[FFmpeg-devel] [PATCH]Simplify mlp_filter_channel_x86()

[Carl Eugen Hoyos]

Hendrik Leppkes <h.leppkes <at> gmail.com> writes:

> > Attached patch by Francisco Blas Izquierdo Riera 
> > fixes a compilation error in mlpdsp_init.c with 
> > -fstack-check and some gcc compilers (I reproduced 
> > the issue with gcc 4.7.3) by simplifying the code.
> > I ran some benchmarks with a very large thd file, 
> > 'time' confirms the results of '-benchmark'.
> 
> Please benchmark ASM changes using STAR/STOP_TIMER 
> macros.

It's 5400 without vs 5500 decicyles with the patch 
attached.

I (still) find it very hard to understand that 
testing a few bytes of input data has more 
meaning than running decoding several GBs 
several times.

> Looking at the code, the change introduces one 
> extra instruction in x64, and is identical in x86, 
> so I don't really see how it can be faster?

I am not claiming it is faster.
Just that a slowdown is not reproducible.

Carl Eugen