From: Richard Biener <richard.guenther@gmail.com>
To: "Martin Liška" <mliska@suse.cz>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>, Jan Hubicka <hubicka@ucw.cz>
Subject: Re: [PATCH] Handle PHI nodes w/o a argument (PR ipa/80205).
Date: Tue, 28 Mar 2017 08:23:00 -0000 [thread overview]
Message-ID: <CAFiYyc0924oF_6X0FFXdfMZ63h6Cei8hijGMn+XX6Awwk6Lzeg@mail.gmail.com> (raw)
In-Reply-To: <d40534cd-9e13-cb67-8e94-da959c136843@suse.cz>
On Tue, Mar 28, 2017 at 9:45 AM, Martin Liška <mliska@suse.cz> wrote:
> On 03/27/2017 04:27 PM, Richard Biener wrote:
>> On Mon, Mar 27, 2017 at 4:14 PM, Richard Biener
>> <richard.guenther@gmail.com> wrote:
>>> On Mon, Mar 27, 2017 at 2:47 PM, Martin Liška <mliska@suse.cz> wrote:
>>>> Hello.
>>>>
>>>> As described in the PR, we can create a PHI node in einline that has no argument.
>>>> That can cause ICE in devirtualization and should be thus handled.
>>>>
>>>> Patch can bootstrap on ppc64le-redhat-linux and survives regression tests.
>>>>
>>>> Ready to be installed?
>>>
>>> We shouldn't ever have a PHI w/o argument.
>>
>> ;; basic block 14, loop depth 0
>> ;; pred:
>> # SR.2_19 = PHI <>
>> <L4> [0.00%]:
>> a::~a (&g);
>> resx 12
>> ;; succ: 17
>>
>> the CFG has not been cleaned up here (this block is unreachable).
>>
>> Hmm, I see we are called from fold_stmt. I suppose we process
>> statements_to_fold before removing unreachable blocks to not
>> walk over dead stmts... chicken-and-egg...
>>
>> Still not creating that PHI should be possible.
>
> I see, thanks for help. Patch fixes the issue, may I install it after
> regression tested?
I think it's progression, still not 100% safe as we have a SSA name with
a GIMPLE_NOP definition that is not in any BB and the SSA name is not
a default def.
But yes, I've written a comment to that effect ;)
Thus ok.
An improvement would be to not copy unreachable regions at all...
Thanks,
Richard.
> Thanks,
> Martin
>
>>
>> Index: gcc/tree-inline.c
>> ===================================================================
>> --- gcc/tree-inline.c (revision 246500)
>> +++ gcc/tree-inline.c (working copy)
>> @@ -2344,6 +2344,13 @@ copy_phis_for_bb (basic_block bb, copy_b
>> if (!virtual_operand_p (res))
>> {
>> walk_tree (&new_res, copy_tree_body_r, id, NULL);
>> + if (EDGE_COUNT (new_bb->preds) == 0)
>> + {
>> + /* Technically we'd want a SSA_DEFAULT_DEF here... */
>> + SSA_NAME_DEF_STMT (new_res) = gimple_build_nop ();
>> + }
>> + else
>> + {
>> new_phi = create_phi_node (new_res, new_bb);
>> FOR_EACH_EDGE (new_edge, ei, new_bb->preds)
>> {
>> @@ -2389,6 +2396,7 @@ copy_phis_for_bb (basic_block bb, copy_b
>>
>> add_phi_arg (new_phi, new_arg, new_edge, locus);
>> }
>> + }
>> }
>> }
>>
>>
>>
>>>> Martin
>
prev parent reply other threads:[~2017-03-28 8:12 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-03-27 12:51 Martin Liška
2017-03-27 14:25 ` Richard Biener
2017-03-27 14:27 ` Jeff Law
2017-03-27 14:28 ` Richard Biener
2017-03-27 15:30 ` Jeff Law
2017-03-28 7:52 ` Martin Liška
2017-03-28 8:23 ` 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=CAFiYyc0924oF_6X0FFXdfMZ63h6Cei8hijGMn+XX6Awwk6Lzeg@mail.gmail.com \
--to=richard.guenther@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=hubicka@ucw.cz \
--cc=mliska@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).