From: Jakub Jelinek <jakub@redhat.com>
To: Drew Ross <drross@redhat.com>
Cc: gcc-patches@gcc.gnu.org
Subject: Re: [PATCH] match.pd: Canonicalize (signed x << c) >> c [PR101955]
Date: Tue, 1 Aug 2023 23:36:38 +0200 [thread overview]
Message-ID: <ZMl65qj856ItuwQR@tucnak> (raw)
In-Reply-To: <20230801192033.432742-1-drross@redhat.com>
On Tue, Aug 01, 2023 at 03:20:33PM -0400, Drew Ross via Gcc-patches wrote:
> Canonicalizes (signed x << c) >> c into the lowest
> precision(type) - c bits of x IF those bits have a mode precision or a
> precision of 1. Also combines this rule with (unsigned x << c) >> c -> x &
> ((unsigned)-1 >> c) to prevent duplicate pattern. Tested successfully on
> x86_64 and x86 targets.
>
> PR middle-end/101955
>
> gcc/ChangeLog:
>
> * match.pd ((signed x << c) >> c): New canonicalization.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.dg/pr101955.c: New test.
> ---
> gcc/match.pd | 20 +++++++----
> gcc/testsuite/gcc.dg/pr101955.c | 63 +++++++++++++++++++++++++++++++++
> 2 files changed, 77 insertions(+), 6 deletions(-)
> create mode 100644 gcc/testsuite/gcc.dg/pr101955.c
>
> diff --git a/gcc/match.pd b/gcc/match.pd
> index 8543f777a28..62f7c84f565 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -3758,13 +3758,21 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> - TYPE_PRECISION (TREE_TYPE (@2)))))
> (bit_and (convert @0) (lshift { build_minus_one_cst (type); } @1))))
>
> -/* Optimize (x << c) >> c into x & ((unsigned)-1 >> c) for unsigned
> - types. */
> +/* For (x << c) >> c, optimize into x & ((unsigned)-1 >> c) for
> + unsigned x OR truncate into the precision(type) - c lowest bits
> + of signed x (if they have mode precision or a precision of 1) */
There should be . between ) and " */" above.
> (simplify
> - (rshift (lshift @0 INTEGER_CST@1) @1)
> - (if (TYPE_UNSIGNED (type)
> - && (wi::ltu_p (wi::to_wide (@1), element_precision (type))))
> - (bit_and @0 (rshift { build_minus_one_cst (type); } @1))))
> + (rshift (nop_convert? (lshift @0 INTEGER_CST@1)) @@1)
> + (if (wi::ltu_p (wi::to_wide (@1), element_precision (type)))
> + (if (TYPE_UNSIGNED (type))
> + (bit_and @0 (rshift { build_minus_one_cst (type); } @1))
This needs to be (convert @0) instead of @0, because now that there is
the nop_convert? in between, @0 could have different type than type.
I certainly see regressions on
gcc.c-torture/compile/950612-1.c
on i686-linux because of this:
/home/jakub/src/gcc/gcc/testsuite/gcc.c-torture/compile/950612-1.c:17:1: error: type mismatch in binary expression
long long unsigned int
long long int
long long unsigned int
_346 = _3 & 4294967295;
during GIMPLE pass: forwprop
/home/jakub/src/gcc/gcc/testsuite/gcc.c-torture/compile/950612-1.c:17:1: internal compiler error: verify_gimple failed
0x9018a4e verify_gimple_in_cfg(function*, bool, bool)
../../gcc/tree-cfg.cc:5646
0x8e81eb5 execute_function_todo
../../gcc/passes.cc:2088
0x8e8234c do_per_function
../../gcc/passes.cc:1687
0x8e82431 execute_todo
../../gcc/passes.cc:2142
Please submit a full bug report, with preprocessed source (by using -freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.
> + (if (INTEGRAL_TYPE_P (type))
> + (with {
> + int width = element_precision (type) - tree_to_uhwi (@1);
> + tree stype = build_nonstandard_integer_type (width, 0);
> + }
> + (if (width == 1 || type_has_mode_precision_p (stype))
> + (convert (convert:stype @0))))))))
just one space before == instead of two
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/pr101955.c
> @@ -0,0 +1,63 @@
> +/* { dg-do compile } */
The above line should be
/* { dg-do compile { target int32 } } */
because the test relies on 32-bit int, some targets have just
16-bit int.
Of course, unless you want to make the testcase more portable, by
using say
#define CHAR_BITS __CHAR_BIT__
#define INT_BITS (__SIZEOF_INT__ * __CHAR_BIT__)
#define LLONG_BITS (__SIZEOF_LONGLONG__ * __CHAR_BIT__)
and replacing all the 31, 24, 56 etc. constants with (INT_BITS - 1),
(INT_BITS - CHAR_BITS), (LLONG_BITS - CHAR_BITS) etc.
Though, it would still fail on some AVR configurations which have
(invalid for C) just 8-bit int, and the question is what to do with
that 16, because (INT_BITS - 2 * CHAR_BITS) is 0 on 16-bit ints, so
it would need to be (INT_BITS / 2) instead. C requires that
long long is at least 64-bit, so that is less problematic (no known
target to have > 64-bit long long, though theoretically possible).
> +/* { dg-options "-O2 -fdump-tree-optimized" } */
> +
Jakub
next prev parent reply other threads:[~2023-08-01 21:36 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-07-21 15:08 [PATCH] match.pd: Implement missed optimization (x << c) >> c -> -(x & 1) [PR101955] Drew Ross
2023-07-21 17:27 ` Andrew Pinski
2023-07-22 6:09 ` Jeff Law
2023-07-24 7:16 ` Richard Biener
2023-07-24 19:29 ` Drew Ross
2023-07-24 19:42 ` Jakub Jelinek
2023-07-25 6:54 ` Richard Biener
2023-07-25 19:25 ` Drew Ross
2023-07-25 19:43 ` Jakub Jelinek
2023-07-26 8:39 ` Richard Biener
2023-07-26 18:18 ` Drew Ross
2023-07-28 6:30 ` Richard Biener
2023-08-01 19:20 ` [PATCH] match.pd: Canonicalize (signed x << c) >> c [PR101955] Drew Ross
2023-08-01 21:36 ` Jakub Jelinek [this message]
[not found] <20230801191657.432375-1-drross@redhat.com>
2023-08-01 19:18 ` Drew Ross
[not found] <20230802162143.447292-1-drross@redhat.com>
2023-08-02 16:23 ` Drew Ross
2023-08-04 7:10 ` Jakub Jelinek
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=ZMl65qj856ItuwQR@tucnak \
--to=jakub@redhat.com \
--cc=drross@redhat.com \
--cc=gcc-patches@gcc.gnu.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).