falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
Browse Source

Making idct factor tables dynamically allocated 

Originally committed as revision 15885 to svn://svn.ffmpeg.org/ffmpeg/trunk
tags/v0.5
Roman Shaposhnik 17 years ago
parent
eeee48b26b
commit
a21fd75f56
2 changed files with 48 additions and 35 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +34
    -35
      libavcodec/dv.c
  2. +14
    -0
      libavcodec/dvdata.h

+ 34
- 35
libavcodec/dv.c View File

@@ -54,8 +54,6 @@ typedef struct DVVideoContext {
uint8_t *buf;

uint8_t dv_zigzag[2][64];
uint32_t dv_idct_factor[2][2][22][64];
uint32_t dv100_idct_factor[4][4][16][64];

void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size);
void (*fdct[2])(DCTELEM *block);
@@ -208,6 +206,8 @@ static inline void dv_calc_mb_coordinates(const DVprofile *d, int chan, int seq,
static int dv_init_dynamic_tables(const DVprofile *d)
{
int j,i,c,s,p;
uint32_t *factor1, *factor2;
const int *iweight1, *iweight2;

if (!d->work_chunks[dv_work_pool_size(d)-1].buf_offset) {
p = i = 0;
@@ -226,42 +226,44 @@ static int dv_init_dynamic_tables(const DVprofile *d)
}
}
}
return 0;
}

static void dv_build_unquantize_tables(DVVideoContext *s, uint8_t* perm)
{
int i, q, a;

/* NOTE: max left shift is 6 */
for (q = 0; q < 22; q++) {
/* 88DCT */
i = 1;
for (a = 0; a < 4; a++) {
for (; i < dv_quant_areas[a]; i++) {
/* 88 table */
s->dv_idct_factor[0][0][q][i] = dv_iweight_88[i] << (dv_quant_shifts[q][a] + 1);
s->dv_idct_factor[1][0][q][i] = s->dv_idct_factor[0][0][q][i] << 1;

/* 248 table */
s->dv_idct_factor[0][1][q][i] = dv_iweight_248[i] << (dv_quant_shifts[q][a] + 1);
s->dv_idct_factor[1][1][q][i] = s->dv_idct_factor[0][1][q][i] << 1;
if (!d->idct_factor[DV_PROFILE_IS_HD(d)?8191:5631]) {
factor1 = &d->idct_factor[0];
factor2 = &d->idct_factor[DV_PROFILE_IS_HD(d)?4096:2816];
if (d->height == 720) {
iweight1 = &dv_iweight_720_y[0];
iweight2 = &dv_iweight_720_c[0];
} else {
iweight1 = &dv_iweight_1080_y[0];
iweight2 = &dv_iweight_1080_c[0];
}
if (DV_PROFILE_IS_HD(d)) {
for (c = 0; c < 4; c++) {
for (s = 0; s < 16; s++) {
for (i = 0; i < 64; i++) {
*factor1++ = (dv100_qstep[s] << (c + 9)) * iweight1[i];
*factor2++ = (dv100_qstep[s] << (c + 9)) * iweight2[i];
}
}
}
} else {
iweight1 = &dv_iweight_88[0];
for (j = 0; j < 2; j++, iweight1 = &dv_iweight_248[0]) {
for (s = 0; s < 22; s++) {
for (i = c = 0; c < 4; c++) {
for (; i < dv_quant_areas[c]; i++) {
*factor1 = iweight1[i] << (dv_quant_shifts[s][c] + 1);
*factor2++ = (*factor1++) << 1;
}
}

for (a = 0; a < 4; a++) {
for (q = 0; q < 16; q++) {
for (i = 1; i < 64; i++) {
s->dv100_idct_factor[0][a][q][i] = (dv100_qstep[q] << (a + 9)) * dv_iweight_1080_y[i];
s->dv100_idct_factor[1][a][q][i] = (dv100_qstep[q] << (a + 9)) * dv_iweight_1080_c[i];
s->dv100_idct_factor[2][a][q][i] = (dv100_qstep[q] << (a + 9)) * dv_iweight_720_y[i];
s->dv100_idct_factor[3][a][q][i] = (dv100_qstep[q] << (a + 9)) * dv_iweight_720_c[i];
}
}
}
}

return 0;
}

static av_cold int dvvideo_init(AVCodecContext *avctx)
{
DVVideoContext *s = avctx->priv_data;
@@ -385,9 +387,6 @@ static av_cold int dvvideo_init(AVCodecContext *avctx)
}else
memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64);

/* XXX: do it only for constant case */
dv_build_unquantize_tables(s, dsp.idct_permutation);

avctx->coded_frame = &s->picture;
s->avctx = avctx;

@@ -556,13 +555,13 @@ static int dv_decode_video_segment(AVCodecContext *avctx, DVwork_chunk *work_chu
if (DV_PROFILE_IS_HD(s->sys)) {
mb->idct_put = s->idct_put[0];
mb->scan_table = s->dv_zigzag[0];
mb->factor_table = s->dv100_idct_factor[((s->sys->height == 720) << 1) | (j >= 4)][class1][quant];
mb->factor_table = &s->sys->idct_factor[(j >= 4)*4*16*64 + class1*16*64 + quant*64];
is_field_mode[mb_index] |= !j && dct_mode;
} else {
mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
mb->scan_table = s->dv_zigzag[dct_mode];
mb->factor_table = s->dv_idct_factor[class1 == 3][dct_mode]
[quant + dv_quant_offset[class1]];
mb->factor_table = &s->sys->idct_factor[(class1 == 3)*2*22*64 + dct_mode*22*64 +
(quant + dv_quant_offset[class1])*64];
}
dc = dc << 2;
/* convert to unsigned because 128 is not added in the


