Benchmarks
Mainly to test the performance of the OpenBlas, Eigen, NNPACK, ComputeLibrary library's matrix multiplication function and convolution function on various devices.
Matrix Multiplication
Compare the matrix multiplication performance of Eigen and OpenBlas library.
[TODO] Add a matrix multiplication function based on ARM ComputeLibrary and its performance test.
Depthwise Convolution
For the depthwise convolution in mobilenet, the performance of the NeonDepthwiseConv(a convolution optimization based on ARM-Neon instruction) is 7-10 times higher than the GemmConv based on matrix multiplication(use openblas).