[CFT] gcc: support for barcelona

[Steve Kargl]

On Wed, May 29, 2013 at 09:47:52AM +0100, David Chisnall wrote:
> On 29 May 2013, at 07:57, Andriy Gapon <avg at FreeBSD.org> wrote:
> 
> > In fact, I am of opinion that while such bugs exist gcc should be crowned back
> > as a default compiler.
> 
> Seriously?  Your show stopper bug is that, very occasionally, clang emits incorrect debug info?  And Steve's is that clang emits code that is fully compliant with the C standard, but gives more floating point precision than he wanted?
> 

Ah, no.  I just pointed to the last example of poor floating point
support from clang.  The other bug I mentioned has sat in the llvm
bug database for nearly 2 years.  But, to be blunt, clang cannot
do complex arithmetic correctly.

> If those are the most serious problems we have with clang, then
> it's time to remove gcc 4.2.1 from the tree right now.  I wish
> the problems that we had with it were so trivial...

The most serious problems IMNSHO are:

1) The dismal compile time of clang.  My buildworlds are 3 to 4
   times slower with clang than gcc.  I certainly understand the
   recent explosion in tinderbox messages.  No one wants to wait
   the better part of a day to do a make world and make universe
   must be painful.

2) Last time I checked, clang could not build a functioning X11
   environment.  Perhaps, having the xserver blow chunks when
   she's starting is consider to a Good Thing.  When I reported
   the problem, I was dutifully informed that the problem wasn't
   clang, it was the x11 loader.  Oddly, gcc does not have this
   issue.  May be this has been fix.  Don't know, don't care,
   because the last time I tried clang as the default compiler
   it took a week to recover a working environment. 

3) Last time I benchmarked a libm built by clang against a libm
   built by gcc.  gcc produced smaller, faster code and in one
   case a gcc compiled function produced a correct result while
   the clang compiled functions produced NaN for all inputs in
   a specific range.  Yeah, I reported that too.  I was told the
   test case was too complicated.

PS: That extra precision can be a big headache when one is trying
    to write high quality software.

-- 
Steve