[Bug middle-end/16237] New: [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[Bug middle-end/16237] New: [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

I am getting an ICE in avail_expr_eq:
/Volumes/UFS_Partition/pinskia/src/devel/objc-darwin/gcc/libiberty/regex.c: In function 
`byte_compile_range':
/Volumes/UFS_Partition/pinskia/src/devel/objc-darwin/gcc/libiberty/regex.c:4550: warning: signed 
and unsigned type in conditional expression
/Volumes/UFS_Partition/pinskia/src/devel/objc-darwin/gcc/libiberty/regex.c:4560: warning: signed 
and unsigned type in conditional expression
/Volumes/UFS_Partition/pinskia/src/devel/objc-darwin/gcc/libiberty/regex.c:4560: warning: signed 
and unsigned type in conditional expression
/Volumes/UFS_Partition/pinskia/src/devel/objc-darwin/gcc/libiberty/regex.c: In function 
`byte_regex_compile':
/Volumes/UFS_Partition/pinskia/src/devel/objc-darwin/gcc/libiberty/regex.c:2321: internal compiler 
error: in avail_expr_eq, at tree-ssa-dom.c:3356
Please submit a full bug report,

This is caused by:


(gdb) p debug_tree(((struct expr_hash_elt *)p1)->rhs)
 <array_ref 0x6bffd0
    type <integer_type 0x40f2845c unsigned char unsigned QI
        size <integer_cst 0x40f26200 constant invariant 8>
        unit size <integer_cst 0x40f26220 constant invariant 1>
        align 8 symtab 5 alias set -1 precision 8 min <integer_cst 0x837fc0 0> max <integer_cst 
0x837fe0 255>
        pointer_to_this <pointer_type 0x40fb883c>>
   
    arg 0 <var_decl 0x6c2c1c str
        type <array_type 0x6c2ba0 type <integer_type 0x40f2845c unsigned char>
            BLK
            size <integer_cst 0x40fb7c00 constant invariant 1024>
            unit size <integer_cst 0x40fb7be0 constant invariant 128>
            align 8 symtab 0 alias set -1 domain <integer_type 0x41815aa8>>
        used BLK file /Volumes/UFS_Partition/pinskia/src/devel/objc-darwin/gcc/libiberty/regex.c line 
3484 size <integer_cst 0x40fb7c00 1024> unit size <integer_cst 0x40fb7be0 128>
        align 8 context <function_decl 0x63407c byte_regex_compile>>
    arg 1 <integer_cst 0x6cfae0 type <integer_type 0x40f2864c int> constant invariant 0>>
$2 = void
(gdb) p debug_tree(((struct expr_hash_elt *)p2)->rhs)
 <array_ref 0x833344
    type <integer_type 0x40f2845c unsigned char unsigned QI
        size <integer_cst 0x40f26200 constant invariant 8>
        unit size <integer_cst 0x40f26220 constant invariant 1>
        align 8 symtab 5 alias set -1 precision 8 min <integer_cst 0x837fc0 0> max <integer_cst 
0x837fe0 255>
        pointer_to_this <pointer_type 0x40fb883c>>
   
    arg 0 <var_decl 0x6c2c1c str
        type <array_type 0x6c2ba0 type <integer_type 0x40f2845c unsigned char>
            BLK
            size <integer_cst 0x40fb7c00 constant invariant 1024>
            unit size <integer_cst 0x40fb7be0 constant invariant 128>
            align 8 symtab 0 alias set -1 domain <integer_type 0x41815aa8>>
        used BLK file /Volumes/UFS_Partition/pinskia/src/devel/objc-darwin/gcc/libiberty/regex.c line 
3484 size <integer_cst 0x40fb7c00 1024> unit size <integer_cst 0x40fb7be0 128>
        align 8 context <function_decl 0x63407c byte_regex_compile>>
    arg 1 <integer_cst 0x6d5960 type <integer_type 0x41815aa8> constant invariant 0>
    arg 2 <integer_cst 0x40f26880 type <integer_type 0x40f736c8 long unsigned int> constant 
invariant 0>
    arg 3 <integer_cst 0x40f267e0 type <integer_type 0x40f736c8 long unsigned int> constant 
invariant 1>>
$3 = void
(gdb)

