astc: Enable parallel CPU astc decoding

Given the issues with GPU accelerated ASTC decoding with NVIDIA's latest drivers, parallelize astc decoding on the CPU.
Uses half the available threads in the system for astc decoding.

externals/CMakeLists.txt

@@ -16,7 +16,6 @@ endif()
 
 # Dynarmic
 if (ARCHITECTURE_x86_64)
-    set(DYNARMIC_TESTS OFF)
     set(DYNARMIC_NO_BUNDLED_FMT ON)
     set(DYNARMIC_IGNORE_ASSERTS ON CACHE BOOL "" FORCE)
     add_subdirectory(dynarmic)

src/audio_core/CMakeLists.txt

@@ -194,6 +194,7 @@ add_library(audio_core STATIC
     sink/sink.h
     sink/sink_details.cpp
     sink/sink_details.h
+    sink/sink_stream.cpp
     sink/sink_stream.h
 )
 

src/audio_core/audio_core.cpp

@@ -57,12 +57,12 @@ void AudioCore::PauseSinks(const bool pausing) const {
     }
 }
 
-u32 AudioCore::GetStreamQueue() const {
-    return estimated_queue.load();
+void AudioCore::SetNVDECActive(bool active) {
+    nvdec_active = active;
 }
 
-void AudioCore::SetStreamQueue(u32 size) {
-    estimated_queue.store(size);
+bool AudioCore::IsNVDECActive() const {
+    return nvdec_active;
 }
 
 } // namespace AudioCore

src/audio_core/audio_core.h

@@ -66,18 +66,16 @@ public:
     void PauseSinks(bool pausing) const;
 
     /**
-     * Get the size of the current stream queue.
+     * Toggle NVDEC state, used to avoid stall in playback.
      *
-     * @return Current stream queue size.
+     * @param active - Set true if nvdec is active, otherwise false.
      */
-    u32 GetStreamQueue() const;
+    void SetNVDECActive(bool active);
 
     /**
-     * Get the size of the current stream queue.
-     *
-     * @param size - New stream size.
+     * Get NVDEC state.
      */
-    void SetStreamQueue(u32 size);
+    bool IsNVDECActive() const;
 
 private:
     /**
@@ -93,8 +91,8 @@ private:
     std::unique_ptr<Sink::Sink> input_sink;
     /// The ADSP in the sysmodule
     std::unique_ptr<AudioRenderer::ADSP::ADSP> adsp;
-    /// Current size of the stream queue
-    std::atomic<u32> estimated_queue{0};
+    /// Is NVDec currently active?
+    bool nvdec_active{false};
 };
 
 } // namespace AudioCore

src/audio_core/device/audio_buffer.h

@@ -8,6 +8,10 @@
 namespace AudioCore {
 
 struct AudioBuffer {
+    /// Timestamp this buffer started playing.
+    u64 start_timestamp;
+    /// Timestamp this buffer should finish playing.
+    u64 end_timestamp;
     /// Timestamp this buffer completed playing.
     s64 played_timestamp;
     /// Game memory address for these samples.

src/audio_core/device/audio_buffers.h

@@ -58,6 +58,7 @@ public:
             if (index < 0) {
                 index += N;
             }
+
             out_buffers.push_back(buffers[index]);
             registered_count++;
             registered_index = (registered_index + 1) % append_limit;
@@ -100,7 +101,7 @@ public:
             }
 
             // Check with the backend if this buffer can be released yet.
-            if (!session.IsBufferConsumed(buffers[index].tag)) {
+            if (!session.IsBufferConsumed(buffers[index])) {
                 break;
             }
 
@@ -280,6 +281,16 @@ public:
         return true;
     }
 
+    u64 GetNextTimestamp() const {
+        // Iterate backwards through the buffer queue, and take the most recent buffer's end
+        std::scoped_lock l{lock};
+        auto index{appended_index - 1};
+        if (index < 0) {
+            index += append_limit;
+        }
+        return buffers[index].end_timestamp;
+    }
+
 private:
     /// Buffer lock
     mutable std::recursive_mutex lock{};

src/audio_core/device/device_session.cpp

@@ -7,11 +7,20 @@
 #include "audio_core/device/device_session.h"
 #include "audio_core/sink/sink_stream.h"
 #include "core/core.h"
+#include "core/core_timing.h"
 #include "core/memory.h"
 
 namespace AudioCore {
 
-DeviceSession::DeviceSession(Core::System& system_) : system{system_} {}
+using namespace std::literals;
+constexpr auto INCREMENT_TIME{5ms};
+
+DeviceSession::DeviceSession(Core::System& system_)
+    : system{system_}, thread_event{Core::Timing::CreateEvent(
+                           "AudioOutSampleTick",
+                           [this](std::uintptr_t, s64 time, std::chrono::nanoseconds) {
+                               return ThreadFunc();
+                           })} {}
 
 DeviceSession::~DeviceSession() {
     Finalize();
@@ -50,20 +59,21 @@ void DeviceSession::Finalize() {
 }
 
 void DeviceSession::Start() {
-    stream->SetPlayedSampleCount(played_sample_count);
-    stream->Start();
+    if (stream) {
+        stream->Start();
+        system.CoreTiming().ScheduleLoopingEvent(std::chrono::nanoseconds::zero(), INCREMENT_TIME,
+                                                 thread_event);
+    }
 }
 
 void DeviceSession::Stop() {
     if (stream) {
-        played_sample_count = stream->GetPlayedSampleCount();
         stream->Stop();
+        system.CoreTiming().UnscheduleEvent(thread_event, {});
     }
 }
 
 void DeviceSession::AppendBuffers(std::span<AudioBuffer> buffers) const {
-    auto& memory{system.Memory()};
-
     for (size_t i = 0; i < buffers.size(); i++) {
         Sink::SinkBuffer new_buffer{
             .frames = buffers[i].size / (channel_count * sizeof(s16)),
@@ -77,7 +87,7 @@ void DeviceSession::AppendBuffers(std::span<AudioBuffer> buffers) const {
             stream->AppendBuffer(new_buffer, samples);
         } else {
             std::vector<s16> samples(buffers[i].size / sizeof(s16));
-            memory.ReadBlockUnsafe(buffers[i].samples, samples.data(), buffers[i].size);
+            system.Memory().ReadBlockUnsafe(buffers[i].samples, samples.data(), buffers[i].size);
             stream->AppendBuffer(new_buffer, samples);
         }
     }
@@ -85,17 +95,13 @@ void DeviceSession::AppendBuffers(std::span<AudioBuffer> buffers) const {
 
 void DeviceSession::ReleaseBuffer(AudioBuffer& buffer) const {
     if (type == Sink::StreamType::In) {
-        auto& memory{system.Memory()};
         auto samples{stream->ReleaseBuffer(buffer.size / sizeof(s16))};
-        memory.WriteBlockUnsafe(buffer.samples, samples.data(), buffer.size);
+        system.Memory().WriteBlockUnsafe(buffer.samples, samples.data(), buffer.size);
     }
 }
 
-bool DeviceSession::IsBufferConsumed(u64 tag) const {
-    if (stream) {
-        return stream->IsBufferConsumed(tag);
-    }
-    return true;
+bool DeviceSession::IsBufferConsumed(AudioBuffer& buffer) const {
+    return played_sample_count >= buffer.end_timestamp;
 }
 
 void DeviceSession::SetVolume(f32 volume) const {
@@ -105,10 +111,22 @@ void DeviceSession::SetVolume(f32 volume) const {
 }
 
 u64 DeviceSession::GetPlayedSampleCount() const {
-    if (stream) {
-        return stream->GetPlayedSampleCount();
+    return played_sample_count;
+}
+
+std::optional<std::chrono::nanoseconds> DeviceSession::ThreadFunc() {
+    // Add 5ms of samples at a 48K sample rate.
+    played_sample_count += 48'000 * INCREMENT_TIME / 1s;
+    if (type == Sink::StreamType::Out) {
+        system.AudioCore().GetAudioManager().SetEvent(Event::Type::AudioOutManager, true);
+    } else {
+        system.AudioCore().GetAudioManager().SetEvent(Event::Type::AudioInManager, true);
     }
-    return 0;
+    return std::nullopt;
+}
+
+void DeviceSession::SetRingSize(u32 ring_size) {
+    stream->SetRingSize(ring_size);
 }
 
 } // namespace AudioCore

src/audio_core/device/device_session.h

@@ -3,6 +3,9 @@
 
 #pragma once
 
+#include <chrono>
+#include <memory>
+#include <optional>
 #include <span>
 
 #include "audio_core/common/common.h"
@@ -11,9 +14,13 @@
 
 namespace Core {
 class System;
-}
+namespace Timing {
+struct EventType;
+} // namespace Timing
+} // namespace Core
 
 namespace AudioCore {
+
 namespace Sink {
 class SinkStream;
 struct SinkBuffer;
@@ -70,7 +77,7 @@ public:
      * @param tag - Unqiue tag of the buffer to check.
      * @return true if the buffer has been consumed, otherwise false.
      */
-    bool IsBufferConsumed(u64 tag) const;
+    bool IsBufferConsumed(AudioBuffer& buffer) const;
 
     /**
      * Start this device session, starting the backend stream.
@@ -96,6 +103,16 @@ public:
      */
     u64 GetPlayedSampleCount() const;
 
+    /*
+     * CoreTiming callback to increment played_sample_count over time.
+     */
+    std::optional<std::chrono::nanoseconds> ThreadFunc();
+
+    /*
+     * Set the size of the ring buffer.
+     */
+    void SetRingSize(u32 ring_size);
+
 private:
     /// System
     Core::System& system;
@@ -118,9 +135,13 @@ private:
     /// Applet resource user id of this device session
     u64 applet_resource_user_id{};
     /// Total number of samples played by this device session
-    u64 played_sample_count{};
+    std::atomic<u64> played_sample_count{};
+    /// Event increasing the played sample count every 5ms
+    std::shared_ptr<Core::Timing::EventType> thread_event;
     /// Is this session initialised?
     bool initialized{};
+    /// Buffer queue
+    std::vector<AudioBuffer> buffer_queue{};
 };
 
 } // namespace AudioCore

src/audio_core/in/audio_in_system.cpp

@@ -93,6 +93,7 @@ Result System::Start() {
     std::vector<AudioBuffer> buffers_to_flush{};
     buffers.RegisterBuffers(buffers_to_flush);
     session->AppendBuffers(buffers_to_flush);
+    session->SetRingSize(static_cast<u32>(buffers_to_flush.size()));
 
     return ResultSuccess;
 }
@@ -112,8 +113,13 @@ bool System::AppendBuffer(const AudioInBuffer& buffer, const u64 tag) {
         return false;
     }
 
