_idwt_slice2(&d, start + h); /* decode */
1897 /* NOTE: add_rect_clamped hasn't been templated hence the shifts.
1898 * idwt.stride is passed as pixels, not in bytes as in the rest of the decoder */
1899 s->diracdsp.add_rect_clamped(frame + start*p->stride, mctmp, p->stride,
1900 (int16_t*)(p->idwt.buf) + start*(p->idwt.stride >> 1), (p->idwt.stride >> 1), p->width, h);
1911 static int get_buffer_with_edge(AVCodecContext *avctx, AVFrame *f, int flags)
1914 int chroma_x_shift, chroma_y_shift;
1915 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_x_shift, &chroma_y_shift);
1917 f->width = avctx->width + 2 * EDGE_WIDTH;
1918 f->height = avctx->height + 2 * EDGE_WIDTH + 2;
1919 ret = ff_get_buffer(avctx, f, flags);
1923 for (i = 0; f->data[i]; i++) {
1924 int offset = (EDGE_WIDTH >> (i && i<3 ? chroma_y_shift : 0)) *
1925 f->linesize[i] + 32;
1926 f->data[i] += offset;
1928 f->width = avctx->width;
1929 f->height = avctx->height;
1935 * Dirac Specification ->
1936 * 11.1.1 Picture Header. picture_header()
1938 static int dirac_decode_picture_header(DiracContext *s)
1940 unsigned retire, picnum;
1942 int64_t refdist, refnum;
1943 GetBitContext *gb = &s->gb;
1945 /* [DIRAC_STD] 11.1.1 Picture Header. picture_header() PICTURE_NUM */
1946 picnum = s->current_picture->avframe->display_picture_number = get_bits_long(gb, 32);
1949 av_log(s->avctx,AV_LOG_DEBUG,"PICTURE_NUM: %d\n",picnum);
1951 /* if this is the first keyframe after a sequence header, start our
1952 reordering from here */
1953 if (s->frame_number < 0)
1954 s->frame_number = picnum;
1956 s->ref_pics[0] = s->ref_pics[1] = NULL;
1957 for (i = 0; i < s->num_refs; i++) {
1958 refnum = (picnum + dirac_get_se_golomb(gb)) & 0xFFFFFFFF;
1959 refdist = INT64_MAX;
1961 /* find the closest reference to the one we want */
1962 /* Jordi: this is needed if the referenced picture hasn't yet arrived */
1963 for (j = 0; j < MAX_REFERENCE_FRAMES && refdist; j++)
1964 if (s->ref_frames[j]
1965 && FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum) < refdist) {
1966 s->ref_pics[i] = s->ref_frames[j];
1967 refdist = FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum);
1970 if (!s->ref_pics[i] || refdist)
1971 av_log(s->avctx, AV_LOG_DEBUG, "Reference not found\n");
1973 /* if there were no references at all, allocate one */
1974 if (!s->ref_pics[i])
1975 for (j = 0; j < MAX_FRAMES; j++)
1976 if (!s->all_frames[j].avframe->data[0]) {
1977 s->ref_pics[i] = &s->all_frames[j];
1978 ret = get_buffer_with_edge(s->avctx, s->ref_pics[i]->avframe, AV_GET_BUFFER_FLAG_REF);
1984 if (!s->ref_pics[i]) {
1985 av_log(s->avctx, AV_LOG_ERROR, "Reference could not be allocated\n");
1986 return AVERROR_INVALIDDATA;
1991 /* retire the reference frames that are not used anymore */
1992 if (s->current_picture->reference) {
1993 retire = (picnum + dirac_get_se_golomb(gb)) & 0xFFFFFFFF;
1994 if (retire != picnum) {
1995 DiracFrame *retire_pic = remove_frame(s->ref_frames, retire);
1998 retire_pic->reference &= DELAYED_PIC_REF;
