expf vs __expf ?

[Noel_Lopes]

Can anyone tell me which one is faster and what are the differences between those functions?

Thanks.

[seibert]

Generally, the function prefixed with ‘__’ is the faster version and is implemented with an instruction using the special function unit in hardware. The tradeoff is performance for is accuracy. The CUDA Programming Guide, Appendix B lists the error for all of the functions, and you can compare expf() (Table B-1) with __expf() (Table B-3).

[CTom]

What does ULP mean ???

Appendix B-1 p86:

The maximum error is stated as the absolute value of the difference

in ulps between a correctly rounded single-precision result and the

result returned by the CUDA library function.

[CTom]

Oups, just found it. :">

Google is my friend…

ULP

Unit in the Last Place is the smallest possible increment or decrement that can be made using the machine’s floating point arithmetic. In Java’s IEEE format, floating point numbers are stored as finite binary fractions. An 0.5 ulp maximum error is the best you could hope for, since this corresponds to always rounding to the nearest representable floating-point number.

http://mindprod.com/jgloss/ulp.html

Sorry i’m quite a newbie to CUDA and fast programmation/ optimization in general.