Re: Interpreters and (math) speed

Stefan Wils wrote:
> I'm kind of new to this area, but I am in the process of designing a
> 'graphics language' : a language which has a very extended set of


> [Depends on the granularity of the operations. If each op is big enough,
> the interpreter overhead vanishes into the noise. That's why interpretive
> matrix math systems are popular. -John]


APL and J are good examples of the array-oriented languages alluded to
by John. The problem with them as interpreted languages is that, even
with big arrays, you usually get Amdahl's-Lawed to death by the scalar
and wee tiny array ops that creep into any computation other than
something like a linear equation solver with no other work (so it
becomes a single primitive in APL or J).


Compiling these is a good way out -- it gives lots of expressiveness,
conciseness( due to array ops and higher-order functions), and rapid
development. WHat we need is a JIT compiler for these interpreters.


For some performance stats, overview on the problem, etc., check my
thesis and tech report at: http://home.ican.net/~bernecky


Bob
--

Return to the comp.compilers page.
Search the comp.compilers archives again.