SciPy Signal is commonly used to build signal processing workflows and achieves machine level speed for common operations like filtering and spectrum estimation. cuSignal GPU accelerates the SciPy Signal library by leveraging the GPU PyData ecosystem libraries like Numba, CuPy, and the _cuda_array_interface_ to provide 10-100x speedup over CPU with the same API. In this talk, we discuss the motivations and background behind cuSignal, including how to build GPU accelerated Python libraries and handling zero-copy memory between CPU and GPU.
*****************
/ enthought
/ enthought
/ enthought
29 авг 2024