Re: [PATCH 2/6] d3d9/tests: Use create_device in test_fpu_setup.

[Henri Verbeet]

On 20 October 2014 13:08, Stefan Dösinger <stefan(a)codeweavers.com> wrote:
> @@ -3598,20 +3604,18 @@ static void test_fpu_setup(void)
>      d3d9 = Direct3DCreate9(D3D_SDK_VERSION);
>      ok(!!d3d9, "Failed to create a D3D object.\n");
>
> -    memset(&present_parameters, 0, sizeof(present_parameters));
> -    present_parameters.Windowed = TRUE;
> -    present_parameters.hDeviceWindow = window;
> -    present_parameters.SwapEffect = D3DSWAPEFFECT_DISCARD;
> +    device_desc.device_window = window;
> +    device_desc.width = 640;
> +    device_desc.height = 480;
> +    device_desc.flags = CREATE_DEVICE_HWVP_ONLY;

This essentially matches the existing code, but does the test really
not work with software vertexprocessing? It seems to work for me.

[PATCH 2/6] d3d9/tests: Use create_device in test_fpu_setup.

[Stefan Dösinger]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 2014-10-21 11:37, schrieb Henri Verbeet:
> This essentially matches the existing code, but does the test 
> really not work with software vertexprocessing? It seems to work 
> for me.
> I wasn't sure if the HWVP requirement caused the test to skip on some
> setups like older Intel GPU drivers, which may or may not have
> different fpu setup behavior. Thus I decided to leave it alone in this
> patch.

create_device() already existed with the hwvp/swvp fallback code when
e6611e22 was added. Do you recall any platform specific problems with
fpusetup?

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
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=3451
-----END PGP SIGNATURE-----

[PATCH 2/6] d3d9/tests: Use create_device in test_fpu_setup.

[Henri Verbeet]

On 21 October 2014 11:51, Stefan Dösinger <stefandoesinger(a)gmail.com> wrote:
> Am 2014-10-21 11:37, schrieb Henri Verbeet:
> > This essentially matches the existing code, but does the test
> > really not work with software vertexprocessing? It seems to work
> > for me.
> > I wasn't sure if the HWVP requirement caused the test to skip on some
> > setups like older Intel GPU drivers, which may or may not have
> > different fpu setup behavior. Thus I decided to leave it alone in this
> > patch.
>
> create_device() already existed with the hwvp/swvp fallback code when
> e6611e22 was added. Do you recall any platform specific problems with
> fpusetup?

No, the test didn't use create_device() because of the effort of
fitting D3DCREATE_FPU_PRESERVE in there on the one hand, and caring
less about platforms without hardware vertexprocessing at the time on
the other hand.