Hardware multiplier access in SPIN2 language

rogloh

Is there a way to make use of the P2 hardware's MUL or MULS instructions in SPIN2 or does it always have to use the CORDIC engine for this?

I've found cases where I know the items being multiplied are both 16 bit quantities and it would be nice to be able to get a fast 2 cycle multiply done instead of the slower / general purpose CORDIC operation. How is this achievable? If it currently is not part of the language could it be extended to support it or is that too hard to fit in these days?

If it needs to be coded in inline PASM is there so much overhead to setting up the inline PASM vs using the CORDIC that it wouldn't be worth it?

pik33

Yes, I saw what the compiler (Flexspin) did with my functions. They may be 20xfaster. The temporary solutions may be

• try and goto C (and hope if 16-bit int declared, no cordic is used)
• make all function in inline PASM so the register saving, etc overhead stuff will be done once for all the function, and not on every loop
• make a dedicated cog to do the work in its RAM - fast(for functions like putpixel, draw line, circle etc)

Wuerfel_21

Flexspin indeed uses MUL for 16 bit values.

PUB main() | long a, long b
a := ina
b := inb
outa := a*b

compiles to

_main
    mov _var01, ina
    mov _var02, inb
    qmul    _var01, _var02
    getqx   outa
_main_ret
    ret

---

and

PUB main() | word a, word b
a := ina
b := inb
outa := a*b

compiles to

_main
    mov _var01, ina
    mov _var02, inb
    mul _var01, _var02
    mov outa, _var01
_main_ret
    ret

Sadly it doesn't seem to have the ability to squeeze other instructions inbetween the QMUL and GETQX.

pik33

Good to remember that Spin2 allows declaring the local variable type I will replace all longs where they are not needed.

cgracey

I would need to make the interpreter look at the input data sizes and dynamically do MUL or QMUL.