| 1 |
As promised, this is my report regarding the benchmarking suite. My |
| 2 |
mid-weekly objectives were: |
| 3 |
|
| 4 |
* Implementing the LAPACK test suite |
| 5 |
* Clean the source code to improve the mainenability |
| 6 |
|
| 7 |
For the lapack testing suite I started considering the routines |
| 8 |
bundled with the sources. Later, I decided to use BTL for lapack too |
| 9 |
-- I already use it for blas and cblas -- and implement a few more |
| 10 |
actions in order to test the following: |
| 11 |
- LU decomposition |
| 12 |
- General linear system of equation solver -- implicit: by means of LU |
| 13 |
- Least squares problem solver -- implicit: by means of QR |
| 14 |
- Cholesky decomposition for SPD matrices |
| 15 |
- Eigenvalues retrieval for symmetric matrices |
| 16 |
|
| 17 |
Everything is now working. As the three basic libraries (blas, cblas, |
| 18 |
lapack) use the same tools for the benchmarking, much was done in |
| 19 |
order to create a single base python script that, with few parameters, |
| 20 |
can benchmark many implementations of a library. The weight of the |
| 21 |
code is now almost the same than some days ago, but the features have |
| 22 |
considerably increased. |
| 23 |
|
| 24 |
Now, beacause of personal reasons, I will be away for ~12 days, which |
| 25 |
was unexpected. I already discussed with my mentor and Donnie. I don't |
| 26 |
know yet what will happen, but I will anyway continue with my project, |
| 27 |
within or not the SoC. |
| 28 |
|
| 29 |
Cheers |
| 30 |
Andrea Arteaga |
Archives