MMX-enabled Dolby Digital decoder

Software decoding of Dolby Digital (an adaptive transform-based coder using a frequency-linear, critically sampled filterbank) allows it to become a baseline capability on the personal computer (PC), with greater flexibility than a hardware approach. Intel´s MMX technology provides instructions that can significantly speed up the execution of the Dolby Digital decoder, freeing up the processor to perform other tasks such as video decoding and/or audio enhancement. Intel´s MMX instructions operate on 8, 16, and 32 bits. The smaller the data size, the more operations per instruction are performed. Using 16 bits of accuracy uniformly through a Dolby Digital decoder is insufficient to pass the test suite. The challenge was to obtain both good execution speed and good audio quality. Although 32-bit floating-point numbers could be used throughout the data path and only use MMX technology for bit manipulation, this would not be the most processor-efficient method. To this end, we used 16-bit SIMD (single-instruction, multiple data) operations during much of the decoder, but performed 8- and 32-bit SIMD operations on certain sections. While we discuss a particular use of MMX technology, the MMX instruction set is general purpose in nature. We provide a description of MMX technology and then describe the major functional blocks of a Dolby Digital decoder and the special techniques used that take advantage of MMX technology. We also include a description of precision enhancements that were implemented to maintain accuracy and a description of other performance enhancements that were made. We conclude with results in terms of efficient processor utilization, numerical accuracy, and audio quality