Uniprocessor optimization

Here follows a set of short but hopefully instructive examples illustrating the problems of getting good performance on one CPU or thread.

Read this before you begin!   (some general rules for constructing good benchmarks)

Use full optimization unless stated otherwise.
 
Use double precision in all floating point computations.

• +, -, *, /, sin etc.
   
• locality
   
• indirect addressing
   
• inlining  
   
• Lapack
     
• profiling
   
  
• MEX-files, Matlab and band matrices 
   

   Include your own code in the report, but do NOT include LAPACK- or BLAS-routines.