On Mon, Feb 06, 2017 at 10:31:49AM +0000, Peter Maydell wrote: > On 6 February 2017 at 08:58, Bharata B Rao <bhar...@linux.vnet.ibm.com> wrote: > > On Fri, Feb 03, 2017 at 03:39:16PM +0000, Peter Maydell wrote: > >> On 3 February 2017 at 15:12, Bharata B Rao <bhar...@linux.vnet.ibm.com> > >> wrote: > >> > As I said above, it is based on float128_to_int64() > >> > >> Ah, right. I think that's probably a bad model to copy because > >> it's a conversion to signed integer, not a conversion to > >> unsigned integer (so the edge cases are different). > > > > I checked the original berkeley implementation and I see that > > float128_to_uint64 implementation there also is based on > > float128_to_int64 implementation with edge cases being different. > > > > To the best of my understanding, the corner cases for unsigned > > int are covered in the above implemenation. Could you please > > take a re-look at this ? > > OK, but that's about 5 times harder to review because I have > to work everything out from scratch rather than being able to > say "yes, this is doing everything the same way that our > existing known-to-be-good other function is doing it".
Fair enough. I will come up with a version of float128_to_uint64_round_to_zero() based on the existing float64_to_uint32_round_to_zero(). Regards, Bharata.
