[Mesa-dev] [PATCH] [RFC] egl: propose simple EGL_MESA_image_dma_buf_export v2.2

[Dave Airlie]

On 4 March 2015 at 05:23, Chad Versace <chad.versace at intel.com> wrote:
> On 03/03/2015 11:13 AM, Dave Airlie wrote:
>> On 4 March 2015 at 05:06, Chad Versace <chad.versace at intel.com> wrote:
>>> On 03/03/2015 11:01 AM, Daniel Stone wrote:
>>>> Hi,
>>>>
>>>> On 3 March 2015 at 18:40, Chad Versace <chad.versace at intel.com> wrote:
>>>>> On 03/03/2015 12:13 AM, Daniel Stone wrote:
>>>>>> On Tuesday, March 3, 2015, Dave Airlie <airlied at gmail.com <mailto:airlied at gmail.com>> wrote:
>>>>>>     +    EGLBoolean eglExportDMABUFImageQueryMESA(EGLDisplay dpy,
>>>>>>     +                                  EGLImageKHR image,
>>>>>>     +                                 int *fourcc,
>>>>>>     +                                 int *num_planes);
>>>>>>
>>>>>>
>>>>>> Shouldn't this contain the modifier(s) as well?
>>>>>
>>>>> What is the motivation for defining two export calls, and exporting
>>>>> a subset of information from each call? I don't understand why the
>>>>> fourcc values are exported by one call, but the strides exported a
>>>>> separate call.
>>>>
>>>> Querying the image gives you what you need in order to know if you can
>>>> actually deal with the image, before you go to all the effort of
>>>> exporting it.
>>>
>>> Thanks. That makes sense.
>>
>> Also you need the number of planes to correctly size the arrays for
>> the second call to put the results into.
>
> Right, I understood that you needed to know the number of planes before
> submitting the array. I was mostly confused by why fourcc was in this call,
> not num_planes.
>
> My real feedback: It's probably best to rename num_planes to num_fds,
> because you're not guaranteed num_fds != num_planes for interleaved formats.
> And, using the name 'num_fds' allows us to avoid awkwardness if we need to
> create future layered extensions
> that export auxilliary buffers that may not correspond YUV planes. (There's
> been a lot of lunch talk about exporting aux buffers at Intel).


But num_planes sizes fds, strides and offsets, which seems like
we should either just return the same fd for each plane or one fd in slot0
and -1, -1 in the others if we need one fd per multi-planar..

Dave.