Abstract: A method for matching neural network layouts to hardware architectures is provided. The method comprises iteratively: holding neural network parameters constant while changing a hardware architecture parameters, calculating a first loss value for a combination of the neural network parameters and hardware architecture parameters according to a gradient-based differentiable function within specified resource constraints, holding the hardware architecture parameters constant while changing the neural network parameters, calculating a second loss value for a new combination of parameters within the specified resource constraints, and combining the first loss value and the second loss value to calculate a combined loss value. The above iterative steps are stopped when the combined loss value reaches a specified threshold, and an optimal combination of neural network parameters and hardware architecture parameters is determined according to the combined loss value.

Abstract: Systems and methods are provided for efficient screen capture and video recording on mobile and other devices. The video is recorded using a graphics rendering pipeline that includes a primary frame buffer, a secondary frame buffer, and a video writer module that encodes and writes data to a video file. The frame buffers include multiple textures to which graphics data can be rendered or copied, with at least the secondary frame buffer textures backed with memory that is quickly accessible by a central processing unit. In operation, a frame is rendered into a texture in the primary frame buffer, and the contents of the rendered frame are copied to a texture of the secondary frame buffer as well as to a default graphics rendering pipeline for output to a display. The contents of the rendered frame are then provided from the secondary frame buffer to the video writer for output to a video file.