[Bug c++/20963] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[Bug c++/20963] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[marcus at jet dot franken dot de]

------- Additional Comments From marcus at jet dot franken dot de  2005-04-12 08:30 -------
Created an attachment (id=8604)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=8604&action=view)
xx.ii

mostly minimized extract from dosbox / dos_mem.cpp,

compile with:
g++ -c -O2

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug c++/20963] New: ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[marcus at jet dot franken dot de]

compiling dosbox results in this error (enable checking) on i386: 
 
g++ -O2 -c xx.ii  
xx.ii: In function 'void DOS_FreeProcessMemory(short unsigned int)': 
xx.ii:43: internal compiler error: tree check: expected value_handle, have 
addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

-- 
           Summary: ICE tree check: expected value_handle, have addr_expr in
                    value_exists_in_set_bitmap, at tree-ssa-pre.c:437
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: marcus at jet dot franken dot de
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.0 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dberlin at gcc dot gnu dot
                   |                            |org
          Component|c++                         |tree-optimization
           Keywords|                            |ice-on-valid-code
            Summary|ICE tree check: expected    |[4.0 Regression] ICE tree
                   |value_handle, have addr_expr|check: expected
                   |in                          |value_handle, have addr_expr
                   |value_exists_in_set_bitmap, |in
                   |at tree-ssa-pre.c:437       |value_exists_in_set_bitmap,
                   |                            |at tree-ssa-pre.c:437
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.0/4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-04-12 08:46 -------
Confirmed.
This started to fail after "4.0.0 20050225" which is after the branch.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2005-04-12 08:46:22
               date|                            |
            Summary|[4.0 Regression] ICE tree   |[4.0/4.1 Regression] ICE
                   |check: expected             |tree check: expected
                   |value_handle, have addr_expr|value_handle, have addr_expr
                   |in                          |in
                   |value_exists_in_set_bitmap, |value_exists_in_set_bitmap,
                   |at tree-ssa-pre.c:437       |at tree-ssa-pre.c:437


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.0/4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-04-12 08:49 -------
Note if the code was changed to used the normal offsetof, it would work.  Why do people try to use the 
non-normal version is behond me.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.0/4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-04-12 09:06 -------
More reduced testcase:
char mem_readb(int pt);
struct sMCB
{
  int type, psp_segment;
};
int f();
void DOS_FreeProcessMemory(int pt)
{
  int addr1, addr;
  while (true)
  {
    if (f())
    {
      addr = (unsigned int)((char*)&(((sMCB*)0)->psp_segment)-(char*)0);
      mem_readb(addr);
    }
    addr1 = (unsigned int)((char*)&(((sMCB*)0)->type)-(char*)0);
    if (mem_readb(addr1))
      return;
  }
};


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.0/4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-04-12 09:52 -------
There is always a bigger fish - an a smaller test case: 
 
struct sMCB { 
  unsigned char type; 
}; 
 
extern void foo (void); 
unsigned char mem_readb(char *pt) __attribute__((nothrow)); 
 
void DOS_FreeProcessMemory(unsigned short pspseg) { 
        while (1) { 
                if (pspseg) 
                        foo (); 
                char *addr = (char*)(&((sMCB*)0)->type); 
                if (mem_readb(addr)==0x5a) break; 
        } 
}; 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.0/4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-04-12 09:56 -------
In addition to ICEing, PRE is also being stupid about moving out a 
loop invariant expression: 
 
Found partial redundancy for expression (char *) &0B->type 
Skipping insertion of phi for partial redundancy: Looks like an induction 
variable 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.0/4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[dberlin at dberlin dot org]

------- Additional Comments From dberlin at gcc dot gnu dot org  2005-04-12 13:26 -------
Subject: Re:  [4.0/4.1 Regression] ICE tree
	check: expected value_handle, have addr_expr in value_exists_in_set_bitmap,
	at tree-ssa-pre.c:437

On Tue, 2005-04-12 at 09:56 +0000, steven at gcc dot gnu dot org wrote:
> ------- Additional Comments From steven at gcc dot gnu dot org  2005-04-12 09:56 -------
> In addition to ICEing, PRE is also being stupid about moving out a 
> loop invariant expression: 
>  
> Found partial redundancy for expression (char *) &0B->type 
> Skipping insertion of phi for partial redundancy: Looks like an induction 
> variable 
>  

This means nothing.
If it's really loop invariant, it will still be eliminated anyway (and
show up on the eliminated list) because the outside the loop expression
it *does* insert will have the right value to replace the
inside-the-loop version.





