bootstrap failure of tree-ssa on armv5l-linux

bootstrap failure of tree-ssa on armv5l-linux

[jeroen dobbelaere]

Hi,

I've added the tree-ssa branch to my 'daily gcc-cvs build platform' which is based on
a i80200 (xscale core, armv5l architecture) processor.

Currently the bootstrap failes with following lines :


stage1/xgcc -Bstage1/ -B/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/bin/ -c   -g -O2  -DIN_GCC   -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common   -DHAVE_CONFIG_H 
   -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config -I../../gcc/gcc/../include -I../../gcc/gcc/../libbanshee/libcompat -I../../gcc/gcc/../libbanshee -I../../gcc/gcc/../libbanshee/points-to ../../gcc/gcc/haifa-sched.c -o 
haifa-sched.o
stage1/xgcc -Bstage1/ -B/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/bin/ -c   -g -O2  -DIN_GCC   -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common   -DHAVE_CONFIG_H 
   -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config -I../../gcc/gcc/../include -I../../gcc/gcc/../libbanshee/libcompat -I../../gcc/gcc/../libbanshee -I../../gcc/gcc/../libbanshee/points-to ../../gcc/gcc/hooks.c -o hooks.o
stage1/xgcc -Bstage1/ -B/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/bin/ -c   -g -O2  -DIN_GCC   -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common   -DHAVE_CONFIG_H 
   -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config -I../../gcc/gcc/../include -I../../gcc/gcc/../libbanshee/libcompat -I../../gcc/gcc/../libbanshee -I../../gcc/gcc/../libbanshee/points-to ../../gcc/gcc/ifcvt.c -o ifcvt.o
stage1/xgcc -Bstage1/ -B/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/bin/ -c   -g -O2  -DIN_GCC   -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common -Wno-error 
-DHAVE_CONFIG_H -DGENERATOR_FILE    -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config -I../../gcc/gcc/../include -I../../gcc/gcc/../libbanshee/libcompat -I../../gcc/gcc/../libbanshee -I../../gcc/gcc/../libbanshee/points-to 
../../gcc/gcc/genattrtab.c -o genattrtab.o
../../gcc/gcc/genattrtab.c: In function `attr_rtx':
../../gcc/gcc/genattrtab.c:726: warning: second parameter of `va_start' not last named argument
../../gcc/gcc/genattrtab.c: In function `attr_rtx_cost':
../../gcc/gcc/genattrtab.c:3201: internal compiler error: in verify_local_live_at_start, at flow.c:578
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.
make[2]: *** [genattrtab.o] Error 1
make[2]: Leaving directory `/usr/src/cvs-others/gcc-trees/gcc-tree-ssa-20020619-branch/build-tmp-0/gcc'
make[1]: *** [stage2_build] Error 2
make[1]: Leaving directory `/usr/src/cvs-others/gcc-trees/gcc-tree-ssa-20020619-branch/build-tmp-0/gcc'
make: *** [bootstrap] Error 2


Configure is done as follows :
../gcc/configure --prefix=/usr/local/gcc-3.5-tree-ssa-cvs --with-cpu=xscale --enable-languages=c,c++,f77,objc


Platform :
Linux xingu-07 2.4.16-rmk1-xng1 #38 Mon Jul 8 16:09:28 CEST 2002 armv5l unknown


Greetings,
-- 
Jeroen Dobbelaere
Embedded Software Engineer

ACUNIA Embedded Solutions
http://www.acunia.com/aes

Re: bootstrap failure of tree-ssa on armv5l-linux

[Diego Novillo]

On Mon, 2003-05-12 at 03:59, jeroen dobbelaere wrote:
> Hi,
> 
> I've added the tree-ssa branch to my 'daily gcc-cvs build platform' which is based on
> a i80200 (xscale core, armv5l architecture) processor.
> 
Thanks.

> Currently the bootstrap failes with following lines :
> 
Hmm, this looks target-specific.  It may be bad RTL coming out of the
tree-ssa passes.  I doubt I can reproduce it on another target, but
could you post a .i file?

If you can reproduce this with -fdisable-simple, then the problem may
not be in the gimplifier nor the tree optimizers.

If you can reproduce this with -fdisable-tree-ssa, then the problem is
in the tree optimizers.

If you can reproduce this with -fdisable-simple but it works with
mainline, then we have a bigger problem because -fdisable-simple
completely disables GIMPLE and the tree optimizers, resulting in a
compiler that should behave exactly like mainline.


Thanks.  Diego.

