On Dec 8, 2012, at 7:15 AM, Erik Verbruggen <[email protected]> wrote:
> > On Nov 27, 2012, at 22:01, Douglas Gregor <[email protected]> wrote: > >> This is expected to be a performance win. Is it? > > I think so. Disclaimer: I didn't do proper statistics, and my laptop is not > super "quiet". I took basetexteditor.cpp and botan.cpp from Qt Creator, > preprocessed them, and ran them through "time clang -fsyntax-only" 10 times. > The best user times: > > botan.cpp (102777 lines preprocessed): > Before patch: > real 0m1.580s > user 0m1.544s > sys 0m0.030s > After patch: > real 0m1.551s > user 0m1.514s > sys 0m0.034s > (so that's 2% faster?) > > basetexteditor.cpp (78717 lines preprocessed): > Before patch: > real 0m0.980s > user 0m0.955s > sys 0m0.023s > After patch: > real 0m0.973s > user 0m0.948s > sys 0m0.021s > (so about .8% faster?) > > Now I don't know if these files are proper tests, and how much the > SourceRange/SourceLocations get queried. The reason I took botan.cpp is that > it's doing public-key crypto, while basetexteditor.cpp has lots of > declarations and short method definitions. Both are attached if someone wants > to repeat the tests. > > Any ideas on a better benchmark? I checked Cocoa.h and it was ~0.6% faster. Patch LGTM, thanks! - Doug _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
