/*
- *
- * Copyright (C) 2003 the ffmpeg project
+ * Copyright (C) 2003-2004 the ffmpeg project
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
- * VP3 Video Decoder by Mike Melanson (melanson@pcisys.net)
- * For more information about the VP3 coding process, visit:
- * http://www.pcisys.net/~melanson/codecs/
- *
*/
/**
* @file vp3.c
* On2 VP3 Video Decoder
+ *
+ * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
+ * For more information about the VP3 coding process, visit:
+ * http://multimedia.cx/
+ *
+ * Theora decoder by Alex Beregszaszi
*/
#include <stdio.h>
#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"
-#include "dsputil.h"
#include "vp3data.h"
static inline void debug_idct(const char *format, ...) { }
#endif
+typedef struct Coeff {
+ struct Coeff *next;
+ DCTELEM coeff;
+ uint8_t index;
+} Coeff;
+
+//FIXME split things out into their own arrays
typedef struct Vp3Fragment {
- DCTELEM coeffs[64];
- int coding_method;
- int coeff_count;
- int last_coeff;
- int motion_x;
- int motion_y;
+ Coeff *next_coeff;
/* address of first pixel taking into account which plane the fragment
* lives on as well as the plane stride */
int first_pixel;
/* this is the macroblock that the fragment belongs to */
- int macroblock;
+ uint16_t macroblock;
+ uint8_t coding_method;
+ uint8_t coeff_count;
+ int8_t motion_x;
+ int8_t motion_y;
} Vp3Fragment;
#define SB_NOT_CODED 0
typedef struct Vp3DecodeContext {
AVCodecContext *avctx;
+ int theora, theora_tables;
+ int version;
int width, height;
AVFrame golden_frame;
AVFrame last_frame;
AVFrame current_frame;
int keyframe;
DSPContext dsp;
+ int flipped_image;
int quality_index;
int last_quality_index;
int fragment_height;
Vp3Fragment *all_fragments;
+ Coeff *coeffs;
+ Coeff *next_coeff;
int u_fragment_start;
int v_fragment_start;
+
+ ScanTable scantable;
+
+ /* tables */
+ uint16_t coded_dc_scale_factor[64];
+ uint32_t coded_ac_scale_factor[64];
+ uint16_t coded_intra_y_dequant[64];
+ uint16_t coded_intra_c_dequant[64];
+ uint16_t coded_inter_dequant[64];
/* this is a list of indices into the all_fragments array indicating
* which of the fragments are coded */
VLC ac_vlc_3[16];
VLC ac_vlc_4[16];
- int16_t intra_y_dequant[64];
- int16_t intra_c_dequant[64];
- int16_t inter_dequant[64];
+ VLC superblock_run_length_vlc;
+ VLC fragment_run_length_vlc;
+ VLC mode_code_vlc;
+ VLC motion_vector_vlc;
+
+ /* these arrays need to be on 16-byte boundaries since SSE2 operations
+ * index into them */
+ int16_t __align16 intra_y_dequant[64];
+ int16_t __align16 intra_c_dequant[64];
+ int16_t __align16 inter_dequant[64];
/* This table contains superblock_count * 16 entries. Each set of 16
* numbers corresponds to the fragment indices 0..15 of the superblock.
int last_coded_y_fragment;
int last_coded_c_fragment;
-} Vp3DecodeContext;
-
-/************************************************************************
- * VP3 I/DCT
- ************************************************************************/
-
-#define IdctAdjustBeforeShift 8
-#define xC1S7 64277
-#define xC2S6 60547
-#define xC3S5 54491
-#define xC4S4 46341
-#define xC5S3 36410
-#define xC6S2 25080
-#define xC7S1 12785
-
-void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix,
- int16_t *output_data)
-{
- int32_t intermediate_data[64];
- int32_t *ip = intermediate_data;
- int16_t *op = output_data;
-
- int32_t _A, _B, _C, _D, _Ad, _Bd, _Cd, _Dd, _E, _F, _G, _H;
- int32_t _Ed, _Gd, _Add, _Bdd, _Fd, _Hd;
- int32_t t1, t2;
-
- int i, j;
-
- debug_idct("raw coefficient block:\n");
- for (i = 0; i < 8; i++) {
- for (j = 0; j < 8; j++) {
- debug_idct(" %5d", input_data[i * 8 + j]);
- }
- debug_idct("\n");
- }
- debug_idct("\n");
-
- for (i = 0; i < 64; i++) {
- j = dezigzag_index[i];
- intermediate_data[j] = dequant_matrix[i] * input_data[i];
- }
-
- debug_idct("dequantized block:\n");
- for (i = 0; i < 8; i++) {
- for (j = 0; j < 8; j++) {
- debug_idct(" %5d", intermediate_data[i * 8 + j]);
- }
- debug_idct("\n");
- }
- debug_idct("\n");
-
- /* Inverse DCT on the rows now */
- for (i = 0; i < 8; i++) {
- /* Check for non-zero values */
- if ( ip[0] | ip[1] | ip[2] | ip[3] | ip[4] | ip[5] | ip[6] | ip[7] ) {
- t1 = (int32_t)(xC1S7 * ip[1]);
- t2 = (int32_t)(xC7S1 * ip[7]);
- t1 >>= 16;
- t2 >>= 16;
- _A = t1 + t2;
-
- t1 = (int32_t)(xC7S1 * ip[1]);
- t2 = (int32_t)(xC1S7 * ip[7]);
- t1 >>= 16;
- t2 >>= 16;
- _B = t1 - t2;
-
- t1 = (int32_t)(xC3S5 * ip[3]);
- t2 = (int32_t)(xC5S3 * ip[5]);
- t1 >>= 16;
- t2 >>= 16;
- _C = t1 + t2;
-
- t1 = (int32_t)(xC3S5 * ip[5]);
- t2 = (int32_t)(xC5S3 * ip[3]);
- t1 >>= 16;
- t2 >>= 16;
- _D = t1 - t2;
-
-
- t1 = (int32_t)(xC4S4 * (_A - _C));
- t1 >>= 16;
- _Ad = t1;
-
- t1 = (int32_t)(xC4S4 * (_B - _D));
- t1 >>= 16;
- _Bd = t1;
-
-
- _Cd = _A + _C;
- _Dd = _B + _D;
-
- t1 = (int32_t)(xC4S4 * (ip[0] + ip[4]));
- t1 >>= 16;
- _E = t1;
-
- t1 = (int32_t)(xC4S4 * (ip[0] - ip[4]));
- t1 >>= 16;
- _F = t1;
+ uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
+ uint8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
- t1 = (int32_t)(xC2S6 * ip[2]);
- t2 = (int32_t)(xC6S2 * ip[6]);
- t1 >>= 16;
- t2 >>= 16;
- _G = t1 + t2;
-
- t1 = (int32_t)(xC6S2 * ip[2]);
- t2 = (int32_t)(xC2S6 * ip[6]);
- t1 >>= 16;
- t2 >>= 16;
- _H = t1 - t2;
-
-
- _Ed = _E - _G;
- _Gd = _E + _G;
-
- _Add = _F + _Ad;
- _Bdd = _Bd - _H;
-
- _Fd = _F - _Ad;
- _Hd = _Bd + _H;
-
- /* Final sequence of operations over-write original inputs. */
- ip[0] = (int16_t)((_Gd + _Cd ) >> 0);
- ip[7] = (int16_t)((_Gd - _Cd ) >> 0);
-
- ip[1] = (int16_t)((_Add + _Hd ) >> 0);
- ip[2] = (int16_t)((_Add - _Hd ) >> 0);
-
- ip[3] = (int16_t)((_Ed + _Dd ) >> 0);
- ip[4] = (int16_t)((_Ed - _Dd ) >> 0);
-
- ip[5] = (int16_t)((_Fd + _Bdd ) >> 0);
- ip[6] = (int16_t)((_Fd - _Bdd ) >> 0);
-
- }
-
- ip += 8; /* next row */
- }
+ /* Huffman decode */
+ int hti;
+ unsigned int hbits;
+ int entries;
+ int huff_code_size;
+ uint16_t huffman_table[80][32][2];
- ip = intermediate_data;
-
- for ( i = 0; i < 8; i++) {
- /* Check for non-zero values (bitwise or faster than ||) */
- if ( ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
- ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8] ) {
-
- t1 = (int32_t)(xC1S7 * ip[1*8]);
- t2 = (int32_t)(xC7S1 * ip[7*8]);
- t1 >>= 16;
- t2 >>= 16;
- _A = t1 + t2;
-
- t1 = (int32_t)(xC7S1 * ip[1*8]);
- t2 = (int32_t)(xC1S7 * ip[7*8]);
- t1 >>= 16;
- t2 >>= 16;
- _B = t1 - t2;
-
- t1 = (int32_t)(xC3S5 * ip[3*8]);
- t2 = (int32_t)(xC5S3 * ip[5*8]);
- t1 >>= 16;
- t2 >>= 16;
- _C = t1 + t2;
-
- t1 = (int32_t)(xC3S5 * ip[5*8]);
- t2 = (int32_t)(xC5S3 * ip[3*8]);
- t1 >>= 16;
- t2 >>= 16;
- _D = t1 - t2;
-
-
- t1 = (int32_t)(xC4S4 * (_A - _C));
- t1 >>= 16;
- _Ad = t1;
-
- t1 = (int32_t)(xC4S4 * (_B - _D));
- t1 >>= 16;
- _Bd = t1;
-
-
- _Cd = _A + _C;
- _Dd = _B + _D;
-
- t1 = (int32_t)(xC4S4 * (ip[0*8] + ip[4*8]));
- t1 >>= 16;
- _E = t1;
-
- t1 = (int32_t)(xC4S4 * (ip[0*8] - ip[4*8]));
- t1 >>= 16;
- _F = t1;
-
- t1 = (int32_t)(xC2S6 * ip[2*8]);
- t2 = (int32_t)(xC6S2 * ip[6*8]);
- t1 >>= 16;
- t2 >>= 16;
- _G = t1 + t2;
-
- t1 = (int32_t)(xC6S2 * ip[2*8]);
- t2 = (int32_t)(xC2S6 * ip[6*8]);
- t1 >>= 16;
- t2 >>= 16;
- _H = t1 - t2;
-
-
- _Ed = _E - _G;
- _Gd = _E + _G;
-
- _Add = _F + _Ad;
- _Bdd = _Bd - _H;
-
- _Fd = _F - _Ad;
- _Hd = _Bd + _H;
-
- _Gd += IdctAdjustBeforeShift;
- _Add += IdctAdjustBeforeShift;
- _Ed += IdctAdjustBeforeShift;
- _Fd += IdctAdjustBeforeShift;
-
- /* Final sequence of operations over-write original inputs. */
- op[0*8] = (int16_t)((_Gd + _Cd ) >> 4);
- op[7*8] = (int16_t)((_Gd - _Cd ) >> 4);
-
- op[1*8] = (int16_t)((_Add + _Hd ) >> 4);
- op[2*8] = (int16_t)((_Add - _Hd ) >> 4);
-
- op[3*8] = (int16_t)((_Ed + _Dd ) >> 4);
- op[4*8] = (int16_t)((_Ed - _Dd ) >> 4);
-
- op[5*8] = (int16_t)((_Fd + _Bdd ) >> 4);
- op[6*8] = (int16_t)((_Fd - _Bdd ) >> 4);
-
- } else {
-
- op[0*8] = 0;
- op[7*8] = 0;
- op[1*8] = 0;
- op[2*8] = 0;
- op[3*8] = 0;
- op[4*8] = 0;
- op[5*8] = 0;
- op[6*8] = 0;
- }
-
- ip++; /* next column */
- op++;
- }
-}
-
-void vp3_idct_put(int16_t *input_data, int16_t *dequant_matrix,
- uint8_t *dest, int stride)
-{
- int16_t transformed_data[64];
- int16_t *op;
- int i, j;
-
- vp3_idct_c(input_data, dequant_matrix, transformed_data);
-
- /* place in final output */
- op = transformed_data;
- for (i = 0; i < 8; i++) {
- for (j = 0; j < 8; j++) {
- if (*op < -128)
- *dest = 0;
- else if (*op > 127)
- *dest = 255;
- else
- *dest = (uint8_t)(*op + 128);
- op++;
- dest++;
- }
- dest += (stride - 8);
- }
-}
-
-void vp3_idct_add(int16_t *input_data, int16_t *dequant_matrix,
- uint8_t *dest, int stride)
-{
- int16_t transformed_data[64];
- int16_t *op;
- int i, j;
- int16_t sample;
+ uint32_t filter_limit_values[64];
+ int bounding_values_array[256];
+} Vp3DecodeContext;
- vp3_idct_c(input_data, dequant_matrix, transformed_data);
-
- /* place in final output */
- op = transformed_data;
- for (i = 0; i < 8; i++) {
- for (j = 0; j < 8; j++) {
- sample = *dest + *op;
- if (sample < 0)
- *dest = 0;
- else if (sample > 255)
- *dest = 255;
- else
- *dest = (uint8_t)(sample & 0xFF);
- op++;
- dest++;
- }
- dest += (stride - 8);
- }
-}
+static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb);
+static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb);
/************************************************************************
* VP3 specific functions
break;
default:
- printf (" vp3: help! Got a bad token: %d > 31\n", token);
+ av_log(NULL, AV_LOG_ERROR, " vp3: help! Got a bad token: %d > 31\n", token);
break;
}
/* zero out all of the fragment information */
s->coded_fragment_list_index = 0;
for (i = 0; i < s->fragment_count; i++) {
- memset(s->all_fragments[i].coeffs, 0, 64 * sizeof(DCTELEM));
s->all_fragments[i].coeff_count = 0;
- s->all_fragments[i].last_coeff = 0;
-s->all_fragments[i].motion_x = 0xbeef;
-s->all_fragments[i].motion_y = 0xbeef;
+ s->all_fragments[i].motion_x = 127;
+ s->all_fragments[i].motion_y = 127;
+ s->all_fragments[i].next_coeff= NULL;
+ s->coeffs[i].index=
+ s->coeffs[i].coeff=0;
+ s->coeffs[i].next= NULL;
}
}
/*
- * This function sets of the dequantization tables used for a particular
+ * This function sets up the dequantization tables used for a particular
* frame.
*/
static void init_dequantizer(Vp3DecodeContext *s)
{
- int quality_scale = vp31_quality_threshold[s->quality_index];
- int dc_scale_factor = vp31_dc_scale_factor[s->quality_index];
+ int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
+ int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
int i, j;
debug_vp3(" vp3: initializing dequantization tables\n");
* 100
*
* where sf = dc_scale_factor for DC quantizer
- * or quality_scale for AC quantizer
+ * or ac_scale_factor for AC quantizer
*
* Then, saturate the result to a lower limit of MIN_DEQUANT_VAL.
*/
#define SCALER 4
/* scale DC quantizers */
- s->intra_y_dequant[0] = vp31_intra_y_dequant[0] * dc_scale_factor / 100;
+ s->intra_y_dequant[0] = s->coded_intra_y_dequant[0] * dc_scale_factor / 100;
if (s->intra_y_dequant[0] < MIN_DEQUANT_VAL * 2)
s->intra_y_dequant[0] = MIN_DEQUANT_VAL * 2;
s->intra_y_dequant[0] *= SCALER;
- s->intra_c_dequant[0] = vp31_intra_c_dequant[0] * dc_scale_factor / 100;
+ s->intra_c_dequant[0] = s->coded_intra_c_dequant[0] * dc_scale_factor / 100;
if (s->intra_c_dequant[0] < MIN_DEQUANT_VAL * 2)
s->intra_c_dequant[0] = MIN_DEQUANT_VAL * 2;
s->intra_c_dequant[0] *= SCALER;
- s->inter_dequant[0] = vp31_inter_dequant[0] * dc_scale_factor / 100;
+ s->inter_dequant[0] = s->coded_inter_dequant[0] * dc_scale_factor / 100;
if (s->inter_dequant[0] < MIN_DEQUANT_VAL * 4)
s->inter_dequant[0] = MIN_DEQUANT_VAL * 4;
s->inter_dequant[0] *= SCALER;
/* scale AC quantizers, zigzag at the same time in preparation for
* the dequantization phase */
for (i = 1; i < 64; i++) {
+ int k= s->scantable.scantable[i];
+ j = s->scantable.permutated[i];
- j = zigzag_index[i];
-
- s->intra_y_dequant[j] = vp31_intra_y_dequant[i] * quality_scale / 100;
+ s->intra_y_dequant[j] = s->coded_intra_y_dequant[k] * ac_scale_factor / 100;
if (s->intra_y_dequant[j] < MIN_DEQUANT_VAL)
s->intra_y_dequant[j] = MIN_DEQUANT_VAL;
s->intra_y_dequant[j] *= SCALER;
- s->intra_c_dequant[j] = vp31_intra_c_dequant[i] * quality_scale / 100;
+ s->intra_c_dequant[j] = s->coded_intra_c_dequant[k] * ac_scale_factor / 100;
if (s->intra_c_dequant[j] < MIN_DEQUANT_VAL)
s->intra_c_dequant[j] = MIN_DEQUANT_VAL;
s->intra_c_dequant[j] *= SCALER;
- s->inter_dequant[j] = vp31_inter_dequant[i] * quality_scale / 100;
+ s->inter_dequant[j] = s->coded_inter_dequant[k] * ac_scale_factor / 100;
if (s->inter_dequant[j] < MIN_DEQUANT_VAL * 2)
s->inter_dequant[j] = MIN_DEQUANT_VAL * 2;
s->inter_dequant[j] *= SCALER;
}
+
+ memset(s->qscale_table, (FFMAX(s->intra_y_dequant[1], s->intra_c_dequant[1])+8)/16, 512); //FIXME finetune
/* print debug information as requested */
debug_dequantizers("intra Y dequantizers:\n");
debug_dequantizers("\n");
}
+/*
+ * This function initializes the loop filter boundary limits if the frame's
+ * quality index is different from the previous frame's.
