------- Comment #4 from dave at hiauly1 dot hia dot nrc dot ca 2010-09-02 16:27 ------- Subject: Re: [4.6 Regression] FAIL: gcc.dg/torture/builtin-cproj-1.c -O1 (test for excess errors)
On Thu, 02 Sep 2010, ghazi at gcc dot gnu dot org wrote: > > > ------- Comment #3 from ghazi at gcc dot gnu dot org 2010-09-02 15:24 ------- > > Please check whether > > +/* { dg-add-options c99_runtime } */ > > +/* { dg-require-effective-target c99_runtime } */ > > You're right that these foldings should succeed anyway, the c99_runtime should > not be necessary. If requiring a c99 effective target eliminates the error, > it's only because the test doesn't run anymore. > > The subject indicates the test fails at -O1, but I assume not -O0 or -O2. > This > leads me to guess that the failing line(s) are in the #ifdef __OPTIMIZE__ > section. These bits rely on some generic optimizations to fully fold away the > relevant code, which may not be happening here at -O1. > > I don't have access to a test infrastructure ATM. So if David could please > narrow down which line is failing to fold, it would help. Each test line here > calls link_error(__LINE__) so if you add -fdump-tree-* you should be able to > see which one(s) aren't folding, and hopefully why. > > I suspect it's something more than the c99 complex stuff. The test also fails at -O1 on hppa-unknown-linux-gnu. This is a c99 target. It doesn't fail at -O0 or at -O2. Attached builtin-cproj-1.c.149t.optimized. ------- Comment #5 from dave at hiauly1 dot hia dot nrc dot ca 2010-09-02 16:27 ------- Created an attachment (id=21671) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21671&action=view) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43959