[PATCH] tree-call-cdce: If !HONOR_NANS do not make code with NaNs (PR88055)

Segher Boessenkool segher@kernel.crashing.org
Fri Apr 19 17:19:00 GMT 2019


On Fri, Apr 19, 2019 at 10:15:31AM -0600, Jeff Law wrote:
> On 4/18/19 3:23 PM, Segher Boessenkool wrote:
> > If we don't HONOR_NANS we should not try to use any unordered
> > comparison results.  Best case those will just be optimized away;
> > realistically, they ICE.  For example, the rs6000 backend has some
> > code that specifically checks we never do this.
> > 
> > This patch fixes it.  Bootstrapped and tested on powerpc64-linux
> > {-m32,-m64}.  Is this okay for trunk?
> > 
> > 2019-04-18  Segher Boessenkool  <segher@kernel.crashing.org>
> > 
> > 	PR tree-optimization/88055
> > 	* tree-call-cdce.c (comparison_code_if_no_nans): New function.
> > 	(gen_one_condition): Use it if !HONOR_NANS.
> ISTM that this should have been marked as a regression since presumably
> at some point we didn't try to eliminate the call and thus wouldn't have
> generated the problematic code and ICE'd.

Yes, I just still don't know how to mark things as regressions.  It's a
GCC 9 regression.

> Thus, OK for gcc-9.

Thanks!


Segher



More information about the Gcc-patches mailing list