Re: [PATCH] wined3d: Fix some videocards property detection

[Stefan Dösinger]

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

Hi,

This patch has a few unrelated changes. The core change is that it
re-introduces a hack we once had in Wine and removed because it broke
newer mac GPUs. The CrossOver version of the hack works a bit better
than the unconditional Wine version, but it's still ugly.

Am 2015-03-28 um 06:15 schrieb Sergey Isakov:
> - /* Non-Apple GL vendors report uniforms correctly */
> - if (!match_apple(gl_info, gl_renderer, gl_vendor, card_vendor, device)) return FALSE;
> - /* All dx9 cards are overreported on OSX */
> - if (!match_dx10_capable(gl_info, gl_renderer, gl_vendor, card_vendor, device)) return TRUE;
> - /* Nvidia and Intel DX10 cards support > 256 uniforms */
> - if (card_vendor != HW_VENDOR_AMD) return FALSE;

It is terrible that OSX lies about the GPU capability and then forces
applications to have a crystal ball that tells them what the GPU can
actually do.

- From a pragmatic point of view I don't think OSX will ever change in
this regard. It was a design decision by Apple to do this, and they
stopped updating drivers for the old hardware a long time ago.

I am open to re-introducing the hack if we can find a way to detect the
bug more directly. E.g. try to declare a shader with 257 uniforms, read
them with indirect addressing so the driver can't optimize the access
and then check for software fallbacks. There's a CGL function for
checking that, but I don't remember from the top of my head how it works.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
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=1uP/
-----END PGP SIGNATURE-----

[PATCH] wined3d: Fix some videocards property detection

[Charles Davis]

> On Mar 28, 2015, at 4:21 AM, Stefan Dösinger  wrote:
> then check for software fallbacks. There's a CGL function for
> checking that, but I don't remember from the top of my head how it works.

You mean:

CGLGetParameter(ctx, kCGLCPGPUVertexProcessing, &gpu_vertex);
  CGLGetParameter(ctx, kCGLCPGPUFragmentProcessing, &gpu_frag);

When each one is false, it means that the corresponding stage fell back to
software.

We may need some kind of Wine-specific WGL extension for this, though.
wined3d doesn’t know about CGL contexts.

By the way, those are “get” only. You can’t force the driver to do it in
hardware by forcing them to true (in case you get that idea).

Chip