-- 
           Summary: [3.5 Regression] ICE caused by Kenner's big untested
                    (regression causing) patch
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code, build
          Severity: critical
          Priority: P2
         Component: middle-end
        AssignedTo: kenner at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-27 21:24 -------
Please revert your patch as this is a boostrap failure causing unless you can figure out how to fix this 
bug.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
 GCC target triplet|                            |powerpc-apple-darwin


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-27 21:27 -------
Note this is blocking me from getting anywork done today, please fix soon otherwise I will revert the 
patch for you because the patch seems like it was not tested at all and was crazierly big and should not 
have been applied in one big chunk.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-06-27 21:28 -------
Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    Please revert your patch as this is a boostrap failure causing unless
    you can figure out how to fix this bug.

Which patch?  There have been quite a number of updates to the orignal
patch, both by myself and RTH.  I don't see a bootstrap problem on my
machine, so I'll need more detailed instructions on how to reproduce this.

I wonder if this was the hash table problem RTH mentioned, but if so, what
he did should have fixed it.  However, your tree dump suggested that you
aren't using the latest sources since we no longer fill in operands 2 and
3 for an ARRAY_REF if they are constant and that's what the dump shows.

Can you verify that you have the latest sources?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-27 21:29 -------
This is the latest sources I just updated this morning.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-27 21:30 -------
you are asking which patch, well the biggest one you applied in the last year:
2004-06-21  Richard Kenner  <kenner@vlsi1.ultra.nyu.edu>

        * Makefile.in (gimplify.o): Add cgraph.h.
        * alias.c (adjust_offset_for_component_ref): Use
        component_ref_field_offset.
        * c-decl.c (build_array_declarator): Add news args for ARRAY_REF.
        * c-gimplify.c (gimplify_expr_stmt): Use alloc_stmt_list.
        (gimplify_decl_stmt): Call gimplify_type_sizes for type.
        For decl, call gimplify_one_sizepos and use statement list.
        (gimplify_compound_literal_expr): New arg PRE_P.
        Add statement to PRE_P list and return DECL.
        (c_gimplify_expr, case COMPOUND_LITERAL_EXPR): Add arg to
        gimplify_compound_literal_expr.
        * c-tree.h (getdecls): Deleted.
        * c-typeck.c (build_component_ref): Add operand for COMPONENT_REF.
        (build_array_ref): Add two operands for ARRAY_REF.
        (build_unary_op): Set TREE_INVARIANT and TREE_CONSTANT for
        COMPOUND_LITERAL_EXPR.
        * coverage.c (tree_coverage_counter_ref): Add new operands
        for ARRAY_REF.
        * emit-rtl.c (component_ref_for_mem_expr): Add new operand
        for COMPONENT_REF.
        (set_mem_attributes_minus_bitpos): Use array_ref_low_bound
        and array_ref_element_size.
        (widen_memory_access):Use component_ref_field_offset.
        * explow.c (update_nonlocal_goto_save_area): Add two operands
        for ARRAY_REF.
        * expr.c (array_ref_element_size, array_ref_low_bound): New functions.
        (component_ref_field_offset): Likewise.
        (get_inner_reference): Use them.
        (expand_expr_real_1, case ARRAY_REF): Use array_ref_low_bound.
        * fold-const.c (fold, case EQ_EXPR): Properly handle DECL_SIZE.
        (fold_read_from_constant_string): Use array_ref_low_bound.
        Verify that result is a character type.
        (build_fold_indirect_ref): Add two operands for ARRAY_REF.
        * function.c (expand_function_start): Likewise.
        * gimple-low.c (expand_var_p): Delete duplicated line.
        * gimplify.c: Add static decls for local functions.
        (cgraph.h): Now included.
        (create_tmp_var): Remove check for ARRAY_TYPE.
        (copy_if_shared_r): Look at bounds and sizes of types.
        (build_and_jump): Return alloc_stmt_list instead of build_empty_stmt.
        (gimplify_exit_expr, shortcut_cond_expr): Likewise.
....

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-06-27 21:33 -------
Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    This is the latest sources I just updated this morning.

Then can you please set a breakpoint and see why operands 2 and 3
of that ARRAY_REF are getting filled in by gimplifcation?

