[Moonlight-list] Fwd: silverlight 3 features

Chris Toshok toshok at gmail.com
Tue Nov 10 16:17:09 EST 2009


On 11/10/2009 11:10 AM, Rodrigo Kumpera wrote:
>
>
> On Tue, Nov 10, 2009 at 4:59 PM, Chris Toshok <toshok at gmail.com 
> <mailto:toshok at gmail.com>> wrote:
>
>     On 11/10/2009 08:19 AM, Rodrigo Kumpera wrote:
>>
>>
>>     On Tue, Nov 10, 2009 at 1:52 PM, Chris Toshok <toshok at gmail.com
>>     <mailto:toshok at gmail.com>> wrote:
>>
>>
>>>         TODO
>>>         ====
>>>
>>>         Pixel shaders, perspective transformations and bitmap caching can be
>>>         implemented in a similar way to how the opacity mask property in the
>>>         UIElement class is handled.
>>>
>>>         I recommend that we use Gallium3D for pixel processing
>>>
>>>         Gallium3D
>>>            + MIT licenced
>>>            + SSE2 optimized software implementation
>>>            + LLVM implementation
>>>            + Can be hardware accelerated if necessary
>>>
>>>         We need to convert .ps effect files to TGSI
>>>
>>>            
>>         Hm, the only possible problem is download size.  how large is
>>         gallium3d (the portions we need, anyway) when built?
>>
>>
>>
>>     There is another issue to consider, the shader has to be verified
>>     to be well formed.
>>
>     right but we have that issue regardless of whether or not we use
>     Gallium3D.
>
>
> Not really, if we emit managed code to apply the pixel shaders, we 
> offload the verification work to the current IL verifier.
The transformation from .ps to IL necessarily includes some sort of 
verification.  Like, if the shader references input 25, but there is no 
input 25, clearly the shader is invalid, but since referencing input 25 
results in a generated IL call to shaderContext.GetInput(25), it's not 
something we'd ever hit as part of IL verification.

Chris
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.ximian.com/pipermail/moonlight-list/attachments/20091110/3d30ff72/attachment.html 


More information about the Moonlight-list mailing list