提交 010f98f9 编写于 作者: J Jason Garrett-Glaser

VC-1/WMV3 decoding: don't clear blocks unless they're actually used.

~8% faster VC-1 decoding.
Possible future optimization: clear blocks after use instead of before, and for
DC-only blocks, only clear the DC coefficient.

Originally committed as revision 19205 to svn://svn.ffmpeg.org/ffmpeg/trunk
上级 4f717c69
......@@ -1800,6 +1800,8 @@ static int vc1_decode_intra_block(VC1Context *v, DCTELEM block[64], int n, int c
int scale;
int q1, q2 = 0;
s->dsp.clear_block(block);
/* XXX: Guard against dumb values of mquant */
mquant = (mquant < 1) ? 0 : ( (mquant>31) ? 31 : mquant );
......@@ -1989,6 +1991,8 @@ static int vc1_decode_p_block(VC1Context *v, DCTELEM block[64], int n, int mquan
int ttblk = ttmb & 7;
int pat = 0;
s->dsp.clear_block(block);
if(ttmb == -1) {
ttblk = ff_vc1_ttblk_to_tt[v->tt_index][get_vlc2(gb, ff_vc1_ttblk_vlc[v->tt_index].table, VC1_TTBLK_VLC_BITS, 1)];
}
......@@ -2166,8 +2170,6 @@ static int vc1_decode_p_mb(VC1Context *v)
else
skipped = v->s.mbskip_table[mb_pos];
s->dsp.clear_blocks(s->block[0]);
apply_loop_filter = s->loop_filter && !(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY);
if (!fourmv) /* 1MV mode */
{
......@@ -2459,7 +2461,6 @@ static void vc1_decode_b_mb(VC1Context *v)
else
skipped = v->s.mbskip_table[mb_pos];
s->dsp.clear_blocks(s->block[0]);
dmv_x[0] = dmv_x[1] = dmv_y[0] = dmv_y[1] = 0;
for(i = 0; i < 6; i++) {
v->mb_type[0][s->block_index[i]] = 0;
......@@ -2851,7 +2852,6 @@ static void vc1_decode_p_blocks(VC1Context *v)
for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
ff_init_block_index(s);
ff_update_block_index(s);
s->dsp.clear_blocks(s->block[0]);
vc1_decode_p_mb(v);
if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
......@@ -2901,7 +2901,6 @@ static void vc1_decode_b_blocks(VC1Context *v)
for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
ff_init_block_index(s);
ff_update_block_index(s);
s->dsp.clear_blocks(s->block[0]);
vc1_decode_b_mb(v);
if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册