- refactoring of converter: unique entry point is now an ONNX file
- generic converter for multi memory layout (NCHW/NHWC). Every models is transformed in NHWC in SADL. Complex graph layouts are still work in progress.
- added minimal support for 2 new layers (Transpose/Flatten) in order to deal with ONNX data layout adaptation
- added MAC counters (accurate only in generic build)
- make gemm more generic on the stride to benefit from compiler vectorization opportunities
- new simpler sample program: two examples (tf2/pytorch) of a simple classifier provided
- updated license headers
- minor corrections and cleaning
- remove the MACRO hack for classes inspection
- improve print of tensors