+ 14
- 0
libavcodec/dvdata.h View File

@@ -53,6 +53,7 @@ typedef struct DVprofile {
int width; /* picture width in pixels */
AVRational sar[2]; /* sample aspect ratios for 4:3 and 16:9 */
DVwork_chunk *work_chunks; /* each thread gets its own chunk of frame to work on */
uint32_t *idct_factor; /* set of iDCT factor tables */
enum PixelFormat pix_fmt; /* picture pixel format */
int bpm; /* blocks per macroblock */
const uint8_t *block_sizes; /* AC block sizes, in bits */
@@ -469,6 +470,10 @@ static DVwork_chunk work_chunks_dv100ntscp[2*10*27];
static DVwork_chunk work_chunks_dv100pali [4*12*27];
static DVwork_chunk work_chunks_dv100ntsci[4*10*27];

static uint32_t dv_idct_factor_sd [2*2*22*64];
static uint32_t dv_idct_factor_hd1080[2*4*16*64];
static uint32_t dv_idct_factor_hd720 [2*4*16*64];

static const DVprofile dv_profiles[] = {
{ .dsf = 0,
.video_stype = 0x0,
@@ -481,6 +486,7 @@ static const DVprofile dv_profiles[] = {
.width = 720,
.sar = {{10, 11}, {40, 33}},
.work_chunks = &work_chunks_dv25ntsc[0],
.idct_factor = &dv_idct_factor_sd[0],
.pix_fmt = PIX_FMT_YUV411P,
.bpm = 6,
.block_sizes = block_sizes_dv2550,
@@ -500,6 +506,7 @@ static const DVprofile dv_profiles[] = {
.width = 720,
.sar = {{59, 54}, {118, 81}},
.work_chunks = &work_chunks_dv25pal[0],
.idct_factor = &dv_idct_factor_sd[0],
.pix_fmt = PIX_FMT_YUV420P,
.bpm = 6,
.block_sizes = block_sizes_dv2550,
@@ -519,6 +526,7 @@ static const DVprofile dv_profiles[] = {
.width = 720,
.sar = {{59, 54}, {118, 81}},
.work_chunks = &work_chunks_dv25pal411[0],
.idct_factor = &dv_idct_factor_sd[0],
.pix_fmt = PIX_FMT_YUV411P,
.bpm = 6,
.block_sizes = block_sizes_dv2550,
@@ -538,6 +546,7 @@ static const DVprofile dv_profiles[] = {
.width = 720,
.sar = {{10, 11}, {40, 33}},
.work_chunks = &work_chunks_dv50ntsc[0],
.idct_factor = &dv_idct_factor_sd[0],
.pix_fmt = PIX_FMT_YUV422P,
.bpm = 6,
.block_sizes = block_sizes_dv2550,
@@ -557,6 +566,7 @@ static const DVprofile dv_profiles[] = {
.width = 720,
.sar = {{59, 54}, {118, 81}},
.work_chunks = &work_chunks_dv50pal[0],
.idct_factor = &dv_idct_factor_sd[0],
.pix_fmt = PIX_FMT_YUV422P,
.bpm = 6,
.block_sizes = block_sizes_dv2550,
@@ -576,6 +586,7 @@ static const DVprofile dv_profiles[] = {
.width = 1280,
.sar = {{1, 1}, {1, 1}},
.work_chunks = &work_chunks_dv100ntsci[0],
.idct_factor = &dv_idct_factor_hd1080[0],
.pix_fmt = PIX_FMT_YUV422P,
.bpm = 8,
.block_sizes = block_sizes_dv100,
@@ -595,6 +606,7 @@ static const DVprofile dv_profiles[] = {
.width = 1440,
.sar = {{1, 1}, {1, 1}},
.work_chunks = &work_chunks_dv100pali[0],
.idct_factor = &dv_idct_factor_hd1080[0],
.pix_fmt = PIX_FMT_YUV422P,
.bpm = 8,
.block_sizes = block_sizes_dv100,
@@ -614,6 +626,7 @@ static const DVprofile dv_profiles[] = {
.width = 960,
.sar = {{1, 1}, {1, 1}},
.work_chunks = &work_chunks_dv100ntscp[0],
.idct_factor = &dv_idct_factor_hd720[0],
.pix_fmt = PIX_FMT_YUV422P,
.bpm = 8,
.block_sizes = block_sizes_dv100,
@@ -633,6 +646,7 @@ static const DVprofile dv_profiles[] = {
.width = 960,
.sar = {{1, 1}, {1, 1}},
.work_chunks = &work_chunks_dv100palp[0],
.idct_factor = &dv_idct_factor_hd720[0],
.pix_fmt = PIX_FMT_YUV422P,
.bpm = 8,
.block_sizes = block_sizes_dv100,


Write Preview
Loading…
Cancel
Save