[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: CLOS Benchmarks

> >PCL runs particularly poorly in AKCL.  This could be fixed with, I
> >believe, only a modest amount of work by someone familiar with the
> >backend of the ACKL compiler.  It would require writing a "LAP Code"
> >generator for AKCL.  See the *lap* files in PCL for more information.
> It may be more than a matter of just PCL not being optimized for
> AKCL.  At least in what I've seen on AKCL on the Sun-4s, the basic
> overhead of function calling in AKCL is nearly an order of magnitude
> greater than in Lucid (Sun CL) and Allegro.  (Non-function-calling
> list-processing and arithmetic operations in AKCL aren't so bad).

Can you suggest a benchmark that would demonstrate this.  KCL
function calls to functions in the same file were C functon 
calls.  In AKCL, they're (usully? always?) slightly more indirect
but can be made to pass args on the C stack.  So I don't see
why they should be tremendously slow.  Does this "nearly an order
of magnitude" mean nearly twice as slow or nearly ten times as