how to implement high performance Affine Transform in AI core(called D-card or Ascend 310) of Huawei with tvm ?
Please be more specific about the questions, i.e. what are the specific Affine transformations and what are the target requirement. As far as I know, nobody in the community can understand the question(thus answer it) given there is little public info about the specific chip you talked about.
we have a project is to do face recognition. before that is face detection, face correction. the face correction need to use 2d affine transform.
see https://en.wikipedia.org/wiki/Affine_transformation.
and https://docs.opencv.org/3.1.0/d4/d61/tutorial_warp_affine.html.
emmmmm…we need to put the whole algorithm into Ascend 310(also called D-card). the bottle neck is the face correction, we use an opencv function to do it. but Huawei has professional compiler called TE, it also supports TVM stuffs.