-    AudioBuffer new_buffer{
-        .played_timestamp = 0, .samples = buffer.samples, .tag = tag, .size = buffer.size};
+    const auto timestamp{buffers.GetNextTimestamp()};
+    AudioBuffer new_buffer{.start_timestamp = timestamp,
+                           .end_timestamp = timestamp + buffer.size / (channel_count * sizeof(s16)),
+                           .played_timestamp = 0,
+                           .samples = buffer.samples,
+                           .tag = tag,
+                           .size = buffer.size};
 
     buffers.AppendBuffer(new_buffer);
     RegisterBuffers();

src/audio_core/out/audio_out_system.cpp

@@ -92,6 +92,7 @@ Result System::Start() {
     std::vector<AudioBuffer> buffers_to_flush{};
     buffers.RegisterBuffers(buffers_to_flush);
     session->AppendBuffers(buffers_to_flush);
+    session->SetRingSize(static_cast<u32>(buffers_to_flush.size()));
 
     return ResultSuccess;
 }
@@ -111,8 +112,13 @@ bool System::AppendBuffer(const AudioOutBuffer& buffer, u64 tag) {
         return false;
     }
 
-    AudioBuffer new_buffer{
-        .played_timestamp = 0, .samples = buffer.samples, .tag = tag, .size = buffer.size};
+    const auto timestamp{buffers.GetNextTimestamp()};
+    AudioBuffer new_buffer{.start_timestamp = timestamp,
+                           .end_timestamp = timestamp + buffer.size / (channel_count * sizeof(s16)),
+                           .played_timestamp = 0,
+                           .samples = buffer.samples,
+                           .tag = tag,
+                           .size = buffer.size};
 
     buffers.AppendBuffer(new_buffer);
     RegisterBuffers();

src/audio_core/renderer/adsp/audio_renderer.cpp

@@ -106,9 +106,6 @@ void AudioRenderer::Start(AudioRenderer_Mailbox* mailbox_) {
 
     mailbox = mailbox_;
     thread = std::thread(&AudioRenderer::ThreadFunc, this);
-    for (auto& stream : streams) {
-        stream->Start();
-    }
     running = true;
 }
 
@@ -130,6 +127,7 @@ void AudioRenderer::CreateSinkStreams() {
         std::string name{fmt::format("ADSP_RenderStream-{}", i)};
         streams[i] =
             sink.AcquireSinkStream(system, channels, name, ::AudioCore::Sink::StreamType::Render);
+        streams[i]->SetRingSize(4);
     }
 }
 
@@ -198,11 +196,6 @@ void AudioRenderer::ThreadFunc() {
                             command_list_processor.Process(index) - start_time;
                     }
 
-                    if (index == 0) {
-                        auto stream{command_list_processor.GetOutputSinkStream()};
-                        system.AudioCore().SetStreamQueue(stream->GetQueueSize());
-                    }
-
                     const auto end_time{system.CoreTiming().GetClockTicks()};
 
                     command_buffer.remaining_command_count =

src/audio_core/renderer/behavior/behavior_info.cpp

@@ -43,13 +43,15 @@ void BehaviorInfo::AppendError(ErrorInfo& error) {
 }
 
 void BehaviorInfo::CopyErrorInfo(std::span<ErrorInfo> out_errors, u32& out_count) {
-    auto error_count_{std::min(error_count, MaxErrors)};
-    std::memset(out_errors.data(), 0, MaxErrors * sizeof(ErrorInfo));
+    out_count = std::min(error_count, MaxErrors);
 
-    for (size_t i = 0; i < error_count_; i++) {
-        out_errors[i] = errors[i];
+    for (size_t i = 0; i < MaxErrors; i++) {
+        if (i < out_count) {
+            out_errors[i] = errors[i];
+        } else {
+            out_errors[i] = {};
+        }
     }
-    out_count = error_count_;
 }
 
 void BehaviorInfo::UpdateFlags(const Flags flags_) {

src/audio_core/renderer/command/sink/device.cpp

@@ -46,6 +46,10 @@ void DeviceSinkCommand::Process(const ADSP::CommandListProcessor& processor) {
 
     out_buffer.tag = reinterpret_cast<u64>(samples.data());
     stream->AppendBuffer(out_buffer, samples);
+
+    if (stream->IsPaused()) {
+        stream->Start();
+    }
 }
 
 bool DeviceSinkCommand::Verify(const ADSP::CommandListProcessor& processor) {

src/audio_core/renderer/system_manager.cpp

@@ -15,8 +15,7 @@ MICROPROFILE_DEFINE(Audio_RenderSystemManager, "Audio", "Render System Manager",
                     MP_RGB(60, 19, 97));
 
 namespace AudioCore::AudioRenderer {
-constexpr std::chrono::nanoseconds BaseRenderTime{5'000'000UL};
-constexpr std::chrono::nanoseconds RenderTimeOffset{400'000UL};
+constexpr std::chrono::nanoseconds RENDER_TIME{5'000'000UL};
 
 SystemManager::SystemManager(Core::System& core_)
     : core{core_}, adsp{core.AudioCore().GetADSP()}, mailbox{adsp.GetRenderMailbox()},
@@ -36,8 +35,8 @@ bool SystemManager::InitializeUnsafe() {
         if (adsp.Start()) {
             active = true;
             thread = std::jthread([this](std::stop_token stop_token) { ThreadFunc(); });
-            core.CoreTiming().ScheduleLoopingEvent(std::chrono::nanoseconds(0),
-                                                   BaseRenderTime - RenderTimeOffset, thread_event);
+            core.CoreTiming().ScheduleLoopingEvent(std::chrono::nanoseconds(0), RENDER_TIME,
+                                                   thread_event);
         }
     }
 
@@ -121,35 +120,9 @@ void SystemManager::ThreadFunc() {
 }
 
 std::optional<std::chrono::nanoseconds> SystemManager::ThreadFunc2(s64 time) {
-    std::optional<std::chrono::nanoseconds> new_schedule_time{std::nullopt};
-    const auto queue_size{core.AudioCore().GetStreamQueue()};
-    switch (state) {
-    case StreamState::Filling:
-        if (queue_size >= 5) {
-            new_schedule_time = BaseRenderTime;
-            state = StreamState::Steady;
-        }
-        break;
-    case StreamState::Steady:
-        if (queue_size <= 2) {
-            new_schedule_time = BaseRenderTime - RenderTimeOffset;
-            state = StreamState::Filling;
-        } else if (queue_size > 5) {
-            new_schedule_time = BaseRenderTime + RenderTimeOffset;
-            state = StreamState::Draining;
-        }
-        break;
-    case StreamState::Draining:
-        if (queue_size <= 5) {
-            new_schedule_time = BaseRenderTime;
-            state = StreamState::Steady;
-        }
-        break;
-    }
-
     update.store(true);
     update.notify_all();
-    return new_schedule_time;
+    return std::nullopt;
 }
 
 void SystemManager::PauseCallback(bool paused) {

src/audio_core/sink/cubeb_sink.cpp

@@ -1,21 +1,13 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <algorithm>
-#include <atomic>
 #include <span>
+#include <vector>
 
-#include "audio_core/audio_core.h"
-#include "audio_core/audio_event.h"
-#include "audio_core/audio_manager.h"
+#include "audio_core/common/common.h"
 #include "audio_core/sink/cubeb_sink.h"
 #include "audio_core/sink/sink_stream.h"
-#include "common/assert.h"
-#include "common/fixed_point.h"
 #include "common/logging/log.h"
-#include "common/reader_writer_queue.h"
-#include "common/ring_buffer.h"
-#include "common/settings.h"
 #include "core/core.h"
 
 #ifdef _WIN32
@@ -42,10 +34,10 @@ public:
      * @param system_          - Core system.
      * @param event            - Event used only for audio renderer, signalled on buffer consume.
      */
-    CubebSinkStream(cubeb* ctx_, const u32 device_channels_, const u32 system_channels_,
+    CubebSinkStream(cubeb* ctx_, u32 device_channels_, u32 system_channels_,
                     cubeb_devid output_device, cubeb_devid input_device, const std::string& name_,
-                    const StreamType type_, Core::System& system_)
-        : ctx{ctx_}, type{type_}, system{system_} {
+                    StreamType type_, Core::System& system_)
+        : SinkStream(system_, type_), ctx{ctx_} {
 #ifdef _WIN32
         CoInitializeEx(nullptr, COINIT_MULTITHREADED);
 #endif
@@ -79,12 +71,10 @@ public:
 
         minimum_latency = std::max(minimum_latency, 256u);
 
-        playing_buffer.consumed = true;
-
-        LOG_DEBUG(Service_Audio,
-                  "Opening cubeb stream {} type {} with: rate {} channels {} (system channels {}) "
-                  "latency {}",
-                  name, type, params.rate, params.channels, system_channels, minimum_latency);
+        LOG_INFO(Service_Audio,
+                 "Opening cubeb stream {} type {} with: rate {} channels {} (system channels {}) "
+                 "latency {}",
+                 name, type, params.rate, params.channels, system_channels, minimum_latency);
 
         auto init_error{0};
         if (type == StreamType::In) {
@@ -111,6 +101,8 @@ public:
     ~CubebSinkStream() override {
         LOG_DEBUG(Service_Audio, "Destructing cubeb stream {}", name);
 
+        Unstall();
+
         if (!ctx) {
             return;
         }
@@ -136,7 +128,7 @@ public:
      * @param resume - Set to true if this is resuming the stream a previously-active stream.
      *                 Default false.
      */
-    void Start(const bool resume = false) override {
+    void Start(bool resume = false) override {
         if (!ctx) {
             return;
         }
@@ -158,6 +150,7 @@ public:
      * Stop the sink stream.
      */
     void Stop() override {
+        Unstall();
         if (!ctx) {
             return;
         }
@@ -170,194 +163,7 @@ public:
         paused = true;
     }
 
