[Bug libgcj/15813] New: Throw_2 test case does not work anymore.

[Bug libgcj/15813] New: Throw_2 test case does not work anymore.

[andreast at gcc dot gnu dot org]

The Throw_2 test case from the libjava suite fails since the patch from 
> 2004-05-19  Jeff Law <law@redhat.com>
>  * tree-into-ssa.c (prepare_operand_for_rename): New argument
> 	is_use.
>          If the operand is for a use, then strip away the SSA_NAME, do
>   	not strip away the SSA_NAME for a set.  Never call
> 	release_ssa_name.
>          (mark_def_sites): Appropriately pass additional argument to
>          prepare_operand_for_rename.  If a VDEF_RESULT is not an
>   	SSA_NAME, then set the VDEF_RESULT to the VDEF_OP.
>          (set_def_block): Strip away any SSA_NAME to get to the real
>          underlying variable.
> 

I have the feeling that something gets miscompiled in the area of boehm-gc.
My tests so far show that just before the patch it works and afterwards the test
case times out. A comparison of the libgcj.a shows some differences in a few
files. I will post the file names soon. One is finalize.o from boehm-gc. This
bug is maybe related to 15812?

-- 
           Summary: Throw_2 test case does not work anymore.
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libgcj
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: andreast at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org,java-prs at gcc dot gnu
                    dot org,law at redhat dot com
 GCC build triplet: powerpc-unknown-linux-gnu
  GCC host triplet: powerpc-unknown-linux-gnu
GCC target triplet: powerpc-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|libgcj                      |tree-optimization
           Keywords|                            |wrong-code
            Summary|Throw_2 test case does not  |[3.5 Regression] Throw_2
                   |work anymore.               |test case does not work
                   |                            |anymore.
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[andreast at gcc dot gnu dot org]

------- Additional Comments From andreast at gcc dot gnu dot org  2004-06-05 20:12 -------
Created an attachment (id=6476)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6476&action=view)
diffed objdump output


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[andreast at gcc dot gnu dot org]

------- Additional Comments From andreast at gcc dot gnu dot org  2004-06-05 20:13 -------
I did build with and without the patch, as said. Then i compared the test case
itself and saw no difference in size nor in content. So I compared the involved
libraries and I found that the libgcj differed in size. (The others were the
same, libgcc & libstdc++)
I then concentrated to find out what the diff is, so I compared the objdump -d
of the files and mostly I had no diff until I systematically dumped and diffed
all files.  
Here the list with files which differ when I do an objdump -dS on linux:

-rw-rw-r--    1 andreas  andreas    263420 Jun  2 10:10 BigInteger.o.diff
-rw-rw-r--    1 andreas  andreas     44207 Jun  2 10:10 Convert.o.diff
-rw-rw-r--    1 andreas  andreas    405080 Jun  2 10:10 DecimalFormat.o.diff
-rw-rw-r--    1 andreas  andreas      4491 Jun  2 10:10 finalize.o.diff
-rw-rw-r--    1 andreas  andreas      7672 Jun  2 10:10 natDouble.o.diff
-rw-rw-r--    1 andreas  andreas    120844 Jun  2 10:10 trees.o.diff
-rw-rw-r--    1 andreas  andreas     13376 Jun  2 10:10 URLDecoder.o.diff
-rw-rw-r--    1 andreas  andreas     88862 Jun  2 10:10 verify.o.diff

It is pretty difficult for me to find a pattern which is not ok.

I also attach the diffs of these files, but I can't say for sure if you get the
same diffs.

Maybe you see something obvious.

(ha, created the attachement in the meaning it comes together with the comment :)



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[law at redhat dot com]

------- Additional Comments From law at redhat dot com  2004-06-10 19:18 -------
Subject: Re:  [3.5 Regression] Throw_2 test
	case does not work anymore.

On Sat, 2004-06-05 at 14:13, andreast at gcc dot gnu dot org wrote:
> ------- Additional Comments From andreast at gcc dot gnu dot org  2004-06-05 20:13 -------
> I did build with and without the patch, as said. Then i compared the test case
> itself and saw no difference in size nor in content. So I compared the involved
> libraries and I found that the libgcj differed in size. (The others were the
> same, libgcc & libstdc++)
> I then concentrated to find out what the diff is, so I compared the objdump -d
> of the files and mostly I had no diff until I systematically dumped and diffed
> all files.  
> Here the list with files which differ when I do an objdump -dS on linux:
> 
> -rw-rw-r--    1 andreas  andreas    263420 Jun  2 10:10 BigInteger.o.diff
> -rw-rw-r--    1 andreas  andreas     44207 Jun  2 10:10 Convert.o.diff
> -rw-rw-r--    1 andreas  andreas    405080 Jun  2 10:10 DecimalFormat.o.diff
> -rw-rw-r--    1 andreas  andreas      4491 Jun  2 10:10 finalize.o.diff
> -rw-rw-r--    1 andreas  andreas      7672 Jun  2 10:10 natDouble.o.diff
> -rw-rw-r--    1 andreas  andreas    120844 Jun  2 10:10 trees.o.diff
> -rw-rw-r--    1 andreas  andreas     13376 Jun  2 10:10 URLDecoder.o.diff
> -rw-rw-r--    1 andreas  andreas     88862 Jun  2 10:10 verify.o.diff
> 
> It is pretty difficult for me to find a pattern which is not ok.
> 
> I also attach the diffs of these files, but I can't say for sure if you get the
> same diffs.
> 
> Maybe you see something obvious.
> 
> (ha, created the attachement in the meaning it comes together with the comment :)
FYI -- I've got my PPC box upgraded to a more modern version of YDL
and I've managed to get the Throw2 test to blow chunks.

