From: Richard Biener <richard.guenther@gmail.com>
To: Martin Jambor <mjambor@suse.cz>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>, Jan Hubicka <hubicka@ucw.cz>
Subject: Re: [PATCH] ipa: Avoid constructing aggregate jump functions with huge offsets (PR 109303)
Date: Fri, 31 Mar 2023 11:24:39 +0200 [thread overview]
Message-ID: <CAFiYyc3uAwRcbvc6uZtawspzqq4u0Jx3=MXLO2_DUS=nT+6bvw@mail.gmail.com> (raw)
In-Reply-To: <ri6h6u146um.fsf@suse.cz>
On Fri, Mar 31, 2023 at 11:18 AM Martin Jambor <mjambor@suse.cz> wrote:
>
> Hi,
>
> On Fri, Mar 31 2023, Richard Biener wrote:
> > On Fri, Mar 31, 2023 at 10:46 AM Martin Jambor <mjambor@suse.cz> wrote:
> >>
> >> Hi,
> >>
> >> we are in the process of changing data structures holding information
> >> about constants passed by reference and in aggregates to use unsigned
> >> int offsets rather than HOST_WIDE_INT (which was selected simply
> >> because that is what fell out of get_ref_base_and_extent at that time)
> >> in order to conserve memory, especially at WPA time.
> >>
> >> PR 109303 testcase discovers that we do not properly check that we
> >> only create jump functions with offsets (plus sizes) that fit into the
> >> smaller type. This patch adds the necessary check.
> >>
> >> Bootstrapped and tested on x86_64-linux. OK for master?
> >>
> >> Thanks,
> >>
> >> Martin
> >>
> >>
> >> gcc/ChangeLog:
> >>
> >> 2023-03-30 Martin Jambor <mjambor@suse.cz>
> >>
> >> PR ipa/109303
> >> * ipa-prop.cc (determine_known_aggregate_parts): Check that the
> >> offset + size will be representable in unsigned int.
> >>
> >> gcc/testsuite/ChangeLog:
> >>
> >> 2023-03-30 Jakub Jelinek <jakub@redhat.com>
> >> Martin Jambor <mjambor@suse.cz>
> >>
> >> PR ipa/109303
> >> * gcc.dg/pr109303.c: New test.
> >> ---
> >> gcc/ipa-prop.cc | 4 +++-
> >> gcc/testsuite/gcc.dg/pr109303.c | 24 ++++++++++++++++++++++++
> >> 2 files changed, 27 insertions(+), 1 deletion(-)
> >> create mode 100644 gcc/testsuite/gcc.dg/pr109303.c
> >>
> >> diff --git a/gcc/ipa-prop.cc b/gcc/ipa-prop.cc
> >> index de45dbccf16..9ffd49b590c 100644
> >> --- a/gcc/ipa-prop.cc
> >> +++ b/gcc/ipa-prop.cc
> >> @@ -2086,7 +2086,9 @@ determine_known_aggregate_parts (struct ipa_func_body_info *fbi,
> >> whether its value is clobbered any other dominating one. */
> >> if ((content->value.pass_through.formal_id >= 0
> >> || content->value.pass_through.operand)
> >> - && !clobber_by_agg_contents_list_p (all_list, content))
> >> + && !clobber_by_agg_contents_list_p (all_list, content)
> >> + && (content->offset + content->size - arg_offset
> >> + <= (HOST_WIDE_INT) UINT_MAX * BITS_PER_UNIT))
> >> {
> >
> > it does seem a bit misplaced since after the if we add the same
> > 'content' to another
> > list anyway.
>
> The other list is a clobber list, as we crawl backwards from the call
> statement searching for stores, we also look whether we have already
> encountered a store of something else to an overlapping area. In theory
> we could have a store to a smaller data type, where the offset + size
> would still fit unsigned int, be followed by a larger store, which would
> not. We want the large store to end up in the clobber list so that the
> smaller one does not.
>
> This is the place where we also calculate the size of the final
> heap-allocated vector, so that is why eventually I put it there.
>
> > Wouldn't a more obvious place be where we end up truncating this sum?
>
> My reasoning was that since we know we would not be able to use it, it
> makes sense to discard the data before we stream it from compilation to
> WPA.
>
> Also, when we shorten the offset type also in ipa_agg_jf_item (which is
> what I want to do next), this is where the check eventually needs to
> be.
OK, maybe add the above as comment then. OK with that change.
Richard.
> Thanks,
>
> Martin
prev parent reply other threads:[~2023-03-31 9:24 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-03-31 8:45 Martin Jambor
2023-03-31 8:57 ` Richard Biener
2023-03-31 9:18 ` Martin Jambor
2023-03-31 9:24 ` Richard Biener [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAFiYyc3uAwRcbvc6uZtawspzqq4u0Jx3=MXLO2_DUS=nT+6bvw@mail.gmail.com' \
--to=richard.guenther@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=hubicka@ucw.cz \
--cc=mjambor@suse.cz \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).