Front page | perl.perl5.porters |
Postings from January 2011
Re: Being more careful with FP on gcc:
Thread Previous
From:
Andy Dougherty
Date:
January 26, 2011 11:54
Subject:
Re: Being more careful with FP on gcc:
Message ID:
alpine.DEB.2.00.1101261411280.13081@fractal.phys.lafayette.edu
On Sat, 22 Jan 2011, Greg Lindahl wrote:
> On Sun, Jan 23, 2011 at 11:36:43AM +0530, Philip Potter wrote:
>
> > Investigating i386 further, it seems only a few i386 processors have
> > double-precision SSE instructions; the gcc docs suggest most intel and
> > all AMD i386 processors only support single-precision SSE, which
> > doesn't solve the problem completely. This pretty much knocks the
> > -mfparch=sse option out for most i386 situations.
>
> I would bet that most i386 machines today are x86-64 machines being
> run in 32-bit mode, which all do support SSE2.
That may be true in some global average sense, but it is most definitely
not the case for any of the i386 machines that I use or administer. They
are really 32 bit machines. Most, but certainly not all, of the machines
I rely on support SSE2.
> Still, distros do build
> very conservatively for 32-bits, the perl rpm in CentOS 5.5 is '.386'
> and not '.686', and I don't recall off the top of my head if 686 is
> actually using SSE2, but I think not. 386, definitely not.
It's certainly reasonable and useful to want easy ways to build perl
either conservatively (e.g. for broad distribution) or less conservatively
(e.g. when you know all the target machines will have certain
capabilities). Or, put another way, it should be easy to turn this sort
of thing either on or off.
Overall, I understand and am sympathetic towards the general idea of
wanting to control the x86 floating point precision. I'm still pondering
what I think the Configure user interface on this sort of thing ought to
look like, and what the defaults ought to be.
--
Andy Dougherty doughera@lafayette.edu
Thread Previous