Multi-Threaded Wavetable Synthesizer clicks. Is this caused by torn reads?
I'm currently working on a very basic wavetable synthesizer. I'll try my best to give an understandable overview.
I have a single lookup-waveform with 4096 points, which is sampled by the audio thread. In general, however, a single oscillator has multiple waveforms, which can be morphed by a position parameter. For example, let's take two sine waves where the second one has half the period of the first one. When we play those back at frequency = 440Hz and position = 0.5, we would hear two equally loud tones at 440Hz and 880Hz. Here's a link to the audio (See Sidenote 2).
Due to performance reasons I generate the lookup table on another thread whenever the position parameter changes. And therein lies my problem. Even when I don't change the table position and just periodically generate my lookup table, I get weird, unpredictable glitches. Since the audio data in the table doesn't change in this case, I can only explain this by torn reads. Is that correct? And if so, how do I get around it? Do I use a double-buffer with an atomic pointer-swap after I finish generating the data?
Again, here's the audio for this recording (See Sidenote 2). Also, some oscilloscope data:
Sidenote 1: Populating my lookup table takes ~50,000ns without optimizations and ~10,000ns with optimizations. But the glitches seem to happen with the same probability.
Sidenote 2: The website I uploaded the audio to seems to have introduced some weird harmonics, which are not audible in my video. Just so you know.
Found it! I accidentally set an old scanner position after generating my lookup table. If the latter took too long my position was basically reset to a few microseconds earlier, which explains the visual setback in the oscilloscope.
- .NET Core application deadlocks when reading stdout and stderr from a child process - how to troubleshoot using WinDBG?
- RuntimeError: You cannot use AsyncToSync in the same thread as an async event loop
- C# IMemoryCache: append to list in one operation?
- Performing calculation over huge dataset in mongoDB with nodejs express.js in backend code | Optimising time and memory
- Timeout on a function call
- Should a return statement be inside or outside a lock?
- Use multiple service or do all operations in single service, which is preferable?
- RESTful framework class field is safe or unsafe
- How to share a boolean flag with a scoped thread in Rust?
- Can compiler optimizations avoid writing to memory by operating exclusively in registers?
- Read Introduction in C# - how to protect against it?
- Perform a function only once in a method call across multiple threads in C#
- My worker threads are not being picked up
- Do threads share local variables?
- FileSystemWatcher losing files in its queue
- Decrease priority of maven build threads/processes
- What resources are shared between threads?
- Do Java 21 virtual threads address the main reason to switch to reactive single-thread frameworks?
- Understanding the volatile Modifier in the Context of x86 Architecture and the Java Memory Model (JMM)
- What is the point for asyncio synchronization primitives not to be thread safe?
- Why use a ReentrantLock if one can use synchronized(this)?
- Winui3 Desktop; ObservableCollection, updating UI when property changes? Updating from different thread
- Is there a good way to distinguish spurious wake up and Thread.interrupt()?
- std::lock_guard or std::scoped_lock?
- Synchronizing on String objects in Java
- Point Cloud Library (PCL) filtering timeout
- Managing multiple exclusive locks for a set of unique values
- Run external command in bg (other thread), capture output when available for rendering in TUI
- Why did entering a lock on a UI thread trigger an OnPaint event?
- What's the difference between "static" and "dynamic" schedule in OpenMP?