Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.

vslzConvExec .. ways to make it faster???

Paul_Margosian
Beginner
285 Views
With considerable help from a Member of this group (Vladimir) have gotten this convolution function to work properly.

In my application, do sets of 2D convolutions (8, 16, etc) and add the results as part of an MRI parallel imaging reconstruction. Currently call vslzConvExec once for each convolution and manage rest of the math externally. It works but, of course, I wish it were faster.

Question: Is there a way to load several sets of data into this function (8, 16, etc) and have the convolutions done all at once? Even if possible don't know whether this would be faster, but it's worth a try.

Paul Margosian
0 Kudos
2 Replies
Victor_Gladkikh
New Contributor I
285 Views

Hi Paul,

MKL don't provide such functionality.

You couldusenext mode:VSL_CONV_MODE_DIRECT.It may give speed-up forsmall problems.

Victor.

0 Kudos
Hsin-ying_L_Intel
285 Views

Hi Paul,

Thanks for bringing the issue to our attention.

Victors recommendation above may serve your need for current release. In addition, we will take this into our consideration for future potential feature enhancement, it has been recorded into our internal tracking system. We will inform you when there is an update on this feature request after next review cycle.

Best regards,

Hsin-Ying

0 Kudos
Reply