Last active
August 11, 2022 05:24
-
-
Save shunghsiyu/a63e08e6231553d1abdece4aef29f70e to your computer and use it in GitHub Desktop.
Fix description of tnum.mask, and use C-style comment
Another ideal would be to brute-force through all the combinations at a lower size (i.e. bandwidth), maybe 16-bit instead of 64-bit, then plot some thing like a closure plot (using the posits paper as an example here)
Scratch that... I cannot represent tnum on a single number line
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
@Werkov thanks for spotting the mistake! Now fixed.
Very interesting indeed, though I wonder if I should eliminate cases where it overflows, otherwise a
Tnum(val=-1, mask=0) + Tnum(val=0, mask=1) == Tnum(val=0, mask=-1)
worst-case might kill all the fun.Another ideal would be to brute-force through all the combinations at a lower size (i.e. bandwidth), maybe 16-bit instead of 64-bit, then plot some thing like a closure plot (using the posits paper as an example here)
In Sound, Precise, and Fast Abstract Interpretation with Tristate Numbers, they used brute-forcing to compare the precision between their proposed multiplication algorithm and the original
tnum_mul()
in the Linux kernel.I hope not :P