Displaying 1 result from an estimated 1 matches for "round_to_zero".
2015 Apr 29
2
[LLVMdev] [RFC][Float2Int] Converting (fcmp Pred, x * F, y) to (ICmp ...)
...to use the
existing
Float2Int infrastructure to track the integer length that would be required
for the conversion.
The main difficulty I can see with this is validating the transformation.
Ideally
the property that we want to prove for the guessed integers a and b would be
that (a*x)/b == round_to_zero(f * to_float(x)), for every integer x in a
given
range. The problem with this is that we would need to come up with a
reasonably
quick method of proving this.
My best guess (still need to work out the exact details) would be that we
only
need to check the first and last b numbers of the rang...