From: Richard Biener <rguenther@suse.de>
To: Tom de Vries <Tom_deVries@mentor.com>
Cc: Jakub Jelinek <jakub@redhat.com>,
"gcc-patches@gnu.org" <gcc-patches@gnu.org>
Subject: Re: [PATCH, 4/16] Implement -foffload-alias
Date: Wed, 16 Dec 2015 13:16:00 -0000 [thread overview]
Message-ID: <alpine.LSU.2.11.1512161410000.3571@t29.fhfr.qr> (raw)
In-Reply-To: <566EE3CA.2010804@mentor.com>
On Mon, 14 Dec 2015, Tom de Vries wrote:
> On 14/12/15 14:26, Richard Biener wrote:
> > On Sun, 13 Dec 2015, Tom de Vries wrote:
> >
> > > On 11/12/15 14:00, Richard Biener wrote:
> > > > On Fri, 11 Dec 2015, Tom de Vries wrote:
> > > >
> > > > > On 13/11/15 12:39, Jakub Jelinek wrote:
> > > > > > We simply have some compiler internal interface between the caller
> > > > > > and
> > > > > > callee of the outlined regions, each interface in between those has
> > > > > > its own structure type used to communicate the info;
> > > > > > we can attach attributes on the fields, or some flags to indicate
> > > > > > some
> > > > > > properties interesting from aliasing POV. We don't really need to
> > > > > > perform
> > > > > > full IPA-PTA, perhaps it would be enough to a) record somewhere in
> > > > > > cgraph
> > > > > > the relationship in between such callers and callees (for offloading
> > > > > > regions
> > > > > > we already have "omp target entrypoint" attribute on the callee and
> > > > > > a
> > > > > > singler caller), tell LTO if possible not to split those into
> > > > > > different
> > > > > > partitions if easily possible, and then just for these pairs perform
> > > > > > aliasing/points-to analysis in the caller and the result record
> > > > > > using
> > > > > > cliques/special attributes/whatever to the callee side, so that the
> > > > > > callee
> > > > > > (outlined OpenMP/OpenACC/Cilk+ region) can then improve its alias
> > > > > > analysis.
> > > > >
> > > > > Hi,
> > > > >
> > > > > This work-in-progress patch allows me to use IPA PTA information in
> > > > > the
> > > > > kernels pass group.
> > > > >
> > > > > Since:
> > > > > - I'm running IPA PTA before ealias, and IPA PTA does not interpret
> > > > > restrict, and
> > > > > - compute_may_alias doesn't run if IPA PTA information is present
> > > > > I needed to convince ealias to do the restrict clique/base annotation.
> > > > >
> > > > > It would be more logical to fit IPA PTA after ealias, but one is an
> > > > > IPA
> > > > > pass,
> > > > > the other a regular one-function pass, so I would have to split the
> > > > > containing
> > > > > pass groups pass_all_early_optimizations and
> > > > > pass_local_optimization_passes.
> > > > > I'll give that a try now.
> > > > >
> > >
> > > I've tried this approach, but realized that this changes the order in
> > > which
> > > non-openacc functions are processed in the compiler, so I've abandoned
> > > this
> > > idea.
> > >
> > > > > Any comments?
> > > >
> > > > I don't think you want to run IPA PTA before early
> > > > optimizations, it (and ealias) rely on some initial cleanup to
> > > > do anything meaningful with well-spent ressources.
> > > >
> > > > The local PTA "hack" also looks more like a waste of resources, but well
> > > > ... teaching IPA PTA to honor restrict might be an impossible task
> > > > though I didn't think much about it other than handling it only for
> > > > nonlocal_p functions (for others we should see all incoming args
> > > > if IPA PTA works optimally). The restrict tags will leak all over
> > > > the place of course and in the end no meaningful cliques may remain.
> > > >
> > >
> > > This patch:
> > > - moves the kernels pass group to the first position in the pass list
> > > after ealias where we're back in ipa mode
> > > - inserts an new ipa pass to contain the gimple pass group called
> > > pass_oacc_ipa
> > > - inserts a version of ipa-pta before the pass group.