Re: bootstrap failure of tree-ssa on armv5l-linux

[jeroen dobbelaere]

[-- Attachment #1: Type: text/plain, Size: 3078 bytes --]

Diego Novillo wrote:
> On Mon, 2003-05-12 at 03:59, jeroen dobbelaere wrote:
> 
>>Hi,
>>
>>I've added the tree-ssa branch to my 'daily gcc-cvs build platform' which is based on
>>a i80200 (xscale core, armv5l architecture) processor.
>>
> 
> Thanks.
> 
> 
>>Currently the bootstrap failes with following lines :
>>
> 
> Hmm, this looks target-specific.  It may be bad RTL coming out of the
> tree-ssa passes.  I doubt I can reproduce it on another target, but
> could you post a .i file?
> 
> If you can reproduce this with -fdisable-simple, then the problem may
> not be in the gimplifier nor the tree optimizers.

With -fdisable-simple, the file compiles fine, and I get :

  stage1/xgcc -Bstage1/ -B/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/bin/ -c   -g -O2  -DIN_GCC   -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common -Wno-error 
-DHAVE_CONFIG_H -DGENERATOR_FILE    -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config -I../../gcc/gcc/../include -I../../gcc/gcc/../libbanshee/libcompat -I../../gcc/gcc/../libbanshee -I../../gcc/gcc/../libbanshee/points-to 
../../gcc/gcc/genattrtab.c -o genattrtab.o -fdisable-simple
../../gcc/gcc/genattrtab.c:0: warning: -fdisable-simple also disables optimizations on trees

> 
> If you can reproduce this with -fdisable-tree-ssa, then the problem is
> in the tree optimizers.

stage1/xgcc -Bstage1/ -B/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/bin/ -c   -g -O2  -DIN_GCC
  -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common
  -Wno-error  -DHAVE_CONFIG_H -DGENERATOR_FILE    -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config
  -I../../gcc/gcc/../include -I../../gcc/gcc/../libbanshee/libcompat -I../../gcc/gcc/../libbanshee
-I../../gcc/gcc/../libbanshee/points-to ../../gcc/gcc/genattrtab.c -o genattrtab.o -fdisable-tree-ssa
../../gcc/gcc/genattrtab.c: In function `attr_rtx':
../../gcc/gcc/genattrtab.c:726: warning: second parameter of `va_start' not last named argument
../../gcc/gcc/genattrtab.c: In function `attr_rtx_cost':
../../gcc/gcc/genattrtab.c:3201: internal compiler error: in verify_local_live_at_start, at flow.c:578
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

> 
> If you can reproduce this with -fdisable-simple but it works with
> mainline, then we have a bigger problem because -fdisable-simple
> completely disables GIMPLE and the tree optimizers, resulting in a
> compiler that should behave exactly like mainline.
> 

current mainline is no problem.
For example :
<http://gcc.gnu.org/ml/gcc-testresults/2003-05/msg00681.html>

But at the moment of the merger (2003/05/04) the mainline also had problems bootstrapping.
Although, '-fdisable-simple' seems to work fine, so I don't know if this is related.

I've appended the .i output.

Greetings,
-- 
Jeroen Dobbelaere
Embedded Software Engineer

ACUNIA Embedded Solutions
http://www.acunia.com/aes


[-- Attachment #2: genattrtab.i.bz2 --]
[-- Type: application/x-bzip2, Size: 47529 bytes --]

Re: bootstrap failure of tree-ssa on armv5l-linux

[Diego Novillo]

On Mon, 2003-05-12 at 09:01, jeroen dobbelaere wrote:
> Diego Novillo wrote:
> > On Mon, 2003-05-12 at 03:59, jeroen dobbelaere wrote:
> > 
> >>Hi,
> >>
> >>I've added the tree-ssa branch to my 'daily gcc-cvs build platform' which is based on
> >>a i80200 (xscale core, armv5l architecture) processor.
> >>
> > 
> > Thanks.
> > 
> > 
> >>Currently the bootstrap failes with following lines :
> >>
> > 
> > Hmm, this looks target-specific.  It may be bad RTL coming out of the
> > tree-ssa passes.  I doubt I can reproduce it on another target, but
> > could you post a .i file?
> > 
> > If you can reproduce this with -fdisable-simple, then the problem may
> > not be in the gimplifier nor the tree optimizers.
> 
> With -fdisable-simple, the file compiles fine, and I get :
> 
>   stage1/xgcc -Bstage1/ -B/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/bin/ -c   -g -O2  -DIN_GCC   -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common -Wno-error 
> -DHAVE_CONFIG_H -DGENERATOR_FILE    -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config -I../../gcc/gcc/../include -I../../gcc/gcc/../libbanshee/libcompat -I../../gcc/gcc/../libbanshee -I../../gcc/gcc/../libbanshee/points-to 
> ../../gcc/gcc/genattrtab.c -o genattrtab.o -fdisable-simple
> ../../gcc/gcc/genattrtab.c:0: warning: -fdisable-simple also disables optimizations on trees
> > 
> > If you can reproduce this with -fdisable-tree-ssa, then the problem is
> > in the tree optimizers.
> 
Drat.  There's a missing 'not' in that sentence.  If you can reproduce
the problem with -fdisable-tree-ssa, the problem is *not* in the
optimizers.  It seems to be a gimplifier problem, then.

