-
Steve Lhomme authored
The core needs to know how long the rendering takes place to estimate when a frame needs to be dropped preemptively or not. So far we waited until the render (plus decoding and filtering?) is done before we allowed the picture swap to happen. Now we only wait until (almost) the swap date. If we still have to wait, we'll do it after the swap, hoping it will still happen in time. It is reversing the strict waiting (all?) commands issued to be processed, including all decoder threads and filters on the ID3D11DeviceContext. We no longer allow waiting past the display date and we do the swap when we're told. We no longer release the ID3D11DeviceContext lock either as it allows more commands in and makes the waiting time/performance worse. We do wait until the initial wait condition is done, if anything to silence some D3D11 warnings saying we wanted to wait for something but never got to the point it was actually finished. It also keeps the waiting in the display thread so it doesn't think it can do more things when it's still busy with the previous call. This will allow it do drop frames more accurately. A cleaner way would be to count the Display call in the time to render of the core, but it's not possible as in some case we wait and the work is actually finished and in some others we wait and the work is not finished. We can't tell the difference in the core. Fixes #21600
42011b9f