> >
> > In principle I like this a lot, but
> >
> > + NEXT_PASS (pass_ipa_pta_oacc_kernels);
> > + NEXT_PASS (pass_oacc_ipa);
> > + PUSH_INSERT_PASSES_WITHIN (pass_oacc_ipa)
> >
> > I think you can put pass_ipa_pta_oacc_kernels into the pass_oacc_ipa
> > group and thus just "clone" ipa_pta?
>
> Done. But using a clone means using the same gate function, and that means
> that this pass_ipa_pta instance no longer runs by default for openacc by
> default.
>
> I've added enabling-by-default of fipa-pta for fopenacc in
> default_options_optimization to fix that.
Hmm, but that enables both IPA PTA passes then? I suppose that's ok,
and if not enabling the "late" IPA PTA you'd want to re-set
gimple_df->ipa_pta.
> > sub-passes of IPA passes can
> > be both ipa passes and non-ipa passes.
>
> Right. It does mean that I need yet another pass (pass_ipa_oacc_kernels) to do
> the IPA/non-IPA transition at pass/sub-pass boundary:
> ...
> NEXT_PASS (pass_ipa_oacc);
> PUSH_INSERT_PASSES_WITHIN (pass_ipa_oacc)
> NEXT_PASS (pass_ipa_pta);
> NEXT_PASS (pass_ipa_oacc_kernels);
> PUSH_INSERT_PASSES_WITHIN (pass_ipa_oacc_kernels)
> /* out-of-ipa */
> NEXT_PASS (pass_oacc_kernels);
> PUSH_INSERT_PASSES_WITHIN (pass_oacc_kernels)
> ...
>
> OK for stage3 if bootstrap and reg-test succeeds?
Ok.
Richard.
next prev parent reply other threads:[~2015-12-16 13:16 UTC|newest]
Thread overview: 133+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-11-09 15:35 [PATCH series, 16] Use parloops to parallelize oacc kernels regions Tom de Vries
2015-11-09 15:44 ` [PATCH, 1/16] Insert new exit block only when needed in transform_to_exit_first_loop_alt Tom de Vries
2015-11-11 10:50 ` Richard Biener
2015-11-09 15:45 ` [PATCH, 2/16] Make create_parallel_loop return void Tom de Vries
2015-11-11 10:50 ` Richard Biener
2015-11-09 15:51 ` [PATCH, 3/16] Ignore reduction clause on kernels directive Tom de Vries
2015-11-24 12:25 ` [PING][PATCH, " Tom de Vries
2016-01-18 14:24 ` [PING^2][PATCH, " Tom de Vries
2016-01-18 14:26 ` Jakub Jelinek
2015-11-09 16:10 ` [PATCH, 4/16] Implement -foffload-alias Tom de Vries
2015-11-11 10:53 ` Richard Biener
2015-11-11 11:01 ` Jakub Jelinek
2015-11-12 16:04 ` Tom de Vries
2015-11-13 8:46 ` Richard Biener
2015-11-13 11:03 ` Tom de Vries
2015-11-13 11:30 ` Richard Biener
2015-11-13 11:39 ` Jakub Jelinek
2015-11-21 12:24 ` Tom de Vries
2015-11-23 11:46 ` Richard Biener
2015-11-27 11:44 ` Tom de Vries
2015-11-27 12:14 ` Tom de Vries
2015-12-02 9:59 ` Jakub Jelinek
2016-03-14 13:16 ` Tom de Vries
2016-03-14 23:18 ` Tom de Vries
2015-12-02 9:46 ` Jakub Jelinek
2015-12-02 13:11 ` Tom de Vries
2015-12-11 12:45 ` Tom de Vries
2015-12-11 13:00 ` Richard Biener
2015-12-13 16:38 ` Tom de Vries
2015-12-14 13:26 ` Richard Biener
2015-12-14 15:44 ` Tom de Vries
2015-12-16 13:16 ` Richard Biener [this message]
