Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Richard Kenner]

    For some reason, I don't seem to get too far when I enable Ada:

    stage1/xgcc -Bstage1/ -B/home/dnovillo/gcc.ada/native/i686-pc-linux-gnu/bin/ -c -g -O2      -gnatpg -gnata -I- -I. -Iada -I/home/dnovillo/gcc.ada/src/gcc/ada /home/dnovillo/gcc.ada/src/gcc/ada/ada.ads -o ada/ada.o
    xgcc: installation problem, cannot exec 'gnat1': No such file or directory
    make[1]: *** [ada/ada.o] Error 1
    make[1]: Leaving directory `/home/dnovillo/notnfs/BLD-gcc.ada-native.topo/gcc'
    make: *** [stage2_build] Error 2
    
    I've configured with '--enable-languages=ada,c'.  I do have a host ada
    compiler as well.  What else should I check?

Sounds like you configured with Ada, but somehow didn't "make" to build it.

Before you do the "make bootstrap" you should have a gnat1 in the same
directory as cc1 and it looks like you don't.  See if it's in stage1.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Diego Novillo]

On Wed, 2004-07-28 at 10:01, Richard Kenner wrote:

> Sounds like you configured with Ada, but somehow didn't "make" to build it.
> 
$ <src>/configure --enable-languages=ada,c
$ make bootstrap

I was missing the changes to ada/config-lang.in.  Which strikes me as
odd, because I was explicitly asking for ada in the command line.  Oh,
well.

I'm now bootstrapping ada with a slightly different patch than the one I
sent originally (that one didn't work).


Diego.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Richard Kenner]

    I'm testing a patch to fix it.  I'll check if it has any effect on ada
    and let you know.

OK.  I'll hold off looking into it until I hear back from you.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Diego Novillo]

On Wed, 2004-07-28 at 11:04, Richard Kenner wrote:
>     stage2/xgcc -Bstage2/ -B/home/dnovillo/gcc.ada/native/i686-pc-linux-gnu/bin/ -c -g -O2      -gnatpg -gnata -I- -I. -Iada -I/home/dnovillo/gcc.ada/src/gcc/ada /home/dnovillo/gcc.ada/src/gcc/ada/ada.ads -o ada/ada.o
>     +===========================GNAT BUG DETECTED==============================+
>     | 3.5.0 20040728 (experimental) (i686-pc-linux-gnu) Storage_Error stack overflow (or erroneous memory access)|
> 
> I was afraid something else would have gotten broken in the week or so
> that I was last able to do a bootstrap.
> 
> I guess I know what I'm working on today.  :-(
>
Note that the .GLOBAL_VAR creation heuristic is buggy.  Causing bad code
generation problems.  I see that Ada triggers the creation of
.GLOBAL_VAR.

I'm testing a patch to fix it.  I'll check if it has any effect on ada
and let you know.


Diego.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Richard Kenner]

    stage2/xgcc -Bstage2/ -B/home/dnovillo/gcc.ada/native/i686-pc-linux-gnu/bin/ -c -g -O2      -gnatpg -gnata -I- -I. -Iada -I/home/dnovillo/gcc.ada/src/gcc/ada /home/dnovillo/gcc.ada/src/gcc/ada/ada.ads -o ada/ada.o
    +===========================GNAT BUG DETECTED==============================+
    | 3.5.0 20040728 (experimental) (i686-pc-linux-gnu) Storage_Error stack overflow (or erroneous memory access)|

I was afraid something else would have gotten broken in the week or so
that I was last able to do a bootstrap.

