Thanks a lot for the quick answer. I'll definitly take a closer look at neko and include it in our benchmarks. I'll keep a close eye on this SSE API too.
Thanks! 2007/12/6, Nicolas Cannasse <[EMAIL PROTECTED]>: > Marc-Antoine Desroches a écrit : > > Hello, > > > > I work for a game developer and we study the possibility of using haxe > > on neko as a scripting platform but we need to have really good > > performance across the board, including floating point arithmetic. Is > > there a way to have floating point numbers that are not boxed or is > > there a plan to provide such a feature in the future? > > Hello, > > Floating point speed is not that bad, but could be improved.I've been > thinking about this for some time already. > > While it's not possible to modify the VM to unbox floats, it would be > possible to provide a lowlevel API for performing optimized floating > point calculus (that would be turned into FPU/SSE instructions by JIT). > > This API would not be very easy to use directly in Neko, but the haXe > compiler could easily target it, at least for floating point calculus > involving local variables (for calls and object storage, boxing would > still be needed). > > Nicolas > > -- > Neko : One VM to run them all > (http://nekovm.org) > -- Neko : One VM to run them all (http://nekovm.org)
