Files
lmms/plugins/SpectrumAnalyzer
Michael Gregorius 286e62adf5 Simplify sample frame operations (make it a class) (#7156)
* Remove the struct StereoSample

Remove the struct `StereoSample`. Let `AudioEngine::getPeakValues` return a `sampleFrame` instead.

Adjust the calls in `Mixer`  and `Oscilloscope`.

* Simplify AudioEngine::getPeakValues

* Remove surroundSampleFrame

Some code assumes that `surroundSampleFrame` is interchangeable with `sampleFrame`. Thus, if the line `#define LMMS_DISABLE_SURROUND` is commented out in `lmms_basics.h` then the code does not compile anymore because `surroundSampleFrame` now is defined to be an array with four values instead of two. There also does not seem to be any support for surround sound (four channels instead of two) in the application. The faders and mixers do not seem to support more that two channels and the instruments and effects all expect a `sampleFrame`, i.e. stereo channels. It therefore makes sense to remove the "feature" because it also hinders the improvement of `sampleFrame`, e.g. by making it a class with some convenience methods that act on `sampleFrame` instances.

All occurrences of `surroundSampleFrame` are replaced with `sampleFrame`.

The version of `BufferManager::clear` that takes a `surroundSampleFrame` is removed completely.

The define `SURROUND_CHANNELS` is removed. All its occurrences are replaced with `DEFAULT_CHANNELS`.

Most of the audio devices classes, i.e. classes that inherit from `AudioDevice`, now clamp the configuration parameter between two values of `DEFAULT_CHANNELS`. This can be improved/streamlined later.

`BYTES_PER_SURROUND_FRAME` has been removed as it was not used anywhere anyway.

* Make sampleFrame a class

Make `sampleFrame` a class with several convenience methods. As a first step and demonstration adjust the follow methods to make use of the new functionality:
* `AudioEngine::getPeakValues`: Much more concise now.
* `lmms::MixHelpers::sanitize`: Better structure, better readable, less dereferencing and juggling with indices.
* `AddOp`, `AddMultipliedOp`, `multiply`: Make use of operators. Might become superfluous in the future.

* More operators and methods for sampleFrame

Add some more operators and methods to `sampleFrame`:
* Constructor which initializes both channels from a single sample value
* Assignment operator from a single sample value
* Addition/multiplication operators
* Scalar product

Adjust some more plugins to the new functionality of `sampleFrame`.

* Adjust DelayEffect to methods in sampleFrame

* Use composition instead of inheritance

Using inheritance was the quickest way to enable adding methods to `sampleFrame` without having to reimpement much of `std::array`s interface.

This is changed with this commit. The array is now a member of `sampleFrame` and the interface is extended with the necessary methods `data` and the index operator.

An `average` method was added so that no iterators need to be implemented (see changes in `SampleWaveform.cpp`).

* Apply suggestions from code review

Apply Veratil's suggestions from the code review

Co-authored-by: Kevin Zander <veratil@gmail.com>

* Fix warnings: zeroing non-trivial type

Fix several warnings of the following form:

Warnung: »void* memset(void*, int, size_t)« Säubern eines Objekts von nichttrivialem Typ »class lmms::sampleFrame«; use assignment or value-initialization instead [-Wclass-memaccess]

* Remove unnecessary reinterpret_casts

Remove some unnecessary reinterpret_casts with regards to `sampleFrame` buffers.

`PlayHandle::m_playHandleBuffer` already is a `sampleFrame*` and does not need a reinterpret_cast anymore.

In `LadspaEffect::processAudioBuffer` the `QVarLengthArray` is now directly initialized as an array of `sampleFrame` instances.

I guess in both places the `sampleFrame` previously was a `surroundSampleFrame` which has been removed.

* Clean up zeroSampleFrames code

* Fix warnings in RemotePlugin

Fix some warnings related to calls to `memcpy` in conjunction with`sampleFrame` which is now a class.

Add the helper functions `copyToSampleFrames` and `copyFromSampleFrames` and use them. The first function copies data from a `float` buffer into a `sampleFrame` buffer and the second copies vice versa.

* Rename "sampleFrame" to "SampleFrame"

Uppercase the name of `sampleFrame` so that it uses UpperCamelCase convention.

* Move SampleFrame into its own file

Move the class `SampleFrame` into its own class and remove it from `lmms_basics.h`.

Add forward includes to all headers where possible or include the `SampleFrame` header if it's not just referenced but used.

Add include to all cpp files where necessary.

It's a bit surprising that the `SampleFrame` header does not need to be included much more often in the implementation/cpp files. This is an indicator that it seems to be included via an include chain that at one point includes one of the headers where an include instead of a forward declaration had to be added in this commit.

* Return reference for += and *=

Return a reference for the compound assignment operators `+=` and `-=`.

