diff options
| author | James Almer <jamrial@gmail.com> | 2019-03-14 16:16:43 -0300 |
|---|---|---|
| committer | James Almer <jamrial@gmail.com> | 2019-03-14 16:16:43 -0300 |
| commit | c6892f59eb0e9f2a9ec1f55b21a5841a60540e1f (patch) | |
| tree | 28e193b07e9fb2c5c58f5b17f505437be5b1eab3 /libavcodec/aarch64 | |
| parent | 79025da3f2e7ab047c8f3c0c817952a98480b26b (diff) | |
| parent | cc7ba00c35faf0478f1f56215e926f70ccb31282 (diff) | |
| download | ffmpeg-streaming-c6892f59eb0e9f2a9ec1f55b21a5841a60540e1f.zip ffmpeg-streaming-c6892f59eb0e9f2a9ec1f55b21a5841a60540e1f.tar.gz | |
Merge commit 'cc7ba00c35faf0478f1f56215e926f70ccb31282'
* commit 'cc7ba00c35faf0478f1f56215e926f70ccb31282':
aarch64: vp8: Port missing epel8 functions from arm version
Merged-by: James Almer <jamrial@gmail.com>
Diffstat (limited to 'libavcodec/aarch64')
| -rw-r--r-- | libavcodec/aarch64/vp8dsp_init_aarch64.c | 4 | ||||
| -rw-r--r-- | libavcodec/aarch64/vp8dsp_neon.S | 87 |
2 files changed, 91 insertions, 0 deletions
diff --git a/libavcodec/aarch64/vp8dsp_init_aarch64.c b/libavcodec/aarch64/vp8dsp_init_aarch64.c index cd74201..d41dee0 100644 --- a/libavcodec/aarch64/vp8dsp_init_aarch64.c +++ b/libavcodec/aarch64/vp8dsp_init_aarch64.c @@ -46,8 +46,12 @@ av_cold void ff_vp78dsp_init_aarch64(VP8DSPContext *dsp) dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_neon; dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon; + dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_neon; + dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_neon; + dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_neon; dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_neon; dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_neon; + dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_neon; dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_neon; dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_neon; } diff --git a/libavcodec/aarch64/vp8dsp_neon.S b/libavcodec/aarch64/vp8dsp_neon.S index 6a727e7..1b79db6 100644 --- a/libavcodec/aarch64/vp8dsp_neon.S +++ b/libavcodec/aarch64/vp8dsp_neon.S @@ -957,6 +957,51 @@ function ff_put_vp8_epel16_h6v6_neon, export=1 ret endfunc +function ff_put_vp8_epel8_v6_neon, export=1 + sub x2, x2, x3, lsl #1 + + movrel x7, subpel_filters, -16 + add x6, x7, w6, uxtw #4 + ld1 {v0.8h}, [x6] +1: + ld1 {v2.8b}, [x2], x3 + ld1 {v3.8b}, [x2], x3 + ld1 {v4.8b}, [x2], x3 + ld1 {v5.8b}, [x2], x3 + ld1 {v6.8b}, [x2], x3 + ld1 {v7.8b}, [x2], x3 + ld1 {v28.8b}, [x2] + + sub x2, x2, x3, lsl #2 + + vp8_epel8_v6_y2 v2, v3, v2, v3, v4, v5, v6, v7, v28 + + st1 {v2.8b}, [x0], x1 + st1 {v3.8b}, [x0], x1 + subs w4, w4, #2 + b.ne 1b + + ret +endfunc + +function ff_put_vp8_epel8_h6_neon, export=1 + sub x2, x2, #2 + + movrel x7, subpel_filters, -16 + add x5, x7, w5, uxtw #4 + ld1 {v0.8h}, [x5] +1: + ld1 {v2.8b, v3.8b}, [x2], x3 + + vp8_epel8_h6 v2, v2, v3 + + st1 {v2.8b}, [x0], x1 + subs w4, w4, #1 + b.ne 1b + + ret +endfunc + function ff_put_vp8_epel8_h6v6_neon, export=1 sub x2, x2, x3, lsl #1 sub x2, x2, #2 @@ -1003,6 +1048,48 @@ function ff_put_vp8_epel8_h6v6_neon, export=1 ret endfunc +function ff_put_vp8_epel8_v4_neon, export=1 + sub x2, x2, x3 + + movrel x7, subpel_filters, -16 + add x6, x7, w6, uxtw #4 + ld1 {v0.8h}, [x6] +1: + ld1 {v2.8b}, [x2], x3 + ld1 {v3.8b}, [x2], x3 + ld1 {v4.8b}, [x2], x3 + ld1 {v5.8b}, [x2], x3 + ld1 {v6.8b}, [x2] + sub x2, x2, x3, lsl #1 + + vp8_epel8_v4_y2 v2, v2, v3, v4, v5, v6 + + st1 {v2.d}[0], [x0], x1 + st1 {v2.d}[1], [x0], x1 + subs w4, w4, #2 + b.ne 1b + + ret +endfunc + +function ff_put_vp8_epel8_h4_neon, export=1 + sub x2, x2, #1 + + movrel x7, subpel_filters, -16 + add x5, x7, w5, uxtw #4 + ld1 {v0.8h}, [x5] +1: + ld1 {v2.8b,v3.8b}, [x2], x3 + + vp8_epel8_h4 v2, v2, v3 + + st1 {v2.8b}, [x0], x1 + subs w4, w4, #1 + b.ne 1b + + ret +endfunc + function ff_put_vp8_epel8_h4v6_neon, export=1 sub x2, x2, x3, lsl #1 sub x2, x2, #1 |
