From: Marcel Vollweiler <marcel@codesourcery.com>
To: Thomas Schwinge <thomas@codesourcery.com>,
Chung-Lin Tang <cltang@codesourcery.com>
Cc: <gcc-patches@gcc.gnu.org>, Jakub Jelinek <jakub@redhat.com>,
Tom de Vries <tdevries@suse.de>,
Andrew Stubbs <ams@codesourcery.com>,
Julian Brown <julian@codesourcery.com>,
Tobias Burnus <tobias@codesourcery.com>
Subject: Re: [PATCH] libgomp, OpenMP: Fix issue for omp_get_device_num on gfx targets.
Date: Tue, 18 Jan 2022 15:31:52 +0100 [thread overview]
Message-ID: <53521b17-79d1-6298-fcbf-e77398239422@codesourcery.com> (raw)
In-Reply-To: <87bl09cjtl.fsf@euler.schwinge.homeip.net>
Hi Thomas,
Am 18.01.2022 um 13:25 schrieb Thomas Schwinge:
> Hi!
>
> Maybe I'm just totally confused -- as so often ;-) -- but things seem
> strange here:
>
> On 2022-01-12T10:43:05+0100, Marcel Vollweiler <marcel@codesourcery.com> wrote:
>> Currently omp_get_device_num does not work on gcn targets with more than
>> one offload device. The reason is that GOMP_DEVICE_NUM_VAR
>
> I understand the 'GOMP_DEVICE_NUM_VAR' "macro indirection" is so that we
> define the actual symbol name ('__gomp_device_num') in one place
> ('libgomp/libgomp-plugin.h'), and then use it (via macro expansion) in
> several places, right?
Yes, as far as I understood.
>
>> is static in
>> icv-device.c and thus "__gomp_device_num" is not visible in the offload
>> image.
>
> That behavior seems correct -- but undesired indeed?
Good question. In contrast to nvptx I observed that __gomp_device_num is
not part of the offload image which we read out in
GOMP_OFFLOAD_load_image ("if (status != HSA_STATUS_SUCCESS)" in
libgomp/plugin/plugin-gcn.c returns false). I validated it with some
additional output in the if-branches.
>
>> This patch removes "static" such that "__gomp_device_num" is now part of
>> the offload image and can now be found in GOMP_OFFLOAD_load_image in the
>> plugin.
>
> That seems correct?
>
> Or, is there a reason to have it 'static', say, so that several such
> local variables can co-exist, instead of just one global one?
>
>> This is not an issue for nvptx. There, "__gomp_device_num" is in the
>> offload image even with "static".
>
> That's unexpected then, and should be looked into?
Actually, I don't see the reason for the different behaviour for nvptx.
I just tested that for nvptx the correct device number is returned by
omp_get_device_num on the device - also if we have more than one device.
>
> Still, should 'static' be removed here, too?
I wouldn't suggest unless it is really necessary? I mean, there we don't
have any issue. Although I aggree with Andrew that we could combine both
icv-device.c files into one common file.
Marcel
-----------------
Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht München, HRB 106955
next prev parent reply other threads:[~2022-01-18 14:32 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-07-23 10:21 [PATCH, libgomp, OpenMP 5.0] Implement omp_get_device_num Chung-Lin Tang
2021-07-23 10:39 ` Jakub Jelinek
2021-08-02 13:10 ` [PATCH, v2, " Chung-Lin Tang
2021-08-03 11:08 ` Jakub Jelinek
2021-08-03 12:07 ` Thomas Schwinge
2021-08-05 15:30 ` [PATCH, v3, libgomp, OpenMP 5.0, committed] " Chung-Lin Tang
2021-08-09 7:16 ` [PATCH, libgomp, OpenMP 5.0, OG11, " Chung-Lin Tang
2021-08-09 9:30 ` Julian Brown
2021-08-09 9:30 ` Julian Brown
2021-08-03 12:22 ` [PATCH, v2, libgomp, OpenMP 5.0] " Thomas Schwinge
2021-08-05 8:34 ` Chung-Lin Tang
2021-07-23 11:01 ` [PATCH, " Tobias Burnus
2021-08-02 13:09 ` Chung-Lin Tang
2022-01-04 9:28 ` [PATCH, OpenMP, libgomp, committed] Fix GOMP_DEVICE_NUM_VAR stringification error Chung-Lin Tang
2022-01-04 14:12 ` [committed] libgomp/testsuite: Improve omp_get_device_num() tests (was: Re: [PATCH, OpenMP, libgomp, committed] Fix GOMP_DEVICE_NUM_VAR stringification error) Tobias Burnus
2022-01-13 12:22 ` [committed] libgomp/testsuite: Improve omp_get_device_num() tests Thomas Schwinge
2022-01-12 9:43 ` [PATCH] libgomp, OpenMP: Fix issue for omp_get_device_num on gfx targets Marcel Vollweiler
2022-01-18 12:25 ` Thomas Schwinge
2022-01-18 13:55 ` Andrew Stubbs
2022-01-18 14:31 ` Marcel Vollweiler [this message]
2022-01-18 13:47 ` Andrew Stubbs
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=53521b17-79d1-6298-fcbf-e77398239422@codesourcery.com \
--to=marcel@codesourcery.com \
--cc=ams@codesourcery.com \
--cc=cltang@codesourcery.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jakub@redhat.com \
--cc=julian@codesourcery.com \
--cc=tdevries@suse.de \
--cc=thomas@codesourcery.com \
--cc=tobias@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).