From: Richard Guenther <richard.guenther@gmail.com>
To: "Stubbs, Andrew" <Andrew_Stubbs@mentor.com>
Cc: Andrew Stubbs <andrew.stubbs@linaro.org>,
gcc-patches@gcc.gnu.org, patches@linaro.org
Subject: Re: [PATCH (3/7)] Widening multiply-and-accumulate pattern matching
Date: Fri, 24 Jun 2011 16:13:00 -0000 [thread overview]
Message-ID: <BANLkTi=1TRM7uHWxLD3Se=S-ibe0C13T-Q@mail.gmail.com> (raw)
In-Reply-To: <1A77B5B39081C241A68E6CF16983025F020906F6@EU1-MAIL.mgc.mentorg.com>
On Fri, Jun 24, 2011 at 3:46 PM, Stubbs, Andrew
<Andrew_Stubbs@mentor.com> wrote:
> On 24/06/11 09:28, Richard Guenther wrote:
>>> > To be clear, it only skips past NOP_EXPR. Is it not the case that what
>>> > you're describing would need a CONVERT_EXPR?
>> NOP_EXPR is the same as CONVERT_EXPR.
>
> Are you sure?
Yes, definitely. They are synonyms of each other (an unfinished merging
process), the usual check for them is via CONVERT_EXPR_P.
> I thought this was safe because the internals manual says:
>
> NOP_EXPR
> These nodes are used to represent conversions that do not require any
> code-generation ....
>
> CONVERT_EXPR
> These nodes are similar to NOP_EXPRs, but are used in those
> situations where code may need to be generated ....
Which is wrong (sorry).
> So, I tried this example:
>
> int
> foo (int a, short b, short c)
> {
> int bc = b * c;
> return a + (short)bc;
> }
>
> Both before and after my patch, GCC gives:
>
> mul r2, r1, r2
> sxtah r0, r0, r2
>
> (where, SXTAH means sign-extend the third operand from HImode to SImode
> and add to the second operand.)
>
> The dump after the widening_mult pass is:
>
> foo (int a, short int b, short int c)
> {
> int bc;
> int D.2018;
> short int D.2017;
> int D.2016;
> int D.2015;
> int D.2014;
>
> <bb 2>:
> D.2014_2 = (int) b_1(D);
> D.2015_4 = (int) c_3(D);
> bc_5 = b_1(D) w* c_3(D);
> D.2017_6 = (short int) bc_5;
> D.2018_7 = (int) D.2017_6;
> D.2016_9 = D.2018_7 + a_8(D);
> return D.2016_9;
>
> }
>
> Where you can clearly see that the addition has not been recognised as a
> multiply-and-accumulate.
>
> When I step through convert_plusminus_to_widen, I can see that the
> reason it has not matched is because "D.2017_6 = (short int) bc_5" is
> encoded with a CONVERT_EXPR, just as the manual said it would be.
A NOP_EXPR in this place would be valid as well. The merging hasn't
been completed and at least the C frontend still generates CONVERT_EXPRs
in some cases.
> So, according to the manual, and my (admittedly limited) experiments,
> skipping over NOP_EXPR does appear to be safe.
>
> But you say that it isn't safe. So now I'm confused. :(
>
> I can certainly add checks to make sure that the skipped operations
> actually don't make any important changes to the value, but do I need to?
Yes.
Thanks,
Richard.
> Andrew
>
next prev parent reply other threads:[~2011-06-24 15:47 UTC|newest]
Thread overview: 107+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-06-23 14:38 [PATCH (0/7)] Improve use of Widening Multiplies Andrew Stubbs
2011-06-23 14:39 ` [PATCH (1/7)] New optab framework for widening multiplies Andrew Stubbs
2011-07-09 15:38 ` Andrew Stubbs
2011-07-14 15:29 ` Andrew Stubbs
2011-07-22 13:01 ` Bernd Schmidt
2011-07-22 13:50 ` Andrew Stubbs
2011-07-22 14:01 ` Bernd Schmidt
2011-07-22 15:52 ` Andrew Stubbs
2011-08-19 14:41 ` Andrew Stubbs
2011-08-19 14:55 ` Richard Guenther
2011-08-19 15:07 ` Andrew Stubbs
2011-08-19 16:40 ` Andrew Stubbs
2011-06-23 14:41 ` [PATCH (2/7)] Widening multiplies by more than one mode Andrew Stubbs
2011-07-12 10:15 ` Andrew Stubbs
2011-07-12 11:05 ` Richard Guenther
2011-07-12 11:14 ` Richard Guenther
2011-07-12 11:38 ` Andrew Stubbs
2011-07-12 11:51 ` Richard Guenther
2011-07-21 19:51 ` Joseph S. Myers
2011-07-22 8:58 ` Andrew Stubbs
