an IEEE754 FP "multiply" pipeline is needed, for FP16/32/64. also needed is the optional ability to specify how many stages the actual multiplication is to take
FP16 mul pipeline bug: * 0xe7bb 0x81ce 0x2afa (returns 0x2af9) * 0x113 0xf569 0xb5d0 (returns 0xb5ce)
https://git.libre-riscv.org/?p=ieee754fpu.git;a=blob;f=src/ieee754/fpmul/align.py;h=6fea67ca13d168b0050cef93591c9d8f2f1056e3;hb=13417cb39c9dc37e5472555934dd27b39aa5b5ed#l72 found source of inaccuracy: alignment (pre-normalisation) of a and b were entirely misssing! unit tests pass, ran several tens of thousands of tests on FP16, FP32 and FP64.