+ */
+static void init_loop_filter(Vp3DecodeContext *s)
+{
+ int *bounding_values= s->bounding_values_array+127;
+ int filter_limit;
+ int x;
+
+ filter_limit = s->filter_limit_values[s->quality_index];
+
+ /* set up the bounding values */
+ memset(s->bounding_values_array, 0, 256 * sizeof(int));
+ for (x = 0; x < filter_limit; x++) {
+ bounding_values[-x - filter_limit] = -filter_limit + x;
+ bounding_values[-x] = -x;
+ bounding_values[x] = x;
+ bounding_values[x + filter_limit] = filter_limit - x;
+ }
+}
+
/*
* This function is used to fetch runs of 1s or 0s from the bitstream for
* use in determining which superblocks are fully and partially coded.
* fetched the bit will be toggled again */
bit ^= 1;
while (current_superblock < s->superblock_count) {
- if (current_run == 0) {
+ if (current_run-- == 0) {
bit ^= 1;
+#if 1
+ current_run = get_vlc2(gb,
+ s->superblock_run_length_vlc.table, 6, 2);
+ if (current_run == 33)
+ current_run += get_bits(gb, 12);
+#else
current_run = get_superblock_run_length(gb);
+#endif
debug_block_coding(" setting superblocks %d..%d to %s\n",
current_superblock,
current_superblock + current_run - 1,
decode_partial_blocks = 1;
}
}
- s->superblock_coding[current_superblock++] =
- (bit) ? SB_PARTIALLY_CODED : SB_NOT_CODED;
- current_run--;
+ s->superblock_coding[current_superblock++] = bit;
}
/* unpack the list of fully coded superblocks if any of the blocks were
/* skip any superblocks already marked as partially coded */
if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
- if (current_run == 0) {
+ if (current_run-- == 0) {
bit ^= 1;
+#if 1
+ current_run = get_vlc2(gb,
+ s->superblock_run_length_vlc.table, 6, 2);
+ if (current_run == 33)
+ current_run += get_bits(gb, 12);
+#else
current_run = get_superblock_run_length(gb);
+#endif
}
debug_block_coding(" setting superblock %d to %s\n",
current_superblock,
(bit) ? "fully coded" : "not coded");
- s->superblock_coding[current_superblock] =
- (bit) ? SB_FULLY_CODED : SB_NOT_CODED;
- current_run--;
+ s->superblock_coding[current_superblock] = 2*bit;
}
current_superblock++;
}
/* 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;
/* if the fragment is in bounds, check its coding status */
current_fragment = s->superblock_fragments[i * 16 + j];
if (current_fragment >= s->fragment_count) {
- printf (" vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
current_fragment, s->fragment_count);
return 1;
}
/* fragment may or may not be coded; this is the case
* that cares about the fragment coding runs */
- if (current_run == 0) {
+ if (current_run-- == 0) {
bit ^= 1;
+#if 1
+ current_run = get_vlc2(gb,
+ s->fragment_run_length_vlc.table, 5, 2);
+#else
current_run = get_fragment_run_length(gb);
+#endif
}
if (bit) {
* 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] =
current_fragment;
if ((current_fragment >= s->u_fragment_start) &&
i, current_fragment);
}
- current_run--;
-
} 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->u_fragment_start) &&
(s->macroblock_coding[current_macroblock] == MODE_COPY))
continue;
if (current_macroblock >= s->macroblock_count) {
- printf (" vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
current_macroblock, s->macroblock_count);
return 1;
}
if (scheme == 7)
coding_mode = get_bits(gb, 3);
else
+{
+#if 1
+ coding_mode = ModeAlphabet[scheme]
+ [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
+#else
coding_mode = ModeAlphabet[scheme][get_mode_code(gb)];
+#endif
+}
s->macroblock_coding[current_macroblock] = coding_mode;
for (k = 0; k < 6; k++) {
if (current_fragment == -1)
continue;
if (current_fragment >= s->fragment_count) {
- printf (" vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
current_fragment, s->fragment_count);
return 1;
}
(s->macroblock_coding[current_macroblock] == MODE_COPY))
continue;
if (current_macroblock >= s->macroblock_count) {
- printf (" vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
current_macroblock, s->macroblock_count);
return 1;
}
current_fragment = s->macroblock_fragments[current_macroblock * 6];
if (current_fragment >= s->fragment_count) {
- printf (" vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
current_fragment, s->fragment_count);
return 1;
}
case MODE_GOLDEN_MV:
/* all 6 fragments use the same motion vector */
if (coding_mode == 0) {
+#if 1
+ motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
+ motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
+#else
motion_x[0] = get_motion_vector_vlc(gb);
motion_y[0] = get_motion_vector_vlc(gb);
+#endif
} else {
+#if 1
+ motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
+ motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
+#else
motion_x[0] = get_motion_vector_fixed(gb);
motion_y[0] = get_motion_vector_fixed(gb);
+#endif
}
+
for (k = 1; k < 6; k++) {
motion_x[k] = motion_x[0];
motion_y[k] = motion_y[0];
motion_x[4] = motion_y[4] = 0;
for (k = 0; k < 4; k++) {
if (coding_mode == 0) {
+#if 1
+ motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
+ motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
+#else
motion_x[k] = get_motion_vector_vlc(gb);
motion_y[k] = get_motion_vector_vlc(gb);
+#endif
} else {
+#if 1
+ motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
+ motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
+#else
motion_x[k] = get_motion_vector_fixed(gb);
motion_y[k] = get_motion_vector_fixed(gb);
+#endif
}
motion_x[4] += motion_x[k];
motion_y[4] += motion_y[k];
if (current_fragment == -1)
continue;
if (current_fragment >= s->fragment_count) {
- printf (" vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
current_fragment, s->fragment_count);
return 1;
}
{
int i;
int token;
- int zero_run;
- DCTELEM coeff;
+ int zero_run = 0;
+ DCTELEM coeff = 0;
Vp3Fragment *fragment;
+ uint8_t *perm= s->scantable.permutated;
+ int bits_to_get;
if ((first_fragment >= s->fragment_count) ||
(last_fragment >= s->fragment_count)) {
- printf (" vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
first_fragment, last_fragment);
return 0;
}
token = get_vlc2(gb, table->table, 5, 3);
debug_vlc(" token = %2d, ", token);
/* use the token to get a zero run, a coefficient, and an eob run */
+#if 1
+ if (token <= 6) {
+ 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 {
+ bits_to_get = coeff_get_bits[token];
+ if (!bits_to_get)
+ coeff = coeff_tables[token][0];
+ else
+ coeff = coeff_tables[token][get_bits(gb, 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]);
+ }
+#else
unpack_token(gb, token, &zero_run, &coeff, &eob_run);
+#endif
}
if (!eob_run) {
fragment->coeff_count += zero_run;
- if (fragment->coeff_count < 64)
- fragment->coeffs[fragment->coeff_count++] = coeff;
+ if (fragment->coeff_count < 64){
+ fragment->next_coeff->coeff= coeff;
+ fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
+ fragment->next_coeff->next= s->next_coeff;
+ s->next_coeff->next=NULL;
+ fragment->next_coeff= s->next_coeff++;
+ }
debug_vlc(" fragment %d coeff = %d\n",
- s->coded_fragment_list[i], fragment->coeffs[coeff_index]);
+ s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
} else {
- fragment->last_coeff = fragment->coeff_count;
- fragment->coeff_count = 64;
+ fragment->coeff_count |= 128;
debug_vlc(" fragment %d eob with %d coefficients\n",
- s->coded_fragment_list[i], fragment->last_coeff);
+ s->coded_fragment_list[i], fragment->coeff_count&127);
eob_run--;
}
}
#define COMPATIBLE_FRAME(x) \
(compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
#define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
+#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
static inline int iabs (int x) { return ((x < 0) ? -x : x); }
static void reverse_dc_prediction(Vp3DecodeContext *s,
predictor_group = (x == 0) + ((y == 0) << 1) +
((x + 1 == fragment_width) << 2);
debug_dc_pred(" frag %d: group %d, orig DC = %d, ",
- i, predictor_group, s->all_fragments[i].coeffs[0]);
