Which version of the library are you using, and which OS? I'm guessing Linux on the OS, and gfortran?
The ifort version of the library are built with -Qprec for windows, and -mp1 for linux. These were chosen to propagate NaN's for programs like matlab and R.
You will have to refer to the examples to get the link commands right, but you might try the ifort library versions to see if they solve the problem.
Can this somehow be fixed in the GNU Fortran version as well? I cannot use Intel Fortran, and I have noticed the NaN propagation issue still persists in ACML 5.3.1.
I'd like to confirm that I see the same (mis)behaviour with ACML (not clMath) 5.3.1 (gfortran64_fma4_mp) in R 3.0.2 built by gcc 4.7.2.
mr. Chip, can you tell whether it's a blocker? no-go for R + ACML?
it does NOT look good for ACML, from what I understand it is said ACML's own problem and R developers, correct me if I'm wrong discourage from using ACML.
it would be a BIG shame cause all the bits that I've linked to ACML performed better and in some cases had ~10 times better performance
would be nice if AMD experts clarified the situation
many thanks & regards