As I said, I've been bootstrapped regularly all week, as are other targets,
so this is something target-specific.

You already gave me the target triple, so if you could send me the .i file
I can build for that target and see if I reproduce it.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-27 21:36 -------
Well I cannot give you the preprocessed source as it is not reproducable with -save-temps.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-27 21:44 -------
I will also note RTH only fixed it for COMPONENT_REFs and not ARRAY_REFs see <http://gcc.gnu.org/
ml/gcc-patches/2004-06/msg01906.html>.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-06-27 21:45 -------
Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    Well I cannot give you the preprocessed source as it is not
    reproducable with -save-temps.

That's certainly consistent with the sort of thing RTH was looking at.

Can you show me what both rhs1 and rhs2 are at that point?  Which is
the one you dumped?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-27 21:47 -------
They are both there already.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at physics dot uc dot edu]

------- Additional Comments From pinskia at physics dot uc dot edu  2004-06-27 21:51 -------
Subject: Re:  [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

> 
>     I will also note RTH only fixed it for COMPONENT_REFs and not
>     ARRAY_REFs see <http://gcc.gnu.org/ ml/gcc-patches/2004-06/msg01906.html>.
> 
> Well, I thought he did both, but that code has been changed a few times
> since then.  Right now, it's certainly doing both cases and only
> filling in the operand if not a constant.
> 
> Try this: if we understand the issue properly, this should fix it. Let
> me know.  But I'd also like to know why you're seeing them filled in!

that will not work as my problem is with ARRAY_REF and not COMPONENT_REF.

Andrew

> 
> *** fold-const.c.old	Sun Jun 27 17:45:23 2004
> --- fold-const.c	Sun Jun 27 17:46:59 2004
> *************** operand_equal_p (tree arg0, tree arg1, u
> *** 2483,2487 ****
>   				  TREE_OPERAND (arg1, 0), flags);
>   
> - 	case COMPONENT_REF:
>   	case ARRAY_REF:
>   	case ARRAY_RANGE_REF:
> --- 2483,2486 ----
> *************** operand_equal_p (tree arg0, tree arg1, u
> *** 2489,2493 ****
>   				   TREE_OPERAND (arg1, 0), flags)
>   		  && operand_equal_p (TREE_OPERAND (arg0, 1),
> ! 				      TREE_OPERAND (arg1, 1), flags));
>   
>   	case BIT_FIELD_REF:
> --- 2488,2506 ----
>   				   TREE_OPERAND (arg1, 0), flags)
>   		  && operand_equal_p (TREE_OPERAND (arg0, 1),
> ! 				      TREE_OPERAND (arg1, 1), flags)
> ! 		  && operand_equal_p (TREE_OPERAND (arg0, 2),
> ! 				      TREE_OPERAND (arg1, 2), flags)
> ! 		  && operand_equal_p (TREE_OPERAND (arg0, 3),
> ! 				      TREE_OPERAND (arg1, 3), flags));
> ! 
> ! 
> ! 	case COMPONENT_REF:
> ! 	  return (operand_equal_p (TREE_OPERAND (arg0, 0),
> ! 				   TREE_OPERAND (arg1, 0), flags)
> ! 		  && operand_equal_p (TREE_OPERAND (arg0, 1),
> ! 				      TREE_OPERAND (arg1, 1), flags)
> ! 		  && operand_equal_p (TREE_OPERAND (arg0, 2),
> ! 				      TREE_OPERAND (arg1, 2), flags));
> ! 
>   
>   	case BIT_FIELD_REF:
> 
> 



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at physics dot uc dot edu]

------- Additional Comments From pinskia at physics dot uc dot edu  2004-06-28 00:16 -------
Subject: Re:  [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

> 
>     that will not work as my problem is with ARRAY_REF and not COMPONENT_REF.
> 
> The first piece fixes the ARRAY_REF case (the label is outside of
> the range patched) and the second the COMPONENT_REF case.
> 
> I now see both dumps in the PR and indeed one has operands 2 and 3 filled in
> and one doesn't.  I think I now understand where the one that does comes
> from.
> 
> But the patch I just sent assumed that operand_equal_p handled NULL.
> It doesn't.  So this is what's needed.  It's a kludge, though, but
> it should get you going.  I'll try to get the correct patch tomorrow.

