From: Jakub Jelinek <jakub@redhat.com>
To: Siddhesh Poyarekar <siddhesh@gotplt.org>
Cc: gcc-patches@gcc.gnu.org
Subject: Re: [PATCH v4 3/6] tree-object-size: Support dynamic sizes in conditions
Date: Wed, 15 Dec 2021 19:52:53 +0100 [thread overview]
Message-ID: <20211215185253.GN2646553@tucnak> (raw)
In-Reply-To: <1c596138-c500-bf62-0e1f-28063fb9edc3@gotplt.org>
On Wed, Dec 15, 2021 at 11:26:48PM +0530, Siddhesh Poyarekar wrote:
> > This makes me a little bit worried. Do you compute the wholesize SSA_NAME
> > at runtime always, or only when it is really needed and known not to always
> > be equal to the size?
> > I mean, e.g. for the cases where there is just const char *p = malloc (size);
> > and the pointer is never increased size == wholesize. For __bos it will
> > just be 2 different INTEGER_CSTs, but if it would at runtime mean we compute
> > something twice and hope we eventually find out during later passes
> > it is the same, it would be bad.
>
> I'm emitting both size and wholesize all the time; wholesize only really
> gets used in size_for_offset and otherwise should get DCE'd. Right now for
> __bos (and constant sizes) wholesize is unused if it is the same as size.
>
> FOR GIMPLE_CALL, GIMPLE_NOP, etc. I return the same tree for size and
> wholesize; maybe a trivial pointer comparison (sz != wholesize) ought to get
> rid of most of the uses in size_for_offset.
Perhaps DCE can handle well when you compute something (wholesize) that isn't really
needed and VN/CSE the case where size and wholesize is equal. I think it
would be worth looking at a few testcases.
> > > + {
> > > + edge e = gimple_phi_arg_edge (obj_phi, i);
> > > +
> > > + /* Put the size definition before the last statement of the source
> > > + block of the PHI edge. This ensures that any branches at the end
> > > + of the source block remain the last statement. We are OK even if
> > > + the last statement is the definition of the object since it will
> > > + succeed any definitions that contribute to its size and the size
> > > + expression will succeed them too. */
> > > + gimple_stmt_iterator gsi = gsi_last_bb (e->src);
> > > + gsi_insert_seq_before (&gsi, seq, GSI_CONTINUE_LINKING);
> >
> > This looks problematic. The last stmt in the bb might not exist at all,
>
> Wouldn't the bb minimally have to contain the definition of the object whose
> size we computed? e.g. for PHI [a(2), b(3)], wouldn't bb 2 at least have a
> statement with the definition of a?
It can e.g. contain just a PHI.
> Or wait, there could be situations where the definition is in a different
> block, e.g. bb 1, which has a single edge going on to bb 2?
> I suppose __bos-like behaviour could be a good compromise, i.e. insert a
> MAX_EXPR (or MIN_EXPR) if we can't find a suitable location to insert on
> edge.
MAX_EXPR or MIN_EXPR? I'd have expect the __bos constant in there.
But I must say I'm right now unsure what kind of PHIs one can have on bbs
reachable from both ab/eh edges and normal edges if we can have such bbs at
all. I guess looking at some sigjmp/longjmp or non-local or computed goto
testcases might show something, perhaps I'll have a look tomorrow.
I'm sure we can have vop PHI.
