[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

[pinskia@physics.uc.edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812


pinskia@physics.uc.edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |3.3.1

[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

[wilson at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812



------- Additional Comments From wilson at gcc dot gnu dot org  2003-06-27 07:07 -------
I built 3.0.4 and 2.95 m68k-linux cross compilers, and both failed for the
testcase same as 3.3.  This does not appear to be a gcc regression to me.

The problem stems from a patch I added in 1995, which makes
LEGITIMATE_PIC_OPERAND_P reject CONST_DOUBLEs that are in the constant pool.  It
isn't clear if that patch is still useful.  No other target does this.  I am
checking to see if I can reproduce the original problem.  Changing this may be
risky though.

I suspect that this check can be made conditional on reload_completed and/or
reload_in_progress.  I haven't looked into this yet.

[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

[wilson at tuliptree dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812



------- Additional Comments From wilson at tuliptree dot org  2003-06-30 06:30 -------
Subject: Re:  [3.3/3.4 regression] [m68k] ICE in
 extract_insn, at recog.c:2148

I googled the original testcase from 1995.
http://groups.google.com/groups?q=Andreas+Schwab+group:gnu.gcc.bug+group:gnu.gcc.bug+group:gnu.gcc.bug+group:gnu.gcc.bug+group:gnu.gcc.bug+group:gnu.gcc.bug+group:gnu.gcc.bug+group:gnu.gcc.bug&start=70&hl=en&lr=&ie=UTF-8&group=gnu.gcc.bug&selm=9511241005.AA04838%40lamothe.informatik.uni-dortmund.de&rnum=79
I don't know what I was thinking when I wrote that patch in 1995.

There is an obvious problem with the XFmode pattern here.  It calls
force_const_mem during reload which is not safe.  We either need to 
reload the address ourselves, or not call it during reload.  Fixing the 
address outselves would require secondary reloads, so that seems 
infeasible.  So the obvious solution here is to guard the call against 
reload_in_progress.

Fixing that makes  the original LEGITIMATE_PIC_OPERAND_P patch 
unnecessary.  So we can remove the CONST_DOUBLE handling from that 
macro.  While doing that, I  noticed there were about 6 duplicates of 
that macro in various m68k files, so I deleted all but one.  They used 
to be different long ago, but those differences have all since 
disappeared.  Since m68k was the only place where mem_for_const_double 
was used, we can now get rid of that function.

While looking at this, I noticed that there is a lot of confusion in the 
XFmode patterns regarding constant support. Most of them disallow 
constants in predicates and constraints, but have output templates that 
handle emitting code for constants.  This looks like it has been broken 
since the beginning.  I suspect that constant support was partially 
disabled because of a bug, and then it was accidentally left in this 
state forever.  Or maybe someone copied the DFmode patterns and did a 
poor job of editing them.  I documented this in a comment.

With those changes, we now get unrecognized insns because reload thinks
that XFmode constants are OK, but we have no patterns at all that accept 
XFmode constants.  We fix that by rejecting them in LEGITIMATE_CONSTANT_P.

The resulting patch works for this testcase.  I suspect we will get
better PIC FP code also, since we aren't needlessly forcing non-XFmode
constants into memory.  This gives a medium size patch that will need a
lot of testing, and may not be appropriate for the gcc-3.3.1 release.

I can't really test this with m68k crosses, and I don't have access to 
native m68k hardware.

Jim

2003-06-27  James E Wilson  <wilson@tuliptree.org>

	* rtl.h (mem_for_const_double): Delete prototype.
	* varasm.c (mem_for_const_double): Delete function.
	* config/m68k/hp320.h, config/m68k/linux.h, config/m68k/m68kelf.h,
	config/m68k/m68kv4.h, config/m68k/netbsd-elf.h
	(LEGITIMATE_PIC_OPERAND_P): Delete duplicate definitions.
	* config/m68k/m68k.h (LEGITIMATE_CONSTANT_P): Disallow XFmode.
	(LEGITIMATE_PIC_OPERAND_P): Delete CONST_DOUBLE tests.
	* config/m68k/m68k.md (movxf): Add reload_in_progress guard.  Add
	comment about confused support for XFmode constants.

