[Mesa-dev] [PATCH] st/mesa: enable GL_KHR_robustness
Bas Nieuwenhuizen
bas at basnieuwenhuizen.nl
Thu Sep 29 09:09:07 UTC 2016
On Thu, Sep 29, 2016 at 10:20 AM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> On 29.09.2016 00:00, Bas Nieuwenhuizen wrote:
>>
>> On Wed, Sep 28, 2016 at 6:27 PM, Nicolai Hähnle <nhaehnle at gmail.com>
>> wrote:
>>>
>>> On 28.09.2016 16:20, Ilia Mirkin wrote:
>>>>
>>>>
>>>> On Wed, Sep 28, 2016 at 6:25 AM, Nicolai Hähnle <nhaehnle at gmail.com>
>>>> wrote:
>>>>>
>>>>>
>>>>> From: Nicolai Hähnle <nicolai.haehnle at amd.com>
>>>>>
>>>>> The difference to the virtually identical ARB_robustness (which is
>>>>> already
>>>>> enabled unconditionally) is miniscule and handled elsewhere, but this
>>>>> set
>>>>> of caps seems like the right thing to require for this extension.
>>>>> ---
>>>>> docs/features.txt | 2 +-
>>>>> docs/relnotes/12.1.0.html | 1 +
>>>>> src/mesa/state_tracker/st_extensions.c | 4 ++++
>>>>> 3 files changed, 6 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/docs/features.txt b/docs/features.txt
>>>>> index fbb3952..52d194e 100644
>>>>> --- a/docs/features.txt
>>>>> +++ b/docs/features.txt
>>>>> @@ -211,21 +211,21 @@ GL 4.5, GLSL 4.50:
>>>>> GL_ARB_ES3_1_compatibility DONE
>>>>> (i965/hsw+,
>>>>> nvc0, radeonsi)
>>>>> GL_ARB_clip_control DONE (i965,
>>>>> nv50, nvc0, r600, radeonsi, llvmpipe, softpipe, swr)
>>>>> GL_ARB_conditional_render_inverted DONE (i965,
>>>>> nv50, nvc0, r600, radeonsi, llvmpipe, softpipe, swr)
>>>>> GL_ARB_cull_distance DONE (i965,
>>>>> nv50, nvc0, radeonsi, llvmpipe, softpipe, swr)
>>>>> GL_ARB_derivative_control DONE (i965,
>>>>> nv50, nvc0, r600, radeonsi)
>>>>> GL_ARB_direct_state_access DONE (all
>>>>> drivers)
>>>>> GL_ARB_get_texture_sub_image DONE (all
>>>>> drivers)
>>>>> GL_ARB_shader_texture_image_samples DONE (i965,
>>>>> nv50, nvc0, r600, radeonsi)
>>>>> GL_ARB_texture_barrier DONE (i965,
>>>>> nv50, nvc0, r600, radeonsi)
>>>>> GL_KHR_context_flush_control DONE (all -
>>>>> but
>>>>> needs GLX/EGL extension to be useful)
>>>>> - GL_KHR_robustness DONE (i965)
>>>>> + GL_KHR_robustness DONE (i965,
>>>>> radeonsi)
>>>>> GL_EXT_shader_integer_mix DONE (all
>>>>> drivers that support GLSL)
>>>>>
>>>>> These are the extensions cherry-picked to make GLES 3.1
>>>>> GLES3.1, GLSL ES 3.1 -- all DONE: i965/hsw+, nvc0, radeonsi
>>>>>
>>>>> GL_ARB_arrays_of_arrays DONE (all
>>>>> drivers that support GLSL 1.30)
>>>>> GL_ARB_compute_shader DONE
>>>>> (i965/gen7+, softpipe)
>>>>> GL_ARB_draw_indirect DONE
>>>>> (i965/gen7+, r600, llvmpipe, softpipe, swr)
>>>>> GL_ARB_explicit_uniform_location DONE (all
>>>>> drivers that support GLSL)
>>>>> GL_ARB_framebuffer_no_attachments DONE
>>>>> (i965/gen7+, r600, softpipe)
>>>>> diff --git a/docs/relnotes/12.1.0.html b/docs/relnotes/12.1.0.html
>>>>> index cdd8909..0c99f19 100644
>>>>> --- a/docs/relnotes/12.1.0.html
>>>>> +++ b/docs/relnotes/12.1.0.html
>>>>> @@ -52,20 +52,21 @@ Note: some of the new features are only available
>>>>> with certain drivers.
