yading@10: /* yading@10: * Copyright (C) 2003-2004 the ffmpeg project 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: * On2 VP3 Video Decoder yading@10: * yading@10: * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx) yading@10: * For more information about the VP3 coding process, visit: yading@10: * http://wiki.multimedia.cx/index.php?title=On2_VP3 yading@10: * yading@10: * Theora decoder by Alex Beregszaszi yading@10: */ yading@10: yading@10: #include yading@10: #include yading@10: #include yading@10: yading@10: #include "libavutil/imgutils.h" yading@10: #include "avcodec.h" yading@10: #include "internal.h" yading@10: #include "dsputil.h" yading@10: #include "get_bits.h" yading@10: #include "hpeldsp.h" yading@10: #include "videodsp.h" yading@10: #include "vp3data.h" yading@10: #include "vp3dsp.h" yading@10: #include "xiph.h" yading@10: #include "thread.h" yading@10: yading@10: #define FRAGMENT_PIXELS 8 yading@10: yading@10: //FIXME split things out into their own arrays yading@10: typedef struct Vp3Fragment { yading@10: int16_t dc; yading@10: uint8_t coding_method; yading@10: uint8_t qpi; yading@10: } Vp3Fragment; yading@10: yading@10: #define SB_NOT_CODED 0 yading@10: #define SB_PARTIALLY_CODED 1 yading@10: #define SB_FULLY_CODED 2 yading@10: yading@10: // This is the maximum length of a single long bit run that can be encoded yading@10: // for superblock coding or block qps. Theora special-cases this to read a yading@10: // bit instead of flipping the current bit to allow for runs longer than 4129. yading@10: #define MAXIMUM_LONG_BIT_RUN 4129 yading@10: yading@10: #define MODE_INTER_NO_MV 0 yading@10: #define MODE_INTRA 1 yading@10: #define MODE_INTER_PLUS_MV 2 yading@10: #define MODE_INTER_LAST_MV 3 yading@10: #define MODE_INTER_PRIOR_LAST 4 yading@10: #define MODE_USING_GOLDEN 5 yading@10: #define MODE_GOLDEN_MV 6 yading@10: #define MODE_INTER_FOURMV 7 yading@10: #define CODING_MODE_COUNT 8 yading@10: yading@10: /* special internal mode */ yading@10: #define MODE_COPY 8 yading@10: yading@10: static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb); yading@10: static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb); yading@10: yading@10: yading@10: /* There are 6 preset schemes, plus a free-form scheme */ yading@10: static const int ModeAlphabet[6][CODING_MODE_COUNT] = yading@10: { yading@10: /* scheme 1: Last motion vector dominates */ yading@10: { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, yading@10: MODE_INTER_PLUS_MV, MODE_INTER_NO_MV, yading@10: MODE_INTRA, MODE_USING_GOLDEN, yading@10: MODE_GOLDEN_MV, MODE_INTER_FOURMV }, yading@10: yading@10: /* scheme 2 */ yading@10: { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, yading@10: MODE_INTER_NO_MV, MODE_INTER_PLUS_MV, yading@10: MODE_INTRA, MODE_USING_GOLDEN, yading@10: MODE_GOLDEN_MV, MODE_INTER_FOURMV }, yading@10: yading@10: /* scheme 3 */ yading@10: { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV, yading@10: MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV, yading@10: MODE_INTRA, MODE_USING_GOLDEN, yading@10: MODE_GOLDEN_MV, MODE_INTER_FOURMV }, yading@10: yading@10: /* scheme 4 */ yading@10: { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV, yading@10: MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST, yading@10: MODE_INTRA, MODE_USING_GOLDEN, yading@10: MODE_GOLDEN_MV, MODE_INTER_FOURMV }, yading@10: yading@10: /* scheme 5: No motion vector dominates */ yading@10: { MODE_INTER_NO_MV, MODE_INTER_LAST_MV, yading@10: MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV, yading@10: MODE_INTRA, MODE_USING_GOLDEN, yading@10: MODE_GOLDEN_MV, MODE_INTER_FOURMV }, yading@10: yading@10: /* scheme 6 */ yading@10: { MODE_INTER_NO_MV, MODE_USING_GOLDEN, yading@10: MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, yading@10: MODE_INTER_PLUS_MV, MODE_INTRA, yading@10: MODE_GOLDEN_MV, MODE_INTER_FOURMV }, yading@10: yading@10: }; yading@10: yading@10: static const uint8_t hilbert_offset[16][2] = { yading@10: {0,0}, {1,0}, {1,1}, {0,1}, yading@10: {0,2}, {0,3}, {1,3}, {1,2}, yading@10: {2,2}, {2,3}, {3,3}, {3,2}, yading@10: {3,1}, {2,1}, {2,0}, {3,0} yading@10: }; yading@10: yading@10: #define MIN_DEQUANT_VAL 2 yading@10: yading@10: typedef struct Vp3DecodeContext { yading@10: AVCodecContext *avctx; yading@10: int theora, theora_tables; yading@10: int version; yading@10: int width, height; yading@10: int chroma_x_shift, chroma_y_shift; yading@10: ThreadFrame golden_frame; yading@10: ThreadFrame last_frame; yading@10: ThreadFrame current_frame; yading@10: int keyframe; yading@10: uint8_t idct_permutation[64]; yading@10: uint8_t idct_scantable[64]; yading@10: HpelDSPContext hdsp; yading@10: VideoDSPContext vdsp; yading@10: VP3DSPContext vp3dsp; yading@10: DECLARE_ALIGNED(16, int16_t, block)[64]; yading@10: int flipped_image; yading@10: int last_slice_end; yading@10: int skip_loop_filter; yading@10: yading@10: int qps[3]; yading@10: int nqps; yading@10: int last_qps[3]; yading@10: yading@10: int superblock_count; yading@10: int y_superblock_width; yading@10: int y_superblock_height; yading@10: int y_superblock_count; yading@10: int c_superblock_width; yading@10: int c_superblock_height; yading@10: int c_superblock_count; yading@10: int u_superblock_start; yading@10: int v_superblock_start; yading@10: unsigned char *superblock_coding; yading@10: yading@10: int macroblock_count; yading@10: int macroblock_width; yading@10: int macroblock_height; yading@10: yading@10: int fragment_count; yading@10: int fragment_width[2]; yading@10: int fragment_height[2]; yading@10: yading@10: Vp3Fragment *all_fragments; yading@10: int fragment_start[3]; yading@10: int data_offset[3]; yading@10: yading@10: int8_t (*motion_val[2])[2]; yading@10: yading@10: /* tables */ yading@10: uint16_t coded_dc_scale_factor[64]; yading@10: uint32_t coded_ac_scale_factor[64]; yading@10: uint8_t base_matrix[384][64]; yading@10: uint8_t qr_count[2][3]; yading@10: uint8_t qr_size [2][3][64]; yading@10: uint16_t qr_base[2][3][64]; yading@10: yading@10: /** yading@10: * This is a list of all tokens in bitstream order. Reordering takes place yading@10: * by pulling from each level during IDCT. As a consequence, IDCT must be yading@10: * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32 yading@10: * otherwise. The 32 different tokens with up to 12 bits of extradata are yading@10: * collapsed into 3 types, packed as follows: yading@10: * (from the low to high bits) yading@10: * yading@10: * 2 bits: type (0,1,2) yading@10: * 0: EOB run, 14 bits for run length (12 needed) yading@10: * 1: zero run, 7 bits for run length yading@10: * 7 bits for the next coefficient (3 needed) yading@10: * 2: coefficient, 14 bits (11 needed) yading@10: * yading@10: * Coefficients are signed, so are packed in the highest bits for automatic yading@10: * sign extension. yading@10: */ yading@10: int16_t *dct_tokens[3][64]; yading@10: int16_t *dct_tokens_base; yading@10: #define TOKEN_EOB(eob_run) ((eob_run) << 2) yading@10: #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1) yading@10: #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2) yading@10: yading@10: /** yading@10: * number of blocks that contain DCT coefficients at the given level or higher yading@10: */ yading@10: int num_coded_frags[3][64]; yading@10: int total_num_coded_frags; yading@10: yading@10: /* this is a list of indexes into the all_fragments array indicating yading@10: * which of the fragments are coded */ yading@10: int *coded_fragment_list[3]; yading@10: yading@10: VLC dc_vlc[16]; yading@10: VLC ac_vlc_1[16]; yading@10: VLC ac_vlc_2[16]; yading@10: VLC ac_vlc_3[16]; yading@10: VLC ac_vlc_4[16]; yading@10: yading@10: VLC superblock_run_length_vlc; yading@10: VLC fragment_run_length_vlc; yading@10: VLC mode_code_vlc; yading@10: VLC motion_vector_vlc; yading@10: yading@10: /* these arrays need to be on 16-byte boundaries since SSE2 operations yading@10: * index into them */ yading@10: DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane] yading@10: yading@10: /* This table contains superblock_count * 16 entries. Each set of 16 yading@10: * numbers corresponds to the fragment indexes 0..15 of the superblock. yading@10: * An entry will be -1 to indicate that no entry corresponds to that yading@10: * index. */ yading@10: int *superblock_fragments; yading@10: yading@10: /* This is an array that indicates how a particular macroblock yading@10: * is coded. */ yading@10: unsigned char *macroblock_coding; yading@10: yading@10: uint8_t *edge_emu_buffer; yading@10: yading@10: /* Huffman decode */ yading@10: int hti; yading@10: unsigned int hbits; yading@10: int entries; yading@10: int huff_code_size; yading@10: uint32_t huffman_table[80][32][2]; yading@10: yading@10: uint8_t filter_limit_values[64]; yading@10: DECLARE_ALIGNED(8, int, bounding_values_array)[256+2]; yading@10: } Vp3DecodeContext; yading@10: yading@10: /************************************************************************ yading@10: * VP3 specific functions yading@10: ************************************************************************/ yading@10: yading@10: static void vp3_decode_flush(AVCodecContext *avctx) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: yading@10: if (s->golden_frame.f) yading@10: ff_thread_release_buffer(avctx, &s->golden_frame); yading@10: if (s->last_frame.f) yading@10: ff_thread_release_buffer(avctx, &s->last_frame); yading@10: if (s->current_frame.f) yading@10: ff_thread_release_buffer(avctx, &s->current_frame); yading@10: } yading@10: yading@10: static av_cold int vp3_decode_end(AVCodecContext *avctx) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: int i; yading@10: yading@10: av_freep(&s->superblock_coding); yading@10: av_freep(&s->all_fragments); yading@10: av_freep(&s->coded_fragment_list[0]); yading@10: av_freep(&s->dct_tokens_base); yading@10: av_freep(&s->superblock_fragments); yading@10: av_freep(&s->macroblock_coding); yading@10: av_freep(&s->motion_val[0]); yading@10: av_freep(&s->motion_val[1]); yading@10: av_freep(&s->edge_emu_buffer); yading@10: yading@10: s->theora_tables = 0; yading@10: yading@10: /* release all frames */ yading@10: vp3_decode_flush(avctx); yading@10: av_frame_free(&s->current_frame.f); yading@10: av_frame_free(&s->last_frame.f); yading@10: av_frame_free(&s->golden_frame.f); yading@10: yading@10: if (avctx->internal->is_copy) yading@10: return 0; yading@10: yading@10: for (i = 0; i < 16; i++) { yading@10: ff_free_vlc(&s->dc_vlc[i]); yading@10: ff_free_vlc(&s->ac_vlc_1[i]); yading@10: ff_free_vlc(&s->ac_vlc_2[i]); yading@10: ff_free_vlc(&s->ac_vlc_3[i]); yading@10: ff_free_vlc(&s->ac_vlc_4[i]); yading@10: } yading@10: yading@10: ff_free_vlc(&s->superblock_run_length_vlc); yading@10: ff_free_vlc(&s->fragment_run_length_vlc); yading@10: ff_free_vlc(&s->mode_code_vlc); yading@10: ff_free_vlc(&s->motion_vector_vlc); yading@10: yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * This function sets up all of the various blocks mappings: yading@10: * superblocks <-> fragments, macroblocks <-> fragments, yading@10: * superblocks <-> macroblocks yading@10: * yading@10: * @return 0 is successful; returns 1 if *anything* went wrong. yading@10: */ yading@10: static int init_block_mapping(Vp3DecodeContext *s) yading@10: { yading@10: int sb_x, sb_y, plane; yading@10: int x, y, i, j = 0; yading@10: yading@10: for (plane = 0; plane < 3; plane++) { yading@10: int sb_width = plane ? s->c_superblock_width : s->y_superblock_width; yading@10: int sb_height = plane ? s->c_superblock_height : s->y_superblock_height; yading@10: int frag_width = s->fragment_width[!!plane]; yading@10: int frag_height = s->fragment_height[!!plane]; yading@10: yading@10: for (sb_y = 0; sb_y < sb_height; sb_y++) yading@10: for (sb_x = 0; sb_x < sb_width; sb_x++) yading@10: for (i = 0; i < 16; i++) { yading@10: x = 4*sb_x + hilbert_offset[i][0]; yading@10: y = 4*sb_y + hilbert_offset[i][1]; yading@10: yading@10: if (x < frag_width && y < frag_height) yading@10: s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x; yading@10: else yading@10: s->superblock_fragments[j++] = -1; yading@10: } yading@10: } yading@10: yading@10: return 0; /* successful path out */ yading@10: } yading@10: yading@10: /* yading@10: * This function sets up the dequantization tables used for a particular yading@10: * frame. yading@10: */ yading@10: static void init_dequantizer(Vp3DecodeContext *s, int qpi) yading@10: { yading@10: int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]]; yading@10: int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]]; yading@10: int i, plane, inter, qri, bmi, bmj, qistart; yading@10: yading@10: for(inter=0; inter<2; inter++){ yading@10: for(plane=0; plane<3; plane++){ yading@10: int sum=0; yading@10: for(qri=0; qriqr_count[inter][plane]; qri++){ yading@10: sum+= s->qr_size[inter][plane][qri]; yading@10: if(s->qps[qpi] <= sum) yading@10: break; yading@10: } yading@10: qistart= sum - s->qr_size[inter][plane][qri]; yading@10: bmi= s->qr_base[inter][plane][qri ]; yading@10: bmj= s->qr_base[inter][plane][qri+1]; yading@10: for(i=0; i<64; i++){ yading@10: int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i] yading@10: - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i] yading@10: + s->qr_size[inter][plane][qri]) yading@10: / (2*s->qr_size[inter][plane][qri]); yading@10: yading@10: int qmin= 8<<(inter + !i); yading@10: int qscale= i ? ac_scale_factor : dc_scale_factor; yading@10: yading@10: s->qmat[qpi][inter][plane][s->idct_permutation[i]] = yading@10: av_clip((qscale * coeff) / 100 * 4, qmin, 4096); yading@10: } yading@10: // all DC coefficients use the same quant so as not to interfere with DC prediction yading@10: s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0]; yading@10: } yading@10: } yading@10: } yading@10: yading@10: /* yading@10: * This function initializes the loop filter boundary limits if the frame's yading@10: * quality index is different from the previous frame's. yading@10: * yading@10: * The filter_limit_values may not be larger than 127. yading@10: */ yading@10: static void init_loop_filter(Vp3DecodeContext *s) yading@10: { yading@10: int *bounding_values= s->bounding_values_array+127; yading@10: int filter_limit; yading@10: int x; yading@10: int value; yading@10: yading@10: filter_limit = s->filter_limit_values[s->qps[0]]; yading@10: av_assert0(filter_limit < 128U); yading@10: yading@10: /* set up the bounding values */ yading@10: memset(s->bounding_values_array, 0, 256 * sizeof(int)); yading@10: for (x = 0; x < filter_limit; x++) { yading@10: bounding_values[-x] = -x; yading@10: bounding_values[x] = x; yading@10: } yading@10: for (x = value = filter_limit; x < 128 && value; x++, value--) { yading@10: bounding_values[ x] = value; yading@10: bounding_values[-x] = -value; yading@10: } yading@10: if (value) yading@10: bounding_values[128] = value; yading@10: bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202; yading@10: } yading@10: yading@10: /* yading@10: * This function unpacks all of the superblock/macroblock/fragment coding yading@10: * information from the bitstream. yading@10: */ yading@10: static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) yading@10: { yading@10: int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start }; yading@10: int bit = 0; yading@10: int current_superblock = 0; yading@10: int current_run = 0; yading@10: int num_partial_superblocks = 0; yading@10: yading@10: int i, j; yading@10: int current_fragment; yading@10: int plane; yading@10: yading@10: if (s->keyframe) { yading@10: memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count); yading@10: yading@10: } else { yading@10: yading@10: /* unpack the list of partially-coded superblocks */ yading@10: bit = get_bits1(gb) ^ 1; yading@10: current_run = 0; yading@10: yading@10: while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) { yading@10: if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN) yading@10: bit = get_bits1(gb); yading@10: else yading@10: bit ^= 1; yading@10: yading@10: current_run = get_vlc2(gb, yading@10: s->superblock_run_length_vlc.table, 6, 2) + 1; yading@10: if (current_run == 34) yading@10: current_run += get_bits(gb, 12); yading@10: yading@10: if (current_superblock + current_run > s->superblock_count) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n"); yading@10: return -1; yading@10: } yading@10: yading@10: memset(s->superblock_coding + current_superblock, bit, current_run); yading@10: yading@10: current_superblock += current_run; yading@10: if (bit) yading@10: num_partial_superblocks += current_run; yading@10: } yading@10: yading@10: /* unpack the list of fully coded superblocks if any of the blocks were yading@10: * not marked as partially coded in the previous step */ yading@10: if (num_partial_superblocks < s->superblock_count) { yading@10: int superblocks_decoded = 0; yading@10: yading@10: current_superblock = 0; yading@10: bit = get_bits1(gb) ^ 1; yading@10: current_run = 0; yading@10: yading@10: while (superblocks_decoded < s->superblock_count - num_partial_superblocks yading@10: && get_bits_left(gb) > 0) { yading@10: yading@10: if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN) yading@10: bit = get_bits1(gb); yading@10: else yading@10: bit ^= 1; yading@10: yading@10: current_run = get_vlc2(gb, yading@10: s->superblock_run_length_vlc.table, 6, 2) + 1; yading@10: if (current_run == 34) yading@10: current_run += get_bits(gb, 12); yading@10: yading@10: for (j = 0; j < current_run; current_superblock++) { yading@10: if (current_superblock >= s->superblock_count) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n"); yading@10: return -1; yading@10: } yading@10: yading@10: /* skip any superblocks already marked as partially coded */ yading@10: if (s->superblock_coding[current_superblock] == SB_NOT_CODED) { yading@10: s->superblock_coding[current_superblock] = 2*bit; yading@10: j++; yading@10: } yading@10: } yading@10: superblocks_decoded += current_run; yading@10: } yading@10: } yading@10: yading@10: /* if there were partial blocks, initialize bitstream for yading@10: * unpacking fragment codings */ yading@10: if (num_partial_superblocks) { yading@10: yading@10: current_run = 0; yading@10: bit = get_bits1(gb); yading@10: /* toggle the bit because as soon as the first run length is yading@10: * fetched the bit will be toggled again */ yading@10: bit ^= 1; yading@10: } yading@10: } yading@10: yading@10: /* figure out which fragments are coded; iterate through each yading@10: * superblock (all planes) */ yading@10: s->total_num_coded_frags = 0; yading@10: memset(s->macroblock_coding, MODE_COPY, s->macroblock_count); yading@10: yading@10: for (plane = 0; plane < 3; plane++) { yading@10: int sb_start = superblock_starts[plane]; yading@10: int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count); yading@10: int num_coded_frags = 0; yading@10: yading@10: for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) { yading@10: yading@10: /* iterate through all 16 fragments in a superblock */ yading@10: for (j = 0; j < 16; j++) { yading@10: yading@10: /* if the fragment is in bounds, check its coding status */ yading@10: current_fragment = s->superblock_fragments[i * 16 + j]; yading@10: if (current_fragment != -1) { yading@10: int coded = s->superblock_coding[i]; yading@10: yading@10: if (s->superblock_coding[i] == SB_PARTIALLY_CODED) { yading@10: yading@10: /* fragment may or may not be coded; this is the case yading@10: * that cares about the fragment coding runs */ yading@10: if (current_run-- == 0) { yading@10: bit ^= 1; yading@10: current_run = get_vlc2(gb, yading@10: s->fragment_run_length_vlc.table, 5, 2); yading@10: } yading@10: coded = bit; yading@10: } yading@10: yading@10: if (coded) { yading@10: /* default mode; actual mode will be decoded in yading@10: * the next phase */ yading@10: s->all_fragments[current_fragment].coding_method = yading@10: MODE_INTER_NO_MV; yading@10: s->coded_fragment_list[plane][num_coded_frags++] = yading@10: current_fragment; yading@10: } else { yading@10: /* not coded; copy this fragment from the prior frame */ yading@10: s->all_fragments[current_fragment].coding_method = yading@10: MODE_COPY; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: s->total_num_coded_frags += num_coded_frags; yading@10: for (i = 0; i < 64; i++) yading@10: s->num_coded_frags[plane][i] = num_coded_frags; yading@10: if (plane < 2) yading@10: s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags; yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: /* yading@10: * This function unpacks all the coding mode data for individual macroblocks yading@10: * from the bitstream. yading@10: */ yading@10: static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb) yading@10: { yading@10: int i, j, k, sb_x, sb_y; yading@10: int scheme; yading@10: int current_macroblock; yading@10: int current_fragment; yading@10: int coding_mode; yading@10: int custom_mode_alphabet[CODING_MODE_COUNT]; yading@10: const int *alphabet; yading@10: Vp3Fragment *frag; yading@10: yading@10: if (s->keyframe) { yading@10: for (i = 0; i < s->fragment_count; i++) yading@10: s->all_fragments[i].coding_method = MODE_INTRA; yading@10: yading@10: } else { yading@10: yading@10: /* fetch the mode coding scheme for this frame */ yading@10: scheme = get_bits(gb, 3); yading@10: yading@10: /* is it a custom coding scheme? */ yading@10: if (scheme == 0) { yading@10: for (i = 0; i < 8; i++) yading@10: custom_mode_alphabet[i] = MODE_INTER_NO_MV; yading@10: for (i = 0; i < 8; i++) yading@10: custom_mode_alphabet[get_bits(gb, 3)] = i; yading@10: alphabet = custom_mode_alphabet; yading@10: } else yading@10: alphabet = ModeAlphabet[scheme-1]; yading@10: yading@10: /* iterate through all of the macroblocks that contain 1 or more yading@10: * coded fragments */ yading@10: for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) { yading@10: for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) { yading@10: if (get_bits_left(gb) <= 0) yading@10: return -1; yading@10: yading@10: for (j = 0; j < 4; j++) { yading@10: int mb_x = 2*sb_x + (j>>1); yading@10: int mb_y = 2*sb_y + (((j>>1)+j)&1); yading@10: current_macroblock = mb_y * s->macroblock_width + mb_x; yading@10: yading@10: if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height) yading@10: continue; yading@10: yading@10: #define BLOCK_X (2*mb_x + (k&1)) yading@10: #define BLOCK_Y (2*mb_y + (k>>1)) yading@10: /* coding modes are only stored if the macroblock has at least one yading@10: * luma