Re: [PATCH,testsuite] fix gcc.dg/vect/vect-{82,83}.c

[Janis Johnson <janis187 at us dot ibm dot com>]

On Wed, 2009-03-11 at 12:40 -0700, Nathan Froyd wrote:
> vect.exp has this bit of code in it for powerpc targets:
> 
>     lappend DEFAULT_VECTCFLAGS "-maltivec"
>     if [check_vmx_hw_available] {
> 	set dg-do-what-default run
>     } else {
> 	if [is-effective-target ilp32] {
> 	    # Specify a cpu that supports VMX for compile-only tests.
> 	    lappend DEFAULT_VECTCFLAGS "-mcpu=970"
> 	}
> 	set dg-do-what-default compile
>     }
> 
> On 32-bit targets that don't have VMX hardware, we'll compile with
> -mcpu=970, which frees up the compiler to use 64-bit instructions where
> appropriate.  vect-{82,83}.c, however, are specified to be runnable
> testcases.  They also use 'long long unsigned int' variables, which get
> stored into via 'std' instructions, since we're compiling for a 64-bit
> CPU.
> 
> These instructions would normally throw SIGILL and cause the testcase to
> exit cleanly...except on bare-metal targets, signal handlers don't work
> so well.  This causes spurious execution failures.
> 
> The solution is to remove the `dg-do run' directives from these
> testcases, as most other vect testcases don't have such directives,
> preferring to use dg-do-what-default in vect.exp instead.
> 
> Tested on powerpc-eabi.  OK to commit?

> -/* { dg-do run { target powerpc*-*-* } } */
>  /* { dg-require-effective-target vect_int } */


By removing the dg-do directive you're also losing the restriction
that the test only run on powerpc*-*-*.  You can combine the two
restrictions by replacing both test directives with

/* { dg-skip-if "" { ! { powerpc*-*-* && vect_int } } { "*" } { "" }  } */

although you should replace the empty comment with a real one.

That change is OK.

Janis