topic Re: Questions about DFT routine in the 8.0 release in Intel® oneAPI Math Kernel Library

Questions about DFT routine in the 8.0 release

jayl — Mon, 10 Oct 2005 21:20:54 GMT

Dear all,

I'm an engineer working on radar imaging. We used MKL 7.2 for our previous product and try to move to the new version 8.0. I met a problem when we used the function DftiSetValue(). We used toset configuration values

DFTI_INITIALIZATION_EFFORT andDFTI_HIGH in the function DftiSetValue(), but these two parameters are not supported by version 8.0

Demo code like

int fft_rv;

int num = 256;

DFTI_DESCRIPTOR *dfti_1D;

fft_rv = DftiCreateDescriptor(&dfti_1D, DFTI_SINGLE, DFTI_COMPLEX,

1, num);

DftiSetValue(dfti_1D, DFTI_INITIALIZATION_EFFORT, DFTI_HIGH);

Could anyone give me a hinton setting the parameterin the new version toachieve the same performance?

Thanks a lot.

Regards,

Jay

Message Edited by JayL on 10-10-2005 07:23 AM

Re: Questions about DFT routine in the 8.0 release

Todd_R_Intel — Tue, 11 Oct 2005 01:36:54 GMT

You will receive the same performance with the new version (8.0) by omitting from your code the call DftiSetValue to set the DFTI_INITIALIZATION_EFFORT. In MKL 7.2, only one level of initialization effort was supported and that remains so in MKL 8.0. I'll submit a request to document the omission of the parameter in the new version.

-Todd

Re: Questions about DFT routine in the 8.0 release

jayl — Tue, 11 Oct 2005 02:11:39 GMT

Hi, Todd,

Thanks a lot for the help. However I remember In MKL 7.2 the default value for DFTI_INITIALIZATION_EFFORT is DFTI_MEDIUM. If I omit to code to call DftiSetValue(), can I still get same precision as DFTI_HIGH? Is DFTI_HIGH is a default value now? Thanks.

Regards,

Jay

Message Edited by JayL on 10-10-2005 12:15 PM

Re: Questions about DFT routine in the 8.0 release

Todd_R_Intel — Tue, 11 Oct 2005 02:56:13 GMT

That's right. Although there was a parameter that could be set, from what I understand there was only one behavior implemented for 7.2. So with 8.0 you should still get that same behavior, but you must remove the call to set this parameter.

If you are getting significantly different results, I would recommend submitting an issue.

Todd

Re: Questions about DFT routine in the 8.0 release

jayl — Tue, 11 Oct 2005 03:19:36 GMT

Hi, Todd,

Thanks for the suggestion. Our product was first released with MKL 6.1 and then was moved to 7.2. We only did a little bit change to get version 7.2 working. The code to set up DFTI_INITIALIZATION_EFFORT as DFTI_HIGH is inherited from code with MKL 6.1. According to your post, do you mean actually there's only one behaviour implemented no matter what value, DFTI_LOW,DFTI_MEDIUMor DFTI_HIGH,I set? Could you please verify if it's the same case for version 6.1? Thank you.

Regards,

Jay

Re: Questions about DFT routine in the 8.0 release

Todd_R_Intel — Tue, 11 Oct 2005 20:15:30 GMT

Yes, it is the same case for all versions including 6.1.

Todd

Re: Questions about DFT routine in the 8.0 release

jayl — Tue, 11 Oct 2005 23:50:42 GMT

Hi, Todd,

Thanks for the information. Have a nice day.

Regards,

Jay