ippsPhase_64f precision

[eliosh]

I could not find any note about the precision of conversion functions.

What can be expected from the ippsPhase_64f's precision ? Simple tests show that this function is less precise than ippsAtan2_64f_A53 or even ippsAtan2_64f_A50. Could you, please, provide some information about conversion functions.



Thank you.

[Chao_Y_Intel]

Hello,

Could you post your test code here? That will be helpful to check the problem.

Thanks,

Chao

[eliosh]

Dear Chao,

I do not see any problem here.

Just wanted to know what precision ippsPhase_64f provides. That is how many correct digits can I expect?


Thank you.

[Chao_Y_Intel]

Hi,

This depends on the data. ippsPhase_64f use 64 bit float point as calculation. unlikely the fixed-accuracy vector functions ( ippsAtan2_64f_A53 etc), it does not guaranteed level of accuracy. In most of the cases, 64f is sufficient, and work well for most of the applications. But in some corner cases and some special values, it does not guarantee to the accuracy to some least important bits.

Fixed-accuracy vector functions have a precisely determined process. Only it reach some precession of some level. Then the calculation will be finished.

Thanks,

Chao