Index: rtl.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/rtl.h,v
retrieving revision 1.419
diff -p -r1.419 rtl.h
*** rtl.h	25 Jun 2003 03:45:13 -0000	1.419
--- rtl.h	27 Jun 2003 09:19:15 -0000
*************** extern void end_full_sequence		PARAMS ((
*** 1494,1500 ****
  
  /* In varasm.c  */
  extern rtx immed_double_const		PARAMS ((HOST_WIDE_INT, HOST_WIDE_INT, enum machine_mode));
- extern rtx mem_for_const_double		PARAMS ((rtx));
  extern rtx force_const_mem		PARAMS ((enum machine_mode, rtx));
  
  /* In varasm.c  */
--- 1494,1499 ----
Index: varasm.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/varasm.c,v
retrieving revision 1.368
diff -p -r1.368 varasm.c
*** varasm.c	25 Jun 2003 22:14:26 -0000	1.368
--- varasm.c	27 Jun 2003 09:19:18 -0000
*************** record_constant_rtx (mode, x)
*** 3033,3056 ****
    return ptr;
  }
  \f
- /* Given a constant rtx X, return a MEM for the location in memory at which
-    this constant has been placed.  Return 0 if it not has been placed yet.  */
- 
- rtx
- mem_for_const_double (x)
-      rtx x;
- {
-   enum machine_mode mode = GET_MODE (x);
-   struct constant_descriptor_rtx *desc;
- 
-   for (desc = const_rtx_hash_table[const_hash_rtx (mode, x)]; desc;
-        desc = desc->next)
-     if (compare_constant_rtx (mode, x, desc))
-       return desc->rtl;
- 
-   return 0;
- }
- 
  /* Given a constant rtx X, make (or find) a memory constant for its value
     and return a MEM rtx to refer to it in memory.  */
  
--- 3033,3038 ----
Index: config/m68k/hp320.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/m68k/hp320.h,v
retrieving revision 1.26
diff -p -r1.26 hp320.h
*** config/m68k/hp320.h	19 Jun 2003 21:47:15 -0000	1.26
--- config/m68k/hp320.h	27 Jun 2003 09:19:20 -0000
*************** do { size_t i, limit = (SIZE);		\
*** 555,572 ****
  
  #endif /* not HPUX_ASM */
  
- /* In m68k svr4, a symbol_ref rtx can be a valid PIC operand if it is an
-    operand of a function call.  */
- #undef LEGITIMATE_PIC_OPERAND_P
- #define LEGITIMATE_PIC_OPERAND_P(X) \
-   ((! symbolic_operand (X, VOIDmode) \
-     && ! (GET_CODE (X) == CONST_DOUBLE && mem_for_const_double (X) != 0	\
- 	  && GET_CODE (mem_for_const_double (X)) == MEM			\
- 	  && symbolic_operand (XEXP (mem_for_const_double (X), 0),	\
- 			       VOIDmode))) 				\
-    || (GET_CODE (X) == SYMBOL_REF && SYMBOL_REF_FLAG (X))       	\
-    || PCREL_GENERAL_OPERAND_OK)
- 
  /* hpux8 and later have C++ compatible include files, so do not
     pretend they are `extern "C"'.  */
  #define NO_IMPLICIT_EXTERN_C
--- 555,560 ----
Index: config/m68k/linux.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/m68k/linux.h,v
retrieving revision 1.33
diff -p -r1.33 linux.h
*** config/m68k/linux.h	7 Jun 2003 17:11:46 -0000	1.33
--- config/m68k/linux.h	27 Jun 2003 09:19:20 -0000
*************** do {									\
*** 283,300 ****
     ? gen_rtx_REG ((MODE), 16)						\
     : gen_rtx_REG ((MODE), 0))
  
- /* In m68k svr4, a symbol_ref rtx can be a valid PIC operand if it is
-    an operand of a function call.  */
- #undef LEGITIMATE_PIC_OPERAND_P
- #define LEGITIMATE_PIC_OPERAND_P(X) \
-   ((! symbolic_operand (X, VOIDmode) \
-     && ! (GET_CODE (X) == CONST_DOUBLE && mem_for_const_double (X) != 0	\
- 	  && GET_CODE (mem_for_const_double (X)) == MEM			\
- 	  && symbolic_operand (XEXP (mem_for_const_double (X), 0),	\
- 			       VOIDmode))) 				\
-    || (GET_CODE (X) == SYMBOL_REF && SYMBOL_REF_FLAG (X))       	\
-    || PCREL_GENERAL_OPERAND_OK)
- 
  /* For m68k SVR4, structures are returned using the reentrant
     technique.  */
  #undef PCC_STATIC_STRUCT_RETURN
--- 283,288 ----
Index: config/m68k/m68k.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/m68k/m68k.h,v
retrieving revision 1.90
diff -p -r1.90 m68k.h
*** config/m68k/m68k.h	19 Jun 2003 21:47:15 -0000	1.90
--- config/m68k/m68k.h	27 Jun 2003 09:19:22 -0000
*************** __transfer_from_trampoline ()					\
*** 997,1003 ****
  /* Nonzero if the constant value X is a legitimate general operand.
     It is given that X satisfies CONSTANT_P or is a CONST_DOUBLE.  */
  
