[Mesa-dev] [PATCH 2/6] gallium: implement clamping controls (ARB_color_buffer_float)
e0425955 at student.tuwien.ac.at
Mon Mar 21 11:25:43 PDT 2011
On 03/21/2011 05:43 PM, Marek Olšák wrote:
> On Mon, Mar 21, 2011 at 4:50 PM, Brian Paul <brianp at vmware.com> wrote:
>> On 03/21/2011 09:10 AM, Keith Whitwell wrote:
>>> On Mon, 2011-03-21 at 02:12 +0100, Marek Olšák wrote:
>>>> diff --git a/src/gallium/include/pipe/p_state.h
>>>> index cf6c5b5..f6ad456 100644
>>>> --- a/src/gallium/include/pipe/p_state.h
>>>> +++ b/src/gallium/include/pipe/p_state.h
>>>> @@ -81,6 +81,8 @@ struct pipe_rasterizer_state
>>>> unsigned flatshade:1;
>>>> unsigned light_twoside:1;
>>>> + unsigned clamp_vertex_color:1;
>>>> + unsigned clamp_fragment_color:1;
>>>> unsigned front_ccw:1;
>>>> unsigned cull_face:2; /**< PIPE_FACE_x */
>>>> unsigned fill_front:2; /**< PIPE_POLYGON_MODE_x */
>>> Don't know if this affects the overall packing of the struct. Have you
>>> been able to check?
>>> Otherwise the interface changes look good to me.
>> How do most GPUs handle this clamping? Is it done with extra shader code
>> or is it a discrete operation? If most hardware does it with shader code,
>> maybe we should just do it that way too in Gallium and avoid the new cap
>> bits, etc.
> There is a bit in the CB_COLOR[0-7]_INFO register on R600 which controls
> color clamping before blending. I am not sure whether it can be used for
> ARB_color_buffer_float or whether it's strictly dependent on the colorbuffer
> Christoph, do you happen to know whether NV50/NVC0 has a hardware state
> for it too?
there's a command to control per-RT fragment colour clamping and one to
enable/disable vertex colour clamping. No fiddling with shaders.
Not sure where exactly the clamping happens, but it's supposed to behave
like OpenGL requires it to.
(So we just call them FRAG_COLOR_CLAMP_EN and VERT_COLOR_CLAMP_EN,
on nv50 the latter is part of the color semantic/linkage setup, on nvc0
it always applies to the dedicated colour outputs.)
>> OpenGL 3.0 deprecates the per-vertex and per-fragment clamp controls so
>> people will have to do it in their shaders eventually.
>> Re: your comment in rasterizer.rst:
>> XXX: this happens _before_ the geometry shader and thus might not belong
>> So is there some question about where the per-vertex clamp is to be done?
>> It seems to me that it should be done after the geometry shader.
> You're right. I will remove that comment.
More information about the mesa-dev