[D] Current state of Native kernels for Deep Learning on different architectures

I am new to Deep Learning, but have some experience with embedded systems, having worked on DSP algorithms like FFT, IFFT, Convolution etc. I have even written those kernels in ARM assembly and on Digital Signal Processors like ADI's Blackfin, TI's C6000, so I know a little bit about optimizing math operations and pipelines on different architectures.

From whatever I have learnt so far, the very core of DL computations just seem to be matrix multiplications and additions. I was …

algorithms architectures arm assembly convolution current deep learning digital dsp embedded etc experience fft machinelearning processors signal state systems