Re: RFA: MIPS paired single support

[James E Wilson <wilson at specifixinc dot com>]

On Mon, 2004-08-30 at 23:03, Richard Sandiford wrote:
> Well, the reason I said there was some disagreement that was that ISTR
> Jim suggesting a new target-specific subdirectory.  I agree that might
> be better.

I think doing this right would mean adding target/$target
subdirectories, and writing .exp code which would automatically pick the
right directory for the current target.  No one has written such code,
and I don't have immediate plans to do so.  I don't think we should
require this before adding the new testcases.

rth said that Red Hat might already have a scheme for this, but he
wasn't sure.  So there was an unanswered question about whether Red Hat
might want to contribute something here.

As a half measure, we could just add a mips subdirectory, populate it
with the new testcases, and optionally move existing mips specific
testcases into it.  We can create the new mips.exp file by copying an
existing one from a sibling directory, such as the weak/weak.exp file. 
An advantage of this is that one can do RUNTESTFLAGS=mips.exp to run
just the mips testcases.  Otherwise, we get no apparent benefit from
this change.  A disadvantage is that this may confuse automatic
testers.  Doing this would require defining which testcases get moved
into the new directory.  Perhaps just all of them that contain mips and
only mips in a dg-* string?

The simplest solution is to just follow existing practice, and add the
testcases to the gcc.dg directory along with all of the others.  This is
what I would prefer as I am lazy.

I have made the suggested changes to my copies, using mipsisa64* and
adding __FAST_MATH__ tests.
-- 
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com