[Libre-soc-bugs] [Bug 1085] parallel prefix sum polynomial interpolation

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=1085

--- Comment #5 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #4)

> parallel reduction is fine here too -- reduce using fadd, though as noted it
> gets less accuracy than using fma (because fma doesn't round between the mul
> and add and because higher powers of x tend to give monomials (includes
> coefficients) that are smaller, so adding a smaller approximate value to a
> larger exact value (the coefficient) gives a value with higher relative
> precision).

sounds to me like /mrr (reverse-gear) would be sensible to use,
at the end where the tinier values would accumulate with better
accuracy.

exactly the kind of thing to put in an app note.

-- 
You are receiving this mail because:
You are on the CC list for the bug.