X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fvp3.c;h=adcecbcf043a20d88530cc2947007f6014bde5a9;hb=282c6a1a0ef3e8c8e180c15b39cfe5b89704d848;hp=e6693327408de6622db9c244275d3f1dcf8f0dd1;hpb=ecb51b25bb3511f3e251d7a3ead653eda9a9179d;p=ffmpeg diff --git a/libavcodec/vp3.c b/libavcodec/vp3.c index e6693327408..adcecbcf043 100644 --- a/libavcodec/vp3.c +++ b/libavcodec/vp3.c @@ -1,25 +1,25 @@ /* * Copyright (C) 2003-2004 the ffmpeg project * - * This file is part of FFmpeg. + * This file is part of Libav. * - * FFmpeg is free software; you can redistribute it and/or + * Libav is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * - * FFmpeg is distributed in the hope that it will be useful, + * Libav is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software + * License along with Libav; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ /** - * @file libavcodec/vp3.c + * @file * On2 VP3 Video Decoder * * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx) @@ -33,29 +33,24 @@ #include #include +#include "libavutil/imgutils.h" #include "avcodec.h" +#include "internal.h" #include "dsputil.h" #include "get_bits.h" - +#include "hpeldsp.h" +#include "videodsp.h" #include "vp3data.h" +#include "vp3dsp.h" #include "xiph.h" +#include "thread.h" #define FRAGMENT_PIXELS 8 -static av_cold int vp3_decode_end(AVCodecContext *avctx); - -typedef struct Coeff { - struct Coeff *next; - DCTELEM coeff; - uint8_t index; -} Coeff; - //FIXME split things out into their own arrays typedef struct Vp3Fragment { - Coeff *next_coeff; + int16_t dc; uint8_t coding_method; - int8_t motion_x; - int8_t motion_y; uint8_t qpi; } Vp3Fragment; @@ -122,6 +117,13 @@ static const int ModeAlphabet[6][CODING_MODE_COUNT] = }; +static const uint8_t hilbert_offset[16][2] = { + {0,0}, {1,0}, {1,1}, {0,1}, + {0,2}, {0,3}, {1,3}, {1,2}, + {2,2}, {2,3}, {3,3}, {3,2}, + {3,1}, {2,1}, {2,0}, {3,0} +}; + #define MIN_DEQUANT_VAL 2 typedef struct Vp3DecodeContext { @@ -129,13 +131,20 @@ typedef struct Vp3DecodeContext { int theora, theora_tables; int version; int width, height; - AVFrame golden_frame; - AVFrame last_frame; - AVFrame current_frame; + int chroma_x_shift, chroma_y_shift; + ThreadFrame golden_frame; + ThreadFrame last_frame; + ThreadFrame current_frame; int keyframe; - DSPContext dsp; + uint8_t idct_permutation[64]; + uint8_t idct_scantable[64]; + HpelDSPContext hdsp; + VideoDSPContext vdsp; + VP3DSPContext vp3dsp; + DECLARE_ALIGNED(16, int16_t, block)[64]; int flipped_image; int last_slice_end; + int skip_loop_filter; int qps[3]; int nqps; @@ -144,8 +153,10 @@ typedef struct Vp3DecodeContext { int superblock_count; int y_superblock_width; int y_superblock_height; + int y_superblock_count; int c_superblock_width; int c_superblock_height; + int c_superblock_count; int u_superblock_start; int v_superblock_start; unsigned char *superblock_coding; @@ -155,17 +166,14 @@ typedef struct Vp3DecodeContext { int macroblock_height; int fragment_count; - int fragment_width; - int fragment_height; + int fragment_width[2]; + int fragment_height[2]; Vp3Fragment *all_fragments; - uint8_t *coeff_counts; - Coeff *coeffs; - Coeff *next_coeff; int fragment_start[3]; int data_offset[3]; - ScanTable scantable; + int8_t (*motion_val[2])[2]; /* tables */ uint16_t coded_dc_scale_factor[64]; @@ -175,18 +183,38 @@ typedef struct Vp3DecodeContext { uint8_t qr_size [2][3][64]; uint16_t qr_base[2][3][64]; + /** + * This is a list of all tokens in bitstream order. Reordering takes place + * by pulling from each level during IDCT. As a consequence, IDCT must be + * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32 + * otherwise. The 32 different tokens with up to 12 bits of extradata are + * collapsed into 3 types, packed as follows: + * (from the low to high bits) + * + * 2 bits: type (0,1,2) + * 0: EOB run, 14 bits for run length (12 needed) + * 1: zero run, 7 bits for run length + * 7 bits for the next coefficient (3 needed) + * 2: coefficient, 14 bits (11 needed) + * + * Coefficients are signed, so are packed in the highest bits for automatic + * sign extension. + */ + int16_t *dct_tokens[3][64]; + int16_t *dct_tokens_base; +#define TOKEN_EOB(eob_run) ((eob_run) << 2) +#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1) +#define TOKEN_COEFF(coeff) (((coeff) << 2) + 2) + + /** + * number of blocks that contain DCT coefficients at the given level or higher + */ + int num_coded_frags[3][64]; + int total_num_coded_frags; + /* this is a list of indexes into the all_fragments array indicating * which of the fragments are coded */ - int *coded_fragment_list; - int coded_fragment_list_index; - - /* track which fragments have already been decoded; called 'fast' - * because this data structure avoids having to iterate through every - * fragment in coded_fragment_list; once a fragment has been fully - * decoded, it is removed from this list */ - int *fast_fragment_list; - int fragment_list_y_head; - int fragment_list_c_head; + int *coded_fragment_list[3]; VLC dc_vlc[16]; VLC ac_vlc_1[16]; @@ -201,7 +229,7 @@ typedef struct Vp3DecodeContext { /* these arrays need to be on 16-byte boundaries since SSE2 operations * index into them */ - DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64]; // fragments, macroblocks <-> fragments, - * superblocks <-> macroblocks - * - * Returns 0 is successful; returns 1 if *anything* went wrong. - */ -static int init_block_mapping(Vp3DecodeContext *s) +static void vp3_decode_flush(AVCodecContext *avctx) { - int i, j; - signed int hilbert_walk_mb[4]; - - int current_fragment = 0; - int current_width = 0; - int current_height = 0; - int right_edge = 0; - int bottom_edge = 0; - int superblock_row_inc = 0; - int mapping_index = 0; - - int current_macroblock; - int c_fragment; + Vp3DecodeContext *s = avctx->priv_data; - static const signed char travel_width[16] = { - 1, 1, 0, -1, - 0, 0, 1, 0, - 1, 0, 1, 0, - 0, -1, 0, 1 - }; + if (s->golden_frame.f) + ff_thread_release_buffer(avctx, &s->golden_frame); + if (s->last_frame.f) + ff_thread_release_buffer(avctx, &s->last_frame); + if (s->current_frame.f) + ff_thread_release_buffer(avctx, &s->current_frame); +} - static const signed char travel_height[16] = { - 0, 0, 1, 0, - 1, 1, 0, -1, - 0, 1, 0, -1, - -1, 0, -1, 0 - }; +static av_cold int vp3_decode_end(AVCodecContext *avctx) +{ + Vp3DecodeContext *s = avctx->priv_data; + int i; - hilbert_walk_mb[0] = 1; - hilbert_walk_mb[1] = s->macroblock_width; - hilbert_walk_mb[2] = 1; - hilbert_walk_mb[3] = -s->macroblock_width; - - /* iterate through each superblock (all planes) and map the fragments */ - for (i = 0; i < s->superblock_count; i++) { - /* time to re-assign the limits? */ - if (i == 0) { - - /* start of Y superblocks */ - right_edge = s->fragment_width; - bottom_edge = s->fragment_height; - current_width = -1; - current_height = 0; - superblock_row_inc = 3 * s->fragment_width - - (s->y_superblock_width * 4 - s->fragment_width); - - /* the first operation for this variable is to advance by 1 */ - current_fragment = -1; - - } else if (i == s->u_superblock_start) { - - /* start of U superblocks */ - right_edge = s->fragment_width / 2; - bottom_edge = s->fragment_height / 2; - current_width = -1; - current_height = 0; - superblock_row_inc = 3 * (s->fragment_width / 2) - - (s->c_superblock_width * 4 - s->fragment_width / 2); - - /* the first operation for this variable is to advance by 1 */ - current_fragment = s->fragment_start[1] - 1; - - } else if (i == s->v_superblock_start) { - - /* start of V superblocks */ - right_edge = s->fragment_width / 2; - bottom_edge = s->fragment_height / 2; - current_width = -1; - current_height = 0; - superblock_row_inc = 3 * (s->fragment_width / 2) - - (s->c_superblock_width * 4 - s->fragment_width / 2); - - /* the first operation for this variable is to advance by 1 */ - current_fragment = s->fragment_start[2] - 1; + av_freep(&s->superblock_coding); + av_freep(&s->all_fragments); + av_freep(&s->coded_fragment_list[0]); + av_freep(&s->dct_tokens_base); + av_freep(&s->superblock_fragments); + av_freep(&s->macroblock_coding); + av_freep(&s->motion_val[0]); + av_freep(&s->motion_val[1]); + av_freep(&s->edge_emu_buffer); - } + /* release all frames */ + vp3_decode_flush(avctx); + av_frame_free(&s->current_frame.f); + av_frame_free(&s->last_frame.f); + av_frame_free(&s->golden_frame.f); - if (current_width >= right_edge - 1) { - /* reset width and move to next superblock row */ - current_width = -1; - current_height += 4; + if (avctx->internal->is_copy) + return 0; - /* fragment is now at the start of a new superblock row */ - current_fragment += superblock_row_inc; - } + for (i = 0; i < 16; i++) { + ff_free_vlc(&s->dc_vlc[i]); + ff_free_vlc(&s->ac_vlc_1[i]); + ff_free_vlc(&s->ac_vlc_2[i]); + ff_free_vlc(&s->ac_vlc_3[i]); + ff_free_vlc(&s->ac_vlc_4[i]); + } - /* iterate through all 16 fragments in a superblock */ - for (j = 0; j < 16; j++) { - current_fragment += travel_width[j] + right_edge * travel_height[j]; - current_width += travel_width[j]; - current_height += travel_height[j]; - - /* check if the fragment is in bounds */ - if ((current_width < right_edge) && - (current_height < bottom_edge)) { - s->superblock_fragments[mapping_index] = current_fragment; - } else { - s->superblock_fragments[mapping_index] = -1; - } + ff_free_vlc(&s->superblock_run_length_vlc); + ff_free_vlc(&s->fragment_run_length_vlc); + ff_free_vlc(&s->mode_code_vlc); + ff_free_vlc(&s->motion_vector_vlc); - mapping_index++; - } - } - return 0; /* successful path out */ + return 0; } /* - * This function wipes out all of the fragment data. + * This function sets up all of the various blocks mappings: + * superblocks <-> fragments, macroblocks <-> fragments, + * superblocks <-> macroblocks + * + * @return 0 is successful; returns 1 if *anything* went wrong. */ -static void init_frame(Vp3DecodeContext *s, GetBitContext *gb) +static int init_block_mapping(Vp3DecodeContext *s) { - int i; + int sb_x, sb_y, plane; + int x, y, i, j = 0; - /* zero out all of the fragment information */ - s->coded_fragment_list_index = 0; - for (i = 0; i < s->fragment_count; i++) { - s->coeff_counts[i] = 0; - s->all_fragments[i].motion_x = 127; - s->all_fragments[i].motion_y = 127; - s->all_fragments[i].next_coeff= NULL; - s->all_fragments[i].qpi = 0; - s->coeffs[i].index= - s->coeffs[i].coeff=0; - s->coeffs[i].next= NULL; + for (plane = 0; plane < 3; plane++) { + int sb_width = plane ? s->c_superblock_width : s->y_superblock_width; + int sb_height = plane ? s->c_superblock_height : s->y_superblock_height; + int frag_width = s->fragment_width[!!plane]; + int frag_height = s->fragment_height[!!plane]; + + for (sb_y = 0; sb_y < sb_height; sb_y++) + for (sb_x = 0; sb_x < sb_width; sb_x++) + for (i = 0; i < 16; i++) { + x = 4*sb_x + hilbert_offset[i][0]; + y = 4*sb_y + hilbert_offset[i][1]; + + if (x < frag_width && y < frag_height) + s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x; + else + s->superblock_fragments[j++] = -1; + } } + + return 0; /* successful path out */ } /* @@ -403,14 +375,13 @@ static void init_dequantizer(Vp3DecodeContext *s, int qpi) int qmin= 8<<(inter + !i); int qscale= i ? ac_scale_factor : dc_scale_factor; - s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096); + s->qmat[qpi][inter][plane][s->idct_permutation[i]] = + av_clip((qscale * coeff) / 100 * 4, qmin, 4096); } // all DC coefficients use the same quant so as not to interfere with DC prediction s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0]; } } - - memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune } /* @@ -427,6 +398,7 @@ static void init_loop_filter(Vp3DecodeContext *s) int value; filter_limit = s->filter_limit_values[s->qps[0]]; + assert(filter_limit < 128); /* set up the bounding values */ memset(s->bounding_values_array, 0, 256 * sizeof(int)); @@ -449,15 +421,15 @@ static void init_loop_filter(Vp3DecodeContext *s) */ static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) { + int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start }; int bit = 0; int current_superblock = 0; int current_run = 0; - int decode_fully_flags = 0; - int decode_partial_blocks = 0; - int first_c_fragment_seen; + int num_partial_superblocks = 0; int i, j; int current_fragment; + int plane; if (s->keyframe) { memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count); @@ -465,8 +437,15 @@ static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) } else { /* unpack the list of partially-coded superblocks */ - bit = get_bits1(gb); - while (current_superblock < s->superblock_count) { + bit = get_bits1(gb) ^ 1; + current_run = 0; + + while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) { + if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN) + bit = get_bits1(gb); + else + bit ^= 1; + current_run = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1; if (current_run == 34) @@ -480,52 +459,51 @@ static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) memset(s->superblock_coding + current_superblock, bit, current_run); current_superblock += current_run; - - /* if any of the superblocks are not partially coded, flag - * a boolean to decode the list of fully-coded superblocks */ - if (bit == 0) { - decode_fully_flags = 1; - } else { - - /* make a note of the fact that there are partially coded - * superblocks */ - decode_partial_blocks = 1; - } - - bit ^= 1; + if (bit) + num_partial_superblocks += current_run; } /* unpack the list of fully coded superblocks if any of the blocks were * not marked as partially coded in the previous step */ - if (decode_fully_flags) { + if (num_partial_superblocks < s->superblock_count) { + int superblocks_decoded = 0; current_superblock = 0; + bit = get_bits1(gb) ^ 1; current_run = 0; - bit = get_bits1(gb); - /* toggle the bit because as soon as the first run length is - * fetched the bit will be toggled again */ - bit ^= 1; - while (current_superblock < s->superblock_count) { - /* skip any superblocks already marked as partially coded */ - if (s->superblock_coding[current_superblock] == SB_NOT_CODED) { + while (superblocks_decoded < s->superblock_count - num_partial_superblocks + && get_bits_left(gb) > 0) { + + if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN) + bit = get_bits1(gb); + else + bit ^= 1; - if (current_run-- == 0) { - bit ^= 1; current_run = get_vlc2(gb, - s->superblock_run_length_vlc.table, 6, 2); - if (current_run == 33) + s->superblock_run_length_vlc.table, 6, 2) + 1; + if (current_run == 34) current_run += get_bits(gb, 12); + + for (j = 0; j < current_run; current_superblock++) { + if (current_superblock >= s->superblock_count) { + av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n"); + return -1; } + + /* skip any superblocks already marked as partially coded */ + if (s->superblock_coding[current_superblock] == SB_NOT_CODED) { s->superblock_coding[current_superblock] = 2*bit; + j++; + } } - current_superblock++; + superblocks_decoded += current_run; } } /* if there were partial blocks, initialize bitstream for * unpacking fragment codings */ - if (decode_partial_blocks) { + if (num_partial_superblocks) { current_run = 0; bit = get_bits1(gb); @@ -537,32 +515,25 @@ static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) /* figure out which fragments are coded; iterate through each * superblock (all planes) */ - s->coded_fragment_list_index = 0; - s->next_coeff= s->coeffs + s->fragment_count; - s->first_coded_y_fragment = s->first_coded_c_fragment = 0; - s->last_coded_y_fragment = s->last_coded_c_fragment = -1; - first_c_fragment_seen = 0; + s->total_num_coded_frags = 0; memset(s->macroblock_coding, MODE_COPY, s->macroblock_count); - for (i = 0; i < s->superblock_count; i++) { + + for (plane = 0; plane < 3; plane++) { + int sb_start = superblock_starts[plane]; + int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count); + int num_coded_frags = 0; + + for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) { /* iterate through all 16 fragments in a superblock */ for (j = 0; j < 16; j++) { /* if the fragment is in bounds, check its coding status */ current_fragment = s->superblock_fragments[i * 16 + j]; - if (current_fragment >= s->fragment_count) { - av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n", - current_fragment, s->fragment_count); - return 1; - } if (current_fragment != -1) { - if (s->superblock_coding[i] == SB_NOT_CODED) { - - /* copy all the fragments from the prior frame */ - s->all_fragments[current_fragment].coding_method = - MODE_COPY; + int coded = s->superblock_coding[i]; - } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) { + if (s->superblock_coding[i] == SB_PARTIALLY_CODED) { /* fragment may or may not be coded; this is the case * that cares about the fragment coding runs */ @@ -571,77 +542,30 @@ static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2); } + coded = bit; + } - if (bit) { + if (coded) { /* default mode; actual mode will be decoded in * the next phase */ s->all_fragments[current_fragment].coding_method = MODE_INTER_NO_MV; - s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment; - s->coded_fragment_list[s->coded_fragment_list_index] = + s->coded_fragment_list[plane][num_coded_frags++] = current_fragment; - if ((current_fragment >= s->fragment_start[1]) && - (s->last_coded_y_fragment == -1) && - (!first_c_fragment_seen)) { - s->first_coded_c_fragment = s->coded_fragment_list_index; - s->last_coded_y_fragment = s->first_coded_c_fragment - 1; - first_c_fragment_seen = 1; - } - s->coded_fragment_list_index++; } else { /* not coded; copy this fragment from the prior frame */ s->all_fragments[current_fragment].coding_method = MODE_COPY; } - - } else { - - /* fragments are fully coded in this superblock; actual - * coding will be determined in next step */ - s->all_fragments[current_fragment].coding_method = - MODE_INTER_NO_MV; - s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment; - s->coded_fragment_list[s->coded_fragment_list_index] = - current_fragment; - if ((current_fragment >= s->fragment_start[1]) && - (s->last_coded_y_fragment == -1) && - (!first_c_fragment_seen)) { - s->first_coded_c_fragment = s->coded_fragment_list_index; - s->last_coded_y_fragment = s->first_coded_c_fragment - 1; - first_c_fragment_seen = 1; - } - s->coded_fragment_list_index++; - } } } } - - if (!first_c_fragment_seen) - /* only Y fragments coded in this frame */ - s->last_coded_y_fragment = s->coded_fragment_list_index - 1; - else - /* end the list of coded C fragments */ - s->last_coded_c_fragment = s->coded_fragment_list_index - 1; - - for (i = 0; i < s->fragment_count - 1; i++) { - s->fast_fragment_list[i] = i + 1; - } - s->fast_fragment_list[s->fragment_count - 1] = -1; - - if (s->last_coded_y_fragment == -1) - s->fragment_list_y_head = -1; - else { - s->fragment_list_y_head = s->first_coded_y_fragment; - s->fast_fragment_list[s->last_coded_y_fragment] = -1; - } - - if (s->last_coded_c_fragment == -1) - s->fragment_list_c_head = -1; - else { - s->fragment_list_c_head = s->first_coded_c_fragment; - s->fast_fragment_list[s->last_coded_c_fragment] = -1; + s->total_num_coded_frags += num_coded_frags; + for (i = 0; i < 64; i++) + s->num_coded_frags[plane][i] = num_coded_frags; + if (plane < 2) + s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags; } - return 0; } @@ -658,6 +582,7 @@ static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb) int coding_mode; int custom_mode_alphabet[CODING_MODE_COUNT]; const int *alphabet; + Vp3Fragment *frag; if (s->keyframe) { for (i = 0; i < s->fragment_count; i++) @@ -682,11 +607,12 @@ static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb) * coded fragments */ for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) { for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) { + if (get_bits_left(gb) <= 0) + return -1; for (j = 0; j < 4; j++) { int mb_x = 2*sb_x + (j>>1); int mb_y = 2*sb_y + (((j>>1)+j)&1); - int frags_coded = 0; current_macroblock = mb_y * s->macroblock_width + mb_x; if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height) @@ -697,7 +623,7 @@ static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb) /* coding modes are only stored if the macroblock has at least one * luma block coded, otherwise it must be INTER_NO_MV */ for (k = 0; k < 4; k++) { - current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X; + current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X; if (s->all_fragments[current_fragment].coding_method != MODE_COPY) break; } @@ -715,20 +641,31 @@ static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb) s->macroblock_coding[current_macroblock] = coding_mode; for (k = 0; k < 4; k++) { - current_fragment = - BLOCK_Y*s->fragment_width + BLOCK_X; - if (s->all_fragments[current_fragment].coding_method != - MODE_COPY) - s->all_fragments[current_fragment].coding_method = - coding_mode; + frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X; + if (frag->coding_method != MODE_COPY) + frag->coding_method = coding_mode; } - for (k = 0; k < 2; k++) { - current_fragment = s->fragment_start[k+1] + - mb_y*(s->fragment_width>>1) + mb_x; - if (s->all_fragments[current_fragment].coding_method != - MODE_COPY) - s->all_fragments[current_fragment].coding_method = - coding_mode; + +#define SET_CHROMA_MODES \ + if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \ + frag[s->fragment_start[1]].coding_method = coding_mode;\ + if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \ + frag[s->fragment_start[2]].coding_method = coding_mode; + + if (s->chroma_y_shift) { + frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x; + SET_CHROMA_MODES + } else if (s->chroma_x_shift) { + frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x; + for (k = 0; k < 2; k++) { + SET_CHROMA_MODES + frag += s->fragment_width[1]; + } + } else { + for (k = 0; k < 4; k++) { + frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X; + SET_CHROMA_MODES + } } } } @@ -746,21 +683,19 @@ static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) { int j, k, sb_x, sb_y; int coding_mode; - int motion_x[6]; - int motion_y[6]; + int motion_x[4]; + int motion_y[4]; int last_motion_x = 0; int last_motion_y = 0; int prior_last_motion_x = 0; int prior_last_motion_y = 0; int current_macroblock; int current_fragment; + int frag; if (s->keyframe) return 0; - memset(motion_x, 0, 6 * sizeof(int)); - memset(motion_y, 0, 6 * sizeof(int)); - /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */ coding_mode = get_bits1(gb); @@ -768,6 +703,8 @@ static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) * coded fragments */ for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) { for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) { + if (get_bits_left(gb) <= 0) + return -1; for (j = 0; j < 4; j++) { int mb_x = 2*sb_x + (j>>1); @@ -808,9 +745,8 @@ static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) /* fetch 4 vectors from the bitstream, one for each * Y fragment, then average for the C fragment vectors */ - motion_x[4] = motion_y[4] = 0; for (k = 0; k < 4; k++) { - current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X; + current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X; if (s->all_fragments[current_fragment].coding_method != MODE_COPY) { if (coding_mode == 0) { motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; @@ -825,14 +761,7 @@ static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) motion_x[k] = 0; motion_y[k] = 0; } - motion_x[4] += motion_x[k]; - motion_y[4] += motion_y[k]; } - - motion_x[5]= - motion_x[4]= RSHIFT(motion_x[4], 2); - motion_y[5]= - motion_y[4]= RSHIFT(motion_y[4], 2); break; case MODE_INTER_LAST_MV: @@ -869,24 +798,55 @@ static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) /* assign the motion vectors to the correct fragments */ for (k = 0; k < 4; k++) { current_fragment = - BLOCK_Y*s->fragment_width + BLOCK_X; + BLOCK_Y*s->fragment_width[0] + BLOCK_X; if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { - s->all_fragments[current_fragment].motion_x = motion_x[k]; - s->all_fragments[current_fragment].motion_y = motion_y[k]; + s->motion_val[0][current_fragment][0] = motion_x[k]; + s->motion_val[0][current_fragment][1] = motion_y[k]; } else { - s->all_fragments[current_fragment].motion_x = motion_x[0]; - s->all_fragments[current_fragment].motion_y = motion_y[0]; + s->motion_val[0][current_fragment][0] = motion_x[0]; + s->motion_val[0][current_fragment][1] = motion_y[0]; } } - for (k = 0; k < 2; k++) { - current_fragment = s->fragment_start[k+1] + - mb_y*(s->fragment_width>>1) + mb_x; + + if (s->chroma_y_shift) { + if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { + motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2); + motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2); + } + motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1); + motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1); + frag = mb_y*s->fragment_width[1] + mb_x; + s->motion_val[1][frag][0] = motion_x[0]; + s->motion_val[1][frag][1] = motion_y[0]; + } else if (s->chroma_x_shift) { if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { - s->all_fragments[current_fragment].motion_x = motion_x[k+4]; - s->all_fragments[current_fragment].motion_y = motion_y[k+4]; + motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1); + motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1); + motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1); + motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1); } else { - s->all_fragments[current_fragment].motion_x = motion_x[0]; - s->all_fragments[current_fragment].motion_y = motion_y[0]; + motion_x[1] = motion_x[0]; + motion_y[1] = motion_y[0]; + } + motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1); + motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1); + + frag = 2*mb_y*s->fragment_width[1] + mb_x; + for (k = 0; k < 2; k++) { + s->motion_val[1][frag][0] = motion_x[k]; + s->motion_val[1][frag][1] = motion_y[k]; + frag += s->fragment_width[1]; + } + } else { + for (k = 0; k < 4; k++) { + frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X; + if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { + s->motion_val[1][frag][0] = motion_x[k]; + s->motion_val[1][frag][1] = motion_y[k]; + } else { + s->motion_val[1][frag][0] = motion_x[0]; + s->motion_val[1][frag][1] = motion_y[0]; + } } } } @@ -899,14 +859,20 @@ static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb) { int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi; - int num_blocks = s->coded_fragment_list_index; + int num_blocks = s->total_num_coded_frags; for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) { i = blocks_decoded = num_blocks_at_qpi = 0; - bit = get_bits1(gb); + bit = get_bits1(gb) ^ 1; + run_length = 0; do { + if (run_length == MAXIMUM_LONG_BIT_RUN) + bit = get_bits1(gb); + else + bit ^= 1; + run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1; if (run_length == 34) run_length += get_bits(gb, 12); @@ -916,20 +882,15 @@ static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb) num_blocks_at_qpi += run_length; for (j = 0; j < run_length; i++) { - if (i >= s->coded_fragment_list_index) + if (i >= s->total_num_coded_frags) return -1; - if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) { - s->all_fragments[s->coded_fragment_list[i]].qpi += bit; + if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) { + s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit; j++; } } - - if (run_length == MAXIMUM_LONG_BIT_RUN) - bit = get_bits1(gb); - else - bit ^= 1; - } while (blocks_decoded < num_blocks); + } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0); num_blocks -= num_blocks_at_qpi; } @@ -951,58 +912,62 @@ static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb) */ static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb, VLC *table, int coeff_index, - int y_plane, + int plane, int eob_run) { - int i; + int i, j = 0; int token; int zero_run = 0; - DCTELEM coeff = 0; - Vp3Fragment *fragment; + int16_t coeff = 0; int bits_to_get; - int next_fragment; - int previous_fragment; - int fragment_num; - int *list_head; + int blocks_ended; + int coeff_i = 0; + int num_coeffs = s->num_coded_frags[plane][coeff_index]; + int16_t *dct_tokens = s->dct_tokens[plane][coeff_index]; /* local references to structure members to avoid repeated deferences */ - uint8_t *perm= s->scantable.permutated; - int *coded_fragment_list = s->coded_fragment_list; + int *coded_fragment_list = s->coded_fragment_list[plane]; Vp3Fragment *all_fragments = s->all_fragments; - uint8_t *coeff_counts = s->coeff_counts; VLC_TYPE (*vlc_table)[2] = table->table; - int *fast_fragment_list = s->fast_fragment_list; - if (y_plane) { - next_fragment = s->fragment_list_y_head; - list_head = &s->fragment_list_y_head; + if (num_coeffs < 0) + av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index); + + if (eob_run > num_coeffs) { + coeff_i = blocks_ended = num_coeffs; + eob_run -= num_coeffs; } else { - next_fragment = s->fragment_list_c_head; - list_head = &s->fragment_list_c_head; + coeff_i = blocks_ended = eob_run; + eob_run = 0; } - i = next_fragment; - previous_fragment = -1; /* this indicates that the previous fragment is actually the list head */ - while (i != -1) { - fragment_num = coded_fragment_list[i]; + // insert fake EOB token to cover the split between planes or zzi + if (blocks_ended) + dct_tokens[j++] = blocks_ended << 2; - if (coeff_counts[fragment_num] > coeff_index) { - previous_fragment = i; - i = fast_fragment_list[i]; - continue; - } - fragment = &all_fragments[fragment_num]; - - if (!eob_run) { + while (coeff_i < num_coeffs && get_bits_left(gb) > 0) { /* decode a VLC into a token */ - token = get_vlc2(gb, vlc_table, 5, 3); + token = get_vlc2(gb, vlc_table, 11, 3); /* use the token to get a zero run, a coefficient, and an eob run */ - if (token <= 6) { + if ((unsigned) token <= 6U) { eob_run = eob_run_base[token]; if (eob_run_get_bits[token]) eob_run += get_bits(gb, eob_run_get_bits[token]); - coeff = zero_run = 0; - } else { + + // record only the number of blocks ended in this plane, + // any spill will be recorded in the next plane. + if (eob_run > num_coeffs - coeff_i) { + dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i); + blocks_ended += num_coeffs - coeff_i; + eob_run -= num_coeffs - coeff_i; + coeff_i = num_coeffs; + } else { + dct_tokens[j++] = TOKEN_EOB(eob_run); + blocks_ended += eob_run; + coeff_i += eob_run; + eob_run = 0; + } + } else if (token >= 0) { bits_to_get = coeff_get_bits[token]; if (bits_to_get) bits_to_get = get_bits(gb, bits_to_get); @@ -1011,34 +976,53 @@ static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb, zero_run = zero_run_base[token]; if (zero_run_get_bits[token]) zero_run += get_bits(gb, zero_run_get_bits[token]); - } - } - if (!eob_run) { - coeff_counts[fragment_num] += zero_run; - if (coeff_counts[fragment_num] < 64){ - fragment->next_coeff->coeff= coeff; - fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already? - fragment->next_coeff->next= s->next_coeff; - s->next_coeff->next=NULL; - fragment->next_coeff= s->next_coeff++; - } - /* previous fragment is now this fragment */ - previous_fragment = i; - } else { - coeff_counts[fragment_num] |= 128; - eob_run--; - /* remove this fragment from the list */ - if (previous_fragment != -1) - fast_fragment_list[previous_fragment] = fast_fragment_list[i]; - else - *list_head = fast_fragment_list[i]; - /* previous fragment remains unchanged */ - } + if (zero_run) { + dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run); + } else { + // Save DC into the fragment structure. DC prediction is + // done in raster order, so the actual DC can't be in with + // other tokens. We still need the token in dct_tokens[] + // however, or else the structure collapses on itself. + if (!coeff_index) + all_fragments[coded_fragment_list[coeff_i]].dc = coeff; + + dct_tokens[j++] = TOKEN_COEFF(coeff); + } - i = fast_fragment_list[i]; + if (coeff_index + zero_run > 64) { + av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with" + " %d coeffs left\n", zero_run, 64-coeff_index); + zero_run = 64 - coeff_index; + } + + // zero runs code multiple coefficients, + // so don't try to decode coeffs for those higher levels + for (i = coeff_index+1; i <= coeff_index+zero_run; i++) + s->num_coded_frags[plane][i]--; + coeff_i++; + } else { + av_log(s->avctx, AV_LOG_ERROR, + "Invalid token %d\n", token); + return -1; + } } + if (blocks_ended > s->num_coded_frags[plane][coeff_index]) + av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n"); + + // decrement the number of blocks that have higher coeffecients for each + // EOB run at this level + if (blocks_ended) + for (i = coeff_index+1; i < 64; i++) + s->num_coded_frags[plane][i] -= blocks_ended; + + // setup the next buffer + if (plane < 2) + s->dct_tokens[plane+1][coeff_index] = dct_tokens + j; + else if (coeff_index < 63) + s->dct_tokens[0][coeff_index+1] = dct_tokens + j; + return eob_run; } @@ -1061,28 +1045,38 @@ static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb) VLC *y_tables[64]; VLC *c_tables[64]; + s->dct_tokens[0][0] = s->dct_tokens_base; + /* fetch the DC table indexes */ dc_y_table = get_bits(gb, 4); dc_c_table = get_bits(gb, 4); /* unpack the Y plane DC coefficients */ residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0, - 1, residual_eob_run); + 0, residual_eob_run); + if (residual_eob_run < 0) + return residual_eob_run; /* reverse prediction of the Y-plane DC coefficients */ - reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height); + reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]); /* unpack the C plane DC coefficients */ residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, - 0, residual_eob_run); + 1, residual_eob_run); + if (residual_eob_run < 0) + return residual_eob_run; + residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, + 2, residual_eob_run); + if (residual_eob_run < 0) + return residual_eob_run; /* reverse prediction of the C-plane DC coefficients */ if (!(s->avctx->flags & CODEC_FLAG_GRAY)) { reverse_dc_prediction(s, s->fragment_start[1], - s->fragment_width / 2, s->fragment_height / 2); + s->fragment_width[1], s->fragment_height[1]); reverse_dc_prediction(s, s->fragment_start[2], - s->fragment_width / 2, s->fragment_height / 2); + s->fragment_width[1], s->fragment_height[1]); } /* fetch the AC table indexes */ @@ -1109,13 +1103,19 @@ static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb) /* decode all AC coefficents */ for (i = 1; i <= 63; i++) { - if (s->fragment_list_y_head != -1) residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i, - 1, residual_eob_run); + 0, residual_eob_run); + if (residual_eob_run < 0) + return residual_eob_run; - if (s->fragment_list_c_head != -1) residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i, - 0, residual_eob_run); + 1, residual_eob_run); + if (residual_eob_run < 0) + return residual_eob_run; + residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i, + 2, residual_eob_run); + if (residual_eob_run < 0) + return residual_eob_run; } return 0; @@ -1128,7 +1128,7 @@ static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb) */ #define COMPATIBLE_FRAME(x) \ (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type) -#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this +#define DC_COEFF(u) s->all_fragments[u].dc static void reverse_dc_prediction(Vp3DecodeContext *s, int first_fragment, @@ -1272,21 +1272,9 @@ static void reverse_dc_prediction(Vp3DecodeContext *s, } /* at long last, apply the predictor */ - if(s->coeffs[i].index){ - *s->next_coeff= s->coeffs[i]; - s->coeffs[i].index=0; - s->coeffs[i].coeff=0; - s->coeffs[i].next= s->next_coeff++; - } - s->coeffs[i].coeff += predicted_dc; + DC_COEFF(i) += predicted_dc; /* save the DC */ last_dc[current_frame_type] = DC_COEFF(i); - if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){ - s->coeff_counts[i]= 129; -// s->all_fragments[i].next_coeff= s->next_coeff; - s->coeffs[i].next= s->next_coeff; - (s->next_coeff++)->next=NULL; - } } } } @@ -1297,11 +1285,11 @@ static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int ye int x, y; int *bounding_values= s->bounding_values_array+127; - int width = s->fragment_width >> !!plane; - int height = s->fragment_height >> !!plane; + int width = s->fragment_width[!!plane]; + int height = s->fragment_height[!!plane]; int fragment = s->fragment_start [plane] + ystart * width; - int stride = s->current_frame.linesize[plane]; - uint8_t *plane_data = s->current_frame.data [plane]; + ptrdiff_t stride = s->current_frame.f->linesize[plane]; + uint8_t *plane_data = s->current_frame.f->data [plane]; if (!s->flipped_image) stride = -stride; plane_data += s->data_offset[plane] + 8*ystart*stride; @@ -1316,14 +1304,14 @@ static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int ye { /* do not perform left edge filter for left columns frags */ if (x > 0) { - s->dsp.vp3_h_loop_filter( + s->vp3dsp.h_loop_filter( plane_data + 8*x, stride, bounding_values); } /* do not perform top edge filter for top row fragments */ if (y > 0) { - s->dsp.vp3_v_loop_filter( + s->vp3dsp.v_loop_filter( plane_data + 8*x, stride, bounding_values); } @@ -1333,7 +1321,7 @@ static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int ye * in this frame (it will be filtered in next iteration) */ if ((x < width - 1) && (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) { - s->dsp.vp3_h_loop_filter( + s->vp3dsp.h_loop_filter( plane_data + 8*x + 8, stride, bounding_values); } @@ -1343,7 +1331,7 @@ static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int ye * in this frame (it will be filtered in the next row) */ if ((y < height - 1) && (s->all_fragments[fragment + width].coding_method == MODE_COPY)) { - s->dsp.vp3_v_loop_filter( + s->vp3dsp.v_loop_filter( plane_data + 8*x + 8*stride, stride, bounding_values); } @@ -1355,88 +1343,176 @@ static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int ye } } +/** + * Pull DCT tokens from the 64 levels to decode and dequant the coefficients + * for the next block in coding order + */ +static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag, + int plane, int inter, int16_t block[64]) +{ + int16_t *dequantizer = s->qmat[frag->qpi][inter][plane]; + uint8_t *perm = s->idct_scantable; + int i = 0; + + do { + int token = *s->dct_tokens[plane][i]; + switch (token & 3) { + case 0: // EOB + if (--token < 4) // 0-3 are token types, so the EOB run must now be 0 + s->dct_tokens[plane][i]++; + else + *s->dct_tokens[plane][i] = token & ~3; + goto end; + case 1: // zero run + s->dct_tokens[plane][i]++; + i += (token >> 2) & 0x7f; + if (i > 63) { + av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n"); + return i; + } + block[perm[i]] = (token >> 9) * dequantizer[perm[i]]; + i++; + break; + case 2: // coeff + block[perm[i]] = (token >> 2) * dequantizer[perm[i]]; + s->dct_tokens[plane][i++]++; + break; + default: // shouldn't happen + return i; + } + } while (i < 64); + // return value is expected to be a valid level + i--; +end: + // the actual DC+prediction is in the fragment structure + block[0] = frag->dc * s->qmat[0][inter][plane][0]; + return i; +} + /** * called when all pixels up to row y are complete */ static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y) { - int h, cy; - int offset[4]; + int h, cy, i; + int offset[AV_NUM_DATA_POINTERS]; + + if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) { + int y_flipped = s->flipped_image ? s->avctx->height-y : y; + + // At the end of the frame, report INT_MAX instead of the height of the frame. + // This makes the other threads' ff_thread_await_progress() calls cheaper, because + // they don't have to clip their values. + ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0); + } if(s->avctx->draw_horiz_band==NULL) return; h= y - s->last_slice_end; + s->last_slice_end= y; y -= h; if (!s->flipped_image) { - if (y == 0) - h -= s->height - s->avctx->height; // account for non-mod16 - y = s->height - y - h; + y = s->avctx->height - y - h; } - cy = y >> 1; - offset[0] = s->current_frame.linesize[0]*y; - offset[1] = s->current_frame.linesize[1]*cy; - offset[2] = s->current_frame.linesize[2]*cy; - offset[3] = 0; + cy = y >> s->chroma_y_shift; + offset[0] = s->current_frame.f->linesize[0]*y; + offset[1] = s->current_frame.f->linesize[1]*cy; + offset[2] = s->current_frame.f->linesize[2]*cy; + for (i = 3; i < AV_NUM_DATA_POINTERS; i++) + offset[i] = 0; emms_c(); - s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h); - s->last_slice_end= y + h; + s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h); +} + +/** + * Wait for the reference frame of the current fragment. + * The progress value is in luma pixel rows. + */ +static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y) +{ + ThreadFrame *ref_frame; + int ref_row; + int border = motion_y&1; + + if (fragment->coding_method == MODE_USING_GOLDEN || + fragment->coding_method == MODE_GOLDEN_MV) + ref_frame = &s->golden_frame; + else + ref_frame = &s->last_frame; + + ref_row = y + (motion_y>>1); + ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border); + + ff_thread_await_progress(ref_frame, ref_row, 0); } /* * Perform the final rendering for a particular slice of data. - * The slice number ranges from 0..(macroblock_height - 1). + * The slice number ranges from 0..(c_superblock_height - 1). */ static void render_slice(Vp3DecodeContext *s, int slice) { - int x; - int16_t *dequantizer; - LOCAL_ALIGNED_16(DCTELEM, block, [64]); + int x, y, i, j, fragment; + int16_t *block = s->block; int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef; int motion_halfpel_index; uint8_t *motion_source; - int plane; + int plane, first_pixel; - if (slice >= s->macroblock_height) + if (slice >= s->c_superblock_height) return; for (plane = 0; plane < 3; plane++) { - uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane]; - uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane]; - uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane]; - int stride = s->current_frame.linesize[plane]; - int plane_width = s->width >> !!plane; - int plane_height = s->height >> !!plane; - int y = slice * FRAGMENT_PIXELS << !plane ; - int slice_height = y + (FRAGMENT_PIXELS << !plane); - int i = s->fragment_start[plane] + (y>>3)*(s->fragment_width>>!!plane); + uint8_t *output_plane = s->current_frame.f->data [plane] + s->data_offset[plane]; + uint8_t * last_plane = s-> last_frame.f->data [plane] + s->data_offset[plane]; + uint8_t *golden_plane = s-> golden_frame.f->data [plane] + s->data_offset[plane]; + ptrdiff_t stride = s->current_frame.f->linesize[plane]; + int plane_width = s->width >> (plane && s->chroma_x_shift); + int plane_height = s->height >> (plane && s->chroma_y_shift); + int8_t (*motion_val)[2] = s->motion_val[!!plane]; + + int sb_x, sb_y = slice << (!plane && s->chroma_y_shift); + int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift); + int slice_width = plane ? s->c_superblock_width : s->y_superblock_width; + + int fragment_width = s->fragment_width[!!plane]; + int fragment_height = s->fragment_height[!!plane]; + int fragment_start = s->fragment_start[plane]; + int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME); if (!s->flipped_image) stride = -stride; if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY)) continue; + /* for each superblock row in the slice (both of them)... */ + for (; sb_y < slice_height; sb_y++) { - if(FFABS(stride) > 2048) - return; //various tables are fixed size + /* for each superblock in a row... */ + for (sb_x = 0; sb_x < slice_width; sb_x++) { - /* for each fragment row in the slice (both of them)... */ - for (; y < slice_height; y += 8) { + /* for each block in a superblock... */ + for (j = 0; j < 16; j++) { + x = 4*sb_x + hilbert_offset[j][0]; + y = 4*sb_y + hilbert_offset[j][1]; + fragment = y*fragment_width + x; - /* for each fragment in a row... */ - for (x = 0; x < plane_width; x += 8, i++) { - int first_pixel = y*stride + x; + i = fragment_start + fragment; - if ((i < 0) || (i >= s->fragment_count)) { - av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i); - return; - } + // bounds check + if (x >= fragment_width || y >= fragment_height) + continue; + + first_pixel = 8*y*stride + 8*x; + + if (do_await && s->all_fragments[i].coding_method != MODE_INTRA) + await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift); /* transform if this block was coded */ if (s->all_fragments[i].coding_method != MODE_COPY) { - if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) || (s->all_fragments[i].coding_method == MODE_GOLDEN_MV)) motion_source= golden_plane; @@ -1451,17 +1527,11 @@ static void render_slice(Vp3DecodeContext *s, int slice) if ((s->all_fragments[i].coding_method > MODE_INTRA) && (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) { int src_x, src_y; - motion_x = s->all_fragments[i].motion_x; - motion_y = s->all_fragments[i].motion_y; - if(plane){ - motion_x= (motion_x>>1) | (motion_x&1); - motion_y= (motion_y>>1) | (motion_y&1); - } + motion_x = motion_val[fragment][0]; + motion_y = motion_val[fragment][1]; - src_x= (motion_x>>1) + x; - src_y= (motion_y>>1) + y; - if ((motion_x == 127) || (motion_y == 127)) - av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y); + src_x= (motion_x>>1) + 8*x; + src_y= (motion_y>>1) + 8*y; motion_halfpel_index = motion_x & 0x01; motion_source += (motion_x >> 1); @@ -1471,10 +1541,13 @@ static void render_slice(Vp3DecodeContext *s, int slice) if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){ uint8_t *temp= s->edge_emu_buffer; - if(stride<0) temp -= 9*stride; - else temp += 9*stride; + if(stride<0) temp -= 8*stride; - ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height); + s->vdsp.emulated_edge_mc(temp, motion_source, + stride, stride, + 9, 9, src_x, src_y, + plane_width, + plane_height); motion_source= temp; } } @@ -1488,68 +1561,58 @@ static void render_slice(Vp3DecodeContext *s, int slice) VP3 source but this would be slower as put_no_rnd_pixels_tab is better optimzed */ if(motion_halfpel_index != 3){ - s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index]( + s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index]( output_plane + first_pixel, motion_source, stride, 8); }else{ int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1 - s->dsp.put_no_rnd_pixels_l2[1]( + s->vp3dsp.put_no_rnd_pixels_l2( output_plane + first_pixel, motion_source - d, motion_source + stride + 1 + d, stride, 8); } - dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane]; - }else{ - dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane]; - } - - /* dequantize the DCT coefficients */ - if(s->avctx->idct_algo==FF_IDCT_VP3){ - Coeff *coeff= s->coeffs + i; - s->dsp.clear_block(block); - while(coeff->next){ - block[coeff->index]= coeff->coeff * dequantizer[coeff->index]; - coeff= coeff->next; - } - }else{ - Coeff *coeff= s->coeffs + i; - s->dsp.clear_block(block); - while(coeff->next){ - block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2; - coeff= coeff->next; - } } /* invert DCT and place (or add) in final output */ if (s->all_fragments[i].coding_method == MODE_INTRA) { - if(s->avctx->idct_algo!=FF_IDCT_VP3) - block[0] += 128<<3; - s->dsp.idct_put( + int index; + index = vp3_dequant(s, s->all_fragments + i, plane, 0, block); + if (index > 63) + continue; + s->vp3dsp.idct_put( output_plane + first_pixel, stride, block); } else { - s->dsp.idct_add( + int index = vp3_dequant(s, s->all_fragments + i, plane, 1, block); + if (index > 63) + continue; + if (index > 0) { + s->vp3dsp.idct_add( output_plane + first_pixel, stride, block); + } else { + s->vp3dsp.idct_dc_add(output_plane + first_pixel, stride, block); + } } } else { /* copy directly from the previous frame */ - s->dsp.put_pixels_tab[1][0]( + s->hdsp.put_pixels_tab[1][0]( output_plane + first_pixel, last_plane + first_pixel, stride, 8); } + } } - // Filter the previous block row. We can't filter the current row yet - // since it needs pixels from the next row - if (y > 0) - apply_loop_filter(s, plane, (y>>3)-1, (y>>3)); + + // Filter up to the last row in the superblock row + if (!s->skip_loop_filter) + apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1)); } } @@ -1561,22 +1624,70 @@ static void render_slice(Vp3DecodeContext *s, int slice) * dispatch (slice - 1); */ - // now that we've filtered the last rows, they're safe to display - if (slice) - vp3_draw_horiz_band(s, 16*slice); + vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16)); +} + +/// Allocate tables for per-frame data in Vp3DecodeContext +static av_cold int allocate_tables(AVCodecContext *avctx) +{ + Vp3DecodeContext *s = avctx->priv_data; + int y_fragment_count, c_fragment_count; + + y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; + c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; + + s->superblock_coding = av_malloc(s->superblock_count); + s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment)); + s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int)); + s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base)); + s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0])); + s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1])); + + /* work out the block mapping tables */ + s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int)); + s->macroblock_coding = av_malloc(s->macroblock_count + 1); + + if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base || + !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding || + !s->motion_val[0] || !s->motion_val[1]) { + vp3_decode_end(avctx); + return -1; + } + + init_block_mapping(s); + + return 0; +} + +static av_cold int init_frames(Vp3DecodeContext *s) +{ + s->current_frame.f = av_frame_alloc(); + s->last_frame.f = av_frame_alloc(); + s->golden_frame.f = av_frame_alloc(); + + if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) { + av_frame_free(&s->current_frame.f); + av_frame_free(&s->last_frame.f); + av_frame_free(&s->golden_frame.f); + return AVERROR(ENOMEM); + } + + return 0; } -/* - * This is the ffmpeg/libavcodec API init function. - */ static av_cold int vp3_decode_init(AVCodecContext *avctx) { Vp3DecodeContext *s = avctx->priv_data; - int i, inter, plane; + int i, inter, plane, ret; int c_width; int c_height; - int y_superblock_count; - int c_superblock_count; + int y_fragment_count, c_fragment_count; + + ret = init_frames(s); + if (ret < 0) + return ret; + + avctx->internal->allocate_progress = 1; if (avctx->codec_tag == MKTAG('V','P','3','0')) s->version = 0; @@ -1586,57 +1697,58 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx) s->avctx = avctx; s->width = FFALIGN(avctx->width, 16); s->height = FFALIGN(avctx->height, 16); - avctx->pix_fmt = PIX_FMT_YUV420P; + if (avctx->pix_fmt == AV_PIX_FMT_NONE) + avctx->pix_fmt = AV_PIX_FMT_YUV420P; avctx->chroma_sample_location = AVCHROMA_LOC_CENTER; - if(avctx->idct_algo==FF_IDCT_AUTO) - avctx->idct_algo=FF_IDCT_VP3; - dsputil_init(&s->dsp, avctx); - - ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct); + ff_hpeldsp_init(&s->hdsp, avctx->flags | CODEC_FLAG_BITEXACT); + ff_videodsp_init(&s->vdsp, 8); + ff_vp3dsp_init(&s->vp3dsp, avctx->flags); + + for (i = 0; i < 64; i++) { +#define T(x) (x >> 3) | ((x & 7) << 3) + s->idct_permutation[i] = T(i); + s->idct_scantable[i] = T(ff_zigzag_direct[i]); +#undef T + } /* initialize to an impossible value which will force a recalculation * in the first frame decode */ for (i = 0; i < 3; i++) s->qps[i] = -1; + av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, + &s->chroma_y_shift); + s->y_superblock_width = (s->width + 31) / 32; s->y_superblock_height = (s->height + 31) / 32; - y_superblock_count = s->y_superblock_width * s->y_superblock_height; + s->y_superblock_count = s->y_superblock_width * s->y_superblock_height; /* work out the dimensions for the C planes */ - c_width = s->width / 2; - c_height = s->height / 2; + c_width = s->width >> s->chroma_x_shift; + c_height = s->height >> s->chroma_y_shift; s->c_superblock_width = (c_width + 31) / 32; s->c_superblock_height = (c_height + 31) / 32; - c_superblock_count = s->c_superblock_width * s->c_superblock_height; + s->c_superblock_count = s->c_superblock_width * s->c_superblock_height; - s->superblock_count = y_superblock_count + (c_superblock_count * 2); - s->u_superblock_start = y_superblock_count; - s->v_superblock_start = s->u_superblock_start + c_superblock_count; - s->superblock_coding = av_malloc(s->superblock_count); + s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2); + s->u_superblock_start = s->y_superblock_count; + s->v_superblock_start = s->u_superblock_start + s->c_superblock_count; s->macroblock_width = (s->width + 15) / 16; s->macroblock_height = (s->height + 15) / 16; s->macroblock_count = s->macroblock_width * s->macroblock_height; - s->fragment_width = s->width / FRAGMENT_PIXELS; - s->fragment_height = s->height / FRAGMENT_PIXELS; + s->fragment_width[0] = s->width / FRAGMENT_PIXELS; + s->fragment_height[0] = s->height / FRAGMENT_PIXELS; + s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift; + s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift; /* fragment count covers all 8x8 blocks for all 3 planes */ - s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2; - s->fragment_start[1] = s->fragment_width * s->fragment_height; - s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4; - - s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment)); - s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts)); - s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65); - s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int)); - s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int)); - if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts || - !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) { - vp3_decode_end(avctx); - return -1; - } + y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; + c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; + s->fragment_count = y_fragment_count + 2*c_fragment_count; + s->fragment_start[1] = y_fragment_count; + s->fragment_start[2] = y_fragment_count + c_fragment_count; if (!s->theora_tables) { @@ -1662,61 +1774,61 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx) for (i = 0; i < 16; i++) { /* DC histograms */ - init_vlc(&s->dc_vlc[i], 5, 32, + init_vlc(&s->dc_vlc[i], 11, 32, &dc_bias[i][0][1], 4, 2, &dc_bias[i][0][0], 4, 2, 0); /* group 1 AC histograms */ - init_vlc(&s->ac_vlc_1[i], 5, 32, + init_vlc(&s->ac_vlc_1[i], 11, 32, &ac_bias_0[i][0][1], 4, 2, &ac_bias_0[i][0][0], 4, 2, 0); /* group 2 AC histograms */ - init_vlc(&s->ac_vlc_2[i], 5, 32, + init_vlc(&s->ac_vlc_2[i], 11, 32, &ac_bias_1[i][0][1], 4, 2, &ac_bias_1[i][0][0], 4, 2, 0); /* group 3 AC histograms */ - init_vlc(&s->ac_vlc_3[i], 5, 32, + init_vlc(&s->ac_vlc_3[i], 11, 32, &ac_bias_2[i][0][1], 4, 2, &ac_bias_2[i][0][0], 4, 2, 0); /* group 4 AC histograms */ - init_vlc(&s->ac_vlc_4[i], 5, 32, + init_vlc(&s->ac_vlc_4[i], 11, 32, &ac_bias_3[i][0][1], 4, 2, &ac_bias_3[i][0][0], 4, 2, 0); } } else { - for (i = 0; i < 16; i++) { + for (i = 0; i < 16; i++) { /* DC histograms */ - if (init_vlc(&s->dc_vlc[i], 5, 32, - &s->huffman_table[i][0][1], 4, 2, - &s->huffman_table[i][0][0], 4, 2, 0) < 0) + if (init_vlc(&s->dc_vlc[i], 11, 32, + &s->huffman_table[i][0][1], 8, 4, + &s->huffman_table[i][0][0], 8, 4, 0) < 0) goto vlc_fail; /* group 1 AC histograms */ - if (init_vlc(&s->ac_vlc_1[i], 5, 32, - &s->huffman_table[i+16][0][1], 4, 2, - &s->huffman_table[i+16][0][0], 4, 2, 0) < 0) + if (init_vlc(&s->ac_vlc_1[i], 11, 32, + &s->huffman_table[i+16][0][1], 8, 4, + &s->huffman_table[i+16][0][0], 8, 4, 0) < 0) goto vlc_fail; /* group 2 AC histograms */ - if (init_vlc(&s->ac_vlc_2[i], 5, 32, - &s->huffman_table[i+16*2][0][1], 4, 2, - &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0) + if (init_vlc(&s->ac_vlc_2[i], 11, 32, + &s->huffman_table[i+16*2][0][1], 8, 4, + &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0) goto vlc_fail; /* group 3 AC histograms */ - if (init_vlc(&s->ac_vlc_3[i], 5, 32, - &s->huffman_table[i+16*3][0][1], 4, 2, - &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0) + if (init_vlc(&s->ac_vlc_3[i], 11, 32, + &s->huffman_table[i+16*3][0][1], 8, 4, + &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0) goto vlc_fail; /* group 4 AC histograms */ - if (init_vlc(&s->ac_vlc_4[i], 5, 32, - &s->huffman_table[i+16*4][0][1], 4, 2, - &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0) + if (init_vlc(&s->ac_vlc_4[i], 11, 32, + &s->huffman_table[i+16*4][0][1], 8, 4, + &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0) goto vlc_fail; } } @@ -1737,41 +1849,117 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx) &motion_vector_vlc_table[0][1], 2, 1, &motion_vector_vlc_table[0][0], 2, 1, 0); - /* work out the block mapping tables */ - s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int)); - s->macroblock_coding = av_malloc(s->macroblock_count + 1); - if (!s->superblock_fragments || !s->macroblock_coding) { - vp3_decode_end(avctx); - return -1; - } - init_block_mapping(s); + return allocate_tables(avctx); - for (i = 0; i < 3; i++) { - s->current_frame.data[i] = NULL; - s->last_frame.data[i] = NULL; - s->golden_frame.data[i] = NULL; +vlc_fail: + av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n"); + return -1; +} + +/// Release and shuffle frames after decode finishes +static int update_frames(AVCodecContext *avctx) +{ + Vp3DecodeContext *s = avctx->priv_data; + int ret = 0; + + + /* shuffle frames (last = current) */ + ff_thread_release_buffer(avctx, &s->last_frame); + ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame); + if (ret < 0) + goto fail; + + if (s->keyframe) { + ff_thread_release_buffer(avctx, &s->golden_frame); + ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame); } +fail: + ff_thread_release_buffer(avctx, &s->current_frame); + return ret; +} + +static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src) +{ + ff_thread_release_buffer(s->avctx, dst); + if (src->f->data[0]) + return ff_thread_ref_frame(dst, src); return 0; +} -vlc_fail: - av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n"); - return -1; +static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src) +{ + int ret; + if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 || + (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 || + (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0) + return ret; + return 0; +} + +static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src) +{ + Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data; + int qps_changed = 0, i, err; + +#define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field) + + if (!s1->current_frame.f->data[0] + ||s->width != s1->width + ||s->height!= s1->height) { + if (s != s1) + ref_frames(s, s1); + return -1; + } + + if (s != s1) { + // init tables if the first frame hasn't been decoded + if (!s->current_frame.f->data[0]) { + int y_fragment_count, c_fragment_count; + s->avctx = dst; + err = allocate_tables(dst); + if (err) + return err; + y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; + c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; + memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0])); + memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1])); + } + + // copy previous frame data + if ((err = ref_frames(s, s1)) < 0) + return err; + + s->keyframe = s1->keyframe; + + // copy qscale data if necessary + for (i = 0; i < 3; i++) { + if (s->qps[i] != s1->qps[1]) { + qps_changed = 1; + memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i])); + } + } + + if (s->qps[0] != s1->qps[0]) + memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array)); + + if (qps_changed) + copy_fields(s, s1, qps, superblock_count); +#undef copy_fields + } + + return update_frames(dst); } -/* - * This is the ffmpeg/libavcodec API frame decode function. - */ static int vp3_decode_frame(AVCodecContext *avctx, - void *data, int *data_size, + void *data, int *got_frame, AVPacket *avpkt) { const uint8_t *buf = avpkt->data; int buf_size = avpkt->size; Vp3DecodeContext *s = avctx->priv_data; GetBitContext gb; - static int counter = 0; - int i; + int i, ret; init_get_bits(&gb, buf, buf_size * 8); @@ -1796,8 +1984,10 @@ static int vp3_decode_frame(AVCodecContext *avctx, if (s->avctx->debug & FF_DEBUG_PICT_INFO) av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n", - s->keyframe?"key":"", counter, s->qps[0]); - counter++; + s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]); + + s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] || + avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY); if (s->qps[0] != s->last_qps[0]) init_loop_filter(s); @@ -1811,6 +2001,15 @@ static int vp3_decode_frame(AVCodecContext *avctx, if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe) return buf_size; + s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P; + if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0) { + av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n"); + goto error; + } + + if (!s->edge_emu_buffer) + s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.f->linesize[0])); + if (s->keyframe) { if (!s->theora) { @@ -1819,7 +2018,7 @@ static int vp3_decode_frame(AVCodecContext *avctx, if (s->version) { s->version = get_bits(&gb, 5); - if (counter == 1) + if (avctx->frame_number == 0) av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version); } } @@ -1829,138 +2028,84 @@ static int vp3_decode_frame(AVCodecContext *avctx, av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n"); skip_bits(&gb, 2); /* reserved? */ } - - if (s->last_frame.data[0] == s->golden_frame.data[0]) { - if (s->golden_frame.data[0]) - avctx->release_buffer(avctx, &s->golden_frame); - s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */ - } else { - if (s->golden_frame.data[0]) - avctx->release_buffer(avctx, &s->golden_frame); - if (s->last_frame.data[0]) - avctx->release_buffer(avctx, &s->last_frame); - } - - s->golden_frame.reference = 3; - if(avctx->get_buffer(avctx, &s->golden_frame) < 0) { - av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n"); - return -1; - } - - /* golden frame is also the current frame */ - s->current_frame= s->golden_frame; } else { - /* allocate a new current frame */ - s->current_frame.reference = 3; - if (!s->golden_frame.data[0]) { - av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n"); - return -1; - } - if(avctx->get_buffer(avctx, &s->current_frame) < 0) { - av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n"); - return -1; + if (!s->golden_frame.f->data[0]) { + av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n"); + + s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I; + if (ff_thread_get_buffer(avctx, &s->golden_frame, AV_GET_BUFFER_FLAG_REF) < 0) { + av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n"); + goto error; + } + ff_thread_release_buffer(avctx, &s->last_frame); + if ((ret = ff_thread_ref_frame(&s->last_frame, &s->golden_frame)) < 0) + goto error; + ff_thread_report_progress(&s->last_frame, INT_MAX, 0); } } - s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame - s->current_frame.qstride= 0; - - init_frame(s, &gb); + memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment)); + ff_thread_finish_setup(avctx); if (unpack_superblocks(s, &gb)){ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n"); - return -1; + goto error; } if (unpack_modes(s, &gb)){ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n"); - return -1; + goto error; } if (unpack_vectors(s, &gb)){ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n"); - return -1; + goto error; } if (unpack_block_qpis(s, &gb)){ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n"); - return -1; + goto error; } if (unpack_dct_coeffs(s, &gb)){ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n"); - return -1; + goto error; } for (i = 0; i < 3; i++) { + int height = s->height >> (i && s->chroma_y_shift); if (s->flipped_image) s->data_offset[i] = 0; else - s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i]; + s->data_offset[i] = (height-1) * s->current_frame.f->linesize[i]; } s->last_slice_end = 0; - for (i = 0; i < s->macroblock_height; i++) + for (i = 0; i < s->c_superblock_height; i++) render_slice(s, i); // filter the last row for (i = 0; i < 3; i++) { - int row = (s->height >> (3+!!i)) - 1; + int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1; apply_loop_filter(s, i, row, row+1); } - vp3_draw_horiz_band(s, s->height); - - *data_size=sizeof(AVFrame); - *(AVFrame*)data= s->current_frame; + vp3_draw_horiz_band(s, s->avctx->height); - /* release the last frame, if it is allocated and if it is not the - * golden frame */ - if ((s->last_frame.data[0]) && - (s->last_frame.data[0] != s->golden_frame.data[0])) - avctx->release_buffer(avctx, &s->last_frame); + if ((ret = av_frame_ref(data, s->current_frame.f)) < 0) + return ret; + *got_frame = 1; - /* shuffle frames (last = current) */ - s->last_frame= s->current_frame; - s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */ + if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME)) { + ret = update_frames(avctx); + if (ret < 0) + return ret; + } return buf_size; -} -/* - * This is the ffmpeg/libavcodec API module cleanup function. - */ -static av_cold int vp3_decode_end(AVCodecContext *avctx) -{ - Vp3DecodeContext *s = avctx->priv_data; - int i; - - av_free(s->superblock_coding); - av_free(s->all_fragments); - av_free(s->coeff_counts); - av_free(s->coeffs); - av_free(s->coded_fragment_list); - av_free(s->fast_fragment_list); - av_free(s->superblock_fragments); - av_free(s->macroblock_coding); +error: + ff_thread_report_progress(&s->current_frame, INT_MAX, 0); - for (i = 0; i < 16; i++) { - free_vlc(&s->dc_vlc[i]); - free_vlc(&s->ac_vlc_1[i]); - free_vlc(&s->ac_vlc_2[i]); - free_vlc(&s->ac_vlc_3[i]); - free_vlc(&s->ac_vlc_4[i]); - } + if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME)) + av_frame_unref(s->current_frame.f); - free_vlc(&s->superblock_run_length_vlc); - free_vlc(&s->fragment_run_length_vlc); - free_vlc(&s->mode_code_vlc); - free_vlc(&s->motion_vector_vlc); - - /* release all frames */ - if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0]) - avctx->release_buffer(avctx, &s->golden_frame); - if (s->last_frame.data[0]) - avctx->release_buffer(avctx, &s->last_frame); - /* no need to release the current_frame since it will always be pointing - * to the same frame as either the golden or last frame */ - - return 0; + return -1; } static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb) @@ -1974,7 +2119,8 @@ static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb) return -1; } token = get_bits(gb, 5); - //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size); + av_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n", + s->hti, s->hbits, token, s->entries, s->huff_code_size); s->huffman_table[s->hti][token][0] = s->hbits; s->huffman_table[s->hti][token][1] = s->huff_code_size; s->entries++; @@ -1997,11 +2143,35 @@ static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb) return 0; } +static int vp3_init_thread_copy(AVCodecContext *avctx) +{ + Vp3DecodeContext *s = avctx->priv_data; + + s->superblock_coding = NULL; + s->all_fragments = NULL; + s->coded_fragment_list[0] = NULL; + s->dct_tokens_base = NULL; + s->superblock_fragments = NULL; + s->macroblock_coding = NULL; + s->motion_val[0] = NULL; + s->motion_val[1] = NULL; + s->edge_emu_buffer = NULL; + + return init_frames(s); +} + #if CONFIG_THEORA_DECODER +static const enum AVPixelFormat theora_pix_fmts[4] = { + AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P +}; + static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb) { Vp3DecodeContext *s = avctx->priv_data; int visible_width, visible_height, colorspace; + int offset_x = 0, offset_y = 0; + int ret; + AVRational fps, aspect; s->theora = get_bits_long(gb, 24); av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora); @@ -2017,24 +2187,32 @@ static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb) visible_width = s->width = get_bits(gb, 16) << 4; visible_height = s->height = get_bits(gb, 16) << 4; - if(avcodec_check_dimensions(avctx, s->width, s->height)){ - av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height); - s->width= s->height= 0; - return -1; - } - if (s->theora >= 0x030200) { visible_width = get_bits_long(gb, 24); visible_height = get_bits_long(gb, 24); - skip_bits(gb, 8); /* offset x */ - skip_bits(gb, 8); /* offset y */ + offset_x = get_bits(gb, 8); /* offset x */ + offset_y = get_bits(gb, 8); /* offset y, from bottom */ } - skip_bits(gb, 32); /* fps numerator */ - skip_bits(gb, 32); /* fps denumerator */ - skip_bits(gb, 24); /* aspect numerator */ - skip_bits(gb, 24); /* aspect denumerator */ + fps.num = get_bits_long(gb, 32); + fps.den = get_bits_long(gb, 32); + if (fps.num && fps.den) { + if (fps.num < 0 || fps.den < 0) { + av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n"); + return AVERROR_INVALIDDATA; + } + av_reduce(&avctx->time_base.num, &avctx->time_base.den, + fps.den, fps.num, 1<<30); + } + + aspect.num = get_bits_long(gb, 24); + aspect.den = get_bits_long(gb, 24); + if (aspect.num && aspect.den) { + av_reduce(&avctx->sample_aspect_ratio.num, + &avctx->sample_aspect_ratio.den, + aspect.num, aspect.den, 1<<30); + } if (s->theora < 0x030200) skip_bits(gb, 5); /* keyframe frequency force */ @@ -2046,17 +2224,20 @@ static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb) if (s->theora >= 0x030200) { skip_bits(gb, 5); /* keyframe frequency force */ - skip_bits(gb, 2); /* pixel format: 420,res,422,444 */ + avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)]; skip_bits(gb, 3); /* reserved */ } // align_get_bits(gb); if ( visible_width <= s->width && visible_width > s->width-16 - && visible_height <= s->height && visible_height > s->height-16) - avcodec_set_dimensions(avctx, visible_width, visible_height); + && visible_height <= s->height && visible_height > s->height-16 + && !offset_x && (offset_y == s->height - visible_height)) + ret = ff_set_dimensions(avctx, visible_width, visible_height); else - avcodec_set_dimensions(avctx, s->width, s->height); + ret = ff_set_dimensions(avctx, s->width, s->height); + if (ret < 0) + return ret; if (colorspace == 1) { avctx->color_primaries = AVCOL_PRI_BT470M; @@ -2079,13 +2260,9 @@ static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb) if (s->theora >= 0x030200) { n = get_bits(gb, 3); /* loop filter limit values table */ - for (i = 0; i < 64; i++) { - s->filter_limit_values[i] = get_bits(gb, n); - if (s->filter_limit_values[i] > 127) { - av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]); - s->filter_limit_values[i] = 127; - } - } + if (n) + for (i = 0; i < 64; i++) + s->filter_limit_values[i] = get_bits(gb, n); } if (s->theora >= 0x030200) @@ -2199,13 +2376,15 @@ static av_cold int theora_decode_init(AVCodecContext *avctx) return -1; } - if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size, + if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size, 42, header_start, header_len) < 0) { av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n"); return -1; } for(i=0;i<3;i++) { + if (header_len[i] <= 0) + continue; init_get_bits(&gb, header_start[i], header_len[i] * 8); ptype = get_bits(&gb, 8); @@ -2245,31 +2424,35 @@ static av_cold int theora_decode_init(AVCodecContext *avctx) return vp3_decode_init(avctx); } -AVCodec theora_decoder = { - "theora", - CODEC_TYPE_VIDEO, - CODEC_ID_THEORA, - sizeof(Vp3DecodeContext), - theora_decode_init, - NULL, - vp3_decode_end, - vp3_decode_frame, - CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND, - NULL, - .long_name = NULL_IF_CONFIG_SMALL("Theora"), +AVCodec ff_theora_decoder = { + .name = "theora", + .long_name = NULL_IF_CONFIG_SMALL("Theora"), + .type = AVMEDIA_TYPE_VIDEO, + .id = AV_CODEC_ID_THEORA, + .priv_data_size = sizeof(Vp3DecodeContext), + .init = theora_decode_init, + .close = vp3_decode_end, + .decode = vp3_decode_frame, + .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | + CODEC_CAP_FRAME_THREADS, + .flush = vp3_decode_flush, + .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy), + .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context) }; #endif -AVCodec vp3_decoder = { - "vp3", - CODEC_TYPE_VIDEO, - CODEC_ID_VP3, - sizeof(Vp3DecodeContext), - vp3_decode_init, - NULL, - vp3_decode_end, - vp3_decode_frame, - CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND, - NULL, - .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"), +AVCodec ff_vp3_decoder = { + .name = "vp3", + .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"), + .type = AVMEDIA_TYPE_VIDEO, + .id = AV_CODEC_ID_VP3, + .priv_data_size = sizeof(Vp3DecodeContext), + .init = vp3_decode_init, + .close = vp3_decode_end, + .decode = vp3_decode_frame, + .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | + CODEC_CAP_FRAME_THREADS, + .flush = vp3_decode_flush, + .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy), + .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context), };