-    /**
-     * Append a new buffer and its samples to a waiting queue to play.
-     *
-     * @param buffer  - Audio buffer information to be queued.
-     * @param samples - The s16 samples to be queue for playback.
-     */
-    void AppendBuffer(::AudioCore::Sink::SinkBuffer& buffer, std::vector<s16>& samples) override {
-        if (type == StreamType::In) {
-            queue.enqueue(buffer);
-            queued_buffers++;
-        } else {
-            constexpr s32 min{std::numeric_limits<s16>::min()};
-            constexpr s32 max{std::numeric_limits<s16>::max()};
-
-            auto yuzu_volume{Settings::Volume()};
-            if (yuzu_volume > 1.0f) {
-                yuzu_volume = 0.6f + 20 * std::log10(yuzu_volume);
-            }
-            auto volume{system_volume * device_volume * yuzu_volume};
-
-            if (system_channels == 6 && device_channels == 2) {
-                // We're given 6 channels, but our device only outputs 2, so downmix.
-                constexpr std::array<f32, 4> down_mix_coeff{1.0f, 0.707f, 0.251f, 0.707f};
-
-                for (u32 read_index = 0, write_index = 0; read_index < samples.size();
-                     read_index += system_channels, write_index += device_channels) {
-                    const auto left_sample{
-                        ((Common::FixedPoint<49, 15>(
-                              samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
-                              down_mix_coeff[0] +
-                          samples[read_index + static_cast<u32>(Channels::Center)] *
-                              down_mix_coeff[1] +
-                          samples[read_index + static_cast<u32>(Channels::LFE)] *
-                              down_mix_coeff[2] +
-                          samples[read_index + static_cast<u32>(Channels::BackLeft)] *
-                              down_mix_coeff[3]) *
-                         volume)
-                            .to_int()};
-
-                    const auto right_sample{
-                        ((Common::FixedPoint<49, 15>(
-                              samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
-                              down_mix_coeff[0] +
-                          samples[read_index + static_cast<u32>(Channels::Center)] *
-                              down_mix_coeff[1] +
-                          samples[read_index + static_cast<u32>(Channels::LFE)] *
-                              down_mix_coeff[2] +
-                          samples[read_index + static_cast<u32>(Channels::BackRight)] *
-                              down_mix_coeff[3]) *
-                         volume)
-                            .to_int()};
-
-                    samples[write_index + static_cast<u32>(Channels::FrontLeft)] =
-                        static_cast<s16>(std::clamp(left_sample, min, max));
-                    samples[write_index + static_cast<u32>(Channels::FrontRight)] =
-                        static_cast<s16>(std::clamp(right_sample, min, max));
-                }
-
-                samples.resize(samples.size() / system_channels * device_channels);
-
-            } else if (system_channels == 2 && device_channels == 6) {
-                // We need moar samples! Not all games will provide 6 channel audio.
-                // TODO: Implement some upmixing here. Currently just passthrough, with other
-                // channels left as silence.
-                std::vector<s16> new_samples(samples.size() / system_channels * device_channels, 0);
-
-                for (u32 read_index = 0, write_index = 0; read_index < samples.size();
-                     read_index += system_channels, write_index += device_channels) {
-                    const auto left_sample{static_cast<s16>(std::clamp(
-                        static_cast<s32>(
-                            static_cast<f32>(
-                                samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
-                            volume),
-                        min, max))};
-
-                    new_samples[write_index + static_cast<u32>(Channels::FrontLeft)] = left_sample;
-
-                    const auto right_sample{static_cast<s16>(std::clamp(
-                        static_cast<s32>(
-                            static_cast<f32>(
-                                samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
-                            volume),
-                        min, max))};
-
-                    new_samples[write_index + static_cast<u32>(Channels::FrontRight)] =
-                        right_sample;
-                }
-                samples = std::move(new_samples);
-
-            } else if (volume != 1.0f) {
-                for (u32 i = 0; i < samples.size(); i++) {
-                    samples[i] = static_cast<s16>(std::clamp(
-                        static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
-                }
-            }
-
-            samples_buffer.Push(samples);
-            queue.enqueue(buffer);
-            queued_buffers++;
-        }
-    }
-
-    /**
-     * Release a buffer. Audio In only, will fill a buffer with recorded samples.
-     *
-     * @param num_samples - Maximum number of samples to receive.
-     * @return Vector of recorded samples. May have fewer than num_samples.
-     */
-    std::vector<s16> ReleaseBuffer(const u64 num_samples) override {
-        static constexpr s32 min = std::numeric_limits<s16>::min();
-        static constexpr s32 max = std::numeric_limits<s16>::max();
-
-        auto samples{samples_buffer.Pop(num_samples)};
-
-        // TODO: Up-mix to 6 channels if the game expects it.
-        // For audio input this is unlikely to ever be the case though.
-
-        // Incoming mic volume seems to always be very quiet, so multiply by an additional 8 here.
-        // TODO: Play with this and find something that works better.
-        auto volume{system_volume * device_volume * 8};
-        for (u32 i = 0; i < samples.size(); i++) {
-            samples[i] = static_cast<s16>(
-                std::clamp(static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
-        }
-
-        if (samples.size() < num_samples) {
-            samples.resize(num_samples, 0);
-        }
-        return samples;
-    }
-
-    /**
-     * Check if a certain buffer has been consumed (fully played).
-     *
-     * @param tag - Unique tag of a buffer to check for.
-     * @return True if the buffer has been played, otherwise false.
-     */
-    bool IsBufferConsumed(const u64 tag) override {
-        if (released_buffer.tag == 0) {
-            if (!released_buffers.try_dequeue(released_buffer)) {
-                return false;
-            }
-        }
-
-        if (released_buffer.tag == tag) {
-            released_buffer.tag = 0;
-            return true;
-        }
-        return false;
-    }
-
-    /**
-     * Empty out the buffer queue.
-     */
-    void ClearQueue() override {
-        samples_buffer.Pop();
-        while (queue.pop()) {
-        }
-        while (released_buffers.pop()) {
-        }
-        queued_buffers = 0;
-        released_buffer = {};
-        playing_buffer = {};
-        playing_buffer.consumed = true;
-    }
-
 private:
-    /**
-     * Signal events back to the audio system that a buffer was played/can be filled.
-     *
-     * @param buffer - Consumed audio buffer to be released.
-     */
-    void SignalEvent(const ::AudioCore::Sink::SinkBuffer& buffer) {
-        auto& manager{system.AudioCore().GetAudioManager()};
-        switch (type) {
-        case StreamType::Out:
-            released_buffers.enqueue(buffer);
-            manager.SetEvent(Event::Type::AudioOutManager, true);
-            break;
-        case StreamType::In:
-            released_buffers.enqueue(buffer);
-            manager.SetEvent(Event::Type::AudioInManager, true);
-            break;
-        case StreamType::Render:
-            break;
-        }
-    }
-
     /**
      * Main callback from Cubeb. Either expects samples from us (audio render/audio out), or will
      * provide samples to be copied (audio in).
@@ -378,106 +184,15 @@ private:
 
         const std::size_t num_channels = impl->GetDeviceChannels();
         const std::size_t frame_size = num_channels;
-        const std::size_t frame_size_bytes = frame_size * sizeof(s16);
         const std::size_t num_frames{static_cast<size_t>(num_frames_)};
-        size_t frames_written{0};
-        [[maybe_unused]] bool underrun{false};
 
         if (impl->type == StreamType::In) {
-            // INPUT
             std::span<const s16> input_buffer{reinterpret_cast<const s16*>(in_buff),
                                               num_frames * frame_size};
-
-            while (frames_written < num_frames) {
-                auto& playing_buffer{impl->playing_buffer};
-
-                // If the playing buffer has been consumed or has no frames, we need a new one
-                if (playing_buffer.consumed || playing_buffer.frames == 0) {
-                    if (!impl->queue.try_dequeue(impl->playing_buffer)) {
-                        // If no buffer was available we've underrun, just push the samples and
-                        // continue.
-                        underrun = true;
-                        impl->samples_buffer.Push(&input_buffer[frames_written * frame_size],
-                                                  (num_frames - frames_written) * frame_size);
-                        frames_written = num_frames;
-                        continue;
-                    } else {
-                        // Successfully got a new buffer, mark the old one as consumed and signal.
-                        impl->queued_buffers--;
-                        impl->SignalEvent(impl->playing_buffer);
-                    }
-                }
-
-                // Get the minimum frames available between the currently playing buffer, and the
-                // amount we have left to fill
-                size_t frames_available{
-                    std::min(playing_buffer.frames - playing_buffer.frames_played,
-                             num_frames - frames_written)};
-
-                impl->samples_buffer.Push(&input_buffer[frames_written * frame_size],
-                                          frames_available * frame_size);
-
-                frames_written += frames_available;
-                playing_buffer.frames_played += frames_available;
-
-                // If that's all the frames in the current buffer, add its samples and mark it as
-                // consumed
-                if (playing_buffer.frames_played >= playing_buffer.frames) {
-                    impl->AddPlayedSampleCount(playing_buffer.frames_played * num_channels);
-                    impl->playing_buffer.consumed = true;
-                }
-            }
-
-            std::memcpy(&impl->last_frame[0], &input_buffer[(frames_written - 1) * frame_size],
-                        frame_size_bytes);
+            impl->ProcessAudioIn(input_buffer, num_frames);
         } else {
-            // OUTPUT
             std::span<s16> output_buffer{reinterpret_cast<s16*>(out_buff), num_frames * frame_size};
-
-            while (frames_written < num_frames) {
-                auto& playing_buffer{impl->playing_buffer};
-
-                // If the playing buffer has been consumed or has no frames, we need a new one
-                if (playing_buffer.consumed || playing_buffer.frames == 0) {
-                    if (!impl->queue.try_dequeue(impl->playing_buffer)) {
-                        // If no buffer was available we've underrun, fill the remaining buffer with
-                        // the last written frame and continue.
-                        underrun = true;
-                        for (size_t i = frames_written; i < num_frames; i++) {
-                            std::memcpy(&output_buffer[i * frame_size], &impl->last_frame[0],
-                                        frame_size_bytes);
-                        }
-                        frames_written = num_frames;
-                        continue;
-                    } else {
-                        // Successfully got a new buffer, mark the old one as consumed and signal.
-                        impl->queued_buffers--;
-                        impl->SignalEvent(impl->playing_buffer);
-                    }
-                }
-
-                // Get the minimum frames available between the currently playing buffer, and the
-                // amount we have left to fill
-                size_t frames_available{
-                    std::min(playing_buffer.frames - playing_buffer.frames_played,
-                             num_frames - frames_written)};
-
-                impl->samples_buffer.Pop(&output_buffer[frames_written * frame_size],
-                                         frames_available * frame_size);
-
-                frames_written += frames_available;
-                playing_buffer.frames_played += frames_available;
-
-                // If that's all the frames in the current buffer, add its samples and mark it as
-                // consumed
-                if (playing_buffer.frames_played >= playing_buffer.frames) {
-                    impl->AddPlayedSampleCount(playing_buffer.frames_played * num_channels);
-                    impl->playing_buffer.consumed = true;
-                }
-            }
-
-            std::memcpy(&impl->last_frame[0], &output_buffer[(frames_written - 1) * frame_size],
-                        frame_size_bytes);
+            impl->ProcessAudioOutAndRender(output_buffer, num_frames);
         }
 
         return num_frames_;
@@ -490,32 +205,12 @@ private:
      * @param user_data   - Custom data pointer passed along, points to a CubebSinkStream.
      * @param state       - New state of the device.
      */
-    static void StateCallback([[maybe_unused]] cubeb_stream* stream,
-                              [[maybe_unused]] void* user_data,
-                              [[maybe_unused]] cubeb_state state) {}
+    static void StateCallback(cubeb_stream*, void*, cubeb_state) {}
 
     /// Main Cubeb context
     cubeb* ctx{};
     /// Cubeb stream backend
     cubeb_stream* stream_backend{};
-    /// Name of this stream
-    std::string name{};
-    /// Type of this stream
-    StreamType type;
-    /// Core system
-    Core::System& system;
-    /// Ring buffer of the samples waiting to be played or consumed
-    Common::RingBuffer<s16, 0x10000> samples_buffer;
-    /// Audio buffers queued and waiting to play
-    Common::ReaderWriterQueue<::AudioCore::Sink::SinkBuffer> queue;
-    /// The currently-playing audio buffer
-    ::AudioCore::Sink::SinkBuffer playing_buffer{};
-    /// Audio buffers which have been played and are in queue to be released by the audio system
-    Common::ReaderWriterQueue<::AudioCore::Sink::SinkBuffer> released_buffers{};
-    /// Currently released buffer waiting to be taken by the audio system
-    ::AudioCore::Sink::SinkBuffer released_buffer{};
-    /// The last played (or received) frame of audio, used when the callback underruns
-    std::array<s16, MaxChannels> last_frame{};
 };
 
