From: Thomas Schwinge <thomas@codesourcery.com>
To: Julian Brown <julian@codesourcery.com>, Jakub Jelinek <jakub@redhat.com>
Cc: <gcc-patches@gcc.gnu.org>,
Chung-Lin Tang <chunglin_tang@mentor.com>,
<Catherine_Moore@mentor.com>, <fortran@gcc.gnu.org>
Subject: Re: [PATCH] OpenACC 2.6 manual deep copy support (attach/detach)
Date: Fri, 18 Oct 2019 17:20:00 -0000 [thread overview]
Message-ID: <87pniuuhkj.fsf@euler.schwinge.homeip.net> (raw)
In-Reply-To: <20181210194137.27720f3e@squid.athome>
[-- Attachment #1: Type: text/plain, Size: 3871 bytes --]
Hi!
While reviewing
<http://mid.mail-archive.com/20191003163505.49997-2-julian@codesourcery.com>
"OpenACC reference count overhaul", I've just now stumbled over one thing
that originally was designed here:
On 2018-12-10T19:41:37+0000, Julian Brown <julian@codesourcery.com> wrote:
> On Fri, 7 Dec 2018 14:50:19 +0100
> Jakub Jelinek <jakub@redhat.com> wrote:
>
>> On Fri, Nov 30, 2018 at 03:41:09AM -0800, Julian Brown wrote:
>> > @@ -918,8 +920,13 @@ struct splay_tree_key_s {
>> > uintptr_t tgt_offset;
>> > /* Reference count. */
>> > uintptr_t refcount;
>> > - /* Dynamic reference count. */
>> > - uintptr_t dynamic_refcount;
>> > + /* Reference counts beyond those that represent genuine references in the
>> > + linked splay tree key/target memory structures, e.g. for multiple OpenACC
>> > + "present increment" operations (via "acc enter data") refering to the same
>> > + host-memory block. */
>> > + uintptr_t virtual_refcount;
>> > + /* For a block with attached pointers, the attachment counters for each. */
>> > + unsigned short *attach_count;
>> > /* Pointer to the original mapping of "omp declare target link" object. */
>> > splay_tree_key link_key;
>> > };
>>
>> This is something I'm worried about a lot, the nodes keep growing way
>> too much.
Is that just a would-be-nice-to-avoid, or is it an actual problem?
If the latter, can we maybe move some data into on-the-side data
structures, say an associative array keyed by [something suitable]? I
would assume that compared to actual host to/from device data movement
(or even lookup etc.), lookup of values from such an associative array
should be relatively cheap?
I'm bringing this up, because:
>> Is there a way to reuse some other field if it is of
>> certain kind?
>
> How about this -- it seems that the link_key is only used for OpenMP,
So, is that actually correct? Per my understanding, for the OpenACC
'link' clause we uses 'GOMP_MAP_LINK', which sets "omp declare target
link", and thus:
> and the attach count is only needed for OpenACC. So the obvious thing
> to do is probably to put those two together into a tagged union. The
> question is where to put the tag?
>
> Options are, I guess:
>
> 1. The high or low bits of the address. Potentially non-portable, ugly.
>
> 2. Or, the virtual refcount is also only needed for OpenACC, so we can
> reserve a magic value for that field to act as a tag.
>
> I've tried implementing the latter in the attached patch, and it seems
> to work OK.
... this is not actually feasible?
It's certainly possible that we're totally lacking sufficient testsuite
coverage, and that there are issues in the 'link' implementation
(<https://gcc.gnu.org/PR81689> "libgomp.c/target-link-1.c fails for
nvptx: #pragma omp target link not implemented" comes to mind
immediatelly, and certainly for OpenACC I used to be aware of additional
issues; I think I intended to use that mechanism for Fortran
'allocatable' with OpenACC 'declare'), but the libgomp handling to me
seems reasonable upon quick review -- just that we need to keep it alive
for OpenACC, too, unless I'm confused?
Simplifying the libgomp code to avoid the 'VREFCOUNT_LINK_KEY' toggle
flag, and not putting 'link_key' into an union together with
'attach_count', that should -- I hope -- resolve/obsolete some of the
questions raised in my late-night pre-Christmas 2018 review,
<https://gcc.gnu.org/ml/gcc-patches/2018-12/msg01620.html>, where I'm now
not sure yet whether all my questions have been addressed (or disputed,
but I didn't hear anything) in the recent -- split-out, thanks! --
version of this patch,
<http://mid.mail-archive.com/20191003163505.49997-2-julian@codesourcery.com>
"OpenACC reference count overhaul".
