yading@10: /* yading@10: * Copyright (C) 2007 Marco Gerards yading@10: * Copyright (C) 2009 David Conrad yading@10: * Copyright (C) 2011 Jordi Ortiz yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: /** yading@10: * @file yading@10: * Dirac Decoder yading@10: * @author Marco Gerards , David Conrad, Jordi Ortiz yading@10: */ yading@10: yading@10: #include "avcodec.h" yading@10: #include "dsputil.h" yading@10: #include "get_bits.h" yading@10: #include "bytestream.h" yading@10: #include "internal.h" yading@10: #include "golomb.h" yading@10: #include "dirac_arith.h" yading@10: #include "mpeg12data.h" yading@10: #include "dirac_dwt.h" yading@10: #include "dirac.h" yading@10: #include "diracdsp.h" yading@10: #include "videodsp.h" // for ff_emulated_edge_mc_8 yading@10: yading@10: /** yading@10: * The spec limits the number of wavelet decompositions to 4 for both yading@10: * level 1 (VC-2) and 128 (long-gop default). yading@10: * 5 decompositions is the maximum before >16-bit buffers are needed. yading@10: * Schroedinger allows this for DD 9,7 and 13,7 wavelets only, limiting yading@10: * the others to 4 decompositions (or 3 for the fidelity filter). yading@10: * yading@10: * We use this instead of MAX_DECOMPOSITIONS to save some memory. yading@10: */ yading@10: #define MAX_DWT_LEVELS 5 yading@10: yading@10: /** yading@10: * The spec limits this to 3 for frame coding, but in practice can be as high as 6 yading@10: */ yading@10: #define MAX_REFERENCE_FRAMES 8 yading@10: #define MAX_DELAY 5 /* limit for main profile for frame coding (TODO: field coding) */ yading@10: #define MAX_FRAMES (MAX_REFERENCE_FRAMES + MAX_DELAY + 1) yading@10: #define MAX_QUANT 68 /* max quant for VC-2 */ yading@10: #define MAX_BLOCKSIZE 32 /* maximum xblen/yblen we support */ yading@10: yading@10: /** yading@10: * DiracBlock->ref flags, if set then the block does MC from the given ref yading@10: */ yading@10: #define DIRAC_REF_MASK_REF1 1 yading@10: #define DIRAC_REF_MASK_REF2 2 yading@10: #define DIRAC_REF_MASK_GLOBAL 4 yading@10: yading@10: /** yading@10: * Value of Picture.reference when Picture is not a reference picture, but yading@10: * is held for delayed output. yading@10: */ yading@10: #define DELAYED_PIC_REF 4 yading@10: yading@10: #define ff_emulated_edge_mc ff_emulated_edge_mc_8 /* Fix: change the calls to this function regarding bit depth */ yading@10: yading@10: #define CALC_PADDING(size, depth) \ yading@10: (((size + (1 << depth) - 1) >> depth) << depth) yading@10: yading@10: #define DIVRNDUP(a, b) (((a) + (b) - 1) / (b)) yading@10: yading@10: typedef struct { yading@10: AVFrame avframe; yading@10: int interpolated[3]; /* 1 if hpel[] is valid */ yading@10: uint8_t *hpel[3][4]; yading@10: uint8_t *hpel_base[3][4]; yading@10: } DiracFrame; yading@10: yading@10: typedef struct { yading@10: union { yading@10: int16_t mv[2][2]; yading@10: int16_t dc[3]; yading@10: } u; /* anonymous unions aren't in C99 :( */ yading@10: uint8_t ref; yading@10: } DiracBlock; yading@10: yading@10: typedef struct SubBand { yading@10: int level; yading@10: int orientation; yading@10: int stride; yading@10: int width; yading@10: int height; yading@10: int quant; yading@10: IDWTELEM *ibuf; yading@10: struct SubBand *parent; yading@10: yading@10: /* for low delay */ yading@10: unsigned length; yading@10: const uint8_t *coeff_data; yading@10: } SubBand; yading@10: yading@10: typedef struct Plane { yading@10: int width; yading@10: int height; yading@10: int stride; yading@10: yading@10: int idwt_width; yading@10: int idwt_height; yading@10: int idwt_stride; yading@10: IDWTELEM *idwt_buf; yading@10: IDWTELEM *idwt_buf_base; yading@10: IDWTELEM *idwt_tmp; yading@10: yading@10: /* block length */ yading@10: uint8_t xblen; yading@10: uint8_t yblen; yading@10: /* block separation (block n+1 starts after this many pixels in block n) */ yading@10: uint8_t xbsep; yading@10: uint8_t ybsep; yading@10: /* amount of overspill on each edge (half of the overlap between blocks) */ yading@10: uint8_t xoffset; yading@10: uint8_t yoffset; yading@10: yading@10: SubBand band[MAX_DWT_LEVELS][4]; yading@10: } Plane; yading@10: yading@10: typedef struct DiracContext { yading@10: AVCodecContext *avctx; yading@10: DSPContext dsp; yading@10: DiracDSPContext diracdsp; yading@10: GetBitContext gb; yading@10: dirac_source_params source; yading@10: int seen_sequence_header; yading@10: int frame_number; /* number of the next frame to display */ yading@10: Plane plane[3]; yading@10: int chroma_x_shift; yading@10: int chroma_y_shift; yading@10: yading@10: int zero_res; /* zero residue flag */ yading@10: int is_arith; /* whether coeffs use arith or golomb coding */ yading@10: int low_delay; /* use the low delay syntax */ yading@10: int globalmc_flag; /* use global motion compensation */ yading@10: int num_refs; /* number of reference pictures */ yading@10: yading@10: /* wavelet decoding */ yading@10: unsigned wavelet_depth; /* depth of the IDWT */ yading@10: unsigned wavelet_idx; yading@10: yading@10: /** yading@10: * schroedinger older than 1.0.8 doesn't store yading@10: * quant delta if only one codebook exists in a band yading@10: */ yading@10: unsigned old_delta_quant; yading@10: unsigned codeblock_mode; yading@10: yading@10: struct { yading@10: unsigned width; yading@10: unsigned height; yading@10: } codeblock[MAX_DWT_LEVELS+1]; yading@10: yading@10: struct { yading@10: unsigned num_x; /* number of horizontal slices */ yading@10: unsigned num_y; /* number of vertical slices */ yading@10: AVRational bytes; /* average bytes per slice */ yading@10: uint8_t quant[MAX_DWT_LEVELS][4]; /* [DIRAC_STD] E.1 */ yading@10: } lowdelay; yading@10: yading@10: struct { yading@10: int pan_tilt[2]; /* pan/tilt vector */ yading@10: int zrs[2][2]; /* zoom/rotate/shear matrix */ yading@10: int perspective[2]; /* perspective vector */ yading@10: unsigned zrs_exp; yading@10: unsigned perspective_exp; yading@10: } globalmc[2]; yading@10: yading@10: /* motion compensation */ yading@10: uint8_t mv_precision; /* [DIRAC_STD] REFS_WT_PRECISION */ yading@10: int16_t weight[2]; /* [DIRAC_STD] REF1_WT and REF2_WT */ yading@10: unsigned weight_log2denom; /* [DIRAC_STD] REFS_WT_PRECISION */ yading@10: yading@10: int blwidth; /* number of blocks (horizontally) */ yading@10: int blheight; /* number of blocks (vertically) */ yading@10: int sbwidth; /* number of superblocks (horizontally) */ yading@10: int sbheight; /* number of superblocks (vertically) */ yading@10: yading@10: uint8_t *sbsplit; yading@10: DiracBlock *blmotion; yading@10: yading@10: uint8_t *edge_emu_buffer[4]; yading@10: uint8_t *edge_emu_buffer_base; yading@10: yading@10: uint16_t *mctmp; /* buffer holding the MC data multipled by OBMC weights */ yading@10: uint8_t *mcscratch; yading@10: yading@10: DECLARE_ALIGNED(16, uint8_t, obmc_weight)[3][MAX_BLOCKSIZE*MAX_BLOCKSIZE]; yading@10: yading@10: void (*put_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h); yading@10: void (*avg_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h); yading@10: void (*add_obmc)(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen); yading@10: dirac_weight_func weight_func; yading@10: dirac_biweight_func biweight_func; yading@10: yading@10: DiracFrame *current_picture; yading@10: DiracFrame *ref_pics[2]; yading@10: yading@10: DiracFrame *ref_frames[MAX_REFERENCE_FRAMES+1]; yading@10: DiracFrame *delay_frames[MAX_DELAY+1]; yading@10: DiracFrame all_frames[MAX_FRAMES]; yading@10: } DiracContext; yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * Parse code values. 9.6.1 Table 9.1 yading@10: */ yading@10: enum dirac_parse_code { yading@10: pc_seq_header = 0x00, yading@10: pc_eos = 0x10, yading@10: pc_aux_data = 0x20, yading@10: pc_padding = 0x30, yading@10: }; yading@10: yading@10: enum dirac_subband { yading@10: subband_ll = 0, yading@10: subband_hl = 1, yading@10: subband_lh = 2, yading@10: subband_hh = 3 yading@10: }; yading@10: yading@10: static const uint8_t default_qmat[][4][4] = { yading@10: { { 5, 3, 3, 0}, { 0, 4, 4, 1}, { 0, 5, 5, 2}, { 0, 6, 6, 3} }, yading@10: { { 4, 2, 2, 0}, { 0, 4, 4, 2}, { 0, 5, 5, 3}, { 0, 7, 7, 5} }, yading@10: { { 5, 3, 3, 0}, { 0, 4, 4, 1}, { 0, 5, 5, 2}, { 0, 6, 6, 3} }, yading@10: { { 8, 4, 4, 0}, { 0, 4, 4, 0}, { 0, 4, 4, 0}, { 0, 4, 4, 0} }, yading@10: { { 8, 4, 4, 0}, { 0, 4, 4, 0}, { 0, 4, 4, 0}, { 0, 4, 4, 0} }, yading@10: { { 0, 4, 4, 8}, { 0, 8, 8, 12}, { 0, 13, 13, 17}, { 0, 17, 17, 21} }, yading@10: { { 3, 1, 1, 0}, { 0, 4, 4, 2}, { 0, 6, 6, 5}, { 0, 9, 9, 7} }, yading@10: }; yading@10: yading@10: static const int qscale_tab[MAX_QUANT+1] = { yading@10: 4, 5, 6, 7, 8, 10, 11, 13, yading@10: 16, 19, 23, 27, 32, 38, 45, 54, yading@10: 64, 76, 91, 108, 128, 152, 181, 215, yading@10: 256, 304, 362, 431, 512, 609, 724, 861, yading@10: 1024, 1218, 1448, 1722, 2048, 2435, 2896, 3444, yading@10: 4096, 4871, 5793, 6889, 8192, 9742, 11585, 13777, yading@10: 16384, 19484, 23170, 27554, 32768, 38968, 46341, 55109, yading@10: 65536, 77936 yading@10: }; yading@10: yading@10: static const int qoffset_intra_tab[MAX_QUANT+1] = { yading@10: 1, 2, 3, 4, 4, 5, 6, 7, yading@10: 8, 10, 12, 14, 16, 19, 23, 27, yading@10: 32, 38, 46, 54, 64, 76, 91, 108, yading@10: 128, 152, 181, 216, 256, 305, 362, 