+ i, predictor_group, DC_COEFF(i));
switch (predictor_group) {
l = i - 1;
/* fetch the DC values for the predicting fragments */
- vul = s->all_fragments[ul].coeffs[0];
- vu = s->all_fragments[u].coeffs[0];
- vur = s->all_fragments[ur].coeffs[0];
- vl = s->all_fragments[l].coeffs[0];
+ vul = DC_COEFF(ul);
+ vu = DC_COEFF(u);
+ vur = DC_COEFF(ur);
+ vl = DC_COEFF(l);
/* figure out which fragments are valid */
ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
ur = i - fragment_width + 1;
/* fetch the DC values for the predicting fragments */
- vu = s->all_fragments[u].coeffs[0];
- vur = s->all_fragments[ur].coeffs[0];
+ vu = DC_COEFF(u);
+ vur = DC_COEFF(ur);
/* figure out which fragments are valid */
fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
l = i - 1;
/* fetch the DC values for the predicting fragments */
- vl = s->all_fragments[l].coeffs[0];
+ vl = DC_COEFF(l);
/* figure out which fragments are valid */
fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
l = i - 1;
/* fetch the DC values for the predicting fragments */
- vul = s->all_fragments[ul].coeffs[0];
- vu = s->all_fragments[u].coeffs[0];
- vl = s->all_fragments[l].coeffs[0];
+ vul = DC_COEFF(ul);
+ vu = DC_COEFF(u);
+ vl = DC_COEFF(l);
/* figure out which fragments are valid */
ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
/* if there were no fragments to predict from, use last
* DC saved */
- s->all_fragments[i].coeffs[0] += last_dc[current_frame_type];
+ predicted_dc = last_dc[current_frame_type];
debug_dc_pred("from last DC (%d) = %d\n",
- current_frame_type, s->all_fragments[i].coeffs[0]);
+ current_frame_type, DC_COEFF(i));
} else {
predicted_dc = vul;
}
- /* at long last, apply the predictor */
- s->all_fragments[i].coeffs[0] += predicted_dc;
debug_dc_pred("from pred DC = %d\n",
- s->all_fragments[i].coeffs[0]);
+ DC_COEFF(i));
}
+ /* 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;
/* save the DC */
- last_dc[current_frame_type] = s->all_fragments[i].coeffs[0];
+ last_dc[current_frame_type] = DC_COEFF(i);
+ if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
+ s->all_fragments[i].coeff_count= 129;
+// s->all_fragments[i].next_coeff= s->next_coeff;
+ s->coeffs[i].next= s->next_coeff;
+ (s->next_coeff++)->next=NULL;
+ }
}
}
}
}
+
+static void horizontal_filter(unsigned char *first_pixel, int stride,
+ int *bounding_values);
+static void vertical_filter(unsigned char *first_pixel, int stride,
+ int *bounding_values);
+
+/*
+ * Perform the final rendering for a particular slice of data.
+ * The slice number ranges from 0..(macroblock_height - 1).
+ */
+static void render_slice(Vp3DecodeContext *s, int slice)
+{
+ int x, y;
+ int m, n;
+ int i; /* indicates current fragment */
+ int16_t *dequantizer;
+ DCTELEM __align16 block[64];
+ unsigned char *output_plane;
+ unsigned char *last_plane;
+ unsigned char *golden_plane;
+ int stride;
+ int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
+ int upper_motion_limit, lower_motion_limit;
+ int motion_halfpel_index;
+ uint8_t *motion_source;
+ int plane;
+ int plane_width;
+ int plane_height;
+ int slice_height;
+ int current_macroblock_entry = slice * s->macroblock_width * 6;
+ int *bounding_values= s->bounding_values_array+127;
+ int fragment_width;
+
+ if (slice >= s->macroblock_height)
+ return;
+
+ for (plane = 0; plane < 3; plane++) {
+
+ /* set up plane-specific parameters */
+ if (plane == 0) {
+ output_plane = s->current_frame.data[0];
+ last_plane = s->last_frame.data[0];
+ golden_plane = s->golden_frame.data[0];
+ stride = s->current_frame.linesize[0];
+ if (!s->flipped_image) stride = -stride;
+ upper_motion_limit = 7 * s->current_frame.linesize[0];
+ lower_motion_limit = s->height * s->current_frame.linesize[0] + s->width - 8;
+ y = slice * FRAGMENT_PIXELS * 2;
+ plane_width = s->width;
+ plane_height = s->height;
+ slice_height = y + FRAGMENT_PIXELS * 2;
+ i = s->macroblock_fragments[current_macroblock_entry + 0];
+ } else if (plane == 1) {
+ output_plane = s->current_frame.data[1];
+ last_plane = s->last_frame.data[1];
+ golden_plane = s->golden_frame.data[1];
+ stride = s->current_frame.linesize[1];
+ if (!s->flipped_image) stride = -stride;
+ upper_motion_limit = 7 * s->current_frame.linesize[1];
+ lower_motion_limit = (s->height / 2) * s->current_frame.linesize[1] + (s->width / 2) - 8;
+ y = slice * FRAGMENT_PIXELS;
+ plane_width = s->width / 2;
+ plane_height = s->height / 2;
+ slice_height = y + FRAGMENT_PIXELS;
+ i = s->macroblock_fragments[current_macroblock_entry + 4];
+ } else {
+ output_plane = s->current_frame.data[2];
+ last_plane = s->last_frame.data[2];
+ golden_plane = s->golden_frame.data[2];
+ stride = s->current_frame.linesize[2];
+ if (!s->flipped_image) stride = -stride;
+ upper_motion_limit = 7 * s->current_frame.linesize[2];
+ lower_motion_limit = (s->height / 2) * s->current_frame.linesize[2] + (s->width / 2) - 8;
+ y = slice * FRAGMENT_PIXELS;
+ plane_width = s->width / 2;
+ plane_height = s->height / 2;
+ slice_height = y + FRAGMENT_PIXELS;
+ i = s->macroblock_fragments[current_macroblock_entry + 5];
+ }
+ fragment_width = plane_width / FRAGMENT_PIXELS;
+
+ if(ABS(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 fragment in a row... */
+ for (x = 0; x < plane_width; x += 8, i++) {
+
+ if ((i < 0) || (i >= s->fragment_count)) {
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
+ return;
+ }
+
+ /* transform if this block was coded */
+ if ((s->all_fragments[i].coding_method != MODE_COPY) &&
+ !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
+
+ if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
+ (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
+ motion_source= golden_plane;
+ else
+ motion_source= last_plane;
+
+ motion_source += s->all_fragments[i].first_pixel;
+ motion_halfpel_index = 0;
+
+ /* sort out the motion vector if this fragment is coded
+ * using a motion vector method */
+ 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);
+ }
+
+ 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);
+
+ motion_halfpel_index = motion_x & 0x01;
+ motion_source += (motion_x >> 1);
+
+ motion_halfpel_index |= (motion_y & 0x01) << 1;
+ motion_source += ((motion_y >> 1) * stride);
+
+ 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;
+
+ ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
+ motion_source= temp;
+ }
+ }
+
+
+ /* first, take care of copying a block from either the
+ * previous or the golden frame */
+ if (s->all_fragments[i].coding_method != MODE_INTRA) {
+ /* Note, it is possible to implement all MC cases with
+ put_no_rnd_pixels_l2 which would look more like the
+ 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](
+ output_plane + s->all_fragments[i].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](
+ output_plane + s->all_fragments[i].first_pixel,
+ motion_source - d,
+ motion_source + stride + 1 + d,
+ stride, 8);
+ }
+ dequantizer = s->inter_dequant;
+ }else{
+ if (plane == 0)
+ dequantizer = s->intra_y_dequant;
+ else
+ dequantizer = s->intra_c_dequant;
+ }
+
+ /* dequantize the DCT coefficients */
+ debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
+ i, s->all_fragments[i].coding_method,
+ DC_COEFF(i), dequantizer[0]);
+
+ if(s->avctx->idct_algo==FF_IDCT_VP3){
+ Coeff *coeff= s->coeffs + i;
+ memset(block, 0, sizeof(block));
+ while(coeff->next){
+ block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
+ coeff= coeff->next;
+ }
+ }else{
+ Coeff *coeff= s->coeffs + i;
+ memset(block, 0, sizeof(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(
+ output_plane + s->all_fragments[i].first_pixel,
+ stride,
+ block);
+ } else {
+ s->dsp.idct_add(
+ output_plane + s->all_fragments[i].first_pixel,
+ stride,
+ block);
+ }
+
+ debug_idct("block after idct_%s():\n",
+ (s->all_fragments[i].coding_method == MODE_INTRA)?