2000 av_log(s->avctx, AV_LOG_DEBUG, "Frame to retire not found\n");
2003 /* if reference array is full, remove the oldest as per the spec */
2004 while (add_frame(s->ref_frames, MAX_REFERENCE_FRAMES, s->current_picture)) {
2005 av_log(s->avctx, AV_LOG_ERROR, "Reference frame overflow\n");
2006 remove_frame(s->ref_frames, s->ref_frames[0]->avframe->display_picture_number)->reference &= DELAYED_PIC_REF;
2011 ret = dirac_unpack_prediction_parameters(s); /* [DIRAC_STD] 11.2 Picture Prediction Data. picture_prediction() */
2014 ret = dirac_unpack_block_motion_data(s); /* [DIRAC_STD] 12. Block motion data syntax */
2018 ret = dirac_unpack_idwt_params(s); /* [DIRAC_STD] 11.3 Wavelet transform data */
2026 static int get_delayed_pic(DiracContext *s, AVFrame *picture, int *got_frame)
2028 DiracFrame *out = s->delay_frames[0];
2032 /* find frame with lowest picture number */
2033 for (i = 1; s->delay_frames[i]; i++)
2034 if (s->delay_frames[i]->avframe->display_picture_number < out->avframe->display_picture_number) {
2035 out = s->delay_frames[i];
2039 for (i = out_idx; s->delay_frames[i]; i++)
2040 s->delay_frames[i] = s->delay_frames[i+1];
2043 out->reference ^= DELAYED_PIC_REF;
2045 if((ret = av_frame_ref(picture, out->avframe)) < 0)
2053 * Dirac Specification ->
2054 * 9.6 Parse Info Header Syntax. parse_info()
2055 * 4 byte start code + byte parse code + 4 byte size + 4 byte previous size
2057 #define DATA_UNIT_HEADER_SIZE 13
2059 /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3
2060 inside the function parse_sequence() */
2061 static int dirac_decode_data_unit(AVCodecContext *avctx, const uint8_t *buf, int size)
2063 DiracContext *s = avctx->priv_data;
2064 DiracFrame *pic = NULL;
2065 AVDiracSeqHeader *dsh;
2070 if (size < DATA_UNIT_HEADER_SIZE)
2071 return AVERROR_INVALIDDATA;
2073 parse_code = buf[4];
2075 init_get_bits(&s->gb, &buf[13], 8*(size - DATA_UNIT_HEADER_SIZE));
2077 if (parse_code == DIRAC_PCODE_SEQ_HEADER) {
2078 if (s->seen_sequence_header)
2081 /* [DIRAC_STD] 10. Sequence header */
2082 ret = av_dirac_parse_sequence_header(&dsh, buf + DATA_UNIT_HEADER_SIZE, size - DATA_UNIT_HEADER_SIZE, avctx);
2084 av_log(avctx, AV_LOG_ERROR, "error parsing sequence header");
2088 ret = ff_set_dimensions(avctx, dsh->width, dsh->height);
2094 ff_set_sar(avctx, dsh->sample_aspect_ratio);
2095 avctx->pix_fmt = dsh->pix_fmt;
2096 avctx->color_range = dsh->color_range;
2097 avctx->color_trc = dsh->color_trc;
2098 avctx->color_primaries = dsh->color_primaries;
2099 avctx->colorspace = dsh->colorspace;
2100 avctx->profile = dsh->profile;
2101 avctx->level = dsh->level;
2102 avctx->framerate = dsh->framerate;
2103 s->bit_depth = dsh->bit_depth;
2104 s->version.major = dsh->version.major;
2105 s->version.minor = dsh->version.minor;
2109 s->pshift = s->bit_depth > 8;
2111 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
2113 ret = alloc_sequence_buffers(s);
2117 s->seen_sequence_header = 1;
2118 } else if (parse_code == DIRAC_PCODE_END_SEQ) { /* [DIRAC_STD] End of Sequence */
2119 free_sequence_buffers(s);
