How to perform floating point computing on matrix with OpenVX?


#1

Hello everybody,

Is there anyway to compute operations such as addition, subtract, multiply, inverse, etc. on matrices of float type?
As I found out in the OpenVX specification 1.2. (https://www.khronos.org/registry/OpenVX/specs/1.2/html/d0/d84/page_design.html#sec_vision_functions), there are too few functions supporting float_32 data type, and none of them are one of the above operations.

Please help me.

Thanks in advance,
Regard


#2

You are correct. Most of the OpenVX standard kernels do not operate on floating point, since OpenVX is intended to be a cross-platform open embedded standard which might be implemented on some hardware accelerators or architectures which may not support floating point operations. You can use a custom user kernel, or perhaps check if your vendor of the OpenVX implementation has an extension which has added floating point versions of these kernels.