spra666_Hand-Tuning Loops and Control Code on the TMS320C6000.pdf

Application Report SPRA666 – August 2006 1 Hand-Tuning Loops and Control Code on the TMS320C6000 Elana Granston Software anization ABSTRACT The piler automatically perfo rms a great deal of performance-related tuning. piler-driven optimization usuall y suffices. For the occasional cases where additional CPU performance is needed, this application report presents strategies and examples for improving performance of C/ C++ applications. Memory-related performance improvements (such as background DMA transfers or cache usage) are outside the scope of this report. The techniques apply to all me mbers of the C6000 architecture family. The target audience is intermediate to advanc ed application developers. Familiarity with the C6000 architecture and experience developing code for this architecture is assumed. Contents 1 Introduction ................................................................................................................... .................. 3 2 Overview of the TMS320C6000 .................................................................................................... .. 3 TMS320C6000 Architecture .......................................................................................... 3 piler............................................................................................... 5 3 General Performance Tuning Strategy.......................................................................................... 6 Selecting the piler Options............................................................................ 6 Basic Performance Tuning Tips..................................................................................... 9 Exploiting ments (-s)............................................................................ 10 4 Tuning Software-Pipelined Loops............................................................................................... 11 Using restrict qualifiers, MUST_ITERATE