From: Ilya Verbin <iverbin@gmail.com>
To: Jakub Jelinek <jakub@redhat.com>
Cc: gcc-patches@gcc.gnu.org, Martin Liska <mliska@suse.cz>
Subject: Re: [hsa merge 07/10] IPA-HSA pass
Date: Fri, 15 Jan 2016 18:06:00 -0000 [thread overview]
Message-ID: <20160115180547.GC48907@msticlxl57.ims.intel.com> (raw)
In-Reply-To: <20160115164522.GA3017@tucnak.redhat.com>
On Fri, Jan 15, 2016 at 17:45:22 +0100, Jakub Jelinek wrote:
> On Fri, Jan 15, 2016 at 07:38:14PM +0300, Ilya Verbin wrote:
> > On Fri, Jan 15, 2016 at 17:09:54 +0100, Jakub Jelinek wrote:
> > > On Fri, Jan 15, 2016 at 05:02:34PM +0100, Martin Jambor wrote:
> > > > How do other accelerators cope with the situation when half of the
> > > > application is compiled with the accelerator disabled? (Would some of
> > > > their calls to GOMP_target_ext lead to abort?)
> > >
> > > GOMP_target_ext should never abort (unless internal error), worst case it
> > > just falls back into the host fallback.
> >
> > Wouldn't that lead to hard-to-find problems in case of nonshared memory?
> > I mean when someone expects that all target regions are executed on the device,
> > but in fact some of them are silently executed on the host with different data
> > environment.
>
> E.g. for HSA it really shouldn't matter, as it is shared memory accelerator.
> For XeonPhi we hopefully can offload anything.
As you said, if compilation of target image fails with ICE or somehow, host
fallback and offloading to other targets should still work:
https://gcc.gnu.org/ml/gcc-patches/2015-02/msg00951.html
That patch was not applied, but it can be simulated by -foffload=disable,
I've created a testcase:
$ cat main.c
#pragma omp declare target
int x;
#pragma omp end declare target
extern int foo ();
int main ()
{
int shared_mem = 0;
#pragma omp target map (alloc: x, shared_mem)
{
x = 10;
shared_mem = 1;
}
x = 20;
int r = foo ();
if (!shared_mem && r != 100)
__builtin_abort ();
return 0;
}
$ cat liba.c
#pragma omp declare target
extern int x;
#pragma omp end declare target
int foo ()
{
int r;
#pragma omp target map (from: r) map (alloc: x)
r = x * x;
return r;
}
$ gcc -fopenmp -fPIC -shared liba.c -o liba.so -foffload=disable
$ gcc -fopenmp -L. -la main.c
Currently it prints "libgomp: Target function wasn't mapped", but after this
change:
--- a/libgomp/target.c
+++ b/libgomp/target.c
@@ -1390,7 +1390,7 @@ gomp_get_target_fn_addr (struct gomp_device_descr *devicep,
splay_tree_key tgt_fn = splay_tree_lookup (&devicep->mem_map, &k);
gomp_mutex_unlock (&devicep->lock);
if (tgt_fn == NULL)
- gomp_fatal ("Target function wasn't mapped");
+ return NULL;
... it will fail at __builtin_abort, but without -foffload=disable it will pass.