2120 s->seen_sequence_header = 0;
2121 } else if (parse_code == DIRAC_PCODE_AUX) {
2122 if (buf[13] == 1) { /* encoder implementation/version */
2124 /* versions older than 1.0.8 don't store quant delta for
2125 subbands with only one codeblock */
2126 if (sscanf(buf+14, "Schroedinger %d.%d.%d", ver, ver+1, ver+2) == 3)
2127 if (ver[0] == 1 && ver[1] == 0 && ver[2] <= 7)
2128 s->old_delta_quant = 1;
2130 } else if (parse_code & 0x8) { /* picture data unit */
2131 if (!s->seen_sequence_header) {
2132 av_log(avctx, AV_LOG_DEBUG, "Dropping frame without sequence header\n");
2133 return AVERROR_INVALIDDATA;
2136 /* find an unused frame */
2137 for (i = 0; i < MAX_FRAMES; i++)
2138 if (s->all_frames[i].avframe->data[0] == NULL)
2139 pic = &s->all_frames[i];
2141 av_log(avctx, AV_LOG_ERROR, "framelist full\n");
2142 return AVERROR_INVALIDDATA;
2145 av_frame_unref(pic->avframe);
2147 /* [DIRAC_STD] Defined in 9.6.1 ... */
2148 tmp = parse_code & 0x03; /* [DIRAC_STD] num_refs() */
2150 av_log(avctx, AV_LOG_ERROR, "num_refs of 3\n");
2151 return AVERROR_INVALIDDATA;
2154 s->is_arith = (parse_code & 0x48) == 0x08; /* [DIRAC_STD] using_ac() */
2155 s->low_delay = (parse_code & 0x88) == 0x88; /* [DIRAC_STD] is_low_delay() */
2156 s->core_syntax = (parse_code & 0x88) == 0x08; /* [DIRAC_STD] is_core_syntax() */
2157 s->ld_picture = (parse_code & 0xF8) == 0xC8; /* [DIRAC_STD] is_ld_picture() */
2158 s->hq_picture = (parse_code & 0xF8) == 0xE8; /* [DIRAC_STD] is_hq_picture() */
2159 s->dc_prediction = (parse_code & 0x28) == 0x08; /* [DIRAC_STD] using_dc_prediction() */
2160 pic->reference = (parse_code & 0x0C) == 0x0C; /* [DIRAC_STD] is_reference() */
2161 pic->avframe->key_frame = s->num_refs == 0; /* [DIRAC_STD] is_intra() */
2162 pic->avframe->pict_type = s->num_refs + 1; /* Definition of AVPictureType in avutil.h */
2164 /* VC-2 Low Delay has a different parse code than the Dirac Low Delay */
2165 if (s->version.minor == 2 && parse_code == 0x88)
2168 if (s->low_delay && !(s->ld_picture || s->hq_picture) ) {
2169 av_log(avctx, AV_LOG_ERROR, "Invalid low delay flag\n");
2170 return AVERROR_INVALIDDATA;
2173 if ((ret = get_buffer_with_edge(avctx, pic->avframe, (parse_code & 0x0C) == 0x0C ? AV_GET_BUFFER_FLAG_REF : 0)) < 0)
2175 s->current_picture = pic;
2176 s->plane[0].stride = pic->avframe->linesize[0];
2177 s->plane[1].stride = pic->avframe->linesize[1];
2178 s->plane[2].stride = pic->avframe->linesize[2];
2180 if (alloc_buffers(s, FFMAX3(FFABS(s->plane[0].stride), FFABS(s->plane[1].stride), FFABS(s->plane[2].stride))) < 0)
2181 return AVERROR(ENOMEM);
2183 /* [DIRAC_STD] 11.1 Picture parse. picture_parse() */
2184 ret = dirac_decode_picture_header(s);
2188 /* [DIRAC_STD] 13.0 Transform data syntax. transform_data() */
2189 ret = dirac_decode_frame_internal(s);
2196 static int dirac_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPacket *pkt)
2198 DiracContext *s = avctx->priv_data;
2199 AVFrame *picture = data;
2200 uint8_t *buf = pkt->data;
2201 int buf_size = pkt->size;
2204 unsigned data_unit_size;
