mcore-elf on 3.3 branch

mcore-elf on 3.3 branch

[Joel Sherrill]

Hi,

mcore-elf has a problem on the 3.3 branch.  I currently have a 
cc1 running with over 17 minutes CPU time on a 2.4 Ghz P4 
compiling _lshrdi3.o.

Nick have you tried to build this target lately?

Is this a bootstrap failure?  I don't know the best way to file this
one.

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
Support Available                (256) 722-9985

Re: mcore-elf on 3.3 branch

[Nick Clifton]

Hi Joel,

> mcore-elf has a problem on the 3.3 branch.  I currently have a 
> cc1 running with over 17 minutes CPU time on a 2.4 Ghz P4 
> compiling _lshrdi3.o.

Still investigating.  Problem is in the second if-conversion pass.
Applying the workaround patch below allows the compiler to build, (and
possibly bootstrap, I have not checked).

Cheers
        Nick

2003-05-13  Nick Clifton  <nickc@redhat.com>

	* config/mcore/mcore.h (OPTIMIZATION_OPTIONS): Disable second
	if-conversion pass.

Index: gcc/config/mcore/mcore.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/mcore/mcore.h,v
retrieving revision 1.41.4.1
diff -c -3 -p -w -r1.41.4.1 mcore.h
*** gcc/config/mcore/mcore.h	29 Apr 2003 14:31:53 -0000	1.41.4.1
--- gcc/config/mcore/mcore.h	13 May 2003 17:20:52 -0000
*************** extern const char * mcore_stack_incremen
*** 194,199 ****
--- 194,200 ----
      {						\
        flag_no_function_cse = 1;			\
        flag_omit_frame_pointer = 1;		\
+       flag_if_conversion2 = 0; \
  						\
        if (LEVEL >= 2)				\
          {					\
        

Re: mcore-elf on 3.3 branch

[Nick Clifton]

Hi Joel,

> mcore-elf has a problem on the 3.3 branch.  I currently have a 
> cc1 running with over 17 minutes CPU time on a 2.4 Ghz P4 
> compiling _lshrdi3.o.

... and here is a proper patch to fix the problem.

A jump insn was being generated by a call to emit_insn() which was
confusing find_basic_blocks(), which then caused the target of the
jump to be deleted (for being in an unreachable block), which then
confused the if-conversion code.

I am going to apply this patch to the mainline sources, but since the
3.3 branch is frozen it will have to wait until after the release.

Cheers
        Nick

2003-05-13  Nick Clifton  <nickc@redhat.com>

	* config/mcore/mcore.md (jump): Use emit_jump_insn.

        
Index: gcc/config/mcore/mcore.md
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/mcore/mcore.md,v
retrieving revision 1.9
diff -c -3 -p -w -r1.9 mcore.md
*** gcc/config/mcore/mcore.md	20 Dec 2002 04:30:51 -0000	1.9
--- gcc/config/mcore/mcore.md	13 May 2003 18:34:40 -0000
***************
*** 1896,1902 ****
   ""
   "
  {
!   emit_insn (gen_jump_real (operand0));
    DONE;
  }
  ")
--- 1896,1902 ----
   ""
   "
  {
!   emit_jump_insn (gen_jump_real (operand0));
    DONE;
  }
  ")