Bug 47391 - [4.5 Regression] read from const volatile incorrectly eliminated
Summary: [4.5 Regression] read from const volatile incorrectly eliminated
Status: RESOLVED FIXED
Alias: None
Product: gcc
Classification: Unclassified
Component: tree-optimization (show other bugs)
Version: 4.6.0
: P1 normal
Target Milestone: 4.5.3
Assignee: Jakub Jelinek
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2011-01-21 05:51 UTC by John Regehr
Modified: 2011-04-07 18:39 UTC (History)
3 users (show)

See Also:
Host:
Target:
Build:
Known to work: 4.4.5, 4.6.0
Known to fail: 4.5.2
Last reconfirmed: 2011-01-21 08:24:13

Attachments
gcc46-pr47391.patch (547 bytes, patch)
2011-01-21 08:36 UTC, Jakub Jelinek 		Details | Diff
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description John Regehr 2011-01-21 05:51:52 UTC 
I'm using r169046 on x86 Linux.

  The problem is that this code:

const volatile int g_2 = 1;
int g_1 = 0;

void func_1 (void) {
   g_1 = g_2;
}

int main (void) {
  func_1();
  return 0;
}

  Compiled like this:

current-gcc -O2 -S small.c

  Gives this main():

main:
	movl	$1, g_1
	xorl	%eax, %eax
	ret

The load from g_2 shouldn't have been removed.  Oddly, the load is not eliminated from func_1(), nor is it eliminated from main if we inline the function by hand.
Comment 1 Jakub Jelinek 2011-01-21 08:24:13 UTC 
Buggy const_value_known_p.
Comment 2 Jakub Jelinek 2011-01-21 08:28:20 UTC 
4.5 actually miscompiles this too, 4.4 works.
Comment 3 Jakub Jelinek 2011-01-21 08:36:08 UTC 
Created attachment 23058 [details]
gcc46-pr47391.patch

Untested fix.
Comment 4 Eric Botcazou 2011-01-21 09:19:42 UTC 
> Untested fix.

tree_invariant_p_1 uses

  if (TREE_CONSTANT (t)
      || (TREE_READONLY (t) && !TREE_SIDE_EFFECTS (t)))
    return true;

rather than TREE_THIS_VOLATILE.
Comment 5 Jakub Jelinek 2011-01-21 09:29:46 UTC 
setup_one_parameter too, on the other side tree_add_const_value_attribute_for_decl, set_mem_attributes_minus_bitpos and
get_callee_fndecl test TREE_READONLY && !TREE_THIS_VOLATILE.
Comment 6 Jakub Jelinek 2011-01-21 12:57:55 UTC 
Author: jakub
Date: Fri Jan 21 12:57:52 2011
New Revision: 169084

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=169084
Log:
	PR tree-optimization/47391
	* varpool.c (const_value_known_p): Return false if
	decl is volatile.

	* gcc.dg/pr47391.c: New test.

Added:
    trunk/gcc/testsuite/gcc.dg/pr47391.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/varpool.c
Comment 7 Jakub Jelinek 2011-01-21 13:02:29 UTC 
Fixed on the trunk so far.
Comment 8 Jakub Jelinek 2011-04-07 18:22:54 UTC 
Author: jakub
Date: Thu Apr  7 18:22:50 2011
New Revision: 172112

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=172112
Log:
	PR tree-optimization/47391
	* tree-ssa-ccp.c (get_symbol_constant_value): Don't optimize
	if sym is volatile.

	Backport from mainline
	2011-01-21  Jakub Jelinek  <jakub@redhat.com>
 
	PR tree-optimization/47391
	* gcc.dg/pr47391.c: New test.

Added:
    branches/gcc-4_5-branch/gcc/testsuite/gcc.dg/pr47391.c
Modified:
    branches/gcc-4_5-branch/gcc/ChangeLog
    branches/gcc-4_5-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_5-branch/gcc/tree-ssa-ccp.c
Comment 9 Jakub Jelinek 2011-04-07 18:39:38 UTC 
Fixed.