splitting special case in scan_one_insn? (regclass)

[Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>]

As per the discussion in http://gcc.gnu.org/ml/gcc/2004-03/msg01711.html 
which ended in nothing.  The comment starts like this:

 /* Improve handling of two-address insns such as
     (set X (ashift CONST Y)) where CONST must be made to
     match X.  */


Summary:

1) At the time the transformation could not be triggered on any of 
x86-linux, mips-elf, sh-elf, arm-elf and I don't see why it should be 
different now.

2) In fact, the instruction is probably invalid on all those targets, 
and they use a target-specific splitter or (more likely) a define_expand 
to do this.  So the insn would probably not even pass the predicates.

3) If anywhere, this option would belong in regmove.


To add my two cents to item (2): anybody knows a reason why a target 
should not use a define_expand (and thus benefit of loop invariant 
hoisting, CSE, etc. etc.) nowadays?  This would yield both benefits 
since the (ashift CONST Y) would still be there as a REG_EQUAL note, at 
least after CSE or fwprop.

The code predates the GCC subversion repository.

Paolo