But does this not cause a performance regression.

Andrew


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-06-28 00:26 -------
Subject: Re:  [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    But does this not cause a performance regression.

It adds a couple of lines of code, but nothing likely measurable: there
is a hash table in use here, after all.

Nevertheless, I'd rather handle the zero case in the ARRAY_REF and
COMPONENT_REF case instead of all cases.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at physics dot uc dot edu]

------- Additional Comments From pinskia at physics dot uc dot edu  2004-06-28 01:38 -------
Subject: Re:  [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

> 
>     But does this not cause a performance regression.
> 
> It adds a couple of lines of code, but nothing likely measurable: there
> is a hash table in use here, after all.
> 
> Nevertheless, I'd rather handle the zero case in the ARRAY_REF and
> COMPONENT_REF case instead of all cases.

I was talking about code generation rather than compile time.




-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-06-28 01:43 -------
Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    I was talking about code generation rather than compile time.

Potentially there will indeed be some cases where duplicate array
references are missed at the tree level (I don't think this can happen
for component references), but hopefully they'll get caught at the RTL
level.  This isn't meant to be a permanent situation, of course.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-02 00:18 -------
Note I now see this on "make -j5 bootstrap".

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-12 06:06 -------
*** Bug 16488 has been marked as a duplicate of this bug. ***

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |billingd at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-12 06:07 -------
Confirming based two another reports.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
 GCC target triplet|powerpc-apple-darwin        |
   Last reconfirmed|0000-00-00 00:00:00         |2004-07-12 06:07:15
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-16 19:39 -------
*** Bug 16594 has been marked as a duplicate of this bug. ***

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gdr at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[gdr at gcc dot gnu dot org]

------- Additional Comments From gdr at gcc dot gnu dot org  2004-07-16 20:11 -------
(In reply to comment #15)
> Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested
(regression causing) patch
> 
>     I was talking about code generation rather than compile time.
> 
> Potentially there will indeed be some cases where duplicate array
> references are missed at the tree level (I don't think this can happen
> for component references), but hopefully they'll get caught at the RTL
> level.  This isn't meant to be a permanent situation, of course.
> 

Kenner --

  This regression causing patch has been in there for more than two weeks now.
When do you plan to look into it?  Please have a look at the PR 16594
I made that is reported duplicate of this.  More than two weeks is
more than what our usual policy tolerates.

Thanks,



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-07-17 14:00 -------
Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    This regression causing patch has been in there for more than two
    weeks now.  When do you plan to look into it?  Please have a look at
    the PR 16594 I made that is reported duplicate of this.  More than two
    weeks is more than what our usual policy tolerates.

I can't reproduce these errors, so I'm very much limited in what I can do.,

Did this go away with the fold-const.c patch I posted on the GCC list a
couple of days ago?

If not, I'm going to need some help in how to reproduce this.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-17 17:53 -------
I can no longer reproduce it with the conditions I had (even without your patch) but it looks like it other 
people are having the same problem.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-20 00:23 -------
*** Bug 16638 has been marked as a duplicate of this bug. ***

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |danglin at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[dave at hiauly1 dot hia dot nrc dot ca]

------- Additional Comments From dave at hiauly1 dot hia dot nrc dot ca  2004-07-20 00:39 -------
Subject: Re:  [3.5 Regression] ICE caused by Kenner's bi

> ------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-20
> 00:23 -------
> *** Bug 16638 has been marked as a duplicate of this bug. ***

Looks the same but I could build with yesterdays cvs ...

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-20 00:42 -------
Note the reasons why all of these are the same bugs come from the fact if you change the memory 
layout slightly, it can make the ICE to show up.  I have not come up with a testcase as any slight 
change even -save-temps will not show the ICE.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-07-20 00:53 -------
Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    Note the reasons why all of these are the same bugs come from the fact
    if you change the memory layout slightly, it can make the ICE to show
    up.  I have not come up with a testcase as any slight change even
    -save-temps will not show the ICE.

