On Thu, Jun 28, 2012 at 5:41 PM, Paul Mullowney <[email protected]> wrote:

I moved all CUSPARSE functions into seqcusparse/aijcusparse.cu (and a similar mpi version). I was asked to do this some months ago. I did my best. I think it is significantly improved from before.

I was also asked by multiple people to attempt to get Complex GPU capabilities working. I succeeded by making changes to petscmath.h. As Jed pointed out, some of the changes may have not been C99 compliant.

There is one commented out function in aijcusparse.cu. That can easily be fixed.

Many of the other changes (especially to .cu) files were made to protect against builds breaking because many of the CUSP preconditioners do not support complex arithmetic.

You seem to be addressing the specifics of the code; Matt and Jed are talking about the process: patch reviews.

I have asked for input from multiple people on several occasions and not gotten responses.

What am I to do if I ask for input/feedback and don't get it?????

I saw no request for reviewing this patch series. How did you ask people to review? I recommend (for everyone here) smaller commits. Commit often. There is a great extension for editing changesets (i.e. folding multiple changesets together, splitting changesets apart, etc.) called histedit: http://bitbucket.org/durin42/histedit In fact, just today it was included into mercurial core for the next version. I also recommend the 'record' extension (bundled with mercurial). The next version of mercurial (2.3) promises better bookmark functionality (for feature branches), built-in history editing, and possibly evolution (for shared, mutable changesets). If you have a bitbucket account (free), I would recommend working on your own fork there so that others could review and you could mutate the history based on said reviews.