Grüße
Thomas
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 658 bytes --]
next prev parent reply other threads:[~2019-10-18 16:47 UTC|newest]
Thread overview: 81+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-11-10 17:11 [PATCH 0/3] " Julian Brown
2018-11-10 17:11 ` [PATCH 1/3] Host-to-device transfer coalescing & magic offset value self-documentation Julian Brown
2018-12-21 10:56 ` libgomp/target.c magic constants self-documentation Thomas Schwinge
2019-05-29 14:48 ` Thomas Schwinge
2018-11-10 17:11 ` [PATCH 2/3] Factor out duplicate code in gimplify_scan_omp_clauses Julian Brown
2018-12-18 14:16 ` Julian Brown
2018-12-18 14:50 ` Jakub Jelinek
2018-11-10 17:12 ` [PATCH 3/3] OpenACC 2.6 manual deep copy support (attach/detach) Julian Brown
2018-11-11 17:04 ` Bernhard Reutner-Fischer
2018-11-30 11:41 ` [PATCH] " Julian Brown
2018-12-03 17:03 ` Julian Brown
2018-12-07 13:50 ` Jakub Jelinek
2018-12-10 19:42 ` Julian Brown
2018-12-13 10:57 ` Jakub Jelinek
2018-12-14 19:00 ` Julian Brown
2018-12-18 12:25 ` Jakub Jelinek
2018-12-22 13:37 ` Thomas Schwinge
2019-10-18 17:20 ` Thomas Schwinge [this message]
2019-11-06 18:44 ` Julian Brown
2019-11-22 23:54 ` Julian Brown
2019-11-25 10:53 ` Tobias Burnus
2019-11-26 2:54 ` Julian Brown
2019-12-17 12:16 ` Thomas Schwinge
2019-12-17 17:28 ` [WIP] OpenACC 'acc_attach*', 'acc_detach*' runtime library routines (was: [PATCH] OpenACC 2.6 manual deep copy support (attach/detach)) Thomas Schwinge
2019-12-18 6:03 ` [PATCH 00/13] OpenACC 2.6 manual deep copy support Julian Brown
2019-12-18 6:03 ` [PATCH 01/13] Use aux struct in libgomp for infrequently-used/API-specific data Julian Brown
2019-12-18 6:03 ` [PATCH 03/13] OpenACC reference count consistency checking Julian Brown
2019-12-18 6:03 ` [PATCH 02/13] OpenACC reference count overhaul Julian Brown
2020-05-19 15:42 ` Thomas Schwinge
2020-06-04 18:13 ` [OpenACC] Use 'tgt' returned from 'gomp_map_vars' (was: [PATCH 02/13] OpenACC reference count overhaul) Thomas Schwinge
2020-05-19 15:49 ` [PATCH 02/13] OpenACC reference count overhaul Thomas Schwinge
2020-05-19 15:58 ` Thomas Schwinge
2020-06-25 11:03 ` Thomas Schwinge
2020-07-03 15:29 ` Thomas Schwinge
2019-12-18 6:04 ` [PATCH 08/13] OpenACC 2.6 deep copy: middle-end parts Julian Brown
2019-12-21 21:51 ` Thomas Schwinge
2019-12-18 6:04 ` [PATCH 06/13] OpenACC 2.6 deep copy: attach/detach API routines Julian Brown
2019-12-18 6:04 ` [PATCH 09/13] OpenACC 2.6 deep copy: C and C++ front-end parts Julian Brown
2019-12-24 5:05 ` Thomas Schwinge
2019-12-26 19:04 ` Jason Merrill
2021-06-10 11:03 ` Thomas Schwinge
2019-12-18 6:04 ` [PATCH 05/13] Factor out duplicate code in gimplify_scan_omp_clauses Julian Brown
2019-12-18 6:04 ` [PATCH 04/13] Use gomp_map_val for OpenACC host-to-device address translation Julian Brown
2019-12-18 6:05 ` [PATCH 11/13] OpenACC 2.6 deep copy: C and C++ execution tests Julian Brown
2020-06-04 18:43 ` Fix 'sizeof' usage in 'libgomp.oacc-c-c++-common/deep-copy-{7, 8}.c' (was: [PATCH 11/13] OpenACC 2.6 deep copy: C and C++ execution tests) Thomas Schwinge
2023-10-31 14:00 ` Add OpenACC 'acc_map_data' variant to 'libgomp.oacc-c-c++-common/deep-copy-8.c' " Thomas Schwinge