I guess I know what I'm working on today.  :-(

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Diego Novillo]

On Wed, 2004-07-28 at 10:40, Richard Kenner wrote:
>     Ironically, that patch went on to cause an ICE in a non-ada file.  I'll
>     commit this version of the patch if I get through a bootstrap cycle.
> 
> I saw that too.  I'm rerunning with your latst.
> 
OK.  I got as far as building stage3.  It fails with:

stage2/xgcc -Bstage2/ -B/home/dnovillo/gcc.ada/native/i686-pc-linux-gnu/bin/ -c -g -O2      -gnatpg -gnata -I- -I. -Iada -I/home/dnovillo/gcc.ada/src/gcc/ada /home/dnovillo/gcc.ada/src/gcc/ada/ada.ads -o ada/ada.o
+===========================GNAT BUG DETECTED==============================+
| 3.5.0 20040728 (experimental) (i686-pc-linux-gnu) Storage_Error stack overflow (or erroneous memory access)|
| Error detected at :1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1:1 [:1
[ ... ]

And goes on printing garbage until I stop it.  But this is not related
to the verifier failure, so I'll commit the patch after doing a default
bootstrap.


Diego.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Richard Kenner]

    Ironically, that patch went on to cause an ICE in a non-ada file.  I'll
    commit this version of the patch if I get through a bootstrap cycle.

I saw that too.  I'm rerunning with your latst.

    Just to make sure, Ada bootstraps but doesn't pass all of its testsuite,
    right?

The last time I ran it, the status was that it bootstrapped, but
wouldn't build it's library without disabling loop optimization.
There are something like 87 test suite failure remaining (with the
largest cluster being the DCE problem I mentioned last week).

Of course, there have been lots of changes since the last time that
anybody bootstrapped, so I wouldn't want to use the present tense just yet.
I'll know shortly what the present situation is.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Diego Novillo]

On Wed, 2004-07-28 at 10:20, Richard Kenner wrote:
>     Try this (not really tested):
> 
> It fixes that file.  It's one of the last to be compiled, so it's probably
> OK now (at least that problem), but I'm running a full bootstrap now.
>
Ironically, that patch went on to cause an ICE in a non-ada file.  I'll
commit this version of the patch if I get through a bootstrap cycle.

Just to make sure, Ada bootstraps but doesn't pass all of its testsuite,
right?


Diego.


Index: tree-ssa.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-ssa.c,v
retrieving revision 2.24
diff -d -c -p -r2.24 tree-ssa.c
*** tree-ssa.c	28 Jul 2004 05:13:08 -0000	2.24
--- tree-ssa.c	28 Jul 2004 14:22:06 -0000
*************** verify_flow_insensitive_alias_info (void
*** 348,375 ****
  
    for (i = 0; i < num_referenced_vars; i++)
      {
        var_ann_t ann;
  
        var = referenced_var (i);
        ann = var_ann (var);
  
!       if (ann->mem_tag_kind == TYPE_TAG || ann->mem_tag_kind == NAME_TAG)
  	{
! 	  size_t j;
! 	  varray_type may_aliases = ann->may_aliases;
! 
! 	  for (j = 0; may_aliases && j < VARRAY_ACTIVE_SIZE (may_aliases); j++)
! 	    {
! 	      tree alias = VARRAY_TREE (may_aliases, j);
  
! 	      bitmap_set_bit (visited, var_ann (alias)->uid);
  
! 	      if (!may_be_aliased (alias))
! 		{
! 		  error ("Non-addressable variable inside an alias set.");
! 		  debug_variable (alias);
! 		  goto err;
! 		}
  	    }
  	}
      }
--- 348,372 ----
  
    for (i = 0; i < num_referenced_vars; i++)
      {
+       size_t j;
        var_ann_t ann;
+       varray_type may_aliases;
  
        var = referenced_var (i);
        ann = var_ann (var);
+       may_aliases = ann->may_aliases;
  
!       for (j = 0; may_aliases && j < VARRAY_ACTIVE_SIZE (may_aliases); j++)
  	{
! 	  tree alias = VARRAY_TREE (may_aliases, j);
  
! 	  bitmap_set_bit (visited, var_ann (alias)->uid);
  
! 	  if (!may_be_aliased (alias))
! 	    {
! 	      error ("Non-addressable variable inside an alias set.");
! 	      debug_variable (alias);
! 	      goto err;
  	    }
  	}
      }

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Richard Kenner]

    Try this (not really tested):

