vc1dsp: introduce cases for 8x8 and 16x16
authorChristophe Gisquet <christophe.gisquet@gmail.com>
Fri, 11 Apr 2014 19:02:08 +0000 (21:02 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Sun, 20 Apr 2014 16:25:36 +0000 (18:25 +0200)
commit319235c67c59d6abaa78d5af57121ab9816f937d
tree4025d69a59c3e70b8cad858d486983b747671236
parentde9cd5884822375d492ff4dcc98e55317a66c196
vc1dsp: introduce cases for 8x8 and 16x16

This allows further unrolling the DSP implementation where possible.

x86 and ARM DSP modified by simply moving the multiple calls from vc1dec
to the DSP code. Decoding improvements should only occurs because of the
compiler actually able to unroll more.

Decoding time: ~8.80s -> 8.64s (ie around 2%)

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
libavcodec/arm/vc1dsp_init_neon.c
libavcodec/vc1dec.c
libavcodec/vc1dsp.c
libavcodec/vc1dsp.h
libavcodec/x86/vc1dsp_init.c
libavcodec/x86/vc1dsp_mmx.c