[Somewhat Off Topic] AI Contest

François Pinard pinard at iro.umontreal.ca
Tue Aug 7 15:46:23 CEST 2001


[Tim Peters]

> it was just unbearable that recompiling the world [on Cray's] should take 6
> minutes instead of 3.  That's the last time I took users seriously <wink>.

Those machines were quite impressive at the time, speed-wise.  I found
astonishing that a batch job could be initiated at a control point, executed,
and terminated within two refresh cycles of the monitor display, and so,
be invisible to the operator.  (Yet, COS was strangely designed underneath.)

> the real gain achieved thru auto-vectorizing/parallelizing compilers
> was that they taught users *how* to write code clearly worth vectorizing
> and parallelizing.

Exactly.  It is a bit strange that compiler-competent people were surprised.
This is how CFT was documented and taught to its end users.  Having intimate
knowledge of the machine was tremendously helping, anyway.

This is not unrelated to how one gets speed out of Python code.  It helps
knowing a bit how it works inside.  _You_, in particular, educated me!

> [...] processor "multimedia extensions" (newspeak for feeble little vector
> registers <0.5 wink>):

All the same story, all the time in the field of computers.  Old ideas
keep resurrecting with an all new terminology.  Newcomers, and even others,
firmly believe that things were just invented yesterday, and that computer
science moves fast.  I'm not so sure. :-)

-- 
François Pinard   http://www.iro.umontreal.ca/~pinard




More information about the Python-list mailing list