> But at the moment of the merger (2003/05/04) the mainline also had problems bootstrapping.
>
Ah, there's hope then.  I will be committing a new merge today or
tomorrow.  Please try with that when it's ready.

> Although, '-fdisable-simple' seems to work fine, so I don't know if this is related.
> 
Hmm, yeah.  I'll see if I can reproduce the problem on my end as well.


Diego.

Re: bootstrap failure of tree-ssa on armv5l-linux

[jeroen dobbelaere]

jeroen dobbelaere wrote:
> Diego Novillo wrote:
> 
>> [..]
> 
> current mainline is no problem.
> For example :
> <http://gcc.gnu.org/ml/gcc-testresults/2003-05/msg00681.html>
> 
> But at the moment of the merger (2003/05/04) the mainline also had 
> problems bootstrapping.
> Although, '-fdisable-simple' seems to work fine, so I don't know if this 
> is related.
> 
[..]

The problem is still present after the 2003-05-11 merger, so this should indicate a
problem in the gimplifier ?

Anything I can do to provide more detailed information ?

Greetings,
-- 
Jeroen Dobbelaere
Embedded Software Engineer

ACUNIA Embedded Solutions
http://www.acunia.com/aes

Re: bootstrap failure of tree-ssa on armv5l-linux

[Diego Novillo]

On Tue, 2003-05-13 at 04:50, jeroen dobbelaere wrote:

> The problem is still present after the 2003-05-11 merger, so this should indicate a
> problem in the gimplifier ?
> 
> Anything I can do to provide more detailed information ?
> 
I will try to reproduce it with a cross compiler.  Could you send me a
.i file and the configuration needed to get a cross cc1 for your target?


Thanks.  Diego.

Re: bootstrap failure of tree-ssa on armv5l-linux

[jeroen dobbelaere]

[-- Attachment #1: Type: text/plain, Size: 1872 bytes --]

Diego Novillo wrote:
[..]
> 
> I will try to reproduce it with a cross compiler.  Could you send me a
> .i file and the configuration needed to get a cross cc1 for your target?
> 
> 
> Thanks.  Diego.
> 

Assuming that you have binutils compiled and installed for arm-linux, in
/opt/test-cross-gcc-arm-linux

and your working directory layout is :

./build        -> for building
./gcc          -> gcc source
./genattrtab.i -> for testing


You can do this :


export PATH=/opt/test-cross-gcc-arm-linux/bin:$PATH
cd build
../gcc/configure --prefix=/opt/test-cross-gcc-arm-linux --target=arm-linux --with-cpu=xscale --enable-languages=c
make
gcc/cc1 -fpreprocessed ../genattrtab.i -quiet -dumpbase genattrtab.c -auxbase-strip genattrtab.o -g -O2 -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -Wno-error -version -fno-common -o 
genattrtab.s

Results in :

GNU C version 3.5-tree-ssa 20030516 (merged 20030511) (arm-linux)
	compiled by GNU C version 3.0.4.
GGC heuristics: --param ggc-min-expand=64 --param ggc-min-heapsize=64015
../../gcc/gcc/genattrtab.c: In function `attr_rtx':
../../gcc/gcc/genattrtab.c:726: warning: second parameter of `va_start' not last named argument
../../gcc/gcc/genattrtab.c: In function `attr_rtx_cost':
../../gcc/gcc/genattrtab.c:3201: internal compiler error: in verify_local_live_at_start, at flow.c:578
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.



Note : In my testsetup, I have in /opt/test-cross-gcc-arm-linux, besides binutils,
  also glibc and a cross-arm-linux gcc-3.0.4 compiler installed.
  This may influence the reproducability of this procedure. (headers, libraries, ...)