block coded, otherwise it must be INTER_NO_MV */ yading@10: for (k = 0; k < 4; k++) { yading@10: current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X; yading@10: if (s->all_fragments[current_fragment].coding_method != MODE_COPY) yading@10: break; yading@10: } yading@10: if (k == 4) { yading@10: s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV; yading@10: continue; yading@10: } yading@10: yading@10: /* mode 7 means get 3 bits for each coding mode */ yading@10: if (scheme == 7) yading@10: coding_mode = get_bits(gb, 3); yading@10: else yading@10: coding_mode = alphabet yading@10: [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)]; yading@10: yading@10: s->macroblock_coding[current_macroblock] = coding_mode; yading@10: for (k = 0; k < 4; k++) { yading@10: frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X; yading@10: if (frag->coding_method != MODE_COPY) yading@10: frag->coding_method = coding_mode; yading@10: } yading@10: yading@10: #define SET_CHROMA_MODES \ yading@10: if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \ yading@10: frag[s->fragment_start[1]].coding_method = coding_mode;\ yading@10: if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \ yading@10: frag[s->fragment_start[2]].coding_method = coding_mode; yading@10: yading@10: if (s->chroma_y_shift) { yading@10: frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x; yading@10: SET_CHROMA_MODES yading@10: } else if (s->chroma_x_shift) { yading@10: frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x; yading@10: for (k = 0; k < 2; k++) { yading@10: SET_CHROMA_MODES yading@10: frag += s->fragment_width[1]; yading@10: } yading@10: } else { yading@10: for (k = 0; k < 4; k++) { yading@10: frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X; yading@10: SET_CHROMA_MODES yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /* yading@10: * This function unpacks all the motion vectors for the individual yading@10: * macroblocks from the bitstream. yading@10: */ yading@10: static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) yading@10: { yading@10: int j, k, sb_x, sb_y; yading@10: int coding_mode; yading@10: int motion_x[4]; yading@10: int motion_y[4]; yading@10: int last_motion_x = 0; yading@10: int last_motion_y = 0; yading@10: int prior_last_motion_x = 0; yading@10: int prior_last_motion_y = 0; yading@10: int current_macroblock; yading@10: int current_fragment; yading@10: int frag; yading@10: yading@10: if (s->keyframe) yading@10: return 0; yading@10: yading@10: /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */ yading@10: coding_mode = get_bits1(gb); yading@10: yading@10: /* iterate through all of the macroblocks that contain 1 or more yading@10: * coded fragments */ yading@10: for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) { yading@10: for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) { yading@10: if (get_bits_left(gb) <= 0) yading@10: return -1; yading@10: yading@10: for (j = 0; j < 4; j++) { yading@10: int mb_x = 2*sb_x + (j>>1); yading@10: int mb_y = 2*sb_y + (((j>>1)+j)&1); yading@10: current_macroblock = mb_y * s->macroblock_width + mb_x; yading@10: yading@10: if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height || yading@10: (s->macroblock_coding[current_macroblock] == MODE_COPY)) yading@10: continue; yading@10: yading@10: switch (s->macroblock_coding[current_macroblock]) { yading@10: yading@10: case MODE_INTER_PLUS_MV: yading@10: case MODE_GOLDEN_MV: yading@10: /* all 6 fragments use the same motion vector */ yading@10: if (coding_mode == 0) { yading@10: motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; yading@10: motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; yading@10: } else { yading@10: motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)]; yading@10: motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)]; yading@10: } yading@10: yading@10: /* vector maintenance, only on MODE_INTER_PLUS_MV */ yading@10: if (s->macroblock_coding[current_macroblock] == yading@10: MODE_INTER_PLUS_MV) { yading@10: prior_last_motion_x = last_motion_x; yading@10: prior_last_motion_y = last_motion_y; yading@10: last_motion_x = motion_x[0]; yading@10: last_motion_y = motion_y[0]; yading@10: } yading@10: break; yading@10: yading@10: case MODE_INTER_FOURMV: yading@10: /* vector maintenance */ yading@10: prior_last_motion_x = last_motion_x; yading@10: prior_last_motion_y = last_motion_y; yading@10: yading@10: /* fetch 4 vectors from the bitstream, one for each yading@10: * Y fragment, then average for the C fragment vectors */ yading@10: for (k = 0; k < 4; k++) { yading@10: current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X; yading@10: if (s->all_fragments[current_fragment].coding_method != MODE_COPY) { yading@10: if (coding_mode == 0) { yading@10: motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; yading@10: motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; yading@10: } else { yading@10: motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)]; yading@10: motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)]; yading@10: } yading@10: last_motion_x = motion_x[k]; yading@10: last_motion_y = motion_y[k]; yading@10: } else { yading@10: motion_x[k] = 0; yading@10: motion_y[k] = 0; yading@10: } yading@10: } yading@10: break; yading@10: yading@10: case MODE_INTER_LAST_MV: yading@10: /* all 6 fragments use the last motion vector */ yading@10: motion_x[0] = last_motion_x; yading@10: motion_y[0] = last_motion_y; yading@10: yading@10: /* no vector maintenance (last vector remains the yading@10: * last vector) */ yading@10: break; yading@10: yading@10: case MODE_INTER_PRIOR_LAST: yading@10: /* all 6 fragments use the motion vector prior to the yading@10: * last motion vector */ yading@10: motion_x[0] = prior_last_motion_x; yading@10: motion_y[0] = prior_last_motion_y; yading@10: yading@10: /* vector maintenance */ yading@10: prior_last_motion_x = last_motion_x; yading@10: prior_last_motion_y = last_motion_y; yading@10: last_motion_x = motion_x[0]; yading@10: last_motion_y = motion_y[0]; yading@10: break; yading@10: yading@10: default: yading@10: /* covers intra, inter without MV, golden without MV */ yading@10: motion_x[0] = 0; yading@10: motion_y[0] = 0; yading@10: yading@10: /* no vector maintenance */ yading@10: break; yading@10: } yading@10: yading@10: /* assign the motion vectors to the correct fragments */ yading@10: for (k = 0; k < 4; k++) { yading@10: current_fragment = yading@10: BLOCK_Y*s->fragment_width[0] + BLOCK_X; yading@10: if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { yading@10: s->motion_val[0][current_fragment][0] = motion_x[k]; yading@10: s->motion_val[0][current_fragment][1] = motion_y[k]; yading@10: } else { yading@10: s->motion_val[0][current_fragment][0] = motion_x[0]; yading@10: s->motion_val[0][current_fragment][1] = motion_y[0]; yading@10: } yading@10: } yading@10: yading@10: if (s->chroma_y_shift) { yading@10: if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { yading@10: motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2); yading@10: motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2); yading@10: } yading@10: motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1); yading@10: motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1); yading@10: frag = mb_y*s->fragment_width[1] + mb_x; yading@10: s->motion_val[1][frag][0] = motion_x[0]; yading@10: s->motion_val[1][frag][1] = motion_y[0]; yading@10: } else if (s->chroma_x_shift) { yading@10: if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { yading@10: motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1); yading@10: motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1); yading@10: motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1); yading@10: motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1); yading@10: } else { yading@10: motion_x[1] = motion_x[0]; yading@10: motion_y[1] = motion_y[0]; yading@10: } yading@10: motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1); yading@10: motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1); yading@10: yading@10: frag = 2*mb_y*s->fragment_width[1] + mb_x; yading@10: for (k = 0; k < 2; k++) { yading@10: s->motion_val[1][frag][0] = motion_x[k]; yading@10: s->motion_val[1][frag][1] = motion_y[k]; yading@10: frag += s->fragment_width[1]; yading@10: } yading@10: } else { yading@10: for (k = 0; k < 4; k++) { yading@10: frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X; yading@10: if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { yading@10: s->motion_val[1][frag][0] = motion_x[k]; yading@10: s->motion_val[1][frag][1] = motion_y[k]; yading@10: } else { yading@10: s->motion_val[1][frag][0] = motion_x[0]; yading@10: s->motion_val[1][frag][1] = motion_y[0]; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb) yading@10: { yading@10: int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi; yading@10: int num_blocks = s->total_num_coded_frags; yading@10: yading@10: for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) { yading@10: i = blocks_decoded = num_blocks_at_qpi = 0; yading@10: yading@10: bit = get_bits1(gb) ^ 1; yading@10: run_length = 0; yading@10: yading@10: do { yading@10: if (run_length == MAXIMUM_LONG_BIT_RUN) yading@10: bit = get_bits1(gb); yading@10: else yading@10: bit ^= 1; yading@10: yading@10: run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1; yading@10: if (run_length == 34) yading@10: run_length += get_bits(gb, 12); yading@10: blocks_decoded += run_length; yading@10: yading@10: if (!bit) yading@10: num_blocks_at_qpi += run_length; yading@10: yading@10: for (j = 0; j < run_length; i++) { yading@10: if (i >= s->total_num_coded_frags) yading@10: return -1; yading@10: yading@10: if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) { yading@10: s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit; yading@10: j++; yading@10: } yading@10: } yading@10: } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0); yading@10: yading@10: num_blocks -= num_blocks_at_qpi; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /* yading@10: * This function is called by unpack_dct_coeffs() to extract the VLCs from yading@10: * the bitstream. The VLCs encode tokens which are used to unpack DCT yading@10: * data. This function unpacks all the VLCs for either the Y plane or both yading@10: * C planes, and is called for DC coefficients or different AC coefficient yading@10: * levels (since different coefficient types require different VLC tables. yading@10: * yading@10: * This function returns a residual eob run. E.g, if a particular token gave yading@10: * instructions to EOB the next 5 fragments and there were only 2 fragments yading@10: * left in the current fragment range, 3 would be returned so that it could yading@10: * be passed into the next call to this same function. yading@10: */ yading@10: static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb, yading@10: VLC *table, int coeff_index, yading@10: int plane, yading@10: int eob_run) yading@10: { yading@10: int i, j = 0; yading@10: int token; yading@10: int zero_run = 0; yading@10: int16_t coeff = 0; yading@10: int bits_to_get; yading@10: int blocks_ended; yading@10: int coeff_i = 0; yading@10: int num_coeffs = s->num_coded_frags[plane][coeff_index]; yading@10: int16_t *dct_tokens = s->dct_tokens[plane][coeff_index]; yading@10: yading@10: /* local references to structure members to avoid repeated deferences */ yading@10: int *coded_fragment_list = s->coded_fragment_list[plane]; yading@10: Vp3Fragment *all_fragments = s->all_fragments; yading@10: VLC_TYPE (*vlc_table)[2] = table->table; yading@10: yading@10: if (num_coeffs < 0) yading@10: av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index); yading@10: yading@10: if (eob_run > num_coeffs) { yading@10: coeff_i = blocks_ended = num_coeffs; yading@10: eob_run -= num_coeffs; yading@10: } else { yading@10: coeff_i = blocks_ended = eob_run; yading@10: eob_run = 0; yading@10: } yading@10: yading@10: // insert fake EOB token to cover the split between planes or zzi yading@10: if (blocks_ended) yading@10: dct_tokens[j++] = blocks_ended << 2; yading@10: yading@10: while (coeff_i < num_coeffs && get_bits_left(gb) > 0) { yading@10: /* decode a VLC into a token */ yading@10: token = get_vlc2(gb, vlc_table, 11, 3); yading@10: /* use the token to get a zero run, a coefficient, and an eob run */ yading@10: if ((unsigned) token <= 6U) { yading@10: eob_run = eob_run_base[token]; yading@10: if (eob_run_get_bits[token]) yading@10: eob_run += get_bits(gb, eob_run_get_bits[token]); yading@10: yading@10: // record only the number of blocks ended in this plane, yading@10: // any spill will be recorded in the next plane. yading@10: if (eob_run > num_coeffs - coeff_i) { yading@10: dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i); yading@10: blocks_ended += num_coeffs - coeff_i; yading@10: eob_run -= num_coeffs - coeff_i; yading@10: coeff_i = num_coeffs; yading@10: } else { yading@10: dct_tokens[j++] = TOKEN_EOB(eob_run); yading@10: blocks_ended += eob_run; yading@10: coeff_i += eob_run; yading@10: eob_run = 0; yading@10: } yading@10: } else if (token >= 0) { yading@10: bits_to_get = coeff_get_bits[token]; yading@10: if (bits_to_get) yading@10: bits_to_get = get_bits(gb, bits_to_get); yading@10: coeff = coeff_tables[token][bits_to_get]; yading@10: yading@10: zero_run = zero_run_base[token]; yading@10: if (zero_run_get_bits[token]) yading@10: zero_run += get_bits(gb, zero_run_get_bits[token]); yading@10: yading@10: if (zero_run) { yading@10: dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run); yading@10: } else { yading@10: // Save DC into the fragment structure. DC prediction is yading@10: // done in raster order, so the actual DC can't be in with yading@10: // other tokens. We still need the token in dct_tokens[] yading@10: // however, or else the structure collapses on itself. yading@10: if (!coeff_index) yading@10: all_fragments[coded_fragment_list[coeff_i]].dc = coeff; yading@10: yading@10: dct_tokens[j++] = TOKEN_COEFF(coeff); yading@10: } yading@10: yading@10: if (coeff_index + zero_run > 64) { yading@10: av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with" yading@10: " %d coeffs left\n", zero_run, 64-coeff_index); yading@10: zero_run = 64 - coeff_index; yading@10: } yading@10: yading@10: // zero runs code multiple coefficients, yading@10: // so don't try to decode coeffs for those higher levels yading@10: for (i = coeff_index+1; i <= coeff_index+zero_run; i++) yading@10: s->num_coded_frags[plane][i]--; yading@10: coeff_i++; yading@10: } else { yading@10: av_log(s->avctx, AV_LOG_ERROR, yading@10: "Invalid token %d\n", token); yading@10: return -1; yading@10: } yading@10: } yading@10: yading@10: if (blocks_ended > s->num_coded_frags[plane][coeff_index]) yading@10: av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n"); yading@10: yading@10: // decrement the number of blocks that have higher coeffecients for each yading@10: // EOB run at this level yading@10: if (blocks_ended) yading@10: for (i = coeff_index+1; i < 64; i++) yading@10: s->num_coded_frags[plane][i] -= blocks_ended; yading@10: yading@10: // setup the next buffer yading@10: if (plane < 2) yading@10: s->dct_tokens[plane+1][coeff_index] = dct_tokens + j; yading@10: else if (coeff_index < 63) yading@10: s->dct_tokens[0][coeff_index+1] = dct_tokens + j; yading@10: yading@10: return eob_run; yading@10: } yading@10: yading@10: static void reverse_dc_prediction(Vp3DecodeContext *s, yading@10: int first_fragment, yading@10: int fragment_width, yading@10: int fragment_height); yading@10: /* yading@10: * This function unpacks all of the DCT coefficient data from the yading@10: * bitstream. yading@10: */ yading@10: static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb) yading@10: { yading@10: int i; yading@10: int dc_y_table; yading@10: int dc_c_table; yading@10: int ac_y_table; yading@10: int ac_c_table; yading@10: int residual_eob_run = 0; yading@10: VLC *y_tables[64]; yading@10: VLC *c_tables[64]; yading@10: yading@10: s->dct_tokens[0][0] = s->dct_tokens_base; yading@10: yading@10: /* fetch the DC table indexes */ yading@10: dc_y_table = get_bits(gb, 4); yading@10: dc_c_table = get_bits(gb, 4); yading@10: yading@10: /* unpack the Y plane DC coefficients */ yading@10: residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0, yading@10: 0, residual_eob_run); yading@10: if (residual_eob_run < 0) yading@10: return residual_eob_run; yading@10: yading@10: /* reverse prediction of the Y-plane DC coefficients */ yading@10: reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]); yading@10: yading@10: /* unpack the C plane DC coefficients */ yading@10: residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, yading@10: 1, residual_eob_run); yading@10: if (residual_eob_run < 0) yading@10: return residual_eob_run; yading@10: residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, yading@10: 2, residual_eob_run); yading@10: if (residual_eob_run < 0) yading@10: return residual_eob_run; yading@10: yading@10: /* reverse prediction of the C-plane DC coefficients */ yading@10: if (!(s->avctx->flags & CODEC_FLAG_GRAY)) yading@10: { yading@10: reverse_dc_prediction(s, s->fragment_start[1], yading@10: s->fragment_width[1], s->fragment_height[1]); yading@10: reverse_dc_prediction(s, s->fragment_start[2], yading@10: s->fragment_width[1], s->fragment_height[1]); yading@10: } yading@10: yading@10: /* fetch the AC table indexes */ yading@10: ac_y_table = get_bits(gb, 4); yading@10: ac_c_table = get_bits(gb, 4); yading@10: yading@10: /* build tables of AC VLC tables */ yading@10: for (i = 1; i <= 5; i++) { yading@10: y_tables[i] = &s->ac_vlc_1[ac_y_table]; yading@10: c_tables[i] = &s->ac_vlc_1[ac_c_table]; yading@10: } yading@10: for (i = 6; i <= 14; i++) { yading@10: y_tables[i] = &s->ac_vlc_2[ac_y_table]; yading@10: c_tables[i] = &s->ac_vlc_2[ac_c_table]; yading@10: } yading@10: for (i = 15; i <= 27; i++) { yading@10: y_tables[i] = &s->ac_vlc_3[ac_y_table]; yading@10: c_tables[i] = &s->ac_vlc_3[ac_c_table]; yading@10: } yading@10: for (i = 28; i <= 63; i++) { yading@10: y_tables[i] = &s->ac_vlc_4[ac_y_table]; yading@10: c_tables[i] = &s->ac_vlc_4[ac_c_table]; yading@10: } yading@10: yading@10: /* decode all AC coefficents */ yading@10: for (i = 1; i <= 63; i++) { yading@10: residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i, yading@10: 0, residual_eob_run); yading@10: if (residual_eob_run < 0) yading@10: return residual_eob_run; yading@10: yading@10: residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i, yading@10: 1, residual_eob_run); yading@10: if (residual_eob_run < 0) yading@10: return residual_eob_run; yading@10: residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i, yading@10: 2, residual_eob_run); yading@10: if (residual_eob_run < 0) yading@10: return residual_eob_run; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /* yading@10: * This function reverses the DC prediction for each coded fragment in yading@10: * the frame. Much of this function is adapted directly from the original yading@10: * VP3 source code. yading@10: */ yading@10: #define COMPATIBLE_FRAME(x) \ yading@10: (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type) yading@10: #define DC_COEFF(u) s->all_fragments[u].dc yading@10: yading@10: static void reverse_dc_prediction(Vp3DecodeContext *s, yading@10: int first_fragment, yading@10: int fragment_width, yading@10: int fragment_height) yading@10: { yading@10: yading@10: #define PUL 8 yading@10: #define PU 4 yading@10: #define PUR 2 yading@10: #define PL 1 yading@10: yading@10: int x, y; yading@10: int i = first_fragment; yading@10: yading@10: int predicted_dc; yading@10: yading@10: /* DC values for the left, up-left, up, and up-right fragments */ yading@10: int vl, vul, vu, vur; yading@10: yading@10: /* indexes for the left, up-left, up, and up-right fragments */ yading@10: int l, ul, u, ur; yading@10: yading@10: /* yading@10: * The 6 fields mean: yading@10: * 0: up-left multiplier yading@10: * 1: up multiplier yading@10: * 2: up-right multiplier yading@10: * 3: left multiplier yading@10: */ yading@10: static const int predictor_transform[16][4] = { yading@10: { 0, 0, 0, 0}, yading@10: { 0, 0, 0,128}, // PL yading@10: { 0, 0,128, 0}, // PUR yading@10: { 0, 0, 53, 75}, // PUR|PL yading@10: { 0,128, 0, 0}, // PU yading@10: { 0, 64, 0, 64}, // PU|PL yading@10: { 0,128, 0, 0}, // PU|PUR yading@10: { 0, 0, 53, 75}, // PU|PUR|PL yading@10: {128, 0, 0, 0}, // PUL yading@10: { 0, 0, 0,128}, // PUL|PL yading@10: { 64, 0, 64, 0}, // PUL|PUR yading@10: { 0, 0, 53, 75}, // PUL|PUR|PL yading@10: { 0,128, 0, 0}, // PUL|PU yading@10: {-104,116, 0,116}, // PUL|PU|PL yading@10: { 24, 80, 24, 0}, // PUL|PU|PUR yading@10: {-104,116, 0,116} // PUL|PU|PUR|PL yading@10: }; yading@10: yading@10: /* This table shows which types of blocks can use other blocks for yading@10: * prediction. For example, INTRA is the only mode in this table to yading@10: * have a frame number of 0. That means INTRA blocks can only predict yading@10: * from other INTRA blocks. There are 2 golden frame coding types; yading@10: * blocks encoding in these modes can only predict from other blocks yading@10: * that were encoded with these 1 of these 2 modes. */ yading@10: static const unsigned char compatible_frame[9] = { yading@10: 1, /* MODE_INTER_NO_MV */ yading@10: 0, /* MODE_INTRA */ yading@10: 1, /* MODE_INTER_PLUS_MV */ yading@10: 1, /* MODE_INTER_LAST_MV */ yading@10: 1, /* MODE_INTER_PRIOR_MV */ yading@10: 2, /* MODE_USING_GOLDEN */ yading@10: 2, /* MODE_GOLDEN_MV */ yading@10: 1, /* MODE_INTER_FOUR_MV */ yading@10: 3 /* MODE_COPY */ yading@10: }; yading@10: int current_frame_type; yading@10: yading@10: /* there is a last DC predictor for each of the 3 frame types */ yading@10: short last_dc[3]; yading@10: yading@10: int transform = 0; yading@10: yading@10: vul = vu = vur = vl = 0; yading@10: last_dc[0] = last_dc[1] = last_dc[2] = 0; yading@10: yading@10: /* for each fragment row... */ yading@10: for (y = 0; y < fragment_height; y++) { yading@10: yading@10: /* for each fragment in a row... */ yading@10: for (x = 0; x < fragment_width; x++, i++) { yading@10: yading@10: /* reverse prediction if this block was coded */ yading@10: if (s->all_fragments[i].coding_method != MODE_COPY) { yading@10: yading@10: current_frame_type = yading@10: compatible_frame[s->all_fragments[i].coding_method]; yading@10: yading@10: transform= 0; yading@10: if(x){ yading@10: l= i-1; yading@10: