[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Infinite loop in GCC?

On Fri, May 2, 2008 at 4:51 PM, Richard B. Kreckel <kreckel@ginac.de>
>  This is most certainly a bug in GCC on arm:

  Nothing is certain in this world :)

The problem is most likely that the arm build daemons are slow and
have little RAM: 128MB I think, whereas the source files in ginac use,
well, I saw one file using 133MB VRAM with 128MB RSS to compile; I
guess that one file needs just enough more to make them thrash.  You
should contact the arm buildd admin to ask for ginac to be scheduled
on the machine with the most RAM, and for the inactivity timeout for
it to be increased.

I just compiled ginac for arm on a 600MHz box with 512MB, and it ran
to completion in
For the function in question, the .lo compilation took 137MB of
virtual memory and 17 minutes, and the final one (that died before)
took 133 MB VM and 18 minutes.
I limited them to 120MB resident physical RAM, and they were fine - I
guess the netwinder only has 64MB, not quite enough to cope.

I noticed a "characters are signed" bug in the first source file,
add.cpp, filed as bug #479163 with patches to fix it. I don't know why
it wasn't flagged in the build log you indicate.

The other "comparison between signed and unsigned integer expressions"
warnings that I've checked are harmless, due to
   if (int i=0; i<limit; i++)
constructs that should be
   if (unsigned i=0; i<limit; i++)
   if (size_t i=0; i<limit; i++)
according to the type of the unsigned limit variable or function. It
would be good to check and fix them all upstream so that the warnings
that highlight real problems will stand out.

Reply to: