Re: [RFC] trailing_wide_ints with runtime variable lengths

[Aldy Hernandez <aldyh@redhat.com>]

On Fri, Jul 1, 2022 at 4:58 PM Jakub Jelinek <jakub@redhat.com> wrote:
>
> On Fri, Jul 01, 2022 at 04:12:55PM +0200, Aldy Hernandez wrote:
> > > --- a/gcc/wide-int.h
> > > +++ b/gcc/wide-int.h
> > > @@ -1373,10 +1373,13 @@ namespace wi
> > >      : public int_traits <wide_int_storage> {};
> > >  }
> > >
> > > -/* An array of N wide_int-like objects that can be put at the end of
> > > -   a variable-sized structure.  Use extra_size to calculate how many
> > > -   bytes beyond the sizeof need to be allocated.  Use set_precision
> > > -   to initialize the structure.  */
> > > +/* A variable-lengthed array of wide_int-like objects that can be put
> > > +   at the end of a variable-sized structure.  The number of objects is
> > > +   at most N and can be set at runtime by using set_precision().
> > > +
> > > +   Use extra_size to calculate how many bytes beyond the
> > > +   sizeof need to be allocated.  Use set_precision to initialize the
> > > +   structure.  */
> > >  template <int N>
> > >  struct GTY((user)) trailing_wide_ints
> > >  {
> > > @@ -1387,6 +1390,9 @@ private:
> > >    /* The shared maximum length of each number.  */
> > >    unsigned char m_max_len;
> > >
> > > +  /* The number of elements.  */
> > > +  unsigned char m_num_elements;
>
> IMNSHO you certainly don't want to change like this existing
> trailing_wide_ints, you don't want to grow unnecessarily existing
> trailing_wide_ints users (e.g. const_poly_int_def).

That's precisely what I avoided...touching existing trailing_wide_ints
users.  As I explained, there is no cost to either const_poly_int_def
or range_info_def (though I'm about to nuke the latter).  There is
some padding that is currently used by m_len[N], and I just took a
byte out to represent the run-time length.  That would affect
trailing_wide_int users that have N > 4, but none are.  The use in
const_poly_int_def uses 2 (and range_info_def uses 3):

#define NUM_POLY_INT_COEFFS 2

struct GTY((variable_size)) const_poly_int_def {
  trailing_wide_ints<NUM_POLY_INT_COEFFS> coeffs;
};

>
> My brief understanding of wide-int.h is that in some cases stuff like this
> is implied from template parameters or exact class instantiation and in
> other cases it is present explicitly and class inheritence is used to hide
> that stuff nicely.

Yeah, it took me a while to decipher it, but I did read it :).

>
> So, you are looking for something like trailing_wide_ints<N> but where that
> N is actually a runtime value?  Then e.g. the
>   struct {unsigned char len;} m_len[N];
> member can't work properly either, because it isn't constant size.

What my patch does is store the run-time length in the aforementioned
byte, while defaulting to N/MAX.  There is no size difference (or code
changes) for existing users.  With my change, set_precision() and
extra_size() now take a runtime parameter, but it defaults to N and is
inlined, so there is no penalty for existing users.  I benchmarked to
make sure :).

I could hack up a variable_length_wide_int for what I want, but I'd
end up duplicating a lot of the trailing_wide_int_storage, etc.
Another option would be to stream out the HOST_WIDE_INTs in the
tree_int_cst and reconstruct things myself, but that smells of
reinventing the wheel.

Is there another way of allocating an n-bit wide-int at run-time?  I'm
happy to entertain other alternatives...

Aldy