 CubebSink::CubebSink(std::string_view target_device_name) {
@@ -569,15 +264,15 @@ CubebSink::~CubebSink() {
 #endif
 }
 
-SinkStream* CubebSink::AcquireSinkStream(Core::System& system, const u32 system_channels,
-                                         const std::string& name, const StreamType type) {
+SinkStream* CubebSink::AcquireSinkStream(Core::System& system, u32 system_channels,
+                                         const std::string& name, StreamType type) {
     SinkStreamPtr& stream = sink_streams.emplace_back(std::make_unique<CubebSinkStream>(
         ctx, device_channels, system_channels, output_device, input_device, name, type, system));
 
     return stream.get();
 }
 
-void CubebSink::CloseStream(const SinkStream* stream) {
+void CubebSink::CloseStream(SinkStream* stream) {
     for (size_t i = 0; i < sink_streams.size(); i++) {
         if (sink_streams[i].get() == stream) {
             sink_streams[i].reset();
@@ -611,19 +306,19 @@ f32 CubebSink::GetDeviceVolume() const {
     return sink_streams[0]->GetDeviceVolume();
 }
 
-void CubebSink::SetDeviceVolume(const f32 volume) {
+void CubebSink::SetDeviceVolume(f32 volume) {
     for (auto& stream : sink_streams) {
         stream->SetDeviceVolume(volume);
     }
 }
 
-void CubebSink::SetSystemVolume(const f32 volume) {
+void CubebSink::SetSystemVolume(f32 volume) {
     for (auto& stream : sink_streams) {
         stream->SetSystemVolume(volume);
     }
 }
 
-std::vector<std::string> ListCubebSinkDevices(const bool capture) {
+std::vector<std::string> ListCubebSinkDevices(bool capture) {
     std::vector<std::string> device_list;
     cubeb* ctx;
 

src/audio_core/sink/cubeb_sink.h

@@ -46,7 +46,7 @@ public:
      *
      * @param stream - The stream to close.
      */
-    void CloseStream(const SinkStream* stream) override;
+    void CloseStream(SinkStream* stream) override;
 
     /**
      * Close all streams.

src/audio_core/sink/null_sink.h

@@ -3,10 +3,29 @@
 
 #pragma once
 
+#include <string>
+#include <string_view>
+#include <vector>
+
 #include "audio_core/sink/sink.h"
 #include "audio_core/sink/sink_stream.h"
 
+namespace Core {
+class System;
+} // namespace Core
+
 namespace AudioCore::Sink {
+class NullSinkStreamImpl final : public SinkStream {
+public:
+    explicit NullSinkStreamImpl(Core::System& system_, StreamType type_)
+        : SinkStream{system_, type_} {}
+    ~NullSinkStreamImpl() override {}
+    void AppendBuffer(SinkBuffer&, std::vector<s16>&) override {}
+    std::vector<s16> ReleaseBuffer(u64) override {
+        return {};
+    }
+};
+
 /**
  * A no-op sink for when no audio out is wanted.
  */
@@ -15,14 +34,15 @@ public:
     explicit NullSink(std::string_view) {}
     ~NullSink() override = default;
 
-    SinkStream* AcquireSinkStream([[maybe_unused]] Core::System& system,
-                                  [[maybe_unused]] u32 system_channels,
-                                  [[maybe_unused]] const std::string& name,
-                                  [[maybe_unused]] StreamType type) override {
-        return &null_sink_stream;
+    SinkStream* AcquireSinkStream(Core::System& system, u32, const std::string&,
+                                  StreamType type) override {
+        if (null_sink == nullptr) {
+            null_sink = std::make_unique<NullSinkStreamImpl>(system, type);
+        }
+        return null_sink.get();
     }
 
-    void CloseStream([[maybe_unused]] const SinkStream* stream) override {}
+    void CloseStream(SinkStream*) override {}
     void CloseStreams() override {}
     void PauseStreams() override {}
     void UnpauseStreams() override {}
@@ -33,20 +53,7 @@ public:
     void SetSystemVolume(f32 volume) override {}
 
 private:
-    struct NullSinkStreamImpl final : SinkStream {
-        void Finalize() override {}
-        void Start(bool resume = false) override {}
-        void Stop() override {}
-        void AppendBuffer([[maybe_unused]] ::AudioCore::Sink::SinkBuffer& buffer,
-                          [[maybe_unused]] std::vector<s16>& samples) override {}
-        std::vector<s16> ReleaseBuffer([[maybe_unused]] u64 num_samples) override {
-            return {};
-        }
-        bool IsBufferConsumed([[maybe_unused]] const u64 tag) {
-            return true;
-        }
-        void ClearQueue() override {}
-    } null_sink_stream;
+    SinkStreamPtr null_sink{};
 };
 
 } // namespace AudioCore::Sink

src/audio_core/sink/sdl2_sink.cpp

@@ -1,20 +1,13 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <algorithm>
-#include <atomic>
+#include <span>
+#include <vector>
 
-#include "audio_core/audio_core.h"
-#include "audio_core/audio_event.h"
-#include "audio_core/audio_manager.h"
+#include "audio_core/common/common.h"
 #include "audio_core/sink/sdl2_sink.h"
 #include "audio_core/sink/sink_stream.h"
-#include "common/assert.h"
-#include "common/fixed_point.h"
 #include "common/logging/log.h"
-#include "common/reader_writer_queue.h"
-#include "common/ring_buffer.h"
-#include "common/settings.h"
 #include "core/core.h"
 
 // Ignore -Wimplicit-fallthrough due to https://github.com/libsdl-org/SDL/issues/4307
@@ -44,10 +37,9 @@ public:
      * @param system_          - Core system.
      * @param event            - Event used only for audio renderer, signalled on buffer consume.
      */
-    SDLSinkStream(u32 device_channels_, const u32 system_channels_,
-                  const std::string& output_device, const std::string& input_device,
-                  const StreamType type_, Core::System& system_)
-        : type{type_}, system{system_} {
+    SDLSinkStream(u32 device_channels_, u32 system_channels_, const std::string& output_device,
+                  const std::string& input_device, StreamType type_, Core::System& system_)
+        : SinkStream{system_, type_} {
         system_channels = system_channels_;
         device_channels = device_channels_;
 
@@ -63,8 +55,6 @@ public:
         spec.callback = &SDLSinkStream::DataCallback;
         spec.userdata = this;
 
-        playing_buffer.consumed = true;
-
         std::string device_name{output_device};
         bool capture{false};
         if (type == StreamType::In) {
@@ -84,31 +74,30 @@ public:
             return;
         }
 
-        LOG_DEBUG(Service_Audio,
-                  "Opening sdl stream {} with: rate {} channels {} (system channels {}) "
-                  " samples {}",
-                  device, obtained.freq, obtained.channels, system_channels, obtained.samples);
+        LOG_INFO(Service_Audio,
+                 "Opening SDL stream {} with: rate {} channels {} (system channels {}) "
+                 " samples {}",
+                 device, obtained.freq, obtained.channels, system_channels, obtained.samples);
     }
 
     /**
      * Destroy the sink stream.
      */
     ~SDLSinkStream() override {
-        if (device == 0) {
-            return;
-        }
-
-        SDL_CloseAudioDevice(device);
+        LOG_DEBUG(Service_Audio, "Destructing SDL stream {}", name);
+        Finalize();
     }
 
     /**
      * Finalize the sink stream.
      */
     void Finalize() override {
+        Unstall();
         if (device == 0) {
             return;
         }
 
+        Stop();
         SDL_CloseAudioDevice(device);
     }
 
@@ -118,7 +107,7 @@ public:
      * @param resume - Set to true if this is resuming the stream a previously-active stream.
      *                 Default false.
      */
-    void Start(const bool resume = false) override {
+    void Start(bool resume = false) override {
         if (device == 0) {
             return;
         }
@@ -135,7 +124,8 @@ public:
     /**
      * Stop the sink stream.
      */
-    void Stop() {
+    void Stop() override {
+        Unstall();
         if (device == 0) {
             return;
         }
@@ -143,191 +133,7 @@ public:
         paused = true;
     }
 