vl = DC_COEFF(l); yading@10: if(COMPATIBLE_FRAME(l)) yading@10: transform |= PL; yading@10: } yading@10: if(y){ yading@10: u= i-fragment_width; yading@10: vu = DC_COEFF(u); yading@10: if(COMPATIBLE_FRAME(u)) yading@10: transform |= PU; yading@10: if(x){ yading@10: ul= i-fragment_width-1; yading@10: vul = DC_COEFF(ul); yading@10: if(COMPATIBLE_FRAME(ul)) yading@10: transform |= PUL; yading@10: } yading@10: if(x + 1 < fragment_width){ yading@10: ur= i-fragment_width+1; yading@10: vur = DC_COEFF(ur); yading@10: if(COMPATIBLE_FRAME(ur)) yading@10: transform |= PUR; yading@10: } yading@10: } yading@10: yading@10: if (transform == 0) { yading@10: yading@10: /* if there were no fragments to predict from, use last yading@10: * DC saved */ yading@10: predicted_dc = last_dc[current_frame_type]; yading@10: } else { yading@10: yading@10: /* apply the appropriate predictor transform */ yading@10: predicted_dc = yading@10: (predictor_transform[transform][0] * vul) + yading@10: (predictor_transform[transform][1] * vu) + yading@10: (predictor_transform[transform][2] * vur) + yading@10: (predictor_transform[transform][3] * vl); yading@10: yading@10: predicted_dc /= 128; yading@10: yading@10: /* check for outranging on the [ul u l] and yading@10: * [ul u ur l] predictors */ yading@10: if ((transform == 15) || (transform == 13)) { yading@10: if (FFABS(predicted_dc - vu) > 128) yading@10: predicted_dc = vu; yading@10: else if (FFABS(predicted_dc - vl) > 128) yading@10: predicted_dc = vl; yading@10: else if (FFABS(predicted_dc - vul) > 128) yading@10: predicted_dc = vul; yading@10: } yading@10: } yading@10: yading@10: /* at long last, apply the predictor */ yading@10: DC_COEFF(i) += predicted_dc; yading@10: /* save the DC */ yading@10: last_dc[current_frame_type] = DC_COEFF(i); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend) yading@10: { yading@10: int x, y; yading@10: int *bounding_values= s->bounding_values_array+127; yading@10: yading@10: int width = s->fragment_width[!!plane]; yading@10: int height = s->fragment_height[!!plane]; yading@10: int fragment = s->fragment_start [plane] + ystart * width; yading@10: int stride = s->current_frame.f->linesize[plane]; yading@10: uint8_t *plane_data = s->current_frame.f->data [plane]; yading@10: if (!s->flipped_image) stride = -stride; yading@10: plane_data += s->data_offset[plane] + 8*ystart*stride; yading@10: yading@10: for (y = ystart; y < yend; y++) { yading@10: yading@10: for (x = 0; x < width; x++) { yading@10: /* This code basically just deblocks on the edges of coded blocks. yading@10: * However, it has to be much more complicated because of the yading@10: * braindamaged deblock ordering used in VP3/Theora. Order matters yading@10: * because some pixels get filtered twice. */ yading@10: if( s->all_fragments[fragment].coding_method != MODE_COPY ) yading@10: { yading@10: /* do not perform left edge filter for left columns frags */ yading@10: if (x > 0) { yading@10: s->vp3dsp.h_loop_filter( yading@10: plane_data + 8*x, yading@10: stride, bounding_values); yading@10: } yading@10: yading@10: /* do not perform top edge filter for top row fragments */ yading@10: if (y > 0) { yading@10: s->vp3dsp.v_loop_filter( yading@10: plane_data + 8*x, yading@10: stride, bounding_values); yading@10: } yading@10: yading@10: /* do not perform right edge filter for right column yading@10: * fragments or if right fragment neighbor is also coded yading@10: * in this frame (it will be filtered in next iteration) */ yading@10: if ((x < width - 1) && yading@10: (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) { yading@10: s->vp3dsp.h_loop_filter( yading@10: plane_data + 8*x + 8, yading@10: stride, bounding_values); yading@10: } yading@10: yading@10: /* do not perform bottom edge filter for bottom row yading@10: * fragments or if bottom fragment neighbor is also coded yading@10: * in this frame (it will be filtered in the next row) */ yading@10: if ((y < height - 1) && yading@10: (s->all_fragments[fragment + width].coding_method == MODE_COPY)) { yading@10: s->vp3dsp.v_loop_filter( yading@10: plane_data + 8*x + 8*stride, yading@10: stride, bounding_values); yading@10: } yading@10: } yading@10: yading@10: fragment++; yading@10: } yading@10: plane_data += 8*stride; yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Pull DCT tokens from the 64 levels to decode and dequant the coefficients yading@10: * for the next block in coding order yading@10: */ yading@10: static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag, yading@10: int plane, int inter, int16_t block[64]) yading@10: { yading@10: int16_t *dequantizer = s->qmat[frag->qpi][inter][plane]; yading@10: uint8_t *perm = s->idct_scantable; yading@10: int i = 0; yading@10: yading@10: do { yading@10: int token = *s->dct_tokens[plane][i]; yading@10: switch (token & 3) { yading@10: case 0: // EOB yading@10: if (--token < 4) // 0-3 are token types, so the EOB run must now be 0 yading@10: s->dct_tokens[plane][i]++; yading@10: else yading@10: *s->dct_tokens[plane][i] = token & ~3; yading@10: goto end; yading@10: case 1: // zero run yading@10: s->dct_tokens[plane][i]++; yading@10: i += (token >> 2) & 0x7f; yading@10: if (i > 63) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n"); yading@10: return i; yading@10: } yading@10: block[perm[i]] = (token >> 9) * dequantizer[perm[i]]; yading@10: i++; yading@10: break; yading@10: case 2: // coeff yading@10: block[perm[i]] = (token >> 2) * dequantizer[perm[i]]; yading@10: s->dct_tokens[plane][i++]++; yading@10: break; yading@10: default: // shouldn't happen yading@10: return i; yading@10: } yading@10: } while (i < 64); yading@10: // return value is expected to be a valid level yading@10: i--; yading@10: end: yading@10: // the actual DC+prediction is in the fragment structure yading@10: block[0] = frag->dc * s->qmat[0][inter][plane][0]; yading@10: return i; yading@10: } yading@10: yading@10: /** yading@10: * called when all pixels up to row y are complete yading@10: */ yading@10: static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y) yading@10: { yading@10: int h, cy, i; yading@10: int offset[AV_NUM_DATA_POINTERS]; yading@10: yading@10: if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) { yading@10: int y_flipped = s->flipped_image ? s->avctx->height-y : y; yading@10: yading@10: // At the end of the frame, report INT_MAX instead of the height of the frame. yading@10: // This makes the other threads' ff_thread_await_progress() calls cheaper, because yading@10: // they don't have to clip their values. yading@10: ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0); yading@10: } yading@10: yading@10: if(s->avctx->draw_horiz_band==NULL) yading@10: return; yading@10: yading@10: h= y - s->last_slice_end; yading@10: s->last_slice_end= y; yading@10: y -= h; yading@10: yading@10: if (!s->flipped_image) { yading@10: y = s->avctx->height - y - h; yading@10: } yading@10: yading@10: cy = y >> s->chroma_y_shift; yading@10: offset[0] = s->current_frame.f->linesize[0]*y; yading@10: offset[1] = s->current_frame.f->linesize[1]*cy; yading@10: offset[2] = s->current_frame.f->linesize[2]*cy; yading@10: for (i = 3; i < AV_NUM_DATA_POINTERS; i++) yading@10: offset[i] = 0; yading@10: yading@10: emms_c(); yading@10: s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h); yading@10: } yading@10: yading@10: /** yading@10: * Wait for the reference frame of the current fragment. yading@10: * The progress value is in luma pixel rows. yading@10: */ yading@10: static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y) yading@10: { yading@10: ThreadFrame *ref_frame; yading@10: int ref_row; yading@10: int border = motion_y&1; yading@10: yading@10: if (fragment->coding_method == MODE_USING_GOLDEN || yading@10: fragment->coding_method == MODE_GOLDEN_MV) yading@10: ref_frame = &s->golden_frame; yading@10: else yading@10: ref_frame = &s->last_frame; yading@10: yading@10: ref_row = y + (motion_y>>1); yading@10: ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border); yading@10: yading@10: ff_thread_await_progress(ref_frame, ref_row, 0); yading@10: } yading@10: yading@10: /* yading@10: * Perform the final rendering for a particular slice of data. yading@10: * The slice number ranges from 0..(c_superblock_height - 1). yading@10: */ yading@10: static void render_slice(Vp3DecodeContext *s, int slice) yading@10: { yading@10: int x, y, i, j, fragment; yading@10: int16_t *block = s->block; yading@10: int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef; yading@10: int motion_halfpel_index; yading@10: uint8_t *motion_source; yading@10: int plane, first_pixel; yading@10: yading@10: if (slice >= s->c_superblock_height) yading@10: return; yading@10: yading@10: for (plane = 0; plane < 3; plane++) { yading@10: uint8_t *output_plane = s->current_frame.f->data [plane] + s->data_offset[plane]; yading@10: uint8_t * last_plane = s-> last_frame.f->data [plane] + s->data_offset[plane]; yading@10: uint8_t *golden_plane = s-> golden_frame.f->data [plane] + s->data_offset[plane]; yading@10: int stride = s->current_frame.f->linesize[plane]; yading@10: int plane_width = s->width >> (plane && s->chroma_x_shift); yading@10: int plane_height = s->height >> (plane && s->chroma_y_shift); yading@10: int8_t (*motion_val)[2] = s->motion_val[!!plane]; yading@10: yading@10: int sb_x, sb_y = slice << (!plane && s->chroma_y_shift); yading@10: int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift); yading@10: int slice_width = plane ? s->c_superblock_width : s->y_superblock_width; yading@10: yading@10: int fragment_width = s->fragment_width[!!plane]; yading@10: int fragment_height = s->fragment_height[!!plane]; yading@10: int fragment_start = s->fragment_start[plane]; yading@10: int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME); yading@10: yading@10: if (!s->flipped_image) stride = -stride; yading@10: if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY)) yading@10: continue; yading@10: yading@10: /* for each superblock row in the slice (both of them)... */ yading@10: for (; sb_y < slice_height; sb_y++) { yading@10: yading@10: /* for each superblock in a row... */ yading@10: for (sb_x = 0; sb_x < slice_width; sb_x++) { yading@10: yading@10: /* for each block in a superblock... */ yading@10: for (j = 0; j < 16; j++) { yading@10: x = 4*sb_x + hilbert_offset[j][0]; yading@10: y = 4*sb_y + hilbert_offset[j][1]; yading@10: fragment = y*fragment_width + x; yading@10: yading@10: i = fragment_start + fragment; yading@10: yading@10: // bounds check yading@10: if (x >= fragment_width || y >= fragment_height) yading@10: continue; yading@10: yading@10: first_pixel = 8*y*stride + 8*x; yading@10: yading@10: if (do_await && s->all_fragments[i].coding_method != MODE_INTRA) yading@10: await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift); yading@10: yading@10: /* transform if this block was coded */ yading@10: if (s->all_fragments[i].coding_method != MODE_COPY) { yading@10: if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) || yading@10: (s->all_fragments[i].coding_method == MODE_GOLDEN_MV)) yading@10: motion_source= golden_plane; yading@10: else yading@10: motion_source= last_plane; yading@10: yading@10: motion_source += first_pixel; yading@10: motion_halfpel_index = 0; yading@10: