Re: PR target/23303 (worse instruction choice)

[Jan Hubicka <jh at suse dot cz>]

> On Tue, Nov 01, 2005 at 09:34:28AM +0100, Jan Hubicka wrote:
> > +;; After splitting up read-modify operations, array accesses with memory operands
> > +      && (GET_MODE (operands[0]) == QImode || GET_MODE (operands[0]) == HImode))
> 
> Long lines and,
> 
> > +"INTVAL (operands[2]) >= 0 && INTVAL (operands[2]) <= 3
> 
> incorrect indentation.
> 
> > +  operands[1] = gen_rtx_PLUS (Pmode, base, gen_rtx_MULT (Pmode, index, GEN_INT (scale)));
> 
> Do you really want Pmode here?  Seems like you may want to
> use SImode on TARGET_64BIT given SImode operations.

Hmm, I probably should add SImode lea generation for x86-64 even if
Pentium4 nor Opteron use !READ_MODIFY, so it should not match.
In future we probably will see pentiumM like 64bit chip that will likely
use it...

I will send updated patch.

Honza
> 
> 
> r~