Greetings,
-- 
Jeroen Dobbelaere
Embedded Software Engineer

ACUNIA Embedded Solutions
http://www.acunia.com/aes


[-- Attachment #2: genattrtab.i.bz2 --]
[-- Type: application/x-bzip2, Size: 47576 bytes --]

Re: bootstrap failure of tree-ssa on armv5l-linux

[jeroen dobbelaere]

[-- Attachment #1: Type: text/plain, Size: 3528 bytes --]

I have done some further investigation on this problem, and I have
attached a reduced testcase, together with the output at stage .31, .32 and .33

stage1/cc1 -fpreprocessed thetest_reduced.i -quiet -dumpbase thetest_reduced.c -auxbase-strip thetest_reduced.o -g -O2 -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -Wno-error -version 
-fno-common -o thetest_reduced.s -da -v
ignoring nonexistent directory "/usr/local/gcc-3.5-tree-ssa-cvs/include"
ignoring nonexistent directory "/usr/local/gcc-3.5-tree-ssa-cvs/lib/gcc-lib/armv5l-unknown-linux-gnu/3.5-tree-ssa/include"
ignoring nonexistent directory "/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/include"
#include "..." search starts here:
#include <...> search starts here:
  /usr/local/include
  /usr/include
End of search list.
GNU C version 3.5-tree-ssa 20030522 (merged 20030511) (armv5l-unknown-linux-gnu)
	compiled by GNU C version 3.0.4.
GGC heuristics: --param ggc-min-expand=38 --param ggc-min-heapsize=15924
thetest_reduced.i: In function `attr_rtx_cost':
thetest_reduced.i:380: internal compiler error: in verify_local_live_at_start, at flow.c:578


stack trace at that point :

#0  fancy_abort (file=0x5ee010 "../../gcc/gcc/flow.c", line=578, function=0x5ee0f0 "verify_local_live_at_start")
     at ../../gcc/gcc/diagnostic.c:1439
#1  0x002778d4 in verify_local_live_at_start (new_live_at_start=0xbfffed80, bb=0x6bdc68) at ../../gcc/gcc/flow.c:578
#2  0x00277e6c in update_life_info (blocks=0x6caaf8, extent=UPDATE_LIFE_LOCAL, prop_flags=1) at ../../gcc/gcc/flow.c:715
#3  0x0059b51c in schedule_insns (dump_file=0x6cab18) at ../../gcc/gcc/sched-rgn.c:2759
#4  0x0048f0c0 in rest_of_compilation (decl=0x401b4310) at ../../gcc/gcc/toplev.c:3661
#5  0x00034c10 in c_expand_body_1 (fndecl=0x401b4310, nested_p=0) at ../../gcc/gcc/c-decl.c:6439
#6  0x0003533c in c_expand_body (fndecl=0x401b4310) at ../../gcc/gcc/c-decl.c:6523
#7  0x000343f4 in finish_function (nested=0, can_defer_p=1) at ../../gcc/gcc/c-decl.c:6277
#8  0x0000b9e0 in yyparse () at c-parse.y:406
#9  0x00084278 in c_common_parse_file (set_yydebug=0) at ../../gcc/gcc/c-opts.c:1634
#10 0x0048bfe8 in compile_file () at ../../gcc/gcc/toplev.c:2252
#11 0x00493bd4 in do_compile () at ../../gcc/gcc/toplev.c:5648
#12 0x00493c50 in toplev_main (argc=25, argv=0xbffffb84) at ../../gcc/gcc/toplev.c:5679
#13 0x000dac68 in main (argc=25, argv=0xbffffb84) at ../../gcc/gcc/main.c:37
#14 0x4003976c in __libc_start_main (main=0xdac44 <main>, argc=25, ubp_av=0xbffffb84, init=0x9514 <_init>, fini=0,
     rtld_fini=0, stack_end=0x0) at ../sysdeps/generic/libc-start.c:129


The problem seems to be that in update_life_info, the registers live_at_start has changed
(register r0  was removed...)

live_at_start mismatch in bb 15, aborting
New:

first = 0x6cbf90 current = 0x6cbf90 indx = 0
	0x6cbf90 next = (nil) prev = (nil) indx = 0
		bits = { 4 5 6 7 8 10 13 14 25 26 }
Old:
;; Basic block 15, loop depth 0, count 0
;; Predecessors:  1 [29.0%]  (can_fallthru)
;; Registers live at start: 0 [r0] 4 [r4] 5 [r5] 6 [r6] 7 [r7] 8 [r8] 10 [sl] 13 [sp] 14 [lr] 25 [sfp] 26 [afp]
(code_label:HI 28 27 258 15 3 "" [1 uses])
(note:HI 258 28 29 15 [bb 15] NOTE_INSN_BASIC_BLOCK)

Although looking at the content of basic block 15 at that point, it seems to be valid to
conclude that 'r0' is not live at the beginning of bb 15.

