|
      
|
|
|
|
This will be an informal talk on the work I've done for my PhD, including the loose-ends that I am currently trying to tie up. From a computing perspective the broad area is compiling, but my project is unusual in that it is quite application specific. The talk comes in two parts, the first being the rationale behind and development of a set of benchmarks from the domain of sparse linear algebra, in a (mostly) functional language called Aldor (although the functional aspects are not really the main thrust of the talk). The second part is the compiler optimisation of the codes, concentrating specifically on minimising the impact of the main-memory bandwidth bottleneck. I'll will give some comparisons with BLAS routines, and describe my current experiments to compare my work against a FORTRAN equivalent.
