Community
cancel
Showing results for 
Search instead for 
Did you mean: 
P__Ella
Beginner
124 Views

ippiDilateBorder_16u_C1R performance issues

Hi,

In the previous ipp versions I used unique function for 3x3 morphology - ippiDilate3x3_16u_C1IR.

I'm now using ipp18. The new general function - ippiDilateBorder_16u_C1R with 3x3 mask is slower than the old one in about 17%.

Any ideas how to improve performances?

thanks,

Ella

 

0 Kudos
5 Replies
Igor_A_Intel
Employee
124 Views

Hi Ella,

could you provide a bit more info: image size, cpu, arch, OS, static/dynamic linking?

the best is the full output from ippGetLibVersion() function.

regards, Igor

P__Ella
Beginner
124 Views

Hi Igor,

This is the output of ippGetLibVersion() function:

    major        2018
    minor        0
    majorBuild    1
    build        56998
    targetCpu    "core...
    Name         "ippCore"
    Version         "2018.0.1 (r56998)"
    BuildDate    "Oct 10 2017"

 

I runs the function with 2048x2048 images.

I tested it with different computers, with different CPUs, windows 7 or 10.

 

Thanks,

Ella

P__Ella
Beginner
124 Views

Hi Igor,

Update - we've upgrade the ipp version to the following version:

    major        2018
    minor        0
    majorBuild    4
    build        60077
    targetCpu    core
    Name         ippCore
    Version       2018.0.4 (r60077)
    BuildDate   Aug 15 2018

This upgrade improved the running time dramatically, and now the dilate function runs faster in my computer (Core(TM) i7-9820HQ). However, in the target computer, the upgrade did not affect the running time, and it still slow.

The target CPU is Xeon E5-6280 b2

Thanks,

Ella

Pavel_B_Intel1
Employee
124 Views

Hi Ella,

could you clarify: do you use DilateBorder as deprecated Dilate algorithm (without operations on border) or with border operations? 

In any way thank you for the issue reporting, we will investigate this case.

Pavel

P__Ella
Beginner
124 Views

Hi Pavel,

I send to the function an ROI of the image and use ippBorderInMem.

thanks,

Ella

Reply