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;
#define SB_PARTIALLY_CODED 1
#define SB_FULLY_CODED 2
+// This is the maximum length of a single long bit run that can be encoded
+// for superblock coding or block qps. Theora special-cases this to read a
+// bit instead of flipping the current bit to allow for runs longer than 4129.
+#define MAXIMUM_LONG_BIT_RUN 4129
+
#define MODE_INTER_NO_MV 0
#define MODE_INTRA 1
#define MODE_INTER_PLUS_MV 2
};
+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 {
int theora, theora_tables;
int version;
int width, height;
+ int chroma_x_shift, chroma_y_shift;
AVFrame golden_frame;
AVFrame last_frame;
AVFrame current_frame;
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;
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];
+ int8_t (*motion_val[2])[2];
+
ScanTable scantable;
/* tables */
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];
/* 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]; //<qmat[qpi][is_inter][plane]
+ DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
/* This table contains superblock_count * 16 entries. Each set of 16
* numbers corresponds to the fragment indexes 0..15 of the superblock.
* is coded. */
unsigned char *macroblock_coding;
- int first_coded_y_fragment;
- int first_coded_c_fragment;
- int last_coded_y_fragment;
- int last_coded_c_fragment;
-
uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
uint16_t huffman_table[80][32][2];
uint8_t filter_limit_values[64];
- DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
+ DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
} Vp3DecodeContext;
/************************************************************************
*/
static int init_block_mapping(Vp3DecodeContext *s)
{
- 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;
-
- static const signed char travel_width[16] = {
- 1, 1, 0, -1,
- 0, 0, 1, 0,
- 1, 0, 1, 0,
- 0, -1, 0, 1
- };
-
- static const signed char travel_height[16] = {
- 0, 0, 1, 0,
- 1, 1, 0, -1,
- 0, 1, 0, -1,
- -1, 0, -1, 0
- };
-
- 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;
-
- }
-
- if (current_width >= right_edge - 1) {
- /* reset width and move to next superblock row */
- current_width = -1;
- current_height += 4;
-
- /* fragment is now at the start of a new superblock row */
- current_fragment += superblock_row_inc;
- }
+ int sb_x, sb_y, plane;
+ int x, y, i, j = 0;
- /* 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;
- }
-
- mapping_index++;
- }
+ 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 */
}
-/*
- * This function wipes out all of the fragment data.
- */
-static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
-{
- int i;
-
- /* 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;
- }
-}
-
/*
* This function sets up the dequantization tables used for a particular
* frame.
*/
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);
/* unpack the list of partially-coded superblocks */
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) {
- 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);
- /* 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;
- }
+ if (current_superblock + current_run > s->superblock_count) {
+ av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
+ return -1;
}
- s->superblock_coding[current_superblock++] = bit;
+
+ memset(s->superblock_coding + current_superblock, bit, current_run);
+
+ current_superblock += current_run;
+ if (bit)
+ num_partial_superblocks += current_run;
+
+ if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
+ bit = get_bits1(gb);
+ else
+ bit ^= 1;
}
/* 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;
- 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) {
-
- if (current_run-- == 0) {
- bit ^= 1;
+ while (superblocks_decoded < s->superblock_count - num_partial_superblocks) {
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 (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
+ bit = get_bits1(gb);
+ else
+ bit ^= 1;
}
}
/* 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);
/* 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; 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) {
+ int coded = s->superblock_coding[i];
- /* copy all the fragments from the prior frame */
- s->all_fragments[current_fragment].coding_method =
- MODE_COPY;
-
- } 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 */
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;
}
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++)
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)
/* 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;
}
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
+ }
}
}
}
{
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);
/* 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)];
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:
/* 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) {
- 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] + 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) {
+ 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];
+ }
}
}
}
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;
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 == 4129)
+ if (run_length == MAXIMUM_LONG_BIT_RUN)
bit = get_bits1(gb);
else
bit ^= 1;
*/
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;
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);
/* use the token to get a zero run, a coefficient, and an eob run */
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;
+
+ // 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 {
bits_to_get = coeff_get_bits[token];
if (bits_to_get)
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++;
+ }
}
+ 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;
}
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);
/* 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);
+ residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
+ 2, 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 */
/* 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 (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);
+ residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
