RepoMirrors/ffmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2025-04-23 07:26:43 +00:00
Code Issues Projects Releases Wiki Activity

gmc_mmx: enable also for large pictures when emu edge isnt needed.

Browse Source 
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Michael Niedermayer 2012-10-06 23:12:57 +02:00
parent e063ffbf4b
commit 19a4e10360
1 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
libavcodec/x86/dsputil_mmx.c
View File

@ -1961,12 +1961,15 @@ static av_always_inline void gmc(uint8_t *dst, uint8_t *src,
const int dyh = (dyy - (1 << (16 + shift))) * (h - 1); const int dyh = (dyy - (1 << (16 + shift))) * (h - 1);
const int dxh = dxy * (h - 1); const int dxh = dxy * (h - 1);
const int dyw = dyx * (w - 1); const int dyw = dyx * (w - 1);
int need_emu = (unsigned)ix >= width - w ||
(unsigned)iy >= height - h;
if ( // non-constant fullpel offset (3% of blocks) if ( // non-constant fullpel offset (3% of blocks)
((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) | ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) |
(oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift)
// uses more than 16 bits of subpel mv (only at huge resolution) // uses more than 16 bits of subpel mv (only at huge resolution)
|| (dxx | dxy | dyx | dyy) & 15 || (dxx | dxy | dyx | dyy) & 15
|| h > MAX_H || stride > MAX_STRIDE) { || (need_emu && (h > MAX_H || stride > MAX_STRIDE))) {
// FIXME could still use mmx for some of the rows // FIXME could still use mmx for some of the rows
ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy,
shift, r, width, height); shift, r, width, height);
@ -1974,8 +1977,7 @@ static av_always_inline void gmc(uint8_t *dst, uint8_t *src,
} }
src += ix + iy * stride; src += ix + iy * stride;
if ((unsigned)ix >= width - w || if (need_emu) {
(unsigned)iy >= height - h) {
emu_edge_fn(edge_buf, src, stride, w + 1, h + 1, ix, iy, width, height); emu_edge_fn(edge_buf, src, stride, w + 1, h + 1, ix, iy, width, height);
src = edge_buf; src = edge_buf;
} }