generic-library/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

jpeg2000: Optimize output sample conversion

Browse Source 
67935 -> 29984 kcycles

Reviewed-by: Nicolas BERTRAND <nicoinattendu@gmail.com>

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
This commit is contained in:
Michael Niedermayer 2013-07-01 10:01:37 +02:00 committed by Luca Barbato
parent f0552e63a6
commit 22e18ea39e
1 changed files with 44 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

69
libavcodec/jpeg2000dec.c
View File

@ -1162,6 +1162,9 @@ static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile,
Jpeg2000Component *comp = tile->comp + compno; Jpeg2000Component *comp = tile->comp + compno;
float *datap = comp->f_data; float *datap = comp->f_data;
int32_t *i_datap = comp->i_data; int32_t *i_datap = comp->i_data;
int cbps = s->cbps[compno];
int w = tile->comp[compno].coord[0][1] - s->image_offset_x;
y = tile->comp[compno].coord[1][0] - s->image_offset_y; y = tile->comp[compno].coord[1][0] - s->image_offset_y;
line = picture->data[0] + y * picture->linesize[0]; line = picture->data[0] + y * picture->linesize[0];
for (; y < tile->comp[compno].coord[1][1] - s->image_offset_y; y += s->cdy[compno]) { for (; y < tile->comp[compno].coord[1][1] - s->image_offset_y; y += s->cdy[compno]) {
@ -1170,18 +1173,24 @@ static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile,
x = tile->comp[compno].coord[0][0] - s->image_offset_x; x = tile->comp[compno].coord[0][0] - s->image_offset_x;
dst = line + x * s->ncomponents + compno; dst = line + x * s->ncomponents + compno;
for (; x < tile->comp[compno].coord[0][1] - s->image_offset_x; x += s->cdx[compno]) { if (tile->codsty->transform == FF_DWT97) {
int val; for (; x < w; x += s->cdx[compno]) {
/* DC level shift and clip see ISO 15444-1:2002 G.1.2 */ int val = lrintf(*datap) + (1 << (cbps - 1));
if (tile->codsty->transform == FF_DWT97) /* DC level shift and clip see ISO 15444-1:2002 G.1.2 */
val = lrintf(*datap) + (1 << (s->cbps[compno] - 1)); val = av_clip(val, 0, (1 << cbps) - 1);
else *dst = val << (8 - cbps);
val = *i_datap + (1 << (s->cbps[compno] - 1)); datap++;
val = av_clip(val, 0, (1 << s->cbps[compno]) - 1); dst += s->ncomponents;
*dst = val << (8 - s->cbps[compno]); }
datap++; } else {
i_datap++; for (; x < w; x += s->cdx[compno]) {
dst += s->ncomponents; int val = *i_datap + (1 << (cbps - 1));
/* DC level shift and clip see ISO 15444-1:2002 G.1.2 */
val = av_clip(val, 0, (1 << cbps) - 1);
*dst = val << (8 - cbps);
i_datap++;
dst += s->ncomponents;
}
} }
line += picture->linesize[0]; line += picture->linesize[0];
} }
@ -1192,6 +1201,8 @@ static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile,
float *datap = comp->f_data; float *datap = comp->f_data;
int32_t *i_datap = comp->i_data; int32_t *i_datap = comp->i_data;
uint16_t *linel; uint16_t *linel;
int cbps = s->cbps[compno];
int w = tile->comp[compno].coord[0][1] - s->image_offset_x;
y = tile->comp[compno].coord[1][0] - s->image_offset_y; y = tile->comp[compno].coord[1][0] - s->image_offset_y;
linel = (uint16_t *)picture->data[0] + y * (picture->linesize[0] >> 1); linel = (uint16_t *)picture->data[0] + y * (picture->linesize[0] >> 1);
@ -1199,24 +1210,32 @@ static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile,
uint16_t *dst; uint16_t *dst;
x = tile->comp[compno].coord[0][0] - s->image_offset_x; x = tile->comp[compno].coord[0][0] - s->image_offset_x;
dst = linel + (x * s->ncomponents + compno); dst = linel + (x * s->ncomponents + compno);
for (; x < tile->comp[compno].coord[0][1] - s->image_offset_x; x += s-> cdx[compno]) { if (tile->codsty->transform == FF_DWT97) {
int val; for (; x < w; x += s-> cdx[compno]) {
/* DC level shift and clip see ISO 15444-1:2002 G.1.2 */ int val = lrintf(*datap) + (1 << (cbps - 1));
if (tile->codsty->transform == FF_DWT97) /* DC level shift and clip see ISO 15444-1:2002 G.1.2 */
val = lrintf(*datap) + (1 << (s->cbps[compno] - 1)); val = av_clip(val, 0, (1 << cbps) - 1);
else /* align 12 bit values in little-endian mode */
val = *i_datap + (1 << (s->cbps[compno] - 1)); *dst = val << (16 - cbps);
val = av_clip(val, 0, (1 << s->cbps[compno]) - 1); datap++;
/* align 12 bit values in little-endian mode */ dst += s->ncomponents;
*dst = val << (16 - s->cbps[compno]); }
datap++; } else {
i_datap++; for (; x < w; x += s-> cdx[compno]) {
dst += s->ncomponents; int val = *i_datap + (1 << (cbps - 1));
/* DC level shift and clip see ISO 15444-1:2002 G.1.2 */
val = av_clip(val, 0, (1 << cbps) - 1);
/* align 12 bit values in little-endian mode */
*dst = val << (16 - cbps);
i_datap++;
dst += s->ncomponents;
}
} }
linel += picture->linesize[0] >> 1; linel += picture->linesize[0] >> 1;
} }
} }
} }
return 0; return 0;
} }