From: Jakub Jelinek <jakub@redhat.com>
To: Uros Bizjak <ubizjak@gmail.com>, Kirill Yukhin <kirill.yukhin@gmail.com>
Cc: gcc-patches@gcc.gnu.org
Subject: [PATCH] Add peephole for -Os lock; dec (PR target/70821)
Date: Thu, 28 Apr 2016 15:42:00 -0000 [thread overview]
Message-ID: <20160428154237.GS26501@tucnak.zalov.cz> (raw)
Hi!
Optimizing atomic_fetch_add followed by comparison into just testing
the flags of the lock; sub is handled by a peephole2, which works usually
fine, except that for -Os we have another peephole2 that transforms
movl $-1, %reg into orl $-1, %reg and that causes the above mentioned
peephole2 not to trigger anymore.
Fixed by adding a peephole2 even for this case.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2016-04-28 Jakub Jelinek <jakub@redhat.com>
PR target/70821
* config/i386/sync.md (define_peephole2 *atomic_fetch_add_cmp<mode>):
Add new peephole2 where the first insn is *mov<mode>_or instead of
*mov<mode>_internal.
* gcc.target/i386/pr70821.c: New test.
--- gcc/config/i386/sync.md.jj 2016-01-04 14:55:56.000000000 +0100
+++ gcc/config/i386/sync.md 2016-04-28 09:40:28.265764880 +0200
@@ -467,6 +467,36 @@ (define_peephole2
(plus:SWI (match_dup 1)
(match_dup 2)))])])
+;; Likewise, but for the -Os special case of *mov<mode>_or.
+(define_peephole2
+ [(parallel [(set (match_operand:SWI 0 "register_operand")
+ (match_operand:SWI 2 "constm1_operand"))
+ (clobber (reg:CC FLAGS_REG))])
+ (parallel [(set (match_dup 0)
+ (unspec_volatile:SWI
+ [(match_operand:SWI 1 "memory_operand")
+ (match_operand:SI 4 "const_int_operand")]
+ UNSPECV_XCHG))
+ (set (match_dup 1)
+ (plus:SWI (match_dup 1)
+ (match_dup 0)))
+ (clobber (reg:CC FLAGS_REG))])
+ (set (reg:CCZ FLAGS_REG)
+ (compare:CCZ (match_dup 0)
+ (match_operand:SWI 3 "const_int_operand")))]
+ "peep2_reg_dead_p (3, operands[0])
+ && (unsigned HOST_WIDE_INT) INTVAL (operands[2])
+ == -(unsigned HOST_WIDE_INT) INTVAL (operands[3])
+ && !reg_overlap_mentioned_p (operands[0], operands[1])"
+ [(parallel [(set (reg:CCZ FLAGS_REG)
+ (compare:CCZ
+ (unspec_volatile:SWI [(match_dup 1) (match_dup 4)]
+ UNSPECV_XCHG)
+ (match_dup 3)))
+ (set (match_dup 1)
+ (plus:SWI (match_dup 1)
+ (match_dup 2)))])])
+
(define_insn "*atomic_fetch_add_cmp<mode>"
[(set (reg:CCZ FLAGS_REG)
(compare:CCZ
--- gcc/testsuite/gcc.target/i386/pr70821.c.jj 2016-04-28 09:56:06.239893613 +0200
+++ gcc/testsuite/gcc.target/i386/pr70821.c 2016-04-28 09:55:23.000000000 +0200
@@ -0,0 +1,16 @@
+/* PR target/70821 */
+/* { dg-do compile } */
+/* { dg-options "-Os" } */
+/* { dg-additional-options "-march=i686" { target ia32 } } */
+
+void bar (void);
+
+void
+foo (int *p)
+{
+ if (__atomic_sub_fetch (p, 1, __ATOMIC_SEQ_CST))
+ bar ();
+}
+
+/* { dg-final { scan-assembler "lock;? dec" } } */
+/* { dg-final { scan-assembler-not "lock;? xadd" } } */
Jakub
next reply other threads:[~2016-04-28 15:42 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-04-28 15:42 Jakub Jelinek [this message]
2016-04-28 16:30 ` Uros Bizjak
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=20160428154237.GS26501@tucnak.zalov.cz \
--to=jakub@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=kirill.yukhin@gmail.com \
--cc=ubizjak@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).