[tuples][patch] Don't create GIMPLE_CONDs that trap

Janis Johnson janis187@us.ibm.com
Wed May 7 00:18:00 GMT 2008


On Tue, 2008-05-06 at 14:47 -0700, Diego Novillo wrote:
> On 5/6/08 1:41 PM, Rafael Espindola wrote:
> 
> > The attached patch does that. I had to add -fno-trapping-math to some
> > tests. I am not familiar with sse2, so I am not sure if this is a fix
> > or if I should open a PR for a missed optimization.
> 
> You can't really use -msse2 here, because these tests need to run on 
> several platforms.  You only want to add -fno-trapping-math and leave 
> the others the same, right?  I'm not quite sure how to do that in dejagnu.
> 
> The non-testsuite changes are fine.

The vect tests already use -msse2 for i*86 and x86_64 but don't
run them if the test machine doesn't support sse2.

There's no way to add -fno-trapping-math to the existing options,
but there are some tests called no-trapping-math*.c that use that
option.  I've copied Dorit, who can help determine if the tests
that Rafael wanted to modify should be renamed.

Janis



More information about the Gcc-patches mailing list