Michael Meissner continued where he left off at the summit regarding Function Specific optimizations. The basic idea is to be able to control the target options initially on the function being compiled and be able to generate different code for the same functions. The idea seems to allow for dynamic dispatch of function versions of very hot functions that could result in better performance on architectures that supported it. The wiki proposal is available here and the mailing list discussion can be found here.
Mark sent out the 4.3 status report and noted that we are still 34 bugs away from a release branch . The full details of the thread can be found here.
The 2 other major topics of discussion this week were regarding improving debug information from GCC and the contentious plugins project with the overall consensus being that plugins are good for quick prototyping and so on but might not be maintainable in the long run given that API stability is not a given with GCC. So, it appears as though productizing a plugin with GCC might not work in the long term given the pace at which things change but might be good enough for a researcher to do a quick prototype. Ofcourse the legal angle is different and it would be interesting to see what the FSF thinks of this project vis-a-vis the GPL. There's more to come surely on this. I didn't quite follow the debug information thread - so no reports on that.
You can find the discussion on debug information here and the discussion on plugins starting here.
No comments:
Post a Comment