diff options
| author | Christophe Gisquet <christophe.gisquet@gmail.com> | 2015-02-07 18:49:36 +0000 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2015-02-08 22:19:34 +0100 |
| commit | 4919b384214ed91c1d81b6a0541f84f0b08a76b2 (patch) | |
| tree | e921559deb08f850dc695212f7c0b1e26577c186 | |
| parent | 1df841f154f80264c803630eef051b814fd7cce0 (diff) | |
| download | ffmpeg-4919b384214ed91c1d81b6a0541f84f0b08a76b2.tar.gz | |
x86: hevc_mc: fewer gpr autoloads for _v filters
In that case, it's just to load my, but mx/r3src is not used.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
| -rw-r--r-- | libavcodec/x86/hevc_mc.asm | 18 |
1 files changed, 12 insertions, 6 deletions
diff --git a/libavcodec/x86/hevc_mc.asm b/libavcodec/x86/hevc_mc.asm index d830a2dfdd..8867738b9f 100644 --- a/libavcodec/x86/hevc_mc.asm +++ b/libavcodec/x86/hevc_mc.asm @@ -789,7 +789,8 @@ cglobal hevc_put_hevc_bi_epel_h%1_%2, 7, 8, 11, dst, dststride, src, srcstride, ; int height, int mx, int my, int width) ; ****************************** -cglobal hevc_put_hevc_epel_v%1_%2, 6, 7, 11, dst, src, srcstride, height, r3src, my, rfilter +cglobal hevc_put_hevc_epel_v%1_%2, 4, 7, 11, dst, src, srcstride, height, r3src, my, rfilter + movifnidn myd, mym lea r3srcq, [srcstrideq*3] sub srcq, srcstrideq EPEL_FILTER %2, my, m4, m5 @@ -800,7 +801,8 @@ cglobal hevc_put_hevc_epel_v%1_%2, 6, 7, 11, dst, src, srcstride, height, r3src, LOOP_END dst, src, srcstride RET -cglobal hevc_put_hevc_uni_epel_v%1_%2, 7, 8, 11, dst, dststride, src, srcstride, height, r3src, my, rfilter +cglobal hevc_put_hevc_uni_epel_v%1_%2, 5, 8, 11, dst, dststride, src, srcstride, height, r3src, my, rfilter + movifnidn myd, mym lea r3srcq, [srcstrideq*3] movdqa m6, [pw_%2] sub srcq, srcstrideq @@ -817,7 +819,8 @@ cglobal hevc_put_hevc_uni_epel_v%1_%2, 7, 8, 11, dst, dststride, src, srcstride, RET -cglobal hevc_put_hevc_bi_epel_v%1_%2, 8, 9, 11, dst, dststride, src, srcstride, src2, height, r3src, my, rfilter +cglobal hevc_put_hevc_bi_epel_v%1_%2, 6, 9, 11, dst, dststride, src, srcstride, src2, height, r3src, my, rfilter + movifnidn myd, mym lea r3srcq, [srcstrideq*3] movdqa m6, [pw_bi_%2] sub srcq, srcstrideq @@ -1112,7 +1115,8 @@ cglobal hevc_put_hevc_bi_qpel_h%1_%2, 7, 8, 16 , dst, dststride, src, srcstride, ; int height, int mx, int my, int width) ; ****************************** -cglobal hevc_put_hevc_qpel_v%1_%2, 6, 8, 16, dst, src, srcstride, height, r3src, my, rfilter +cglobal hevc_put_hevc_qpel_v%1_%2, 4, 8, 16, dst, src, srcstride, height, r3src, my, rfilter + movifnidn myd, mym lea r3srcq, [srcstrideq*3] QPEL_FILTER %2, my .loop @@ -1125,7 +1129,8 @@ cglobal hevc_put_hevc_qpel_v%1_%2, 6, 8, 16, dst, src, srcstride, height, r3src, LOOP_END dst, src, srcstride RET -cglobal hevc_put_hevc_uni_qpel_v%1_%2, 7, 9, 16, dst, dststride, src, srcstride, height, r3src, my, rfilter +cglobal hevc_put_hevc_uni_qpel_v%1_%2, 5, 9, 16, dst, dststride, src, srcstride, height, r3src, my, rfilter + movifnidn myd, mym movdqa m9, [pw_%2] lea r3srcq, [srcstrideq*3] QPEL_FILTER %2, my @@ -1143,7 +1148,8 @@ cglobal hevc_put_hevc_uni_qpel_v%1_%2, 7, 9, 16, dst, dststride, src, srcstride, jnz .loop ; height loop RET -cglobal hevc_put_hevc_bi_qpel_v%1_%2, 8, 10, 16, dst, dststride, src, srcstride, src2, height, r3src, my, rfilter +cglobal hevc_put_hevc_bi_qpel_v%1_%2, 6, 10, 16, dst, dststride, src, srcstride, src2, height, r3src, my, rfilter + movifnidn myd, mym movdqa m9, [pw_bi_%2] lea r3srcq, [srcstrideq*3] QPEL_FILTER %2, my |