It fixes that file.  It's one of the last to be compiled, so it's probably
OK now (at least that problem), but I'm running a full bootstrap now.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Duncan Sands]

> For some reason, I don't seem to get too far when I enable Ada:
> 
> stage1/xgcc -Bstage1/ -B/home/dnovillo/gcc.ada/native/i686-pc-linux-gnu/bin/ -c -g -O2      -gnatpg -gnata -I- -I. -Iada -I/home/dnovillo/gcc.ada/src/gcc/ada /home/dnovillo/gcc.ada/src/gcc/ada/ada.ads -o ada/ada.o
> xgcc: installation problem, cannot exec 'gnat1': No such file or directory
> make[1]: *** [ada/ada.o] Error 1
> make[1]: Leaving directory `/home/dnovillo/notnfs/BLD-gcc.ada-native.topo/gcc'
> make: *** [stage2_build] Error 2
> 
> I've configured with '--enable-languages=ada,c'.  I do have a host ada
> compiler as well.  What else should I check?

You need

boot_language=yes

in gcc/ada/config-lang.in

I have no idea why it was set to "no"...

All the best,

Duncan.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Andreas Schwab]

Diego Novillo <dnovillo@redhat.com> writes:

> For some reason, I don't seem to get too far when I enable Ada:
>
> stage1/xgcc -Bstage1/ -B/home/dnovillo/gcc.ada/native/i686-pc-linux-gnu/bin/ -c -g -O2      -gnatpg -gnata -I- -I. -Iada -I/home/dnovillo/gcc.ada/src/gcc/ada /home/dnovillo/gcc.ada/src/gcc/ada/ada.ads -o ada/ada.o
> xgcc: installation problem, cannot exec 'gnat1': No such file or directory
> make[1]: *** [ada/ada.o] Error 1
> make[1]: Leaving directory `/home/dnovillo/notnfs/BLD-gcc.ada-native.topo/gcc'
> make: *** [stage2_build] Error 2

You need to set boot_language=yes in ada/config-lang.in.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Diego Novillo]

On Wed, 2004-07-28 at 01:20, Richard Kenner wrote:

> stage1/xgcc -Bstage1/ -B/usr/local/x86_64-unknown-linux-gnu/bin/ -c -g -O2
> -gnatpg -gnata -I- -I. -Iada -I/gcc/gcc/gcc-head/gcc/ada /gcc/gcc/gcc-head/gcc/a
> da/sprint.adb -o ada/sprint.o
> /gcc/gcc/gcc-head/gcc/ada/sprint.adb: In function `Sprint.Sprint_Node_Actual':
> /gcc/gcc/gcc-head/gcc/ada/sprint.adb:615: error: Addressable variable that is an
>  alias tag but is not in any alias set.
> .GLOBAL_VAR, UID 1581, is an alias tag, is static, call clobbered, default def:
> .GLOBAL_VAR_396
> +===========================GNAT BUG DETECTED==============================+
> | 5.03h (20040725) (x86_64-unknown-linux-gnu) GCC error:                   |
> | verify_flow_insensitive_alias_info failed.
> 
> Peculiar that it would move like that.
>
I *think* this ia a bug in the verifier.  I could trigger it on a C-only
bootstrap by forcing the creation of .GLOBAL_VAR always (--params
global-var-threshold).

For some reason, I don't seem to get too far when I enable Ada:

stage1/xgcc -Bstage1/ -B/home/dnovillo/gcc.ada/native/i686-pc-linux-gnu/bin/ -c -g -O2      -gnatpg -gnata -I- -I. -Iada -I/home/dnovillo/gcc.ada/src/gcc/ada /home/dnovillo/gcc.ada/src/gcc/ada/ada.ads -o ada/ada.o
xgcc: installation problem, cannot exec 'gnat1': No such file or directory
make[1]: *** [ada/ada.o] Error 1
make[1]: Leaving directory `/home/dnovillo/notnfs/BLD-gcc.ada-native.topo/gcc'
make: *** [stage2_build] Error 2

I've configured with '--enable-languages=ada,c'.  I do have a host ada
compiler as well.  What else should I check?

