Re: PATCH: Simplify x86 load/store builtin expand

[Uros Bizjak <ubizjak at gmail dot com>]

H.J. Lu wrote:

> > > Here is the updated patch. OK to install?
> > >
> > >       
> >  For the moment, please leave builtins, defined explicitly with
> > builtin_const () function where they are. I propose that const builtins only
> > are converted in this patch, and non-const builtins to be handled in the
> > last step, in a separate patch when all const builtins are converted. This
> > way, they won't mix in any case.
> >     
>
> Did you mean I should break it into 2 patches:
>
> 1. Use an array loop on def_builtin_const.
> 2. Use an array loop on def_builtin.
>   

Yes, but ATM, only commit the first patch.

> >  BTW: I think that usign "... complex arguments" is not good choice as this
> > combination usually refers to complex numbers.
> >
> >
> >     
> > > -/* SSE */
> > > -enum sse_builtin_type
> > > +/* Complex builtin types */
> > > +enum ix86_complex_builtin_type
> > >  {
> > > -  SSE_CTYPE_UNKNOWN,
> > > +  COMPLEX_FTYPE_UNKNOWN,
> > > +  V16QI_FTYPE_PCCHAR,
> > > +  V4SF_FTYPE_PCFLOAT,
> > > +  V2DF_FTYPE_PCDOUBLE,
> > > +  V2DI_FTYPE_PV2DI,
> > > +  VOID_FTYPE_PV2DI_V2DI,
> > > +  VOID_FTYPE_PCHAR_V16QI,
> > > +  VOID_FTYPE_PFLOAT_V4SF,
> > > +  VOID_FTYPE_PDOUBLE_V2DF,
> > > +  VOID_FTYPE_PDI_DI,
> > > +  VOID_FTYPE_PINT_INT,
> > > +};
> > >
> > >
> > >       
> >  The part above should be committed as a separate patch to avoid mixing
> > const and non-const builtins. (Also, we should use some other name than
> > "complex" for the reason described above).
> >
> >     
>
> How about ix86_memory_builtin_type,  bdesc_memory_args
> and ix86_expand_memory_args_builtin?
>   
Hm, perhaps ..._pointer_...  instead of ..._memory_...?

Uros.