+ "put" : "add");
+ for (m = 0; m < 8; m++) {
+ for (n = 0; n < 8; n++) {
+ debug_idct(" %3d", *(output_plane +
+ s->all_fragments[i].first_pixel + (m * stride + n)));
+ }
+ debug_idct("\n");
+ }
+ debug_idct("\n");
+
+ } else {
+
+ /* copy directly from the previous frame */
+ s->dsp.put_pixels_tab[1][0](
+ output_plane + s->all_fragments[i].first_pixel,
+ last_plane + s->all_fragments[i].first_pixel,
+ stride, 8);
+
+ }
+#if 0
+ /* perform the left edge filter if:
+ * - the fragment is not on the left column
+ * - the fragment is coded in this frame
+ * - the fragment is not coded in this frame but the left
+ * fragment is coded in this frame (this is done instead
+ * of a right edge filter when rendering the left fragment
+ * since this fragment is not available yet) */
+ if ((x > 0) &&
+ ((s->all_fragments[i].coding_method != MODE_COPY) ||
+ ((s->all_fragments[i].coding_method == MODE_COPY) &&
+ (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
+ horizontal_filter(
+ output_plane + s->all_fragments[i].first_pixel + 7*stride,
+ -stride, bounding_values);
+ }
+
+ /* perform the top edge filter if:
+ * - the fragment is not on the top row
+ * - the fragment is coded in this frame
+ * - the fragment is not coded in this frame but the above
+ * fragment is coded in this frame (this is done instead
+ * of a bottom edge filter when rendering the above
+ * fragment since this fragment is not available yet) */
+ if ((y > 0) &&
+ ((s->all_fragments[i].coding_method != MODE_COPY) ||
+ ((s->all_fragments[i].coding_method == MODE_COPY) &&
+ (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
+ vertical_filter(
+ output_plane + s->all_fragments[i].first_pixel - stride,
+ -stride, bounding_values);
+ }
+#endif
+ }
+ }
+ }
+
+ /* this looks like a good place for slice dispatch... */
+ /* algorithm:
+ * if (slice == s->macroblock_height - 1)
+ * dispatch (both last slice & 2nd-to-last slice);
+ * else if (slice > 0)
+ * dispatch (slice - 1);
+ */
+
+ emms_c();
+}
+
/*
* This function performs the final rendering of each fragment's data
* onto the output frame.
int m, n;
int i = first_fragment;
int16_t *dequantizer;
+ DCTELEM __align16 block[64];
unsigned char *output_plane;
unsigned char *last_plane;
unsigned char *golden_plane;
int stride;
- int motion_x, motion_y;
+ int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
int upper_motion_limit, lower_motion_limit;
int motion_halfpel_index;
- unsigned int motion_source;
+ uint8_t *motion_source;
debug_vp3(" vp3: rendering final fragments for %s\n",
(plane == 0) ? "Y plane" : (plane == 1) ? "U plane" : "V plane");
/* set up plane-specific parameters */
if (plane == 0) {
- dequantizer = s->intra_y_dequant;
output_plane = s->current_frame.data[0];
last_plane = s->last_frame.data[0];
golden_plane = s->golden_frame.data[0];
- stride = -s->current_frame.linesize[0];
+ stride = s->current_frame.linesize[0];
+ if (!s->flipped_image) stride = -stride;
upper_motion_limit = 7 * s->current_frame.linesize[0];
lower_motion_limit = height * s->current_frame.linesize[0] + width - 8;
} else if (plane == 1) {
- dequantizer = s->intra_c_dequant;
output_plane = s->current_frame.data[1];
last_plane = s->last_frame.data[1];
golden_plane = s->golden_frame.data[1];
- stride = -s->current_frame.linesize[1];
+ stride = s->current_frame.linesize[1];
+ if (!s->flipped_image) stride = -stride;
upper_motion_limit = 7 * s->current_frame.linesize[1];
lower_motion_limit = height * s->current_frame.linesize[1] + width - 8;
} else {
- dequantizer = s->intra_c_dequant;
output_plane = s->current_frame.data[2];
last_plane = s->last_frame.data[2];
golden_plane = s->golden_frame.data[2];
- stride = -s->current_frame.linesize[2];
+ stride = s->current_frame.linesize[2];
+ if (!s->flipped_image) stride = -stride;
upper_motion_limit = 7 * s->current_frame.linesize[2];
lower_motion_limit = height * s->current_frame.linesize[2] + width - 8;
}
+
+ if(ABS(stride) > 2048)
+ return; //various tables are fixed size
/* for each fragment row... */
for (y = 0; y < height; y += 8) {
for (x = 0; x < width; x += 8, i++) {
if ((i < 0) || (i >= s->fragment_count)) {
- printf (" vp3:render_fragments(): bad fragment number (%d)\n", i);
+ av_log(s->avctx, AV_LOG_ERROR, " vp3:render_fragments(): bad fragment number (%d)\n", i);
return;
}
/* transform if this block was coded */
- if (s->all_fragments[i].coding_method != MODE_COPY) {
+ if ((s->all_fragments[i].coding_method != MODE_COPY) &&
+ !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
+
+ if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
+ (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
+ motion_source= golden_plane;
+ else
+ motion_source= last_plane;
- motion_source = s->all_fragments[i].first_pixel;
+ motion_source += s->all_fragments[i].first_pixel;
motion_halfpel_index = 0;
/* sort out the motion vector if this fragment is coded
* using a motion vector method */
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 ((motion_x == 0xbeef) || (motion_y == 0xbeef))
-printf (" help! got beefy vector! (%X, %X)\n", motion_x, motion_y);
-
- if (motion_x >= 0) {
- motion_halfpel_index = motion_x & 0x01;
- motion_source += (motion_x >> 1);
- } else {
- motion_x = -motion_x;
- motion_halfpel_index = motion_x & 0x01;
- motion_source -= ((motion_x + 1) >> 1);
+ if(plane){
+ motion_x= (motion_x>>1) | (motion_x&1);
+ motion_y= (motion_y>>1) | (motion_y&1);
}
-// motion_y = -motion_y;
- if (motion_y >= 0) {
- motion_halfpel_index |= (motion_y & 0x01) << 1;
- motion_source += ((motion_y >> 1) * stride);
- } else {
- motion_y = -motion_y;
- motion_halfpel_index |= (motion_y & 0x01) << 1;
- motion_source -= (((motion_y + 1) >> 1) * stride);
- }
+ 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);
- /* if the are any problems with a motion vector, refuse
- * to render the block */
- if ((motion_source < upper_motion_limit) ||
- (motion_source > lower_motion_limit)) {
- printf (" vp3: help! motion source (%d) out of range (%d..%d), fragment %d\n",
- motion_source, upper_motion_limit, lower_motion_limit, i);
- continue;
+ motion_halfpel_index = motion_x & 0x01;
+ motion_source += (motion_x >> 1);
+
+ motion_halfpel_index |= (motion_y & 0x01) << 1;
+ motion_source += ((motion_y >> 1) * stride);
+
+ if(src_x<0 || src_y<0 || src_x + 9 >= width || src_y + 9 >= height){
+ uint8_t *temp= s->edge_emu_buffer;
+ if(stride<0) temp -= 9*stride;
+ else temp += 9*stride;
+
+ ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, width, height);
+ motion_source= temp;
}
}
+
/* first, take care of copying a block from either the
* previous or the golden frame */
- if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
- (s->all_fragments[i].coding_method == MODE_GOLDEN_MV)) {
-
- s->dsp.put_pixels_tab[1][motion_halfpel_index](
- output_plane + s->all_fragments[i].first_pixel,
- golden_plane + motion_source,
- stride, 8);
-
- } else
if (s->all_fragments[i].coding_method != MODE_INTRA) {
-
- s->dsp.put_pixels_tab[1][motion_halfpel_index](
- output_plane + s->all_fragments[i].first_pixel,
