d3d9/tests: Fix some test failures on Windows 2000.

[Henri Verbeet]

On 27 February 2014 03:27, Francois Gouget <fgouget at codeweavers.com> wrote:
> The remaining CreateDevice() calls seem to be using pretty different
> parameters.
>
> For instance in test_mipmap_levels() it uses a different backbuffer
> format, does not set its size, sets EnableAutoDepthStencil to FALSE and
> passes D3DDEVTYPE_NULLREF. Will the test really still make sense if all
> that is changed?
>
Yeah.

> Another example is test_depthstenciltest() which sets
> EnableAutoDepthStencil to FALSE but still sets AutoDepthStencilFormat.
> That's different from create_device() and intentionally so.
>
> (test_reset() does so once too without a comment which looks more
> questionable)
>
Yeah, both of those intentionally don't set EnableAutoDepthStencil.

> The problem is there's about 20 remaining CreateDevice() calls and I
> would be hard pressed to tell if they are crucially different from what
> create_device() does.
>
Ok, I'll look into it a bit.