That's all nice and applicable to 3D programming, but it's not really relevant. No matter how fancy your shaders are, and how powerful your GPU is -- you will still need to manipulate matrices in your app.
Doing 3D in a managed language should not impose any undue / silly penalties, especially when those penalties are easily avoidable (ref Matrix foo). Frans, I've seen enough of your posts on this list to think that you have a fair share of clues about CLR. But this post seems below your usual quality. You do, of course, realize the difference between reference and value types in CLR, don't you? Allocating reference types is *cheap*, but it is far from *free*. -- arlie -----Original Message----- From: Unmoderated discussion of advanced .NET topics. [mailto:[EMAIL PROTECTED] On Behalf Of Frans Bouma Sent: Monday, May 17, 2004 4:19 AM To: [EMAIL PROTECTED] Subject: Re: [ADVANCED-DOTNET] Large structures > Taken > > But in this particular case, being heavy matrix opreations in 3d, this > kills your performance. If you want performance, write a shader and do your matrix manipulations on the GPU or call into a C++ matrix lib. I also don't really see why classes would kill performance really. Admitted, in C++ I used structs for coordinates and some own written math lib for working with those structs, however a struct is a class in C++ anyway (they're synonyms, IIRC) so why would: public class Vertex3 { public float X; public float Y; public float Z; } be slow in .NET? Graphic programming can only be done fast if you parallellize operations by delegating the tasks to specialized units, often a library written to do a given kind of operation really fast (optimized math lib) or for example a DX helper lib, or better: a shader which does manipulation of your data down the pipe in hardware. I don't think 3D vector math in C# will be that performant over a C++ version. And in the end, if you insist doing the math in C#, you can of course fall back to the good old days method when we all used assembler: 3 arrays of floats or a 3 dim. array of floats. Do your math in place and convert it to an array of vertex objects to send it down the pipe to the hardware. But if you ask me, see the .NET layer in a DX program which relies heavily on graphics performance as a thin layer to keep the program together, not the core application: the power is not in the .NET layer, it's in the DX layer so don't try to do more in the .NET layer than absolutely necessary. > Purity is nice, but here we talk of criticals, too. if performance is critical, use the right objects in the layer-stack to do the work :) So, I'd suggest to call into C++ math libs from C# and pass on the result to DX, or better: write a shader. Much faster :) FB =================================== This list is hosted by DevelopMentor® http://www.develop.com Some .NET courses you may be interested in: NEW! Guerrilla ASP.NET, 17 May 2004, in Los Angeles http://www.develop.com/courses/gaspdotnetls View archives and manage your subscription(s) at http://discuss.develop.com
