This paper investigates the use of idle graphics processors to accelerate audio DSP for real-time algorithms. Several common algorithms have been identified for acceleration and were executed in multiple thread and block configurations to ascertain the desired configuration for the different algorithms. The GPU and CPU performing on the same data sizes and algorithm are compared against each other. From these results the paper discusses the importance of optimizing the code for GPU operation including the allocating shared resources, optimizing memory transfers, and forced serialization of feedback loops. It also introduces a new method for audio processing using GPU’s as the default processor instead of an accelerator.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=17443
(1060KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
