Re: IA64 inline division question

[Jim Wilson <wilson at specifixinc dot com>]

Steve Ellcey wrote:
> Currently I am looking at two issues that seem suspicious to me, one is
> that *recip_approx (and *sqrt_approx) are defined as the itanium_class
> fmisc.  I am wondering if this is correct or if they should be fmac
> instead.  Or perhaps they need their own type.  The other issue is with
> the attribute itanium_requires_unit0.  This treats fmac and fmisc
> differently but I don't see where it is used.  Is itanium_requires_unit0
> obsolete?

Wrong itanium_class attributes can't cause miscompilation, they can only 
cause unoptimal code to be generated.

I see that the itanium1 docs put frcpa/frsqrta in class fmisc, and the 
Itanium2 docs put them in class fmac, so this certainly could be 
improved for Itanium2.  Creating a separate class for them to solve this 
problem would be reasonable.

Yes, the itanium_requires_unit0 attribute is obsolete.  It was used in 
gcc-3.3, but not in current sources.  The code using it disappeared 
2003-01-09 when Vlad merged the itanium-sched-branch to mainline.

> I am going to try and create a small test case but in the mean time I
> thought I would see if anyone has any ideas on what the problem might be
> or had any suggestions on what else I should be looking at.

You can try playing with optimization options to narrow down the 
problem.  If it is a scheduler problem, then it should disappear if you 
use -fno-schedule-insns -fno-schedule-insns2.  Alternatively, try 
compiling with -O and then adding -fschedule-insns -fschedule-insns2 to 
see if that creates the problem.
--
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com