431, yading@10: 512, 609, 724, 861, 1024, 1218, 1448, 1722, yading@10: 2048, 2436, 2897, 3445, 4096, 4871, 5793, 6889, yading@10: 8192, 9742, 11585, 13777, 16384, 19484, 23171, 27555, yading@10: 32768, 38968 yading@10: }; yading@10: yading@10: static const int qoffset_inter_tab[MAX_QUANT+1] = { yading@10: 1, 2, 2, 3, 3, 4, 4, 5, yading@10: 6, 7, 9, 10, 12, 14, 17, 20, yading@10: 24, 29, 34, 41, 48, 57, 68, 81, yading@10: 96, 114, 136, 162, 192, 228, 272, 323, yading@10: 384, 457, 543, 646, 768, 913, 1086, 1292, yading@10: 1536, 1827, 2172, 2583, 3072, 3653, 4344, 5166, yading@10: 6144, 7307, 8689, 10333, 12288, 14613, 17378, 20666, yading@10: 24576, 29226 yading@10: }; yading@10: yading@10: /* magic number division by 3 from schroedinger */ yading@10: static inline int divide3(int x) yading@10: { yading@10: return ((x+1)*21845 + 10922) >> 16; yading@10: } yading@10: yading@10: static DiracFrame *remove_frame(DiracFrame *framelist[], int picnum) yading@10: { yading@10: DiracFrame *remove_pic = NULL; yading@10: int i, remove_idx = -1; yading@10: yading@10: for (i = 0; framelist[i]; i++) yading@10: if (framelist[i]->avframe.display_picture_number == picnum) { yading@10: remove_pic = framelist[i]; yading@10: remove_idx = i; yading@10: } yading@10: yading@10: if (remove_pic) yading@10: for (i = remove_idx; framelist[i]; i++) yading@10: framelist[i] = framelist[i+1]; yading@10: yading@10: return remove_pic; yading@10: } yading@10: yading@10: static int add_frame(DiracFrame *framelist[], int maxframes, DiracFrame *frame) yading@10: { yading@10: int i; yading@10: for (i = 0; i < maxframes; i++) yading@10: if (!framelist[i]) { yading@10: framelist[i] = frame; yading@10: return 0; yading@10: } yading@10: return -1; yading@10: } yading@10: yading@10: static int alloc_sequence_buffers(DiracContext *s) yading@10: { yading@10: int sbwidth = DIVRNDUP(s->source.width, 4); yading@10: int sbheight = DIVRNDUP(s->source.height, 4); yading@10: int i, w, h, top_padding; yading@10: yading@10: /* todo: think more about this / use or set Plane here */ yading@10: for (i = 0; i < 3; i++) { yading@10: int max_xblen = MAX_BLOCKSIZE >> (i ? s->chroma_x_shift : 0); yading@10: int max_yblen = MAX_BLOCKSIZE >> (i ? s->chroma_y_shift : 0); yading@10: w = s->source.width >> (i ? s->chroma_x_shift : 0); yading@10: h = s->source.height >> (i ? s->chroma_y_shift : 0); yading@10: yading@10: /* we allocate the max we support here since num decompositions can yading@10: * change from frame to frame. Stride is aligned to 16 for SIMD, and yading@10: * 1<0) in arith decoding yading@10: * MAX_BLOCKSIZE padding for MC: blocks can spill up to half of that yading@10: * on each side */ yading@10: top_padding = FFMAX(1<plane[i].idwt_buf_base = av_mallocz((w+max_xblen)*h * sizeof(IDWTELEM)); yading@10: s->plane[i].idwt_tmp = av_malloc((w+16) * sizeof(IDWTELEM)); yading@10: s->plane[i].idwt_buf = s->plane[i].idwt_buf_base + top_padding*w; yading@10: if (!s->plane[i].idwt_buf_base || !s->plane[i].idwt_tmp) yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: w = s->source.width; yading@10: h = s->source.height; yading@10: yading@10: /* fixme: allocate using real stride here */ yading@10: s->sbsplit = av_malloc(sbwidth * sbheight); yading@10: s->blmotion = av_malloc(sbwidth * sbheight * 16 * sizeof(*s->blmotion)); yading@10: s->edge_emu_buffer_base = av_malloc((w+64)*MAX_BLOCKSIZE); yading@10: yading@10: s->mctmp = av_malloc((w+64+MAX_BLOCKSIZE) * (h+MAX_BLOCKSIZE) * sizeof(*s->mctmp)); yading@10: s->mcscratch = av_malloc((w+64)*MAX_BLOCKSIZE); yading@10: yading@10: if (!s->sbsplit || !s->blmotion || !s->mctmp || !s->mcscratch) yading@10: return AVERROR(ENOMEM); yading@10: return 0; yading@10: } yading@10: yading@10: static void free_sequence_buffers(DiracContext *s) yading@10: { yading@10: int i, j, k; yading@10: yading@10: for (i = 0; i < MAX_FRAMES; i++) { yading@10: if (s->all_frames[i].avframe.data[0]) { yading@10: av_frame_unref(&s->all_frames[i].avframe); yading@10: memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated)); yading@10: } yading@10: yading@10: for (j = 0; j < 3; j++) yading@10: for (k = 1; k < 4; k++) yading@10: av_freep(&s->all_frames[i].hpel_base[j][k]); yading@10: } yading@10: yading@10: memset(s->ref_frames, 0, sizeof(s->ref_frames)); yading@10: memset(s->delay_frames, 0, sizeof(s->delay_frames)); yading@10: yading@10: for (i = 0; i < 3; i++) { yading@10: av_freep(&s->plane[i].idwt_buf_base); yading@10: av_freep(&s->plane[i].idwt_tmp); yading@10: } yading@10: yading@10: av_freep(&s->sbsplit); yading@10: av_freep(&s->blmotion); yading@10: av_freep(&s->edge_emu_buffer_base); yading@10: yading@10: av_freep(&s->mctmp); yading@10: av_freep(&s->mcscratch); yading@10: } yading@10: yading@10: static av_cold int dirac_decode_init(AVCodecContext *avctx) yading@10: { yading@10: DiracContext *s = avctx->priv_data; yading@10: s->avctx = avctx; yading@10: s->frame_number = -1; yading@10: yading@10: if (avctx->flags&CODEC_FLAG_EMU_EDGE) { yading@10: av_log(avctx, AV_LOG_ERROR, "Edge emulation not supported!\n"); yading@10: return AVERROR_PATCHWELCOME; yading@10: } yading@10: yading@10: ff_dsputil_init(&s->dsp, avctx); yading@10: ff_diracdsp_init(&s->diracdsp); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static void dirac_decode_flush(AVCodecContext *avctx) yading@10: { yading@10: DiracContext *s = avctx->priv_data; yading@10: free_sequence_buffers(s); yading@10: s->seen_sequence_header = 0; yading@10: s->frame_number = -1; yading@10: } yading@10: yading@10: static av_cold int dirac_decode_end(AVCodecContext *avctx) yading@10: { yading@10: dirac_decode_flush(avctx); yading@10: return 0; yading@10: } yading@10: yading@10: #define SIGN_CTX(x) (CTX_SIGN_ZERO + ((x) > 0) - ((x) < 0)) yading@10: yading@10: static inline void coeff_unpack_arith(DiracArith *c, int qfactor, int qoffset, yading@10: SubBand *b, IDWTELEM *buf, int x, int y) yading@10: { yading@10: int coeff, sign; yading@10: int sign_pred = 0; yading@10: int pred_ctx = CTX_ZPZN_F1; yading@10: yading@10: /* Check if the parent subband has a 0 in the corresponding position */ yading@10: if (b->parent) yading@10: pred_ctx += !!b->parent->ibuf[b->parent->stride * (y>>1) + (x>>1)] << 1; yading@10: yading@10: if (b->orientation == subband_hl) yading@10: sign_pred = buf[-b->stride]; yading@10: yading@10: /* Determine if the pixel has only zeros in its neighbourhood */ yading@10: if (x) { yading@10: pred_ctx += !(buf[-1] | buf[-b->stride] | buf[-1-b->stride]); yading@10: if (b->orientation == subband_lh) yading@10: sign_pred = buf[-1]; yading@10: } else { yading@10: pred_ctx += !buf[-b->stride]; yading@10: } yading@10: yading@10: coeff = dirac_get_arith_uint(c, pred_ctx, CTX_COEFF_DATA); yading@10: if (coeff) { yading@10: coeff = (coeff * qfactor + qoffset + 2) >> 2; yading@10: sign = dirac_get_arith_bit(c, SIGN_CTX(sign_pred)); yading@10: coeff = (coeff ^ -sign) + sign; yading@10: } yading@10: *buf = coeff; yading@10: } yading@10: yading@10: static inline int coeff_unpack_golomb(GetBitContext *gb, int qfactor, int qoffset) yading@10: { yading@10: int sign, coeff; yading@10: yading@10: coeff = svq3_get_ue_golomb(gb); yading@10: if (coeff) { yading@10: coeff = (coeff * qfactor + qoffset + 2) >> 2; yading@10: sign = get_bits1(gb); yading@10: coeff = (coeff ^ -sign) + sign; yading@10: } yading@10: return coeff; yading@10: } yading@10: yading@10: /** yading@10: * Decode the coeffs in the rectangle defined by left, right, top, bottom yading@10: * [DIRAC_STD] 13.4.3.2 Codeblock unpacking loop. codeblock() yading@10: */ yading@10: static inline void codeblock(DiracContext *s, SubBand *b, yading@10: GetBitContext *gb, DiracArith *c, yading@10: int left, int right, int top, int bottom, yading@10: int blockcnt_one, int is_arith) yading@10: { yading@10: int x, y, zero_block; yading@10: int qoffset, qfactor; yading@10: IDWTELEM *buf; yading@10: yading@10: /* check for any coded coefficients in this codeblock */ yading@10: if (!blockcnt_one) { yading@10: if (is_arith) yading@10: zero_block = dirac_get_arith_bit(c, CTX_ZERO_BLOCK); yading@10: else yading@10: zero_block = get_bits1(gb); yading@10: yading@10: if (zero_block) yading@10: return; yading@10: } yading@10: yading@10: if (s->codeblock_mode && !