* Explicit float constructor

Make the  constructor that takes a `float` explicit.

Remove the assignment operator that takes a `float`. Clients must use the
explicit `float` constructor and assign the result.

Adjust the code in "BitInvader" accordingly.

* Use std::fill in zeroSampleFrames

* Use zeroSampleFrames in sanitize

* Replace max with absMax

Replace `SampleFrame::max` with `SampleFrame::absMax`.

Use `absMax` in `DelayEffect::processAudioBuffer`. This should also fix
a buggy implementation of the peak computation.

Add the function `getAbsPeakValues`. It  computes the absolute peak
values for a buffer.

Remove `AudioEngine::getPeakValues`. It's not really the business of the
audio engine. Let `Mixer` and `Oscilloscope` use `getAbsPeakValues`.

* Replace scalarProduct

Replace the rather mathematical method `scalarProduct` with
`sumOfSquaredAmplitudes`. It was always called on itself anyway.

* Remove comment/TODO

* Simplify sanitize

Simplify the `sanitize` function by getting rid of the `bool found` and
by zeroing the buffer as soon as a problem is found.

* Put pointer symbols next to type

* Code review adjustments

* Remove "#pragme once"
* Adjust name of include guard
* Remove superfluous includes (leftovers from previous code changes)

---------

Co-authored-by: Kevin Zander <veratil@gmail.com>
2024-06-30 20:21:19 +02:00
..
2019-07-17 22:45:26 +02:00
2019-07-17 22:45:26 +02:00
2019-07-17 22:45:26 +02:00
2016-07-29 00:40:02 +02:00
2019-07-17 22:45:26 +02:00
2019-07-17 22:45:26 +02:00
2019-11-21 14:44:18 +01:00
2023-08-24 19:16:02 +01:00
2019-07-17 22:45:26 +02:00
2019-07-17 22:45:26 +02:00
2019-07-17 22:45:26 +02:00
2019-07-17 22:45:26 +02:00
2019-07-17 22:45:26 +02:00

Spectrum Analyzer plugin

Overview

This plugin consists of three widgets and back-end code to provide them with required data.

The top-level widget is SaControlDialog. It populates configuration widgets (created dynamically) and instantiates spectrum display widgets. Its main back-end class is SaControls, which holds all configuration values.

SaSpectrumView and SaWaterfallView widgets show the result of spectrum analysis. Their main back-end class is SaProcessor, which performs FFT analysis on data received from the Analyzer class, which in turn handles the interface with LMMS.

Threads

The Spectrum Analyzer is involved in three different threads:

  • Effect mixer thread: periodically calls Analyzer::processAudioBuffer() to provide the plugin with more data. This thread is real-time sensitive -- any latency spikes can potentially cause interruptions in the audio stream. For this reason, Analyzer::processAudioBuffer() must finish as fast as possible and must not call any functions that could cause it to be delayed for unpredictable amount of time. A lock-less ring buffer is used to safely feed data to the FFT analysis thread without risking any latency spikes due to a shared mutex being unavailable at the time of writing.
  • FFT analysis thread: a standalone thread formed by the SaProcessor::analyze() function. Takes in data from the ring buffer, performs FFT analysis and prepares results for display. This thread is not real-time sensitive but excessive locking is discouraged to maintain good performance.
  • GUI thread: periodically triggers paintEvent() of all Qt widgets, including SaSpectrumView and SaWaterfallView. While it is not as sensitive to latency spikes as the effect mixer thread, the paintEvent()s appear to be called sequentially and the execution time of each widget therefore adds to the total time needed to complete one full refresh cycle. This means the maximum frame rate of the Qt GUI will be limited to 1 / total_execution_time. Good performance of the paintEvent() functions should be therefore kept in mind.

Changelog

1.1.2	2019-11-18
	- waterfall is no longer cut short when width limit is reached
	- various small tweaks based on final review
1.1.1	2019-10-13
	- improved interface for accessing SaProcessor private data
	- readme file update
	- other small improvements based on reviews
1.1.0	2019-08-29
	- advanced config: expose hidden constants to user
	- advanced config: add support for FFT window overlapping
	- waterfall: display at native resolution on high-DPI screens
	- waterfall: add cursor and improve label density
	- FFT: fix normalization so that 0 dBFS matches full-scale sinewave
	- FFT: decouple data acquisition from processing and display
	- FFT: separate lock for reallocation (to avoid some needless waiting)
	- moved ranges and other constants to a separate file
	- debug: better performance measurements
	- various performance optimizations
1.0.3	2019-07-25
	- rename and tweak amplitude ranges based on feedback
1.0.2	2019-07-12
	- variety of small changes based on code review
1.0.1	2019-06-02
	- code style changes
	- added tool-tips
	- use const for unmodified arrays passed to fft_helpers
1.0.0	2019-04-07
	- initial release