Livermore Loop09 poor performance with sparse-vector/matrix multiply
| Bugzilla Link | 907 |
| Created on | Jan 22, 2012 21:50 |
| Version | svn |
| OS | Linux |
| Architecture | PC |
| Attachments | loop09.sac, crud.simple.sac, condfun.sac |
Extended Description
Created an attachment (id=845) source code to reproduce failure I rewrote loop09 to use a sparse-vector vs. dense matrix inner product function, which made the code much more readable and maintainable. Unfortunately, it's also much slower than the scalar-oriented C code. It should not be, so I had a look into it. There are several LACFUN-related problems, and AWLF suffers from all of them: 1. The loop09 code uses a FOR-loop of the form: (i=0; i
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information