! #define LEGITIMATE_CONSTANT_P(X) 1
  
  /* Nonzero if the constant value X is a legitimate general operand
     when generating PIC code.  It is given that flag_pic is on and 
--- 997,1003 ----
  /* Nonzero if the constant value X is a legitimate general operand.
     It is given that X satisfies CONSTANT_P or is a CONST_DOUBLE.  */
  
! #define LEGITIMATE_CONSTANT_P(X) (GET_MODE (X) != XFmode)
  
  /* Nonzero if the constant value X is a legitimate general operand
     when generating PIC code.  It is given that flag_pic is on and 
*************** __transfer_from_trampoline ()					\
*** 1014,1025 ****
  #endif
  
  #define LEGITIMATE_PIC_OPERAND_P(X)	\
!   ((! symbolic_operand (X, VOIDmode)				\
!     && ! (GET_CODE (X) == CONST_DOUBLE && mem_for_const_double (X) != 0	\
! 	  && GET_CODE (mem_for_const_double (X)) == MEM			\
! 	  && symbolic_operand (XEXP (mem_for_const_double (X), 0),	\
! 			       VOIDmode))) 				\
!    || (GET_CODE (X) == SYMBOL_REF && SYMBOL_REF_FLAG (X))		\
     || PCREL_GENERAL_OPERAND_OK)
  
  /* The macros REG_OK_FOR..._P assume that the arg is a REG rtx
--- 1014,1021 ----
  #endif
  
  #define LEGITIMATE_PIC_OPERAND_P(X)	\
!   (! symbolic_operand (X, VOIDmode)				\
!    || (GET_CODE (X) == SYMBOL_REF && SYMBOL_REF_FLAG (X))	\
     || PCREL_GENERAL_OPERAND_OK)
  
  /* The macros REG_OK_FOR..._P assume that the arg is a REG rtx
Index: config/m68k/m68k.md
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/m68k/m68k.md,v
retrieving revision 1.59
diff -p -r1.59 m68k.md
*** config/m68k/m68k.md	12 Jun 2003 21:57:31 -0000	1.59
--- config/m68k/m68k.md	27 Jun 2003 09:40:12 -0000
***************
*** 999,1025 ****
    "TARGET_5200"
    "* return output_move_double (operands);")
  
  (define_expand "movxf"
    [(set (match_operand:XF 0 "nonimmediate_operand" "")
  	(match_operand:XF 1 "general_operand" ""))]
    ""
    "
  {
!   if (CONSTANT_P (operands[1]))
      {
!       operands[1] = force_const_mem (XFmode, operands[1]);
!       if (! memory_address_p (XFmode, XEXP (operands[1], 0))
! 	  && ! reload_in_progress)
! 	operands[1] = adjust_address (operands[1], XFmode, 0);
!     }
!   if (flag_pic && TARGET_PCREL && ! reload_in_progress)
!     {
!       /* Don't allow writes to memory except via a register;
! 	 the m68k doesn't consider PC-relative addresses to be writable.  */
!       if (GET_CODE (operands[0]) == MEM
! 	  && symbolic_operand (XEXP (operands[0], 0), SImode))
! 	operands[0] = gen_rtx (MEM, XFmode,
! 			       force_reg (SImode, XEXP (operands[0], 0)));
      }
  }")
  
--- 999,1033 ----
    "TARGET_5200"
    "* return output_move_double (operands);")
  