Does the fold-const.c change I posted last week address this?  I've heard
both "yes" and "no" and am confused.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[dave at hiauly1 dot hia dot nrc dot ca]

------- Additional Comments From dave at hiauly1 dot hia dot nrc dot ca  2004-07-20 01:13 -------
Subject: Re:  [3.5 Regression] ICE caused by Kenner's bi

> Does the fold-const.c change I posted last week address this?  I've heard
> both "yes" and "no" and am confused.

Would you attach the patch to the PR?  I can't find it in gcc-patches.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-07-20 01:28 -------
Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    Would you attach the patch to the PR?  I can't find it in gcc-patches.

I'll put it here.  It was on the GCC list itself, not gcc-patches.

*** fold-const.c	8 Jul 2004 17:40:22 -0000	1.421
--- fold-const.c	12 Jul 2004 14:39:08 -0000
*************** int
*** 2348,2353 ****
  operand_equal_p (tree arg0, tree arg1, unsigned int flags)
  {
    /* If either is ERROR_MARK, they aren't equal.  */
!   if (TREE_CODE (arg0) == ERROR_MARK || TREE_CODE (arg1) == ERROR_MARK)
      return 0;
  
--- 2348,2359 ----
  operand_equal_p (tree arg0, tree arg1, unsigned int flags)
  {
+   /* If one is specified and the other isn't, they aren't equal and if
+      neither is specified, they are.  */
+   if ((arg0 && !arg1) || (!arg0 && arg1))
+     return 0;
+   else if (!arg0 && !arg1)
+     return 1;
    /* If either is ERROR_MARK, they aren't equal.  */
!   else if (TREE_CODE (arg0) == ERROR_MARK || TREE_CODE (arg1) == ERROR_MARK)
      return 0;
  
*************** operand_equal_p (tree arg0, tree arg1, u
*** 2483,2487 ****
  				  TREE_OPERAND (arg1, 0), flags);
  
- 	case COMPONENT_REF:
  	case ARRAY_REF:
  	case ARRAY_RANGE_REF:
--- 2489,2492 ----
*************** operand_equal_p (tree arg0, tree arg1, u
*** 2489,2493 ****
  				   TREE_OPERAND (arg1, 0), flags)
  		  && operand_equal_p (TREE_OPERAND (arg0, 1),
! 				      TREE_OPERAND (arg1, 1), flags));
  
  	case BIT_FIELD_REF:
--- 2494,2512 ----
  				   TREE_OPERAND (arg1, 0), flags)
  		  && operand_equal_p (TREE_OPERAND (arg0, 1),
! 				      TREE_OPERAND (arg1, 1), flags)
! 		  && operand_equal_p (TREE_OPERAND (arg0, 2),
! 				      TREE_OPERAND (arg1, 2), flags)
! 		  && operand_equal_p (TREE_OPERAND (arg0, 3),
! 				      TREE_OPERAND (arg1, 3), flags));
! 
! 
! 	case COMPONENT_REF:
! 	  return (operand_equal_p (TREE_OPERAND (arg0, 0),
! 				   TREE_OPERAND (arg1, 0), flags)
! 		  && operand_equal_p (TREE_OPERAND (arg0, 1),
! 				      TREE_OPERAND (arg1, 1), flags)
! 		  && operand_equal_p (TREE_OPERAND (arg0, 2),
! 				      TREE_OPERAND (arg1, 2), flags));
! 
  
  	case BIT_FIELD_REF:


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-21 01:30 -------
*** Bug 16652 has been marked as a duplicate of this bug. ***

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |reichelt at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-21 01:32 -------
The workaround was installed but there is a still an issue of how are the ARRAY_REFs with CONSTANT 
expressions coming about.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

[kenner at vlsi1 dot ultra dot nyu dot edu]

------- Additional Comments From kenner at vlsi1 dot ultra dot nyu dot edu  2004-07-21 01:33 -------
Subject: Re:   [3.5 Regression] ICE caused by Kenner's big untested (regression causing) patch

    The workaround was installed but there is a still an issue of how are
    the ARRAY_REFs with CONSTANT expressions coming about.

Not an issue.  The gimplifier makes them.  For example in
canonicalize_addr_expr.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237

[Bug middle-end/16237] [4.0 Regression] ICE caused by Kenner's big untested (regression causing) patch

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-22 18:24 -------
Fixed for a while now, if I can come up with a testcase for the missed optimization I will file it.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16237