Extended precision

[soren_poulsen]

Dear reader,

I am currently using CVF 6.6 which does not support extended precision under Windows. Since I have an existing code that for some reason requires extended precison, I am in the progress of evaluating IVF.

I understand that the IVF compiler supports extended or quad precision under windows, for all IA-32 Intel processors. I assume that the extended precision calculations have to be performed by software routines, in case the hardware does not support extended precision. Moreover, I assume that the IVF compiler takes advantage of the hardware if present. My question is for which Intel processors/windows versions the IVF compler can take advatange of the extended precision hardware. Is extended precision support associated with EM64T? Itanium? What about the computation time for (a) Extended precison with vs. without hardware support and (b) extended vs. double precision? Rough estimates are sufficient.

Best regards,

Soren, Sweden

[Steven_L_Intel1]

There are no processors supported by the Intel compilers which have quadruple precision support in hardware, so most of your question is moot. The time it takes for individual operations vary and we don't have detailed timing information available. It is much slower than hardware instructions. Of course, your application does not spend 100% of its time in these operations, so the effect on your application run-time can be less.

If performance is important to you, try to restrict use of quadruple precision to only where you need it, and don't do something like /real_)size:128 to make all reals quadruple.