2015-12-16 14:43 ` Tom de Vries
2015-12-17 12:03 ` [gomp4] " Thomas Schwinge
2015-12-03 11:53 ` Tom de Vries
2015-11-09 16:31 ` [PATCH, 5/16] Add in_oacc_kernels_region in struct loop Tom de Vries
2015-11-11 10:57 ` Richard Biener
2015-11-16 11:39 ` Tom de Vries
2015-11-16 12:41 ` Richard Biener
2015-11-16 11:39 ` Tom de Vries
2015-11-09 17:39 ` [PATCH, 6/16] Add pass_oacc_kernels Tom de Vries
2015-11-11 10:59 ` Richard Biener
2015-11-19 13:51 ` Tom de Vries
2015-11-24 12:17 ` Tom de Vries
2015-11-25 10:42 ` Richard Biener
2016-02-05 12:06 ` Use plain -fopenacc to enable OpenACC kernels processing (was: [PATCH, 6/16] Add pass_oacc_kernels) Thomas Schwinge
2016-02-10 14:40 ` Use plain -fopenacc to enable OpenACC kernels processing Thomas Schwinge
2016-02-15 16:54 ` Tom de Vries
2016-02-23 15:19 ` Thomas Schwinge
2015-11-09 18:14 ` [PATCH, 7/16] Add pass_dominator_oacc_kernels Tom de Vries
2015-11-11 11:05 ` Richard Biener
2015-11-16 12:04 ` Tom de Vries
2015-11-09 18:34 ` [PATCH, 8/16] Add pass_ch_oacc_kernels Tom de Vries
2015-11-11 20:29 ` Tom de Vries
2015-11-30 12:12 ` [gomp4] Use pass_ch instead of pass_ch_oacc_kernels (was: [PATCH, 8/16] Add pass_ch_oacc_kernels) Thomas Schwinge
2015-11-09 19:53 ` [PATCH, 9/16] Add pass_parallelize_loops_oacc_kernels Tom de Vries
2015-11-16 11:59 ` Tom de Vries
2015-11-24 12:27 ` Tom de Vries
2015-12-13 16:58 ` [PIING][PATCH, " Tom de Vries
2015-12-14 15:23 ` Richard Biener
2016-01-16 22:41 ` [Committed] Move pass_expand_omp_ssa out of pass_parallelize_loops Tom de Vries
2016-01-18 12:59 ` [Committed] Allow pass_parallelize_loops to be run outside the loop pipeline Tom de Vries
2016-01-18 13:07 ` [committed] Add oacc_kernels_p argument to pass_parallelize_loops Tom de Vries
2016-01-18 13:30 ` [committed] Add pass_parallelize_loops to pass_oacc_kernels Tom de Vries
2016-01-20 8:54 ` [committed] Add oacc_kernels_p argument to pass_parallelize_loops Thomas Schwinge
2016-01-20 10:31 ` Tom de Vries
2015-11-09 19:59 ` [PATCH, 10/16] Add pass_oacc_kernels pass group in passes.def Tom de Vries
2015-11-11 11:03 ` Richard Biener
2015-11-16 11:55 ` Tom de Vries
2015-11-16 12:45 ` Richard Biener
2015-11-16 23:21 ` Tom de Vries
2015-11-17 10:05 ` Richard Biener
2015-11-17 14:54 ` Tom de Vries
2015-11-17 15:18 ` Richard Biener
2015-11-17 15:39 ` Tom de Vries
2015-11-17 22:21 ` [PATCH, PR68373 ] Call scev_const_prop in pass_parallelize_loops::execute Tom de Vries
2015-11-19 9:36 ` Tom de Vries
2015-11-20 10:15 ` Richard Biener
2015-11-18 8:30 ` [PATCH, 10/16] Add pass_oacc_kernels pass group in passes.def Richard Biener
2015-11-18 16:22 ` Bernhard Reutner-Fischer
2015-11-20 12:53 ` [committed, trivial] Fix typo and trailing whitespace in dump-file strings in parloops Tom de Vries