- last_plane + motion_source,
- stride, 8);
+ //Note, it is possible to implement all MC cases with put_no_rnd_pixels_l2 which would look more like the 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](
+ output_plane + s->all_fragments[i].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](
+ output_plane + s->all_fragments[i].first_pixel,
+ motion_source - d,
+ motion_source + stride + 1 + d,
+ stride, 8);
+ }
+ dequantizer = s->inter_dequant;
+ }else{
+ if (plane == 0)
+ dequantizer = s->intra_y_dequant;
+ else
+ dequantizer = s->intra_c_dequant;
}
/* dequantize the DCT coefficients */
debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
i, s->all_fragments[i].coding_method,
- s->all_fragments[i].coeffs[0], dequantizer[0]);
+ DC_COEFF(i), dequantizer[0]);
+
+ if(s->avctx->idct_algo==FF_IDCT_VP3){
+ Coeff *coeff= s->coeffs + i;
+ memset(block, 0, sizeof(block));
+ while(coeff->next){
+ block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
+ coeff= coeff->next;
+ }
+ }else{
+ Coeff *coeff= s->coeffs + i;
+ memset(block, 0, sizeof(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) {
- vp3_idct_put(s->all_fragments[i].coeffs, dequantizer,
+ if(s->avctx->idct_algo!=FF_IDCT_VP3)
+ block[0] += 128<<3;
+ s->dsp.idct_put(
output_plane + s->all_fragments[i].first_pixel,
- stride);
+ stride,
+ block);
} else {
- vp3_idct_add(s->all_fragments[i].coeffs, dequantizer,
+ s->dsp.idct_add(
output_plane + s->all_fragments[i].first_pixel,
- stride);
+ stride,
+ block);
}
debug_idct("block after idct_%s():\n",
}
emms_c();
+}
+
+static void horizontal_filter(unsigned char *first_pixel, int stride,
+ int *bounding_values)
+{
+ unsigned char *end;
+ int filter_value;
+
+ for (end= first_pixel + 8*stride; first_pixel < end; first_pixel += stride) {
+ filter_value =
+ (first_pixel[-2] - first_pixel[ 1])
+ +3*(first_pixel[ 0] - first_pixel[-1]);
+ filter_value = bounding_values[(filter_value + 4) >> 3];
+ first_pixel[-1] = clip_uint8(first_pixel[-1] + filter_value);
+ first_pixel[ 0] = clip_uint8(first_pixel[ 0] - filter_value);
+ }
+}
+static void vertical_filter(unsigned char *first_pixel, int stride,
+ int *bounding_values)
+{
+ unsigned char *end;
+ int filter_value;
+ const int nstride= -stride;
+
+ for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
+ filter_value =
+ (first_pixel[2 * nstride] - first_pixel[ stride])
+ +3*(first_pixel[0 ] - first_pixel[nstride]);
+ filter_value = bounding_values[(filter_value + 4) >> 3];
+ first_pixel[nstride] = clip_uint8(first_pixel[nstride] + filter_value);
+ first_pixel[0] = clip_uint8(first_pixel[0] - filter_value);
+ }
+}
+
+static void apply_loop_filter(Vp3DecodeContext *s)
+{
+ int x, y, plane;
+ int width, height;
+ int fragment;
+ int stride;
+ unsigned char *plane_data;
+ int *bounding_values= s->bounding_values_array+127;
+
+#if 0
+ int bounding_values_array[256];
+ int filter_limit;
+
+ /* find the right loop limit value */
+ for (x = 63; x >= 0; x--) {
+ if (vp31_ac_scale_factor[x] >= s->quality_index)
+ break;
+ }
+ filter_limit = vp31_filter_limit_values[s->quality_index];
+
+ /* set up the bounding values */
+ memset(bounding_values_array, 0, 256 * sizeof(int));
+ for (x = 0; x < filter_limit; x++) {
+ bounding_values[-x - filter_limit] = -filter_limit + x;
+ bounding_values[-x] = -x;
+ bounding_values[x] = x;
+ bounding_values[x + filter_limit] = filter_limit - x;
+ }
+#endif
+
+ for (plane = 0; plane < 3; plane++) {
+
+ if (plane == 0) {
+ /* Y plane parameters */
+ fragment = 0;
+ width = s->fragment_width;
+ height = s->fragment_height;
+ stride = s->current_frame.linesize[0];
+ plane_data = s->current_frame.data[0];
+ } else if (plane == 1) {
+ /* U plane parameters */
+ fragment = s->u_fragment_start;
+ width = s->fragment_width / 2;
+ height = s->fragment_height / 2;
+ stride = s->current_frame.linesize[1];
+ plane_data = s->current_frame.data[1];
+ } else {
+ /* V plane parameters */
+ fragment = s->v_fragment_start;
+ width = s->fragment_width / 2;
+ height = s->fragment_height / 2;
+ stride = s->current_frame.linesize[2];
+ plane_data = s->current_frame.data[2];
+ }
+
+ for (y = 0; y < height; y++) {
+
+ for (x = 0; x < width; x++) {
+START_TIMER
+ /* do not perform left edge filter for left columns frags */
+ if ((x > 0) &&
+ (s->all_fragments[fragment].coding_method != MODE_COPY)) {
+ horizontal_filter(
+ plane_data + s->all_fragments[fragment].first_pixel - 7*stride,
+ stride, bounding_values);
+ }
+
+ /* do not perform top edge filter for top row fragments */
+ if ((y > 0) &&
+ (s->all_fragments[fragment].coding_method != MODE_COPY)) {
+ vertical_filter(
+ plane_data + s->all_fragments[fragment].first_pixel + stride,
+ stride, bounding_values);
+ }
+
+ /* do not perform right edge filter for right column
+ * fragments or if right fragment neighbor is also coded
+ * in this frame (it will be filtered in next iteration) */
+ if ((x < width - 1) &&
+ (s->all_fragments[fragment].coding_method != MODE_COPY) &&
+ (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
+ horizontal_filter(
+ plane_data + s->all_fragments[fragment + 1].first_pixel - 7*stride,
+ stride, bounding_values);
+ }
+
+ /* do not perform bottom edge filter for bottom row
+ * fragments or if bottom fragment neighbor is also coded
+ * in this frame (it will be filtered in the next row) */
+ if ((y < height - 1) &&
+ (s->all_fragments[fragment].coding_method != MODE_COPY) &&
+ (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
+ vertical_filter(
+ plane_data + s->all_fragments[fragment + width].first_pixel + stride,
+ stride, bounding_values);
+ }
+
+ fragment++;
+STOP_TIMER("loop filter")
+ }
+ }
+ }
}
/*
}
}
+/* FIXME: this should be merged with the above! */
+static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
+{
+
+ int i, x, y;
+
+ /* figure out the first pixel addresses for each of the fragments */
+ /* Y plane */
+ i = 0;
+ for (y = 1; y <= s->fragment_height; y++) {
+ for (x = 0; x < s->fragment_width; x++) {
+ s->all_fragments[i++].first_pixel =
+ s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
+ s->golden_frame.linesize[0] +
+ x * FRAGMENT_PIXELS;
+ debug_init(" fragment %d, first pixel @ %d\n",
+ i-1, s->all_fragments[i-1].first_pixel);
+ }
+ }
+
+ /* U plane */
+ i = s->u_fragment_start;
+ for (y = 1; y <= s->fragment_height / 2; y++) {
+ for (x = 0; x < s->fragment_width / 2; x++) {
+ s->all_fragments[i++].first_pixel =
+ s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
+ s->golden_frame.linesize[1] +
+ x * FRAGMENT_PIXELS;
+ debug_init(" fragment %d, first pixel @ %d\n",
+ i-1, s->all_fragments[i-1].first_pixel);
+ }
+ }
+
+ /* V plane */
+ i = s->v_fragment_start;
+ for (y = 1; y <= s->fragment_height / 2; y++) {
+ for (x = 0; x < s->fragment_width / 2; x++) {
+ s->all_fragments[i++].first_pixel =
+ s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
+ s->golden_frame.linesize[2] +
+ x * FRAGMENT_PIXELS;
+ debug_init(" fragment %d, first pixel @ %d\n",
+ i-1, s->all_fragments[i-1].first_pixel);
+ }
+ }
+}
+
/*
* This is the ffmpeg/libavcodec API init function.
