On Wed, 7 Jan 2026, Rainer Orth wrote:

> Hi Richard,
> 
> > On Wed, 7 Jan 2026, Rainer Orth wrote:
> >
> >> Hi Xi,
> >> 
> >> > On Tue, 2026-01-06 at 09:54 +0100, Rainer Orth wrote:
> >> >> The gcc.dg/tree-ssa/forwprop-43.c test currently FAILs on 32 and 64-bit
> >> >> SPARC:
> >> >> 
> >> >> FAIL: gcc.dg/tree-ssa/forwprop-43.c scan-tree-dump-times forwprop1 
> >> >> "VEC_PERM_EXPR" 10
> >> >> 
> >> >> The dump has no reference to VEC_PERM_EXPR, so the scan needs to guarded
> >> >> by vect_perm.
> >> >
> >> > It will need to be moved to vect/ directory then.  See
> >> > https://gcc.gnu.org/PR113418.
> >> 
> >> unfortunately this is not enough: when doing so the test wasn't run at
> >> all.  This happens because vect.exp only selects part of the files in
> >> gcc.dg/vect based on patterns.  Apart from forwprop-43.c, four more fall
> >> through the cracks:
> >> 
> >> gcc.dg/vect/gimplefe-40.c
> >> gcc.dg/vect/gimplefe-41.c
> >> gcc.dg/vect/group-no-gaps-1.c
> >> gcc.dg/vect/if-cvt-stores-vect-ifcvt-18.c
> >> 
> >> I wonder how best to address this: one could rename them to match one of
> >> the existing patterns, probably just vect-*.c.  However, future files
> >> could well be missed again.
> >> 
> >> Comments?
> >
> > I had started to transition the various pattern dependent extra
> > options to dg-addtional-options but didn't manage to finish.  So
> > that's the solution, and then simply glob all *.c files.
> 
> that's certainly the cleanest solution.  Though largely mechanical, I'm
> not sure we want to do this before the GCC 16 release, though.

IMO it's perfect stage4 work.

> > But IMO some effective targets should maybe made to work outside
> > of vect.exp?  I think not all really require it, only some, and it's
> > not well documented which?
> 
> That would most certainly be welcome.
> 
> Besides, there's quite a number of effective-target keywords not even
> documented in sourcebuild.texi.  I mean to add those that have
> documentation in target-support.exp, create a PR for the rest to prompt
> the authors to fix it, and add a test like gcc.src/maintainers.exp to
> ensure things remain that way.  No idea when I'll get to that, though.

Yeah.  Possibly also get an idea on the number of uses of the various
effective targets - I'd expect a majority to be only used very few
times, thus easy candidates for renaming.

The issue with outside-of-vect.exp usage is the use of
check_cached_effective_target[_indexed]?  I do wonder why we include
target-supports.exp from testsuites where it isn't supported?  It's
name also does not suggest it's not suitable in general :/

Richard.

>       Rainer
> 
> 

-- 
Richard Biener <[email protected]>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Jochen Jaser, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)

Reply via email to