[Mesa-dev] [PATCH 1/3] mesa: Disallow deprecated SNORM formats for renderbuffers

[Marek Olšák]

On Wed, Dec 5, 2012 at 8:39 PM, Eric Anholt <eric at anholt.net> wrote:
> Marek Olšák <maraeo at gmail.com> writes:
>
>> On Tue, Dec 4, 2012 at 7:56 PM, Eric Anholt <eric at anholt.net> wrote:
>>> Ian Romanick <idr at freedesktop.org> writes:
>>>
>>>> From: Ian Romanick <ian.d.romanick at intel.com>
>>>>
>>>> The OpenGL 3.2 core profile spec says:
>>>>
>>>>     "The following base internal formats from table 3.11 are
>>>>     color-renderable: RED, RG, RGB, and RGBA. The sized internal formats
>>>>     from table 3.12 that have a color-renderable base internal format
>>>>     are also color-renderable. No other formats, including compressed
>>>>     internal formats, are color-renderable."
>>>>
>>>> The OpenGL 3.2 compatibility profile spec says (only ALPHA is added):
>>>>
>>>>     "The following base internal formats from table 3.16 are
>>>>     color-renderable: ALPHA, RED, RG, RGB, and RGBA. The sized internal formats
>>>>     from table 3.17 that have a color-renderable base internal format
>>>>     are also color-renderable. No other formats, including compressed
>>>>     internal formats, are color-renderable."
>>>>
>>>> Table 3.12 in the core profile spec and table 3.17 in the compatibility
>>>> profile spec list SNORM formats as having a base internal format of RED,
>>>> RG, RGB, or RGBA.  From this we infer that they should also be color
>>>> renderable.
>>>
>>> This sounds sort of like the description of the bottom half of the
>>> change, to remove L/LA/I snorm formats.  (ALPHA being left in place,
>>> which I missed at first).
>>>
>>> I wonder, do they also not support FBOs with unorm L/LA/I, despite
>>> ARB_framebuffer_object?  fbo-clear-formats should tell.  I'd be happy to
>>> see those go if nobody else supports them either, and it would make this
>>> change make more sense to me.
>>
>> What do you mean by that? R600 has full renderbuffer support for all
>> R, RG, RGBX, RGBA, A, L, LA, and I formats, all can be UNORM, SNORM,
>> UINT, SINT, and FLOAT, and blending and MSAA are a sure thing. I'm
>> only okay with disallowing the formats if all specifications (GL and
>> extensions) agree they *shouldn't* be supported.
>
> ARB_fbo is the only thing that allows L/LA/I to be color-renderable.
> EXT_fbo and all the GL core/compat specs say only R/RG/RGB/RGBA/A are.
> The ARB_fbo spec makes it pretty clear that they expect drivers to say
> "no" for things despite that, just that they don't want to explicitly
> ban it in that spec.  In the core/compat specs, they decided to do the
> reasonable thing instead.
>
> For intel, we need either mesa core to filter out the stupid formats, or
> we can in the driver.  (unfortunately, we can't quite just look at the
> MESA_FORMAT, because those are sometimes promoted from an internalformat
> of LA to a mesa format of RGBA, for example).

I think you can always look at the original "internalformat", which we
need to keep around for glGet.

Anyway, what you say about L/LA/I sounds reasonable. It's not worth
maintaining those formats just because of ARB_fbo. You have my ack for
removal of GL renderbuffer support of L/LA/I.

Marek