Nowadays, it's inevitable to select programmable DSP processor for the developement of DSP Application system, if you must get fast time-to market. DSP compilers are software tools which enables programming of DSP processors in a high-level language (HLL). Instead of using asssembly language, the programmer expresses the computation in a language which is more easily understood and used.
Unfortunately, the relatively low efficiency of DSP compilers is a serious limitation. We must find techniques about code quality improvement for overcome the situation. Traditional compiler optimizations are insufficient for generating efficient and compact code. So it is necessary to use target dependent optimization technique for better compiler code performance.
This thesis describes three code optimization techniques newly implemented for MetaCore DSP compiler. Lowering loop overhead with H/W DO block, Using memory operand at arithmetic operation for accumulator based processor and Applying complex instructions instead of several primitive instructions are these techniques. With above code optimizationmethods, the optimizer generates code which is comparale to the code developed by programmers by hand.