-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.0/4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-04-12 13:38 -------
FWIW, I think the problem might be that there are cases where  
force_gimple_operand generates more sub-expressions, but we do 
not add value handles for them: 
 
(gdb) 
1369      folded = fold (build (TREE_CODE (expr), TREE_TYPE (expr), 
(gdb) p expr->common.code 
$36 = NOP_EXPR 
(gdb) next 
1371      newexpr = force_gimple_operand (folded, &forced_stmts, false, NULL); 
(gdb) p debug_generic_expr(folded) 
(charD.3 *) &0B->typeD.1931 
$37 = void 
(gdb) next 
1372      if (forced_stmts) 
(gdb) p debug_generic_expr(newexpr) 
(charD.3 *) D.1971_2 
$38 = void 
(gdb) p debug_generic_stmt(forced_stmts) 
D.1970_10 = 0B; 
D.1971_2 = &D.1970_10->typeD.1931; 
 
$39 = void 
(gdb)     
 
We never add the value handles for the expressions in those two 
"forced_stmts", and we ICE when we look for a leader for one of 
them... 
 
But I'll leave it to Dan to figure out the details ;-) 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[dberlin at gcc dot gnu dot org]

------- Additional Comments From dberlin at gcc dot gnu dot org  2005-04-12 15:10 -------
That's not where the real problem lies, unfortunately, Steven.
I'm working on fixing this.

This is only a 4.1 regression, not a 4.0 regression.
It doesn't fail on the 4.0 branch (try it!)


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |dberlin at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2005-04-12 08:46:22         |2005-04-12 15:10:29
               date|                            |
            Summary|[4.0/4.1 Regression] ICE    |[4.1 Regression] ICE tree
                   |tree check: expected        |check: expected
                   |value_handle, have addr_expr|value_handle, have addr_expr
                   |in                          |in
                   |value_exists_in_set_bitmap, |value_exists_in_set_bitmap,
                   |at tree-ssa-pre.c:437       |at tree-ssa-pre.c:437
   Target Milestone|4.0.0                       |4.1.0
            Version|4.0.0                       |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[marcus at jet dot franken dot de]

------- Additional Comments From marcus at jet dot franken dot de  2005-04-12 15:19 -------
but it failed for me on a 4.0 branch based compiler (only some patches added) 
 
(current SUSE snapshot of gcc 4.0) 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[dberlin at dberlin dot org]

------- Additional Comments From dberlin at gcc dot gnu dot org  2005-04-12 16:55 -------
Subject: Re:  [4.1 Regression] ICE tree check:
	expected value_handle, have addr_expr in value_exists_in_set_bitmap, at
	tree-ssa-pre.c:437

On Tue, 2005-04-12 at 15:19 +0000, marcus at jet dot franken dot de
wrote:
> ------- Additional Comments From marcus at jet dot franken dot de  2005-04-12 15:19 -------
> but it failed for me on a 4.0 branch based compiler (only some patches added) 
>  
> (current SUSE snapshot of gcc 4.0) 
> 

Then it is one of your patches.

If you have patches to fold-const or tree-ssa-pre/tree-vn, it is likely
one of these.

I have tried this on a clean gcc 4.0 branch compiler and it does not
crash.




-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-04-12 17:00 -------
(In reply to comment #11)
> I have tried this on a clean gcc 4.0 branch compiler and it does not
> crash.

And I just did the same too, it does not crash on a 4.0 compiler even with --enable-checking=all.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-04-13 06:28 -------
Maybe 4.0 with the patch for PR20929 triggers the ICE. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-04-14 15:25 -------
Subject: Bug 20963

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	dberlin@gcc.gnu.org	2005-04-14 15:25:02

Modified files:
	gcc            : ChangeLog tree-ssa-pre.c 

Log message:
	2005-04-14  Daniel Berlin  <dberlin@dberlin.org>
	
	Fix PR tree-optimization/20963
	* tree-ssa-pre.c (compute_avail): Remove special case for
	TREE_INVARIANT.
	(create_expression_by_pieces): Add value numbers for forced out
	statements.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.8288&r2=2.8289
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-pre.c.diff?cvsroot=gcc&r1=2.74&r2=2.75



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963

[Bug tree-optimization/20963] [4.1 Regression] ICE tree check: expected value_handle, have addr_expr in value_exists_in_set_bitmap, at tree-ssa-pre.c:437

[dberlin at gcc dot gnu dot org]

------- Additional Comments From dberlin at gcc dot gnu dot org  2005-04-15 15:51 -------
Fixed

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20963