2011-07-14 14:17 ` Andrew Stubbs
2011-07-14 14:24 ` Richard Guenther
2011-08-19 14:45 ` Andrew Stubbs
2011-06-23 14:42 ` [PATCH (3/7)] Widening multiply-and-accumulate pattern matching Andrew Stubbs
2011-06-23 16:28 ` Richard Guenther
2011-06-24 8:14 ` Andrew Stubbs
2011-06-24 9:31 ` Richard Guenther
2011-06-24 14:08 ` Stubbs, Andrew
2011-06-24 16:13 ` Richard Guenther [this message]
2011-06-24 18:22 ` Stubbs, Andrew
2011-06-25 9:58 ` Richard Guenther
2011-06-28 11:32 ` Andrew Stubbs
2011-06-28 12:48 ` Richard Guenther
2011-06-28 16:37 ` Michael Matz
2011-06-28 16:48 ` Andrew Stubbs
2011-06-28 17:09 ` Michael Matz
2011-07-01 11:58 ` Stubbs, Andrew
2011-07-01 12:25 ` Richard Guenther
2011-07-04 14:23 ` Andrew Stubbs
2011-07-07 10:00 ` Richard Guenther
2011-07-07 10:27 ` Andrew Stubbs
2011-07-07 12:18 ` Andrew Stubbs
2011-07-07 12:34 ` Richard Guenther
2011-07-07 12:49 ` Richard Guenther
2011-07-08 12:55 ` Andrew Stubbs
2011-07-08 13:22 ` Richard Guenther
2011-07-11 17:01 ` Andrew Stubbs
2011-07-12 11:05 ` Richard Guenther
2011-08-19 14:50 ` Andrew Stubbs
2011-07-14 14:26 ` Andrew Stubbs
2011-07-19 0:36 ` Janis Johnson
2011-07-19 9:01 ` Andrew Stubbs
2011-07-01 12:33 ` Paolo Bonzini
2011-07-01 13:31 ` Stubbs, Andrew
2011-07-01 14:41 ` Paolo Bonzini
2011-07-01 14:55 ` Stubbs, Andrew
2011-07-01 15:54 ` Paolo Bonzini
2011-07-01 18:18 ` Stubbs, Andrew
2011-07-01 15:10 ` Stubbs, Andrew
2011-07-01 16:40 ` Bernd Schmidt
2011-06-23 21:55 ` Janis Johnson
2011-06-23 14:43 ` [PATCH (4/7)] Unsigned multiplies using wider signed multiplies Andrew Stubbs
2011-06-28 13:28 ` Andrew Stubbs
2011-06-28 14:49 ` Andrew Stubbs
2011-07-04 14:27 ` Andrew Stubbs
2011-07-07 10:10 ` Richard Guenther
2011-07-07 10:42 ` Andrew Stubbs
2011-07-07 11:08 ` Richard Guenther
2011-07-12 14:10 ` Andrew Stubbs
2011-07-14 14:28 ` Andrew Stubbs
2011-07-14 14:31 ` Richard Guenther
2011-08-19 14:51 ` Andrew Stubbs
2011-06-28 13:30 ` Paolo Bonzini
2011-06-23 14:44 ` [PATCH (5/7)] Widening multiplies for mis-matched mode inputs Andrew Stubbs
2011-06-28 15:44 ` Andrew Stubbs
2011-07-04 14:29 ` Andrew Stubbs
2011-07-07 10:11 ` Richard Guenther
2011-07-14 14:34 ` Andrew Stubbs
2011-07-14 14:35 ` Richard Guenther
2011-08-19 14:54 ` Andrew Stubbs
2011-06-23 14:51 ` [PATCH (6/7)] More widening multiply-and-accumulate pattern matching Andrew Stubbs
2011-06-28 15:49 ` Andrew Stubbs
2011-07-04 14:32 ` Andrew Stubbs
2011-07-07 10:20 ` Richard Guenther
2011-07-14 14:35 ` Andrew Stubbs
2011-07-14 14:41 ` Richard Guenther
2011-08-19 15:03 ` Andrew Stubbs
2011-10-13 16:25 ` Matthew Gretton-Dann
2011-06-23 14:54 ` [PATCH (7/7)] Mixed-sign multiplies using narrowest mode Andrew Stubbs
2011-06-28 17:02 ` Andrew Stubbs
2011-07-14 14:44 ` Andrew Stubbs
2011-07-14 14:48 ` Richard Guenther
2011-08-19 15:56 ` Andrew Stubbs
2011-06-25 16:14 ` [PATCH (0/7)] Improve use of Widening Multiplies Bernd Schmidt
2011-06-27 9:16 ` Andrew Stubbs
2011-07-18 14:34 ` [PATCH (8/7)] Fix a bug in multiply-and-accumulate Andrew Stubbs
2011-07-18 16:09 ` Richard Guenther
2011-07-21 13:48 ` Andrew Stubbs
2011-08-19 16:22 ` Andrew Stubbs
2011-07-21 13:14 ` [PATCH (9/7)] Widening multiplies with constant inputs Andrew Stubbs
2011-07-21 14:34 ` Richard Guenther
2011-07-22 12:28 ` Andrew Stubbs
2011-07-22 12:32 ` Andrew Stubbs
2011-07-22 12:34 ` Richard Guenther
2011-07-22 16:06 ` Andrew Stubbs
2011-08-19 16:24 ` Andrew Stubbs
2011-08-19 16:52 ` H.J. Lu
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='BANLkTi=1TRM7uHWxLD3Se=S-ibe0C13T-Q@mail.gmail.com' \
--to=richard.guenther@gmail.com \
--cc=Andrew_Stubbs@mentor.com \
--cc=andrew.stubbs@linaro.org \
--cc=gcc-patches@gcc.gnu.org \
--cc=patches@linaro.org \
/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).