-    /**
-     * Append a new buffer and its samples to a waiting queue to play.
-     *
-     * @param buffer  - Audio buffer information to be queued.
-     * @param samples - The s16 samples to be queue for playback.
-     */
-    void AppendBuffer(::AudioCore::Sink::SinkBuffer& buffer, std::vector<s16>& samples) override {
-        if (type == StreamType::In) {
-            queue.enqueue(buffer);
-            queued_buffers++;
-        } else {
-            constexpr s32 min = std::numeric_limits<s16>::min();
-            constexpr s32 max = std::numeric_limits<s16>::max();
-
-            auto yuzu_volume{Settings::Volume()};
-            auto volume{system_volume * device_volume * yuzu_volume};
-
-            if (system_channels == 6 && device_channels == 2) {
-                // We're given 6 channels, but our device only outputs 2, so downmix.
-                constexpr std::array<f32, 4> down_mix_coeff{1.0f, 0.707f, 0.251f, 0.707f};
-
-                for (u32 read_index = 0, write_index = 0; read_index < samples.size();
-                     read_index += system_channels, write_index += device_channels) {
-                    const auto left_sample{
-                        ((Common::FixedPoint<49, 15>(
-                              samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
-                              down_mix_coeff[0] +
-                          samples[read_index + static_cast<u32>(Channels::Center)] *
-                              down_mix_coeff[1] +
-                          samples[read_index + static_cast<u32>(Channels::LFE)] *
-                              down_mix_coeff[2] +
-                          samples[read_index + static_cast<u32>(Channels::BackLeft)] *
-                              down_mix_coeff[3]) *
-                         volume)
-                            .to_int()};
-
-                    const auto right_sample{
-                        ((Common::FixedPoint<49, 15>(
-                              samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
-                              down_mix_coeff[0] +
-                          samples[read_index + static_cast<u32>(Channels::Center)] *
-                              down_mix_coeff[1] +
-                          samples[read_index + static_cast<u32>(Channels::LFE)] *
-                              down_mix_coeff[2] +
-                          samples[read_index + static_cast<u32>(Channels::BackRight)] *
-                              down_mix_coeff[3]) *
-                         volume)
-                            .to_int()};
-
-                    samples[write_index + static_cast<u32>(Channels::FrontLeft)] =
-                        static_cast<s16>(std::clamp(left_sample, min, max));
-                    samples[write_index + static_cast<u32>(Channels::FrontRight)] =
-                        static_cast<s16>(std::clamp(right_sample, min, max));
-                }
-
-                samples.resize(samples.size() / system_channels * device_channels);
-
-            } else if (system_channels == 2 && device_channels == 6) {
-                // We need moar samples! Not all games will provide 6 channel audio.
-                // TODO: Implement some upmixing here. Currently just passthrough, with other
-                // channels left as silence.
-                std::vector<s16> new_samples(samples.size() / system_channels * device_channels, 0);
-
-                for (u32 read_index = 0, write_index = 0; read_index < samples.size();
-                     read_index += system_channels, write_index += device_channels) {
-                    const auto left_sample{static_cast<s16>(std::clamp(
-                        static_cast<s32>(
-                            static_cast<f32>(
-                                samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
-                            volume),
-                        min, max))};
-
-                    new_samples[write_index + static_cast<u32>(Channels::FrontLeft)] = left_sample;
-
-                    const auto right_sample{static_cast<s16>(std::clamp(
-                        static_cast<s32>(
-                            static_cast<f32>(
-                                samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
-                            volume),
-                        min, max))};
-
-                    new_samples[write_index + static_cast<u32>(Channels::FrontRight)] =
-                        right_sample;
-                }
-                samples = std::move(new_samples);
-
-            } else if (volume != 1.0f) {
-                for (u32 i = 0; i < samples.size(); i++) {
-                    samples[i] = static_cast<s16>(std::clamp(
-                        static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
-                }
-            }
-
-            samples_buffer.Push(samples);
-            queue.enqueue(buffer);
-            queued_buffers++;
-        }
-    }
-
-    /**
-     * Release a buffer. Audio In only, will fill a buffer with recorded samples.
-     *
-     * @param num_samples - Maximum number of samples to receive.
-     * @return Vector of recorded samples. May have fewer than num_samples.
-     */
-    std::vector<s16> ReleaseBuffer(const u64 num_samples) override {
-        static constexpr s32 min = std::numeric_limits<s16>::min();
-        static constexpr s32 max = std::numeric_limits<s16>::max();
-
-        auto samples{samples_buffer.Pop(num_samples)};
-
-        // TODO: Up-mix to 6 channels if the game expects it.
-        // For audio input this is unlikely to ever be the case though.
-
-        // Incoming mic volume seems to always be very quiet, so multiply by an additional 8 here.
-        // TODO: Play with this and find something that works better.
-        auto volume{system_volume * device_volume * 8};
-        for (u32 i = 0; i < samples.size(); i++) {
-            samples[i] = static_cast<s16>(
-                std::clamp(static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
-        }
-
-        if (samples.size() < num_samples) {
-            samples.resize(num_samples, 0);
-        }
-        return samples;
-    }
-
-    /**
-     * Check if a certain buffer has been consumed (fully played).
-     *
-     * @param tag - Unique tag of a buffer to check for.
-     * @return True if the buffer has been played, otherwise false.
-     */
-    bool IsBufferConsumed(const u64 tag) override {
-        if (released_buffer.tag == 0) {
-            if (!released_buffers.try_dequeue(released_buffer)) {
-                return false;
-            }
-        }
-
-        if (released_buffer.tag == tag) {
-            released_buffer.tag = 0;
-            return true;
-        }
-        return false;
-    }
-
-    /**
-     * Empty out the buffer queue.
-     */
-    void ClearQueue() override {
-        samples_buffer.Pop();
-        while (queue.pop()) {
-        }
-        while (released_buffers.pop()) {
-        }
-        released_buffer = {};
-        playing_buffer = {};
-        playing_buffer.consumed = true;
-        queued_buffers = 0;
-    }
-
 private:
-    /**
-     * Signal events back to the audio system that a buffer was played/can be filled.
-     *
-     * @param buffer - Consumed audio buffer to be released.
-     */
-    void SignalEvent(const ::AudioCore::Sink::SinkBuffer& buffer) {
-        auto& manager{system.AudioCore().GetAudioManager()};
-        switch (type) {
-        case StreamType::Out:
-            released_buffers.enqueue(buffer);
-            manager.SetEvent(Event::Type::AudioOutManager, true);
-            break;
-        case StreamType::In:
-            released_buffers.enqueue(buffer);
-            manager.SetEvent(Event::Type::AudioInManager, true);
-            break;
-        case StreamType::Render:
-            break;
-        }
-    }
-
     /**
      * Main callback from SDL. Either expects samples from us (audio render/audio out), or will
      * provide samples to be copied (audio in).
@@ -345,122 +151,20 @@ private:
 
         const std::size_t num_channels = impl->GetDeviceChannels();
         const std::size_t frame_size = num_channels;
-        const std::size_t frame_size_bytes = frame_size * sizeof(s16);
         const std::size_t num_frames{len / num_channels / sizeof(s16)};
-        size_t frames_written{0};
-        [[maybe_unused]] bool underrun{false};
 
         if (impl->type == StreamType::In) {
-            std::span<s16> input_buffer{reinterpret_cast<s16*>(stream), num_frames * frame_size};
-
-            while (frames_written < num_frames) {
-                auto& playing_buffer{impl->playing_buffer};
-
-                // If the playing buffer has been consumed or has no frames, we need a new one
-                if (playing_buffer.consumed || playing_buffer.frames == 0) {
-                    if (!impl->queue.try_dequeue(impl->playing_buffer)) {
-                        // If no buffer was available we've underrun, just push the samples and
-                        // continue.
-                        underrun = true;
-                        impl->samples_buffer.Push(&input_buffer[frames_written * frame_size],
-                                                  (num_frames - frames_written) * frame_size);
-                        frames_written = num_frames;
-                        continue;
-                    } else {
-                        impl->queued_buffers--;
-                        impl->SignalEvent(impl->playing_buffer);
-                    }
-                }
-
-                // Get the minimum frames available between the currently playing buffer, and the
-                // amount we have left to fill
-                size_t frames_available{
-                    std::min(playing_buffer.frames - playing_buffer.frames_played,
-                             num_frames - frames_written)};
-
-                impl->samples_buffer.Push(&input_buffer[frames_written * frame_size],
-                                          frames_available * frame_size);
-
-                frames_written += frames_available;
-                playing_buffer.frames_played += frames_available;
-
-                // If that's all the frames in the current buffer, add its samples and mark it as
-                // consumed
-                if (playing_buffer.frames_played >= playing_buffer.frames) {
-                    impl->AddPlayedSampleCount(playing_buffer.frames_played * num_channels);
-                    impl->playing_buffer.consumed = true;
-                }
-            }
-
-            std::memcpy(&impl->last_frame[0], &input_buffer[(frames_written - 1) * frame_size],
-                        frame_size_bytes);
+            std::span<const s16> input_buffer{reinterpret_cast<const s16*>(stream),
+                                              num_frames * frame_size};
+            impl->ProcessAudioIn(input_buffer, num_frames);
         } else {
             std::span<s16> output_buffer{reinterpret_cast<s16*>(stream), num_frames * frame_size};
-
-            while (frames_written < num_frames) {
-                auto& playing_buffer{impl->playing_buffer};
-
-                // If the playing buffer has been consumed or has no frames, we need a new one
-                if (playing_buffer.consumed || playing_buffer.frames == 0) {
-                    if (!impl->queue.try_dequeue(impl->playing_buffer)) {
-                        // If no buffer was available we've underrun, fill the remaining buffer with
-                        // the last written frame and continue.
-                        underrun = true;
-                        for (size_t i = frames_written; i < num_frames; i++) {
-                            std::memcpy(&output_buffer[i * frame_size], &impl->last_frame[0],
-                                        frame_size_bytes);
-                        }
-                        frames_written = num_frames;
-                        continue;
-                    } else {
-                        impl->queued_buffers--;
-                        impl->SignalEvent(impl->playing_buffer);
-                    }
-                }
-
-                // Get the minimum frames available between the currently playing buffer, and the
-                // amount we have left to fill
-                size_t frames_available{
-                    std::min(playing_buffer.frames - playing_buffer.frames_played,
-                             num_frames - frames_written)};
-
-                impl->samples_buffer.Pop(&output_buffer[frames_written * frame_size],
-                                         frames_available * frame_size);
-
-                frames_written += frames_available;
-                playing_buffer.frames_played += frames_available;
-
-                // If that's all the frames in the current buffer, add its samples and mark it as
-                // consumed
-                if (playing_buffer.frames_played >= playing_buffer.frames) {
-                    impl->AddPlayedSampleCount(playing_buffer.frames_played * num_channels);
-                    impl->playing_buffer.consumed = true;
-                }
-            }
-
-            std::memcpy(&impl->last_frame[0], &output_buffer[(frames_written - 1) * frame_size],
-                        frame_size_bytes);
+            impl->ProcessAudioOutAndRender(output_buffer, num_frames);
         }
     }
 
     /// SDL device id of the opened input/output device
     SDL_AudioDeviceID device{};
-    /// Type of this stream
-    StreamType type;
-    /// Core system
-    Core::System& system;
-    /// Ring buffer of the samples waiting to be played or consumed
-    Common::RingBuffer<s16, 0x10000> samples_buffer;
-    /// Audio buffers queued and waiting to play
-    Common::ReaderWriterQueue<::AudioCore::Sink::SinkBuffer> queue;
-    /// The currently-playing audio buffer
-    ::AudioCore::Sink::SinkBuffer playing_buffer{};
-    /// Audio buffers which have been played and are in queue to be released by the audio system
-    Common::ReaderWriterQueue<::AudioCore::Sink::SinkBuffer> released_buffers{};
-    /// Currently released buffer waiting to be taken by the audio system
-    ::AudioCore::Sink::SinkBuffer released_buffer{};
-    /// The last played (or received) frame of audio, used when the callback underruns
-    std::array<s16, MaxChannels> last_frame{};
 };
 
 SDLSink::SDLSink(std::string_view target_device_name) {
@@ -482,14 +186,14 @@ SDLSink::SDLSink(std::string_view target_device_name) {
 
 SDLSink::~SDLSink() = default;
 
-SinkStream* SDLSink::AcquireSinkStream(Core::System& system, const u32 system_channels,
-                                       const std::string&, const StreamType type) {
+SinkStream* SDLSink::AcquireSinkStream(Core::System& system, u32 system_channels,
+                                       const std::string&, StreamType type) {
     SinkStreamPtr& stream = sink_streams.emplace_back(std::make_unique<SDLSinkStream>(
         device_channels, system_channels, output_device, input_device, type, system));
     return stream.get();
 }
 
-void SDLSink::CloseStream(const SinkStream* stream) {
+void SDLSink::CloseStream(SinkStream* stream) {
     for (size_t i = 0; i < sink_streams.size(); i++) {
         if (sink_streams[i].get() == stream) {
             sink_streams[i].reset();
@@ -523,19 +227,19 @@ f32 SDLSink::GetDeviceVolume() const {
     return sink_streams[0]->GetDeviceVolume();
 }
 
-void SDLSink::SetDeviceVolume(const f32 volume) {
+void SDLSink::SetDeviceVolume(f32 volume) {
     for (auto& stream : sink_streams) {
         stream->SetDeviceVolume(volume);
     }
 }
 
-void SDLSink::SetSystemVolume(const f32 volume) {
+void SDLSink::SetSystemVolume(f32 volume) {
     for (auto& stream : sink_streams) {
         stream->SetSystemVolume(volume);
     }
 }
 
-std::vector<std::string> ListSDLSinkDevices(const bool capture) {
+std::vector<std::string> ListSDLSinkDevices(bool capture) {
     std::vector<std::string> device_list;
 
     if (!SDL_WasInit(SDL_INIT_AUDIO)) {

src/audio_core/sink/sdl2_sink.h

@@ -44,7 +44,7 @@ public:
      *
      * @param stream - The stream to close.
      */
-    void CloseStream(const SinkStream* stream) override;
+    void CloseStream(SinkStream* stream) override;
 
