--- Comment #27 from Witold Baryluk <> 2012-02-03 
01:09:31 PST ---
Just to add some additional context.

I think many CUDA and OpenCL compilers, as well some tools which uses
OpenGL/Cg, whetever, when asked to do double calculations actually uses
float-float (2x single precission) to have almost same precission of double (49
bits compared to 52 bit of mantisa, which is big win compared to 23 bigs of
single precission floats). So it is common to use float-float or double-double
schemes actually to boost precission. I know software and library will need
much more testing by developer to make sure lack of IEEE-754 comppilance is not
a problem in given library or application, but I think D should in theory allow
some compiler to diverge from IEEE-754 compilance if user really wants to.

Configure issuemail:
------- You are receiving this mail because: -------

Reply via email to