2019-12-18 6:05 ` [PATCH 13/13] Fortran polymorphic class-type support for OpenACC Julian Brown
2019-12-18 6:05 ` [PATCH 07/13] OpenACC 2.6 deep copy: libgomp parts Julian Brown
2019-12-21 23:37 ` Thomas Schwinge
2020-01-03 12:26 ` Julian Brown
2020-05-20 9:37 ` Thomas Schwinge
2020-06-05 16:23 ` [OpenACC 'exit data'] Simplify 'GOMP_MAP_STRUCT' handling (was: [PATCH 07/13] OpenACC 2.6 deep copy: libgomp parts) Thomas Schwinge
2020-06-05 16:36 ` [OpenACC 'exit data'] Strip 'GOMP_MAP_STRUCT' mappings " Thomas Schwinge
2020-05-20 14:52 ` [PATCH 07/13] OpenACC 2.6 deep copy: libgomp parts Thomas Schwinge
2020-05-20 19:11 ` Julian Brown
2020-06-04 18:35 ` [OpenACC] Repair/restore 'is_tgt_unmapped' checking (was: [PATCH 07/13] OpenACC 2.6 deep copy: libgomp parts) Thomas Schwinge
2020-06-04 18:53 ` [PATCH 07/13] OpenACC 2.6 deep copy: libgomp parts Thomas Schwinge
2020-06-05 10:39 ` Thomas Schwinge
2020-06-05 20:28 ` Julian Brown
2020-06-05 11:17 ` Thomas Schwinge
2020-06-05 20:31 ` Julian Brown
2020-06-09 10:41 ` OpenACC 'attach'/'detach' has no business affecting user-visible reference counting (was: [PATCH 07/13] OpenACC 2.6 deep copy: libgomp parts) Thomas Schwinge
2020-06-09 12:23 ` Julian Brown
2020-06-18 18:21 ` Julian Brown
2020-07-16 8:35 ` OpenACC 'attach'/'detach' has no business affecting user-visible reference counting Thomas Schwinge
2020-06-26 9:20 ` [PATCH 07/13] OpenACC 2.6 deep copy: libgomp parts Thomas Schwinge
2020-07-16 9:35 ` Thomas Schwinge
2020-07-16 21:21 ` Julian Brown
2020-07-17 9:12 ` Thomas Schwinge
2020-06-30 15:58 ` Thomas Schwinge
2019-12-18 6:05 ` [PATCH 12/13] OpenACC 2.6 deep copy: Fortran execution tests Julian Brown
2019-12-18 7:20 ` [PATCH 10/13] OpenACC 2.6 deep copy: Fortran front-end parts Julian Brown
2019-12-18 23:30 ` Tobias Burnus
2019-12-20 12:25 ` [committed] Improve is-coindexed check for OpenACC/OpenMP (was: [PATCH 10/13] OpenACC 2.6 deep copy: Fortran front-end parts) Tobias Burnus
2019-12-20 13:25 ` [PATCH 10/13] OpenACC 2.6 deep copy: Fortran front-end parts Tobias Burnus
2019-12-20 10:08 ` [patch,committed] Fix testsuite-fallout of OpenACC deep-copy patch (was: [PATCH 10/13] OpenACC 2.6 deep copy: Fortran front-end parts) Tobias Burnus
2019-12-18 18:24 ` [PATCH 00/13] OpenACC 2.6 manual deep copy support Thomas Schwinge
2019-12-20 1:21 ` Julian Brown
2019-12-20 14:36 ` OpenACC regression and development pace Thomas Koenig
2020-06-04 18:07 ` [OpenACC] XFAIL behavior of over-eager 'finalize' clause (was: [PATCH 00/13] OpenACC 2.6 manual deep copy support) Thomas Schwinge
2019-12-17 16:53 ` In 'libgomp/target.c', 'struct splay_tree_key_s', use 'struct splay_tree_aux' for infrequently-used or API-specific data (was: [PATCH] OpenACC 2.6 manual deep copy support (attach/detach)) Thomas Schwinge
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=87pniuuhkj.fsf@euler.schwinge.homeip.net \
--to=thomas@codesourcery.com \
--cc=Catherine_Moore@mentor.com \
--cc=chunglin_tang@mentor.com \
--cc=fortran@gcc.gnu.org \
--cc=gcc-patches@gcc.gnu.org \
--cc=jakub@redhat.com \
--cc=julian@codesourcery.com \
/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).