(s->old_delta_quant && blockcnt_one)) { yading@10: int quant = b->quant; yading@10: if (is_arith) yading@10: quant += dirac_get_arith_int(c, CTX_DELTA_Q_F, CTX_DELTA_Q_DATA); yading@10: else yading@10: quant += dirac_get_se_golomb(gb); yading@10: if (quant < 0) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Invalid quant\n"); yading@10: return; yading@10: } yading@10: b->quant = quant; yading@10: } yading@10: yading@10: b->quant = FFMIN(b->quant, MAX_QUANT); yading@10: yading@10: qfactor = qscale_tab[b->quant]; yading@10: /* TODO: context pointer? */ yading@10: if (!s->num_refs) yading@10: qoffset = qoffset_intra_tab[b->quant]; yading@10: else yading@10: qoffset = qoffset_inter_tab[b->quant]; yading@10: yading@10: buf = b->ibuf + top * b->stride; yading@10: for (y = top; y < bottom; y++) { yading@10: for (x = left; x < right; x++) { yading@10: /* [DIRAC_STD] 13.4.4 Subband coefficients. coeff_unpack() */ yading@10: if (is_arith) yading@10: coeff_unpack_arith(c, qfactor, qoffset, b, buf+x, x, y); yading@10: else yading@10: buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); yading@10: } yading@10: buf += b->stride; yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 13.3 intra_dc_prediction(band) yading@10: */ yading@10: static inline void intra_dc_prediction(SubBand *b) yading@10: { yading@10: IDWTELEM *buf = b->ibuf; yading@10: int x, y; yading@10: yading@10: for (x = 1; x < b->width; x++) yading@10: buf[x] += buf[x-1]; yading@10: buf += b->stride; yading@10: yading@10: for (y = 1; y < b->height; y++) { yading@10: buf[0] += buf[-b->stride]; yading@10: yading@10: for (x = 1; x < b->width; x++) { yading@10: int pred = buf[x - 1] + buf[x - b->stride] + buf[x - b->stride-1]; yading@10: buf[x] += divide3(pred); yading@10: } yading@10: buf += b->stride; yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 13.4.2 Non-skipped subbands. subband_coeffs() yading@10: */ yading@10: static av_always_inline void decode_subband_internal(DiracContext *s, SubBand *b, int is_arith) yading@10: { yading@10: int cb_x, cb_y, left, right, top, bottom; yading@10: DiracArith c; yading@10: GetBitContext gb; yading@10: int cb_width = s->codeblock[b->level + (b->orientation != subband_ll)].width; yading@10: int cb_height = s->codeblock[b->level + (b->orientation != subband_ll)].height; yading@10: int blockcnt_one = (cb_width + cb_height) == 2; yading@10: yading@10: if (!b->length) yading@10: return; yading@10: yading@10: init_get_bits(&gb, b->coeff_data, b->length*8); yading@10: yading@10: if (is_arith) yading@10: ff_dirac_init_arith_decoder(&c, &gb, b->length); yading@10: yading@10: top = 0; yading@10: for (cb_y = 0; cb_y < cb_height; cb_y++) { yading@10: bottom = (b->height * (cb_y+1)) / cb_height; yading@10: left = 0; yading@10: for (cb_x = 0; cb_x < cb_width; cb_x++) { yading@10: right = (b->width * (cb_x+1)) / cb_width; yading@10: codeblock(s, b, &gb, &c, left, right, top, bottom, blockcnt_one, is_arith); yading@10: left = right; yading@10: } yading@10: top = bottom; yading@10: } yading@10: yading@10: if (b->orientation == subband_ll && s->num_refs == 0) yading@10: intra_dc_prediction(b); yading@10: } yading@10: yading@10: static int decode_subband_arith(AVCodecContext *avctx, void *b) yading@10: { yading@10: DiracContext *s = avctx->priv_data; yading@10: decode_subband_internal(s, b, 1); yading@10: return 0; yading@10: } yading@10: yading@10: static int decode_subband_golomb(AVCodecContext *avctx, void *arg) yading@10: { yading@10: DiracContext *s = avctx->priv_data; yading@10: SubBand **b = arg; yading@10: decode_subband_internal(s, *b, 0); yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * [DIRAC_STD] 13.4.1 core_transform_data() yading@10: */ yading@10: static void decode_component(DiracContext *s, int comp) yading@10: { yading@10: AVCodecContext *avctx = s->avctx; yading@10: SubBand *bands[3*MAX_DWT_LEVELS+1]; yading@10: enum dirac_subband orientation; yading@10: int level, num_bands = 0; yading@10: yading@10: /* Unpack all subbands at all levels. */ yading@10: for (level = 0; level < s->wavelet_depth; level++) { yading@10: for (orientation = !!level; orientation < 4; orientation++) { yading@10: SubBand *b = &s->plane[comp].band[level][orientation]; yading@10: bands[num_bands++] = b; yading@10: yading@10: align_get_bits(&s->gb); yading@10: /* [DIRAC_STD] 13.4.2 subband() */ yading@10: b->length = svq3_get_ue_golomb(&s->gb); yading@10: if (b->length) { yading@10: b->quant = svq3_get_ue_golomb(&s->gb); yading@10: align_get_bits(&s->gb); yading@10: b->coeff_data = s->gb.buffer + get_bits_count(&s->gb)/8; yading@10: b->length = FFMIN(b->length, FFMAX(get_bits_left(&s->gb)/8, 0)); yading@10: skip_bits_long(&s->gb, b->length*8); yading@10: } yading@10: } yading@10: /* arithmetic coding has inter-level dependencies, so we can only execute one level at a time */ yading@10: if (s->is_arith) yading@10: avctx->execute(avctx, decode_subband_arith, &s->plane[comp].band[level][!!level], yading@10: NULL, 4-!!level, sizeof(SubBand)); yading@10: } yading@10: /* golomb coding has no inter-level dependencies, so we can execute all subbands in parallel */ yading@10: if (!s->is_arith) yading@10: avctx->execute(avctx, decode_subband_golomb, bands, NULL, num_bands, sizeof(SubBand*)); yading@10: } yading@10: yading@10: /* [DIRAC_STD] 13.5.5.2 Luma slice subband data. luma_slice_band(level,orient,sx,sy) --> if b2 == NULL */ yading@10: /* [DIRAC_STD] 13.5.5.3 Chroma slice subband data. chroma_slice_band(level,orient,sx,sy) --> if b2 != NULL */ yading@10: static void lowdelay_subband(DiracContext *s, GetBitContext *gb, int quant, yading@10: int slice_x, int slice_y, int bits_end, yading@10: SubBand *b1, SubBand *b2) yading@10: { yading@10: int left = b1->width * slice_x / s->lowdelay.num_x; yading@10: int right = b1->width *(slice_x+1) / s->lowdelay.num_x; yading@10: int top = b1->height * slice_y / s->lowdelay.num_y; yading@10: int bottom = b1->height *(slice_y+1) / s->lowdelay.num_y; yading@10: yading@10: int qfactor = qscale_tab[FFMIN(quant, MAX_QUANT)]; yading@10: int qoffset = qoffset_intra_tab[FFMIN(quant, MAX_QUANT)]; yading@10: yading@10: IDWTELEM *buf1 = b1->ibuf + top * b1->stride; yading@10: IDWTELEM *buf2 = b2 ? b2->ibuf + top * b2->stride : NULL; yading@10: int x, y; yading@10: /* we have to constantly check for overread since the spec explictly yading@10: requires this, with the meaning that all remaining coeffs are set to 0 */ yading@10: if (get_bits_count(gb) >= bits_end) yading@10: return; yading@10: yading@10: for (y = top; y < bottom; y++) { yading@10: for (x = left; x < right; x++) { yading@10: buf1[x] = coeff_unpack_golomb(gb, qfactor, qoffset); yading@10: if (get_bits_count(gb) >= bits_end) yading@10: return; yading@10: if (buf2) { yading@10: buf2[x] = coeff_unpack_golomb(gb, qfactor, qoffset); yading@10: if (get_bits_count(gb) >= bits_end) yading@10: return; yading@10: } yading@10: } yading@10: buf1 += b1->stride; yading@10: if (buf2) yading@10: buf2 += b2->stride; yading@10: } yading@10: } yading@10: yading@10: struct lowdelay_slice { yading@10: GetBitContext gb; yading@10: int slice_x; yading@10: int slice_y; yading@10: int bytes; yading@10: }; yading@10: yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 13.5.2 Slices. slice(sx,sy) yading@10: */ yading@10: static int decode_lowdelay_slice(AVCodecContext *avctx, void *arg) yading@10: { yading@10: DiracContext *s = avctx->priv_data; yading@10: struct lowdelay_slice *slice = arg; yading@10: GetBitContext *gb = &slice->gb; yading@10: enum dirac_subband orientation; yading@10: int level, quant, chroma_bits, chroma_end; yading@10: yading@10: int quant_base = get_bits(gb, 7); /*[DIRAC_STD] qindex */ yading@10: int length_bits = av_log2(8 * slice->bytes)+1; yading@10: int luma_bits = get_bits_long(gb, length_bits); yading@10: int luma_end = get_bits_count(gb) + FFMIN(luma_bits, get_bits_left(gb)); yading@10: yading@10: /* [DIRAC_STD] 13.5.5.2 luma_slice_band */ yading@10: for (level = 0; level < s->wavelet_depth; level++) yading@10: for (orientation = !!level; orientation < 4; orientation++) { yading@10: quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0); yading@10: lowdelay_subband(s, gb, quant, slice->slice_x, slice->slice_y, luma_end, yading@10: &s->plane[0].band[level][orientation], NULL); yading@10: } yading@10: yading@10: /* consume any unused bits from luma */ yading@10: skip_bits_long(gb, get_bits_count(gb) - luma_end); yading@10: yading@10: chroma_bits = 8*slice->bytes - 7 - length_bits - luma_bits; yading@10: chroma_end = get_bits_count(gb) + FFMIN(chroma_bits, get_bits_left(gb)); yading@10: /* [DIRAC_STD] 13.5.5.3 chroma_slice_band */ yading@10: for (level = 0; level < s->wavelet_depth; level++) yading@10: for (orientation = !!level; orientation < 4; orientation++) { yading@10: quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0); yading@10: lowdelay_subband(s, gb, quant, slice->slice_x, slice->slice_y, chroma_end, yading@10: &s->plane[1].band[level][orientation], yading@10: &s->plane[2].band[level][orientation]); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 13.5.1 low_delay_transform_data() yading@10: */ yading@10: static void decode_lowdelay(DiracContext *s) yading@10: { yading@10: AVCodecContext *avctx = s->avctx; yading@10: int slice_x, slice_y, bytes, bufsize; yading@10: const uint8_t *buf; yading@10: struct lowdelay_slice *slices; yading@10: int slice_num = 0; yading@10: yading@10: slices = av_mallocz(s->lowdelay.num_x * s->lowdelay.num_y * sizeof(struct lowdelay_slice)); yading@10: yading@10: align_get_bits(&s->gb); yading@10: /*[DIRAC_STD] 13.5.2 Slices. slice(sx,sy) */ yading@10: buf = s->gb.buffer + get_bits_count(&s->gb)/8; yading@10: bufsize = get_bits_left(&s->gb); yading@10: yading@10: for (slice_y = 0; bufsize > 0 && slice_y < s->lowdelay.num_y; slice_y++) yading@10: for (slice_x = 0; bufsize > 0 && slice_x < s->lowdelay.num_x; slice_x++) { yading@10: bytes = (slice_num+1) * s->lowdelay.bytes.num / s->lowdelay.bytes.den yading@10: - slice_num * s->lowdelay.bytes.num / s->lowdelay.bytes.den; yading@10: yading@10: slices[slice_num].bytes = bytes; yading@10: slices[slice_num].slice_x = slice_x; yading@10: slices[slice_num].slice_y = slice_y; yading@10: init_get_bits(&slices[slice_num].gb, buf, bufsize); yading@10: slice_num++; yading@10: yading@10: buf += bytes; yading@10: bufsize -= bytes*8; yading@10: } yading@10: yading@10: avctx->execute(avctx, decode_lowdelay_slice, slices, NULL, slice_num, yading@10: sizeof(struct lowdelay_slice)); /* [DIRAC_STD] 13.5.2 Slices */ yading@10: intra_dc_prediction(&s->plane[0].