Jeff




-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[andreast at gcc dot gnu dot org]

------- Additional Comments From andreast at gcc dot gnu dot org  2004-06-12 22:02 -------
translated for me, does it work for you?

What binutils and what kernel are you running ?

I have a ydl 3.0.1 and the so called latest hjl binutils. And I can reproduce it
in every combination. The kernel I run is 2.4.2x. Also, on darwin I encounter
also a Throw_2 timeout, but it is not directly related to your patch. Means,
reverting your patch does not help there.
On solaris it's worse, 32 bit fails, 64 bit times out but not only Throw_2, all
executable tc's. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[mark at gcc dot gnu dot org]

------- Additional Comments From mark at gcc dot gnu dot org  2004-06-24 20:40 -------
I also get the Throw2 timeouts on:

Debian GNU/Linux 3.0
Linux 2.4.18-newpmac
binutils Version: 2.12.90.0.15-1
glibc Version: 2.2.5-15

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mark at klomp dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[aph at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |aph at redhat dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[pinskia at gcc dot gnu dot org]

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[law at redhat dot com]

------- Additional Comments From law at redhat dot com  2004-06-29 15:10 -------
Subject: Re:  [3.5 Regression] Throw_2 test
	case does not work anymore.

On Sat, 2004-06-26 at 06:52, aph at gcc dot gnu dot org wrote:
[ ... nothing ... ]

FWIW, we're mis-compiling natDouble in libjava.  I haven't figured
out how/why yet, but it's clearly natDouble that's causing the 
problems.

jeff



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-29 15:16 -------
Confirmed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-06-29 15:16:53
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[geoffk at gcc dot gnu dot org]

------- Additional Comments From geoffk at gcc dot gnu dot org  2004-06-30 08:50 -------
This is just like a similar problem I'm fixing on Darwin.  The problem is that the unwind info for 
parseDouble is not completely correct in the presence of asynchronous exceptions; there's one point at 
which the link register is saved in a GPR (and no longer in the link register) but the unwind info doesn't 
reflect this.  Patch to follow.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[law at redhat dot com]

------- Additional Comments From law at redhat dot com  2004-06-30 15:57 -------
Subject: Re:  [3.5 Regression] Throw_2 test
	case does not work anymore.

On Wed, 2004-06-30 at 02:50, geoffk at gcc dot gnu dot org wrote:
> ------- Additional Comments From geoffk at gcc dot gnu dot org  2004-06-30 08:50 -------
> This is just like a similar problem I'm fixing on Darwin.  The problem is that the unwind info for 
> parseDouble is not completely correct in the presence of asynchronous exceptions; there's one point at 
> which the link register is saved in a GPR (and no longer in the link register) but the unwind info doesn't 
> reflect this.  Patch to follow.
When you've got something to poke at, feel free to send it along.

I guess your explanation makes sense -- the patch which exposed this
bug can certainly change how values are coalesced in the out-of-ssa
pass.  Changing the order of coalescing in turn can change the set
of copies which appear at the RTL level, which could in turn cause
the LR->GPR, use LR to hold some other value behavior we're seeing.
Jeff



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-07-01 04:09 -------
Subject: Bug 15813

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	geoffk@gcc.gnu.org	2004-07-01 04:09:07

Modified files:
	gcc            : ChangeLog Makefile.in dwarf2out.c unwind-dw2.c 
	gcc/config/rs6000: darwin.h rs6000.c t-darwin 
	gcc/testsuite  : ChangeLog 
	gcc/testsuite/gcc.dg: cleanup-10.c cleanup-11.c cleanup-5.c 
	                      cleanup-8.c cleanup-9.c 
	libjava        : ChangeLog configure configure.host configure.in 
Added files:
	gcc            : unwind-dw2.h 
	gcc/config/rs6000: darwin-fallback.c 
	libjava/include: darwin-signal.h 

Log message:
	Index: gcc/ChangeLog
	2004-06-28  Geoffrey Keating  <geoffk@apple.com>
	Andreas Tobler  <a.tobler@schweiz.ch>
	
