Hi, As a number of you might know, there is a performance opportunity in Ruby when run on SPARC. The quick summary of the issue is that the Ruby code (both 1.8* and 1.9* branches) contains flush register windows trap instructions in a hot part of the code. These traps do not appear to be necessary, and removing them leads to 10% gains on a number of the standard benchmarks, and for some of the benchmarks the performance nearly doubles.
I've documented details of the change here: http://blogs.sun.com/d/entry/ruby_performance_gains_on_sparc I'd love to have someone work with me on confirming that these instructions are indeed unnecessary, and also in getting the changes back into the trunk. Many thanks, Darryl. -- Darryl Gove Compiler Performance Engineering Blog: http://blogs.sun.com/d/ Book: http://www.sun.com/books/catalog/solaris_app_programming.xml