*/
int y_superblock_count;
int c_superblock_count;
+ if (avctx->codec_tag == MKTAG('V','P','3','0'))
+ s->version = 0;
+ else
+ s->version = 1;
+
s->avctx = avctx;
-#if 0
- s->width = avctx->width;
- s->height = avctx->height;
-#else
s->width = (avctx->width + 15) & 0xFFFFFFF0;
s->height = (avctx->height + 15) & 0xFFFFFFF0;
-#endif
avctx->pix_fmt = PIX_FMT_YUV420P;
avctx->has_b_frames = 0;
+ 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);
/* initialize to an impossible value which will force a recalculation
* in the first frame decode */
s->v_fragment_start);
s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
+ s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
s->pixel_addresses_inited = 0;
- /* init VLC tables */
- for (i = 0; i < 16; i++) {
-
- /* DC histograms */
- init_vlc(&s->dc_vlc[i], 5, 32,
- &dc_bias[i][0][1], 4, 2,
- &dc_bias[i][0][0], 4, 2);
-
- /* group 1 AC histograms */
- init_vlc(&s->ac_vlc_1[i], 5, 32,
- &ac_bias_0[i][0][1], 4, 2,
- &ac_bias_0[i][0][0], 4, 2);
-
- /* group 2 AC histograms */
- init_vlc(&s->ac_vlc_2[i], 5, 32,
- &ac_bias_1[i][0][1], 4, 2,
- &ac_bias_1[i][0][0], 4, 2);
-
- /* group 3 AC histograms */
- init_vlc(&s->ac_vlc_3[i], 5, 32,
- &ac_bias_2[i][0][1], 4, 2,
- &ac_bias_2[i][0][0], 4, 2);
-
- /* group 4 AC histograms */
- init_vlc(&s->ac_vlc_4[i], 5, 32,
- &ac_bias_3[i][0][1], 4, 2,
- &ac_bias_3[i][0][0], 4, 2);
+ if (!s->theora_tables)
+ {
+ for (i = 0; i < 64; i++)
+ s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
+ for (i = 0; i < 64; i++)
+ s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
+ for (i = 0; i < 64; i++)
+ s->coded_intra_y_dequant[i] = vp31_intra_y_dequant[i];
+ for (i = 0; i < 64; i++)
+ s->coded_intra_c_dequant[i] = vp31_intra_c_dequant[i];
+ for (i = 0; i < 64; i++)
+ s->coded_inter_dequant[i] = vp31_inter_dequant[i];
+ for (i = 0; i < 64; i++)
+ s->filter_limit_values[i] = vp31_filter_limit_values[i];
+
+ /* init VLC tables */
+ for (i = 0; i < 16; i++) {
+
+ /* DC histograms */
+ init_vlc(&s->dc_vlc[i], 5, 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,
+ &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,
+ &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,
+ &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,
+ &ac_bias_3[i][0][1], 4, 2,
+ &ac_bias_3[i][0][0], 4, 2, 0);
+ }
+ } else {
+ for (i = 0; i < 16; i++) {
+
+ /* DC histograms */
+ 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);
+
+ /* group 1 AC histograms */
+ 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);
+
+ /* group 2 AC histograms */
+ 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);
+
+ /* group 3 AC histograms */
+ 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);
+
+ /* group 4 AC histograms */
+ 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);
+ }
}
- /* build quantization zigzag table */
- for (i = 0; i < 64; i++)
- zigzag_index[dezigzag_index[i]] = i;
+ init_vlc(&s->superblock_run_length_vlc, 6, 34,
+ &superblock_run_length_vlc_table[0][1], 4, 2,
+ &superblock_run_length_vlc_table[0][0], 4, 2, 0);
+
+ init_vlc(&s->fragment_run_length_vlc, 5, 31,
+ &fragment_run_length_vlc_table[0][1], 4, 2,
+ &fragment_run_length_vlc_table[0][0], 4, 2, 0);
+
+ init_vlc(&s->mode_code_vlc, 3, 8,
+ &mode_code_vlc_table[0][1], 2, 1,
+ &mode_code_vlc_table[0][0], 2, 1, 0);
+
+ init_vlc(&s->motion_vector_vlc, 6, 63,
+ &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));
Vp3DecodeContext *s = avctx->priv_data;
GetBitContext gb;
static int counter = 0;
-
- *data_size = 0;
+ int i;
init_get_bits(&gb, buf, buf_size * 8);
+
+ if (s->theora && get_bits1(&gb))
+ {
+ int ptype = get_bits(&gb, 7);
+
+ skip_bits(&gb, 6*8); /* "theora" */
+
+ switch(ptype)
+ {
+ case 1:
+ theora_decode_comments(avctx, gb);
+ break;
+ case 2:
+ theora_decode_tables(avctx, gb);
+ init_dequantizer(s);
+ break;
+ default:
+ av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype);
+ }
+ return buf_size;
+ }
- s->keyframe = get_bits(&gb, 1);
- s->keyframe ^= 1;
- skip_bits(&gb, 1);
+ s->keyframe = !get_bits1(&gb);
+ if (!s->theora)
+ skip_bits(&gb, 1);
s->last_quality_index = s->quality_index;
s->quality_index = get_bits(&gb, 6);
+ if (s->theora >= 0x030200)
+ skip_bits1(&gb);
- debug_vp3(" VP3 frame #%d: Q index = %d", counter, s->quality_index);
+ 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->quality_index);
counter++;
- if (s->quality_index != s->last_quality_index)
+ if (s->quality_index != s->last_quality_index) {
init_dequantizer(s);
+ init_loop_filter(s);
+ }
if (s->keyframe) {
-
- debug_vp3(", keyframe\n");
- /* skip the other 2 header bytes for now */
- skip_bits(&gb, 16);
+ if (!s->theora)
+ {
+ skip_bits(&gb, 4); /* width code */
+ skip_bits(&gb, 4); /* height code */
+ if (s->version)
+ {
+ s->version = get_bits(&gb, 5);
+ if (counter == 1)
+ av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
+ }
+ }
+ if (s->version || s->theora)
+ {
+ if (get_bits1(&gb))
+ 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);
avctx->release_buffer(avctx, &s->last_frame);
}
- s->golden_frame.reference = 0;
+ s->golden_frame.reference = 3;
if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
- printf("vp3: get_buffer() failed\n");
+ av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
return -1;
}
/* time to figure out pixel addresses? */
if (!s->pixel_addresses_inited)
- vp3_calculate_pixel_addresses(s);
-
+ {
+ if (!s->flipped_image)
+ vp3_calculate_pixel_addresses(s);
+ else
+ theora_calculate_pixel_addresses(s);
+ }
} else {
-
- debug_vp3("\n");
-
/* allocate a new current frame */
- s->current_frame.reference = 0;
+ s->current_frame.reference = 3;
if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
- printf("vp3: get_buffer() failed\n");
+ av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
return -1;
}
}
+ s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
+ s->current_frame.qstride= 0;
+
+ {START_TIMER
init_frame(s, &gb);
+ STOP_TIMER("init_frame")}
#if KEYFRAMES_ONLY
if (!s->keyframe) {
} else {
#endif
- if (unpack_superblocks(s, &gb) ||
- unpack_modes(s, &gb) ||
- unpack_vectors(s, &gb) ||
- unpack_dct_coeffs(s, &gb)) {
-
- printf(" vp3: could not decode frame\n");
+ {START_TIMER
+ if (unpack_superblocks(s, &gb)){
+ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
return -1;
}
+ STOP_TIMER("unpack_superblocks")}
+ {START_TIMER
+ if (unpack_modes(s, &gb)){
+ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
+ return -1;
+ }
+ STOP_TIMER("unpack_modes")}
+ {START_TIMER
+ if (unpack_vectors(s, &gb)){
+ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
+ return -1;
+ }
+ STOP_TIMER("unpack_vectors")}
+ {START_TIMER
+ if (unpack_dct_coeffs(s, &gb)){
+ av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
+ return -1;
+ }
+ STOP_TIMER("unpack_dct_coeffs")}
+ {START_TIMER
reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
- render_fragments(s, 0, s->width, s->height, 0);
-
if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
reverse_dc_prediction(s, s->u_fragment_start,