     /**
      * Close all streams.

src/audio_core/sink/sink.h

@@ -32,7 +32,7 @@ public:
      *
      * @param stream - The stream to close.
      */
-    virtual void CloseStream(const SinkStream* stream) = 0;
+    virtual void CloseStream(SinkStream* stream) = 0;
 
     /**
      * Close all streams.

src/audio_core/sink/sink_details.cpp

@@ -5,7 +5,7 @@
 #include <memory>
 #include <string>
 #include <vector>
-#include "audio_core/sink/null_sink.h"
+
 #include "audio_core/sink/sink_details.h"
 #ifdef HAVE_CUBEB
 #include "audio_core/sink/cubeb_sink.h"
@@ -13,6 +13,7 @@
 #ifdef HAVE_SDL2
 #include "audio_core/sink/sdl2_sink.h"
 #endif
+#include "audio_core/sink/null_sink.h"
 #include "common/logging/log.h"
 
 namespace AudioCore::Sink {
@@ -59,8 +60,7 @@ const SinkDetails& GetOutputSinkDetails(std::string_view sink_id) {
 
     if (sink_id == "auto" || iter == std::end(sink_details)) {
         if (sink_id != "auto") {
-            LOG_ERROR(Audio, "AudioCore::Sink::GetOutputSinkDetails given invalid sink_id {}",
-                      sink_id);
+            LOG_ERROR(Audio, "Invalid sink_id {}", sink_id);
         }
         // Auto-select.
         // sink_details is ordered in terms of desirability, with the best choice at the front.

src/audio_core/sink/sink_stream.cpp

@@ -0,0 +1,263 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <array>
+#include <atomic>
+#include <memory>
+#include <span>
+#include <vector>
+
+#include "audio_core/audio_core.h"
+#include "audio_core/common/common.h"
+#include "audio_core/sink/sink_stream.h"
+#include "common/common_types.h"
+#include "common/fixed_point.h"
+#include "common/settings.h"
+#include "core/core.h"
+
+namespace AudioCore::Sink {
+
+void SinkStream::AppendBuffer(SinkBuffer& buffer, std::vector<s16>& samples) {
+    if (type == StreamType::In) {
+        queue.enqueue(buffer);
+        queued_buffers++;
+        return;
+    }
+
+    constexpr s32 min{std::numeric_limits<s16>::min()};
+    constexpr s32 max{std::numeric_limits<s16>::max()};
+
+    auto yuzu_volume{Settings::Volume()};
+    if (yuzu_volume > 1.0f) {
+        yuzu_volume = 0.6f + 20 * std::log10(yuzu_volume);
+    }
+    auto volume{system_volume * device_volume * yuzu_volume};
+
+    if (system_channels == 6 && device_channels == 2) {
+        // We're given 6 channels, but our device only outputs 2, so downmix.
+        constexpr std::array<f32, 4> down_mix_coeff{1.0f, 0.707f, 0.251f, 0.707f};
+
+        for (u32 read_index = 0, write_index = 0; read_index < samples.size();
+             read_index += system_channels, write_index += device_channels) {
+            const auto left_sample{
+                ((Common::FixedPoint<49, 15>(
+                      samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
+                      down_mix_coeff[0] +
+                  samples[read_index + static_cast<u32>(Channels::Center)] * down_mix_coeff[1] +
+                  samples[read_index + static_cast<u32>(Channels::LFE)] * down_mix_coeff[2] +
+                  samples[read_index + static_cast<u32>(Channels::BackLeft)] * down_mix_coeff[3]) *
+                 volume)
+                    .to_int()};
+
+            const auto right_sample{
+                ((Common::FixedPoint<49, 15>(
+                      samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
+                      down_mix_coeff[0] +
+                  samples[read_index + static_cast<u32>(Channels::Center)] * down_mix_coeff[1] +
+                  samples[read_index + static_cast<u32>(Channels::LFE)] * down_mix_coeff[2] +
+                  samples[read_index + static_cast<u32>(Channels::BackRight)] * down_mix_coeff[3]) *
+                 volume)
+                    .to_int()};
+
+            samples[write_index + static_cast<u32>(Channels::FrontLeft)] =
+                static_cast<s16>(std::clamp(left_sample, min, max));
+            samples[write_index + static_cast<u32>(Channels::FrontRight)] =
+                static_cast<s16>(std::clamp(right_sample, min, max));
+        }
+
+        samples.resize(samples.size() / system_channels * device_channels);
+
+    } else if (system_channels == 2 && device_channels == 6) {
+        // We need moar samples! Not all games will provide 6 channel audio.
+        // TODO: Implement some upmixing here. Currently just passthrough, with other
+        // channels left as silence.
+        std::vector<s16> new_samples(samples.size() / system_channels * device_channels, 0);
+
+        for (u32 read_index = 0, write_index = 0; read_index < samples.size();
+             read_index += system_channels, write_index += device_channels) {
+            const auto left_sample{static_cast<s16>(std::clamp(
+                static_cast<s32>(
+                    static_cast<f32>(samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
+                    volume),
+                min, max))};
+
+            new_samples[write_index + static_cast<u32>(Channels::FrontLeft)] = left_sample;
+
+            const auto right_sample{static_cast<s16>(std::clamp(
+                static_cast<s32>(
+                    static_cast<f32>(samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
+                    volume),
+                min, max))};
+
+            new_samples[write_index + static_cast<u32>(Channels::FrontRight)] = right_sample;
+        }
+        samples = std::move(new_samples);
+
+    } else if (volume != 1.0f) {
+        for (u32 i = 0; i < samples.size(); i++) {
+            samples[i] = static_cast<s16>(
+                std::clamp(static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
+        }
+    }
+
+    samples_buffer.Push(samples);
+    queue.enqueue(buffer);
+    queued_buffers++;
+}
+
+std::vector<s16> SinkStream::ReleaseBuffer(u64 num_samples) {
+    constexpr s32 min = std::numeric_limits<s16>::min();
+    constexpr s32 max = std::numeric_limits<s16>::max();
+
+    auto samples{samples_buffer.Pop(num_samples)};
+
+    // TODO: Up-mix to 6 channels if the game expects it.
+    // For audio input this is unlikely to ever be the case though.
+
+    // Incoming mic volume seems to always be very quiet, so multiply by an additional 8 here.
+    // TODO: Play with this and find something that works better.
+    auto volume{system_volume * device_volume * 8};
+    for (u32 i = 0; i < samples.size(); i++) {
+        samples[i] = static_cast<s16>(
+            std::clamp(static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
+    }
+
+    if (samples.size() < num_samples) {
+        samples.resize(num_samples, 0);
+    }
+    return samples;
+}
+
+void SinkStream::ClearQueue() {
+    samples_buffer.Pop();
+    while (queue.pop()) {
+    }
+    queued_buffers = 0;
+    playing_buffer = {};
+    playing_buffer.consumed = true;
+}
+
+void SinkStream::ProcessAudioIn(std::span<const s16> input_buffer, std::size_t num_frames) {
+    const std::size_t num_channels = GetDeviceChannels();
+    const std::size_t frame_size = num_channels;
+    const std::size_t frame_size_bytes = frame_size * sizeof(s16);
+    size_t frames_written{0};
+
+    if (queued_buffers > max_queue_size) {
+        Stall();
+    }
+
+    while (frames_written < num_frames) {
+        // If the playing buffer has been consumed or has no frames, we need a new one
+        if (playing_buffer.consumed || playing_buffer.frames == 0) {
+            if (!queue.try_dequeue(playing_buffer)) {
+                // If no buffer was available we've underrun, just push the samples and
+                // continue.
+                samples_buffer.Push(&input_buffer[frames_written * frame_size],
+                                    (num_frames - frames_written) * frame_size);
+                frames_written = num_frames;
+                continue;
+            }
+            // Successfully dequeued a new buffer.
+            queued_buffers--;
+        }
+
+        // Get the minimum frames available between the currently playing buffer, and the
+        // amount we have left to fill
+        size_t frames_available{std::min(playing_buffer.frames - playing_buffer.frames_played,
+                                         num_frames - frames_written)};
+
+        samples_buffer.Push(&input_buffer[frames_written * frame_size],
+                            frames_available * frame_size);
+
+        frames_written += frames_available;
+        playing_buffer.frames_played += frames_available;
+
+        // If that's all the frames in the current buffer, add its samples and mark it as
+        // consumed
+        if (playing_buffer.frames_played >= playing_buffer.frames) {
+            playing_buffer.consumed = true;
+        }
+    }
+
+    std::memcpy(&last_frame[0], &input_buffer[(frames_written - 1) * frame_size], frame_size_bytes);
+
+    if (queued_buffers <= max_queue_size) {
+        Unstall();
+    }
+}
+
+void SinkStream::ProcessAudioOutAndRender(std::span<s16> output_buffer, std::size_t num_frames) {
+    const std::size_t num_channels = GetDeviceChannels();
+    const std::size_t frame_size = num_channels;
+    const std::size_t frame_size_bytes = frame_size * sizeof(s16);
+    size_t frames_written{0};
+
+    // Due to many frames being queued up with nvdec (5 frames or so?), a lot of buffers also get
+    // queued up (30+) but not all at once, which causes constant stalling here, so just let the
+    // video play out without attempting to stall.
+    // Can hopefully remove this later with a more complete NVDEC implementation.
+    const auto nvdec_active{system.AudioCore().IsNVDECActive()};
+    if (!nvdec_active && queued_buffers > max_queue_size) {
+        Stall();
+    }
+
+    while (frames_written < num_frames) {
+        // If the playing buffer has been consumed or has no frames, we need a new one
+        if (playing_buffer.consumed || playing_buffer.frames == 0) {
+            if (!queue.try_dequeue(playing_buffer)) {
+                // If no buffer was available we've underrun, fill the remaining buffer with
+                // the last written frame and continue.
+                for (size_t i = frames_written; i < num_frames; i++) {
+                    std::memcpy(&output_buffer[i * frame_size], &last_frame[0], frame_size_bytes);
+                }
+                frames_written = num_frames;
+                continue;
+            }
+            // Successfully dequeued a new buffer.
+            queued_buffers--;
+        }
+
+        // Get the minimum frames available between the currently playing buffer, and the
+        // amount we have left to fill
+        size_t frames_available{std::min(playing_buffer.frames - playing_buffer.frames_played,
+                                         num_frames - frames_written)};
+
+        samples_buffer.Pop(&output_buffer[frames_written * frame_size],
+                           frames_available * frame_size);
+
+        frames_written += frames_available;
+        playing_buffer.frames_played += frames_available;
+
+        // If that's all the frames in the current buffer, add its samples and mark it as
+        // consumed
+        if (playing_buffer.frames_played >= playing_buffer.frames) {
+            playing_buffer.consumed = true;
+        }
+    }
+
+    std::memcpy(&last_frame[0], &output_buffer[(frames_written - 1) * frame_size],
+                frame_size_bytes);
+
+    if (stalled && queued_buffers <= max_queue_size) {
+        Unstall();
+    }
+}
+
+void SinkStream::Stall() {
+    if (stalled) {
+        return;
+    }
+    stalled = true;
+    system.StallProcesses();
+}
+
+void SinkStream::Unstall() {
+    if (!stalled) {
+        return;
+    }
+    system.UnstallProcesses();
+    stalled = false;
+}
+
+} // namespace AudioCore::Sink

src/audio_core/sink/sink_stream.h

@@ -3,12 +3,20 @@
 
 #pragma once
 
+#include <array>
 #include <atomic>
 #include <memory>
+#include <span>
 #include <vector>
 
