Abstract: A GPGPU accelerated audio mixing algorithm for multi-party collaboration namely GAM is introduced. The mixing is performed by pixel shader. To minimize time used while buffer switching, traditional mixing algorithm was rearranged to a two-pass rendering. By vertex shader, the projection of 3-D model was moved between the passes. So multiple functions resulting different sized output set were performed on one render buffer without multiple shader executions on one pixel. The audio units to be mixed were inputted in one texture to remedy the issue that the I/O bandwidth of GPGPU was not match to the computational capacity of GPGPU. The experimental result shows that the GAM algorithm achieves high performance contrasting to CPU-based algorithms on mainstream hardware.

Key words: GPGPU, audio mixing, parallel algorithm

摘要： 提出了一种名为GAM的基于GPGPU的多目的混音算法。该算法通过像素渲染器完成混音。为了减少缓冲区切换造成的时间占用，该算法将传统算法优化为两遍渲染。三维模型的投影位置由顶点渲染器控制，使得多个具有不同输出集合维度的算法可以在同一个渲染缓冲区上并存，并且避免了对同一像素的多次渲染。待混音的数据单元以单一纹理的形式输入，缓解了GPGPU I/O能力及其计算能力间不匹配的问题。实验结果表明，在主流硬件平台上，该算法较基于CPU实现的算法具有较高的加速比。

关键词: GPGPU, 混音, 并行算法