[patches] Re: x86_64 merger part 17 - split-long-move update

[Jan Hubicka]

> > 
> > I'd prefer you used change_address instead of PUT_MODE throughout.
> > I know that existing code in this function does it, but I don't
> > like that either.
> > 
> > Again, please use UNITS_PER_WORD.  It's more concise than the
> > explicit conditional, and it documents exactly what you want.
> > 
> > > +       /* In 64bit mode we don't have 32bit push available.  In case this is
> > > + 	 register, it is OK - we will just use larger counterpart.  We also
> > > + 	 retype memory - these comes from attempt to avoid REX prefix on
> > > + 	 moving of second half of TFmode value.  */
> > > +       if (TARGET_64BIT && GET_MODE (part[1][1]) == SImode)
> > 
> > This part is confusingly arranged with multiple tests for TARGET_64BIT.
> > 
> > >     [(const_int 0)]
> > > !   "ix86_split_long_move (operands);")
> > 
> > You've lost the DONE.
> 
> Hi
> I am just testing following updated version of the patch. OK assuming
> that it passes bootstrap/testsuite?
Ignore it for the moment - the testsuite found number of problems I am hunting
right now.

Honza