[arjvik]

Not to take away from this cool project, but its design decisions are incredibly impractical.

[refulgentis]

I honestly can’t tell if it’s a cool project or just a md file someone with 0 experience had an LLM output.

[imtringued]

Considering the use of 32-bit fixed point numbers, you can bet on that person having zero clues.

Even the crappiest FPGA has at least 18x18 multipliers, meaning that you could add a small exponent on top to get a floating point type with a slightly worse precision than single precision floats.

32 bit fixed point doesn't map to any DSP I'm aware of.