target/9090: arm ICE with >= -O2; regression from 2.95

target/9090: arm ICE with >= -O2; regression from 2.95

[Matthias Klose]

>Number:         9090
>Category:       target
>Synopsis:       arm ICE with >= -O2; regression from 2.95
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          ice-on-legal-code
>Submitter-Id:   net
>Arrival-Date:   Sun Dec 29 07:36:01 PST 2002
>Closed-Date:
>Last-Modified:
>Originator:     James Troup <james@nocrew.org>
>Release:        3.2.1 (Debian) (Debian unstable)
>Organization:
The Debian Project
>Environment:
System: Debian GNU/Linux (unstable)
Architecture: arm
host: arm-linux
Configured with: /home/packages/gcc/3.2/gcc-3.2-3.2.1ds5/src/configure -v --enable-languages=c,c++,java,f77,proto,objc --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --with-gxx-include-dir=/usr/include/c++/3.2 --enable-shared --with-system-zlib --enable-nls --without-included-gettext --enable-__cxa_atexit --enable-clocale=gnu --enable-java-gc=boehm --enable-objc-gc arm-linux
>Description:
[ Reported to the Debian BTS as report #168086.
  Please CC 168086@bugs.debian.org on replies.
  Log of report can be found at http://bugs.debian.org/168086 ]

The attached testcase ICEs on arm-linux with 3.2 branch 20021212 (-O2 
only, not -O). As the 3.3 branch doesn't bootstrap, I cannot check with 3.3. 
g++-2.95 did work ok, so this is a regression. 
 
$ gcc-3.2  -O2 -c bug-168086-3.2.ii  
GLinkGLvar/FGLvarHigh.C: In member function `virtual void  
   FGLvarHigh::xIdentify(short int)': 
GLinkGLvar/FGLvarHigh.C:736: Internal compiler error in elimination_effects, at   reload1.c:2837 
Please submit a full bug report, 
with preprocessed source if appropriate. 
See <URL:http://www.gnu.org/software/gcc/bugs.html> for instructions. 
 
$ gcc-3.2  -O -c bug-168086-3.2.ii  
$ 

>How-To-Repeat:
	
>Fix:
	
>Release-Note:
>Audit-Trail:
>Unformatted:

Re: target/9090: arm ICE with >= -O2; regression from 2.95

[rearnsha]

Synopsis: arm ICE with >= -O2; regression from 2.95

State-Changed-From-To: open->analyzed
State-Changed-By: rearnsha
State-Changed-When: Sat Jan 11 09:04:58 2003
State-Changed-Why:
    The cause of PR 9090 comes from the fact that when we have
    
    (insn 127 126 128 1 0x40c653c8 (set (reg:SI 1 r1)
            (reg:SI 59)) 161 {*arm_movsi_insn} (nil)
        (expr_list:REG_EQUAL (addressof:SI (reg/v:DI 56) 55 0x40aa48c0)
            (nil)))
    
    purge_addressof_1, when applied to the notes fails to purge the address, 
    instead giving just
    
    (insn 127 126 128 1 0x40c653c8 (set (reg:SI 1 r1)
            (reg:SI 59)) 161 {*arm_movsi_insn} (nil)
        (expr_list:REG_EQUAL (addressof:SI (mem/s:DI (plus:SI (reg/f:SI 25 sfp)
                        (const_int -16 [0xfffffff0])) [14 S8 A32]) 55 
    0x40aa48c0)
    
    That is, it's pushed the referenced register onto the stack, but then 
    failed to simplify the resulting addressof expression.  However, the patch 
    seems so simple that I'm concerned that I've missed somthing obvious.
    
    Would someone care to comment?
    
    No regressions shown by an arm-elf cross build.
    
    <date>  Richard Earnshaw  <rearnsha@arm.com>
    
    	* function.c (purge_addressof_1): After pushing an addressed register
    	onto the stack, simplify the result.
    
    
    
    2.	addressof.patch	(text/x-patch)
    This is a text/x-patch
    It might be displayable with metamail.	(Invoke menu with right button.)
    Index: function.c
    ===================================================================
    RCS file: /cvs/gcc/gcc/gcc/function.c,v
    retrieving revision 1.392
    diff -p -r1.392 function.c
    *** function.c	10 Jan 2003 02:22:01 -0000	1.392
    --- function.c	11 Jan 2003 16:57:16 -0000
    *************** purge_addressof_1 (loc, insn, force, sto
    *** 3007,3016 ****
            rtx sub, insns;
      
            if (GET_CODE (XEXP (x, 0)) != MEM)
    ! 	{
    ! 	  put_addressof_into_stack (x, ht);
    ! 	  return true;
    ! 	}
      
            /* We must create a copy of the rtx because it was created by
      	 overwriting a REG rtx which is always shared.  */
    --- 3007,3013 ----
            rtx sub, insns;
      
            if (GET_CODE (XEXP (x, 0)) != MEM)
    ! 	put_addressof_into_stack (x, ht);
      
            /* We must create a copy of the rtx because it was created by
      	 overwriting a REG rtx which is always shared.  */
    

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=9090