Lpc2000

> 
> > From: "James Dabbs" <jdabbs@...>
> > Subject: RE: LPC2132 as DSP processor
> 
> >Compared to dedicated DSP's, I've found the bigger limitation to be
> >ARM's lack of circular buffer management. In a 4-pole IIR filter, I saw
> >most of the CPU BW spent managing pointers and ldr/str'ing rather than
> >doing the actual math.

 Agreed! Whenever I optimise DSP-like code, I concentrate on managing the 
data movement; try to use LDM and STM instructions rather than LDR and STR, 
possibly interleave history and coefficients, and perform as many operations 
in registers as possible between loads and stores - it's possible to perform 
four or five FIR or IIR operations in parallel within the ARM's registers.
 Following these principles I was able to get a five-times speedup of a 
quite reasonable assembler implementation of a notch filter, though managing 
register re-use so that the STM instruction left the results in the right 
order for the next call to the function required a couple of days of 
skull-sweat.
 This is one of the few situations I've seen where I would justify the use 
of assembler over a C code implementation - though I still like having a 
proven C implementation available so that I can validate the results 
returned by the assembler function. "All software is guilty until proven 
innocent".

Peter.


[Non-text portions of this message have been removed]