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */ yading@10: intra_dc_prediction(&s->plane[1].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */ yading@10: intra_dc_prediction(&s->plane[2].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */ yading@10: av_free(slices); yading@10: } yading@10: yading@10: static void init_planes(DiracContext *s) yading@10: { yading@10: int i, w, h, level, orientation; yading@10: yading@10: for (i = 0; i < 3; i++) { yading@10: Plane *p = &s->plane[i]; yading@10: yading@10: p->width = s->source.width >> (i ? s->chroma_x_shift : 0); yading@10: p->height = s->source.height >> (i ? s->chroma_y_shift : 0); yading@10: p->idwt_width = w = CALC_PADDING(p->width , s->wavelet_depth); yading@10: p->idwt_height = h = CALC_PADDING(p->height, s->wavelet_depth); yading@10: p->idwt_stride = FFALIGN(p->idwt_width, 8); yading@10: yading@10: for (level = s->wavelet_depth-1; level >= 0; level--) { yading@10: w = w>>1; yading@10: h = h>>1; yading@10: for (orientation = !!level; orientation < 4; orientation++) { yading@10: SubBand *b = &p->band[level][orientation]; yading@10: yading@10: b->ibuf = p->idwt_buf; yading@10: b->level = level; yading@10: b->stride = p->idwt_stride << (s->wavelet_depth - level); yading@10: b->width = w; yading@10: b->height = h; yading@10: b->orientation = orientation; yading@10: yading@10: if (orientation & 1) yading@10: b->ibuf += w; yading@10: if (orientation > 1) yading@10: b->ibuf += b->stride>>1; yading@10: yading@10: if (level) yading@10: b->parent = &p->band[level-1][orientation]; yading@10: } yading@10: } yading@10: yading@10: if (i > 0) { yading@10: p->xblen = s->plane[0].xblen >> s->chroma_x_shift; yading@10: p->yblen = s->plane[0].yblen >> s->chroma_y_shift; yading@10: p->xbsep = s->plane[0].xbsep >> s->chroma_x_shift; yading@10: p->ybsep = s->plane[0].ybsep >> s->chroma_y_shift; yading@10: } yading@10: yading@10: p->xoffset = (p->xblen - p->xbsep)/2; yading@10: p->yoffset = (p->yblen - p->ybsep)/2; yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Unpack the motion compensation parameters yading@10: * Dirac Specification -> yading@10: * 11.2 Picture prediction data. picture_prediction() yading@10: */ yading@10: static int dirac_unpack_prediction_parameters(DiracContext *s) yading@10: { yading@10: static const uint8_t default_blen[] = { 4, 12, 16, 24 }; yading@10: static const uint8_t default_bsep[] = { 4, 8, 12, 16 }; yading@10: yading@10: GetBitContext *gb = &s->gb; yading@10: unsigned idx, ref; yading@10: yading@10: align_get_bits(gb); yading@10: /* [DIRAC_STD] 11.2.2 Block parameters. block_parameters() */ yading@10: /* Luma and Chroma are equal. 11.2.3 */ yading@10: idx = svq3_get_ue_golomb(gb); /* [DIRAC_STD] index */ yading@10: yading@10: if (idx > 4) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Block prediction index too high\n"); yading@10: return -1; yading@10: } yading@10: yading@10: if (idx == 0) { yading@10: s->plane[0].xblen = svq3_get_ue_golomb(gb); yading@10: s->plane[0].yblen = svq3_get_ue_golomb(gb); yading@10: s->plane[0].xbsep = svq3_get_ue_golomb(gb); yading@10: s->plane[0].ybsep = svq3_get_ue_golomb(gb); yading@10: } else { yading@10: /*[DIRAC_STD] preset_block_params(index). Table 11.1 */ yading@10: s->plane[0].xblen = default_blen[idx-1]; yading@10: s->plane[0].yblen = default_blen[idx-1]; yading@10: s->plane[0].xbsep = default_bsep[idx-1]; yading@10: s->plane[0].ybsep = default_bsep[idx-1]; yading@10: } yading@10: /*[DIRAC_STD] 11.2.4 motion_data_dimensions() yading@10: Calculated in function dirac_unpack_block_motion_data */ yading@10: yading@10: 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) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Block separation too small\n"); yading@10: return -1; yading@10: } yading@10: if (s->plane[0].xbsep > s->plane[0].xblen || s->plane[0].ybsep > s->plane[0].yblen) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Block separation greater than size\n"); yading@10: return -1; yading@10: } yading@10: if (FFMAX(s->plane[0].xblen, s->plane[0].yblen) > MAX_BLOCKSIZE) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Unsupported large block size\n"); yading@10: return -1; yading@10: } yading@10: yading@10: /*[DIRAC_STD] 11.2.5 Motion vector precision. motion_vector_precision() yading@10: Read motion vector precision */ yading@10: s->mv_precision = svq3_get_ue_golomb(gb); yading@10: if (s->mv_precision > 3) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "MV precision finer than eighth-pel\n"); yading@10: return -1; yading@10: } yading@10: yading@10: /*[DIRAC_STD] 11.2.6 Global motion. global_motion() yading@10: Read the global motion compensation parameters */ yading@10: s->globalmc_flag = get_bits1(gb); yading@10: if (s->globalmc_flag) { yading@10: memset(s->globalmc, 0, sizeof(s->globalmc)); yading@10: /* [DIRAC_STD] pan_tilt(gparams) */ yading@10: for (ref = 0; ref < s->num_refs; ref++) { yading@10: if (get_bits1(gb)) { yading@10: s->globalmc[ref].pan_tilt[0] = dirac_get_se_golomb(gb); yading@10: s->globalmc[ref].pan_tilt[1] = dirac_get_se_golomb(gb); yading@10: } yading@10: /* [DIRAC_STD] zoom_rotate_shear(gparams) yading@10: zoom/rotation/shear parameters */ yading@10: if (get_bits1(gb)) { yading@10: s->globalmc[ref].zrs_exp = svq3_get_ue_golomb(gb); yading@10: s->globalmc[ref].zrs[0][0] = dirac_get_se_golomb(gb); yading@10: s->globalmc[ref].zrs[0][1] = dirac_get_se_golomb(gb); yading@10: s->globalmc[ref].zrs[1][0] = dirac_get_se_golomb(gb); yading@10: s->globalmc[ref].zrs[1][1] = dirac_get_se_golomb(gb); yading@10: } else { yading@10: s->globalmc[ref].zrs[0][0] = 1; yading@10: s->globalmc[ref].zrs[1][1] = 1; yading@10: } yading@10: /* [DIRAC_STD] perspective(gparams) */ yading@10: if (get_bits1(gb)) { yading@10: s->globalmc[ref].perspective_exp = svq3_get_ue_golomb(gb); yading@10: s->globalmc[ref].perspective[0] = dirac_get_se_golomb(gb); yading@10: s->globalmc[ref].perspective[1] = dirac_get_se_golomb(gb); yading@10: } yading@10: } yading@10: } yading@10: yading@10: /*[DIRAC_STD] 11.2.7 Picture prediction mode. prediction_mode() yading@10: Picture prediction mode, not currently used. */ yading@10: if (svq3_get_ue_golomb(gb)) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Unknown picture prediction mode\n"); yading@10: return -1; yading@10: } yading@10: yading@10: /* [DIRAC_STD] 11.2.8 Reference picture weight. reference_picture_weights() yading@10: just data read, weight calculation will be done later on. */ yading@10: s->weight_log2denom = 1; yading@10: s->weight[0] = 1; yading@10: s->weight[1] = 1; yading@10: yading@10: if (get_bits1(gb)) { yading@10: s->weight_log2denom = svq3_get_ue_golomb(gb); yading@10: s->weight[0] = dirac_get_se_golomb(gb); yading@10: if (s->num_refs == 2) yading@10: s->weight[1] = dirac_get_se_golomb(gb); yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 11.3 Wavelet transform data. wavelet_transform() yading@10: */ yading@10: static int dirac_unpack_idwt_params(DiracContext *s) yading@10: { yading@10: GetBitContext *gb = &s->gb; yading@10: int i, level; yading@10: unsigned tmp; yading@10: yading@10: #define CHECKEDREAD(dst, cond, errmsg) \ yading@10: tmp = svq3_get_ue_golomb(gb); \ yading@10: if (cond) { \ yading@10: av_log(s->avctx, AV_LOG_ERROR, errmsg); \ yading@10: return -1; \ yading@10: }\ yading@10: dst = tmp; yading@10: yading@10: align_get_bits(gb); yading@10: yading@10: s->zero_res = s->num_refs ? get_bits1(gb) : 0; yading@10: if (s->zero_res) yading@10: return 0; yading@10: yading@10: /*[DIRAC_STD] 11.3.1 Transform parameters. transform_parameters() */ yading@10: CHECKEDREAD(s->wavelet_idx, tmp > 6, "wavelet_idx is too big\n") yading@10: yading@10: CHECKEDREAD(s->wavelet_depth, tmp > MAX_DWT_LEVELS || tmp < 1, "invalid number of DWT decompositions\n") yading@10: yading@10: if (!s->low_delay) { yading@10: /* Codeblock parameters (core syntax only) */ yading@10: if (get_bits1(gb)) { yading@10: for (i = 0; i <= s->wavelet_depth; i++) { yading@10: CHECKEDREAD(s->codeblock[i].width , tmp < 1, "codeblock width invalid\n") yading@10: CHECKEDREAD(s->codeblock[i].height, tmp < 1, "codeblock height invalid\n") yading@10: } yading@10: yading@10: CHECKEDREAD(s->codeblock_mode, tmp > 1, "unknown codeblock mode\n") yading@10: } else yading@10: for (i = 0; i <= s->wavelet_depth; i++) yading@10: s->codeblock[i].width = s->codeblock[i].height = 1; yading@10: } else { yading@10: /* Slice parameters + quantization matrix*/ yading@10: /*[DIRAC_STD] 11.3.4 Slice coding Parameters (low delay syntax only). slice_parameters() */ yading@10: s->lowdelay.num_x = svq3_get_ue_golomb(gb); yading@10: s->lowdelay.num_y = svq3_get_ue_golomb(gb); yading@10: s->lowdelay.bytes.num = svq3_get_ue_golomb(gb); yading@10: s->lowdelay.bytes.den = svq3_get_ue_golomb(gb); yading@10: yading@10: if (s->lowdelay.bytes.den <= 0) { yading@10: av_log(s->avctx,AV_LOG_ERROR,"Invalid lowdelay.bytes.den\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: /* [DIRAC_STD] 11.3.5 Quantisation matrices (low-delay syntax). quant_matrix() */ yading@10: if (get_bits1(gb)) { yading@10: av_log(s->avctx,AV_LOG_DEBUG,"Low Delay: Has Custom Quantization Matrix!