topic Re: Fixed-Accuracy Arithmetic Functions in Intel® Integrated Performance Primitives

Fixed-Accuracy Arithmetic Functions

constantine-vassilev — Thu, 11 Dec 2008 21:54:09 GMT

From the Fixed-Accuracy Arithmetic Functions listed in Vol 1/Signal Processing/Chapter 12 I do not see multiplication, addition and subtraction - do I miss something?

Also I suppose Ipp32f notation which generally means floating point, to be interpred as fixed point in the context of Fixed-Accuracy Arithmetic Functions. Am I correct?

Re: Fixed-Accuracy Arithmetic Functions

gol — Mon, 15 Dec 2008 03:42:54 GMT

those operations are so basic that it would surely be less efficient to achieve less accuracy, there would be no point

Re: Fixed-Accuracy Arithmetic Functions

gol — Mon, 15 Dec 2008 03:47:36 GMT

Also, no it's not fixed point, but floating point. Those functions are functions that eat a lot of CPU but can be approximated with less accuracy for less CPU, that's why they are there. Even though there are native instructions to divide, it's still a very slow process, whilethe CPU can do several multiplications per cycle, hence nothing to speed up here.

Re: Fixed-Accuracy Arithmetic Functions

constantine-vassilev — Mon, 15 Dec 2008 04:57:28 GMT

Quoting - gol

looking at their header files I see all oprations, obviously documentation is not full.

Re: Fixed-Accuracy Arithmetic Functions

Vladimir_Dudnik — Mon, 15 Dec 2008 05:49:29 GMT

Hello,

all IPP fnctions should be listed in documentation. Could you please point to which function you did not find reference in IPP manual?

Regards,
Vladimir