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.

TDA4VM-Q1: Any restriction of input size of texture to OpenGL?

Part Number: TDA4VM-Q1

Hi, expert

Do GPU have any restrictions of input size of texture, like length or bandh must be multiple of 64 pixel?

Regards

Zekun

  • Zekun,

    Texture's can be different sizes, they don't need to be a multiple of 64 pixels. But you will get better performance when the length is a multiple of 2. What usecase are you trying? If you have a specific example, we can test it out.

    Regards,

    Erick

  • Hi Erick

    Customer test this function in with different input.

    EGLImageKHR appEglWindowCreateIMG(app_egl_obj_t *obj, app_egl_tex_prop_t *prop)

    RTOS SDK 8.5 

    J721S2

    Crop_w: 

    4 failed

    8 failed

    16 failed

    32 failed

    64 succeed

    800 failed

    768 succeed

    Here is the failed log:

    Regards

    Zekun

  • Zekun,

    We will need to investigate this limitation.

    Does the customer need a particular resolution?

    One limitation I am aware of is the need for 64-bit alignment for the memory. This might be the reason resolutions that are not multiples of 64 are not supported, the driver implementation will not do padding.

    Does this answer your question?

    Regards,

    Erick

  • Hi, Erick

    Thanks, I will try to explain to customer.

    Further more, 64 bit alignment requirement for GPU or DMA?

    Regards

    Zekun

  • Hi, Erick

    Customer test on 8.0,  if not align with 64 bit, it will failed.

    But on SDK 8.4, if will succeed without alignment of 64 bits. 

    GPU had a upgrade from 8.0 to 8.4, so is it related to GPU?

    Customer want to know if this need is mandatory or it is just for high performance. If for high performance, is there any way to avoid this requirement?

    Regards

    Zekun

  • Zekun,

    This should not have been the case before as well, I'm surprised the behavior changed. The GPU driver version did change between those two releases (1.13 vs 1.15 DDK).

    As I understand, it is mandatory because going lower will be going smaller than the GPU cache line size, which can cause other issues in the driver. So please stick to using 64-bit alignment or higher.

    Thanks,

    Erick