Cain: Automatic code generation for simultaneous convolutional kernels on focal-plane sensor-processors

Focal-plane Sensor-processors (FPSPs) are a camera technology that enables low power, high frame rate computation, making the device suitable for edge computation. Unfortunately, the device’s limited instruction set and registers make the development of complex algorithms challenging. In this work, we present Cain – a compiler that targets SCAMP-5, a general-purpose FPSP – which generates SCAMP-5 code from multiple convolutional kernels. As an example, given the convolutional kernels for an MNIST digit recognition neural network, Cain produces code that is half as long, when compared to the other available compilers for SCAMP-5