Re: Contributing FRV

[Graham Stott <graham dot stott at btinternet dot com>]

Graham Stott wrote:
> Graham Stott wrote:
>
> > Graham Stott wrote:
> > Time to run the testsuite to check my patches :-)
> >
> > Graham
> 1st problem testcase gcc.c-torture/compile/20000326-2.c is referencing
> a deleted label within rtl this shows up as an enable checking failure
> in mark_label_ref. Obviously this label should not have been deleted,
> time to figure out why it was thought safe to delete the label
> (insn 14 37 15 (nil) (set (reg:SI 9 gr9)^M
>         (label_ref:SI [11 deleted])) 7 {*movsi_internal} (nil)^M
>     (insn_list:REG_LABEL 11 (nil)))^M
> ^M
> (call_insn 15 14 29 (nil) (parallel [^M
>             (set (reg:SI 8 gr8)^M
>                 (call (mem:QI (symbol_ref:SI ("printk")) [0 S1 A8])^M
>                     (const_int 0 [0x0])))^M
>             (use (const_int 0 [0x0]))^M
>             (clobber (reg:SI 161 lr))^M
>         ]) 156 {call_value_internal} (nil)^M
>     (nil)^M
>     (expr_list (use (reg:SI 9 gr9))^M
>         (expr_list (use (reg:SI 8 gr8))^M
>             (nil))))^M
Looking at the testcase it's one of those ucky uses of &&label which
because the label isn't a goto/jump target it gets deleted and changed
into a NOTE_DELETED_LABEL_INSN.

I'm not sure why the mark_label_ref enable-checking abort triggers on FRV
and not on x86.

Graham