aglew@urbana.mcd.mot.com (Andy-Krazy-Glew) (09/26/89)
Would anyone feel competent to make a list of features that
distinguish DSP chips from general purpose microprocessors?
DSP chips seem to converge on general purpose microprocessors. Will
they eventually merge at the high end?
---
More particular questions:
I have been doing a literature survey of computer arithmetic techniques.
First, there are a number of alternatives to standard "binary" computer
arithmetic:
(1) redundant positional systems, which have the advantage of
constant carry propagation time (but then most DSP seems to be small
word-width, at least the commercial DSP chips) [widely used in
multipliers and dividers, but results are converted back to "standard"
form.
(2) logarithmic (homomorphic) number systems, which are obviously
cheap for *, /, and sqrt(). They are more expensive for +-, but apparently
you can build a system such that logarithmic +- is not so much slower
than */ that there is a net speed improvement (assuming +:* in typical
ratios 4:1 -- not so true for general purpose, but possible for DSP
applications)
(3) residue systems.
...
Are any of these alternate arithmetic systems actually used?
Much of the literature in Computer Arithmetic is concerned with digit
online and serial techniques, as opposed to the parallel-parallel
techniques which always seem the goal of general purpose computation.
I imagine that some of these are appropriate to DSP, eg.
embedded radar signal processing applications. How far off am I?
Are there "standard" components for this sort of thing?
---
As I mentioned above, I am doing a literature survey of computer
arithmetic; arithmetic is closely related to DSP (and computer
architecture in general); and I invite correspondence.
--
Andy "Krazy" Glew, Motorola MCD, aglew@urbana.mcd.mot.com
1101 E. University, Urbana, IL 61801, USA. {uunet!,}uiucuxc!udc!aglew
My opinions are my own; I indicate my company only so that the reader
may account for any possible bias I may have towards our products.