Hi Anh, The short answer, which applies to almost everything: do the experiment! :-)
The long answer: For data that are truly two-dimensional, use a 2D region. If the X10 compiler fails to generate efficient executable code from this, then the compiler probably needs to be improved. I assume you have read http://x10-lang.org/documentation/performance-tuning.html - specifically, the final section about for loop indexing? Further to that, X10 arrays are optimized for the special cases of rectangular and zero-based arrays. For example, indexing a zero-based array: val x = new Array[Int](0..(N-1) * 0..(N-1)); will perform better than a rectangular, non-zero-based array: val x = new Array[Int](1..N * 1..N); which will in turn perform better than: val r:Region(2) = // arbitrary region val x = new Array[Int](r); It's also worth considering that the X10 Array isn't just a C-style array (i.e. a pointer that allows subscripting). It is a proper object with fields representing the Region, backing storage and some properties. So an Array of Arrays will require additional storage for each internal Array object. Depending on the internal dimension, this could be a significant overhead. Cheers, Josh Anh Trinh wrote: > which one is better performance > 2D region array or Array[Array[Int]]? > > Thanks > > > ------------------------------------------------------------------------------ > RSA® Conference 2012 > Save $700 by Nov 18 > Register now! > http://p.sf.net/sfu/rsa-sfdev2dev1 > _______________________________________________ > X10-users mailing list > X10-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/x10-users > ------------------------------------------------------------------------------ RSA® Conference 2012 Save $700 by Nov 18 Register now! http://p.sf.net/sfu/rsa-sfdev2dev1 _______________________________________________ X10-users mailing list X10-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/x10-users
