From: Raymond Toy
Sent: Monday, December 27, 2010 5:01 PM
To: maxima at math.utexas.edu
Subject: Re: [Maxima] Multi-threaded Maxima
On 12/27/10 4:11 PM, Richard Fateman wrote:
It would require an underlying lisp that supports multiprocessing, for a start. Not all host lisps
support multiple threads. (Do any?)
I know that ccl, ecl, and sbcl support threads. I assume Lispworks and Allegro do.
The Intel I7 has seven cores (using hyper-threading, I think). Maxima could be as
Probably 4 cores with two threads each. And if you're rich enough, I think there are 6-core i7's.
There are a number of basic algorithms that could be made faster, e.g. polynomial multiplication in
the rational function representation, if memory bandwidth were not an issue.
It may be possible to call library routines (written with threads in mind) to do stuff faster,
e.g. bignum arithmetic using GMP.
I would be surprised if any obvious opportunities for parallelism sped up the test suite.
You could always run each individual test in a separate thread. That would speed up the testsuite quite a bit.
But this would also be a great source of hard-to-find and hard-to-fix bugs. There are enough bugs in maxima that we don't need to go intentionally adding even more. :-)
I still wonder, though, on what kind of symbolic computation would need to run that fast. Everytime I've tried to do some computation that took more than a few minutes, the result was such a huge mess that it was useless. A little thought on my part and some work with maxima produced some manageable symbolic result.
Maybe I do the wrong kind of symbolic work?
I have written a convolution program for computing convolutions of piecewise continuous functions that are coming from cspline() in interpol.mac that uses matrix convolution. This program would benefit from threads greatly if matrix multiplication were multi-threaded. The parallelism is so obvious it is my main interest for multi threading.
Ray
--------------------------------------------------------------------------------
_______________________________________________
Maxima mailing list
Maxima at math.utexas.edu
http://www.math.utexas.edu/mailman/listinfo/maxima