Re: The recent regclass.c change

[Bernd Schmidt <crux at Pool dot Informatik dot RWTH-Aachen dot DE>]

> 
>   > This patch undoes the regclass.c change and disables duplicate_loop_exit_te  > st
>   > if a REG_WAS_0 note is found.
> Instead of disabling duplicate_loop_exit_test, just remove the REG_WAS_0 note.
> That's likely less intrusive from an optimization standpoint.  Note that cse2
> might put it back, but it should do so in a safe way :-)
> 
>   > Do we actually need REG_WAS_0 still?  As far as I can see it is used in
>   > three machine descriptions to turn movl 0,reg to incl reg.  The same
>   > transformation ought to be done by the new reload_move2add pass, oughtn't
>   > it?
> Excellent question.  I'd say it's probably no longer needed, but we'd have to
> do a little testing.
> 
> In theory, the code in i386.md which uses REG_WAS_0 should never trigger
> anymore (since the constant load should have been replaced with a mov2add
> sequence).  I'd say pick a few big hunks of code (compiler, glibc, 2.1.xxx
> kernel?) and see if the REG_WAS_0 code in i386.md ever triggers.  If not, then
> I'd say kill REG_WAS_0.

I tried with the kernel, and the REG_WAS_0 optimization does trigger still.
It appears that the move2add optimization requires RTX_COSTS to be tweaked.
This ought to be investigated, but for now I'm afraid I might break the
other cost calculations if I go mucking about with that macro, so this fix
here seems like a safer solution for now.

Note: the patch is untested.

Bernd

	* jump.c (duplicate_loop_exit_test): Strip REG_WAS_0 notes off all
	insns we're going to copy.
	* regclass.c (reg_scan_mark_refs): Don't test X for NULL_RTX.

Index: jump.c
===================================================================
RCS file: /egcs/carton/cvsfiles/egcs/gcc/jump.c,v
retrieving revision 1.37
diff -u -p -r1.37 jump.c
--- jump.c	1998/09/02 14:49:17	1.37
+++ jump.c	1998/10/07 13:19:36
@@ -2434,8 +2434,7 @@ duplicate_loop_exit_test (loop_start)
 	 has a REG_RETVAL or REG_LIBCALL note (hard to adjust)
 	 is a NOTE_INSN_LOOP_BEG because this means we have a nested loop
 	 is a NOTE_INSN_BLOCK_{BEG,END} because duplicating these notes
-	      are not valid
-	    
+	      is not valid.
 
      We also do not do this if we find an insn with ASM_OPERANDS.  While
      this restriction should not be necessary, copying an insn with
@@ -2480,6 +2479,10 @@ duplicate_loop_exit_test (loop_start)
 	  break;
 	case JUMP_INSN:
 	case INSN:
+	  /* The code below would grossly mishandle REG_WAS_0 notes,
+	     so get rid of them here.  */
+	  while ((p = find_reg_note (insn, REG_WAS_0, NULL_RTX)) != 0)
+	    remove_note (insn, p);
 	  if (++num_insns > 20
 	      || find_reg_note (insn, REG_RETVAL, NULL_RTX)
 	      || find_reg_note (insn, REG_LIBCALL, NULL_RTX)
Index: regclass.c
===================================================================
RCS file: /egcs/carton/cvsfiles/egcs/gcc/regclass.c,v
retrieving revision 1.31
diff -u -p -r1.31 regclass.c
--- regclass.c	1998/10/02 01:34:34	1.31
+++ regclass.c	1998/10/07 13:19:51
@@ -1995,13 +1995,6 @@ reg_scan_mark_refs (x, insn, note_flag, 
   register rtx dest;
   register rtx note;
 
-  /* This can happen when scanning insns referenced by certain notes.
-
-     It is unclear if we should be scanning such insns; until someone can
-     say for sure this seems like the safest fix.  */
-  if (x == NULL_RTX)
-    return;
-
   code = GET_CODE (x);
   switch (code)
     {