Hi,
i try to generate floating point cores using arithmetic functions IP in Aria 10. I have generated and used successfully 32-bit floating point adders, multipliers, comparators and sqroot blocks. However when i try to customize the floating point width in order to get half precision accuracy (16-bit) the resulting blocks use a lot more cycles and use a lot more ALMs than the respective 32-bit ones (except from the sqroor function). I was assuming that if my design would migrate from 32 to 16 bit floating point blocks it would be smaller for sure and possibly faster. Any idea if this is true in Aria 10? has anybody used customised arithmetic fp functions with 16-bit width?
