Perl_do_print: stringify an SVt_IV IV/UV more efficiently

[richardleach]

https://stackoverflow.com/questions/79244888 points out that pp_print
is really slow at stringifying integers. It's bizarrely much faster to add
additional operations to do the stringification first.

Here are some sample timings:

1003 ms | 'for (my $i=1; $i<=10000000;$i++) { say $i }'
 630 ms | 'for (my $i=1; $i<=10000000;$i++) { say "$i" }'
 695 ms | 'for (my $i=1; $i<=10000000;$i++) { say $i."" }'
 553 ms | 'my $i = "str"; for ($i=1; $i<=10000000;$i++) { say $i }'

Note from the last timing that stringification in pp_print is not
always slow, only when the IV is stored in a SVt_IV. This traces back
to Perl_do_print which has a special path for just this case:

    if (SvTYPE(sv) == SVt_IV && SvIOK(sv)) {
        assert(!SvGMAGICAL(sv));
        if (SvIsUV(sv))
            PerlIO_printf(fp, "%" UVuf, (UV)SvUVX(sv));
        else
            PerlIO_printf(fp, "%" IVdf, (IV)SvIVX(sv));
        return !PerlIO_error(fp);
    }

(There are no code comments to say why it does this. Perhaps deliberately
preserving the type for some reason?)

PerlIO_printf calls PerlIO_vprintf, which calls Perl_vnewSVpvf, which
creates a new (effectively) temporary SV, then sends that to
Perl_sv_vcatpvfn_flags for stringification and storage of the IV.
PerlIO_vprintf then writes out the buffer from the temp SV and frees it.

All the other routes essentially call SvPV_const on $i or a PADTMP.
This invokes sv_2pv_flags and its helper, S_uiv_2buf, does the work.
No temporary SVs are created and freed in such cases.

This PR changes Perl_do_print to also use a small buffer and
S_uiv_2buf to do the stringification more efficiently, whilst still
avoiding a type upgrade, Re-measuring after this PR:

 470 ms 'for (my $i=1; $i<=10000000;$i++) { say $i }'

---

• This set of changes requires a perldelta entry, and it is included.

[mauke]

Will this symbol be visible in external (C/XS) code? Because at the moment it is an unprefixed identifier (no Perl_ or anything) in the global namespace.

[richardleach]

Good question. I'm not sure how best to handle it. Happy to take advice.

[richardleach]

Guarding it inside a #if defined(PERL_CORE) || defined (PERL_EXT)?

[mauke]

Why is happy declared so far away from its first use?

[richardleach]

Mirroring the else branch of Perl_do_print. I kept it it the same for consistency. Happy to change both instances in a follow-up commit.

[mauke]

This could be

bool happy = !(len && PerlIO_write(fp, ptr, len) == 0);

[richardleach]

This was copied from the else branch of Perl_do_print. I kept it it the same for consistency. Happy to change both instances in a follow-up commit.

[mauke]

This could be

return happy && !PerlIO_error(fp);

[richardleach]

This was copied from the else branch of Perl_do_print. I kept it it the same for consistency. Happy to change both instances in a follow-up commit.