Any hints ?

Greetings,
-- 
Jeroen Dobbelaere
Embedded Software Engineer

ACUNIA Embedded Solutions
http://www.acunia.com/aes


[-- Attachment #2: thetest_reduced.c.31.bbro.bz2 --]
[-- Type: application/x-bzip2, Size: 5726 bytes --]

[-- Attachment #3: thetest_reduced.c.32.ce3.bz2 --]
[-- Type: application/x-bzip2, Size: 4030 bytes --]

[-- Attachment #4: thetest_reduced.c.33.sched2.bz2 --]
[-- Type: application/x-bzip2, Size: 2302 bytes --]

[-- Attachment #5: thetest_reduced.i.bz2 --]
[-- Type: application/x-bzip2, Size: 2020 bytes --]

Re: bootstrap failure of tree-ssa on armv5l-linux

[Diego Novillo]

On Fri, 2003-05-23 at 04:56, jeroen dobbelaere wrote:
> I have done some further investigation on this problem, and I have
> attached a reduced testcase, together with the output at stage .31, .32 and .33
> 
Jeroen,

Sorry it's taken so long.  This seems to have been fixed now.  Could you
try bootstrapping again?


Thanks.  Diego.

Re: bootstrap failure of tree-ssa on armv5l-linux

[jeroen dobbelaere]

Hi Diego,

Diego Novillo wrote:
[..]

> Jeroen,
> 
> Sorry it's taken so long.  This seems to have been fixed now.  Could you
> try bootstrapping again?
> 
> 
> Thanks.  Diego.
> 

Nope, still the same problem.

stage1/xgcc -Bstage1/ -B/usr/local/gcc-3.5-tree-ssa-cvs/armv5l-unknown-linux-gnu/bin/ -c   -g -O2  -DIN_GCC   -W -Wall -Wwrite-
strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common -Wno-error  -DHAVE_CONFIG_H -DGEN
ERATOR_FILE    -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config -I../../gcc/gcc/../include -I../../gcc/gcc/../l
ibbanshee/libcompat -I../../gcc/gcc/../libbanshee -I../../gcc/gcc/../libbanshee/points-to ../../gcc/gcc/genattrtab.c -o genattr
tab.o
../../gcc/gcc/genattrtab.c: In function `attr_rtx_cost':
../../gcc/gcc/genattrtab.c:3200: internal compiler error: in verify_local_live_at_start, at flow.c:578
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.
make[2]: *** [genattrtab.o] Error 1


Greetings,
-- 
Jeroen Dobbelaere
Embedded Software Engineer

ACUNIA Embedded Solutions
http://www.acunia.com/aes

Re: bootstrap failure of tree-ssa on armv5l-linux

[Diego Novillo]

On Fri, 2003-06-13 at 09:39, jeroen dobbelaere wrote:

> Nope, still the same problem.
> 
OK.  I think I may have access to an ARM box.  Send me the .i file?


Thanks.  Diego.

Re: bootstrap failure of tree-ssa on armv5l-linux

[jeroen dobbelaere]

Diego Novillo wrote:
> On Fri, 2003-06-13 at 09:39, jeroen dobbelaere wrote:
> 
> 
>>Nope, still the same problem.
>>
> 
> OK.  I think I may have access to an ARM box.  Send me the .i file?
> 
> 
> Thanks.  Diego.
>

See :

<http://gcc.gnu.org/ml/gcc/2003-05/msg02062.html>

Note : I have it also reproduced with a cross-compiler :

./gcc/cc1 -fpreprocessed ../thetest_reduced.i -quiet -dumpbase genattrtab.c -auxbase-strip genattrtab.o -g -O2 -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -Wno-error -version 
-fno-common -o genattrtab.s
GNU C version 3.5-tree-ssa 20030613 (merged 20030525) (arm-linux)
         compiled by GNU C version 3.2.2 (Mandrake Linux 9.1 3.2.2-3mdk).
GGC heuristics: --param ggc-min-expand=47 --param ggc-min-heapsize=31960
../thetest_reduced.i: In function `attr_rtx_cost':
../thetest_reduced.i:380: internal compiler error: in verify_local_live_at_start, at flow.c:578
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

I'll also be looking at this problem in the weekend.

Greetings,
-- 
Jeroen Dobbelaere
Embedded Software Engineer

ACUNIA Embedded Solutions
http://www.acunia.com/aes