64 bit floats

[kavermeer]

I am using the IPP library for both scientific and commercial applications. In the latter case, I'm perfectly sattisfied with the currently supported data formats. However, for scientific use, I really miss 64 bit floats. Is there a reason these are not included? Or is it simply because nobody ever asked for it?

Best,
Koen Vermeer

[Vladimir_Dudnik]

Hi Koen,

what exactly do you mean? We do support Ipp64f data type in IPP. Would be nice if before scientific use of library you at least briefly looked trough IPP documentation.

Regards,
Vladimir

[kavermeer]

Hi Vladimir,

I'm sorry that I wasn't that clear. I didn't want to give a specific example, as I was afraid of getting the 'report to premier support' reply...

If I take a look at the image functions, none (or maybe almost none) of them support the Ipp64f data type. For example, the ippiAdd function has support for 32f and 32fc floats, but no 64f. Same applies to all other image functions I took a quick look at.

Hopefully it's clear now! If not, I apparently really missed something in the documentation; in that case, could you point me to the right chapter/page?

Best,
Koen

[Vladimir_Dudnik]

Hi Koen,

in that sense you are correct, we did not introduce many functions with 64-bit FP data type support in image processing. We do have rich set of such functions in signal processing library, so you can process 2D images on row-by-row basis.

Regards,
Vladimir

[kavermeer]

Hi Vladimir,

Thanks for the reply. I like working with the image processing functions because they allow me to process a full image at once, instead of row-by-row computations. So, yes, I could do that, but no, I don't really want to :-)
Is this a design decision, as far as you know? In that case, there's no need to bug Premier Support on this. Otherwise, I will let them know that I'd appreciate extending the image processing functions to Ipp64f.

Best,
Koen

[Vladimir_Dudnik]

Koen,

this way we try to optimize our resources on IPP development. It is huge library already andwe are trying to avoid additional entry points with duplicated functionality. It should be more convenient to write some C++ image class,which may hide all this complexity from application, rather then to double library size.

But please fill free to submit your request to Intel Premier Support, it is important for us to have feedback and understand what might be improved in future versions.

Regards,
Vladimir