Re: WineD3D: Implement detection of ATI cards with Mesa DRI drivers [try 4]

[Roderick Colenbrander]

> This is an updated version of my previous patch for the same purpose.
>
> The proposed code now uses the WINE_D3DX_CAPABLE macros instead of DRI 
> module names for guessing the exact card model which is to be returned. 
> The patch also no longer includes manual addition of an entry to the 
> ChangeLog. Thanks to Roderick Colenbrander and Stefan Dösinger for the 
> suggestions. I have tested and confirmed that this patch fixes the bug 
> #7267 (http://bugs.winehq.org/show_bug.cgi?id=7267 ) to the same extent 
> as the previous version of the patch did.

I think the code should be like this (in pseudo code) this way also other cards will at least be classified:

case VENDOR_MESA:
    if(D3D9_CAPABLE(gl_info)
       if(strstr(gl_info->gl_renderer, "R300"))
           card = radeon_9500;
       else
           card = generic_nv_card;
    if(D3D8_CAPABLE(...)
       if(strstr(gl_info->gl_renderer, "R200"))
           card = ...

Roderick

WineD3D: Implement detection of ATI cards with Mesa DRI drivers [try 4]

[Roman Mamedov]

Roderick Colenbrander wrote:
> I think the code should be like this (in pseudo code) this way also other cards will at least be classified:
>
> case VENDOR_MESA:
>     if(D3D9_CAPABLE(gl_info)
>        if(strstr(gl_info->gl_renderer, "R300"))
>            card = radeon_9500;
>        else
>            card = generic_nv_card;
>     if(D3D8_CAPABLE(...)
>        if(strstr(gl_info->gl_renderer, "R200"))
>            card = ...

You are right, but please notice that with my latest patch, the same 
behavior is already achieved, by not "break"-ing at the end of 
VENDOR_MESA case if we didn't match one of ATI module names. This is the 
code that proceeds to be executed then:

...
                break;
            }
            /* Card is probably one of the non-ATI cards which are also 
supported by Mesa.
             * Continue to the fallback behavior of VENDOR_WINE.
             */
        case VENDOR_WINE:
        default:
            /* Default to generic Nvidia hardware based on the supported 
OpenGL extensions. The choice
             * for Nvidia was because the hardware and drivers they make 
are of good quality. This makes
             * them a good generic choice.
             */
            gl_info->gl_vendor = VENDOR_NVIDIA;
            if(WINE_D3D9_CAPABLE(gl_info))
                gl_info->gl_card = CARD_NVIDIA_GEFORCEFX_5600;
            else if(WINE_D3D8_CAPABLE(gl_info))
                gl_info->gl_card = CARD_NVIDIA_GEFORCE3;
            else if(WINE_D3D7_CAPABLE(gl_info))
                gl_info->gl_card = CARD_NVIDIA_GEFORCE;
            else if(WINE_D3D6_CAPABLE(gl_info))
                gl_info->gl_card = CARD_NVIDIA_RIVA_TNT;
            else
                gl_info->gl_card = CARD_NVIDIA_RIVA_128;
...

WineD3D: Implement detection of ATI cards with Mesa DRI drivers [try 4]

[Roderick Colenbrander]

> Roderick Colenbrander wrote:
> > I think the code should be like this (in pseudo code) this way also
>
> other cards will at least be classified:
> > case VENDOR_MESA:
> >     if(D3D9_CAPABLE(gl_info)
> >        if(strstr(gl_info->gl_renderer, "R300"))
> >            card = radeon_9500;
> >        else
> >            card = generic_nv_card;
> >     if(D3D8_CAPABLE(...)
> >        if(strstr(gl_info->gl_renderer, "R200"))
> >            card = ...
>
> You are right, but please notice that with my latest patch, the same 
> behavior is already achieved, by not "break"-ing at the end of 
> VENDOR_MESA case if we didn't match one of ATI module names.

The main reason why the D3DX_CAPABLE macros should be used before checking the renderer string is that not in all cases all extensions are around while the rendering string can still be the same.

Roderick

WineD3D: Implement detection of ATI cards with Mesa DRI drivers [try 4]

[Roman Mamedov]

Roderick Colenbrander wrote:
> The main reason why the D3DX_CAPABLE macros should be used before checking the renderer string is that not in all cases all extensions are around while the rendering string can still be the same.

I think it's is quite difficult to say for sure what will be better in 
most or all cases.

The approach proposed by me, is that if the code can be reasonably sure 
that the card is some model from ATI, it should never report the card as 
being nVIDIA.

That is why the patch first looks at the renderer strings (which for 
Mesa can provide a reliable indication if it's an ATI card at all, or 
not), and only then slides inside the model range for exact model, 
according to the extensions/capabilities available. And in case if some 
of them are not available for some reason, it will return a step-lower 
generation of an ATI card, and not nVIDIA one.

I.e. reporting Radeon 9500 as being Radeon 8500 should be OK, but 
reporting it as GeForce 3 might be not, since a program might then use 
some nVIDIA-specific code, or just bail out completely (see issue #7267).