[PATCH] Fix PR optimization/10157

[Eric Botcazou]

> My feeling is that this is ok, we can tolerate it for the rare cases
> where it is an issue.  The extra test coverage is worth it.  And it
> doesn't even come into play with your recent testsuite entries.  So
> please do use sparc*, including for tests you've already checked in.

Ok, but I'd like us to first settle on a well-defined set of options before I 
start to modify anything.

The gcc.dg/ultrasp* tests come with two flavors of options:
- tests #1, #3, #5, #6, #7 and #8 use

/* { dg-do compile { target sparc-*-* } } */
/* { dg-options "-mcpu=ultrasparc" } */

- tests #2 and #4 use

/* { dg-do compile } */
/* { dg-options "" } */
/* { dg-options "-O0 -m64" { target sparc64-*-* } } */
/* { dg-options "-O0 -m64" { target sparcv9-*-* } } */
/* { dg-options "" { target sparc-*-solaris2.[0-6] } } */
/* { dg-options "" { target sparc-*-solaris2.[0-6].* } } */


I propose to modify the first 6 tests on mainline to use:

/* { dg-do compile { target sparc*-*-* } } */
/* { dg-options "-mcpu=ultrasparc" { target sparc-*-* } } */
/* { dg-options "-m32" { target sparc64-*-* } } */

if it is a 32-bit test and

/* { dg-do compile { target sparc*-*-* } } */
/* { dg-options "-mcpu=ultrasparc -m64" { target sparc-*-* } } */
/* { dg-options "" { target sparc64-*-* } } */

if it is a 64-bit test, so that you don't get failures by adding -m64 to 
sparc32 or -m32 to sparc64.


And I propose to remove the "-m64" switch and the "sparcv9" line in the two 
other testcases, in order to avoid failures with "-m32".

-- 
Eric Botcazou