\n"); yading@10: /* custom quantization matrix */ yading@10: s->lowdelay.quant[0][0] = svq3_get_ue_golomb(gb); yading@10: for (level = 0; level < s->wavelet_depth; level++) { yading@10: s->lowdelay.quant[level][1] = svq3_get_ue_golomb(gb); yading@10: s->lowdelay.quant[level][2] = svq3_get_ue_golomb(gb); yading@10: s->lowdelay.quant[level][3] = svq3_get_ue_golomb(gb); yading@10: } yading@10: } else { yading@10: if (s->wavelet_depth > 4) { yading@10: av_log(s->avctx,AV_LOG_ERROR,"Mandatory custom low delay matrix missing for depth %d\n", s->wavelet_depth); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: /* default quantization matrix */ yading@10: for (level = 0; level < s->wavelet_depth; level++) yading@10: for (i = 0; i < 4; i++) { yading@10: s->lowdelay.quant[level][i] = default_qmat[s->wavelet_idx][level][i]; yading@10: /* haar with no shift differs for different depths */ yading@10: if (s->wavelet_idx == 3) yading@10: s->lowdelay.quant[level][i] += 4*(s->wavelet_depth-1 - level); yading@10: } yading@10: } yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: static inline int pred_sbsplit(uint8_t *sbsplit, int stride, int x, int y) yading@10: { yading@10: static const uint8_t avgsplit[7] = { 0, 0, 1, 1, 1, 2, 2 }; yading@10: yading@10: if (!(x|y)) yading@10: return 0; yading@10: else if (!y) yading@10: return sbsplit[-1]; yading@10: else if (!x) yading@10: return sbsplit[-stride]; yading@10: yading@10: return avgsplit[sbsplit[-1] + sbsplit[-stride] + sbsplit[-stride-1]]; yading@10: } yading@10: yading@10: static inline int pred_block_mode(DiracBlock *block, int stride, int x, int y, int refmask) yading@10: { yading@10: int pred; yading@10: yading@10: if (!(x|y)) yading@10: return 0; yading@10: else if (!y) yading@10: return block[-1].ref & refmask; yading@10: else if (!x) yading@10: return block[-stride].ref & refmask; yading@10: yading@10: /* return the majority */ yading@10: pred = (block[-1].ref & refmask) + (block[-stride].ref & refmask) + (block[-stride-1].ref & refmask); yading@10: return (pred >> 1) & refmask; yading@10: } yading@10: yading@10: static inline void pred_block_dc(DiracBlock *block, int stride, int x, int y) yading@10: { yading@10: int i, n = 0; yading@10: yading@10: memset(block->u.dc, 0, sizeof(block->u.dc)); yading@10: yading@10: if (x && !(block[-1].ref & 3)) { yading@10: for (i = 0; i < 3; i++) yading@10: block->u.dc[i] += block[-1].u.dc[i]; yading@10: n++; yading@10: } yading@10: yading@10: if (y && !(block[-stride].ref & 3)) { yading@10: for (i = 0; i < 3; i++) yading@10: block->u.dc[i] += block[-stride].u.dc[i]; yading@10: n++; yading@10: } yading@10: yading@10: if (x && y && !(block[-1-stride].ref & 3)) { yading@10: for (i = 0; i < 3; i++) yading@10: block->u.dc[i] += block[-1-stride].u.dc[i]; yading@10: n++; yading@10: } yading@10: yading@10: if (n == 2) { yading@10: for (i = 0; i < 3; i++) yading@10: block->u.dc[i] = (block->u.dc[i]+1)>>1; yading@10: } else if (n == 3) { yading@10: for (i = 0; i < 3; i++) yading@10: block->u.dc[i] = divide3(block->u.dc[i]); yading@10: } yading@10: } yading@10: yading@10: static inline void pred_mv(DiracBlock *block, int stride, int x, int y, int ref) yading@10: { yading@10: int16_t *pred[3]; yading@10: int refmask = ref+1; yading@10: int mask = refmask | DIRAC_REF_MASK_GLOBAL; /* exclude gmc blocks */ yading@10: int n = 0; yading@10: yading@10: if (x && (block[-1].ref & mask) == refmask) yading@10: pred[n++] = block[-1].u.mv[ref]; yading@10: yading@10: if (y && (block[-stride].ref & mask) == refmask) yading@10: pred[n++] = block[-stride].u.mv[ref]; yading@10: yading@10: if (x && y && (block[-stride-1].ref & mask) == refmask) yading@10: pred[n++] = block[-stride-1].u.mv[ref]; yading@10: yading@10: switch (n) { yading@10: case 0: yading@10: block->u.mv[ref][0] = 0; yading@10: block->u.mv[ref][1] = 0; yading@10: break; yading@10: case 1: yading@10: block->u.mv[ref][0] = pred[0][0]; yading@10: block->u.mv[ref][1] = pred[0][1]; yading@10: break; yading@10: case 2: yading@10: block->u.mv[ref][0] = (pred[0][0] + pred[1][0] + 1) >> 1; yading@10: block->u.mv[ref][1] = (pred[0][1] + pred[1][1] + 1) >> 1; yading@10: break; yading@10: case 3: yading@10: block->u.mv[ref][0] = mid_pred(pred[0][0], pred[1][0], pred[2][0]); yading@10: block->u.mv[ref][1] = mid_pred(pred[0][1], pred[1][1], pred[2][1]); yading@10: break; yading@10: } yading@10: } yading@10: yading@10: static void global_mv(DiracContext *s, DiracBlock *block, int x, int y, int ref) yading@10: { yading@10: int ez = s->globalmc[ref].zrs_exp; yading@10: int ep = s->globalmc[ref].perspective_exp; yading@10: int (*A)[2] = s->globalmc[ref].zrs; yading@10: int *b = s->globalmc[ref].pan_tilt; yading@10: int *c = s->globalmc[ref].perspective; yading@10: yading@10: int m = (1<u.mv[ref][0] = (mx + (1<<(ez+ep))) >> (ez+ep); yading@10: block->u.mv[ref][1] = (my + (1<<(ez+ep))) >> (ez+ep); yading@10: } yading@10: yading@10: static void decode_block_params(DiracContext *s, DiracArith arith[8], DiracBlock *block, yading@10: int stride, int x, int y) yading@10: { yading@10: int i; yading@10: yading@10: block->ref = pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF1); yading@10: block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF1); yading@10: yading@10: if (s->num_refs == 2) { yading@10: block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF2); yading@10: block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF2) << 1; yading@10: } yading@10: yading@10: if (!block->ref) { yading@10: pred_block_dc(block, stride, x, y); yading@10: for (i = 0; i < 3; i++) yading@10: block->u.dc[i] += dirac_get_arith_int(arith+1+i, CTX_DC_F1, CTX_DC_DATA); yading@10: return; yading@10: } yading@10: yading@10: if (s->globalmc_flag) { yading@10: block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_GLOBAL); yading@10: block->ref ^= dirac_get_arith_bit(arith, CTX_GLOBAL_BLOCK) << 2; yading@10: } yading@10: yading@10: for (i = 0; i < s->num_refs; i++) yading@10: if (block->ref & (i+1)) { yading@10: if (block->ref & DIRAC_REF_MASK_GLOBAL) { yading@10: global_mv(s, block, x, y, i); yading@10: } else { yading@10: pred_mv(block, stride, x, y, i); yading@10: block->u.mv[i][0] += dirac_get_arith_int(arith + 4 + 2 * i, CTX_MV_F1, CTX_MV_DATA); yading@10: block->u.mv[i][1] += dirac_get_arith_int(arith + 5 + 2 * i, CTX_MV_F1, CTX_MV_DATA); yading@10: } yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Copies the current block to the other blocks covered by the current superblock split mode yading@10: */ yading@10: static void propagate_block_data(DiracBlock *block, int stride, int size) yading@10: { yading@10: int x, y; yading@10: DiracBlock *dst = block; yading@10: yading@10: for (x = 1; x < size; x++) yading@10: dst[x] = *block; yading@10: yading@10: for (y = 1; y < size; y++) { yading@10: dst += stride; yading@10: for (x = 0; x < size; x++) yading@10: dst[x] = *block; yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 12. Block motion data syntax yading@10: */ yading@10: static int dirac_unpack_block_motion_data(DiracContext *s) yading@10: { yading@10: GetBitContext *gb = &s->gb; yading@10: uint8_t *sbsplit = s->sbsplit; yading@10: int i, x, y, q, p; yading@10: DiracArith arith[8]; yading@10: yading@10: align_get_bits(gb); yading@10: yading@10: /* [DIRAC_STD] 11.2.4 and 12.2.1 Number of blocks and superblocks */ yading@10: s->sbwidth = DIVRNDUP(s->source.width, 4*s->plane[0].xbsep); yading@10: s->sbheight = DIVRNDUP(s->source.height, 4*s->plane[0].ybsep); yading@10: s->blwidth = 4 * s->sbwidth; yading@10: s->blheight = 4 * s->sbheight; yading@10: yading@10: /* [DIRAC_STD] 12.3.1 Superblock splitting modes. superblock_split_modes() yading@10: decode superblock split modes */ yading@10: ff_dirac_init_arith_decoder(arith, gb, svq3_get_ue_golomb(gb)); /* svq3_get_ue_golomb(gb) is the length */ yading@10: for (y = 0; y < s->sbheight; y++) { yading@10: for (x = 0; x < s->sbwidth; x++) { yading@10: unsigned int split = dirac_get_arith_uint(arith, CTX_SB_F1, CTX_SB_DATA); yading@10: if (split > 2) yading@10: return -1; yading@10: sbsplit[x] = (split + pred_sbsplit(sbsplit+x, s->sbwidth, x, y)) % 3; yading@10: } yading@10: sbsplit += s->sbwidth; yading@10: } yading@10: yading@10: /* setup arith decoding */ yading@10: ff_dirac_init_arith_decoder(arith, gb, svq3_get_ue_golomb(gb)); yading@10: for (i = 0; i < s->num_refs; i++) { yading@10: ff_dirac_init_arith_decoder(arith + 4 + 2 * i, gb, svq3_get_ue_golomb(gb)); yading@10: ff_dirac_init_arith_decoder(arith + 5 + 2 * i, gb, svq3_get_ue_golomb(gb)); yading@10: } yading@10: for (i = 0; i < 3; i++) yading@10: ff_dirac_init_arith_decoder(arith+1+i, gb, svq3_get_ue_golomb(gb)); yading@10: yading@10: for (y = 0; y < s->sbheight; y++) yading@10: for (x = 0; x < s->sbwidth; x++) { yading@10: int blkcnt = 1 << s->sbsplit[y * s->sbwidth + x]; yading@10: int step = 4 >> s->sbsplit[y * s->sbwidth + x]; yading@10: yading@10: for (q = 0; q < blkcnt; q++) yading@10: for (p = 0; p < blkcnt; p++) { yading@10: int bx = 4 * x + p*step; yading@10: int by = 4 * y + q*step; yading@10: DiracBlock *block = &s->blmotion[by*s->blwidth + bx]; yading@10: decode_block_params(s, arith, block, s->blwidth, bx, by); yading@10: propagate_block_data(block, s->blwidth, step); yading@10: } yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int weight(int i, int blen, int offset) yading@10: { yading@10: #define ROLLOFF(i) offset == 1 ? ((i) ? 