Re: [patch RFC] Tweak gcc.c-torture/execute/nest-align-1.c

[David Daney <ddaney at avtrex dot com>]

Janis Johnson wrote:
> On Sat, 2008-02-23 at 09:23 +0900, Kaz Kojima wrote:
> > Hi,
> >
> > gcc.c-torture/execute/nest-align-1.c which is recently added
> > fails on SH.  From testresults posting, it fails also on ARM
> >   http://gcc.gnu.org/ml/gcc-testresults/2008-02/msg01396.html
> > and MIPS
> >   http://gcc.gnu.org/ml/gcc-testresults/2008-02/msg01409.html
> > at least.
> >
> > The testcase defines a 16-byte aligned type and uses an automatic
> > variable with that type.  The test checks if the alignment of
> > this variable on stack matches with __alignof__ of that type.
> > It seems that this fails on targets with PREFERRED_STACK_BOUNDARY
> > less than 128, because the real alignment of that variable is
> > limited to PREFERRED_STACK_BOUNDARY.
> > Since our doc/extend.texi says
> >
> >  Whenever you leave out the alignment factor in an @code{aligned} attribute
> >  specification, the compiler automatically sets the alignment for the declared
> >  variable or field to the largest alignment which is ever used for any data
> >  type on the target machine you are compiling for. 
> >
> > how about the patch below?
> > It's tested on i686-pc-linux-gnu with "make -k check".
>
> OK unless the test's author, Joey Ye, objects and has a
> different suggestion for the targets where the test fails.

Note that Richard Sandiford fixed the failure on mips by not running the 
test on that target.

If we are fixing the test so that it passes (or should pass) on all 
targets, Richard's patch should probably be reverted.

An alternative to Kaz's patch would be to add SH and ARM to the targets 
that don't run the test.

David Daney