x86/vp9lpf: add ff_vp9_loop_filter_[vh]_88_16_{ssse3,avx}.
authorClément Bœsch <u@pkh.me>
Sat, 25 Jan 2014 16:38:42 +0000 (17:38 +0100)
committerClément Bœsch <u@pkh.me>
Tue, 28 Jan 2014 06:36:38 +0000 (07:36 +0100)
commit222c46c531089dab3009a0e1e7938d51af2b494e
tree771cbe18b809db586b5080e1807ab8f955c62bf4
parent2a9c50798b798057f82f55bc7564356f66e64db5
x86/vp9lpf: add ff_vp9_loop_filter_[vh]_88_16_{ssse3,avx}.

9680 decicycles in loop_filter_v_88_16_c, 4193765 runs, 539 skips
9233 decicycles in loop_filter_h_88_16_c, 4193751 runs, 553 skips

1929 decicycles in ff_vp9_loop_filter_v_88_16_ssse3, 4194118 runs, 186 skips
2738 decicycles in ff_vp9_loop_filter_h_88_16_ssse3, 4193861 runs, 443 skips

5.978 → 5.417 overall decode time on ped1080p.webm (-threads 1)

Adding SSE2 support should be relatively trivial (just a matter of
changing the pshufb [mask_mix] with something else), patch welcome.
libavcodec/x86/vp9dsp_init.c
libavcodec/x86/vp9lpf.asm