From: Segher Boessenkool <segher@kernel.crashing.org>
To: "Kewen.Lin" <linkw@linux.ibm.com>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>,
David Edelsohn <dje.gcc@gmail.com>,
Peter Bergner <bergner@linux.ibm.com>
Subject: Re: [PATCH] rs6000/test: Update some cases with -mdejagnu-tune
Date: Fri, 22 Jul 2022 13:17:13 -0500 [thread overview]
Message-ID: <20220722181712.GO25951@gate.crashing.org> (raw)
In-Reply-To: <11061a26-e6db-2f61-065e-b1c9a32d3181@linux.ibm.com>
On Fri, Jul 22, 2022 at 10:22:51AM +0800, Kewen.Lin wrote:
> on 2022/7/22 02:48, Segher Boessenkool wrote:
> > On Wed, Jul 20, 2022 at 05:31:11PM +0800, Kewen.Lin wrote:
> >> As PR106345 shows, some test cases should be updated with
> >> -mdejagnu-tune, since their test points are sensitive to
> >> rs6000_tune, such as: group_ending_nop, loop align (ic),
> >> float conversion cost etc.
> >
> > It does not make sense to require -mdejagnu-tune= if -mdejagnu-cpu= is
> > already given? What is the failure case?
> >
>
> I think cpu setting only sets tune setting when tune setting isn't
> explicitly provided as:
>
> if (rs6000_tune_index >= 0)
> tune_index = rs6000_tune_index;
> else if (cpu_index >= 0)
> rs6000_tune_index = tune_index = cpu_index;
>
> As PR106345 shows, GCC can use an explicit tune setting when it's
> configured, even if there is one "-mdejagnu-cpu=", it doesn't
> override the explicit given one, so we need one explicit
> "-mdejagnu-tune=".
And that is the problem. GCC's automatic setting is *not* an explicit
option, not given by the user. --with-tune= should not result in adding
an -mtune= option in the resulting compiler, it should not set command-
line options.
> Although the test case has adopted option "-mdejagnu-cpu=power7", but
> the configured "--with-tune-64=power9" takes effect and make it
> return align_loops instead of align_flags (5).
And it should not do that.
> >> This patch is to replace -mdejagnu-cpu with -mdejagnu-tune
> >> or append -mdejagnu-tune (keep the original -mdejagnu-cpu
> >> when it's required) accordingly.
> >
> > It is *always* required. Testcases with -mtune= but unspecified -mcpu=
> > make no sense.
>
> The loop_align.c testings made me think if we know the insn count for
> the loop on all cpus is in range (4,8] then the cpu setting doesn't matter.
Sure, it probably works without -mcpu=, but it does not make sense :-)
Only using -mtune= while not having -mcpu= serves no purpose in any
"normal" use, so we shouldn't do that in the testsuite either.
> > This should only make a difference if you have -mtune= in your
> > RUNTEST_FLAGS, and you shouldn't do silly things like that. I suspect
> > you see it in other cases, and those are actual bugs then, that need
> > actual fixing instead of sweeping under the carper.
>
> Unfortunately it's due to the explicit tune setting in configuration.
So that needs some actual fixes. Something in how --with-tune= works
is suboptimal?
> > The testcase suggests this is with a compiler configured with
> > --with-cpu= --with-tune=, which should just work, and -mcpu= should
> > override both of those!
>
> Unfortunately -mcpu= (-mdejagnu-cpu=) doesn't actually override here.
... or that.
Segher
next prev parent reply other threads:[~2022-07-22 18:18 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-07-20 9:31 Kewen.Lin
2022-07-21 18:48 ` Segher Boessenkool
2022-07-22 2:22 ` Kewen.Lin
2022-07-22 18:17 ` Segher Boessenkool [this message]
2022-07-22 18:53 ` Peter Bergner
2022-07-22 19:28 ` Peter Bergner
2022-07-25 6:09 ` Kewen.Lin
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=20220722181712.GO25951@gate.crashing.org \
--to=segher@kernel.crashing.org \
--cc=bergner@linux.ibm.com \
--cc=dje.gcc@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=linkw@linux.ibm.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).