[plt-scheme] List loop timing

From: Matthias Felleisen (matthias at ccs.neu.edu)
Date: Fri Feb 15 12:16:19 EST 2008

On Feb 15, 2008, at 12:10 PM, Prabhakar Ragde wrote:

> Matthias Felleisen wrote:
>> * It is a COMMON misunderstanding that TR versions of a function
>> do something good for your timing.
>>   ************************************
>>   TR is a space-saving device. PERIOD.
>>   ************************************
>> It does nothing for time.
> Except when it does something for time.

Sometimes. It is NOT uniform!!!!

> It is more accurate to say that between TR and non-TR versions of a  
> function with the same asymptotic running time, the constant factor  
> is not necessarily smaller for the TR version


;; ---

And the only way to benchmark such universal claims properly is to  
use high-quality implementations. With that I don't mean something  
easy to install and run  or cheap to buy. I am referring to compilers  
with highly tuned optimizers. While there are more actively  
maintained Scheme implementations than SML users, only Chez and  
Larceny come to my mind when I use this kind of constraint. (I am  
sure I am overlooking one or two compilers with similar qualities,  
but that's only true on highly optimistic days.)

-- Matthias

P.S. Plus watch out for which platform the compilers were tuned. They  
may not correspond to the popular ones.

Posted on the users mailing list.