Hello,

i'm doing some research to evaluate the BF-527.

i already did code to evaluate / benchmark the perfomance of the core in float point operations. but i just used "include <math.h>" (it's C-program)

but only yesterday I found document "EE.185.Rev.4.08.07" which describes fast-float operations in BlackFin. but all the examples are in ASM.

- is there any library to support fast-float in C? i did find some refferences in Visual DSP 5 -> Help (about '-fast-fp' option) but nothing exactly about types and so on.

i found discussion about FastFloat from 2009 http://ez.analog.com/message/2781#2781

so i got another quesitons:

- is there news / fixes for this?

- is there any fastfloat32 support in C? since "float16.h" doesn't sound like do this support.

- are there only Addition, Substract, Multiply and Division operations supported there?

Thanks in advance,

Dmitry K.

There are no plans to support any further non-standard floating-point types. The support offered by float16.h is a 15-bits exponent and 16-bits mantissa and one sign bit. You are right that only primitive arithmetic is supported, there is no extended library support for functions like sin and cos for example. There are no open problems any longer with float16.h support so if you are using the latest VisualDSP++ 5.0 Update 10 you'll have fixes for the problems mentioned at the forum link you looked at. However, you should be aware that it is possible that float16 support will be removed from the next major release of the tools in the future.

Regarding the -fast-fp switch, that's something different. VisualDSP++ comes with two floating-point arithmetic libraries. The default one used is the fast one so that will be the one you've tested with. The non-default library is fuller featured but larger and slower. We can provide you further details on this if needed.

Regards,

Stuart.