AudioBuffer.h

/*
 * AudioBuffer.h
 *
 * Copyright (c) 2026 Dalton Messmer <messmer.dalton/at/gmail.com>
 *
 * This file is part of LMMS - https://lmms.io
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program (see COPYING); if not, write to the
 * Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
 * Boston, MA 02110-1301 USA.
 *
 */

#ifndef LMMS_AUDIO_BUFFER_H
#define LMMS_AUDIO_BUFFER_H

#include <bitset>
#include <memory_resource>

#include "AudioBufferView.h"
#include "ArrayVector.h"
#include "LmmsTypes.h"
#include "lmms_constants.h"
#include "lmms_export.h"

namespace lmms
{

/**
 * An owning collection of audio channels for an instrument track, mixer channel, or audio processor.
 *
 * Features:
 * - Up to `MaxChannelsPerAudioBuffer` total channels
 * - Audio data in planar format (plus a temporary interleaved buffer for conversions until we use planar only)
 * - All planar buffers are sourced from the same large buffer for better cache locality
 * - Custom allocator support
 * - Silence tracking for each channel (NOTE: requires careful use so that non-silent data is not written to a
 *       channel marked silent without updating that channel's silence flag afterward)
 * - Methods for sanitizing, silencing, and calculating the absolute peak value of channels, and doing so more
 *       efficiently using the data from silence tracking
 * - Can organize channels into arbitrary groups. For example, you could have 6 total channels divided into 2 groups
 *       where the 1st group contains 2 channels (stereo) and the 2nd contains 4 channels (quadraphonic).
 * - Extensive unit testing - See AudioBufferTest.cpp
 *
 * Audio data layout explanation:
 * - All planar audio data for all channels in an AudioBuffer is sourced from the same large contiguous
 *       buffer called the source buffer (m_sourceBuffer).
 * - The source buffer consists of the buffer for 1st channel followed by the buffer for the 2nd channel, and so on
 *       for all channels. In total, the number of elements is `channels * frames`.
 * - A separate vector of non-owning pointers to channel buffers is also maintained. In this vector, each index
 *       corresponds to a channel, providing a mapping from the channel index to a pointer to the start of that
 *       channel's buffer within the source buffer. This is called the access buffer (m_accessBuffer).
 * - The purpose of the access buffer is to provide channel-wise access to buffers within the source buffer, so
 *       it's `m_accessBuffer[channelIdx][frameIdx]` instead of `m_sourceBuffer[channelIdx * frames + frameIdx]`.
 *       This is very important since many APIs dealing with planar audio expect it in this `float**` 2D array form.
 * - Groups have no effect on the audio data layout in the source/access buffers and are merely a layer built on top.
 *       Conveniently, if you take `m_accessBuffer` and offset it by `channelIndex`, you get another `float**`
 *       starting at that channel. This what the `float**` buffer stored in each ChannelGroup is.
 *
 * Naming notes:
 * - When this class is used in an instrument track or mixer channel, its channels could be referred to
 *       as "track channels" or "internal channels", since they are equivalent to the "track channels" used
 *       in other DAWs such as REAPER.
 * - When this class is used in an audio processor or audio plugin, its channels could be referred to
 *       as "processor channels" or "plugin channels".
 */
class LMMS_EXPORT AudioBuffer
{
public:
	using ChannelFlags = std::bitset<MaxChannelsPerAudioBuffer>;

	//! Non-owning collection of audio channels + metadata
	class ChannelGroup
	{
	public:
		ChannelGroup() = default;
		ChannelGroup(float** buffers, ch_cnt_t channels)
			: m_buffers{buffers}
			, m_channels{channels}
		{}

		auto buffers() const -> const float* const* { return m_buffers; }
		auto buffers() -> float** { return m_buffers; }

		auto buffer(ch_cnt_t channel) const -> const float*
		{
			assert(channel < m_channels);
			return m_buffers[channel];
		}

		auto buffer(ch_cnt_t channel) -> float*
		{
			assert(channel < m_channels);
			return m_buffers[channel];
		}

		auto channels() const -> ch_cnt_t { return m_channels; }

		void setBuffers(float** newBuffers) { m_buffers = newBuffers; }
		void setChannels(ch_cnt_t channels) { m_channels = channels; }

