It is a value type. But I think MS assumes that the function calls will be inlined anyway, and the no copy ends up on the stack.
Thomas Tomiczek THONA Software & Consulting Ltd. (Microsoft MVP C#/.NET) (CTO PowerNodes Ltd.) > -----Original Message----- > From: Unmoderated discussion of advanced .NET topics. > [mailto:[EMAIL PROTECTED] On Behalf Of > Shawn A. Van Ness > Sent: Montag, 17. Mai 2004 07:29 > To: [EMAIL PROTECTED] > Subject: Re: [ADVANCED-DOTNET] Large structures > > > This seems like a big waste of time > > Arlie, are you concerned more about the time (ie, the memcpy > operation) or excess growth of the stack (eg, in recursive > scenarios)? Or both? > > I don't have the managed dx sdk handy -- is their Matrix > really a valuetype? > In System.Drawing.Drawing2D, it's a class. Somewhat ironic, > considering that one's only 2x3 floats. :/ > > I'd be shocked to learn that the JIT we have today gives > special treatment to large valuetypes -- it'd be a pretty > obscure opt, if not for the DirectX calls you cite. But I > agree it sounds like it would be feasible. > (Especially in scenarios where the JIT also decides to inline > the function in question -- wouldn't even need to pass a > pointer, on x86 -- just emit opcodes which read the fields > from somewhere above the ebp, instead of below > it.) > > Cheers, > -Shawn > http://www.windojitsu.com/ > > > -----Original Message----- > From: Arlie Davis [mailto:[EMAIL PROTECTED] > Sent: Sunday, May 16, 2004 11:51 > Subject: Large structures > > I have a few questions for the advanced CLR people here. I > haven't been able to find any satisfactory answers elsewhere. > > Consider value types / structures in CLR, and passing them as > arguments to methods. Does the MS CLR *always* copy the > entire structure onto the stack? > Or does it ever simply pass a pointer to the structure? > > I like value types. However, structures of any significant > size are just impractical, if they must always be copied. > For example, in managed DirectX, nearly all of the matrix > manipulation methods take "Matrix foo" as an argument, never > "ref Matrix foo". These matrices are 4x4 floats, which means > that 64 bytes must be copied for every single matrix that is > passed into a method. > > This seems like a big waste of time, compared to the same > thing in C++, where you would simply always pass a pointer to > the matrix. Does the CLR actually pass an entire Matrix on > the stack? Or just a pointer? > > It seems like it would be within the specification / > requirements of the CLR to allow a CLR implementation to do > this, if it realizes that the method never alters the > structure, only examines its fields. > > Any insight is appreciated. > > -- arlie > > =================================== > 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 > > =================================== > This list is hosted by DevelopMentor(r) 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 > > =================================== 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