Try this (not really tested):

Index: tree-ssa.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-ssa.c,v
retrieving revision 2.24
diff -d -c -p -r2.24 tree-ssa.c
*** tree-ssa.c	28 Jul 2004 05:13:08 -0000	2.24
--- tree-ssa.c	28 Jul 2004 13:18:23 -0000
*************** verify_flow_insensitive_alias_info (void
*** 348,375 ****
  
    for (i = 0; i < num_referenced_vars; i++)
      {
        var_ann_t ann;
  
        var = referenced_var (i);
        ann = var_ann (var);
  
!       if (ann->mem_tag_kind == TYPE_TAG || ann->mem_tag_kind == NAME_TAG)
  	{
! 	  size_t j;
! 	  varray_type may_aliases = ann->may_aliases;
! 
! 	  for (j = 0; may_aliases && j < VARRAY_ACTIVE_SIZE (may_aliases); j++)
! 	    {
! 	      tree alias = VARRAY_TREE (may_aliases, j);
  
! 	      bitmap_set_bit (visited, var_ann (alias)->uid);
  
! 	      if (!may_be_aliased (alias))
! 		{
! 		  error ("Non-addressable variable inside an alias set.");
! 		  debug_variable (alias);
! 		  goto err;
! 		}
  	    }
  	}
      }
--- 348,372 ----
  
    for (i = 0; i < num_referenced_vars; i++)
      {
+       size_t j;
        var_ann_t ann;
+       varray_type may_aliases;
  
        var = referenced_var (i);
        ann = var_ann (var);
+       may_aliases = ann->may_aliases;
  
!       for (j = 0; may_aliases && j < VARRAY_ACTIVE_SIZE (may_aliases); j++)
  	{
! 	  tree alias = VARRAY_TREE (may_aliases, j);
  
! 	  bitmap_set_bit (visited, var_ann (alias)->uid);
  
! 	  if (!may_be_aliased (alias))
! 	    {
! 	      error ("Non-addressable variable inside an alias set.");
! 	      debug_variable (alias);
! 	      goto err;
  	    }
  	}
      }
*************** verify_flow_insensitive_alias_info (void
*** 381,389 ****
        var = referenced_var (i);
        ann = var_ann (var);
  
!       if (ann->mem_tag_kind == NOT_A_TAG
! 	  && ann->is_alias_tag
! 	  && !bitmap_bit_p (visited, ann->uid))
  	{
  	  error ("Addressable variable that is an alias tag but is not in any alias set.");
  	  goto err;
--- 378,384 ----
        var = referenced_var (i);
        ann = var_ann (var);
  
!       if (ann->is_alias_tag && !bitmap_bit_p (visited, ann->uid))
  	{
  	  error ("Addressable variable that is an alias tag but is not in any alias set.");
  	  goto err;

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Laurent GUERBY]

I confirm that I get the same ICE on x86_64-linux
with a clean HEAD checkout.

Laurent

On Wed, 2004-07-28 at 07:20, Richard Kenner wrote:
>     This patches fixes this problem.  Essentially, we should always
>     consider .GLOBAL_VAR as potentially aliased.  This may have been
>     related to the Ada bootstrap problems.  Kenner, could you check?
> 
> It indeed fixed it in that file, but now a similar thing happens
> in another one:
> 
> stage1/xgcc -Bstage1/ -B/usr/local/x86_64-unknown-linux-gnu/bin/ -c -g -O2
> -gnatpg -gnata -I- -I. -Iada -I/gcc/gcc/gcc-head/gcc/ada /gcc/gcc/gcc-head/gcc/a
> da/sprint.adb -o ada/sprint.o
> /gcc/gcc/gcc-head/gcc/ada/sprint.adb: In function `Sprint.Sprint_Node_Actual':
> /gcc/gcc/gcc-head/gcc/ada/sprint.adb:615: error: Addressable variable that is an
>  alias tag but is not in any alias set.
> .GLOBAL_VAR, UID 1581, is an alias tag, is static, call clobbered, default def:
> .GLOBAL_VAR_396
> +===========================GNAT BUG DETECTED==============================+
> | 5.03h (20040725) (x86_64-unknown-linux-gnu) GCC error:                   |
> | verify_flow_insensitive_alias_info failed.
> 
> Peculiar that it would move like that.
> 

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Richard Kenner]

    This patches fixes this problem.  Essentially, we should always
    consider .GLOBAL_VAR as potentially aliased.  This may have been
    related to the Ada bootstrap problems.  Kenner, could you check?

