diff options
| author | Christophe GISQUET <christophe.gisquet@gmail.com> | 2012-03-19 22:46:28 +0100 |
|---|---|---|
| committer | Ronald S. Bultje <rsbultje@gmail.com> | 2012-04-10 10:06:48 -0700 |
| commit | 272b252c0110225188c7d7f31167941210aac197 (patch) | |
| tree | 47bea5996c88057a418e8872a655bac8f261736e /libavcodec/rv34.c | |
| parent | d3c59d5003a483f1a23e225fc71c19bd1116d11c (diff) | |
rv40dsp: implement prescaled versions for biweight.
Quite often, the original weights are multiple of 512. By prescaling them
by 1/512 when they are computed (once per frame), no intermediate shifting
is needed, and no prescaling on each call either.
The x86 code already used that trick.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
Diffstat (limited to 'libavcodec/rv34.c')
| -rw-r--r-- | libavcodec/rv34.c | 58 |
1 files changed, 34 insertions, 24 deletions
diff --git a/libavcodec/rv34.c b/libavcodec/rv34.c index 3ad1717d13..12475692c6 100644 --- a/libavcodec/rv34.c +++ b/libavcodec/rv34.c @@ -521,7 +521,7 @@ static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int */ static int calc_add_mv(RV34DecContext *r, int dir, int val) { - int mul = dir ? -r->weight2 : r->weight1; + int mul = dir ? -r->mv_weight2 : r->mv_weight1; return (val * mul + 0x2000) >> 14; } @@ -776,24 +776,24 @@ static void rv34_mc_1mv(RV34DecContext *r, const int block_type, static void rv4_weight(RV34DecContext *r) { - r->rdsp.rv40_weight_pixels_tab[0](r->s.dest[0], - r->tmp_b_block_y[0], - r->tmp_b_block_y[1], - r->weight1, - r->weight2, - r->s.linesize); - r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[1], - r->tmp_b_block_uv[0], - r->tmp_b_block_uv[2], - r->weight1, - r->weight2, - r->s.uvlinesize); - r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[2], - r->tmp_b_block_uv[1], - r->tmp_b_block_uv[3], - r->weight1, - r->weight2, - r->s.uvlinesize); + r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0], + r->tmp_b_block_y[0], + r->tmp_b_block_y[1], + r->weight1, + r->weight2, + r->s.linesize); + r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1], + r->tmp_b_block_uv[0], + r->tmp_b_block_uv[2], + r->weight1, + r->weight2, + r->s.uvlinesize); + r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2], + r->tmp_b_block_uv[1], + r->tmp_b_block_uv[3], + r->weight1, + r->weight2, + r->s.uvlinesize); } static void rv34_mc_2mv(RV34DecContext *r, const int block_type) @@ -1703,11 +1703,21 @@ int ff_rv34_decode_frame(AVCodecContext *avctx, int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts); int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts); - if (!refdist) { - r->weight1 = r->weight2 = 8192; - } else { - r->weight1 = (dist0 << 14) / refdist; - r->weight2 = (dist1 << 14) / refdist; + if(!refdist){ + r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192; + r->scaled_weight = 0; + }else{ + r->mv_weight1 = (dist0 << 14) / refdist; + r->mv_weight2 = (dist1 << 14) / refdist; + if((r->mv_weight1|r->mv_weight2) & 511){ + r->weight1 = r->mv_weight1; + r->weight2 = r->mv_weight2; + r->scaled_weight = 0; + }else{ + r->weight1 = r->mv_weight1 >> 9; + r->weight2 = r->mv_weight2 >> 9; + r->scaled_weight = 1; + } } } s->mb_x = s->mb_y = 0; |