5 : 3) : \ yading@10: (1 + (6*(i) + offset - 1) / (2*offset - 1)) yading@10: yading@10: if (i < 2*offset) yading@10: return ROLLOFF(i); yading@10: else if (i > blen-1 - 2*offset) yading@10: return ROLLOFF(blen-1 - i); yading@10: return 8; yading@10: } yading@10: yading@10: static void init_obmc_weight_row(Plane *p, uint8_t *obmc_weight, int stride, yading@10: int left, int right, int wy) yading@10: { yading@10: int x; yading@10: for (x = 0; left && x < p->xblen >> 1; x++) yading@10: obmc_weight[x] = wy*8; yading@10: for (; x < p->xblen >> right; x++) yading@10: obmc_weight[x] = wy*weight(x, p->xblen, p->xoffset); yading@10: for (; x < p->xblen; x++) yading@10: obmc_weight[x] = wy*8; yading@10: for (; x < stride; x++) yading@10: obmc_weight[x] = 0; yading@10: } yading@10: yading@10: static void init_obmc_weight(Plane *p, uint8_t *obmc_weight, int stride, yading@10: int left, int right, int top, int bottom) yading@10: { yading@10: int y; yading@10: for (y = 0; top && y < p->yblen >> 1; y++) { yading@10: init_obmc_weight_row(p, obmc_weight, stride, left, right, 8); yading@10: obmc_weight += stride; yading@10: } yading@10: for (; y < p->yblen >> bottom; y++) { yading@10: int wy = weight(y, p->yblen, p->yoffset); yading@10: init_obmc_weight_row(p, obmc_weight, stride, left, right, wy); yading@10: obmc_weight += stride; yading@10: } yading@10: for (; y < p->yblen; y++) { yading@10: init_obmc_weight_row(p, obmc_weight, stride, left, right, 8); yading@10: obmc_weight += stride; yading@10: } yading@10: } yading@10: yading@10: static void init_obmc_weights(DiracContext *s, Plane *p, int by) yading@10: { yading@10: int top = !by; yading@10: int bottom = by == s->blheight-1; yading@10: yading@10: /* don't bother re-initing for rows 2 to blheight-2, the weights don't change */ yading@10: if (top || bottom || by == 1) { yading@10: init_obmc_weight(p, s->obmc_weight[0], MAX_BLOCKSIZE, 1, 0, top, bottom); yading@10: init_obmc_weight(p, s->obmc_weight[1], MAX_BLOCKSIZE, 0, 0, top, bottom); yading@10: init_obmc_weight(p, s->obmc_weight[2], MAX_BLOCKSIZE, 0, 1, top, bottom); yading@10: } yading@10: } yading@10: yading@10: static const uint8_t epel_weights[4][4][4] = { yading@10: {{ 16, 0, 0, 0 }, yading@10: { 12, 4, 0, 0 }, yading@10: { 8, 8, 0, 0 }, yading@10: { 4, 12, 0, 0 }}, yading@10: {{ 12, 0, 4, 0 }, yading@10: { 9, 3, 3, 1 }, yading@10: { 6, 6, 2, 2 }, yading@10: { 3, 9, 1, 3 }}, yading@10: {{ 8, 0, 8, 0 }, yading@10: { 6, 2, 6, 2 }, yading@10: { 4, 4, 4, 4 }, yading@10: { 2, 6, 2, 6 }}, yading@10: {{ 4, 0, 12, 0 }, yading@10: { 3, 1, 9, 3 }, yading@10: { 2, 2, 6, 6 }, yading@10: { 1, 3, 3, 9 }} yading@10: }; yading@10: yading@10: /** yading@10: * For block x,y, determine which of the hpel planes to do bilinear yading@10: * interpolation from and set src[] to the location in each hpel plane yading@10: * to MC from. yading@10: * yading@10: * @return the index of the put_dirac_pixels_tab function to use yading@10: * 0 for 1 plane (fpel,hpel), 1 for 2 planes (qpel), 2 for 4 planes (qpel), and 3 for epel yading@10: */ yading@10: static int mc_subpel(DiracContext *s, DiracBlock *block, const uint8_t *src[5], yading@10: int x, int y, int ref, int plane) yading@10: { yading@10: Plane *p = &s->plane[plane]; yading@10: uint8_t **ref_hpel = s->ref_pics[ref]->hpel[plane]; yading@10: int motion_x = block->u.mv[ref][0]; yading@10: int motion_y = block->u.mv[ref][1]; yading@10: int mx, my, i, epel, nplanes = 0; yading@10: yading@10: if (plane) { yading@10: motion_x >>= s->chroma_x_shift; yading@10: motion_y >>= s->chroma_y_shift; yading@10: } yading@10: yading@10: mx = motion_x & ~(-1 << s->mv_precision); yading@10: my = motion_y & ~(-1 << s->mv_precision); yading@10: motion_x >>= s->mv_precision; yading@10: motion_y >>= s->mv_precision; yading@10: /* normalize subpel coordinates to epel */ yading@10: /* TODO: template this function? */ yading@10: mx <<= 3 - s->mv_precision; yading@10: my <<= 3 - s->mv_precision; yading@10: yading@10: x += motion_x; yading@10: y += motion_y; yading@10: epel = (mx|my)&1; yading@10: yading@10: /* hpel position */ yading@10: if (!((mx|my)&3)) { yading@10: nplanes = 1; yading@10: src[0] = ref_hpel[(my>>1)+(mx>>2)] + y*p->stride + x; yading@10: } else { yading@10: /* qpel or epel */ yading@10: nplanes = 4; yading@10: for (i = 0; i < 4; i++) yading@10: src[i] = ref_hpel[i] + y*p->stride + x; yading@10: yading@10: /* if we're interpolating in the right/bottom halves, adjust the planes as needed yading@10: we increment x/y because the edge changes for half of the pixels */ yading@10: if (mx > 4) { yading@10: src[0] += 1; yading@10: src[2] += 1; yading@10: x++; yading@10: } yading@10: if (my > 4) { yading@10: src[0] += p->stride; yading@10: src[1] += p->stride; yading@10: y++; yading@10: } yading@10: yading@10: /* hpel planes are: yading@10: [0]: F [1]: H yading@10: [2]: V [3]: C */ yading@10: if (!epel) { yading@10: /* check if we really only need 2 planes since either mx or my is yading@10: a hpel position. (epel weights of 0 handle this there) */ yading@10: if (!(mx&3)) { yading@10: /* mx == 0: average [0] and [2] yading@10: mx == 4: average [1] and [3] */ yading@10: src[!mx] = src[2 + !!mx]; yading@10: nplanes = 2; yading@10: } else if (!(my&3)) { yading@10: src[0] = src[(my>>1) ]; yading@10: src[1] = src[(my>>1)+1]; yading@10: nplanes = 2; yading@10: } yading@10: } else { yading@10: /* adjust the ordering if needed so the weights work */ yading@10: if (mx > 4) { yading@10: FFSWAP(const uint8_t *, src[0], src[1]); yading@10: FFSWAP(const uint8_t *, src[2], src[3]); yading@10: } yading@10: if (my > 4) { yading@10: FFSWAP(const uint8_t *, src[0], src[2]); yading@10: FFSWAP(const uint8_t *, src[1], src[3]); yading@10: } yading@10: src[4] = epel_weights[my&3][mx&3]; yading@10: } yading@10: } yading@10: yading@10: /* fixme: v/h _edge_pos */ yading@10: if (x + p->xblen > p->width +EDGE_WIDTH/2 || yading@10: y + p->yblen > p->height+EDGE_WIDTH/2 || yading@10: x < 0 || y < 0) { yading@10: for (i = 0; i < nplanes; i++) { yading@10: ff_emulated_edge_mc(s->edge_emu_buffer[i], src[i], p->stride, yading@10: p->xblen, p->yblen, x, y, yading@10: p->width+EDGE_WIDTH/2, p->height+EDGE_WIDTH/2); yading@10: src[i] = s->edge_emu_buffer[i]; yading@10: } yading@10: } yading@10: return (nplanes>>1) + epel; yading@10: } yading@10: yading@10: static void add_dc(uint16_t *dst, int dc, int stride, yading@10: uint8_t *obmc_weight, int xblen, int yblen) yading@10: { yading@10: int x, y; yading@10: dc += 128; yading@10: yading@10: for (y = 0; y < yblen; y++) { yading@10: for (x = 0; x < xblen; x += 2) { yading@10: dst[x ] += dc * obmc_weight[x ]; yading@10: dst[x+1] += dc * obmc_weight[x+1]; yading@10: } yading@10: dst += stride; yading@10: obmc_weight += MAX_BLOCKSIZE; yading@10: } yading@10: } yading@10: yading@10: static void block_mc(DiracContext *s, DiracBlock *block, yading@10: uint16_t *mctmp, uint8_t *obmc_weight, yading@10: int plane, int dstx, int dsty) yading@10: { yading@10: Plane *p = &s->plane[plane]; yading@10: const uint8_t *src[5]; yading@10: int idx; yading@10: yading@10: switch (block->ref&3) { yading@10: case 0: /* DC */ yading@10: add_dc(mctmp, block->u.dc[plane], p->stride, obmc_weight, p->xblen, p->yblen); yading@10: return; yading@10: case 1: yading@10: case 2: yading@10: idx = mc_subpel(s, block, src, dstx, dsty, (block->ref&3)-1, plane); yading@10: s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen); yading@10: if (s->weight_func) yading@10: s->weight_func(s->mcscratch, p->stride, s->weight_log2denom, yading@10: s->weight[0] + s->weight[1], p->yblen); yading@10: break; yading@10: case 3: yading@10: idx = mc_subpel(s, block, src, dstx, dsty, 0, plane); yading@10: s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen); yading@10: idx = mc_subpel(s, block, src, dstx, dsty, 1, plane); yading@10: if (s->biweight_func) { yading@10: /* fixme: +32 is a quick hack */ yading@10: s->put_pixels_tab[idx](s->mcscratch + 32, src, p->stride, p->yblen); yading@10: s->biweight_func(s->mcscratch, s->mcscratch+32, p->stride, s->weight_log2denom, yading@10: s->weight[0], s->weight[1], p->yblen); yading@10: } else yading@10: s->avg_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen); yading@10: break; yading@10: } yading@10: s->add_obmc(mctmp, s->mcscratch, p->stride, obmc_weight, p->yblen); yading@10: } yading@10: yading@10: static void mc_row(DiracContext *s, DiracBlock *block, uint16_t *mctmp, int plane, int dsty) yading@10: { yading@10: Plane *p = &s->plane[plane]; yading@10: int x, dstx = p->xbsep - p->xoffset; yading@10: yading@10: block_mc(s, block, mctmp, s->obmc_weight[0], plane, -p->xoffset, dsty); yading@10: mctmp += p->xbsep; yading@10: yading@10: for (x = 1; x < s->blwidth-1; x++) { yading@10: