This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH][RFC] Consolidate -O3 torture options
- From: Richard Biener <rguenther at suse dot de>
- To: Jeff Law <law at redhat dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Wed, 15 Jul 2015 10:01:12 +0200 (CEST)
- Subject: Re: [PATCH][RFC] Consolidate -O3 torture options
- Authentication-results: sourceware.org; auth=none
- References: <alpine dot LSU dot 2 dot 11 dot 1507141353530 dot 9923 at zhemvz dot fhfr dot qr> <55A52948 dot 90505 at redhat dot com>
On Tue, 14 Jul 2015, Jeff Law wrote:
> On 07/14/2015 05:58 AM, Richard Biener wrote:
> >
> > The following patch tries to consolidate the -O3 torture testing
> > options in the attempt to reduce testing time while not losing
> > coverage.
> >
> > It drops testing of -funroll-all-loops (which nobody should use)
> > and retains only one non-default -O3 set of options - namely
> > -O3 plus those flags that would be enabled by -fprofile-use.
> >
> > One should hope for ~20% less time in the C and dg tortures this way.
> >
> > Didn't look into other tortures to apply the same yet (objc-torture?)
> >
> > Currently testing on x86_64-unknown-linux-gnu.
> >
> > For weird flag combinations we do have contributors that test
> > them and regularly report bugzillas.
> >
> > Ok?
> >
> > Comments?
> >
> > Thanks,
> > Richard.
> >
> > 2015-07-14 Richard Biener <rguenther@suse.de>
> >
> > * lib/c-torture.exp (C_TORTURE_OPTIONS): Remove
> > { -O3 -fomit-frame-pointer },
> > { -O3 -fomit-frame-pointer -funroll-loops } and
> > { -O3 -fomit-frame-pointer -funroll-all-loops -finline-functions }
> > in favor of
> > { -O3 -fomit-frame-pointer -funroll-loops -fpeel-loops -ftracer
> > -finline-functions }
> > * lib/gcc-dg.exp (DG_TORTURE_OPTIONS): Likewise.
> I think this is OK -- I've occasionally wondered about the additional coverage
> we get vs the amount of time spent for the various options.
>
> I can't recall specific cases where one of those 3 options would trigger a
> failure, but the two didn't. I'm sure it's happened, but it's just common
> enough to warrant the amount of time we spend testing it.
>
> This patch has the additional benefit that I think we can eliminate scanning
> the source for loops and eliminating the -funroll[-all]-loops options. Hmm,
> that code may have already been dead... Hmmm.
Testing reveals one fallout:
FAIL: gcc.c-torture/execute/builtins/snprintf-chk.c execution, -O3
-fomit-frame
-pointer -funroll-loops -fpeel-loops -ftracer -finline-functions
FAIL: gcc.c-torture/execute/builtins/vsnprintf-chk.c execution, -O3
-fomit-fram
e-pointer -funroll-loops -fpeel-loops -ftracer -finline-functions
this is because -ftracer manages to thread things in a way that we
optimize away a _chk variant at compile-time and hit
if (chk_calls != 5)
abort ();
with chk_calls == 4. I'm going to install the following alongside
(we've had similar issues with -ftree-loop-distribute-patterns enabled
by -O3).
Richard.
Index: gcc/testsuite/gcc.c-torture/execute/builtins/builtins.exp
===================================================================
--- gcc/testsuite/gcc.c-torture/execute/builtins/builtins.exp (revision
225768)
+++ gcc/testsuite/gcc.c-torture/execute/builtins/builtins.exp (working
copy)
@@ -37,7 +37,7 @@ load_lib c-torture.exp
torture-init
set-torture-options $C_TORTURE_OPTIONS {{}} $LTO_TORTURE_OPTIONS
-set additional_flags "-fno-tree-loop-distribute-patterns"
+set additional_flags "-fno-tree-loop-distribute-patterns -fno-tracer"
if [istarget "powerpc-*-darwin*"] {
lappend additional_flags "-Wl,-multiply_defined,suppress"
}