It indeed fixed it in that file, but now a similar thing happens
in another one:

stage1/xgcc -Bstage1/ -B/usr/local/x86_64-unknown-linux-gnu/bin/ -c -g -O2
-gnatpg -gnata -I- -I. -Iada -I/gcc/gcc/gcc-head/gcc/ada /gcc/gcc/gcc-head/gcc/a
da/sprint.adb -o ada/sprint.o
/gcc/gcc/gcc-head/gcc/ada/sprint.adb: In function `Sprint.Sprint_Node_Actual':
/gcc/gcc/gcc-head/gcc/ada/sprint.adb:615: error: Addressable variable that is an
 alias tag but is not in any alias set.
.GLOBAL_VAR, UID 1581, is an alias tag, is static, call clobbered, default def:
.GLOBAL_VAR_396
+===========================GNAT BUG DETECTED==============================+
| 5.03h (20040725) (x86_64-unknown-linux-gnu) GCC error:                   |
| verify_flow_insensitive_alias_info failed.

Peculiar that it would move like that.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Richard Kenner]

    This patches fixes this problem.  Essentially, we should always consider
    .GLOBAL_VAR as potentially aliased.  This may have been related to the
    Ada bootstrap problems.  Kenner, could you check?  Thanks.

It fixes the file I saw fail.  I started another bootstrap; if something else
turns up that looks related, I'll let you know.  I figured it was something
simple like this, but wasn't sure exactly what.

Thanks.

Re: [Bug tree-optimization/16705] ICE in verify_flow_insensitive_alias_info

[Diego Novillo]

On Tue, 2004-07-27 at 02:20, pinskia at gcc dot gnu dot org wrote:

> I see this with a fortran to C compiler for SPEC CPU 2000FP code.
> Can you please look into this bug, Diego?
>
This patches fixes this problem.  Essentially, we should always consider
.GLOBAL_VAR as potentially aliased.  This may have been related to the
Ada bootstrap problems.  Kenner, could you check?  Thanks.

I'm not quite sure whether this test can be added as-is to the fortran
testsuite.  Could anyone confirm if I can just drop it in gfortran.dg? 
How should I change the compile flags used?

Bootstrapped and tested x86 and ppc.


Diego.

	PR tree-optimization/16705
	* tree-ssa-alias.c (create_global_var): Set DECL_EXTERNAL on
	.GLOBAL_VAR.


Index: tree-ssa-alias.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-ssa-alias.c,v
retrieving revision 2.17
diff -d -c -p -r2.17 tree-ssa-alias.c
*** tree-ssa-alias.c	27 Jul 2004 15:32:44 -0000	2.17
--- tree-ssa-alias.c	28 Jul 2004 04:15:26 -0000
*************** create_global_var (void)
*** 2202,2208 ****
                             size_type_node);
    DECL_ARTIFICIAL (global_var) = 1;
    TREE_READONLY (global_var) = 0;
!   DECL_EXTERNAL (global_var) = 0;
    TREE_STATIC (global_var) = 1;
    TREE_USED (global_var) = 1;
    DECL_CONTEXT (global_var) = NULL_TREE;
--- 2205,2211 ----
                             size_type_node);
    DECL_ARTIFICIAL (global_var) = 1;
    TREE_READONLY (global_var) = 0;
!   DECL_EXTERNAL (global_var) = 1;
    TREE_STATIC (global_var) = 1;
    TREE_USED (global_var) = 1;
    DECL_CONTEXT (global_var) = NULL_TREE;