[FFmpeg-devel] dvdata.h mess, are there still objections to fixing it

[Måns Rullgård]

Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:

> On Thu, Oct 15, 2009 at 07:21:36PM +0100, M?ns Rullg?rd wrote:
>> Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:
>> > I remember that dvdata.h was discussed in the past.
>> > There is a huge amount of data in that header, and it is included 3
>> > times.
>> > That means that the FFmpeg binary contains
>> > dv_idct_factor_hd1080, dv_idct_factor_hd720, work_chunks_dv100pali,
>> > work_chunks_dv100ntsci, dv_idct_factor_sd etc. 3 times, making about
>> > 233kB of .bss wasted for no good reason at all.
>> > Preferably someone should check which ones are actually, really,
>> > performance relevant and that compiler optimizes away and keep only
>> > those in the header, but in case nobody volunteers, are there any
>> > objections to moving dv_profiles and all the non-const stuff to a
>> > separate .c file (while adding ff_ prefix)?
>> 
>> Just do it.
>
> Yeah, after looking at its effects I started working on it.
> It only moves out the profiles, which I really expect not
> to make any performance difference.
> .rodata size reduced by 3264 bytes, .bss size reduced by
> 312416 bytes (on x86_64).
> While I can't imagine that I could have messed it up, I'd appreciate a
> quick verification of the Makefile part.

I can't read michael-diffs.

-- 
M?ns Rullg?rd
mans at mansr.com