subreg rtl documentation

[Paolo Bonzini]

Bernd Schmidt wrote:
> Jim Wilson wrote:
>> On Tue, 2008-04-08 at 10:02 -0600, Jeff Law wrote:
>>> I don't think we do.  reload makes a pass over all the insns calling 
>>> cleanup_subreg_operands which ought to be zapping all the (subreg 
>>> (mem)) thingies and replacing (subreg (hardreg)) with  (hardreg')
>>
>> But we also call cleanup_subreg_operands in final_scan_insn, right
>> before emitting assembly code, which implies that some subregs might be
>> slipping through.
> 
> Are there any ports that do (subreg (match_operand "m"))?  We should 
> probably explicitly outlaw that (or subregs of match_operand in general).

h8300 and s390, both for paradoxical subregs.  subreg of match_operand 
is more widespread.

Paolo