yading@10: /* sort out the motion vector if this fragment is coded yading@10: * using a motion vector method */ yading@10: if ((s->all_fragments[i].coding_method > MODE_INTRA) && yading@10: (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) { yading@10: int src_x, src_y; yading@10: motion_x = motion_val[fragment][0]; yading@10: motion_y = motion_val[fragment][1]; yading@10: yading@10: src_x= (motion_x>>1) + 8*x; yading@10: src_y= (motion_y>>1) + 8*y; yading@10: yading@10: motion_halfpel_index = motion_x & 0x01; yading@10: motion_source += (motion_x >> 1); yading@10: yading@10: motion_halfpel_index |= (motion_y & 0x01) << 1; yading@10: motion_source += ((motion_y >> 1) * stride); yading@10: yading@10: if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){ yading@10: uint8_t *temp= s->edge_emu_buffer; yading@10: if(stride<0) temp -= 8*stride; yading@10: yading@10: s->vdsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height); yading@10: motion_source= temp; yading@10: } yading@10: } yading@10: yading@10: yading@10: /* first, take care of copying a block from either the yading@10: * previous or the golden frame */ yading@10: if (s->all_fragments[i].coding_method != MODE_INTRA) { yading@10: /* Note, it is possible to implement all MC cases with yading@10: put_no_rnd_pixels_l2 which would look more like the yading@10: VP3 source but this would be slower as yading@10: put_no_rnd_pixels_tab is better optimzed */ yading@10: if(motion_halfpel_index != 3){ yading@10: s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index]( yading@10: output_plane + first_pixel, yading@10: motion_source, stride, 8); yading@10: }else{ yading@10: int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1 yading@10: s->vp3dsp.put_no_rnd_pixels_l2( yading@10: output_plane + first_pixel, yading@10: motion_source - d, yading@10: motion_source + stride + 1 + d, yading@10: stride, 8); yading@10: } yading@10: } yading@10: yading@10: /* invert DCT and place (or add) in final output */ yading@10: yading@10: if (s->all_fragments[i].coding_method == MODE_INTRA) { yading@10: vp3_dequant(s, s->all_fragments + i, plane, 0, block); yading@10: s->vp3dsp.idct_put( yading@10: output_plane + first_pixel, yading@10: stride, yading@10: block); yading@10: } else { yading@10: if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) { yading@10: s->vp3dsp.idct_add( yading@10: output_plane + first_pixel, yading@10: stride, yading@10: block); yading@10: } else { yading@10: s->vp3dsp.idct_dc_add(output_plane + first_pixel, stride, block); yading@10: } yading@10: } yading@10: } else { yading@10: yading@10: /* copy directly from the previous frame */ yading@10: s->hdsp.put_pixels_tab[1][0]( yading@10: output_plane + first_pixel, yading@10: last_plane + first_pixel, yading@10: stride, 8); yading@10: yading@10: } yading@10: } yading@10: } yading@10: yading@10: // Filter up to the last row in the superblock row yading@10: if (!s->skip_loop_filter) yading@10: apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1)); yading@10: } yading@10: } yading@10: yading@10: /* this looks like a good place for slice dispatch... */ yading@10: /* algorithm: yading@10: * if (slice == s->macroblock_height - 1) yading@10: * dispatch (both last slice & 2nd-to-last slice); yading@10: * else if (slice > 0) yading@10: * dispatch (slice - 1); yading@10: */ yading@10: yading@10: vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16)); yading@10: } yading@10: yading@10: /// Allocate tables for per-frame data in Vp3DecodeContext yading@10: static av_cold int allocate_tables(AVCodecContext *avctx) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: int y_fragment_count, c_fragment_count; yading@10: yading@10: y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; yading@10: c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; yading@10: yading@10: s->superblock_coding = av_malloc(s->superblock_count); yading@10: s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment)); yading@10: s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int)); yading@10: s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base)); yading@10: s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0])); yading@10: s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1])); yading@10: yading@10: /* work out the block mapping tables */ yading@10: s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int)); yading@10: s->macroblock_coding = av_malloc(s->macroblock_count + 1); yading@10: yading@10: if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base || yading@10: !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding || yading@10: !s->motion_val[0] || !s->motion_val[1]) { yading@10: vp3_decode_end(avctx); yading@10: return -1; yading@10: } yading@10: yading@10: init_block_mapping(s); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int init_frames(Vp3DecodeContext *s) yading@10: { yading@10: s->current_frame.f = av_frame_alloc(); yading@10: s->last_frame.f = av_frame_alloc(); yading@10: s->golden_frame.f = av_frame_alloc(); yading@10: yading@10: if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) { yading@10: av_frame_free(&s->current_frame.f); yading@10: av_frame_free(&s->last_frame.f); yading@10: av_frame_free(&s->golden_frame.f); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int vp3_decode_init(AVCodecContext *avctx) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: int i, inter, plane, ret; yading@10: int c_width; yading@10: int c_height; yading@10: int y_fragment_count, c_fragment_count; yading@10: yading@10: ret = init_frames(s); yading@10: if (ret < 0) yading@10: return ret; yading@10: yading@10: avctx->internal->allocate_progress = 1; yading@10: yading@10: if (avctx->codec_tag == MKTAG('V','P','3','0')) yading@10: s->version = 0; yading@10: else yading@10: s->version = 1; yading@10: yading@10: s->avctx = avctx; yading@10: s->width = FFALIGN(avctx->width, 16); yading@10: s->height = FFALIGN(avctx->height, 16); yading@10: if (avctx->codec_id != AV_CODEC_ID_THEORA) yading@10: avctx->pix_fmt = AV_PIX_FMT_YUV420P; yading@10: avctx->chroma_sample_location = AVCHROMA_LOC_CENTER; yading@10: ff_hpeldsp_init(&s->hdsp, avctx->flags | CODEC_FLAG_BITEXACT); yading@10: ff_videodsp_init(&s->vdsp, 8); yading@10: ff_vp3dsp_init(&s->vp3dsp, avctx->flags); yading@10: yading@10: for (i = 0; i < 64; i++) { yading@10: #define T(x) (x >> 3) | ((x & 7) << 3) yading@10: s->idct_permutation[i] = T(i); yading@10: s->idct_scantable[i] = T(ff_zigzag_direct[i]); yading@10: #undef T yading@10: } yading@10: yading@10: /* initialize to an impossible value which will force a recalculation yading@10: * in the first frame decode */ yading@10: for (i = 0; i < 3; i++) yading@10: s->qps[i] = -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: s->y_superblock_width = (s->width + 31) / 32; yading@10: s->y_superblock_height = (s->height + 31) / 32; yading@10: s->y_superblock_count = s->y_superblock_width * s->y_superblock_height; yading@10: yading@10: /* work out the dimensions for the C planes */ yading@10: c_width = s->width >> s->chroma_x_shift; yading@10: c_height = s->height >> s->chroma_y_shift; yading@10: s->c_superblock_width = (c_width + 31) / 32; yading@10: s->c_superblock_height = (c_height + 31) / 32; yading@10: s->c_superblock_count = s->c_superblock_width * s->c_superblock_height; yading@10: yading@10: s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2); yading@10: s->u_superblock_start = s->y_superblock_count; yading@10: s->v_superblock_start = s->u_superblock_start + s->c_superblock_count; yading@10: yading@10: s->macroblock_width = (s->width + 15) / 16; yading@10: s->macroblock_height = (s->height + 15) / 16; yading@10: s->macroblock_count = s->macroblock_width * s->macroblock_height; yading@10: yading@10: s->fragment_width[0] = s->width / FRAGMENT_PIXELS; yading@10: s->fragment_height[0] = s->height / FRAGMENT_PIXELS; yading@10: s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift; yading@10: s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift; yading@10: yading@10: /* fragment count covers all 8x8 blocks for all 3 planes */ yading@10: y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; yading@10: c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; yading@10: s->fragment_count = y_fragment_count + 2*c_fragment_count; yading@10: s->fragment_start[1] = y_fragment_count; yading@10: s->fragment_start[2] = y_fragment_count + c_fragment_count; yading@10: yading@10: if (!s->theora_tables) yading@10: { yading@10: for (i = 0; i < 64; i++) { yading@10: s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i]; yading@10: s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i]; yading@10: s->base_matrix[0][i] = vp31_intra_y_dequant[i]; yading@10: s->base_matrix[1][i] = vp31_intra_c_dequant[i]; yading@10: s->base_matrix[2][i] = vp31_inter_dequant[i]; yading@10: s->filter_limit_values[i] = vp31_filter_limit_values[i]; yading@10: } yading@10: yading@10: for(inter=0; inter<2; inter++){ yading@10: for(plane=0; plane<3; plane++){ yading@10: s->qr_count[inter][plane]= 1; yading@10: s->qr_size [inter][plane][0]= 63; yading@10: s->qr_base [inter][plane][0]= yading@10: s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter; yading@10: } yading@10: } yading@10: yading@10: /* init VLC tables */ yading@10: for (i = 0; i < 16; i++) { yading@10: yading@10: /* DC histograms */ yading@10: init_vlc(&s->dc_vlc[i], 11, 32, yading@10: &dc_bias[i][0][1], 4, 2, yading@10: &dc_bias[i][0][0], 4, 2, 0); yading@10: yading@10: /* group 1 AC histograms */ yading@10: init_vlc(&s->ac_vlc_1[i], 11, 32, yading@10: &ac_bias_0[i][0][1], 4, 2, yading@10: &ac_bias_0[i][0][0], 4, 2, 0); yading@10: yading@10: /* group 2 AC histograms */ yading@10: init_vlc(&s->ac_vlc_2[i], 11, 32, yading@10: &ac_bias_1[i][0][1], 4, 2, yading@10: &ac_bias_1[i][0][0], 4, 2, 0); yading@10: yading@10: /* group 3 AC histograms */ yading@10: init_vlc(&s->ac_vlc_3[i], 11, 32, yading@10: &ac_bias_2[i][0][1], 4, 2, yading@10: &ac_bias_2[i][0][0], 4, 2, 0); yading@10: yading@10: /* group 4 AC histograms */ yading@10: init_vlc(&s->ac_vlc_4[i], 11, 32, yading@10: &ac_bias_3[i][0][1], 4, 2, yading@10: &ac_bias_3[i][0][0], 4, 2, 0); yading@10: } yading@10: } else { yading@10: yading@10: for (i = 0; i < 16; i++) { yading@10: /* DC histograms */ yading@10: if (init_vlc(&s->dc_vlc[i], 11, 32, yading@10: &s->huffman_table[i][0][1], 8, 4, yading@10: &s->huffman_table[i][0][0], 8, 4, 0) < 0) yading@10: goto vlc_fail; yading@10: yading@10: /* group 1 AC histograms */ yading@10: if (init_vlc(&s->ac_vlc_1[i], 11, 32, yading@10: &s->huffman_table[i+16][0][1], 8, 4, yading@10: &s->huffman_table[i+16][0][0], 8, 4, 0) < 0) yading@10: goto vlc_fail; yading@10: yading@10: /* group 2 AC histograms */ yading@10: if (init_vlc(&s->ac_vlc_2[i], 11, 32, yading@10: &s->huffman_table[i+16*2][0][1], 8, 4, yading@10: &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0) yading@10: goto vlc_fail; yading@10: yading@10: /* group 3 AC histograms */ yading@10: if (init_vlc(&s->ac_vlc_3[i], 11, 32, yading@10: &s->huffman_table[i+16*3][0][1], 8, 4, yading@10: &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0) yading@10: goto vlc_fail; yading@10: yading@10: /* group 4 AC histograms */ yading@10: if (init_vlc(&s->ac_vlc_4[i], 11, 32, yading@10: &s->huffman_table[i+16*4][0][1], 8, 4, yading@10: &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0) yading@10: goto vlc_fail; yading@10: } yading@10: } yading@10: yading@10: init_vlc(&s->superblock_run_length_vlc, 6, 34, yading@10: &superblock_run_length_vlc_table[0][1], 4, 2, yading@10: &superblock_run_length_vlc_table[0][0], 4, 2, 0); yading@10: yading@10: init_vlc(&s->fragment_run_length_vlc, 5, 30, yading@10: &fragment_run_length_vlc_table[0][1], 4, 2, yading@10: &fragment_run_length_vlc_table[0][0], 4, 2, 0); yading@10: yading@10: init_vlc(&s->mode_code_vlc, 3, 8, yading@10: &mode_code_vlc_table[0][1], 2, 1, yading@10: &mode_code_vlc_table[0][0], 2, 1, 0); yading@10: yading@10: init_vlc(&s->motion_vector_vlc, 6, 63, yading@10: &motion_vector_vlc_table[0][1], 2, 1, yading@10: &motion_vector_vlc_table[0][0], 2, 1, 0); yading@10: yading@10: return allocate_tables(avctx); yading@10: yading@10: vlc_fail: yading@10: av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n"); yading@10: return -1; yading@10: } yading@10: yading@10: /// Release and shuffle frames after decode finishes yading@10: static int update_frames(AVCodecContext *avctx) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: int ret = 0; yading@10: yading@10: yading@10: /* shuffle frames (last = current) */ yading@10: ff_thread_release_buffer(avctx, &s->last_frame); yading@10: ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame); yading@10: if (ret < 0) yading@10: goto fail; yading@10: yading@10: if (s->keyframe) { yading@10: ff_thread_release_buffer(avctx, &s->golden_frame); yading@10: ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame); yading@10: } yading@10: yading@10: fail: yading@10: ff_thread_release_buffer(avctx, &s->current_frame); yading@10: return ret; yading@10: } yading@10: yading@10: static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src) yading@10: { yading@10: ff_thread_release_buffer(s->avctx, dst); yading@10: if (src->f->data[0]) yading@10: return ff_thread_ref_frame(dst, src); yading@10: return 0; yading@10: } yading@10: yading@10: static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src) yading@10: { yading@10: int ret; yading@10: if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 || yading@10: (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 || yading@10: (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0) yading@10: return ret; yading@10: return 0; yading@10: } yading@10: yading@10: static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src) yading@10: { yading@10: Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data; yading@10: int qps_changed = 0, i, err; yading@10: yading@10: #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field) yading@10: yading@10: if (!s1->current_frame.f->data[0] yading@10: ||s->width != s1->width yading@10: ||s->height!= s1->height) { yading@10: if (s != s1) yading@10: ref_frames(s, s1); yading@10: return -1; yading@10: } yading@10: yading@10: if (s != s1) { yading@10: // init tables if the first frame hasn't been decoded yading@10: if (!s->current_frame.f->data[0]) { yading@10: int y_fragment_count, c_fragment_count; yading@10: s->avctx = dst; yading@10: err = allocate_tables(dst); yading@10: if (err) yading@10: return err; yading@10: y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; yading@10: c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; yading@10: memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0])); yading@10: memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1])); yading@10: } yading@10: yading@10: // copy previous frame data yading@10: if ((err = ref_frames(s, s1)) < 0) yading@10: return err; yading@10: yading@10: s->keyframe = s1->keyframe; yading@10: yading@10: // copy qscale data if necessary yading@10: for (i = 0; i < 3; i++) { yading@10: if (s->qps[i] != s1->qps[1]) { yading@10: qps_changed = 1; yading@10: memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i])); yading@10: } yading@10: } yading@10: yading@10: if (s->qps[0] != s1->qps[0]) yading@10: memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array)); yading@10: yading@10: if (qps_changed) yading@10: copy_fields(s, s1, qps, superblock_count); yading@10: #undef copy_fields yading@10: } yading@10: yading@10: return update_frames(dst); yading@10: } yading@10: yading@10: static int vp3_decode_frame(AVCodecContext *avctx, yading@10: void *data, int *got_frame, yading@10: AVPacket *avpkt) yading@10: { yading@10: const uint8_t *buf = avpkt->data; yading@10: int buf_size = avpkt->size; yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: GetBitContext gb; yading@10: int i, ret; yading@10: yading@10: init_get_bits(&gb, buf, buf_size * 8); yading@10: yading@10: #if CONFIG_THEORA_DECODER yading@10: if (s->theora && get_bits1(&gb)) yading@10: { yading@10: int type = get_bits(&gb, 7); yading@10: skip_bits_long(&gb, 6*8); /* "theora" */ yading@10: yading@10: if (s->avctx->active_thread_type&FF_THREAD_FRAME) { yading@10: av_log(avctx, AV_LOG_ERROR, "midstream reconfiguration with multithreading is unsupported, try -threads 1\n"); yading@10: return AVERROR_PATCHWELCOME; yading@10: } yading@10: if (type == 0) { yading@10: vp3_decode_end(avctx); yading@10: ret = theora_decode_header(avctx, &gb); yading@10: yading@10: if (ret < 0) { yading@10: vp3_decode_end(avctx); yading@10: } else yading@10: ret = vp3_decode_init(avctx); yading@10: return ret; yading@10: } else if (type == 2) { yading@10: ret = theora_decode_tables(avctx, &gb); yading@10: if (ret < 0) { yading@10: vp3_decode_end(avctx); yading@10: } else yading@10: ret = vp3_decode_init(avctx); yading@10: return ret; yading@10: } yading@10: yading@10: av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n"); yading@10: return -1; yading@10: } yading@10: #endif yading@10: yading@10: s->keyframe = !get_bits1(&gb); yading@10: if (!s->all_fragments) { yading@10: av_log(avctx, AV_LOG_ERROR, "Data packet without prior valid headers\n"); yading@10: return -1; yading@10: } yading@10: if (!s->theora) yading@10: skip_bits(&gb, 1); yading@10: for (i = 0; i < 3; i++) yading@10: s->last_qps[i] = s->qps[i]; yading@10: yading@10: s->nqps=0; yading@10: do{ yading@10: s->qps[s->nqps++]= get_bits(&gb, 6); yading@10: } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb)); yading@10: for (i = s->nqps; i < 3; i++) yading@10: s->qps[i] = -1; yading@10: yading@10: if (s->avctx->debug & FF_DEBUG_PICT_INFO) yading@10: av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n", yading@10: s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]); yading@10: yading@10: s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] || yading@10: avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY); yading@10: yading@10: if (s->qps[0] != s->last_qps[0]) yading@10: init_loop_filter(s); yading@10: yading@10: for (i = 0; i < s->nqps; i++) yading@10: // reinit all dequantizers if the first one changed, because yading@10: // the DC of the first quantizer must be used for all matrices yading@10: if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0]) yading@10: init_dequantizer(s, i); yading@10: yading@10: if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe) yading@10: return buf_size; yading@10: yading@10: s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P; yading@10: s->current_frame.f->key_frame = s->keyframe; yading@10: if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0) yading@10: goto error; yading@10: yading@10: if (!s->edge_emu_buffer) yading@10: s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.f->linesize[0])); yading@10: yading@10: if (s->keyframe) { yading@10: if (!s->theora) yading@10: { yading@10: skip_bits(&gb, 4); /* width code */ yading@10: skip_bits(&gb, 4); /* height code */ yading@10: if (s->version) yading@10: { yading@10: s->version = get_bits(&gb, 5); yading@10: if (avctx->frame_number == 0) yading@10: av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version); yading@10: } yading@10: } yading@10: if (s->version || s->theora) yading@10: { yading@10: if (get_bits1(&gb)) yading@10: av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n"); yading@10: skip_bits(&gb, 2); /* reserved? */ yading@10: } yading@10: } else { yading@10: if (!s->golden_frame.f->data[0]) { yading@10: av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n"); yading@10: yading@10: s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I; yading@10: if (ff_thread_get_buffer(avctx, &s->golden_frame, AV_GET_BUFFER_FLAG_REF) < 0) yading@10: goto error; yading@10: ff_thread_release_buffer(avctx, &s->last_frame); yading@10: if ((ret = ff_thread_ref_frame(&s->last_frame, &s->golden_frame)) < 0) yading@10: goto error; yading@10: ff_thread_report_progress(&s->last_frame, INT_MAX, 0); yading@10: } yading@10: } yading@10: yading@10: memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment)); yading@10: ff_thread_finish_setup(avctx); yading@10: yading@10: if (unpack_superblocks(s, &gb)){ yading@10: av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n"); yading@10: goto error; yading@10: } yading@10: if (unpack_modes(s, &gb)){ yading@10: av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n"); yading@10: goto error; yading@10: } yading@10: if (unpack_vectors(s, &gb)){ yading@10: av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n"); yading@10: goto error; yading@10: } yading@10: if (unpack_block_qpis(s, &gb)){ yading@10: av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n"); yading@10: goto error; yading@10: } yading@10: if (unpack_dct_coeffs(s, &gb)){ yading@10: av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n"); yading@10: goto error; yading@10: } yading@10: yading@10: for (i = 0; i < 3; i++) { yading@10: int height = s->height >> (i && s->chroma_y_shift); yading@10: if (s->flipped_image) yading@10: s->data_offset[i] = 0; yading@10: else yading@10: s->data_offset[i] = (height-1) * s->current_frame.f->linesize[i]; yading@10: } yading@10: yading@10: s->last_slice_end = 0; yading@10: for (i = 0; i < s->c_superblock_height; i++) yading@10: render_slice(s, i); yading@10: yading@10: // filter the last row yading@10: for (i = 0; i < 3; i++) { yading@10: int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1; yading@10: apply_loop_filter(s, i, row, row+1); yading@10: } yading@10: vp3_draw_horiz_band(s, s->avctx->height); yading@10: yading@10: if ((ret = av_frame_ref(data, s->current_frame.f)) < 0) yading@10: return ret; yading@10: *got_frame = 1; yading@10: yading@10: if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME)) { yading@10: ret = update_frames(avctx); yading@10: if (ret < 0) yading@10: return ret; yading@10: } yading@10: yading@10: return buf_size; yading@10: yading@10: error: yading@10: ff_thread_report_progress(&s->current_frame, INT_MAX, 0); yading@10: yading@10: if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME)) yading@10: av_frame_unref(s->current_frame.f); yading@10: yading@10: return -1; yading@10: } yading@10: yading@10: static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: yading@10: if (get_bits1(gb)) { yading@10: int token; yading@10: if (s->entries >= 32) { /* overflow */ yading@10: av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n"); yading@10: return -1; yading@10: } yading@10: token = get_bits(gb, 5); yading@10: av_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n", yading@10: s->hti, s->hbits, token, s->entries, s->huff_code_size); yading@10: s->huffman_table[s->hti][token][0] = s->hbits; yading@10: s->huffman_table[s->hti][token][1] = s->huff_code_size; yading@10: s->entries++; yading@10: } yading@10: else { yading@10: if (s->huff_code_size >= 32) {/* overflow */ yading@10: av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n"); yading@10: return -1; yading@10: } yading@10: s->huff_code_size++; yading@10: s->hbits <<= 1; yading@10: if (read_huffman_tree(avctx, gb)) yading@10: return -1; yading@10: s->hbits |= 1; yading@10: if (read_huffman_tree(avctx, gb)) yading@10: return -1; yading@10: s->hbits >>= 1; yading@10: s->huff_code_size--; yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: static int vp3_init_thread_copy(AVCodecContext *avctx) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: yading@10: s->superblock_coding = NULL; yading@10: s->all_fragments = NULL; yading@10: s->coded_fragment_list[0] = NULL; yading@10: s->dct_tokens_base = NULL; yading@10: s->superblock_fragments = NULL; yading@10: s->macroblock_coding = NULL; yading@10: s->motion_val[0] = NULL; yading@10: s->motion_val[1] = NULL; yading@10: s->edge_emu_buffer = NULL; yading@10: yading@10: return init_frames(s); yading@10: } yading@10: yading@10: #if CONFIG_THEORA_DECODER yading@10: static const enum AVPixelFormat theora_pix_fmts[4] = { yading@10: AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P yading@10: }; yading@10: yading@10: static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: int visible_width, visible_height, colorspace; yading@10: int offset_x = 0, offset_y = 0; yading@10: AVRational fps, aspect; yading@10: yading@10: s->theora = get_bits_long(gb, 24); yading@10: av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora); yading@10: yading@10: /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */ yading@10: /* but previous versions have the image flipped relative to vp3 */ yading@10: if (s->theora < 0x030200) yading@10: { yading@10: s->flipped_image = 1; yading@10: av_log(avctx, AV_LOG_DEBUG, "Old (width = get_bits(gb, 16) << 4; yading@10: visible_height = s->height = get_bits(gb, 16) << 4; yading@10: yading@10: if(av_image_check_size(s->width, s->height, 0, avctx)){ yading@10: av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height); yading@10: s->width= s->height= 0; yading@10: return -1; yading@10: } yading@10: yading@10: if (s->theora >= 0x030200) { yading@10: visible_width = get_bits_long(gb, 24); yading@10: visible_height = get_bits_long(gb, 24); yading@10: yading@10: offset_x = get_bits(gb, 8); /* offset x */ yading@10: offset_y = get_bits(gb, 8); /* offset y, from bottom */ yading@10: } yading@10: yading@10: fps.num = get_bits_long(gb, 32); yading@10: fps.den = get_bits_long(gb, 32); yading@10: if (fps.num>0 && fps.den>0) { yading@10: av_reduce(&avctx->time_base.num, &avctx->time_base.den, yading@10: fps.den, fps.num, 1<<30); yading@10: } yading@10: yading@10: aspect.num = get_bits_long(gb, 24); yading@10: aspect.den = get_bits_long(gb, 24); yading@10: if (aspect.num && aspect.den) { yading@10: av_reduce(&avctx->sample_aspect_ratio.num, yading@10: &avctx->sample_aspect_ratio.den, yading@10: aspect.num, aspect.den, 1<<30); yading@10: } yading@10: yading@10: if (s->theora < 0x030200) yading@10: skip_bits(gb, 5); /* keyframe frequency force */ yading@10: colorspace = get_bits(gb, 8); yading@10: skip_bits(gb, 24); /* bitrate */ yading@10: yading@10: skip_bits(gb, 6); /* quality hint */ yading@10: yading@10: if (s->theora >= 0x030200) yading@10: { yading@10: skip_bits(gb, 5); /* keyframe frequency force */ yading@10: avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)]; yading@10: if (avctx->pix_fmt == AV_PIX_FMT_NONE) { yading@10: av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: skip_bits(gb, 3); /* reserved */ yading@10: } yading@10: yading@10: // align_get_bits(gb); yading@10: yading@10: if ( visible_width <= s->width && visible_width > s->width-16 yading@10: && visible_height <= s->height && visible_height > s->height-16 yading@10: && !offset_x && (offset_y == s->height - visible_height)) yading@10: avcodec_set_dimensions(avctx, visible_width, visible_height); yading@10: else yading@10: avcodec_set_dimensions(avctx, s->width, s->height); yading@10: yading@10: if (colorspace == 1) { yading@10: avctx->color_primaries = AVCOL_PRI_BT470M; yading@10: } else if (colorspace == 2) { yading@10: avctx->color_primaries = AVCOL_PRI_BT470BG; yading@10: } yading@10: if (colorspace == 1 || colorspace == 2) { yading@10: avctx->colorspace = AVCOL_SPC_BT470BG; yading@10: avctx->color_trc = AVCOL_TRC_BT709; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: int i, n, matrices, inter, plane; yading@10: yading@10: if (s->theora >= 0x030200) { yading@10: n = get_bits(gb, 3); yading@10: /* loop filter limit values table */ yading@10: if (n) yading@10: for (i = 0; i < 64; i++) yading@10: s->filter_limit_values[i] = get_bits(gb, n); yading@10: } yading@10: yading@10: if (s->theora >= 0x030200) yading@10: n = get_bits(gb, 4) + 1; yading@10: else yading@10: n = 16; yading@10: /* quality threshold table */ yading@10: for (i = 0; i < 64; i++) yading@10: s->coded_ac_scale_factor[i] = get_bits(gb, n); yading@10: yading@10: if (s->theora >= 0x030200) yading@10: n = get_bits(gb, 4) + 1; yading@10: else yading@10: n = 16; yading@10: /* dc scale factor table */ yading@10: for (i = 0; i < 64; i++) yading@10: s->coded_dc_scale_factor[i] = get_bits(gb, n); yading@10: yading@10: if (s->theora >= 0x030200) yading@10: matrices = get_bits(gb, 9) + 1; yading@10: else yading@10: matrices = 3; yading@10: yading@10: if(matrices > 384){ yading@10: av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n"); yading@10: return -1; yading@10: } yading@10: yading@10: for(n=0; nbase_matrix[n][i]= get_bits(gb, 8); yading@10: } yading@10: yading@10: for (inter = 0; inter <= 1; inter++) { yading@10: for (plane = 0; plane <= 2; plane++) { yading@10: int newqr= 1; yading@10: if (inter || plane > 0) yading@10: newqr = get_bits1(gb); yading@10: if (!newqr) { yading@10: int qtj, plj; yading@10: if(inter && get_bits1(gb)){ yading@10: qtj = 0; yading@10: plj = plane; yading@10: }else{ yading@10: qtj= (3*inter + plane - 1) / 3; yading@10: plj= (plane + 2) % 3; yading@10: } yading@10: s->qr_count[inter][plane]= s->qr_count[qtj][plj]; yading@10: memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0])); yading@10: memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0])); yading@10: } else { yading@10: int qri= 0; yading@10: int qi = 0; yading@10: yading@10: for(;;){ yading@10: i= get_bits(gb, av_log2(matrices-1)+1); yading@10: if(i>= matrices){ yading@10: av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n"); yading@10: return -1; yading@10: } yading@10: s->qr_base[inter][plane][qri]= i; yading@10: if(qi >= 63) yading@10: break; yading@10: i = get_bits(gb, av_log2(63-qi)+1) + 1; yading@10: s->qr_size[inter][plane][qri++]= i; yading@10: qi += i; yading@10: } yading@10: yading@10: if (qi > 63) { yading@10: av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi); yading@10: return -1; yading@10: } yading@10: s->qr_count[inter][plane]= qri; yading@10: } yading@10: } yading@10: } yading@10: yading@10: /* Huffman tables */ yading@10: for (s->hti = 0; s->hti < 80; s->hti++) { yading@10: s->entries = 0; yading@10: s->huff_code_size = 1; yading@10: if (!get_bits1(gb)) { yading@10: s->hbits = 0; yading@10: if(read_huffman_tree(avctx, gb)) yading@10: return -1; yading@10: s->hbits = 1; yading@10: if(read_huffman_tree(avctx, gb)) yading@10: return -1; yading@10: } yading@10: } yading@10: yading@10: s->theora_tables = 1; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int theora_decode_init(AVCodecContext *avctx) yading@10: { yading@10: Vp3DecodeContext *s = avctx->priv_data; yading@10: GetBitContext gb; yading@10: int ptype; yading@10: uint8_t *header_start[3]; yading@10: int header_len[3]; yading@10: int i; yading@10: yading@10: avctx->pix_fmt = AV_PIX_FMT_YUV420P; yading@10: yading@10: s->theora = 1; yading@10: yading@10: if (!avctx->extradata_size) yading@10: { yading@10: av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n"); yading@10: return -1; yading@10: } yading@10: yading@10: if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size, yading@10: 42, header_start, header_len) < 0) { yading@10: av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n"); yading@10: return -1; yading@10: } yading@10: yading@10: for(i=0;i<3;i++) { yading@10: if (header_len[i] <= 0) yading@10: continue; yading@10: init_get_bits(&gb, header_start[i], header_len[i] * 8); yading@10: yading@10: ptype = get_bits(&gb, 8); yading@10: yading@10: if (!(ptype & 0x80)) yading@10: { yading@10: av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n"); yading@10: // return -1; yading@10: } yading@10: yading@10: // FIXME: Check for this as well. yading@10: skip_bits_long(&gb, 6*8); /* "theora" */ yading@10: yading@10: switch(ptype) yading@10: { yading@10: case 0x80: yading@10: if (theora_decode_header(avctx, &gb) < 0) yading@10: return -1; yading@10: break; yading@10: case 0x81: yading@10: // FIXME: is this needed? it breaks sometimes yading@10: // theora_decode_comments(avctx, gb); yading@10: break; yading@10: case 0x82: yading@10: if (theora_decode_tables(avctx, &gb)) yading@10: return -1; yading@10: break; yading@10: default: yading@10: av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80); yading@10: break; yading@10: } yading@10: if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb)) yading@10: av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype); yading@10: if (s->theora < 0x030200) yading@10: break; yading@10: } yading@10: yading@10: return vp3_decode_init(avctx); yading@10: } yading@10: yading@10: AVCodec ff_theora_decoder = { yading@10: .name = "theora", yading@10: .type = AVMEDIA_TYPE_VIDEO, yading@10: .id = AV_CODEC_ID_THEORA, yading@10: .priv_data_size = sizeof(Vp3DecodeContext), yading@10: .init = theora_decode_init, yading@10: .close = vp3_decode_end, yading@10: .decode = vp3_decode_frame, yading@10: .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | yading@10: CODEC_CAP_FRAME_THREADS, yading@10: .flush = vp3_decode_flush, yading@10: .long_name = NULL_IF_CONFIG_SMALL("Theora"), yading@10: .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy), yading@10: .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context) yading@10: }; yading@10: #endif yading@10: yading@10: AVCodec ff_vp3_decoder = { yading@10: .name = "vp3", yading@10: .type = AVMEDIA_TYPE_VIDEO, yading@10: .id = AV_CODEC_ID_VP3, yading@10: .priv_data_size = sizeof(Vp3DecodeContext), yading@10: .init = vp3_decode_init, yading@10: .close = vp3_decode_end, yading@10: .decode = vp3_decode_frame, yading@10: .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | yading@10: CODEC_CAP_FRAME_THREADS, yading@10: .flush = vp3_decode_flush, yading@10: .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"), yading@10: .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy), yading@10: .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context), yading@10: };