mikiobraun
commented
10 years ago Hi Sean,
Thanks! I'll have a look tomorrow (CET), but I don't think there will be issues. I'm also open to any API additions which don't require touching the Fortran parts ;)
-M Am 17.02.2014 18:53 schrieb "Sean Owen" notifications@github.com:
Mikio I was noodling around the matrix code to see if I could optimize a bit the parts used in Spark MLlib. I found one non-trivial optimization that avoids an extra array allocation. I couldn't help fiddling with a few other things, which you can reject.
I'm also checking the appetite for some other optimizations that might involve adding a few more methods to the API (like an "increment value"
method, and "self times transpose") but that's for another time.
You can merge this Pull Request by running
git pull https://github.com/srowen/jblas MatrixConstructorOptimization
Or view, comment on, or merge it at:
https://github.com/mikiobraun/jblas/pull/38 Commit Summary
- Optimize matrix constructor to avoid extra array allocation, and other more trivial quick ones like not compiling regexes repeatedly in loops.
File Changes
- M src/main/java/org/jblas/ComplexDoubleMatrix.javahttps://github.com/mikiobraun/jblas/pull/38/files#diff-0(8)
- M src/main/java/org/jblas/ComplexFloatMatrix.javahttps://github.com/mikiobraun/jblas/pull/38/files#diff-1(8)
- M src/main/java/org/jblas/DoubleMatrix.javahttps://github.com/mikiobraun/jblas/pull/38/files#diff-2(34)
- M src/main/java/org/jblas/FloatMatrix.javahttps://github.com/mikiobraun/jblas/pull/38/files#diff-3(34)
Patch Links:
Reply to this email directly or view it on GitHubhttps://github.com/mikiobraun/jblas/pull/38 .
Mikio I was noodling around the matrix code to see if I could optimize a bit the parts used in Spark MLlib. I found one non-trivial optimization that avoids an extra array allocation. I couldn't help fiddling with a few other things, which you can reject.
I'm also checking the appetite for some other optimizations that might involve adding a few more methods to the API (like an "increment value" method, and "self times transpose") but that's for another time.