BAndViG
2014-12-30 19:02:02 UTC
Hello all!
I'm working to improve floating point division performance. To achieve that
I'm trying to implement division of fractional parts with Goldshmidt's or
Newton-Rawson algorithm. Additionally I'm planning to share multiplier for
MUL and DIV floating point instructions. However, I've faced with the
following problem. It is quite difficult (for Goldshmidt's or Newton-Rawson
iterations) to handle cases for which A could be divided by B without
reminder (including A==B as particular case). I haven't found a robust
solution yet. Does any body know how to deal with such cases?
BR
Andrey
I'm working to improve floating point division performance. To achieve that
I'm trying to implement division of fractional parts with Goldshmidt's or
Newton-Rawson algorithm. Additionally I'm planning to share multiplier for
MUL and DIV floating point instructions. However, I've faced with the
following problem. It is quite difficult (for Goldshmidt's or Newton-Rawson
iterations) to handle cases for which A could be divided by B without
reminder (including A==B as particular case). I haven't found a robust
solution yet. Does any body know how to deal with such cases?
BR
Andrey