[FFmpeg-devel] [PATCH] h264: assembly version of get_cabac for x86_64 with PIC

[Roland Scheidegger]

Am 13.04.2012 19:34, schrieb Reimar Döffinger:
> On Fri, Apr 13, 2012 at 04:34:34PM +0200, Roland Scheidegger wrote:
>> There are similar functions which could get the same treatment but they
>> are less frequently used and since this isn't very nice as we can't use the
>> same assembly template focus on this function alone for now.
> 
> Huh? The only other ones seem to be decode_significance as far as I can
> tell, and the only reason seems to be that they use
> BRANCHLESS_GET_CABAC.
> Have you tried making them use your BRANCHLESS_GET_CABAC macro?
> Code is in libavcodec/x86/h264_i386.h.
Yes those are the ones I was talking about. decode_significance_8x8_x86
(but not decode_significance_x86) also uses MANGLE itself so needs a
change there. Other than that, the macro arguments are quite different
hence these need some more ifdefs to make it work.
If it's worth it I can give it a try - I just didn't see them in a
profile because they are part of decode_cabac_residual_internal (which
itself does show up quite prominently).

> since the macros are used multiple times the label might cause issues.
> So far that reason and
> for better readability it might be better to require the surrounding
> code to have such a label instead of having it in the macro?

I don't think it should cause problems, the code already had a label
before. Maybe if moving the label outside the lea should be moved out of
the macro too (and instead the reg holding the offset passed in) which
would save the decode_significance functions using BRANCHLESS_GET_CABAC
twice from performing this lea twice. I don't think that helps
readability, though...

Roland