From: David Edelsohn <dje.gcc@gmail.com>
To: Alan Lawrence <alan.lawrence@arm.com>,
"William J. Schmidt" <wschmidt@linux.vnet.ibm.com>
Cc: "gcc-patches@gcc.gnu.org" <gcc-patches@gcc.gnu.org>,
Segher Boessenkool <segher@kernel.crashing.org>
Subject: Re: [PATCH 10/11][RS6000] Migrate reduction optabs to reduc_..._scal
Date: Sat, 25 Oct 2014 00:08:00 -0000 [thread overview]
Message-ID: <CAGWvnykkUGK9YCdjKgC9xdCmWAxMRCbmd+d1Y5rYhWJSBfmN4Q@mail.gmail.com> (raw)
In-Reply-To: <544A40D1.1040605@arm.com>
On Fri, Oct 24, 2014 at 8:06 AM, Alan Lawrence <alan.lawrence@arm.com> wrote:
> This migrates the reduction patterns in altivec.md and vector.md to the new
> names. I've not touched paired.md as I wasn't really sure how to fix that
> (how do I vec_extractv2sf ?), moreover the testing I did didn't seem to
> exercise any of those patterns (iow: I'm not sure what would be an
> appropriate target machine?).
>
> I note the reduc_uplus_v16qi (which I've removed, as unsigned and signed
> addition should be equivalent) differed from reduc_splus_v16qi in using
> gen_altivec_vsum4ubs rather than gen_altivec_vsum4sbs. Testcases
> gcc.dg/vect/{slp-24-big-array.c,slp-24.c,vect-reduc-1char-big-array.c,vert-reduc-1char.c}
> thus produce assembly which differs from previously (only) in that
> "vsum4ubs" becomes "vsum4sbs". These tests are still passing so I assume
> this is OK.
>
> The combining of signed and unsigned addition also improves
> gcc.dg/vect/{vect-outer-4i.c,vect-reduc-1short.c,vect-reduc-dot-u8b.c,vect-reduc-pattern-1c-big-array.c,vect-reduc-pattern-1c.c}
> : these are now reduced using direct vector reduction, rather than with
> shifts as previously (because there was only a reduc_splus rather than the
> reduc_uplus these tests looked for).
>
> ((Side note: the RTL changes to vector.md are to match the combine patterns
> in vsx.md; now that we now longer depend upon combine to generate those
> patterns (as the optab outputs them directly), one might wish to remove the
> smaller pattern from vsx.md, and/or simplify the RTL. I theorize that a
> reduction of a two-element vector is just adding the first element to the
> second, so maybe to something like
>
> [(parallel [(set (match_operand:DF 0 "vfloat_operand" "")
> (VEC_reduc:V2DF
> (vec_select:DF
> (match_operand:V2DF 1 "vfloat_operand" "")
> (parallel [(const_int 1)]))
> (vec_select:DF
> (match_dup 1)
> (parallel [(const_int 0)]))))
> (clobber (match_scratch:V2DF 2 ""))])]
>
> but I think it's best for me to leave that to the port maintainers.))
>
> Bootstrapped and check-gcc on powerpc64-none-linux-gnu
> (gcc110.fsffrance.org, with thanks to the GCC Compile Farm).
>
> gcc/ChangeLog:
>
> * config/rs6000/altivec.md (reduc_splus_<mode>): Rename to...
> (reduc_plus_scal_<mode>): ...this, and rs6000_expand_vector_extract.
> (reduc_uplus_v16qi): Remove.
>
> * config/rs6000/vector.md (VEC_reduc_name): change "splus" to "plus"
> (reduc_<VEC_reduc_name>_v2df): Rename to...
> (reduc_<VEC_reduc_name>_scal_v2df): ...this, wrap VEC_reduc in a
> vec_select of element 1.
> (reduc_<VEC_reduc_name>_v4sf): Rename to...
> (reduc_<VEC_reduc_name>_scal_v4sf): ...this, wrap VEC_reduc in a
> vec_select of element 3, add scratch register.
>
This needs some input from Bill, but he will be busy with a conference
this week.
- David
next prev parent reply other threads:[~2014-10-24 23:49 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-10-24 11:57 [PATCH v2 0-6/11] Fix PR/61114, make direct vector reductions endianness-neutral Alan Lawrence
2014-10-24 11:58 ` [PATCH 7/11][ARM] Migrate to new reduc_plus_scal_optab Alan Lawrence
2014-11-03 17:32 ` Ramana Radhakrishnan
2014-10-24 12:01 ` [PATCH v2 0-6/11] Fix PR/61114, make direct vector reductions endianness-neutral Richard Biener
2014-10-24 12:05 ` [PATCH 8/11][ARM] Migrate to new reduc_[us](min|max)_scal_optab Alan Lawrence
2014-11-04 11:08 ` Ramana Radhakrishnan
2014-10-24 12:06 ` [PATCH 9/11][i386] Migrate reduction optabs to reduc_..._scal Alan Lawrence
2014-10-24 12:07 ` [PATCH 10/11][RS6000] " Alan Lawrence
2014-10-24 12:14 ` Alan Lawrence
2014-10-25 0:08 ` David Edelsohn [this message]
2014-11-03 17:51 ` Bill Schmidt
2014-11-06 16:44 ` Alan Lawrence
2014-11-06 18:57 ` Bill Schmidt
2014-11-07 10:09 ` Alan Lawrence
2014-11-10 22:39 ` Michael Meissner
2014-11-11 7:10 ` Segher Boessenkool
2014-11-12 1:54 ` Michael Meissner
2014-11-12 9:26 ` Segher Boessenkool
2014-11-12 19:20 ` Michael Meissner
2014-11-12 12:32 ` Alan Lawrence
2014-11-12 18:53 ` Alan Lawrence
2014-12-11 15:59 ` Ping: " Alan Lawrence
2014-12-11 18:37 ` Alan Lawrence
2014-10-24 12:12 ` [Protopatch 11/11][IA64] Migrate to reduc_(plus|min|max)_scal_v2df optab Alan Lawrence
2014-10-24 12:50 ` Alan Lawrence
2014-10-24 15:19 ` [PATCH v2 0-6/11] Fix PR/61114, make direct vector reductions endianness-neutral Matthew Fortune
2014-10-27 11:48 ` Richard Biener
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=CAGWvnykkUGK9YCdjKgC9xdCmWAxMRCbmd+d1Y5rYhWJSBfmN4Q@mail.gmail.com \
--to=dje.gcc@gmail.com \
--cc=alan.lawrence@arm.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=segher@kernel.crashing.org \
--cc=wschmidt@linux.vnet.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).