Only use D3DLOCK_DISCARD if dynamic memory (was Re: d3dx9: CloneMesh test and improvements)

[Michael Mc Donnell]

On Fri, Aug 19, 2011 at 2:20 PM, Stefan Dösinger  wrote:
> Am 19.08.2011 um 14:04 schrieb Michael Mc Donnell:
> > Ok I see that in dlls/d3d9/tests/buffers.c. So it will fail to lock on
> > Win7 if D3DLOCK_DISCARD is specified and the vertex buffer wasn't
> > created with D3DUSAGE_DYNAMIC?
>
> Yes, unless Microsoft changed this again.
> > Am I correct in D3DLOCK_DISCARD makes a copy of the vertex buffer so
> > that the video card can keep reading the old one, and the old vertex
> > buffer memory is released after the video card releases its lock [1]?
>
> Sort of. It doesn't copy the old data, it just gives you a fresh, uninitialized block of memory. So if you read from the pointer you get from a DISCARD lock you'll get garbage.
> > In that case it does not make sense to use D3DLOCK_DISCARD at all in
> > CloneMesh, as the video card has not begun to use the new vertex
> > buffer, and it will just add memory overhead. The easiest fix would be
> > to just pass 0 instead of D3DLOCK_DISCARD. I've done that in the
> > updated patch.
>
> Right, the buffer is newly created in CloneMesh. Just not using DISCARD is the best fix then.
> > > I noticed a few more D3DLOCK_DISCARD locks in the mesh.c code.
> >
> > I guess the other methods should check the mesh options for
> > D3DXMESH_DYNAMIC, D3DXMESH_VB_DYNAMIC, and D3DXMESH_IB_DYNAMIC [2]
> > before using D3DLOCK_DISCARD? I'll write patches for those.
>
> Most likely yes. I haven't looked at the details. If the buffer is new in those situations as well just don't pass any flags, otherwise pass DISCARD if you lock a dynamic buffer and are going to rewrite the entire buffer contents.

All of the places with D3DLOCK_DISCARD in mesh.c were working on newly
created meshes or vertex buffers, so they should all pass 0. I've
attached patches for it.