 #include "audio_core/common/common.h"
 #include "common/common_types.h"
+#include "common/reader_writer_queue.h"
+#include "common/ring_buffer.h"
+
+namespace Core {
+class System;
+} // namespace Core
 
 namespace AudioCore::Sink {
 
@@ -34,20 +42,24 @@ struct SinkBuffer {
  * You should regularly call IsBufferConsumed with the unique SinkBuffer tag to check if the buffer
  * has been consumed.
  *
- * Since these are a FIFO queue, always check IsBufferConsumed in the same order you appended the
- * buffers, skipping a buffer will result in all following buffers to never release.
+ * Since these are a FIFO queue, IsBufferConsumed must be checked in the same order buffers were
+ * appended, skipping a buffer will result in the queue getting stuck, and all following buffers to
+ * never release.
  *
  * If the buffers appear to be stuck, you can stop and re-open an IAudioIn/IAudioOut service (this
  * is what games do), or call ClearQueue to flush all of the buffers without a full restart.
  */
 class SinkStream {
 public:
-    virtual ~SinkStream() = default;
+    explicit SinkStream(Core::System& system_, StreamType type_) : system{system_}, type{type_} {}
+    virtual ~SinkStream() {
+        Unstall();
+    }
 
     /**
      * Finalize the sink stream.
      */
-    virtual void Finalize() = 0;
+    virtual void Finalize() {}
 
     /**
      * Start the sink stream.
@@ -55,48 +67,19 @@ public:
      * @param resume - Set to true if this is resuming the stream a previously-active stream.
      *                 Default false.
      */
-    virtual void Start(bool resume = false) = 0;
+    virtual void Start(bool resume = false) {}
 
     /**
      * Stop the sink stream.
      */
-    virtual void Stop() = 0;
-
-    /**
-     * Append a new buffer and its samples to a waiting queue to play.
-     *
-     * @param buffer  - Audio buffer information to be queued.
-     * @param samples - The s16 samples to be queue for playback.
-     */
-    virtual void AppendBuffer(SinkBuffer& buffer, std::vector<s16>& samples) = 0;
-
-    /**
-     * Release a buffer. Audio In only, will fill a buffer with recorded samples.
-     *
-     * @param num_samples - Maximum number of samples to receive.
-     * @return Vector of recorded samples. May have fewer than num_samples.
-     */
-    virtual std::vector<s16> ReleaseBuffer(u64 num_samples) = 0;
-
-    /**
-     * Check if a certain buffer has been consumed (fully played).
-     *
-     * @param tag - Unique tag of a buffer to check for.
-     * @return True if the buffer has been played, otherwise false.
-     */
-    virtual bool IsBufferConsumed(u64 tag) = 0;
-
-    /**
-     * Empty out the buffer queue.
-     */
-    virtual void ClearQueue() = 0;
+    virtual void Stop() {}
 
     /**
      * Check if the stream is paused.
      *
      * @return True if paused, otherwise false.
      */
-    bool IsPaused() {
+    bool IsPaused() const {
         return paused;
     }
 
@@ -127,34 +110,6 @@ public:
         return device_channels;
     }
 
-    /**
-     * Get the total number of samples played by this stream.
-     *
-     * @return Number of samples played.
-     */
-    u64 GetPlayedSampleCount() const {
-        return played_sample_count;
-    }
-
-    /**
-     * Set the number of samples played.
-     * This is started and stopped on system start/stop.
-     *
-     * @param played_sample_count_ - Number of samples to set.
-     */
-    void SetPlayedSampleCount(u64 played_sample_count_) {
-        played_sample_count = played_sample_count_;
-    }
-
-    /**
-     * Add to the played sample count.
-     *
-     * @param num_samples - Number of samples to add.
-     */
-    void AddPlayedSampleCount(u64 num_samples) {
-        played_sample_count += num_samples;
-    }
-
     /**
      * Get the system volume.
      *
@@ -200,15 +155,65 @@ public:
         return queued_buffers.load();
     }
 
+    /**
+     * Set the maximum buffer queue size.
+     */
+    void SetRingSize(u32 ring_size) {
+        max_queue_size = ring_size;
+    }
+
+    /**
+     * Append a new buffer and its samples to a waiting queue to play.
+     *
+     * @param buffer  - Audio buffer information to be queued.
+     * @param samples - The s16 samples to be queue for playback.
+     */
+    virtual void AppendBuffer(SinkBuffer& buffer, std::vector<s16>& samples);
+
+    /**
+     * Release a buffer. Audio In only, will fill a buffer with recorded samples.
+     *
+     * @param num_samples - Maximum number of samples to receive.
+     * @return Vector of recorded samples. May have fewer than num_samples.
+     */
+    virtual std::vector<s16> ReleaseBuffer(u64 num_samples);
+
+    /**
+     * Empty out the buffer queue.
+     */
+    void ClearQueue();
+
+    /**
+     * Callback for AudioIn.
+     *
+     * @param input_buffer - Input buffer to be filled with samples.
+     * @param num_frames - Number of frames to be filled.
+     */
+    void ProcessAudioIn(std::span<const s16> input_buffer, std::size_t num_frames);
+
+    /**
+     * Callback for AudioOut and AudioRenderer.
+     *
+     * @param output_buffer - Output buffer to be filled with samples.
+     * @param num_frames - Number of frames to be filled.
+     */
+    void ProcessAudioOutAndRender(std::span<s16> output_buffer, std::size_t num_frames);
+
+    /**
+     * Stall core processes if the audio thread falls too far behind.
+     */
+    void Stall();
+
+    /**
+     * Unstall core processes.
+     */
+    void Unstall();
+
 protected:
-    /// Number of buffers waiting to be played
-    std::atomic<u32> queued_buffers{};
-    /// Total samples played by this stream
-    std::atomic<u64> played_sample_count{};
-    /// Set by the audio render/in/out system which uses this stream
-    f32 system_volume{1.0f};
-    /// Set via IAudioDevice service calls
-    f32 device_volume{1.0f};
+    /// Core system
+    Core::System& system;
+    /// Type of this stream
+    StreamType type;
     /// Set by the audio render/in/out systen which uses this stream
     u32 system_channels{2};
     /// Channels supported by hardware
@@ -217,6 +222,28 @@ protected:
     std::atomic<bool> paused{true};
     /// Was this stream previously playing?
     std::atomic<bool> was_playing{false};
+    /// Name of this stream
+    std::string name{};
+
+private:
+    /// Ring buffer of the samples waiting to be played or consumed
+    Common::RingBuffer<s16, 0x10000> samples_buffer;
+    /// Audio buffers queued and waiting to play
+    Common::ReaderWriterQueue<SinkBuffer> queue;
+    /// The currently-playing audio buffer
+    SinkBuffer playing_buffer{};
+    /// The last played (or received) frame of audio, used when the callback underruns
+    std::array<s16, MaxChannels> last_frame{};
+    /// Number of buffers waiting to be played
+    std::atomic<u32> queued_buffers{};
+    /// The ring size for audio out buffers (usually 4, rarely 2 or 8)
+    u32 max_queue_size{};
+    /// Set by the audio render/in/out system which uses this stream
+    f32 system_volume{1.0f};
+    /// Set via IAudioDevice service calls
+    f32 device_volume{1.0f};
+    /// True if coretiming has been stalled
+    bool stalled{false};
 };
 
 using SinkStreamPtr = std::unique_ptr<SinkStream>;

src/common/input.h

@@ -102,6 +102,8 @@ struct AnalogProperties {
     float offset{};
     // Invert direction of the sensor data
     bool inverted{};
+    // Press once to activate, press again to release
+    bool toggle{};
 };
 
 // Single analog sensor data
@@ -115,8 +117,11 @@ struct AnalogStatus {
 struct ButtonStatus {
     Common::UUID uuid{};
     bool value{};
+    // Invert value of the button
     bool inverted{};
+    // Press once to activate, press again to release
     bool toggle{};
+    // Internal lock for the toggle status
     bool locked{};
 };
 

src/common/settings.cpp

@@ -195,6 +195,7 @@ void RestoreGlobalState(bool is_powered_on) {
     values.shader_backend.SetGlobal(true);
     values.use_asynchronous_shaders.SetGlobal(true);
     values.use_fast_gpu_time.SetGlobal(true);
+    values.use_pessimistic_flushes.SetGlobal(true);
     values.bg_red.SetGlobal(true);
     values.bg_green.SetGlobal(true);
     values.bg_blue.SetGlobal(true);

src/common/settings.h

@@ -446,6 +446,7 @@ struct Values {
                                                           ShaderBackend::SPIRV, "shader_backend"};
     SwitchableSetting<bool> use_asynchronous_shaders{false, "use_asynchronous_shaders"};
     SwitchableSetting<bool> use_fast_gpu_time{true, "use_fast_gpu_time"};
+    SwitchableSetting<bool> use_pessimistic_flushes{false, "use_pessimistic_flushes"};
 
     SwitchableSetting<u8> bg_red{0, "bg_red"};
     SwitchableSetting<u8> bg_green{0, "bg_green"};

src/core/hid/emulated_controller.cpp

@@ -562,6 +562,16 @@ void EmulatedController::SetButton(const Common::Input::CallbackStatus& callback
         return;
     }
 
+    // GC controllers have triggers not buttons
+    if (npad_type == NpadStyleIndex::GameCube) {
+        if (index == Settings::NativeButton::ZR) {
+            return;
+        }
+        if (index == Settings::NativeButton::ZL) {
+            return;
+        }
+    }
+
     switch (index) {
     case Settings::NativeButton::A:
         controller.npad_button_state.a.Assign(current_status.value);
@@ -738,6 +748,11 @@ void EmulatedController::SetTrigger(const Common::Input::CallbackStatus& callbac
         return;
     }
 
+    // Only GC controllers have analog triggers
+    if (npad_type != NpadStyleIndex::GameCube) {
+        return;
+    }
+
     const auto& trigger = controller.trigger_values[index];
 
     switch (index) {

src/core/hid/input_converter.cpp

@@ -52,6 +52,9 @@ Common::Input::ButtonStatus TransformToButton(const Common::Input::CallbackStatu
     Common::Input::ButtonStatus status{};
     switch (callback.type) {
     case Common::Input::InputType::Analog:
+        status.value = TransformToTrigger(callback).pressed.value;
+        status.toggle = callback.analog_status.properties.toggle;
+        break;
     case Common::Input::InputType::Trigger:
         status.value = TransformToTrigger(callback).pressed.value;
         break;

src/core/hle/result.h

@@ -117,6 +117,7 @@ union Result {
     BitField<0, 9, ErrorModule> module;
     BitField<9, 13, u32> description;
 
