How to do "fixed size" (wrapping) math in Python?

[AlexanderWulf]

Okay, I have studied the Python implementation of FNV-1a (64-Bit) hash, which uses the unit64_t (64-bit unsigned integer) type in the "C" version:
https://de.wikipedia.org/wiki/FNV_(Infor...%C3%BCssel
https://en.wikipedia.org/wiki/Fowler%E2%...NV-1a_hash

Note the requirements:

All variables, except for byte_of_data, are 64-bit unsigned integers.
The multiply returns the lower 64-bits of the product.

In Python's implementation, they effectively do:

for char in text:
    n = (n ^ ord(char)) * prime
    n = n & mask

...with mask defined as 0xffffffffffff. I think the bit-wise AND with mask is done in order to get the "lower 64-bits of the product", as needed.

So, this is probably more efficient than "modulus" operation and I think I can adopt this technique for my needs.

Nonetheless, numpy looks interesting and I will definitely give it a try! However, in my test, multiplying two np.uint64 values produced an overflow warning

[deanhystad]

Generally overflow is a bad thing. You can turn the warning ogf.

[AlexanderWulf]

I agree, that overflow is bad in general. But it is perfectly fine when doing "wrapping" math.

That is why, for example, in Rust the standard × and + operators throw an error on overflow, whereas the wrapping_add() or wrapping_mul() intrinsics do not.

Does Python/numpy have a similar thing? Turning off the warnings "globally" probably isn't a good idea...

[deanhystad]

You turn off the warning for numpy, not Python.

https://numpy.org/doc/stable/reference/g...nings.html