[2/5] d3d9: The depth clamp test fails on Nvidia Windows drivers

[Stefan Dösinger]

Am Montag 05 Juli 2010 17:41:36 schrieb Henri Verbeet:
> On 5 July 2010 17:11, Stefan Dösinger <stefandoesinger at gmx.at> wrote:
> > I tried a lot of things, and I couldn't get these tests working on this
> > box(gf9600,win7 x86). d3drs_clipping doesn't do what the test expects, in
> > either d3d version. If anyone knows what's wrong feel free to bring up a
> > better solution.
> 
> There's probably an issue with depth buffer precision and the
> direction of the depth compare function.
I tested various depth test parameters, including disabling the Z buffer and 
changing the depth compare logic. Under no conditions I managed to get any 
geometry with Z > 1.0 drawn.

> > Wrt ddraw/d3d8, I'd prefer to get the d3d9 testsuite working on my box
> > first. Currently I get thousands of test failures, making  development of
> > new bugfixes pretty much impossible.
> 
> Sounds like a problem with your setup. I get about 90 failures in Win
> 7 with an AMD card, with an nvidia card you should be getting
> something like 25. What makes fixing tests difficult is that you need
> to understand the failure before you mark something broken.
The Stream source tests break the device, and no draw after them succeeds, 
causing all other tests to fail. I haven't figured out what exactly is wrong in 
those tests, so far I disabled them in my tree to attend to other tests. Maybe 
it is just some invalid state change that sticks, maybe 

Among the other failures:
* My system refuses ps_3_0 without a vs_3_0 vertex shader and vice versa, 
breaking many pixel shader tests. Easy to fix by providing a simple vertex 
shader. Consequently the shader version varying tests are moot.
* d3d10 sRGB+Fog. That issue is well known.
* Problems with viewport > framebuffer sizes(unlike MinZ/MaxZ it doesn't reject 
the SetViewport call, but rather the following draw call)
* Clears with a NULL rectangle array need a clear count of 0, not 1. Easy to 
fix.

I have all the patches from the Windows tree attached. I think there are no 
test failures left with them, but patch 8 comments out a number of tests. 
Patch 12 needs to move the check for vertex shader support.

The Stream source tests, among other things, test what happens when a vertex 
shader reads an attribute that is not listed in the vertex declaration, and 
what happens when you use nonstandard attribute formats for fixed function 
attribs(e.g. D3DCOLOR texcoords, FLOAT4 colors). As far as I can see either of 
the 3 tests is enough to break the device.
 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: w0.zip
Type: application/zip
Size: 21986 bytes
Desc: not available
URL: <http://www.winehq.org/pipermail/wine-devel/attachments/20100705/cf316491/attachment-0001.zip>