freedv-gui

History

Mooneer Salem 439e54035f Request real-time priority from the operating system for audio pipeline (#866 ) * Bound requested frames between both min and max. * Request real-time scheduling for pipeline threads (currently macOS-only). * Ensure pipeline threads never execute at the same time as audio threads. * macOS: ask for 5ms audio blocks. * Temporarily re-enable stress test. * Suppress codespell for .mm files. * Swap times/buffer sizes. * Fix compiler error. * Go back to previous frame size and time. * Re-disable stress test. * Try adjusting time quantum. * Move audio priority handling to audio layer. * Set helper thread to real-time on Windows. * Temporarily disable audio workgroup code. * Disable periodicity in macOS RT request * Forgot to define variable. * Only allow a max of 50% CPU time for audio helpers. * More tuning of the duty cycle. * Make sure we sleep some amount of time every time through the loop. * Go back to smaller time quantum. * Another quantum tweak. * Increase audio sample block size to 20ms. * Use 2.5ms for audio block time. * Try increasing FIFO sizes for RX. * Oops, forgot to comment out the / 1000 too. * Get faster turnaround for macOS GH builds. * Revert FIFO changes. * Add additional debugging. * Fix compiler error. * Fix typo. * Use semaphores to notify TX/RX worker threads. * Try dividing duty cycle by 2 to avoid starvation. * Reenable audio workgroups. * No point in having parallel threads if only one mode is active. * Ensure that ParallelStep gets added to the audio workgroup too. * Anything for GUI consumption shouldn't be in the RT path. * Go back to 10ms audio blocks to see if reporting tests can more reliably pass. * Fix issue causing audio to not work on certain Macs. * Support real-time threading on Linux when using PulseAudio/pipewire. * Fix misspelling. * dbus needs to be installed in the environment. * macOS: try a smaller block size again. * Revert "macOS: try a smaller block size again." This reverts commit `1d21ad6934`. * Try architecture-specific audio block times. * rtkit itself also needs to be installed in the environment. * Revert ordering changes to macOS CI builds. * GH user needs to be added to rtkit group. * Implement semaphores on Windows. * Don't exit the render/capture threads unless requested by higher level code. * Don't take address of a pointer. * macOS: set thread QoS for RT threads. * Move RADE RX/TX memory allocations out of real-time path. * Moving additional memory allocations outside of RT context. * Move all remaining allocations in pipeline outside of RT except for legacy FreeDV modes. * Longer audio block times don't seem to be making a difference. * Move legacy mode memory allocations out of RT context. * Prevent unbounded mutex locks inside pipeline. * Windows: fallback to a simple sleep if we can't lock the semaphore. * Increase maximum wait time to match what Windows returns for buffer size. * PulseAudio: fallback to simple sleep if we can't wait on semaphore. * Prevent unbounded locking in TapStep.	2025-04-23 00:18:24 -07:00
..
ISSUE_TEMPLATE	OK, trying again.	2024-10-21 07:07:13 -07:00
workflows	Request real-time priority from the operating system for audio pipeline (#866 )	2025-04-23 00:18:24 -07:00

Request real-time priority from the operating system for audio pipeline (#866 )

* Bound requested frames between both min and max.

* Request real-time scheduling for pipeline threads (currently macOS-only).

* Ensure pipeline threads never execute at the same time as audio threads.

* macOS: ask for 5ms audio blocks.

* Temporarily re-enable stress test.

* Suppress codespell for .mm files.

* Swap times/buffer sizes.

* Fix compiler error.

* Go back to previous frame size and time.

* Re-disable stress test.

* Try adjusting time quantum.

* Move audio priority handling to audio layer.

* Set helper thread to real-time on Windows.

* Temporarily disable audio workgroup code.

* Disable periodicity in macOS RT request

* Forgot to define variable.

* Only allow a max of 50% CPU time for audio helpers.

* More tuning of the duty cycle.

* Make sure we sleep some amount of time every time through the loop.

* Go back to smaller time quantum.

* Another quantum tweak.

* Increase audio sample block size to 20ms.

* Use 2.5ms for audio block time.

* Try increasing FIFO sizes for RX.

* Oops, forgot to comment out the / 1000 too.

* Get faster turnaround for macOS GH builds.

* Revert FIFO changes.

* Add additional debugging.

* Fix compiler error.

* Fix typo.

* Use semaphores to notify TX/RX worker threads.

* Try dividing duty cycle by 2 to avoid starvation.

* Reenable audio workgroups.

* No point in having parallel threads if only one mode is active.

* Ensure that ParallelStep gets added to the audio workgroup too.

* Anything for GUI consumption shouldn't be in the RT path.

* Go back to 10ms audio blocks to see if reporting tests can more reliably pass.

* Fix issue causing audio to not work on certain Macs.

* Support real-time threading on Linux when using PulseAudio/pipewire.

* Fix misspelling.

* dbus needs to be installed in the environment.

* macOS: try a smaller block size again.

* Revert "macOS: try a smaller block size again."

This reverts commit 1d21ad6934.

* Try architecture-specific audio block times.

* rtkit itself also needs to be installed in the environment.

* Revert ordering changes to macOS CI builds.

* GH user needs to be added to rtkit group.

* Implement semaphores on Windows.

* Don't exit the render/capture threads unless requested by higher level code.

* Don't take address of a pointer.

* macOS: set thread QoS for RT threads.

* Move RADE RX/TX memory allocations out of real-time path.

* Moving additional memory allocations outside of RT context.

* Move all remaining allocations in pipeline outside of RT except for legacy FreeDV modes.

* Longer audio block times don't seem to be making a difference.

* Move legacy mode memory allocations out of RT context.

* Prevent unbounded mutex locks inside pipeline.

* Windows: fallback to a simple sleep if we can't lock the semaphore.

* Increase maximum wait time to match what Windows returns for buffer size.

* PulseAudio: fallback to simple sleep if we can't wait on semaphore.

* Prevent unbounded locking in TapStep.

2025-04-23 00:18:24 -07:00

ISSUE_TEMPLATE OK, trying again. 2024-10-21 07:07:13 -07:00

workflows Request real-time priority from the operating system for audio pipeline (#866 ) 2025-04-23 00:18:24 -07:00