Jakub
next prev parent reply other threads:[~2021-12-15 18:53 UTC|newest]
Thread overview: 97+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-11-09 19:01 [PATCH 00/10] __builtin_dynamic_object_size Siddhesh Poyarekar
2021-11-09 19:01 ` [PATCH 01/10] tree-object-size: Replace magic numbers with enums Siddhesh Poyarekar
2021-11-19 16:00 ` Jakub Jelinek
2021-11-09 19:01 ` [PATCH 02/10] tree-object-size: Abstract object_sizes array Siddhesh Poyarekar
2021-11-19 16:18 ` Jakub Jelinek
2021-11-19 16:53 ` Siddhesh Poyarekar
2021-11-09 19:01 ` [PATCH 03/10] tree-object-size: Use tree instead of HOST_WIDE_INT Siddhesh Poyarekar
2021-11-19 17:06 ` Jakub Jelinek
2021-11-19 19:01 ` Siddhesh Poyarekar
2021-11-19 19:16 ` Jakub Jelinek
2021-11-22 8:41 ` Richard Biener
2021-11-22 10:11 ` Siddhesh Poyarekar
2021-11-22 10:31 ` Jakub Jelinek
2021-11-22 12:00 ` Siddhesh Poyarekar
2021-11-22 12:31 ` Siddhesh Poyarekar
2021-11-22 12:32 ` Jakub Jelinek
2021-11-23 11:58 ` Jakub Jelinek
2021-11-23 13:33 ` Siddhesh Poyarekar
2021-11-09 19:01 ` [PATCH 04/10] tree-object-size: Single pass dependency loop resolution Siddhesh Poyarekar
2021-11-23 12:07 ` Jakub Jelinek
2021-11-23 13:44 ` Siddhesh Poyarekar
2021-11-23 14:22 ` Jakub Jelinek
2021-11-09 19:01 ` [PATCH 05/10] __builtin_dynamic_object_size: Recognize builtin Siddhesh Poyarekar
2021-11-23 12:41 ` Jakub Jelinek
2021-11-23 13:53 ` Siddhesh Poyarekar
2021-11-23 14:00 ` Jakub Jelinek
2021-11-09 19:01 ` [PATCH 06/10] tree-object-size: Support dynamic sizes in conditions Siddhesh Poyarekar
2021-11-23 15:12 ` Jakub Jelinek
2021-11-23 15:36 ` Siddhesh Poyarekar
2021-11-23 15:38 ` Siddhesh Poyarekar
2021-11-23 16:17 ` Jakub Jelinek
2021-11-23 15:52 ` Jakub Jelinek
2021-11-23 16:00 ` Siddhesh Poyarekar
2021-11-23 16:19 ` Jakub Jelinek
2021-11-09 19:01 ` [PATCH 07/10] tree-object-size: Handle function parameters Siddhesh Poyarekar
2021-11-09 19:01 ` [PATCH 08/10] tree-object-size: Handle GIMPLE_CALL Siddhesh Poyarekar
2021-11-09 19:01 ` [PATCH 09/10] tree-object-size: Dynamic sizes for ADDR_EXPR Siddhesh Poyarekar
2021-11-09 19:01 ` [PATCH 10/10] tree-object-size: Handle dynamic offsets Siddhesh Poyarekar
2021-11-19 15:56 ` [PATCH 00/10] __builtin_dynamic_object_size Jakub Jelinek
2021-11-26 5:28 ` [PATCH v3 0/8] __builtin_dynamic_object_size Siddhesh Poyarekar
2021-11-26 5:28 ` [PATCH v3 1/8] tree-object-size: Replace magic numbers with enums Siddhesh Poyarekar
2021-11-26 16:46 ` Jakub Jelinek
2021-11-26 17:53 ` Siddhesh Poyarekar
2021-11-26 18:01 ` Jakub Jelinek
2021-11-26 5:28 ` [PATCH v3 2/8] tree-object-size: Abstract object_sizes array Siddhesh Poyarekar
2021-11-26 16:47 ` Jakub Jelinek
2021-11-26 5:28 ` [PATCH v3 3/8] tree-object-size: Save sizes as trees and support negative offsets Siddhesh Poyarekar
2021-11-26 16:56 ` Jakub Jelinek
2021-11-26 17:59 ` Siddhesh Poyarekar
2021-11-26 18:04 ` Jakub Jelinek
2021-11-26 18:07 ` Siddhesh Poyarekar
2021-11-26 5:28 ` [PATCH v3 4/8] __builtin_dynamic_object_size: Recognize builtin Siddhesh Poyarekar
2021-11-26 5:28 ` [PATCH v3 5/8] tree-object-size: Support dynamic sizes in conditions Siddhesh Poyarekar
