While i am using the Opencv3's UMat for TAPI, the result shows that some algorithm of opencv isn't support by TAPI.
I have test the Canny function and GaussianBlur function and so on, It greatly optimized the algorithm, this is the result i test under TI's DSP Accelerator:
DSP sides almost six times more efficient than cpu side.
But if i use the erode or dilate it can't been accelerate by opencl, it almost the same time cost under cpu or DSP side.
So,i really don't know the reason that the dsp can't accelerate the algorithm.I guess that the TI's DSP is not the same as the GPU structure, the TI's C66 serial dsp processor only have about ten processor,but the dsp have so many adders and multiplying units, this will be the CU while accelerate the algorithm, while the erode algorithm is a logical decision algorithm which don't need the adder and multiplying units, so it could not accelerate the algs.
Is there anyone who has test the erode functions under Opencv3-TAPI with GPU accelerator? Is the erode algorithm been accelerated greatly? Thanks a lot!