[Mesa-dev] [PATCH 06/22] gallium: add semaphore_server_signal()

[Andres Rodriguez]

On 2017-12-22 10:56 AM, Roland Scheidegger wrote:
> Am 22.12.2017 um 01:41 schrieb Andres Rodriguez:
>> Calling this function will emit a semaphore signal operation into the
>> GPU's command stream.
>>
>> Signed-off-by: Andres Rodriguez <andresx7 at gmail.com>
>> ---
>>   src/gallium/docs/source/context.rst  | 31 +++++++++++++++++++++++++++++++
>>   src/gallium/include/pipe/p_context.h |  6 ++++++
>>   2 files changed, 37 insertions(+)
>>
>> diff --git a/src/gallium/docs/source/context.rst b/src/gallium/docs/source/context.rst
>> index a8ff3dc..bab9613 100644
>> --- a/src/gallium/docs/source/context.rst
>> +++ b/src/gallium/docs/source/context.rst
>> @@ -567,7 +567,38 @@ by a single pipe_screen and is not shared with another process.
>>   (i.e. you shouldn't use it to flush caches explicitly if you want to e.g.
>>   use the resource for texturing)
>>   
>> +Semaphores
>> +^^^^^^^^^^
>>   
>> +``pipe_semaphore_handle``, and related methods, are used to synchronize
>> +execution between multiple parties. Examples include CPU <-> GPU syncrhonization,
>> +rederer <-> windowing system, multiple external APIs, etc.
>> +
>> +A ``pipe_semaphore_handle`` can either be 'one time use' or 're-usable'. A 'one time use'
>> +semaphore behaves like a traditional GPU fence. Once it reaches the signaled state it
>> +is forever considered to be signaled.
>> +
>> +Once a re-usable ``pipe_semaphore_handle`` becomes signaled, it can be reset
>> +back into an unsignaled state. The ``pipe_semaphore_handle`` will be reset to
>> +the unsignaled state by performing a wait operation on said object, i.e.
>> +``semaphore_server_sync``. As a corollary to this behaviour, a re-usable
>> +``pipe_semaphore_handle`` can only have one waiter.
> 
> I've only skimmed through this, but how are one-time use semaphores
> (which can have multiple waiters) and re-usable ones (only one waiter)
> distinguished? Just by the use of semaphore_server_sync?
> 

Yes, the difference is that only re-usable semaphores can be used for 
semaphore_server_sync(). I should probably document that better and add 
an assert so that it is also clear on the code side.

If anything else that is unclear let me know.

-Andres

> Roland
> 
> 
> 
>> +
>> +This behaviour is useful in producer <-> consumer chains. It helps avoid
>> +unecessarily sharing a new ``pipe_semaphore_handle`` each time a new frame is
>> +ready. Instead, the semaphores are exchanged once ahead of time, and access is synchronized
>> +through GPU signaling instead of direct producer <-> consumer communication.
>> +
>> +``semaphore_server_sync`` inserts a wait command into the GPU's command stream.
>> +
>> +``semaphore_server_signal`` inserts a signal command into the GPU's command stream.
>> +
>> +There are no guarantees that the wait/signal commands will be flushed when
>> +calling ``semaphore_server_sync`` or ``semaphore_server_signal``. An explicit
>> +call to ``flush`` is required to make sure the commands are emitted to the GPU.
>> +
>> +The Gallium implementation may implicitly ``flush`` the command stream during a
>> +``semaphore_server_sync`` or ``semaphore_server_signal`` call if necessary.
>>   
>>   Resource Busy Queries
>>   ^^^^^^^^^^^^^^^^^^^^^
>> diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
>> index 72e4b9d..5827491 100644
>> --- a/src/gallium/include/pipe/p_context.h
>> +++ b/src/gallium/include/pipe/p_context.h
>> @@ -527,6 +527,12 @@ struct pipe_context {
>>                                    struct pipe_semaphore_handle *semaphore);
>>   
>>      /**
>> +    * Insert commands to have the GPU signal a semaphore.
>> +    */
>> +   void (*semaphore_server_signal)(struct pipe_context *pipe,
>> +                                   struct pipe_semaphore_handle *semaphore);
>> +
>> +   /**
>>       * Create a view on a texture to be used by a shader stage.
>>       */
>>      struct pipe_sampler_view * (*create_sampler_view)(struct pipe_context *ctx,
>>
>