2021-11-26 5:28 ` [PATCH v3 6/8] tree-object-size: Handle function parameters Siddhesh Poyarekar
2021-11-26 5:28 ` [PATCH v3 7/8] tree-object-size: Handle GIMPLE_CALL Siddhesh Poyarekar
2021-11-26 5:28 ` [PATCH v3 8/8] tree-object-size: Dynamic sizes for ADDR_EXPR Siddhesh Poyarekar
2021-11-26 5:38 ` [PATCH v3 0/8] __builtin_dynamic_object_size Siddhesh Poyarekar
2021-12-01 14:27 ` [PATCH v4 0/6] __builtin_dynamic_object_size Siddhesh Poyarekar
2021-12-01 14:27 ` [PATCH v4 1/6] tree-object-size: Use trees and support negative offsets Siddhesh Poyarekar
2021-12-15 15:21 ` Jakub Jelinek
2021-12-15 17:12 ` Siddhesh Poyarekar
2021-12-15 18:43 ` Jakub Jelinek
2021-12-16 0:41 ` Siddhesh Poyarekar
2021-12-16 15:49 ` Jakub Jelinek
2021-12-16 18:56 ` Siddhesh Poyarekar
2021-12-16 21:16 ` Jakub Jelinek
2021-12-01 14:27 ` [PATCH v4 2/6] __builtin_dynamic_object_size: Recognize builtin Siddhesh Poyarekar
2021-12-15 15:24 ` Jakub Jelinek
2021-12-16 2:16 ` Siddhesh Poyarekar
2021-12-01 14:27 ` [PATCH v4 3/6] tree-object-size: Support dynamic sizes in conditions Siddhesh Poyarekar
2021-12-15 16:24 ` Jakub Jelinek
2021-12-15 17:56 ` Siddhesh Poyarekar
2021-12-15 18:52 ` Jakub Jelinek [this message]
2021-12-01 14:27 ` [PATCH v4 4/6] tree-object-size: Handle function parameters Siddhesh Poyarekar
2021-12-01 14:27 ` [PATCH v4 5/6] tree-object-size: Handle GIMPLE_CALL Siddhesh Poyarekar
2021-12-01 14:27 ` [PATCH v4 6/6] tree-object-size: Dynamic sizes for ADDR_EXPR Siddhesh Poyarekar
2021-12-18 12:35 ` [PATCH v5 0/4] __builtin_dynamic_object_size Siddhesh Poyarekar
2021-12-18 12:35 ` [PATCH v5 1/4] tree-object-size: Support dynamic sizes in conditions Siddhesh Poyarekar
2022-01-10 10:37 ` Jakub Jelinek
2022-01-10 23:55 ` Siddhesh Poyarekar
2021-12-18 12:35 ` [PATCH v5 2/4] tree-object-size: Handle function parameters Siddhesh Poyarekar
2022-01-10 10:50 ` Jakub Jelinek
2022-01-11 0:32 ` Siddhesh Poyarekar
2021-12-18 12:35 ` [PATCH v5 3/4] tree-object-size: Handle GIMPLE_CALL Siddhesh Poyarekar
2022-01-10 11:03 ` Jakub Jelinek
2021-12-18 12:35 ` [PATCH v5 4/4] tree-object-size: Dynamic sizes for ADDR_EXPR Siddhesh Poyarekar
2022-01-10 11:09 ` Jakub Jelinek
2022-01-04 3:24 ` [PING][PATCH v5 0/4] __builtin_dynamic_object_size Siddhesh Poyarekar
2022-01-11 8:57 ` [PATCH v6 " Siddhesh Poyarekar
2022-01-11 8:57 ` [PATCH v6 1/4] tree-object-size: Support dynamic sizes in conditions Siddhesh Poyarekar
2022-01-11 9:43 ` Jakub Jelinek
2022-01-11 9:44 ` Siddhesh Poyarekar
2022-01-11 8:57 ` [PATCH v6 2/4] tree-object-size: Handle function parameters Siddhesh Poyarekar
2022-01-11 9:44 ` Jakub Jelinek
2022-01-11 8:57 ` [PATCH v6 3/4] tree-object-size: Handle GIMPLE_CALL Siddhesh Poyarekar
2022-01-11 8:57 ` [PATCH v6 4/4] tree-object-size: Dynamic sizes for ADDR_EXPR Siddhesh Poyarekar
2022-01-11 9:47 ` Jakub Jelinek
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=20211215185253.GN2646553@tucnak \
--to=jakub@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=siddhesh@gotplt.org \
/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).