Hi!
On Fri, Sep 25, 2020 at 09:07:46AM -0500, Paul A. Clarke wrote:
> On Thu, Sep 24, 2020 at 06:22:10PM -0500, Segher Boessenkool wrote:
> > > + result [(__N & 0b1111)] = __D;
> >
> > Hrm, GCC supports binary constants like this since 2007, so okay. But I
> > have to wonder if this improves anything over hex (or decimal even!)
> > The parens are superfluous (and only hinder legibility), fwiw.
> > > + result [(__N & 0b1)] = __D;
> >
> > Especially single-digit numbers look really goofy (like 0x0, but even
> > worse for binary somehow).
> >
> > Anyway, okay for trunk, with or without those things improved. Thanks!
>
> I was trying to obviously and consistently convey the sizes of the masks,
> but I really want to convey _why_ there are masks, so let me try a
> different approach, below.
> +extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__,
> __artificial__))
> +_mm_insert_epi8 (__m128i const __A, int const __D, int const __N)
> +{
> + __v16qi result = (__v16qi)__A;
> +
> + result [__N % (sizeof result / sizeof result[0])] = __D;
> +
> + return (__m128i) result;
> +}
I don't think this helps explain things, sorry. Just add a comment
if you want to explain things? Simpler and works perfectly always.
To read these files I always open the x86 ISA docs anyway; I think
everyone will have to anyway, so you do not have to explain all details
you emulate, only the very tricky ones, or implementation choices, that
kind of thing.
Segher
