Re: [patch] sh: Move predicates to predicates.md. (take 2)

[Joern RENNECKE <joern dot rennecke at st dot com>]

Kazu Hirata wrote:

> Hi Joern,
>
>  
>
> > dp-bit.c:952: error: unrecognizable insn:
> > (insn 506 505 511 16 dp-bit.c:863 (set (reg:DI 3 r3)
> > (mult:DI (zero_extend:DI (truncate:SI (reg:DI 7 r7 [orig:177 D.2362 ] 
> > [177])))
> > (zero_extend:DI (truncate:SI (reg:DI 2 r2 [orig:176 D.2363 ] [176]))))) 
> > -1 (nil)
> > (nil))
> > dp-bit.c:952: internal compiler error: in extract_insn, at recog.c:2082
> >    
>
> This turned out to be because "truncate" was missing from match_code
> of arith_operand.  In other words, we needed
>
> (define_predicate "arith_operand"
> -  (match_code "subreg,reg,const_int")
> +  (match_code "subreg,reg,const_int,truncate")
>
> Here is the updated patch.  With this patch, I was able to build
> sh64-elf completely.  Should I build a few more variants?  Or OK to
> apply?
>  
I think it is probably OK, but I want to see some sh64 regression test 
results first.
I understand that Kaz Kojima has already tested x86 cross 
sh4-unknown-linux-gnu.
Kaz, did this include execution tests?

On one of our machines I'm currently generating an sh64-elf sim baseline 
with the
code from CVS today 12:00 UTC and regtesting Richard Sandiford's option
reorganization patch set.  I expect the baseline to be finised in about 
an hour,
then I can start a regtest of your patch without slowing down the other 
test;
but the result is not expected before tomorrow.