[Mesa-dev] EGL/android: pbuffer implementation.

[Tomasz Figa]

On Tue, Nov 22, 2016 at 2:44 PM, Tapani Pälli <tapani.palli at intel.com> wrote:
>
>
> On 11/21/2016 04:15 PM, Emil Velikov wrote:
>>
>> On 21 November 2016 at 07:23, Tomasz Figa <tfiga at chromium.org> wrote:
>>>
>>> Hi,
>>>
>>> On Wed, Nov 16, 2016 at 11:11 AM, Liu Zhiquan <zhiquan.liu at intel.com>
>>> wrote:
>>>> @@ -775,14 +837,6 @@ droid_add_configs_for_visuals(_EGLDriver *drv,
>>>> _EGLDisplay *dpy)
>>>>     for (i = 0; dri2_dpy->driver_configs[i]; i++) {
>>>>        const EGLint surface_type = EGL_WINDOW_BIT | EGL_PBUFFER_BIT;
>>>>        struct dri2_egl_config *dri2_conf;
>>>> -      unsigned int double_buffered = 0;
>>>> -
>>>> -      dri2_dpy->core->getConfigAttrib(dri2_dpy->driver_configs[i],
>>>> -         __DRI_ATTRIB_DOUBLE_BUFFER, &double_buffered);
>>>> -
>>>> -      /* support only double buffered configs */
>>>> -      if (!double_buffered)
>>>> -         continue;
>>>
>>>
>>> Does it really just work like this? Last time I checked the list of
>>> configs generated after such change it contained single-buffered-only
>>> configs with window surface bit set and double-buffered-only configs
>>> with pbuffer bit set, both of which are invalid. Moreover, this caused
>>> some dEQP tests to fail. How was this patch tested?
>>>
>> Afaict the patch has (should have?) gone through the Intel CI,
>> although I'm not sure if the latter builds/runs Android/ARC.
>
>
> It did go through CI but it does not run any Android/ARC tests. AFAIK
> Zhiquan did run pbuffer related tests and reported them passing on Android.
> Which dEQP tests were regressed?

The tests I'm concerned with are all
dEQP-EGL.functional.query_context.get_current_context.*_pbuffer tests.
In my testing on BSW and BDW, a change identical to the part quoted
above made eglGetConfigs() return double buffered configs for
pbuffers, which made the mentioned tests fail because of invalid
config at surface creation time (or SIGSEGV, before my NULL pointer
fix - 94282b6dd0 egl/android: Check return value of
dri2_get_dri_config()).

Note that I haven't tested exactly this patch, just a similar change,
identical in terms of the quoted part above. It is also possible that
something changed in other part of the EGL stack, which fixed the
problem. I will retest current Mesa master a bit later.

Best regards,
Tomasz