+ ;; ??? The XFmode patterns are schizophrenic about whether constants are
+ ;; allowed.  Most but not all have predicates and constraint that disallow
+ ;; constants.  Most but not all have output templates that handle constants.
+ ;; See also LEGITIMATE_CONSTANT_P.
+ 
  (define_expand "movxf"
    [(set (match_operand:XF 0 "nonimmediate_operand" "")
  	(match_operand:XF 1 "general_operand" ""))]
    ""
    "
  {
!   /* We can't rewrite operands during reload.  */
!   if (! reload_in_progress)
      {
!       if (CONSTANT_P (operands[1]))
! 	{
! 	  operands[1] = force_const_mem (XFmode, operands[1]);
! 	  if (! memory_address_p (XFmode, XEXP (operands[1], 0)))
! 	    operands[1] = adjust_address (operands[1], XFmode, 0);
! 	}
!       if (flag_pic && TARGET_PCREL)
! 	{
! 	  /* Don't allow writes to memory except via a register; the
! 	     m68k doesn't consider PC-relative addresses to be writable.  */
! 	  if (GET_CODE (operands[0]) == MEM
! 	      && symbolic_operand (XEXP (operands[0], 0), SImode))
! 	    operands[0] = gen_rtx (MEM, XFmode,
! 				   force_reg (SImode, XEXP (operands[0], 0)));
! 	}
      }
  }")
  
Index: config/m68k/m68kelf.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/m68k/m68kelf.h,v
retrieving revision 1.18
diff -p -r1.18 m68kelf.h
*** config/m68k/m68kelf.h	12 May 2003 09:51:29 -0000	1.18
--- config/m68k/m68kelf.h	27 Jun 2003 09:19:30 -0000
*************** extern int switch_table_difference_label
*** 246,261 ****
  #undef ASM_OUTPUT_BEFORE_CASE_LABEL
  #define ASM_OUTPUT_BEFORE_CASE_LABEL(FILE,PREFIX,NUM,TABLE)		\
    fprintf ((FILE), "%s&%d\n", SWBEG_ASM_OP, XVECLEN (PATTERN (TABLE), 1));
- 
- /* In m68k svr4, a symbol_ref rtx can be a valid PIC operand if it is an
-    operand of a function call.  */
- #undef LEGITIMATE_PIC_OPERAND_P
- 
- #define LEGITIMATE_PIC_OPERAND_P(X)	\
-   (! symbolic_operand (X, VOIDmode)				\
-    || (GET_CODE (X) == SYMBOL_REF && SYMBOL_REF_FLAG (X))	\
-    || PCREL_GENERAL_OPERAND_OK)
- 
  /* end of stuff from m68kv4.h */
  
  #undef SGS_CMP_ORDER
--- 246,251 ----
Index: config/m68k/m68kv4.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/m68k/m68kv4.h,v
retrieving revision 1.22
diff -p -r1.22 m68kv4.h
*** config/m68k/m68kv4.h	17 May 2003 21:57:38 -0000	1.22
--- config/m68k/m68kv4.h	27 Jun 2003 09:19:30 -0000
*************** int switch_table_difference_label_flag;
*** 277,294 ****
  #define ASM_OUTPUT_BEFORE_CASE_LABEL(FILE,PREFIX,NUM,TABLE)		\
    fprintf ((FILE), "%s&%d\n", SWBEG_ASM_OP, XVECLEN (PATTERN (TABLE), 1));
  
- /* In m68k svr4, a symbol_ref rtx can be a valid PIC operand if it is an
-    operand of a function call.  */
- #undef LEGITIMATE_PIC_OPERAND_P
- #define LEGITIMATE_PIC_OPERAND_P(X) \
-   ((! symbolic_operand (X, VOIDmode) \
-     && ! (GET_CODE (X) == CONST_DOUBLE && mem_for_const_double (X) != 0	\
- 	  && GET_CODE (mem_for_const_double (X)) == MEM			\
- 	  && symbolic_operand (XEXP (mem_for_const_double (X), 0),	\
- 			       VOIDmode))) 				\
-    || (GET_CODE (X) == SYMBOL_REF && SYMBOL_REF_FLAG (X))       	\
-    || PCREL_GENERAL_OPERAND_OK)
- 
  /* Output assembler code for a block containing the constant parts
     of a trampoline, leaving space for the variable parts.  */
  
--- 277,282 ----
Index: config/m68k/netbsd-elf.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/m68k/netbsd-elf.h,v
retrieving revision 1.14
diff -p -r1.14 netbsd-elf.h
*** config/m68k/netbsd-elf.h	3 Jun 2003 09:06:54 -0000	1.14
--- config/m68k/netbsd-elf.h	27 Jun 2003 09:19:30 -0000
*************** while (0)
*** 385,404 ****
  #define BIGGEST_ALIGNMENT 64
  
  
- /* In m68k svr4, a symbol_ref rtx can be a valid PIC operand if it is
-    an operand of a function call. */
- 
- #undef LEGITIMATE_PIC_OPERAND_P
- #define LEGITIMATE_PIC_OPERAND_P(X)					\
-   ((! symbolic_operand (X, VOIDmode)					\
-     && ! (GET_CODE (X) == CONST_DOUBLE && mem_for_const_double (X)	\
- 	  && GET_CODE (mem_for_const_double (X)) == MEM			\
- 	  && symbolic_operand (XEXP (mem_for_const_double (X), 0),	\
- 			       VOIDmode)))				\
-    || (GET_CODE (X) == SYMBOL_REF && SYMBOL_REF_FLAG (X))		\
-    || PCREL_GENERAL_OPERAND_OK)
- 
- 
  /* For m68k SVR4, structures are returned using the reentrant
     technique. */
  
--- 385,390 ----

[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

[doko at cs dot tu-berlin dot de]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812



------- Additional Comments From doko at cs dot tu-berlin dot de  2003-07-04 20:30 -------
Subject: Re: Bug#177840:  [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

wilson at tuliptree dot org writes:
> The resulting patch works for this testcase.  I suspect we will get
> better PIC FP code also, since we aren't needlessly forcing non-XFmode
> constants into memory.  This gives a medium size patch that will need a
> lot of testing, and may not be appropriate for the gcc-3.3.1 release.
> 
> I can't really test this with m68k crosses, and I don't have access to 
> native m68k hardware.

bootstrapped on m68k-linux and i386-linux (together with your patch
for PR10021) with no new regressions on the 3.3 branch.

[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

[cvs-commit at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812



------- Additional Comments From cvs-commit at gcc dot gnu dot org  2003-07-07 03:42 -------
Subject: Bug 9812

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	wilson@gcc.gnu.org	2003-07-07 03:42:23

Modified files:
	gcc            : ChangeLog rtl.h varasm.c 
	gcc/config/m68k: hp320.h linux.h m68k.h m68k.md m68kelf.h 
	                 m68kv4.h netbsd-elf.h 

Log message:
	PR optimization/9812
	* rtl.h (mem_for_const_double): Delete prototype.
	* varasm.c (mem_for_const_double): Delete function.
	* config/m68k/hp320.h, config/m68k/linux.h, config/m68k/m68kelf.h,
	config/m68k/m68kv4.h, config/m68k/netbsd-elf.h
	(LEGITIMATE_PIC_OPERAND_P): Delete duplicate definitions.
	* config/m68k/m68k.h (LEGITIMATE_CONSTANT_P): Disallow XFmode.
	(LEGITIMATE_PIC_OPERAND_P): Delete CONST_DOUBLE tests.
	* config/m68k/m68k.md (movxf): Add reload_in_progress guard.  Add
	comment about confused support for XFmode constants.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.368&r2=2.369
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/rtl.h.diff?cvsroot=gcc&r1=1.426&r2=1.427
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/varasm.c.diff?cvsroot=gcc&r1=1.370&r2=1.371
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/m68k/hp320.h.diff?cvsroot=gcc&r1=1.26&r2=1.27
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/m68k/linux.h.diff?cvsroot=gcc&r1=1.33&r2=1.34
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/m68k/m68k.h.diff?cvsroot=gcc&r1=1.91&r2=1.92
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/m68k/m68k.md.diff?cvsroot=gcc&r1=1.60&r2=1.61
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/m68k/m68kelf.h.diff?cvsroot=gcc&r1=1.18&r2=1.19
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/m68k/m68kv4.h.diff?cvsroot=gcc&r1=1.22&r2=1.23
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/m68k/netbsd-elf.h.diff?cvsroot=gcc&r1=1.14&r2=1.15

[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

[wilson at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812



------- Additional Comments From wilson at gcc dot gnu dot org  2003-07-07 03:55 -------
I have added the patch to the mainline.  I have not added the patch the 3.3
branch because it is not a gcc regression according to my analysis.  Also, I am
concerned about the size of the patch.

[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

[mmitchel at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812


mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.3.1                       |3.3.2


------- Additional Comments From mmitchel at gcc dot gnu dot org  2003-07-11 22:58 -------
Postponed until GCC 3.3.2; m68k is not a primary platform.

[Bug optimization/9812] [m68k] ICE in extract_insn, at recog.c:2148

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
            Summary|[3.3/3.4 regression] [m68k] |[m68k] ICE in extract_insn,
                   |ICE in extract_insn, at     |at recog.c:2148
                   |recog.c:2148                |
   Target Milestone|3.3.2                       |3.4


------- Additional Comments From pinskia at physics dot uc dot edu  2003-08-05 04:29 -------
Closing as this is not a regression accroding to  Jim Wilson who caused the problem in 
the first place (back in 1995).