Faster fractional multiply

Bobb Fwed

I want to multiply a 16-bit number by 1.054. Is there a faster way to do it other than float math and not using any additional cogs.

ericball

Yes. Use fixed point.


Use PhiPi's solution. I always forget that SPIN uses signed values. (Plus his is probably slightly faster anyway.)

Phil Pilgrim (PhiPi)

Similar to Eric's:

y += y ** 231928234   'Multiply y by 1.054

This multiplies y by 0.054 * 2³², returning the high 32 bits, then adds the result to y (i.e. the "1" part).

-Phil

Ale

Well... if you use a 16.16 format, and multiply by 1.0000110111010010 then you get a passable approximation (1.0539856)

result = (x << 16) * $1_0DD2

Bobb Fwed

Gives me negative numbers ('cause the "fraction" is signed).

Changed it to

result := input << 2 ** $4374_BC6A

Thanks!