laughlin@fornax.UUCP (Bob Laughlin) (11/10/90)
(This should really be mailed to next-prog but mail to
next-prog@cpac.washinton.edu bounces---just from here?)
I'm using the DSP to multiply 2 vectors. They are big vectors,
on the order of 200K. I get the correct results but it is very
slow (slower than 68882 FP), and it turns out that most of the overhead
is in converting the vectors to Fix24 format and transferring them
to and from DSP memory with DSPAPWriteFloatArray()
(all values are on the range -1.0 to +.999999) and DSAPWriteShortArray()
and transferring the results back with DSAPReadShortArray().
For example, it might take 20 seconds to transfer all the data
and 5 seconds to do all the multiplication. The maximum amount of data
that DSP memory will hold (about 6k) is being transferred each time.
Is there some way to speed up the transfer of data to DSP memory?
--
Bob Laughlin laughlin@cs.sfu.ca