>>>>> <li>GL_ARB_cull_distance on radeonsi</li>
>>>>> <li>GL_ARB_enhanced_layouts on i965</li>
>>>>> <li>GL_ARB_indirect_parameters on radeonsi</li>
>>>>> <li>GL_ARB_shader_draw_parameters on radeonsi</li>
>>>>> <li>GL_ARB_shader_group_vote on nvc0</li>
>>>>> <li>GL_ARB_shader_viewport_layer_array on i965/gen6+</li>
>>>>> <li>GL_ARB_stencil_texturing on i965/hsw</li>
>>>>> <li>GL_ARB_texture_stencil8 on i965/hsw</li>
>>>>> <li>GL_EXT_window_rectangles on nv50, nvc0</li>
>>>>> <li>GL_KHR_blend_equation_advanced on i965</li>
>>>>> +<li>GL_KHR_robustness on radeonsi</li>
>>>>> <li>GL_KHR_texture_compression_astc_sliced_3d on i965</li>
>>>>> <li>GL_OES_copy_image on nv50, nvc0, r600, radeonsi, softpipe,
>>>>> llvmpipe</li>
>>>>> <li>GL_OES_geometry_shader on i965/gen8+, nvc0, radeonsi</li>
>>>>> <li>GL_OES_primitive_bounding_box on i965/gen7+, nvc0, radeonsi</li>
>>>>> <li>GL_OES_texture_cube_map_array on i965/gen8+, nvc0, radeonsi</li>
>>>>> <li>GL_OES_tessellation_shader on i965/gen7+, nvc0, radeonsi</li>
>>>>> <li>GL_OES_viewport_array on nvc0, radeonsi</li>
>>>>> <li>GL_ANDROID_extension_pack_es31a on i965/gen9+</li>
>>>>> </ul>
>>>>>
>>>>> diff --git a/src/mesa/state_tracker/st_extensions.c
>>>>> b/src/mesa/state_tracker/st_extensions.c
>>>>> index 4f42217..4789f2c 100644
>>>>> --- a/src/mesa/state_tracker/st_extensions.c
>>>>> +++ b/src/mesa/state_tracker/st_extensions.c
>>>>> @@ -1192,20 +1192,24 @@ void st_init_extensions(struct pipe_screen
>>>>> *screen,
>>>>> consts->MaxComputeWorkGroupCount[i] = grid_size[i];
>>>>> consts->MaxComputeWorkGroupSize[i] = block_size[i];
>>>>> }
>>>>>
>>>>> extensions->ARB_compute_shader =
>>>>>
>>>>> extensions->ARB_shader_image_load_store &&
>>>>>
>>>>> extensions->ARB_shader_atomic_counters;
>>>>> }
>>>>> }
>>>>>
>>>>> + extensions->KHR_robustness =
>>>>> + extensions->ARB_robust_buffer_access_behavior &&
>>>>> + screen->get_param(screen, PIPE_CAP_DEVICE_RESET_STATUS_QUERY);
>>>>
>>>>
>>>>
>>>> Is that necessary? From what I can tell, a no-op implementation is
>>>> sufficient for KHR_robustness.
>>>
>>>
>>>
>>> The extension does talk about buffer robustness. For the reset
>>> notification
>>> business, I also don't think the spec really has a lot of teeth. I just
>>> thought that checking this cap is what's most in line with the spirit of
>>> the
>>> extension.
>>
>>
>> I think it has some teeth though:
>>
>> "After a graphics reset has occurred on a context, subsequent GL commands
>> on that context (or any context which shares with that context) will
>> generate a CONTEXT_LOST error."
>> (https://www.opengl.org/registry/specs/KHR/robustness.txt)
>>
>> Since IIRC gallium uses LOSE_CONTEXT_ON_RESET as reset notification
>> behavior, we should at least generate the CONTEXT_LOST errors after a
>> GetGraphicsResetStatus call that indicates a reset.
>>
>> I agree that this can be somewhat theoretical though.
>
>
> Mesa core does that for us, already -- see _mesa_GetGraphicsResetStatusARB.
>
> Now to be honest, I was also wondering about how to interpret that spec
> fragment precisely. What Mesa does is start the CONTEXT_LOST behavior only
> if the application calls GetGraphicsResetStatus. Perhaps we should start
> returning CONTEXT_LOST also if an application doesn't do that? Checking for
> resets on every command is clearly too much; the intel driver does it on
> flush, I guess we could do the same.
I would also check when a fence wait has timed out and it has been too
long since submission of that CS. That way we get all places where the
result of a reset could become visible.
- Bas
>
> Cheers,
> Nicolai
>
>
>> - Bas
>>
>>
>>
>>>
>>> Right now, I think it only makes a difference for nvc0. I can drop that
>>> part
>>> of the check if you prefer. I really don't have a strong opinion about
>>> this.
>>>
>>> Cheers,
>>> Nicolai
>>>
>>>
>>>>
>>>>> +
>>>>> /* If we support ES 3.1, we support the ES3_1_compatibility ext.
>>>>> However
>>>>> * there's no clean way of telling whether we would support ES 3.1
>>>>> from
>>>>> * here, so copy the condition from compute_version_es2 here. A lot
>>>>> of
>>>>> * these are redunant, but simpler to just have a (near-)exact copy
>>>>> here.
>>>>> */
>>>>> extensions->ARB_ES3_1_compatibility =
>>>>> extensions->ARB_ES3_compatibility &&
>>>>> extensions->ARB_arrays_of_arrays &&
>>>>> extensions->ARB_compute_shader &&
>>>>> extensions->ARB_draw_indirect &&
>>>>> --
>>>>> 2.7.4
>>>>>
>>>>> _______________________________________________
>>>>> mesa-dev mailing list
>>>>> mesa-dev at lists.freedesktop.org
>>>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>>
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list