2015-11-19 0:35 ` [PATCH, 10/16] Add pass_oacc_kernels pass group in passes.def Tom de Vries
2015-11-20 10:28 ` Richard Biener
2015-11-21 8:42 ` Tom de Vries
2015-11-23 11:31 ` Richard Biener
2015-11-23 15:53 ` Tom de Vries
2015-11-23 16:38 ` Richard Biener
2015-11-19 10:31 ` Tom de Vries
2015-11-20 10:37 ` Richard Biener
2015-11-20 13:27 ` Tom de Vries
2015-11-20 13:29 ` Richard Biener
2015-11-20 16:34 ` Tom de Vries
2015-11-23 10:11 ` Richard Biener
2015-11-24 12:22 ` Tom de Vries
2015-11-24 13:19 ` Richard Biener
2015-11-24 14:33 ` Tom de Vries
2015-11-24 14:36 ` Richard Biener
2015-11-24 15:05 ` Tom de Vries
2015-11-25 10:43 ` Richard Biener
2015-11-25 10:44 ` Richard Biener
2015-11-30 17:48 ` [gomp4] " Thomas Schwinge
2015-11-22 23:37 ` [PATCH] Don't reapply loops flags if unnecessary in loop_optimizer_init Tom de Vries
2015-11-23 10:33 ` Richard Biener
2015-11-23 11:27 ` Tom de Vries
2015-11-09 20:02 ` [PATCH, 11/16] Update testcases after adding kernels pass group Tom de Vries
2015-11-11 11:03 ` Richard Biener
2015-11-12 14:32 ` Tom de Vries
2015-11-12 14:43 ` Richard Biener
2015-11-12 15:42 ` David Malcolm
2015-11-13 9:44 ` Richard Biener
2015-11-09 20:06 ` [PATCH, 12/16] Handle acc loop directive Tom de Vries
2015-11-24 12:30 ` [PING][PATCH, " Tom de Vries
2016-01-18 14:27 ` [PING^2][PATCH, " Tom de Vries
2016-01-26 12:38 ` [PING^3][PATCH, " Tom de Vries
2016-01-26 12:50 ` Jakub Jelinek
2016-02-12 11:11 ` Tom de Vries
2016-02-22 10:55 ` Tom de Vries
2016-02-22 10:58 ` Jakub Jelinek
2016-02-29 3:27 ` Tom de Vries
2016-03-07 8:22 ` [PING][PATCH, " Tom de Vries
2016-03-14 6:21 ` [PING^2][PATCH, " Tom de Vries
2015-11-09 20:08 ` [PATCH, 13/16] Add c-c++-common/goacc/kernels-*.c Tom de Vries
2016-01-18 13:33 ` [committed] Add oacc kernels tests in goacc Tom de Vries
2015-11-09 20:09 ` [PATCH, 14/16] Add gfortran.dg/goacc/kernels-*.f95 Tom de Vries
2015-11-09 20:11 ` [PATCH, 15/16] Add libgomp.oacc-c-c++-common/kernels-*.c Tom de Vries
2016-01-18 13:39 ` [comitted] Add oacc kernels test in libgomp Tom de Vries
2016-03-09 9:18 ` [PATCH, 15/16] Add libgomp.oacc-c-c++-common/kernels-*.c Tom de Vries
2016-03-18 12:46 ` Scan for parallelization of the oacc kernels test-cases in gfortran.dg/goacc (was: [PATCH, 15/16] Add libgomp.oacc-c-c++-common/kernels-*.c) Thomas Schwinge
2016-04-05 9:13 ` Scan for parallelization of the oacc kernels test-cases in gfortran.dg/goacc Tom de Vries
2016-04-07 15:26 ` Thomas Schwinge
2015-11-09 20:12 ` [PATCH, 16/16] Add libgomp.oacc-fortran/kernels-*.f95 Tom de Vries
2016-03-09 9:19 ` Tom de Vries
2016-03-16 13:12 ` 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=alpine.LSU.2.11.1512161410000.3571@t29.fhfr.qr \
--to=rguenther@suse.de \
--cc=Tom_deVries@mentor.com \
--cc=gcc-patches@gnu.org \
--cc=jakub@redhat.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).