		// TODO: Future additions: Group names, type (main/aux), speaker arrangements (for surround sound), ...

	private:
		/**
		 * Provides access to individual channel buffers.
		 * [channel index][frame index]
		 */
		float** m_buffers = nullptr;

		//! Number of channels in `m_buffers` - currently only 2 is used
		ch_cnt_t m_channels = 0;
	};

	AudioBuffer() = delete;

	AudioBuffer(const AudioBuffer&) = delete;
	AudioBuffer(AudioBuffer&&) noexcept = default;
	auto operator=(const AudioBuffer&) -> AudioBuffer& = delete;
	auto operator=(AudioBuffer&&) noexcept -> AudioBuffer& = default;

	/**
	 * Creates AudioBuffer with a 1st (main) channel group.
	 *
	 * Silence tracking is enabled or disabled depending on the auto-quit setting.
	 *
	 * @param frames frame count for each channel
	 * @param channels channel count for the 1st group, or zero to skip adding the 1st group
	 * @param resource memory resource for all buffers
	 */
	explicit AudioBuffer(f_cnt_t frames, ch_cnt_t channels = DEFAULT_CHANNELS,
		std::pmr::memory_resource* resource = std::pmr::get_default_resource());

	/**
	 * Creates AudioBuffer with groups defined.
	 *
	 * Silence tracking is enabled or disabled depending on the auto-quit setting.
	 *
	 * @param frames frame count for each channel
	 * @param channels total channel count
	 * @param groups group count
	 * @param resource memory resource for all buffers
	 * @param groupVisitor see @ref setGroups
	 */
	template<class F>
	AudioBuffer(f_cnt_t frames, ch_cnt_t channels, group_cnt_t groups,
		std::pmr::memory_resource* resource, F&& groupVisitor)
		: AudioBuffer{frames, channels, resource}
	{
		setGroups(groups, std::forward<F>(groupVisitor));
	}

	//! The presence of the temporary interleaved buffer is opt-in. Call this to create it.
	void allocateInterleavedBuffer();

	auto hasInterleavedBuffer() const -> bool { return !m_interleavedBuffer.empty(); }

	/**
	 * @returns the number of bytes needed to allocate buffers with given frame and channel counts.
	 *          Useful for preallocating a buffer for a shared memory resource.
	 */
	static auto allocationSize(f_cnt_t frames, ch_cnt_t channels,
		bool withInterleavedBuffer = false) -> std::size_t;

	//! @returns current number of channel groups
	auto groupCount() const -> group_cnt_t { return static_cast<group_cnt_t>(m_groups.size()); }

	auto group(group_cnt_t index) const -> const ChannelGroup& { return m_groups[index]; }
	auto group(group_cnt_t index) -> ChannelGroup& { return m_groups[index]; }

	//! @returns the buffers for all channel groups
	auto allBuffers() const -> PlanarBufferView<const float>
	{
		return {m_accessBuffer.data(), totalChannels(), m_frames};
	}

	//! @returns the buffers for all channel groups
	auto allBuffers() -> PlanarBufferView<float>
	{
		return {m_accessBuffer.data(), totalChannels(), m_frames};
	}

	//! @returns the buffers of the given channel group
	auto groupBuffers(group_cnt_t index) const -> PlanarBufferView<const float>
	{
		assert(index < groupCount());
		const ChannelGroup& g = m_groups[index];
		return {g.buffers(), g.channels(), m_frames};
	}

	//! @returns the buffers of the given channel group
	auto groupBuffers(group_cnt_t index) -> PlanarBufferView<float>
	{
		assert(index < groupCount());
		ChannelGroup& g = m_groups[index];
		return {g.buffers(), g.channels(), m_frames};
	}

	//! @returns the buffer for the given channel
	auto buffer(ch_cnt_t channel) const -> std::span<const float>
	{
		return {m_accessBuffer[channel], m_frames};
	}

