On Thursday 12 December 2013 Dec 19:06:20 Pau Garcia i Quiles wrote: > On Wed, Dec 11, 2013 at 10:21 AM, Simon Hausmann > <simon.hausm...@digia.com>wrote: > > > > > The other option for 5.2 would be to do runtime detection of SSE2 and > > > enable/disable the JIT according to that. > > > > I think the main issue with run-time detection is that we need -msse2 > > -mfpmath=sse in order to get reliable math results (no 80-bit temps) that > > match with our test expectations, when using the interpreter or non-inline > > math functions. > > > > I have not looked into the actual code, I know everybody hates dlopen and I > haven't even though about the performance implications but, what about this? > > - Move "maybe-SSE2" code to a separate library > - Compile that library twice, one with SSE2 (-msse2 -mfmath=sse), the other > without SSE2 > - Ship both versions of the library: libwhatever_sse2.so, > libwhatever_i486.so > - On runtime, check for SSE2 and load the appropriate version > > The same approach could be used in other parts of Qt, or even for having > SSE2, SSE2, etc versions of that library/plugin >
In Krita, we use Vc to abstract away all this stuff -- we code our vectorization optimizations using Vc and then Vc takes care to select the best alternative run-time. I don't know how well it would fit with Qt, but I'm very glad with it: http://code.compeng.uni-frankfurt.de/projects/vc For Krita 2.8, we added support for AVX2, but that's still in a separate branch. -- Boudewijn Rempt http://www.valdyas.org, http://www.krita.org, http://www.boudewijnrempt.nl _______________________________________________ Development mailing list Development@qt-project.org http://lists.qt-project.org/mailman/listinfo/development
