Re: Inline and XS benchmarks

[Sam Tregar]

On Sat, 30 Jun 2001, Brian Ingerson wrote:

> void
> add4 (...)
>     PPCODE:
>         {
>             SV* sum = sv_newmortal();
>             sv_setiv(sum, (IV)((int)SvIV(ST(0)) + (int)SvIV(ST(1))));
>             ST(0) = sum;
>             XSRETURN(1);
>         }

That can get a little better (10% or so):

  void
  add4 (...)
      PPCODE:
          {
              dXSTARG;
              sv_setiv(TARG, (IV)((int)SvIV(ST(0)) + (int)SvIV(ST(1))));
              XSprePUSH;
              PUSHTARG;
              XSRETURN(1);
          }

That's a trick I picked up from Doug MacEachern.  It saves a malloc by
using the same TARG register that opcodes use, as I understand it.

Of course, your point stands.

-sam