Re: buildworld panic on ia64

On Jul 9, 2009, at 1:50 AM, Anton Shterenlikht wrote:
>> By all means: experiment. But be very careful not to make
>> the assumption that if the code compiles, it'll also run.
>> The weirder the set of compiler options, the more likely
>> you trip over optimization bugs and end up with an unstable
>> system. And I'm not even talking about whether the set
>> of options give you more optimal code in general.
>
> I see..
>
> Is there any advice for compiler options on ia64?

My advise at this time is to not change from the default.
I haven't done any kind of experimentation or know of any-
one else who did, to make any kind of claim as to the
effectiveness or harm of various compiler options.

I'm not talking cleanroom experiments here. I'm sure that
there have been plenty of people looking at SPECcpu and
who came up with a very creative set of compiler options
that make SPECcpu perform "optimally" (for each program).
This normally also includes fixing the compiler (and even
adding special case code) to have correct code generated
in that case.

I'm talking about a safe set of options that people can
use and that yields correct code 99.9% of the time and
gives acceptable (if not good) code. I cannot stress the
importance of having the toolchain generate correct code
when working on a FreeBSD port to a different architecture.

> I'm sorry if I'm asking obvious questions.
> Perhaps this is documented/disucced somewhere
> already? I'm new to ia64, most of my FBSD
> experience is from alpha and i386.

These aren't obvious questions. Compiler options, if they
are being discussed, are primarily discussed for i386 or
amd64.

-- 
Marcel Moolenaar
xcllnt_at_mac.com