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

Re: K6 Optimized Debian Binaries



On Wed, Feb 16, 2000 at 12:22:04PM -0800, Sean 'Shaleh' Perry generated a stream of 1s and 0s:
> 
> On 16-Feb-2000 dan@cg619985-a.adubn1.nj.home.com wrote:
> > I have compiled some of recent 'frozen' sources optimized for K6 with PGCC.
> > Anyone is welcome to get 'em. FTP to cg619985-a.adubn.nj.home.com or 
> > 24.11.49.110. I would not use them on important or production machines
> > since some stuff might break with compiler options like -malign-double.
> > Use at your own risk, though have been stable for me. Enlightenment seems
> > speedier, and so should be others.
> > 
> 
> Just an FYI -- many groups will ask you to recompile a binary before they will
> help you with bugs if they discover the item was compiled with pgcc.  pgcc has
> been known to improperly compile code or break assumptions some code makes. 
> SuSE for example will not touch things compiled with pgcc, or so I am told by
> several of their key programmers.
> 
> 
> -- 

Hence the warning. Though I must say 2.95.3 has been very nice so far
compared to previous releases. Of course gcc 2.95.2, or 2.7.2.3 is
safer. I am only using these binaries on my home desktop. An analogy
might be using an overclocked machine for a mission critical operation
-- not good. Though there are whole distributions completely compiled
with PGCC, Mandrake is a good example. BTW, even if you use GCC 2.95.2,
optimizing for your CPU makes a difference, sometimes very noticeable compared to just -O2 -g. Unfortunately, AFAIK PGCC is a one man work ATM, so the project is kind of slow. For example PGCC does worse floating point code. We need a nice CPU optimizing compiler. Many people don't realize just how much more can be squeezed out of code if a compiler is good enough. Why lose performance when you don't have to. I really see no point in compiling all software in a distro with -O2 -g if that particular software is known to be stable. Most people have Pentiums and better, so something like -fomit-frame-pointer -mcpu=i586 -O3 could be useful. An exception would be a case where your code has to be absolutely debugable.
-- 
Get the truth or risk frying your brains! --> www.truthinlabeling.org <--


Reply to: