Mainline bootstrap fails on powerpc-apple-darwin7.5.0 with --enable-checking=all

Mainline bootstrap fails on powerpc-apple-darwin7.5.0 with --enable-checking=all

[Matt Austern]

The failure is:
./genattrtab ../../fsf.u/gcc/config/rs6000/rs6000.md > tmp-attrtab.c
genattrtab: Internal error: RTL check: expected elt 0 type 's' or 'S', 
have 'E' (rtx define_peephole) in write_attr_case, at genattrtab.c:3908
make[2]: *** [s-attrtab] Error 1
make[1]: *** [stage1_build] Error 2
make: *** [bootstrap] Error 2

Is this known and/or expected?

			--Matt

Re: Mainline bootstrap fails on powerpc-apple-darwin7.5.0 with --enable-checking=all

[Andrew Pinski]

On Sep 8, 2004, at 11:18 AM, Matt Austern wrote:

> The failure is:
> ./genattrtab ../../fsf.u/gcc/config/rs6000/rs6000.md > tmp-attrtab.c
> genattrtab: Internal error: RTL check: expected elt 0 type 's' or 'S', 
> have 'E' (rtx define_peephole) in write_attr_case, at 
> genattrtab.c:3908
> make[2]: *** [s-attrtab] Error 1
> make[1]: *** [stage1_build] Error 2
> make: *** [bootstrap] Error 2
>
> Is this known and/or expected?

I thought that was fixed by:
2004-08-31  Richard Sandiford  <rsandifo@redhat.com>

         * read-rtl.c: Disable RTL checking.
         (apply_macro_to_rtx): Use XTMPL to access 'T' fields.


Maybe there is another problem.  This would be a RTL checking failure.


-- Pinski

Re: Mainline bootstrap fails on powerpc-apple-darwin7.5.0 with --enable-checking=all

[Graham Stott]

Andrew,

 --- Andrew Pinski <pinskia@physics.uc.edu> wrote: 
>
[snip] 
> 
> I thought that was fixed by:
> 2004-08-31  Richard Sandiford  <rsandifo@redhat.com>
> 
>          * read-rtl.c: Disable RTL checking.
>          (apply_macro_to_rtx): Use XTMPL to access 'T' fields.
> 
> 
> Maybe there is another problem.  This would be a RTL checking failure.
> 
> 
> -- Pinski
>

Yes it's a different problem, probably fallout from Richard's eariler patch
on 2004-08-13

Graham

Re: Mainline bootstrap fails on powerpc-apple-darwin7.5.0 with --enable-checking=all

[Richard Sandiford]

Graham Stott <graham.stott@btinternet.com> writes:
> Yes it's a different problem, probably fallout from Richard's eariler patch
> on 2004-08-13

Me again eh? ;(  Oh well.  I'm really sorry for the breakage.  The way
things are going, I think it's time I jumped off a cliff and left gcc
development to people who are vaguely competent. ;/

Anyway, the attached patch should fix things.  It at least gets past
the rtl checking failure in genattrtab for a cross-compile.

Hmm.  Maybe it's time to reconsider including rtl checking in the
default --enable-checking flags?  Clearly I should be doing that
locally anyway if I'm making stupid mistakes like this.

Richard


	* genattrtab.c (write_attr_case, write_const_num_delay_slots): Check
	for DEFINE_PEEPHOLEs before writing out the insn name.

Index: genattrtab.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/genattrtab.c,v
retrieving revision 1.151
diff -c -p -F^\([(a-zA-Z0-9_]\|#define\) -r1.151 genattrtab.c
*** genattrtab.c	13 Aug 2004 16:43:04 -0000	1.151
--- genattrtab.c	8 Sep 2004 21:46:40 -0000
*************** write_attr_case (struct attr_desc *attr,
*** 3905,3911 ****
  	  {
  	    write_indent (indent);
  	    printf ("case %d:  /* %s */\n",
! 		    ie->def->insn_code, XSTR (ie->def->def, 0));
  	  }
      }
    else
--- 3905,3914 ----
  	  {
  	    write_indent (indent);
  	    printf ("case %d:  /* %s */\n",
! 		    ie->def->insn_code,
! 		    GET_CODE (ie->def->def) == DEFINE_PEEPHOLE
! 		    ? "peephole"
! 		    : XSTR (ie->def->def, 0));
  	  }
      }
    else
*************** write_const_num_delay_slots (void)
*** 4383,4389 ****
  	      for (ie = av->first_insn; ie; ie = ie->next)
  		if (ie->def->insn_code != -1)
  		  printf ("    case %d:  /* %s */\n",
! 			  ie->def->insn_code, XSTR (ie->def->def, 0));
  	      printf ("      return 0;\n");
  	    }
  	}
--- 4386,4394 ----
  	      for (ie = av->first_insn; ie; ie = ie->next)
  		if (ie->def->insn_code != -1)
  		  printf ("    case %d:  /* %s */\n",
! 			  GET_CODE (ie->def->def) == DEFINE_PEEPHOLE
! 			  ? "peephole"
! 			  : XSTR (ie->def->def, 0));
  	      printf ("      return 0;\n");
  	    }
  	}

Re: Mainline bootstrap fails on powerpc-apple-darwin7.5.0 with --enable-checking=all

[Richard Sandiford]

Richard Sandiford <rsandifo@redhat.com> writes:
> Anyway, the attached patch should fix things.  It at least gets past
> the rtl checking failure in genattrtab for a cross-compile.

But, to add insult to injury, the write_const_num_delay_slots part
was clearly bogus.  Sigh.

Fixed version attached.

Richard


	* genattrtab.c (write_attr_case, write_const_num_delay_slots): Check
	for DEFINE_PEEPHOLEs before writing out the insn name.

Index: genattrtab.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/genattrtab.c,v
retrieving revision 1.151
diff -c -p -F^\([(a-zA-Z0-9_]\|#define\) -r1.151 genattrtab.c
*** genattrtab.c	13 Aug 2004 16:43:04 -0000	1.151
--- genattrtab.c	8 Sep 2004 22:07:05 -0000
*************** write_attr_case (struct attr_desc *attr,
*** 3905,3911 ****
  	  {
  	    write_indent (indent);
  	    printf ("case %d:  /* %s */\n",
! 		    ie->def->insn_code, XSTR (ie->def->def, 0));
  	  }
      }
    else
--- 3905,3914 ----
  	  {
  	    write_indent (indent);
  	    printf ("case %d:  /* %s */\n",
! 		    ie->def->insn_code,
! 		    GET_CODE (ie->def->def) == DEFINE_PEEPHOLE
! 		    ? "peephole"
! 		    : XSTR (ie->def->def, 0));
  	  }
      }
    else
*************** write_const_num_delay_slots (void)
*** 4383,4389 ****
  	      for (ie = av->first_insn; ie; ie = ie->next)
  		if (ie->def->insn_code != -1)
  		  printf ("    case %d:  /* %s */\n",
! 			  ie->def->insn_code, XSTR (ie->def->def, 0));
  	      printf ("      return 0;\n");
  	    }
  	}
--- 4386,4395 ----
  	      for (ie = av->first_insn; ie; ie = ie->next)
  		if (ie->def->insn_code != -1)
  		  printf ("    case %d:  /* %s */\n",
! 			  ie->def->insn_code,
! 			  GET_CODE (ie->def->def) == DEFINE_PEEPHOLE
! 			  ? "peephole"
! 			  : XSTR (ie->def->def, 0));
  	      printf ("      return 0;\n");
  	    }
  	}