	PR 15813
	* dwarf2out.c (reg_save): Output DW_CFA_same_value when a
	register is saved in itself.
	(initial_return_save): If the return address is a register,
	it's already there, don't bother to mention it in the CFI.
	(struct queued_reg_save): Add field saved_reg.
	(struct reg_saved_in_data): New.
	(regs_saved_in_regs): New.
	(num_regs_saved_in_regs): New.
	(queue_reg_save): Add extra parameter to specify register saved
	in register.  Remove duplicate entries from queue.  Add comment
	for function.
	(flush_queued_reg_saves): Handle registers saved in registers.
	Update regs_saved_in_regs.  Add comment for function.
	(clobbers_queued_reg_save): Add comment for function.  Allow
	for regs_saved_in_regs.
	(reg_saved_in): New.
	(dwarf2out_frame_debug_expr): Handle saving registers in other
	registers.
	(dwarf2out_frame_debug): Reset regs_saved_in_regs.
	* unwind-dw2.c (execute_cfa_program): Correct handling of
	DW_CFA_same_value.  Add FIXME comment about incorrect implementation
	of DW_CFA_restore_extended.
	* config/rs6000/rs6000.c (rs6000_emit_prologue): Let
	dwarf2out_frame_debug_expr see instructions that save registers
	in other registers or save those other registers in memory.
	
	* unwind-dw2.c (DWARF_FRAME_REGISTERS): Move to unwind-dw2.h.
	(_Unwind_FrameState): Likewise.
	* unwind-dw2.h: New.
	* Makefile.in (LIB2ADDEHDEP): Add unwind-dw2.h.
	* config/rs6000/darwin-fallback.c: New file.
	* config/rs6000/darwin.h (MD_FALLBACK_FRAME_STATE_FOR): Define.
	* config/rs6000/t-darwin (LIB2FUNCS_EXTRA): Add darwin-fallback.o.
	
	Index: gcc/testsuite/ChangeLog
	2004-06-26  Geoffrey Keating  <geoffk@apple.com>
	Andreas Tobler  <a.tobler@schweiz.ch>
	
	* gcc.dg/cleanup-10.c: Run on all Linux platforms and powerpc-darwin.
	Use SA_RESETHAND rather than SA_ONESHOT.  Trap SIGBUS as well
	as SIGSEGV.
	* gcc.dg/cleanup-11.c: Likewise.
	* gcc.dg/cleanup-8.c: Likewise.
	* gcc.dg/cleanup-9.c: Likewise.
	* gcc.dg/cleanup-5.c: Run on all platforms.
	
	Index: libjava/ChangeLog
	2004-06-26  Geoffrey Keating  <geoffk@apple.com>
	Andreas Tobler  <a.tobler@schweiz.ch>
	
	* configure.host (powerpc-*-darwin*): New case, define
	can_unwind_signal.
	* configure.in (*-*-darwin*): New case, point to darwin-signal.h.
	* configure: Regenerate.
	* include/darwin-signal.h: New.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/unwind-dw2.h.diff?cvsroot=gcc&r1=NONE&r2=2.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.4232&r2=2.4233
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/Makefile.in.diff?cvsroot=gcc&r1=1.1315&r2=1.1316
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/dwarf2out.c.diff?cvsroot=gcc&r1=1.523&r2=1.524
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/unwind-dw2.c.diff?cvsroot=gcc&r1=1.41&r2=1.42
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/rs6000/darwin-fallback.c.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/rs6000/darwin.h.diff?cvsroot=gcc&r1=1.53&r2=1.54
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/rs6000/rs6000.c.diff?cvsroot=gcc&r1=1.657&r2=1.658
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/rs6000/t-darwin.diff?cvsroot=gcc&r1=1.12&r2=1.13
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.3927&r2=1.3928
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/cleanup-10.c.diff?cvsroot=gcc&r1=1.1&r2=1.2
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/cleanup-11.c.diff?cvsroot=gcc&r1=1.1&r2=1.2
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/cleanup-5.c.diff?cvsroot=gcc&r1=1.3&r2=1.4
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/cleanup-8.c.diff?cvsroot=gcc&r1=1.2&r2=1.3
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/cleanup-9.c.diff?cvsroot=gcc&r1=1.3&r2=1.4
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/libjava/ChangeLog.diff?cvsroot=gcc&r1=1.2888&r2=1.2889
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/libjava/configure.diff?cvsroot=gcc&r1=1.208&r2=1.209
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/libjava/configure.host.diff?cvsroot=gcc&r1=1.57&r2=1.58
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/libjava/configure.in.diff?cvsroot=gcc&r1=1.185&r2=1.186
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/libjava/include/darwin-signal.h.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813

[Bug tree-optimization/15813] [3.5 Regression] Throw_2 test case does not work anymore.

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-07-02 17:26 -------
Fixed.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15813