block_mc(s, block+x, mctmp, s->obmc_weight[1], plane, dstx, dsty); yading@10: dstx += p->xbsep; yading@10: mctmp += p->xbsep; yading@10: } yading@10: block_mc(s, block+x, mctmp, s->obmc_weight[2], plane, dstx, dsty); yading@10: } yading@10: yading@10: static void select_dsp_funcs(DiracContext *s, int width, int height, int xblen, int yblen) yading@10: { yading@10: int idx = 0; yading@10: if (xblen > 8) yading@10: idx = 1; yading@10: if (xblen > 16) yading@10: idx = 2; yading@10: yading@10: memcpy(s->put_pixels_tab, s->diracdsp.put_dirac_pixels_tab[idx], sizeof(s->put_pixels_tab)); yading@10: memcpy(s->avg_pixels_tab, s->diracdsp.avg_dirac_pixels_tab[idx], sizeof(s->avg_pixels_tab)); yading@10: s->add_obmc = s->diracdsp.add_dirac_obmc[idx]; yading@10: if (s->weight_log2denom > 1 || s->weight[0] != 1 || s->weight[1] != 1) { yading@10: s->weight_func = s->diracdsp.weight_dirac_pixels_tab[idx]; yading@10: s->biweight_func = s->diracdsp.biweight_dirac_pixels_tab[idx]; yading@10: } else { yading@10: s->weight_func = NULL; yading@10: s->biweight_func = NULL; yading@10: } yading@10: } yading@10: yading@10: static void interpolate_refplane(DiracContext *s, DiracFrame *ref, int plane, int width, int height) yading@10: { yading@10: /* chroma allocates an edge of 8 when subsampled yading@10: which for 4:2:2 means an h edge of 16 and v edge of 8 yading@10: just use 8 for everything for the moment */ yading@10: int i, edge = EDGE_WIDTH/2; yading@10: yading@10: ref->hpel[plane][0] = ref->avframe.data[plane]; yading@10: s->dsp.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 */ yading@10: yading@10: /* no need for hpel if we only have fpel vectors */ yading@10: if (!s->mv_precision) yading@10: return; yading@10: yading@10: for (i = 1; i < 4; i++) { yading@10: if (!ref->hpel_base[plane][i]) yading@10: ref->hpel_base[plane][i] = av_malloc((height+2*edge) * ref->avframe.linesize[plane] + 32); yading@10: /* we need to be 16-byte aligned even for chroma */ yading@10: ref->hpel[plane][i] = ref->hpel_base[plane][i] + edge*ref->avframe.linesize[plane] + 16; yading@10: } yading@10: yading@10: if (!ref->interpolated[plane]) { yading@10: s->diracdsp.dirac_hpel_filter(ref->hpel[plane][1], ref->hpel[plane][2], yading@10: ref->hpel[plane][3], ref->hpel[plane][0], yading@10: ref->avframe.linesize[plane], width, height); yading@10: s->dsp.draw_edges(ref->hpel[plane][1], ref->avframe.linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM); yading@10: s->dsp.draw_edges(ref->hpel[plane][2], ref->avframe.linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM); yading@10: s->dsp.draw_edges(ref->hpel[plane][3], ref->avframe.linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM); yading@10: } yading@10: ref->interpolated[plane] = 1; yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 13.0 Transform data syntax. transform_data() yading@10: */ yading@10: static int dirac_decode_frame_internal(DiracContext *s) yading@10: { yading@10: DWTContext d; yading@10: int y, i, comp, dsty; yading@10: yading@10: if (s->low_delay) { yading@10: /* [DIRAC_STD] 13.5.1 low_delay_transform_data() */ yading@10: for (comp = 0; comp < 3; comp++) { yading@10: Plane *p = &s->plane[comp]; yading@10: memset(p->idwt_buf, 0, p->idwt_stride * p->idwt_height * sizeof(IDWTELEM)); yading@10: } yading@10: if (!s->zero_res) yading@10: decode_lowdelay(s); yading@10: } yading@10: yading@10: for (comp = 0; comp < 3; comp++) { yading@10: Plane *p = &s->plane[comp]; yading@10: uint8_t *frame = s->current_picture->avframe.data[comp]; yading@10: yading@10: /* FIXME: small resolutions */ yading@10: for (i = 0; i < 4; i++) yading@10: s->edge_emu_buffer[i] = s->edge_emu_buffer_base + i*FFALIGN(p->width, 16); yading@10: yading@10: if (!s->zero_res && !s->low_delay) yading@10: { yading@10: memset(p->idwt_buf, 0, p->idwt_stride * p->idwt_height * sizeof(IDWTELEM)); yading@10: decode_component(s, comp); /* [DIRAC_STD] 13.4.1 core_transform_data() */ yading@10: } yading@10: if (ff_spatial_idwt_init2(&d, p->idwt_buf, p->idwt_width, p->idwt_height, p->idwt_stride, yading@10: s->wavelet_idx+2, s->wavelet_depth, p->idwt_tmp)) yading@10: return -1; yading@10: yading@10: if (!s->num_refs) { /* intra */ yading@10: for (y = 0; y < p->height; y += 16) { yading@10: ff_spatial_idwt_slice2(&d, y+16); /* decode */ yading@10: s->diracdsp.put_signed_rect_clamped(frame + y*p->stride, p->stride, yading@10: p->idwt_buf + y*p->idwt_stride, p->idwt_stride, p->width, 16); yading@10: } yading@10: } else { /* inter */ yading@10: int rowheight = p->ybsep*p->stride; yading@10: yading@10: select_dsp_funcs(s, p->width, p->height, p->xblen, p->yblen); yading@10: yading@10: for (i = 0; i < s->num_refs; i++) yading@10: interpolate_refplane(s, s->ref_pics[i], comp, p->width, p->height); yading@10: yading@10: memset(s->mctmp, 0, 4*p->yoffset*p->stride); yading@10: yading@10: dsty = -p->yoffset; yading@10: for (y = 0; y < s->blheight; y++) { yading@10: int h = 0, yading@10: start = FFMAX(dsty, 0); yading@10: uint16_t *mctmp = s->mctmp + y*rowheight; yading@10: DiracBlock *blocks = s->blmotion + y*s->blwidth; yading@10: yading@10: init_obmc_weights(s, p, y); yading@10: yading@10: if (y == s->blheight-1 || start+p->ybsep > p->height) yading@10: h = p->height - start; yading@10: else yading@10: h = p->ybsep - (start - dsty); yading@10: if (h < 0) yading@10: break; yading@10: yading@10: memset(mctmp+2*p->yoffset*p->stride, 0, 2*rowheight); yading@10: mc_row(s, blocks, mctmp, comp, dsty); yading@10: yading@10: mctmp += (start - dsty)*p->stride + p->xoffset; yading@10: ff_spatial_idwt_slice2(&d, start + h); /* decode */ yading@10: s->diracdsp.add_rect_clamped(frame + start*p->stride, mctmp, p->stride, yading@10: p->idwt_buf + start*p->idwt_stride, p->idwt_stride, p->width, h); yading@10: yading@10: dsty += p->ybsep; yading@10: } yading@10: } yading@10: } yading@10: yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 11.1.1 Picture Header. picture_header() yading@10: */ yading@10: static int dirac_decode_picture_header(DiracContext *s) yading@10: { yading@10: int retire, picnum; yading@10: int i, j, refnum, refdist; yading@10: GetBitContext *gb = &s->gb; yading@10: yading@10: /* [DIRAC_STD] 11.1.1 Picture Header. picture_header() PICTURE_NUM */ yading@10: picnum = s->current_picture->avframe.display_picture_number = get_bits_long(gb, 32); yading@10: yading@10: yading@10: av_log(s->avctx,AV_LOG_DEBUG,"PICTURE_NUM: %d\n",picnum); yading@10: yading@10: /* if this is the first keyframe after a sequence header, start our yading@10: reordering from here */ yading@10: if (s->frame_number < 0) yading@10: s->frame_number = picnum; yading@10: yading@10: s->ref_pics[0] = s->ref_pics[1] = NULL; yading@10: for (i = 0; i < s->num_refs; i++) { yading@10: refnum = picnum + dirac_get_se_golomb(gb); yading@10: refdist = INT_MAX; yading@10: yading@10: /* find the closest reference to the one we want */ yading@10: /* Jordi: this is needed if the referenced picture hasn't yet arrived */ yading@10: for (j = 0; j < MAX_REFERENCE_FRAMES && refdist; j++) yading@10: if (s->ref_frames[j] yading@10: && FFABS(s->ref_frames[j]->avframe.display_picture_number - refnum) < refdist) { yading@10: s->ref_pics[i] = s->ref_frames[j]; yading@10: refdist = FFABS(s->ref_frames[j]->avframe.display_picture_number - refnum); yading@10: } yading@10: yading@10: if (!s->ref_pics[i] || refdist) yading@10: av_log(s->avctx, AV_LOG_DEBUG, "Reference not found\n"); yading@10: yading@10: /* if there were no references at all, allocate one */ yading@10: if (!s->ref_pics[i]) yading@10: for (j = 0; j < MAX_FRAMES; j++) yading@10: if (!s->all_frames[j].avframe.data[0]) { yading@10: s->ref_pics[i] = &s->all_frames[j]; yading@10: ff_get_buffer(s->avctx, &s->ref_pics[i]->avframe, AV_GET_BUFFER_FLAG_REF); yading@10: break; yading@10: } yading@10: } yading@10: yading@10: /* retire the reference frames that are not used anymore */ yading@10: if (s->current_picture->avframe.reference) { yading@10: retire = picnum + dirac_get_se_golomb(gb); yading@10: if (retire != picnum) { yading@10: DiracFrame *retire_pic = remove_frame(s->ref_frames, retire); yading@10: yading@10: if (retire_pic) yading@10: retire_pic->avframe.reference &= DELAYED_PIC_REF; yading@10: else yading@10: av_log(s->avctx, AV_LOG_DEBUG, "Frame to retire not found\n"); yading@10: } yading@10: yading@10: /* if reference array is full, remove the oldest as per the spec */ yading@10: while (add_frame(s->ref_frames, MAX_REFERENCE_FRAMES, s->current_picture)) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Reference frame overflow\n"); yading@10: remove_frame(s->ref_frames, s->ref_frames[0]->avframe.display_picture_number)->avframe.reference &= DELAYED_PIC_REF; yading@10: } yading@10: } yading@10: yading@10: if (s->num_refs) { yading@10: if (dirac_unpack_prediction_parameters(s)) /* [DIRAC_STD] 11.2 Picture Prediction Data. picture_prediction() */ yading@10: return -1; yading@10: if (dirac_unpack_block_motion_data(s)) /* [DIRAC_STD] 12. Block motion data syntax */ yading@10: return -1; yading@10: } yading@10: if (dirac_unpack_idwt_params(s)) /* [DIRAC_STD] 11.3 Wavelet transform data */ yading@10: return -1; yading@10: yading@10: init_planes(s); yading@10: return 0; yading@10: } yading@10: yading@10: static int get_delayed_pic(DiracContext *s, AVFrame *picture, int *got_frame) yading@10: { yading@10: DiracFrame *out = s->delay_frames[0]; yading@10: int i, out_idx = 0; yading@10: int ret; yading@10: yading@10: /* find frame with lowest picture number */ yading@10: for (i = 1; s->delay_frames[i]; i++) yading@10: if (s->delay_frames[i]->avframe.display_picture_number < out->avframe.display_picture_number) { yading@10: out = s->delay_frames[i]; yading@10: out_idx = i; yading@10: } yading@10: yading@10: for (i = out_idx; s->delay_frames[i]; i++) yading@10: s->delay_frames[i] = s->delay_frames[i+1]; yading@10: yading@10: if (out) { yading@10: out->avframe.reference ^= DELAYED_PIC_REF; yading@10: *got_frame = 1; yading@10: if((ret = av_frame_ref(picture, &out->avframe)) < 0) yading@10: return ret; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Dirac Specification -> yading@10: * 9.6 Parse Info Header Syntax. parse_info() yading@10: * 4 byte start code + byte parse code + 4 byte size + 4 byte previous size yading@10: */ yading@10: #define DATA_UNIT_HEADER_SIZE 13 yading@10: yading@10: /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3 yading@10: inside the function parse_sequence() */ yading@10: static int dirac_decode_data_unit(AVCodecContext *avctx, const uint8_t *buf, int size) yading@10: { yading@10: DiracContext *s = avctx->priv_data; yading@10: DiracFrame *pic = NULL; yading@10: int ret, i, parse_code = buf[4]; yading@10: unsigned tmp; yading@10: yading@10: if (size < DATA_UNIT_HEADER_SIZE) yading@10: return -1; yading@10: yading@10: init_get_bits(&s->gb, &buf[13], 8*(size - DATA_UNIT_HEADER_SIZE)); yading@10: yading@10: if (parse_code == pc_seq_header) { yading@10: if (s->seen_sequence_header) yading@10: return 0; yading@10: yading@10: /* [DIRAC_STD] 10. Sequence header */ yading@10: if (avpriv_dirac_parse_sequence_header(avctx, &s->gb, &s->source)) yading@10: return -1; yading@10: yading@10: avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift); yading@10: yading@10: if (alloc_sequence_buffers(s)) yading@10: return -1; yading@10: yading@10: s->seen_sequence_header = 1; yading@10: } else if (parse_code == pc_eos) { /* [DIRAC_STD] End of Sequence */ yading@10: free_sequence_buffers(s); yading@10: s->seen_sequence_header = 0; yading@10: } else if (parse_code == pc_aux_data) { yading@10: if (buf[13] == 1) { /* encoder implementation/version */ yading@10: int ver[3]; yading@10: /* versions older than 1.0.8 don't store quant delta for yading@10: subbands with only one codeblock */ yading@10: if (sscanf(buf+14, "Schroedinger %d.%d.%d", ver, ver+1, ver+2) == 3) yading@10: if (ver[0] == 1 && ver[1] == 0 && ver[2] <= 7) yading@10: s->old_delta_quant = 1; yading@10: } yading@10: } else if (parse_code & 0x8) { /* picture data unit */ yading@10: if (!s->seen_sequence_header) { yading@10: av_log(avctx, AV_LOG_DEBUG, "Dropping frame without sequence header\n"); yading@10: return -1; yading@10: } yading@10: yading@10: /* find an unused frame */ yading@10: for (i = 0; i < MAX_FRAMES; i++) yading@10: if (s->all_frames[i].avframe.data[0] == NULL) yading@10: pic = &s->all_frames[i]; yading@10: if (!pic) { yading@10: av_log(avctx, AV_LOG_ERROR, "framelist full\n"); yading@10: return -1; yading@10: } yading@10: yading@10: avcodec_get_frame_defaults(&pic->avframe); yading@10: yading@10: /* [DIRAC_STD] Defined in 9.6.1 ... */ yading@10: tmp = parse_code & 0x03; /* [DIRAC_STD] num_refs() */ yading@10: if (tmp > 2) { yading@10: av_log(avctx, AV_LOG_ERROR, "num_refs of 3\n"); yading@10: return -1; yading@10: } yading@10: s->num_refs = tmp; yading@10: s->is_arith = (parse_code & 0x48) == 0x08; /* [DIRAC_STD] using_ac() */ yading@10: s->low_delay = (parse_code & 0x88) == 0x88; /* [DIRAC_STD] is_low_delay() */ yading@10: pic->avframe.reference = (parse_code & 0x0C) == 0x0C; /* [DIRAC_STD] is_reference() */ yading@10: pic->avframe.key_frame = s->num_refs == 0; /* [DIRAC_STD] is_intra() */ yading@10: pic->avframe.pict_type = s->num_refs + 1; /* Definition of AVPictureType in avutil.h */ yading@10: yading@10: if ((ret = ff_get_buffer(avctx, &pic->avframe, (parse_code & 0x0C) == 0x0C ? AV_GET_BUFFER_FLAG_REF : 0)) < 0) yading@10: return ret; yading@10: s->current_picture = pic; yading@10: s->plane[0].stride = pic->avframe.linesize[0]; yading@10: s->plane[1].stride = pic->avframe.linesize[1]; yading@10: s->plane[2].stride = pic->avframe.linesize[2]; yading@10: yading@10: /* [DIRAC_STD] 11.1 Picture parse. picture_parse() */ yading@10: if (dirac_decode_picture_header(s)) yading@10: return -1; yading@10: yading@10: /* [DIRAC_STD] 13.0 Transform data syntax. transform_data() */ yading@10: if (dirac_decode_frame_internal(s)) yading@10: return -1; yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: static int dirac_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPacket *pkt) yading@10: { yading@10: DiracContext *s = avctx->priv_data; yading@10: DiracFrame *picture = data; yading@10: uint8_t *buf = pkt->data; yading@10: int buf_size = pkt->size; yading@10: int i, data_unit_size, buf_idx = 0; yading@10: int ret; yading@10: yading@10: /* release unused frames */ yading@10: for (i = 0; i < MAX_FRAMES; i++) yading@10: if (s->all_frames[i].avframe.data[0] && !s->all_frames[i].avframe.reference) { yading@10: av_frame_unref(&s->all_frames[i].avframe); yading@10: memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated)); yading@10: } yading@10: yading@10: s->current_picture = NULL; yading@10: *got_frame = 0; yading@10: yading@10: /* end of stream, so flush delayed pics */ yading@10: if (buf_size == 0) yading@10: return get_delayed_pic(s, (AVFrame *)data, got_frame); yading@10: yading@10: for (;;) { yading@10: /*[DIRAC_STD] Here starts the code from parse_info() defined in 9.6 yading@10: [DIRAC_STD] PARSE_INFO_PREFIX = "BBCD" as defined in ISO/IEC 646 yading@10: BBCD start code search */ yading@10: for (; buf_idx + DATA_UNIT_HEADER_SIZE < buf_size; buf_idx++) { yading@10: if (buf[buf_idx ] == 'B' && buf[buf_idx+1] == 'B' && yading@10: buf[buf_idx+2] == 'C' && buf[buf_idx+3] == 'D') yading@10: break; yading@10: } yading@10: /* BBCD found or end of data */ yading@10: if (buf_idx + DATA_UNIT_HEADER_SIZE >= buf_size) yading@10: break; yading@10: yading@10: data_unit_size = AV_RB32(buf+buf_idx+5); yading@10: if (buf_idx + data_unit_size > buf_size || !data_unit_size) { yading@10: if(buf_idx + data_unit_size > buf_size) yading@10: av_log(s->avctx, AV_LOG_ERROR, yading@10: "Data unit with size %d is larger than input buffer, discarding\n", yading@10: data_unit_size); yading@10: buf_idx += 4; yading@10: continue; yading@10: } yading@10: /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3 inside the function parse_sequence() */ yading@10: if (dirac_decode_data_unit(avctx, buf+buf_idx, data_unit_size)) yading@10: { yading@10: av_log(s->avctx, AV_LOG_ERROR,"Error in dirac_decode_data_unit\n"); yading@10: return -1; yading@10: } yading@10: buf_idx += data_unit_size; yading@10: } yading@10: yading@10: if (!s->current_picture) yading@10: return buf_size; yading@10: yading@10: if (s->current_picture->avframe.display_picture_number > s->frame_number) { yading@10: DiracFrame *delayed_frame = remove_frame(s->delay_frames, s->frame_number); yading@10: yading@10: s->current_picture->avframe.reference |= DELAYED_PIC_REF; yading@10: yading@10: if (add_frame(s->delay_frames, MAX_DELAY, s->current_picture)) { yading@10: int min_num = s->delay_frames[0]->avframe.display_picture_number; yading@10: /* Too many delayed frames, so we display the frame with the lowest pts */ yading@10: av_log(avctx, AV_LOG_ERROR, "Delay frame overflow\n"); yading@10: delayed_frame = s->delay_frames[0]; yading@10: yading@10: for (i = 1; s->delay_frames[i]; i++) yading@10: if (s->delay_frames[i]->avframe.display_picture_number < min_num) yading@10: min_num = s->delay_frames[i]->avframe.display_picture_number; yading@10: yading@10: delayed_frame = remove_frame(s->delay_frames, min_num); yading@10: add_frame(s->delay_frames, MAX_DELAY, s->current_picture); yading@10: } yading@10: yading@10: if (delayed_frame) { yading@10: delayed_frame->avframe.reference ^= DELAYED_PIC_REF; yading@10: if((ret=av_frame_ref(data, &delayed_frame->avframe)) < 0) yading@10: return ret; yading@10: *got_frame = 1; yading@10: } yading@10: } else if (s->current_picture->avframe.display_picture_number == s->frame_number) { yading@10: /* The right frame at the right time :-) */ yading@10: if((ret=av_frame_ref(data, &s->current_picture->avframe)) < 0) yading@10: return ret; yading@10: *got_frame = 1; yading@10: } yading@10: yading@10: if (*got_frame) yading@10: s->frame_number = picture->avframe.display_picture_number + 1; yading@10: yading@10: return buf_idx; yading@10: } yading@10: yading@10: AVCodec ff_dirac_decoder = { yading@10: .name = "dirac", yading@10: .type = AVMEDIA_TYPE_VIDEO, yading@10: .id = AV_CODEC_ID_DIRAC, yading@10: .priv_data_size = sizeof(DiracContext), yading@10: .init = dirac_decode_init, yading@10: .close = dirac_decode_end, yading@10: .decode = dirac_decode_frame, yading@10: .capabilities = CODEC_CAP_DELAY, yading@10: .flush = dirac_decode_flush, yading@10: .long_name = NULL_IF_CONFIG_SMALL("BBC Dirac VC-2"), yading@10: };