-- Ilya
next prev parent reply other threads:[~2016-01-15 18:06 UTC|newest]
Thread overview: 77+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-01-13 17:55 [hsa merge 00/10] Merge of HSA branch Martin Jambor
2016-01-13 17:55 ` [hsa merge 03/10] HSA libgomp plugin Martin Jambor
2016-01-14 12:37 ` Jakub Jelinek
2016-01-13 17:55 ` [hsa merge 06/10] Pass manager changes Martin Jambor
2016-01-14 12:52 ` Jakub Jelinek
2016-01-13 17:55 ` [hsa merge 10/10] HSA register allocator Martin Jambor
2016-01-14 14:41 ` Jakub Jelinek
2016-01-15 15:09 ` Martin Jambor
2016-01-13 17:55 ` [hsa merge 04/10] Avoid extraneous remapping in copy_gimple_seq_and_replace_locals Martin Jambor
2016-01-13 18:02 ` Jakub Jelinek
2016-01-13 17:55 ` [hsa merge 01/10] Configury changes and new options Martin Jambor
2016-01-14 12:35 ` Jakub Jelinek
2016-02-02 12:53 ` Thomas Schwinge
2016-01-13 17:55 ` [hsa merge 02/10] Modifications to libgomp proper Martin Jambor
2016-01-20 13:37 ` Ilya Verbin
2016-01-20 14:48 ` Ilya Verbin
2016-01-20 14:55 ` Jakub Jelinek
2016-01-13 17:55 ` [hsa merge 07/10] IPA-HSA pass Martin Jambor
2016-01-14 12:59 ` Jakub Jelinek
2016-01-15 11:48 ` Martin Liška
2016-01-15 14:53 ` Martin Jambor
2016-01-15 15:01 ` Jakub Jelinek
2016-01-15 16:02 ` Martin Jambor
2016-01-15 16:10 ` Jakub Jelinek
2016-01-15 16:38 ` Ilya Verbin
2016-01-15 16:45 ` Jakub Jelinek
2016-01-15 18:06 ` Ilya Verbin [this message]
2016-01-20 18:54 ` Ilya Verbin
2016-01-21 11:39 ` Alexander Monakov
2016-01-22 10:19 ` Jakub Jelinek
2016-01-28 11:36 ` Partial Offloading (was: [hsa merge 07/10] IPA-HSA pass) Thomas Schwinge
2016-02-17 16:41 ` Ilya Verbin
2016-01-15 19:19 ` [hsa merge 07/10] IPA-HSA pass Alexander Monakov
2016-01-15 19:32 ` Jakub Jelinek
2016-01-15 19:46 ` Alexander Monakov
2016-01-15 19:55 ` Jakub Jelinek
2016-01-15 9:52 ` Jan Hubicka
2016-01-15 11:47 ` Martin Liška
2016-01-16 10:00 ` Jan Hubicka
2016-01-25 15:22 ` Martin Liška
2016-01-25 15:24 ` Jakub Jelinek
2016-01-25 23:41 ` Jan Hubicka
2016-01-27 12:38 ` Martin Liška
2016-01-13 18:11 ` [hsa merge 09/10] Majority of the HSA back-end Martin Jambor
2016-01-14 14:05 ` Jakub Jelinek
2016-01-15 15:08 ` Martin Jambor
2016-01-15 15:18 ` Jakub Jelinek
2016-01-15 23:49 ` Martin Jambor
2016-01-16 8:58 ` Jakub Jelinek
2016-01-16 9:23 ` Jakub Jelinek
2016-01-18 15:33 ` Martin Jambor
2016-01-18 15:48 ` Jakub Jelinek
2016-01-13 18:12 ` [hsa merge 08/10] HSAIL BRIG description header file Martin Jambor
2016-01-14 13:04 ` Jakub Jelinek
2016-01-15 1:19 ` Ian Lance Taylor
2016-01-15 10:03 ` Richard Biener
2016-01-15 10:14 ` Martin Jambor
2016-01-15 10:17 ` Richard Biener
2016-01-15 10:37 ` Jakub Jelinek
2016-01-15 12:03 ` Jakub Jelinek
2016-01-15 17:23 ` Martin Jambor
2016-01-16 11:43 ` Jakub Jelinek
2016-01-18 15:22 ` Martin Jambor
2016-01-15 17:53 ` Mike Stump
2016-01-15 20:23 ` Richard Biener
2016-02-26 18:58 ` Martin Jambor
2016-02-26 19:28 ` Mike Stump
2016-01-13 18:17 ` [hsa merge 05/10] OpenMP lowering/expansion changes (gridification) Martin Jambor
2016-01-14 12:52 ` Jakub Jelinek
2016-01-15 14:52 ` Martin Jambor
2016-01-14 14:22 ` Alexander Monakov
2016-01-19 10:45 ` [hsa merge 00/10] Merge of HSA branch Martin Jambor
2016-01-19 14:25 ` Richard Biener
2016-01-21 23:10 ` Gerald Pfeifer
2016-01-27 14:20 ` Martin Jambor
2016-01-28 15:18 ` Gerald Pfeifer
2016-02-02 14:20 ` 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=20160115180547.GC48907@msticlxl57.ims.intel.com \
--to=iverbin@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jakub@redhat.com \
--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).