Regarding BF609 UDS

Hello,

BF60x's hardware reference says UDS supports the output frame size which is selectable in steps of 16: 16 x n x 16 x m where n,m  = 1,2,...8.

I don't understand what that writing means. I would like to know whether or not scaling down to 640 x 480 from 1280 x 720 is supported.

Best Regards,

Mack. 

  • 0
    •  Analog Employees 
    on Jul 4, 2012 2:15 AM

    Hi Mack,

    As your question relates to the BF60x, I have moved it from the Blackfin Processors Community to the BF60x sub-community. Let's continue the discussion here.

     

    Thanks,

    Prashant

  • 0
    •  Analog Employees 
    on Jul 4, 2012 2:33 AM

    Hi Mack,

    In the Up-Down Scaler module information, it is mentioned that,

    • the module supports input frame sizes from 10x10 to 1280x 960.

    • Supports independent scaling in both directions. The output frame size is selectable in steps of 16. The values supported are: 16 × n × 16 × m where n,m = 1,2,..8.

    The meaning of these statements is simple...., any frame of size from 10x10 to 1280x 960 can be upscaled or downscaled to the frame of size from 16x16 to 128x128 (in the steps of 16).

    So, if your frame size is 1280 x 720, then using UDS module, it can be down-scaled frame of size 128x128 max.

    I am not sure if the frame of size 1280x720 can be downscaled to 640x480 size by using any other PVP module (or their combination).

    But it can be downscaled to 640x360 frame size by using Convolution module of PVP, which can be also support downscaling.

    The Horizontal Scaling factor (HSF) and Vertical Scaling factor (VSF) fields of CNVn_SCALE register can be used for this purpose. Refer Convolution Module Description given in the HRM.

    Let me know if you have any questions.

  • 0
    •  Analog Employees 
    on Jul 4, 2012 3:44 AM

    Hi Mac,

    Sorry I don't have information about it. I guess, one reason could be complexity in implementing it in hardware. Suppose you have frame of size 10x10, then converting it (upscaling) into 16x16 or 128x128 is not so easy because multipliers are not intergral multiple. Similarly downscaling it from any frame size to 128x128 or other frame size could also be difficult.

    The downscaling in convolution block is different. It supports only downscaling using integer dividers. So, downscaling in hardware is easy there.

    This is what my opinion is.

    Thanks,

    Prashant

  • Hi Prashant,

    I got it. Thanks for your answer.

    One more question. Is there any reason to restrict the size of output frame to 128 x 128 max? I think that is so small.

    Regards,

    Mack.