2206 /* release unused frames */
2207 for (i = 0; i < MAX_FRAMES; i++)
2208 if (s->all_frames[i].avframe->data[0] && !s->all_frames[i].reference) {
2209 av_frame_unref(s->all_frames[i].avframe);
2210 memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));
2213 s->current_picture = NULL;
2216 /* end of stream, so flush delayed pics */
2218 return get_delayed_pic(s, (AVFrame *)data, got_frame);
2221 /*[DIRAC_STD] Here starts the code from parse_info() defined in 9.6
2222 [DIRAC_STD] PARSE_INFO_PREFIX = "BBCD" as defined in ISO/IEC 646
2223 BBCD start code search */
2224 for (; buf_idx + DATA_UNIT_HEADER_SIZE < buf_size; buf_idx++) {
2225 if (buf[buf_idx ] == 'B' && buf[buf_idx+1] == 'B' &&
2226 buf[buf_idx+2] == 'C' && buf[buf_idx+3] == 'D')
2229 /* BBCD found or end of data */
2230 if (buf_idx + DATA_UNIT_HEADER_SIZE >= buf_size)
2233 data_unit_size = AV_RB32(buf+buf_idx+5);
2234 if (data_unit_size > buf_size - buf_idx || !data_unit_size) {
2235 if(data_unit_size > buf_size - buf_idx)
2236 av_log(s->avctx, AV_LOG_ERROR,
2237 "Data unit with size %d is larger than input buffer, discarding\n",
2242 /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3 inside the function parse_sequence() */
2243 ret = dirac_decode_data_unit(avctx, buf+buf_idx, data_unit_size);
2246 av_log(s->avctx, AV_LOG_ERROR,"Error in dirac_decode_data_unit\n");
2249 buf_idx += data_unit_size;
2252 if (!s->current_picture)
2255 if (s->current_picture->avframe->display_picture_number > s->frame_number) {
2256 DiracFrame *delayed_frame = remove_frame(s->delay_frames, s->frame_number);
2258 s->current_picture->reference |= DELAYED_PIC_REF;
2260 if (add_frame(s->delay_frames, MAX_DELAY, s->current_picture)) {
2261 int min_num = s->delay_frames[0]->avframe->display_picture_number;
2262 /* Too many delayed frames, so we display the frame with the lowest pts */
2263 av_log(avctx, AV_LOG_ERROR, "Delay frame overflow\n");
2265 for (i = 1; s->delay_frames[i]; i++)
2266 if (s->delay_frames[i]->avframe->display_picture_number < min_num)
2267 min_num = s->delay_frames[i]->avframe->display_picture_number;
2269 delayed_frame = remove_frame(s->delay_frames, min_num);
2270 add_frame(s->delay_frames, MAX_DELAY, s->current_picture);
2273 if (delayed_frame) {
2274 delayed_frame->reference ^= DELAYED_PIC_REF;
2275 if((ret=av_frame_ref(data, delayed_frame->avframe)) < 0)
2279 } else if (s->current_picture->avframe->display_picture_number == s->frame_number) {
2280 /* The right frame at the right time :-) */
2281 if((ret=av_frame_ref(data, s->current_picture->avframe)) < 0)
2287 s->frame_number = picture->display_picture_number + 1;
2292 AVCodec ff_dirac_decoder = {
2294 .long_name = NULL_IF_CONFIG_SMALL("BBC Dirac VC-2"),
2295 .type = AVMEDIA_TYPE_VIDEO,
2296 .id = AV_CODEC_ID_DIRAC,
2297 .priv_data_size = sizeof(DiracContext),
2298 .init = dirac_decode_init,
2299 .close = dirac_decode_end,
2300 .decode = dirac_decode_frame,
2301 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_DR1,
2302 .flush = dirac_decode_flush,