Docs: insufficient explanation of raw value editing

[jrmoserbaltimore]

The below does not do what I expect:

a=Fxp(0xe25398fa9f, dtype='Q2.38')
a.set_val(0xe25398fa9f, raw=True)

The result is to set a to a positive value of all bits on (except the sine bit).

The result I expect is that Fxp.bin(frac_dot=True) return 11.10001001010011100110001111101010011111

I've thus far been unable to determine from the docs what, if anything, would allow me to put in some figure and get that figure to be Fxp.bin(). Apparently, it's not set_val(). As you may notice, this binary string is a negative number.

[francof2a]

Hello @jrmoserbaltimore

In your example, there is a "problem" about how python represents hexadecimal values, because 0xe25398fa9f is just an int (signed) equals to 972065143455 (not a negative number). When you create a Fxp with this value, its sign information was lost.

You have two options at least:

1. You can pass the hexadecimal value as a string:

a = Fxp('0xe25398fa9f', dtype='Q2.38', raw=True)
# or
a.set_val('0xe25398fa9f', raw=True)

a.bin(frac_dot=True)

'11.10001001010011100110001111101010011111'

2. Pass the value with sign information using twos-complement (a function to do that is available in fxpmath.utils):

from fxpmath.utils import twos_complement_repr

v = twos_complement_repr(0xe25398fa9f, nbits=40)
a = Fxp(v, dtype='Q2.38', raw=True)

a.bin(frac_dot=True)

'11.10001001010011100110001111101010011111'

I hope this helps.

[jrmoserbaltimore]

Thanks, that helps. I may have not read the documents closely enough, I'll check again. Didn't think to pass the number as a string.

[javValverde]

Hello,

First of all congrats for the project! It's been a great discovery

I'm trying to understand why a string is needed when converting from integer back to fixed point. Given dtype=Q2.38, it is already known that the number is signed with 40 bits. If the MSB (i.e. bit 39) is one, then the number should be negative. Am I missing any other use case?

I can open a new issue if you prefer it