From: Kyrill Tkachov <kyrylo.tkachov@arm.com>
To: GCC Patches <gcc-patches@gcc.gnu.org>
Cc: Richard Biener <rguenther@suse.de>
Subject: [PATCH][RFC][match.pd] optimize (X & C) == N when C is power of 2
Date: Fri, 24 Jul 2015 08:21:00 -0000 [thread overview]
Message-ID: <55B1F2C3.2000903@arm.com> (raw)
[-- Attachment #1: Type: text/plain, Size: 1225 bytes --]
Hi all,
This transformation folds (X % C) == N into
X & ((1 << (size - 1)) | (C - 1))) == N
for constants C and N where N is positive and C is a power of 2.
The idea is similar to the existing X % C -> X & (C - 1) transformation
for unsigned values but this time when we have the comparison we use the
C - 1 mask (all 1s for power of 2 N) orred with the sign bit.
At runtime, if X is positive then X & ((1 << (size - 1)) | (C - 1)))
calculates X % C, which is compared against the positive N.
If X is negative then X & ((1 << (size - 1)) | (C - 1))) doesn't calculate
X % C but since we also catch the set sign bit, it will never compare equal
to N (which is positive), so we still get the right comparison result.
I don't have much experience with writing match.pd patterns, so I appreciate
any feedback on how to write this properly.
Bootstrapped and tested on arm, aarch64, x86_64.
Thanks,
Kyrill
2015-07-24 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
* match.pd ((X % C) == N -> (X & ((1 << (size - 1)) | (C - 1))) == N):
Transform when N is positive and C is a power of 2.
2015-07-24 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
* gcc.dg/fold-mod-cmp-1.c: New test.
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: matchpd-mod-2-cmp.patch --]
[-- Type: text/x-patch; name=matchpd-mod-2-cmp.patch, Size: 2197 bytes --]
commit af785bad5cb32ef2bc35503ebbe65414b67ef8b1
Author: Kyrylo Tkachov <kyrylo.tkachov@arm.com>
Date: Tue Jul 14 18:20:28 2015 +0100
[match.pd] optimize (X & C) == N when C is power of 2
diff --git a/gcc/match.pd b/gcc/match.pd
index 9cf0278..02ad708 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -266,7 +266,9 @@ along with GCC; see the file COPYING3. If not see
/* Optimize TRUNC_MOD_EXPR by a power of two into a BIT_AND_EXPR,
i.e. "X % C" into "X & (C - 1)", if X and C are positive.
Also optimize A % (C << N) where C is a power of 2,
- to A & ((C << N) - 1). */
+ to A & ((C << N) - 1).
+ Optimize (X % C) == N into (X & ((1 << (size - 1)) | (C - 1))) == N
+ when C is signed, a power of 2 and N is positive. */
(match (power_of_two_cand @1)
INTEGER_CST@1)
(match (power_of_two_cand @1)
@@ -278,7 +280,17 @@ along with GCC; see the file COPYING3. If not see
|| tree_expr_nonnegative_p (@0))
&& tree_nop_conversion_p (type, TREE_TYPE (@3))
&& integer_pow2p (@2) && tree_int_cst_sgn (@2) > 0)
- (bit_and @0 (convert (minus @1 { build_int_cst (TREE_TYPE (@1), 1); }))))))
+ (bit_and @0 (convert (minus @1 { build_int_cst (TREE_TYPE (@1), 1); })))))
+
+ (simplify
+ (eq (mod @0 (INTEGER_CST@1)) INTEGER_CST@2)
+ (if (integer_pow2p (@1) && tree_expr_nonnegative_p (@2)
+ && tree_to_uhwi (@2) < tree_to_uhwi (@1))
+ (eq (bit_and @0
+ (bit_ior (minus @1 { build_int_cst (TREE_TYPE (@1), 1); })
+ { build_int_cst (type, 1 << (tree_to_uhwi (TYPE_SIZE (type)) - 1)); })
+ )
+ @2))))
/* X % Y is smaller than Y. */
(for cmp (lt ge)
diff --git a/gcc/testsuite/gcc.dg/fold-mod-cmp-1.c b/gcc/testsuite/gcc.dg/fold-mod-cmp-1.c
new file mode 100644
index 0000000..9dcf313
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/fold-mod-cmp-1.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-options "-fdump-tree-original" } */
+
+int
+modcmp (int a)
+{
+ return (a % 32) == 6;
+}
+
+/* The above should be simplified to (a & -2147483617) == 6. */
+/* { dg-final { scan-tree-dump "& -2147483617" "original" } } */
+/* { dg-final { scan-tree-dump "== 6" "original" } } */
next reply other threads:[~2015-07-24 8:09 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-07-24 8:21 Kyrill Tkachov [this message]
2015-07-24 8:36 ` Jakub Jelinek
2015-07-24 8:54 ` Kyrill Tkachov
2015-07-24 9:09 ` Richard Biener
2015-07-24 9:09 ` Kyrill Tkachov
2015-07-24 9:22 ` Ramana Radhakrishnan
2015-07-24 9:16 ` Jakub Jelinek
2015-07-24 9:31 ` Kyrill Tkachov
2015-07-24 9:36 ` Richard Biener
2015-07-24 10:11 ` Jakub Jelinek
2015-07-24 10:25 ` Ramana Radhakrishnan
2015-07-24 18:44 ` Jeff Law
2015-07-27 8:17 ` Richard Biener
2015-07-27 16:00 ` Jeff Law
2015-07-25 2:37 ` Segher Boessenkool
2015-07-27 8:23 ` Kyrill Tkachov
2015-07-27 15:54 ` Segher Boessenkool
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=55B1F2C3.2000903@arm.com \
--to=kyrylo.tkachov@arm.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=rguenther@suse.de \
/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).