RE: [PATCH V2] Generalized value pass-through for self-recursive function (ipa/pr93203)

[Tamar Christina <Tamar.Christina@arm.com>]

Hi Feng,

It looks like the option that causes this to trigger is `--param ipa-cp-eval-threshold=1`

So on AArch64 I get it to trigger with -mcpu=native -Ofast -fomit-frame-pointer -flto --param ipa-cp-eval-threshold=1

Kind Regards,
Tamar

> -----Original Message-----
> From: Feng Xue OS <fxue@os.amperecomputing.com>
> Sent: Tuesday, February 11, 2020 13:11
> To: Tamar Christina <Tamar.Christina@arm.com>; Martin Jambor
> <mjambor@suse.cz>; Jan Hubicka <hubicka@ucw.cz>; gcc-
> patches@gcc.gnu.org
> Cc: nd <nd@arm.com>
> Subject: Re: [PATCH V2] Generalized value pass-through for self-recursive
> function (ipa/pr93203)
> 
> Hi Tarmar,
> 
>   Since I could not reproduce this ICE with my local testing on Spec2017, would
> you share your config, or command line options extracted from compilation
> of perlbench?
> 
> Thanks,
> Feng
> 
> ________________________________________
> From: gcc-patches-owner@gcc.gnu.org <gcc-patches-owner@gcc.gnu.org>
> on behalf of Tamar Christina <Tamar.Christina@arm.com>
> Sent: Tuesday, February 11, 2020 6:05 PM
> To: Feng Xue OS; Martin Jambor; Jan Hubicka; gcc-patches@gcc.gnu.org
> Cc: nd
> Subject: RE: [PATCH V2] Generalized value pass-through for self-recursive
> function (ipa/pr93203)
> 
> Hi Feng,
> 
> This patch (commit a0f6a8cb414b687f22c9011a894d5e8e398c4be0) is causing
> ICEs in the GCC and perlbench benchmark in Spec2017.
> 
> during IPA pass: cp
> lto1: internal compiler error: in find_more_scalar_values_for_callers_subset,
> at ipa-cp.c:4709
> 0x1698187 find_more_scalar_values_for_callers_subset
>         ../.././gcc/ipa-cp.c:4709
> 0x169f7d3 decide_about_value<tree_node*>
>         ../.././gcc/ipa-cp.c:5490
> 0x169fdc3 decide_whether_version_node
>         ../.././gcc/ipa-cp.c:5537
> 0x169fdc3 ipcp_decision_stage
>         ../.././gcc/ipa-cp.c:5718
> 0x169fdc3 ipcp_driver
>         ../.././gcc/ipa-cp.c:5901
> Please submit a full bug report,
> 
> Thanks,
> Tamar
> 
> > -----Original Message-----
> > From: gcc-patches-owner@gcc.gnu.org <gcc-patches-owner@gcc.gnu.org>
> On
> > Behalf Of Feng Xue OS
> > Sent: Monday, February 10, 2020 03:29
> > To: Martin Jambor <mjambor@suse.cz>; Jan Hubicka <hubicka@ucw.cz>;
> > gcc-patches@gcc.gnu.org
> > Subject: Re: [PATCH V2] Generalized value pass-through for
> > self-recursive function (ipa/pr93203)
> >
> > >> -           gcc_checking_assert (item->value);
> >
> > > I've been staring at this for quite a while, trying to figure out
> > > how your patch can put NULL here before I realized it was just a
> > > clean-up
> > > :-)  Sending such changes independently or pointing them out in the
> > > email/ChangeLog makes review easier.
> >
> > Ok. I'll add some description on this cleanup on ChangeLog.
> >
> > >> @@ -5564,7 +5610,6 @@ decide_whether_version_node (struct
> > cgraph_node *node)
> > >>       }
> > >>        clone = create_specialized_node (node, known_csts,
> known_contexts,
> > >>                                      aggvals, callers);
> > >> -      info = IPA_NODE_REF (node);
> >
> > > please either drop this change or change it to:
> > >
> > >   gcc_checking_assert (info == IPA_NODE_REF (node));
> >
> > > this line of code was actually necessary when adding nodes possibly
> > > invalidated addresses of all summaries - like fast_function_summary
> > > classes still do.  So if we ever decide to use fast summaries we
> > > need a test to remind us that info address must be obtained again.
> > Ok. I'm not aware that, will keep the line as original.
> >
> > Thanks,
> > Feng