Why so many cycles to execute code with dspic30F

[buchtsucht]

Hi,
I am used to work with PICs like 12Fxx or 18Fxxx and have some rules of thumb, how much time some code needs to execute. As code get more complex, speed of dspic30F drecreases in a terrible way.

Did I have to use special comands to activate the fast DSP-functions of the dsPIC? I can´t read assembler - so maybe somebody can give me a hint.

[ckielstra]

I'm not too familiar with the dsPIC, but I'd say you have to get your rules of thumb right.

The PIC18 is an 8 bit processor, the dsPIC is 16 bit. Being a 16 bit processor makes the dsPIC faster, but you can only compare the two when using their native bit sizes.
For example a 32-bit multiplication takes 228 cycles on a PIC18f458 and CCS compiler v4.077, not the 20-25 cycles you claim.
The whole expression takes 313 cycles on the PIC18, so the 116 for the dsPIC seem fine to me.

For fastest processing use 16-bit variables, not 32-bit!

The 3 instructions for the assignment I can't explain. Two are expected because you are moving a 32 bit variable on a 16 bit processor. But it seems like there is an unnecessary temp location used.
Again, use 16-bit variables!

What is your compiler version number? Newer versions can be expected to have improved efficiency.

[bkamen]

BTW,

Reading assembler is simply a matter of getting the datasheet (or in this case, the dataSHEETS that are the dsPIC family reference) and looking up the opcodes in the section that discussed the instruction set.


If you work with micro controllers, it's a good thing (almost invaluable) to be able to read code -- because sometimes library functions don't do what you want.

Might be your code.
Might be the compiler.

With new PIC's, I've helped CCS work out kinks in handling the new device because I've taken the time to learn the instruction set for PIC's.

It's not as hard as you think.

Cheers,

-Ben
_________________
Dazed and confused? I don't think so. Just "plain lost" will do. :D

[bkamen]

If you look on PDF page 354 of the dsPIC33F family reference, Table 25-2

You'll see there is no MOV instruction that allows #LIT to F

Only
MOV #LIT -> Wn
MOV Wn -> F

Now, your code shows this for a word (16bits) but I'm assuming you made CC a 32bit long word.

Soooo... if you look at the options, you can't set a 32bit literal.

So you could do
(2) 16bit words and then a double move or
(1) 16bit word, a move including a clear (your case)
(1) 16bit word, a clear, then a double move.

I bet if you change your 'cc=512' to something that uses both words, those instructions would change.


Try it out and see,

Cheers,

-Ben
_________________
Dazed and confused? I don't think so. Just "plain lost" will do. :D

[buchtsucht]

Thanks!
With your help, I found the (idiotic) mistake - I have forgotten:
8 Bit PIC => INT= 8Bit
16 Bit Pic => INT= 16 Bit
Additional the CCS compiler for dspic could be better. But playing a little bit with my C-code (having the PIC-functions in mind), I get the expected speed for DSP.

[FvM]