	//! @returns the buffer for the given channel
	auto buffer(ch_cnt_t channel) -> std::span<float>
	{
		return {m_accessBuffer[channel], m_frames};
	}

	//! @returns the total channel count (never exceeds MaxChannelsPerAudioBuffer)
	auto totalChannels() const -> ch_cnt_t { return static_cast<ch_cnt_t>(m_accessBuffer.size()); }

	//! @returns the frame count for each channel buffer
	auto frames() const -> f_cnt_t { return m_frames; }

	//! @returns scratch buffer for conversions between interleaved and planar TODO: Remove once using planar only
	auto interleavedBuffer() const -> InterleavedBufferView<const float, 2>
	{
		assert(hasInterleavedBuffer());
		return {m_interleavedBuffer.data(), m_frames};
	}

	//! @returns scratch buffer for conversions between interleaved and planar TODO: Remove once using planar only
	auto interleavedBuffer() -> InterleavedBufferView<float, 2>
	{
		assert(hasInterleavedBuffer());
		return {m_interleavedBuffer.data(), m_frames};
	}

	/**
	 * @brief Adds a new channel group at the end of the list.
	 *
	 * If the memory resource is `SharedMemoryResource`, all buffers (source, channels,
	 * and interleaved) will be reallocated. The number of bytes allocated will be
	 * `allocationSize(frames(), totalChannels() + channels, hasInterleavedBuffer())`.
	 *
	 * @param channels how many channels the new group should have
	 * @returns the newly created group, or nullptr upon failure
	 */
	auto addGroup(ch_cnt_t channels) -> ChannelGroup*;

	/**
	 * @brief Changes the channel grouping without changing the channel count.
	 *        Does not reallocate any buffers.
	 *
	 * @param groups the new group count
	 * @param groupVisitor called for each new group, passed the index and group reference, and is
	 *                     expected to return the channel count for that group. The visitor may
	 *                     also set the group's metadata.
	 */
	template<class F>
	void setGroups(group_cnt_t groups, F&& groupVisitor)
	{
		static_assert(std::is_invocable_r_v<ch_cnt_t, F, group_cnt_t, ChannelGroup&>,
			"groupVisitor is passed the group index + group reference and must return the group's channel count");

		m_groups.clear();
		ch_cnt_t ch = 0;
		for (group_cnt_t idx = 0; idx < groups; ++idx)
		{
			auto& group = m_groups.emplace_back();

			const auto channels = groupVisitor(idx, group);
			if (channels == 0) { throw std::runtime_error{"group cannot have zero channels"}; }

			group.setBuffers(&m_accessBuffer[ch]);
			group.setChannels(channels);

			ch += channels;
			if (ch > this->totalChannels())
			{
				throw std::runtime_error{"sum of group channel counts exceeds total channels"};
			}
		}
	}

	/**
	 * Channels which are known to be quiet, AKA the silence status.
	 * 1 = channel is known to be silent
	 * 0 = channel is assumed to be non-silent (or, when silence tracking
	 *     is enabled, *known* to be non-silent)
	 *
	 * NOTE: If any channel buffers are used and their data modified outside of this class,
	 *       their silence flags will be invalidated until `updateSilenceFlags()` is called.
	 *       Therefore, calling code must be careful to always keep the silence flags up-to-date.
	 */
	auto silenceFlags() const -> const ChannelFlags& { return m_silenceFlags; }

	//! Forcibly pessimizes silence tracking for a specific channel
	void assumeNonSilent(ch_cnt_t channel) { m_silenceFlags[channel] = false; }

	/**
	 * When silence tracking is enabled, channels will be checked for silence whenever their data may
	 * have changed, so it'll always be known whether they are silent or non-silent. There is a performance cost
	 * to this, but it is likely worth it since this information allows many effects to be put to sleep
	 * when their inputs are silent ("auto-quit"). When a channel is known to be silent, it also
	 * enables optimizations in buffer sanitization, buffer zeroing, and finding the absolute peak sample value.
	 *
	 * When silence tracking is disabled, channels are not checked for silence, so a silence flag may be
	 * unset despite the channel being silent. Non-silence must be assumed whenever the silence status is not
	 * known, so the optimizations which silent buffers allow will not be possible as often.
	 */
	void enableSilenceTracking(bool enabled);
	auto silenceTrackingEnabled() const -> bool { return m_silenceTrackingEnabled; }

