IACA is by far the best tool I've seen for computation kernel development - you can iterate quickly and see the bottlenecks exactly. Once everything is balanced you check on real hardware with the performance counters. It also has the advantage of being able to test against architectures you don't have real hardware for.
Tim, do I interpret your answer correctly when I say it seems to imply that IACA will no longer be developed? Any chance of open sourcing it? I assume Parallel Studio Advisor is not a free tool. Those of us working for free on open source projects cannot always spend large amounts of money on proprietary software, but more importantly we can't require others to. I can bring the results of an IACA analysis to a discussion since anyone else can verify it. It is much tougher, outside of a commercial environment, to bring results from a proprietary tool, since the others involved often cannot reproduce it without an expensive license for said tool.
