From: "Roger Sayle" <roger@nextmovesoftware.com>
To: "'Jeff Law'" <jeffreyalaw@gmail.com>,
"'Andrew MacLeod'" <amacleod@redhat.com>
Cc: "'GCC Patches'" <gcc-patches@gcc.gnu.org>,
"'Richard Biener'" <richard.guenther@gmail.com>
Subject: RE: [PATCH] Ignore (possible) signed zeros in operands of FP comparisons.
Date: Fri, 18 Mar 2022 07:43:42 -0000 [thread overview]
Message-ID: <00dc01d83a9b$e50e2140$af2a63c0$@nextmovesoftware.com> (raw)
In-Reply-To: <de57d6a4-fc6e-de43-f41a-f63ce3039698@gmail.com>
Hi Jeff/Andrew,
> If you're going to do more work in this space, you might want to reach out to
> Aldy and Andrew to see if there's space for collaboration.
One (clever?) suggestion that I do have for ranger would be to add support for
an additional value_range_kind, VR_NONZEROBITS, which would be a variant of
VR_RANGE (for unsigned types?) and require very few changes to the existing
code. Just like VR_RANGE all values would lie in [MIN, MAX], so by default
treating this value_range_kind identically to VR_RANGE there should be no
visible changes, but the change in semantics is that MIN has the minimum bits
set, and MAX, the maximum bits set [equivalent to the RVAL and RMASK pairs
from CCP's bit_value_{bin,un}op]. Hence, the VR_NONZEROBITS range [2,7]
would represent the possible values {2, 3, 6, 7} rather than {2, 3, 4, 5, 6, 7}.
For a small number of bits, int_range can already handle this with multiple
irange spans, but adding this representation would allow the unification of the
bit-based propagation performed in tree-ssa-ccp with the range-value based
propagation performed in EVRP/ranger, allowing the clever forwards/backwards
functionality.
As Andrew's recent (partial) review points out, tracking the effect of operations
like BIT_XOR_EXPR on VR_RANGE is much more complicated than on the
proposed VR_NONZEROBITS.
Alas, I'm not the sort of contributor to make large infrastructure changes
myself, but if the above functionality were in place, I/the compiler would
be able to make use of it.
Cheers,
Roger
--
> -----Original Message-----
> From: Jeff Law <jeffreyalaw@gmail.com>
> Sent: 17 March 2022 23:28
> To: Roger Sayle <roger@nextmovesoftware.com>; 'Richard Biener'
> <richard.guenther@gmail.com>
> Cc: 'GCC Patches' <gcc-patches@gcc.gnu.org>
> Subject: Re: [PATCH] Ignore (possible) signed zeros in operands of FP
> comparisons.
>
>
> On 3/15/2022 2:03 AM, Roger Sayle wrote:
> >> -----Original Message-----
> >> From: Richard Biener <richard.guenther@gmail.com>
> >> Sent: 15 March 2022 07:29
> >> To: Roger Sayle <roger@nextmovesoftware.com>
> >> Cc: GCC Patches <gcc-patches@gcc.gnu.org>
> >> Subject: Re: [PATCH] Ignore (possible) signed zeros in operands of FP
> >> comparisons.
> >>
> >> On Mon, Mar 14, 2022 at 8:26 PM Roger Sayle
> >> <roger@nextmovesoftware.com> wrote:
> >>>
> >>> I've been wondering about the possible
> >>> performance/missed-optimization impact of my patch for PR
> >>> middle-end/98420 and similar IEEE correctness fixes that disable
> >>> constant folding optimizations when worrying
> >> about -0.0.
> >>> In the common situation where the floating point result is used by a
> >>> FP comparison, there's no distinction between +0.0 and -0.0, so some
> >>> HONOR_SIGNED_ZEROS optimizations that we'd usually disable, are safe.
> >>>
> >>> Consider the following interesting example:
> >>>
> >>> int foo(int x, double y) {
> >>> return (x * 0.0) < y;
> >>> }
> >>>
> >>> Although we know that x (when converted to double) can't be NaN or
> >>> Inf, we still worry that for negative values of x that (x * 0.0) may
> >>> be -0.0 and so perform the multiplication at run-time. But in this
> >>> case, the result of the comparison (-0.0 < y) will be exactly the
> >>> same as (+0.0 < y) for any y, hence the above may be safely constant
> >>> folded to "0.0 <
> >> y"
> >>> avoiding the multiplication at run-time.
> >>>
> >>> This patch has been tested on x86_64-pc-linux-gnu with make
> >>> bootstrap and make -k check with no new failures, and allows GCC to
> >>> continue to optimize cases that we optimized in GCC 11 (without regard to
> correctness).
> >>> Ok for mainline?
> >> Isn't that something that gimple-ssa-backprop.c is designed to
> >> handle? I wonder if you can see whether the signed zero speciality can be
> retrofitted there?
> >> It currently tracks "sign does not matter", so possibly another
> >> state, "sign of zero does not matter" could be introduced there.
> > Two questions. Would adding tracking of "sign of zero does not matter"
> > to gimple-ssa-backprop.c be suitable for stage4? Secondly, even if
> > gimple-ssa-backprop.c performed this kind of optimization, would that
> > be a reason not to support these transformations in match.pd? Perhaps
> > someone could open a missed optimization PR for backprop in Bugzilla,
> > but the above patch still needs to be reviewed on its own merits.
>
> Can't see how it's appropriate for stage4, but definitely interesting for gcc-13.
>
> It'd fit well into some of the Ranger plans too -- Aldy and Andrew have been
> talking about tracking the special FP values in Ranger. This is related, though
> not exactly the same since rather than tracking the special value, you're tracking
> if those special values actually matter. If you're going to do more work in this
> space, you might want to reach out to Aldy and Andrew to see if there's space
> for collaboration.
>
>
> >
> > Speaking of tree-ssa passes that could be improved, I was wondering
> > whether you could review my EVRP patch to fix regression PR/102950. Pretty
> please?
> > https://gcc.gnu.org/pipermail/gcc-patches/2022-February/589569.html
>
> I forwarded this to Aldy & Andrew. I suspect they missed it.
>
>
> >
> > Thanks (as always),
>
> No, thank you. I'm so happy to see you contributing to GCC regularly again!
>
>
> Jeff
>
> >
next prev parent reply other threads:[~2022-03-18 7:43 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-03-14 19:25 Roger Sayle
2022-03-15 7:29 ` Richard Biener
2022-03-15 8:03 ` Roger Sayle
2022-03-15 10:50 ` Richard Biener
2022-03-17 23:27 ` Jeff Law
2022-03-18 2:12 ` Andrew MacLeod
2022-03-18 7:43 ` Roger Sayle [this message]
2022-03-18 13:07 ` Andrew MacLeod
2022-03-18 18:07 ` Aldy Hernandez
2022-03-18 13:16 ` Andrew MacLeod
2022-03-18 16:01 ` Jeff Law
2022-03-18 18:33 ` Aldy Hernandez
2022-03-21 15:56 ` Aldy Hernandez
2022-03-26 18:52 ` Roger Sayle
2022-03-16 9:44 ` Richard Sandiford
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='00dc01d83a9b$e50e2140$af2a63c0$@nextmovesoftware.com' \
--to=roger@nextmovesoftware.com \
--cc=amacleod@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jeffreyalaw@gmail.com \
--cc=richard.guenther@gmail.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).