s->fragment_width / 2, s->fragment_height / 2);
reverse_dc_prediction(s, s->v_fragment_start,
s->fragment_width / 2, s->fragment_height / 2);
+ }
+ STOP_TIMER("reverse_dc_prediction")}
+ {START_TIMER
+
+#if 1
+ for (i = 0; i < s->macroblock_height; i++)
+ render_slice(s, i);
+#else
+ render_fragments(s, 0, s->width, s->height, 0);
+ if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
render_fragments(s, s->u_fragment_start, s->width / 2, s->height / 2, 1);
render_fragments(s, s->v_fragment_start, s->width / 2, s->height / 2, 2);
} else {
memset(s->current_frame.data[1], 0x80, s->width * s->height / 4);
memset(s->current_frame.data[2], 0x80, s->width * s->height / 4);
}
+#endif
+ STOP_TIMER("render_fragments")}
+ {START_TIMER
+ apply_loop_filter(s);
+ STOP_TIMER("apply_loop_filter")}
#if KEYFRAMES_ONLY
}
#endif
/* shuffle frames (last = current) */
memcpy(&s->last_frame, &s->current_frame, sizeof(AVFrame));
+ s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
return buf_size;
}
Vp3DecodeContext *s = avctx->priv_data;
av_free(s->all_fragments);
+ av_free(s->coeffs);
av_free(s->coded_fragment_list);
av_free(s->superblock_fragments);
av_free(s->superblock_macroblocks);
av_free(s->macroblock_fragments);
av_free(s->macroblock_coding);
-
+
/* release all frames */
- if (s->golden_frame.data[0])
+ 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);
return 0;
}
+static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
+{
+ Vp3DecodeContext *s = avctx->priv_data;
+
+ if (get_bits(gb, 1)) {
+ int token;
+ if (s->entries >= 32) { /* overflow */
+ av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
+ 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);
+ s->huffman_table[s->hti][token][0] = s->hbits;
+ s->huffman_table[s->hti][token][1] = s->huff_code_size;
+ s->entries++;
+ }
+ else {
+ if (s->huff_code_size >= 32) {/* overflow */
+ av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
+ return -1;
+ }
+ s->huff_code_size++;
+ s->hbits <<= 1;
+ read_huffman_tree(avctx, gb);
+ s->hbits |= 1;
+ read_huffman_tree(avctx, gb);
+ s->hbits >>= 1;
+ s->huff_code_size--;
+ }
+ return 0;
+}
+
+static int theora_decode_header(AVCodecContext *avctx, GetBitContext gb)
+{
+ Vp3DecodeContext *s = avctx->priv_data;
+ int major, minor, micro;
+
+ major = get_bits(&gb, 8); /* version major */
+ minor = get_bits(&gb, 8); /* version minor */
+ micro = get_bits(&gb, 8); /* version micro */
+ av_log(avctx, AV_LOG_INFO, "Theora bitstream version %d.%d.%d\n",
+ major, minor, micro);
+
+ /* FIXME: endianess? */
+ s->theora = (major << 16) | (minor << 8) | micro;
+
+ /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
+ /* but previous versions have the image flipped relative to vp3 */
+ if (s->theora < 0x030200)
+ {
+ s->flipped_image = 1;
+ av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
+ }
+
+ s->width = get_bits(&gb, 16) << 4;
+ s->height = get_bits(&gb, 16) << 4;
+
+ if(avcodec_check_dimensions(avctx, s->width, s->height)){
+ s->width= s->height= 0;
+ return -1;
+ }
+
+ skip_bits(&gb, 24); /* frame width */
+ skip_bits(&gb, 24); /* frame height */
+
+ skip_bits(&gb, 8); /* offset x */
+ skip_bits(&gb, 8); /* offset y */
+
+ skip_bits(&gb, 32); /* fps numerator */
+ skip_bits(&gb, 32); /* fps denumerator */
+ skip_bits(&gb, 24); /* aspect numerator */
+ skip_bits(&gb, 24); /* aspect denumerator */
+
+ if (s->theora < 0x030200)
+ skip_bits(&gb, 5); /* keyframe frequency force */
+ skip_bits(&gb, 8); /* colorspace */
+ skip_bits(&gb, 24); /* bitrate */
+
+ skip_bits(&gb, 6); /* last(?) quality index */
+
+ if (s->theora >= 0x030200)
+ {
+ skip_bits(&gb, 5); /* keyframe frequency force */
+ skip_bits(&gb, 5); /* spare bits */
+ }
+
+// align_get_bits(&gb);
+
+ avctx->width = s->width;
+ avctx->height = s->height;
+
+ return 0;
+}
+
+static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb)
+{
+ int nb_comments, i, tmp;
+
+ tmp = get_bits_long(&gb, 32);
+ tmp = be2me_32(tmp);
+ while(tmp--)
+ skip_bits(&gb, 8);
+
+ nb_comments = get_bits_long(&gb, 32);
+ nb_comments = be2me_32(nb_comments);
+ for (i = 0; i < nb_comments; i++)
+ {
+ tmp = get_bits_long(&gb, 32);
+ tmp = be2me_32(tmp);
+ while(tmp--)
+ skip_bits(&gb, 8);
+ }
+
+ return 0;
+}
+
+static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb)
+{
+ Vp3DecodeContext *s = avctx->priv_data;
+ int i, n;
+
+ 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->theora >= 0x030200)
+ n = get_bits(&gb, 4) + 1;
+ else
+ n = 16;
+ /* quality threshold table */
+ for (i = 0; i < 64; i++)
+ s->coded_ac_scale_factor[i] = get_bits(&gb, n);
+
+ if (s->theora >= 0x030200)
+ n = get_bits(&gb, 4) + 1;
+ else
+ n = 16;
+ /* dc scale factor table */
+ for (i = 0; i < 64; i++)
+ s->coded_dc_scale_factor[i] = get_bits(&gb, n);
+
+ if (s->theora >= 0x030200)
+ n = get_bits(&gb, 9) + 1;
+ else
+ n = 3;
+ if (n != 3) {
+ av_log(NULL,AV_LOG_ERROR, "unsupported nbms : %d\n", n);
+ return -1;
+ }
+ /* y coeffs */
+ for (i = 0; i < 64; i++)
+ s->coded_intra_y_dequant[i] = get_bits(&gb, 8);
+
+ /* uv coeffs */
+ for (i = 0; i < 64; i++)
+ s->coded_intra_c_dequant[i] = get_bits(&gb, 8);
+
+ /* inter coeffs */
+ for (i = 0; i < 64; i++)
+ s->coded_inter_dequant[i] = get_bits(&gb, 8);
+
+ /* Huffman tables */
+ for (i = 0; i <= 1; i++) {
+ for (n = 0; n <= 2; n++) {
+ int newqr;
+ if (i > 0 || n > 0)
+ newqr = get_bits(&gb, 1);
+ else
+ newqr = 1;
+ if (!newqr) {
+ if (i > 0)
+ get_bits(&gb, 1);
+ }
+ else {
+ int qi = 0;
+ skip_bits(&gb, av_log2(2)+1);
+ while (qi < 63) {
+ qi += get_bits(&gb, av_log2(63-qi)+1) + 1;
+ skip_bits(&gb, av_log2(2)+1);
+ }
+ if (qi > 63)
+ av_log(NULL, AV_LOG_ERROR, "error...\n");
+ }
+ }
+ }
+
+ for (s->hti = 0; s->hti < 80; s->hti++) {
+ s->entries = 0;
+ s->huff_code_size = 1;
+ if (!get_bits(&gb, 1)) {
+ s->hbits = 0;
+ read_huffman_tree(avctx, &gb);
+ s->hbits = 1;
+ read_huffman_tree(avctx, &gb);
+ }
+ }
+
+ s->theora_tables = 1;
+
+ return 0;
+}
+
+static int theora_decode_init(AVCodecContext *avctx)
+{
+ Vp3DecodeContext *s = avctx->priv_data;
+ GetBitContext gb;
+ int ptype;
+ uint8_t *p= avctx->extradata;
+ int op_bytes, i;
+
+ s->theora = 1;
+
+ if (!avctx->extradata_size)
+ return -1;
+
+ for(i=0;i<3;i++) {
+ op_bytes = *(p++)<<8;
+ op_bytes += *(p++);
+
+ init_get_bits(&gb, p, op_bytes);
+ p += op_bytes;
+
+ ptype = get_bits(&gb, 8);
+ debug_vp3("Theora headerpacket type: %x\n", ptype);
+
+ if (!(ptype & 0x80))
+ return -1;
+
+ skip_bits(&gb, 6*8); /* "theora" */
+
+ switch(ptype)
+ {
+ case 0x80:
+ theora_decode_header(avctx, gb);
+ break;
+ case 0x81:
+ theora_decode_comments(avctx, gb);
+ break;
+ case 0x82:
+ theora_decode_tables(avctx, gb);
+ break;
+ }
+ }
+
+ vp3_decode_init(avctx);
+ return 0;
+}
+
AVCodec vp3_decoder = {
"vp3",
CODEC_TYPE_VIDEO,
0,
NULL
};
+
+#ifndef CONFIG_LIBTHEORA
+AVCodec theora_decoder = {
+ "theora",
+ CODEC_TYPE_VIDEO,
+ CODEC_ID_THEORA,
+ sizeof(Vp3DecodeContext),
+ theora_decode_init,
+ NULL,
+ vp3_decode_end,
+ vp3_decode_frame,
+ 0,
+ NULL
+};
+#endif
projects / ffmpeg / blobdiff