>> FAST ASM,             C VERSION,            SMALL ASM
>> 1603  cycles,             1889 cycles,              3016 cycles
>> 132 instructions,        146 instructions,         89 instructions
>
> Wouter van Ooijen

Sorry for the (very) late reply. Just saw this message laying in my inbox.

FAST ASM - Probably close to a day including several improvements to the=20
version reported above. It now runs in 1442 cycles and takes 125=20
instructions on a dsPIC30/dsPIC33 (not including the dsPIC33EXXXX, which=20
takes more cycles, but allows for a much faster clock). Three of the=20
speedups: (A) using the dsPIC "DO" instruction, (B) removing a couple stall=
s=20
because of read-after-write hazards, and (C) using the mul.uu reg,#0,reg=20
instruction to clear a double register in one cycle/one instruction.

SMALL ASM - About an hour, basing it on the code in FAST ASM. Now down to 8=
3=20
instructions and a little faster at 2925 cycles.

C VERSION - Maybe 20 minutes.

So, was it worth it to write the FAST ASM version? While the performance is=
=20
all that much better than the C implementation it was a pleasant challenge.=
=20
Also, I implemented the C version last, just for laughs, and was very(!)=20
surprised at the job done by the optimizer. In fact, I stole the "mul.uu=20
reg,#0,reg" trick from the C compiler's output.

-- Bob Ammerman
RAm Systems

--=20
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist
.