Intel® Integrated Performance Primitives
Deliberate problems developing high-performance vision, signal, security, and storage applications.

Modf allow null input (feature request)

jeffc111
Beginner
551 Views

It would be nice if Modf allowed a null input for its integer output. In my use I only need the fractional part.

0 Kudos
5 Replies
Ying_H_Intel
Employee
551 Views
Hi Jeffc111,

Thanks for submitting. It has been recorded in IPP productdatabase for IPPengineerteam furtherreview.

BestRegards,
Ying
0 Kudos
Ying_H_Intel
Employee
551 Views

Hi Jeffc111,

We discussedyour requestinternally, itmay notmake sense to change behavior of ippModf or add new function that returns only fractional part into IPP VM without visible perfrormance gain. But considering it is normal for MKL routine to accept the null pointer as input, we think it is make sense to add new routine for fractional part into MKL, so esclate the feature request to MKL team.MKL team have agreed to add such functionality in MKL. Could you please let me know if a new MKL routine can help you?

MKL VML: http://software.intel.com/sites/products/documentation/hpc/mkl/vml/vmldata.htm

Best Regards,
Ying

0 Kudos
jeffc111
Beginner
551 Views
Hi Ying,
I appreciate your consideration. I don't quite follow your explanation though. Do you mean that if you implemented a version of Modf that only returns the remainder it could not be made to work more efficiently than the instructions the compiler would generate?
I also think there are ipp functions that take null as input, though perhaps not for output (RealToCplx functions come to mind).
In any case, while it's interesting, we do not license MKL. IPP has served us very well, so we'll continue to enjoy using just it.
Best Regards,
Jeff
0 Kudos
Sergey_M_Intel2
Employee
551 Views
Hello Jeff,

We don't expect the modf tobecome measurably faster if second argument is NULL. The only possible (slight) improvement may be due to reduced memory footprint (in factit will be minimal if the vector size is relatively small).

Ying may have something in addition to that.

Thanks,
Sergey
0 Kudos
Ying_H_Intel
Employee
551 Views

Hi Jeff,

I heard from our IPP team. two new functions ippsFrac_32f and ippsFrac_64f have been added to ippVM to support the feature you mentioned here.  The release will be targeted to ready at the Sept.(we shall have a notification here when new version release) . You are welcomed to try by that time.

Best Regards,

Ying

0 Kudos
Reply