[Mesa-dev] [RFC PATCH shader-db 2/2] run: request a debug context

[Marek Olšák]

On Wed, Dec 2, 2015 at 7:25 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> On Wed, Dec 2, 2015 at 1:15 PM, Marek Olšák <maraeo at gmail.com> wrote:
>> On Wed, Dec 2, 2015 at 6:45 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>> On Mon, Nov 9, 2015 at 2:02 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>>> On Mon, Nov 9, 2015 at 1:56 PM, Matt Turner <mattst88 at gmail.com> wrote:
>>>>> On Mon, Nov 9, 2015 at 10:46 AM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>>>>> On Mon, Nov 9, 2015 at 1:35 PM, Matt Turner <mattst88 at gmail.com> wrote:
>>>>>>> On Sun, Nov 8, 2015 at 8:53 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>>>>>>> st/mesa only prints messages in a debug context. Without always enabling
>>>>>>>> the message generation, I don't see a way to hook into the glEnable() to
>>>>>>>> turn it on/off.
>>>>>>>> ---
>>>>>>>>  run.c | 1 +
>>>>>>>>  1 file changed, 1 insertion(+)
>>>>>>>>
>>>>>>>> diff --git a/run.c b/run.c
>>>>>>>> index 73e468d..1d8d3b1 100644
>>>>>>>> --- a/run.c
>>>>>>>> +++ b/run.c
>>>>>>>> @@ -417,6 +417,7 @@ main(int argc, char **argv)
>>>>>>>>          EGL_CONTEXT_OPENGL_CORE_PROFILE_BIT_KHR,
>>>>>>>>          EGL_CONTEXT_MAJOR_VERSION_KHR, 3,
>>>>>>>>          EGL_CONTEXT_MINOR_VERSION_KHR, 2,
>>>>>>>> +        EGL_CONTEXT_FLAGS_KHR, EGL_CONTEXT_OPENGL_DEBUG_BIT_KHR,
>>>>>>>>          EGL_NONE
>>>>>>>>      };
>>>>>>>>      EGLContext core_ctx = eglCreateContext(egl_dpy, cfg, EGL_NO_CONTEXT,
>>>>>>>
>>>>>>> Is this a limitation of st/mesa, or am I using the KHR_debug API
>>>>>>> incorrectly? It's my understanding that you can get KHR_debug messages
>>>>>>> without a debug context by calling glEnable(GL_DEBUG_OUTPUT), and it
>>>>>>> looks like this is supported by the KHR_debug spec.
>>>>>>
>>>>>> But I don't want to be generating the debug info in my driver for no
>>>>>> reason. For example I have a timing-type debug message which gets the
>>>>>> current time (to report how long a sync fence wait takes). This is a
>>>>>> non-free operation that I want to avoid if no one's looking.
>>>>>
>>>>> Huh, looks like we do quite a bit of work before we throw the message
>>>>> away. We could probably make that process a lot simpler.
>>>>>
>>>>>> I can see this both ways... perhaps the more expensive messages should
>>>>>> be keyed on whether it's a debug context and the compiler message
>>>>>> should always be reported. Or perhaps we can just create a debug
>>>>>> context here and be done with it.
>>>>>
>>>>> I guess you can determine that based on
>>>>> DEBUG_SEVERITY_{HIGH,MEDIUM,LOW,NOTIFICATION}?
>>>>
>>>> There's no driver notification about that (that I'm aware). Should there be?
>>>>
>>>> I don't really see how that can work, since I think you're supposed to
>>>> be able to have multiple ones in a group. But I'm new to that API...
>>>> if there's a driver hook that I missed or variable I should be looking
>>>> at, please educate :)
>>>>
>>>>   -ilia
>>>
>>> So what's the conclusion here? Request debug contexts, or I keep this
>>> patch local for now until such a time that I can redo the gallium impl
>>> to also pass messages through for non-debug contexts and teach nouveau
>>> to look at context type to determine whether to do potentially
>>> expensive debug things instead of whether there's a debug callback
>>> set? [Aka for the foreseeable future.]
>>
>> A gallium state flag that enables the compiler messages would be a
>> cleaner solution I think.
>
> Not sure how that maps onto the two options I outlined... I guess the
> latter? (i.e. I keep this patch local)

Yes. Note that PIPE_CONTEXT_DEBUG is meant of hang debugging, not for
OpenGL debug contexts.

Marek