This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

[TIOVX] About Remap

Hello There,

In J7ES Linux Performance Report:

Index Kernel Variant Frame Size (Pixels) Graph Performance (msec)

Node Performance

(msec)

86 Remap Bilinear Interpolation *640x480 (307200) 9.176000 9.093000

This remap C66X kernel is fully optimized? 

Besides, Do we have other remap kernel which based on C71X?or VPAC?

Best Regards,

Samuel

 

  • Samuel,

    From the C66x DSP Core code point of view, it has been optimized within the parameters of the OpenVX spec definition.  Since this could be *any* remap using float coordinates from OpenVX spec, it is somewhat generic and flexible, so the operation operates fully through the L2 cache instead of using DMA acceleration.  That is the main bottleneck, so using some kind of DMA on  the reads could be scope for further optimization.

    However, as you mentioned, the VPAC LDC module does a hardware accelerated remap, so that is recommended.  Please refer to tivxVpacLdcNode API.

    Best Regards,

    Jesse

  • Hi Jesse,

    Thanks for the reply, Do you know where i can found the performance of VPAC in remap function.

    Thanks a lot!

    Best Regards,

    SAMUEL

  • We haven't added the performance report code for the unit tests for VPAC LDC node yet.  If you want to play with it, you can also using the OpenVX performance API to retrieve the node performance.

    For your reference, we have measured 380 output MPixels/s in bilateral mode in single cam usecase. I think this was at 500MHz clock.  Bicubic mode runs at half the rate of bilateral mode.  This can vary a little based on DDR loading and clock frequencies.

    Regards,

    Jesse