Simple D3D/WGL demos for wine

[Matteo Bruni]

2016-08-05 22:00 GMT+02:00 Ruslan Kabatsayev <b7.10110111 at gmail.com>:
> On Fri, Aug 5, 2016 at 10:09 PM, Matteo Bruni <matteo.mystral at gmail.com> wrote:
>> 2016-08-04 19:29 GMT+02:00 Ruslan Kabatsayev <b7.10110111 at gmail.com>:
>>> Hello all. I'd like to implement a set of simple demos which would be
>>> something like glxgears for Wine — a simple way to check that WGL,
>>> D3D8, D3D9 and later D3D10 and D3D11 are operational, without any need
>>> to look for a real app to test. I suppose it's not very interesting to
>>> have just flat untextured gears, so I'd instead draw a textured cube
>>> (similarly to what the tests in dxdiag do).
>>
>> Do you plan to include these demos in Wine or maintain them
>> separately? I'd be in favor of including them in Wine proper (others
>> might disagree though) but that introduces a few additional
>> constraints, like no C++ code allowed.
> I intended them to eventually get into Wine source, after I bring them
> to more or less final form. It seems Wine does provide some macros to
> call Direct3D virtual functions from C code — like
> IDirect3D9_CreateDevice.

Actually those macros are also in the original MS headers. Except when
they aren't, like for a bunch of d3dx9 interfaces. In those cases we
shouldn't have them either.
If there is no macro for a method you need you can just manually call
the function pointer through the vtable; see the relevant test sources
for examples of that if it turns out to be the case.

>> BTW, actually we do have a (VERY sketchy) dxdiag implementation in
>> Wine, under programs/dxdiag/. You could integrate those demos into it
>> in principle.
> Good to know it exists, will look into integration of the demos there.
> Do you think the WGL demo still fits there? I'd like to have a demo
> for WGL since it's the simplest 3D interface for Wine, it's used by
> wined3d, and if even it's not working then the users would be better
> directed to where their problems might be (i.e. graphics
> configuration, not wined3d code).

I think it would be fine and nice to have, yes.

>>> I have some prototype code for this (standalone and written in C++ for
>>> now), and it uses D3DXCreateTextureFromFile. This is all OK when I
>>> compile for D3D9 with wineg++ or when I compile for D3D8 and D3D9 with
>>> ming32. But when I try to compile for D3D8 with wineg++, there's a
>>> problem, since there's no d3dx8*.h nor d3dx8.dll in Wine.
>>>
>>> Is it undesirable for Wine to have d3dx8 (I remember some commits
>>> removing the library and headers related to it several years ago)?
>>
>> You remember correctly. It turns out d3dx8 in practice was always
>> statically compiled into the game executables. IIRC the DX8 SDK had no
>> d3dx8.dll at all, just a d3dx8.lib with the full library. There was
>> also a d3dx8_d.dll with a debug version of the library but that wasn't
>> installed by the usual DirectX redistributable so applications can't
>> depend on its presence either. In any case, we've had 0 applications
>> using our d3dx8.dll, which means our implementation wasn't actually
>> used or tested and there was no point in keeping it.
>>
>>> Should I just drop usage of D3DXCreateTextureFromFile and use plain
>>> IDirect3DDeviceX::CreateTexture?
>>
>> You could use some other API to load the image files. I imagine that
>> mixing d3dx9.h with d3d8 won't be a great experience. There are other
>> options though, depending on the texture file format you want to use;
>> WIC (windowscodecs) might be a good choice generally speaking.
>> Another possibility is to create the textures you need procedurally,
>> without needing to load them from a file or a resource.
> For WGL version I used GdiPlus (flat) API, so I guess this is the way
> to go for D3D versions.
> The texture I thought of using was something like the following:
> http://6g6.eu/sih0-winehq-logo-glass.png
> I don't think it'd be easy to generate procedurally, although if
> there's a good idea on what image to use instead, I'd favor the
> procedural version due to its better reliability (i.e. no dependency
> on external file being readable, residing in the expected location
> etc.).

The procedural generation thing was just an idea, loading from file
or, probably better, from resource is okay too.

>>> Also I use some matrix functions, which are identical between d3dx8
>>> and d3dx9. Would it be OK if a wine demo would link to d3dx9 both for
>>> d3d9 and d3d8 versions of it? Or should I just reimplement all the
>>> needed functions?
>>
>> As I mentioned above, I suspect you would get issues when trying to
>> include d3dx9.h together with d3d8.h, although you can probably
>> workaround that in this case by carefully dividing your code over
>> multiple .c files.
>> Depending on what you exactly need, just copying or reimplementing the
>> relevant functions might be simpler.
> Yeah there were some issues with including d3dx9.h directly, I worked
> around them by manually declaring the necessary functions and types
> and still linking to d3dx9 DLL.

I guess that works too. It doesn't seem critical for the time being
and it will probably be easier to give more sensible suggestions
during an actual patch review.
Looking forward to your patches :)