+ 2, residual_eob_run);
}
return 0;
*/
#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,
}
/* 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;
- }
}
}
}
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];
}
}
+/**
+ * Pulls 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, DCTELEM block[64])
+{
+ int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
+ uint8_t *perm = s->scantable.permutated;
+ 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;
+ 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:
+ av_log(s->avctx, AV_LOG_ERROR, "internal: invalid token type\n");
+ return i;
+ }
+ } while (i < 64);
+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
*/
/*
* 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;
- DECLARE_ALIGNED_16(DCTELEM, block)[64];
+ int x, y, i, j;
+ LOCAL_ALIGNED_16(DCTELEM, block, [64]);
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 * 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);
+ 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];
if (!s->flipped_image) stride = -stride;
if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
if(FFABS(stride) > 2048)
return; //various tables are fixed size
- /* for each fragment row in the slice (both of them)... */
- for (; y < slice_height; y += 8) {
+ /* for each superblock row in the slice (both of them)... */
+ for (; sb_y < slice_height; sb_y++) {
- /* for each fragment in a row... */
- for (x = 0; x < plane_width; x += 8, i++) {
- int first_pixel = y*stride + x;
+ /* for each superblock in a row... */
+ for (sb_x = 0; sb_x < slice_width; sb_x++) {
- if ((i < 0) || (i >= s->fragment_count)) {
- av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
- return;
- }
+ /* 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];
+
+ i = fragment_start + y*fragment_width + x;
+
+ // bounds check
+ if (x >= fragment_width || y >= fragment_height)
+ continue;
+
+ first_pixel = 8*y*stride + 8*x;
/* transform if this block was coded */
if (s->all_fragments[i].coding_method != MODE_COPY) {
+ int intra = s->all_fragments[i].coding_method == MODE_INTRA;
if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
(s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
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[y*fragment_width + x][0];
+ motion_y = motion_val[y*fragment_width + x][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);
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;
- }
- }
+ vp3_dequant(s, s->all_fragments + i, plane, !intra, block);
/* invert DCT and place (or add) in final output */
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
+ apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
}
}
* 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(64*slice + 64-16, s->height-16));
}
/*
int i, inter, plane;
int c_width;
int c_height;
- int y_superblock_count;
- int c_superblock_count;
+ int y_fragment_count, c_fragment_count;
if (avctx->codec_tag == MKTAG('V','P','3','0'))
s->version = 0;
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 == PIX_FMT_NONE)
+ avctx->pix_fmt = PIX_FMT_YUV420P;
avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
if(avctx->idct_algo==FF_IDCT_AUTO)
avctx->idct_algo=FF_IDCT_VP3;
for (i = 0; i < 3; i++)
s->qps[i] = -1;
+ avcodec_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_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->superblock_coding = av_malloc(s->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;
+ 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;
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) {
+ 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]));
+
+ if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
+ !s->coded_fragment_list[0] || !s->motion_val[0] || !s->motion_val[1]) {
vp3_decode_end(avctx);
return -1;
}
if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
return buf_size;
+ s->current_frame.reference = 3;
+ s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
+ if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
+ av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
+ goto error;
+ }
+
if (s->keyframe) {
if (!s->theora)
{
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;
+ av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
+
+ s->golden_frame.reference = 3;
+ s->golden_frame.pict_type = FF_I_TYPE;
+ if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
+ av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
+ goto error;
+ }
+ s->last_frame = s->golden_frame;
+ s->last_frame.type = FF_BUFFER_TYPE_COPY;
}
}
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));
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.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);
/* 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]))
+ if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
avctx->release_buffer(avctx, &s->last_frame);
/* shuffle frames (last = current) */
s->last_frame= s->current_frame;
+
+ if (s->keyframe) {
+ if (s->golden_frame.data[0])
+ avctx->release_buffer(avctx, &s->golden_frame);
+ s->golden_frame = s->current_frame;
+ s->last_frame.type = FF_BUFFER_TYPE_COPY;
+ }
+
s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
return buf_size;
+
+error:
+ if (s->current_frame.data[0])
+ avctx->release_buffer(avctx, &s->current_frame);
+ return -1;
}
/*
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->coded_fragment_list[0]);
+ av_free(s->dct_tokens_base);
av_free(s->superblock_fragments);
av_free(s->macroblock_coding);
+ av_free(s->motion_val[0]);
+ av_free(s->motion_val[1]);
for (i = 0; i < 16; i++) {
free_vlc(&s->dc_vlc[i]);
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])
+ if (s->golden_frame.data[0])
avctx->release_buffer(avctx, &s->golden_frame);
- if (s->last_frame.data[0])
+ if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
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 */
}
#if CONFIG_THEORA_DECODER
+static const enum PixelFormat theora_pix_fmts[4] = {
+ PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
+};
+
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
{
Vp3DecodeContext *s = avctx->priv_data;
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 */
}