SK-TDA4VM: how to accelerate scale

that almost is matching with the HW timing. Since it is around 6.5MP, it is going to take around 14ms for processing nv12 image.. We can bring it down to around 10ms by using both the scalars, but other than this, there is no other solution.. 

Regards,

Brijesh

how to using both the scalars, can you describe details?

There are two MSC scalars, we can process luma in one scalar and chroma in another scalar. That can reduce the total time, but not sure about sw support.. 

how to process luma in one scalar and chroma in another scalar, if the input image type is nv12.

Its not supported in the SW. So not possible to improve this process time.. 

Why not supported in the SW, which api is not supported?

No existing usecase to support it. 

if I set input image type as VX_DF_IMAGE_U8, the buff is uv plane, can tivxVpacMscScaleNode work correctly?

Hi,

But driver will assume it as luma plane. This will not give correct result. 

Probably, you can update the scalar node to support both luma only and chroma only operations and then you can have two nodes, one for luma only and the other for chroma only operations, both can be running on different MSC module and can then improve the total performance..  But it requires change in the current MSC node.. 

Regards,

Brijesh

Can you give some suggestions how to update the scale node to support both luma only and chroma only operations?

You can add some control command in the scalar node and then in this control command processing, you can change the format to chroma only mode..