From: Richard Biener <rguenther@suse.de>
To: Jan Hubicka via Gcc-patches <gcc-patches@gcc.gnu.org>
Cc: Martin Jambor <mjambor@suse.cz>
Subject: Re: [PATCH 8/9] ipa-sra: Make scan_expr_access bail out on uninteresting expressions
Date: Tue, 13 Dec 2022 09:40:42 +0100 [thread overview]
Message-ID: <1AEB9854-AF72-4D81-9172-EB04856E3846@suse.de> (raw)
In-Reply-To: <Y5ekE5o7EmwxJvki@kam.mff.cuni.cz>
> Am 12.12.2022 um 22:59 schrieb Jan Hubicka via Gcc-patches <gcc-patches@gcc.gnu.org>:
>
>
>>
>>> Hi,
>>>
>>> I'm re-posting patches which I have posted at the end of stage 1 but
>>> which have not passed review yet.
>>>
>>> 8<--------------------------------------------------------------------
>>>
>>> I have noticed that scan_expr_access passes all the expressions it
>>> gets to get_ref_base_and_extent even when we are really only
>>> interested in memory accesses. So bail out when the expression is
>>> something clearly uninteresting.
>>>
>>> Bootstrapped and tested individually when I originally posted it and
>>> now bootstrapped and LTO-bootstrapped and tested as part of the whole
>>> series. OK for master?
>>>
>>>
>>> gcc/ChangeLog:
>>>
>>> 2021-12-14 Martin Jambor <mjambor@suse.cz>
>>>
>>> * ipa-sra.c (scan_expr_access): Bail out early if expr is something we
>>> clearly do not need to pass to get_ref_base_and_extent.
>>> ---
>>> gcc/ipa-sra.cc | 5 +++++
>>> 1 file changed, 5 insertions(+)
>>>
>>> diff --git a/gcc/ipa-sra.cc b/gcc/ipa-sra.cc
>>> index 93fceeafc73..3646d71468c 100644
>>> --- a/gcc/ipa-sra.cc
>>> +++ b/gcc/ipa-sra.cc
>>> @@ -1748,6 +1748,11 @@ scan_expr_access (tree expr, gimple *stmt, isra_scan_context ctx,
>>> || TREE_CODE (expr) == REALPART_EXPR)
>>> expr = TREE_OPERAND (expr, 0);
>>>
>>> + if (!handled_component_p (expr)
>>> + && !DECL_P (expr)
>>> + && TREE_CODE (expr) != MEM_REF)
>>> + return;
>> Is this needed because get_ref_base_and_extend crashes if given SSA_NAME
>> or something else or is it just optimization?
>> Perhaps Richi will know if there is better test for this.
> Looking at:
>
> static inline bool
> gimple_assign_load_p (const gimple *gs)
> {
> tree rhs;
> if (!gimple_assign_single_p (gs))
> return false;
> rhs = gimple_assign_rhs1 (gs);
> if (TREE_CODE (rhs) == WITH_SIZE_EXPR)
> return true;
> rhs = get_base_address (rhs);
> return (DECL_P (rhs)
> || TREE_CODE (rhs) == MEM_REF || TREE_CODE (rhs) == TARGET_MEM_REF);
> }
>
> I wonder if we don't want to avoid get_base_address (which is loopy) and
> use same check and move it into a new predicate that is more convenient
> to use?
We can simplify the above to a single stripping of a handled component and considering another handled component as load (register ops are always single)
Richard
>
> Honza
>>
>> Honza
>>> +
>>> base = get_ref_base_and_extent (expr, &poffset, &psize, &pmax_size, &reverse);
>>>
>>> if (TREE_CODE (base) == MEM_REF)
>>> --
>>> 2.38.1
>>>
next prev parent reply other threads:[~2022-12-13 8:40 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-12-12 16:53 Martin Jambor
2022-12-12 21:56 ` Jan Hubicka
2022-12-12 21:58 ` Jan Hubicka
2022-12-13 8:40 ` Richard Biener [this message]
2022-12-13 12:53 ` Richard Biener
2022-12-14 13:20 ` Martin Jambor
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=1AEB9854-AF72-4D81-9172-EB04856E3846@suse.de \
--to=rguenther@suse.de \
--cc=gcc-patches@gcc.gnu.org \
--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).