Re: RTL sharing tester (for testing)

[Jan Hubicka <hubicka at ucw dot cz>]

> Jan Hubicka <jh@suse.cz> wrote:
> > The problem with is that the nature of problem is very target specific.
> > All the remaining problems are either very weird side cases, or target
> > issues.  This is why I am posting the patch so people can test it on
> > their favorite development platform.
> 
> FYI, I've tried your tester on SH and got an error which looks
> similar to PPC's one:
> 
> ../../../../../../TMP/trunk/libjava/classpath/native/fdlibm/e_exp.c:164: error: invalid rtl sharing found in the insn
> (insn:HI 207 303 256 24 ../../../../../../TMP/trunk/libjava/classpath/native/fdlibm/e_exp.c:162 (parallel [
>             (set (reg/v:DF 76 fr12 [orig:180 x ] [180])
>                 (mult:DF (reg/v:DF 76 fr12 [orig:180 x ] [180])
>                     (reg:DF 66 fr2)))
>             (use (reg/v:PSI 151 ))
>         ]) 340 {muldf3_i} (expr_list:REG_DEAD (reg/v:PSI 151 )
>         (expr_list:REG_DEAD (subreg:DF (reg/v:DI 1 r1 [orig:178 sh_u ] [178]) 0)            (expr_list:REG_DEAD (reg:DF 66 fr2)
>                 (expr_list:REG_EQUAL (mult:DF (subreg:DF (reg/v:DI 1 r1 [orig:178 sh_u ] [178]) 0)
>                         (const_double:DF 9.33263618503218878990089544723817169617091446372e-302 [0x0.8p-999]))
>                     (nil))))))
> ../../../../../../TMP/trunk/libjava/classpath/native/fdlibm/e_exp.c:164: error: shared rtx
> (subreg:DF (reg/v:DI 1 r1 [orig:178 sh_u ] [178]) 0)
> ../../../../../../TMP/trunk/libjava/classpath/native/fdlibm/e_exp.c:164: internal compiler error: internal consistency failure
> 
> and SH bootstraps without Java.  As an experiment, I've tried a one
> liner:
> 
> --- ORIG/trunk/gcc/df-problems.c        2007-06-22 09:16:00.000000000 +0900
> +++ TMP/trunk/gcc/df-problems.c 2007-06-26 13:53:47.000000000 +0900
> @@ -3713,7 +3713,8 @@ df_set_note (enum reg_note note_type, rt
>        }
> 
>    /* Did not find the note.  */
> -  REG_NOTES (insn) = alloc_EXPR_LIST (note_type, reg, REG_NOTES (insn));
> +  REG_NOTES (insn) = alloc_EXPR_LIST (note_type, copy_rtx (reg),
> +                                     REG_NOTES (insn));
>    return old;
>  }
> 
> Then the above error on SH went away.  This one liner may be
> an overkill, but I hope that this experiment will help experts
> to narrow down the problem.

Actually this seems like a correct fix for me. In fact I was just
looking into instance of identical problem on PPC bootstrap (different
file, same directory of libjava :).  We will copy iff the reg actually
is subreg and in that case we are required to copy.

I am re-testing PPC targets with this fix in now.

Thanks a lot!
Honza
> 
> Regards,
> 	kaz