	//! Mixes the silence flags of the other `AudioBuffer` with this `AudioBuffer`
	void mixSilenceFlags(const AudioBuffer& other);

	/**
	 * Checks whether any of the selected channels are non-silent (has a signal).
	 *
	 * If silence tracking is disabled, all channels that aren't marked
	 * as silent are assumed to be non-silent.
	 *
	 * A processor could check for a signal present at any of its inputs by
	 * calling this method selecting all of the track channels that are routed
	 * to at least one of its inputs.
	 *
	 * @param channels channels to check for a signal; 1 = selected, 0 = ignore
	 */
	auto hasSignal(const ChannelFlags& channels) const -> bool;

	//! Checks whether any channel is non-silent (has a signal). @see hasSignal
	auto hasAnySignal() const -> bool;

	/**
	 * @brief Sanitizes specified channels of any Inf/NaN values if "nanhandler" setting is enabled
	 *
	 * @param channels channels to sanitize; 1 = selected, 0 = skip
	 * @param upperBound any channel indexes at or above this are skipped
	 */
	void sanitize(const ChannelFlags& channels, ch_cnt_t upperBound = MaxChannelsPerAudioBuffer);

	//! Sanitizes all channels. @see sanitize
	void sanitizeAll();

	/**
	 * @brief Updates the silence status of the given channels, up to the upperBound index.
	 *
	 * @param channels channels to update; 1 = selected, 0 = skip
	 * @param upperBound any channel indexes at or above this are skipped
	 * @returns true if all selected channels were silent
	 */
	auto updateSilenceFlags(const ChannelFlags& channels, ch_cnt_t upperBound = MaxChannelsPerAudioBuffer) -> bool;

	//! Updates the silence status of all channels. @see updateSilenceFlags
	auto updateAllSilenceFlags() -> bool;

	/**
	 * @brief Silences (zeroes) the given channels
	 *
	 * @param channels channels to silence; 1 = selected, 0 = skip
	 * @param upperBound any channel indexes at or above this are skipped
	 */
	void silenceChannels(const ChannelFlags& channels, ch_cnt_t upperBound = MaxChannelsPerAudioBuffer);

	//! Silences (zeroes) all channels. @see silenceChannels
	void silenceAllChannels();

	//! @returns absolute peak sample value for the given channel
	auto absPeakValue(ch_cnt_t channel) const -> float;

private:
	/**
	 * Large buffer that all channel buffers are sourced from.
	 *
	 * [channel index]
	 */
	std::pmr::vector<float> m_sourceBuffer;

	/**
	 * Provides access to individual channel buffers within the source buffer.
	 *
	 * [channel index][frame index]
	 */
	std::pmr::vector<float*> m_accessBuffer;

	/**
	 * Interleaved scratch buffer for conversions between interleaved and planar.
	 *
	 * TODO: Remove once using planar only
	 */
	std::pmr::vector<float> m_interleavedBuffer;

	//! Divides channels into arbitrary groups
	ArrayVector<ChannelGroup, MaxGroupsPerAudioBuffer> m_groups;

	//! Frame count for every channel buffer
	f_cnt_t m_frames = 0;

	/**
	 * Stores which channels are known to be quiet, AKA the silence status.
	 *
	 * This must always be kept in sync with the buffer data when enabled - at minimum
	 * avoiding any false positives where a channel is marked as "silent" when it isn't.
	 * Any channel bits at or above `totalChannels()` must always be marked silent.
	 *
	 * 1 = channel is known to be silent
	 * 0 = channel is assumed to be non-silent (or, when silence tracking
	 *     is enabled, *known* to be non-silent)
	 */
	ChannelFlags m_silenceFlags;

	bool m_silenceTrackingEnabled = false;
};

} // namespace lmms

#endif // LMMS_AUDIO_BUFFER_H