Removing dependency to libgcc_s.so

David Daney ddaney at avtrex.com
Mon Jul 24 17:54:24 UTC 2006


Michael Hunold wrote:
> Hi Rich,
> 
> on 24.07.2006 14:59 Rich Felker said the following:
> 
>>On Mon, Jul 24, 2006 at 12:38:10PM +0200, Michael Hunold wrote:
>>
>>>Hi Rob,
>>>
>>>on 22.07.2006 20:57 Rob Landley said the following:
>>>
>>>>On Friday 21 July 2006 8:56 am, Michael Hunold wrote:
>>>>
>>>>>Is removing the dependency to libgcc a good thing to do?
>>>>
>>>>Yes.  What does your patch look like?
>>>
>>>Very simple. It uses a loop and substract 64bit division I found on the
>>>net. I attached the patch to this mail.
>>
>>Um, this works but incredibly slow. Why not optimize the case (that
>>occurs 100% of the time in practice) where the denominator is 32 bits
>>or less and you can just do 2 32bit divides and only use this horribly
>>slow code when the denominator is truely 64bit?
> 
> 
> By all means we should try to speed up the division, I agree.
> 
> But I fear that we definately need to answer the question first where to
> put these functions, if they can substitue libgcc.so and how to use them
> in busybox or if we should use them at all.
> 

If your mail system didn't bounce my mail as 'spam', you might have seen 
my message suggesting abandoning your libgcc replacement, and just 
linking with -static-libgcc.  I think in the short term that is likely 
to get the best results.

The long term fix, is of course to fix the linker.

David Daney



More information about the busybox mailing list