Re: Stuff in core (was Re: date interface, on language (wasRe: perl6 requirements, on bootstrap))

[Tim Bunce]

On Wed, Aug 02, 2000 at 12:20:00PM -0600, Nathan Torkington wrote:
> Ken Fox writes:
> > pipeline stalls, cache misses and a whole bunch of interesting things. One
> > of the reasons Perl performed well is that it spent a lot of time in what
> > they called native code, i.e. not decoding and dispatching ops.

I think the key point there is that many perl ops are 'heavy'. They
tend to do a lot of work.

> One thing we could do is look at the op paths produced by perl5 code
> and work out which sequences of ops occur so often that they deserve
> to be encapsulated in their own op.  This might give us new ideas on
> how to design the ops.

And the other way around: see which multi-function ops (those that
have important 'if's near the top) could usefully be split into seperate
ops to avoid instruction pipeline bubbles etc.

Tim.