public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [patch] Fix PR rtl-optimization/102306
@ 2021-09-16  9:02 Eric Botcazou
  2021-09-16 15:22 ` Jeff Law
  0 siblings, 1 reply; 2+ messages in thread
From: Eric Botcazou @ 2021-09-16  9:02 UTC (permalink / raw)
  To: gcc-patches

[-- Attachment #1: Type: text/plain, Size: 844 bytes --]

Hi,

this is a duplication of volatile loads introduced during GCC 9 development by 
the new 2->2 mechanism of the RTL combiner.  There is already a substantial 
checking for volatile references in can_combine_p but it implicitly assumes 
that the combination reduces the number of instructions, which is of course 
not the case here.  So the fix teaches try_combine to abort the combination 
when it is about to make a copy of volatile references to preserve them.

Bootstrapped/regtested on x86-64/Linux, OK for mainline and release branches?


2021-09-16  Eric Botcazou  <ebotcazou@adacore.com>

	PR rtl-optimization/102306
	* combine.c (try_combine): Abort the combination if we are about
	to duplicate volatile references.


2021-09-16  Eric Botcazou  <ebotcazou@adacore.com>

	* gcc.target/sparc/20210916-1.c: New test.

-- 
Eric Botcazou

[-- Attachment #2: pr102306.diff --]
[-- Type: text/x-patch, Size: 686 bytes --]

diff --git a/gcc/combine.c b/gcc/combine.c
index 290a3667c65..892c834a160 100644
--- a/gcc/combine.c
+++ b/gcc/combine.c
@@ -3063,6 +3063,16 @@ try_combine (rtx_insn *i3, rtx_insn *i2, rtx_insn *i1, rtx_insn *i0,
       return 0;
     }
 
+  /* We cannot safely duplicate volatile references in any case.  */
+
+  if ((added_sets_2 && volatile_refs_p (PATTERN (i2)))
+      || (added_sets_1 && volatile_refs_p (PATTERN (i1)))
+      || (added_sets_0 && volatile_refs_p (PATTERN (i0))))
+    {
+      undo_all ();
+      return 0;
+    }
+
   /* Count how many auto_inc expressions there were in the original insns;
      we need to have the same number in the resulting patterns.  */
 

[-- Attachment #3: 20210916-1.c --]
[-- Type: text/x-csrc, Size: 289 bytes --]

/* { dg-do compile } */
/* { dg-require-effective-target ilp32 } */
/* { dg-options "-O -mcpu=v8" } */

extern void foo (void);

void test (volatile unsigned char *a) 
{ 
  char b = *a;
  if (!b)
    return;
  if (b & 2)
    foo ();
}

/* { dg-final { scan-assembler-times "ldub" 1 } } */

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [patch] Fix PR rtl-optimization/102306
  2021-09-16  9:02 [patch] Fix PR rtl-optimization/102306 Eric Botcazou
@ 2021-09-16 15:22 ` Jeff Law
  0 siblings, 0 replies; 2+ messages in thread
From: Jeff Law @ 2021-09-16 15:22 UTC (permalink / raw)
  To: Eric Botcazou, gcc-patches



On 9/16/2021 3:02 AM, Eric Botcazou wrote:
> Hi,
>
> this is a duplication of volatile loads introduced during GCC 9 development by
> the new 2->2 mechanism of the RTL combiner.  There is already a substantial
> checking for volatile references in can_combine_p but it implicitly assumes
> that the combination reduces the number of instructions, which is of course
> not the case here.  So the fix teaches try_combine to abort the combination
> when it is about to make a copy of volatile references to preserve them.
>
> Bootstrapped/regtested on x86-64/Linux, OK for mainline and release branches?
>
>
> 2021-09-16  Eric Botcazou  <ebotcazou@adacore.com>
>
> 	PR rtl-optimization/102306
> 	* combine.c (try_combine): Abort the combination if we are about
> 	to duplicate volatile references.
>
>
> 2021-09-16  Eric Botcazou  <ebotcazou@adacore.com>
>
> 	* gcc.target/sparc/20210916-1.c: New test.
OK
jeff


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2021-09-16 15:22 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-09-16  9:02 [patch] Fix PR rtl-optimization/102306 Eric Botcazou
2021-09-16 15:22 ` Jeff Law

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).