+    Result() = default;
     constexpr explicit Result(u32 raw_) : raw(raw_) {}
 
     constexpr Result(ErrorModule module_, u32 description_)
@@ -130,6 +131,7 @@ union Result {
         return !IsSuccess();
     }
 };
+static_assert(std::is_trivial_v<Result>);
 
 [[nodiscard]] constexpr bool operator==(const Result& a, const Result& b) {
     return a.raw == b.raw;

src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp

@@ -1,6 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include "audio_core/audio_core.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "core/core.h"
@@ -65,7 +66,10 @@ NvResult nvhost_nvdec::Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>&
     return NvResult::NotImplemented;
 }
 
-void nvhost_nvdec::OnOpen(DeviceFD fd) {}
+void nvhost_nvdec::OnOpen(DeviceFD fd) {
+    LOG_INFO(Service_NVDRV, "NVDEC video stream started");
+    system.AudioCore().SetNVDECActive(true);
+}
 
 void nvhost_nvdec::OnClose(DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream ended");
@@ -73,6 +77,7 @@ void nvhost_nvdec::OnClose(DeviceFD fd) {
     if (iter != fd_to_id.end()) {
         system.GPU().ClearCdmaInstance(iter->second);
     }
+    system.AudioCore().SetNVDECActive(false);
 }
 
 } // namespace Service::Nvidia::Devices

src/input_common/input_poller.cpp

@@ -824,6 +824,7 @@ std::unique_ptr<Common::Input::InputDevice> InputFactory::CreateAnalogDevice(
         .threshold = std::clamp(params.Get("threshold", 0.5f), 0.0f, 1.0f),
         .offset = std::clamp(params.Get("offset", 0.0f), -1.0f, 1.0f),
         .inverted = params.Get("invert", "+") == "-",
+        .toggle = static_cast<bool>(params.Get("toggle", false)),
     };
     input_engine->PreSetController(identifier);
     input_engine->PreSetAxis(identifier, axis);

src/video_core/buffer_cache/buffer_base.h

@@ -12,6 +12,7 @@
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/div_ceil.h"
+#include "common/settings.h"
 #include "core/memory.h"
 
 namespace VideoCommon {
@@ -219,7 +220,9 @@ public:
             NotifyRasterizer<false>(word_index, untracked_words[word_index], cached_bits);
             untracked_words[word_index] |= cached_bits;
             cpu_words[word_index] |= cached_bits;
-            cached_words[word_index] = 0;
+            if (!Settings::values.use_pessimistic_flushes) {
+                cached_words[word_index] = 0;
+            }
         }
     }
 

src/video_core/textures/astc.cpp

@@ -13,7 +13,9 @@
 
 #include <boost/container/static_vector.hpp>
 
+#include "common/alignment.h"
 #include "common/common_types.h"
+#include "common/thread_worker.h"
 #include "video_core/textures/astc.h"
 
 class InputBitStream {
@@ -1650,29 +1652,41 @@ static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth,
 
 void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
                 uint32_t block_width, uint32_t block_height, std::span<uint8_t> output) {
-    u32 block_index = 0;
-    std::size_t depth_offset = 0;
-    for (u32 z = 0; z < depth; z++) {
-        for (u32 y = 0; y < height; y += block_height) {
-            for (u32 x = 0; x < width; x += block_width) {
-                const std::span<const u8, 16> blockPtr{data.subspan(block_index * 16, 16)};
+    const u32 rows = Common::DivideUp(height, block_height);
+    const u32 cols = Common::DivideUp(width, block_width);
 
-                // Blocks can be at most 12x12
-                std::array<u32, 12 * 12> uncompData;
-                DecompressBlock(blockPtr, block_width, block_height, uncompData);
+    Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2,
+                                 "yuzu:ASTCDecompress"};
 
-                u32 decompWidth = std::min(block_width, width - x);
-                u32 decompHeight = std::min(block_height, height - y);
+    for (u32 z = 0; z < depth; ++z) {
+        const u32 depth_offset = z * height * width * 4;
+        for (u32 y_index = 0; y_index < rows; ++y_index) {
+            auto decompress_stride = [data, width, height, depth, block_width, block_height, output,
+                                      rows, cols, z, depth_offset, y_index] {
+                const u32 y = y_index * block_height;
+                for (u32 x_index = 0; x_index < cols; ++x_index) {
+                    const u32 block_index = (z * rows * cols) + (y_index * cols) + x_index;
+                    const u32 x = x_index * block_width;
 
-                const std::span<u8> outRow = output.subspan(depth_offset + (y * width + x) * 4);
-                for (u32 jj = 0; jj < decompHeight; jj++) {
-                    std::memcpy(outRow.data() + jj * width * 4,
-                                uncompData.data() + jj * block_width, decompWidth * 4);
+                    const std::span<const u8, 16> blockPtr{data.subspan(block_index * 16, 16)};
+
+                    // Blocks can be at most 12x12
+                    std::array<u32, 12 * 12> uncompData;
+                    DecompressBlock(blockPtr, block_width, block_height, uncompData);
+
+                    u32 decompWidth = std::min(block_width, width - x);
+                    u32 decompHeight = std::min(block_height, height - y);
+
+                    const std::span<u8> outRow = output.subspan(depth_offset + (y * width + x) * 4);
+                    for (u32 h = 0; h < decompHeight; ++h) {
+                        std::memcpy(outRow.data() + h * width * 4,
+                                    uncompData.data() + h * block_width, decompWidth * 4);
+                    }
                 }
-                ++block_index;
-            }
+            };
+            workers.QueueWork(std::move(decompress_stride));
         }
-        depth_offset += height * width * 4;
+        workers.WaitForRequests();
     }
 }
 

src/yuzu/configuration/config.cpp

@@ -684,6 +684,7 @@ void Config::ReadRendererValues() {
     ReadGlobalSetting(Settings::values.shader_backend);
     ReadGlobalSetting(Settings::values.use_asynchronous_shaders);
     ReadGlobalSetting(Settings::values.use_fast_gpu_time);
+    ReadGlobalSetting(Settings::values.use_pessimistic_flushes);
     ReadGlobalSetting(Settings::values.bg_red);
     ReadGlobalSetting(Settings::values.bg_green);
     ReadGlobalSetting(Settings::values.bg_blue);
@@ -1300,6 +1301,7 @@ void Config::SaveRendererValues() {
                  Settings::values.shader_backend.UsingGlobal());
     WriteGlobalSetting(Settings::values.use_asynchronous_shaders);
     WriteGlobalSetting(Settings::values.use_fast_gpu_time);
+    WriteGlobalSetting(Settings::values.use_pessimistic_flushes);
     WriteGlobalSetting(Settings::values.bg_red);
     WriteGlobalSetting(Settings::values.bg_green);
     WriteGlobalSetting(Settings::values.bg_blue);

src/yuzu/configuration/configure_graphics_advanced.cpp

@@ -28,6 +28,7 @@ void ConfigureGraphicsAdvanced::SetConfiguration() {
     ui->use_vsync->setChecked(Settings::values.use_vsync.GetValue());
     ui->use_asynchronous_shaders->setChecked(Settings::values.use_asynchronous_shaders.GetValue());
     ui->use_fast_gpu_time->setChecked(Settings::values.use_fast_gpu_time.GetValue());
+    ui->use_pessimistic_flushes->setChecked(Settings::values.use_pessimistic_flushes.GetValue());
 
     if (Settings::IsConfiguringGlobal()) {
         ui->gpu_accuracy->setCurrentIndex(
@@ -55,6 +56,8 @@ void ConfigureGraphicsAdvanced::ApplyConfiguration() {
                                              use_asynchronous_shaders);
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_fast_gpu_time,
                                              ui->use_fast_gpu_time, use_fast_gpu_time);
+    ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_pessimistic_flushes,
+                                             ui->use_pessimistic_flushes, use_pessimistic_flushes);
 }
 
 void ConfigureGraphicsAdvanced::changeEvent(QEvent* event) {
@@ -77,6 +80,8 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
         ui->use_asynchronous_shaders->setEnabled(
             Settings::values.use_asynchronous_shaders.UsingGlobal());
         ui->use_fast_gpu_time->setEnabled(Settings::values.use_fast_gpu_time.UsingGlobal());
+        ui->use_pessimistic_flushes->setEnabled(
+            Settings::values.use_pessimistic_flushes.UsingGlobal());
         ui->anisotropic_filtering_combobox->setEnabled(
             Settings::values.max_anisotropy.UsingGlobal());
 
@@ -89,6 +94,9 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
                                             use_asynchronous_shaders);
     ConfigurationShared::SetColoredTristate(ui->use_fast_gpu_time,
                                             Settings::values.use_fast_gpu_time, use_fast_gpu_time);
+    ConfigurationShared::SetColoredTristate(ui->use_pessimistic_flushes,
+                                            Settings::values.use_pessimistic_flushes,
+                                            use_pessimistic_flushes);
     ConfigurationShared::SetColoredComboBox(
         ui->gpu_accuracy, ui->label_gpu_accuracy,
         static_cast<int>(Settings::values.gpu_accuracy.GetValue(true)));

src/yuzu/configuration/configure_graphics_advanced.h

@@ -39,6 +39,7 @@ private:
     ConfigurationShared::CheckState use_vsync;
     ConfigurationShared::CheckState use_asynchronous_shaders;
     ConfigurationShared::CheckState use_fast_gpu_time;
+    ConfigurationShared::CheckState use_pessimistic_flushes;
 
     const Core::System& system;
 };

src/yuzu/configuration/configure_graphics_advanced.ui

@@ -99,6 +99,16 @@
           </property>
          </widget>
         </item>
+        <item>
+         <widget class="QCheckBox" name="use_pessimistic_flushes">
+          <property name="toolTip">
+            <string>Enables pessimistic buffer flushes. This option will force unmodified buffers to be flushed, which can cost performance.</string>
+          </property>
+          <property name="text">
+           <string>Use pessimistic buffer flushes (Hack)</string>
+          </property>
+         </widget>
+        </item>
         <item>
          <widget class="QWidget" name="af_layout" native="true">
           <layout class="QHBoxLayout" name="horizontalLayout_1">

src/yuzu/configuration/configure_input_player.cpp

@@ -161,6 +161,7 @@ QString ConfigureInputPlayer::ButtonToText(const Common::ParamPackage& param) {
 
     const QString toggle = QString::fromStdString(param.Get("toggle", false) ? "~" : "");
     const QString inverted = QString::fromStdString(param.Get("inverted", false) ? "!" : "");
+    const QString invert = QString::fromStdString(param.Get("invert", "+") == "-" ? "-" : "");
     const auto common_button_name = input_subsystem->GetButtonName(param);
 
     // Retrieve the names from Qt
@@ -184,7 +185,7 @@ QString ConfigureInputPlayer::ButtonToText(const Common::ParamPackage& param) {
         }
         if (param.Has("axis")) {
             const QString axis = QString::fromStdString(param.Get("axis", ""));
-            return QObject::tr("%1%2Axis %3").arg(toggle, inverted, axis);
+            return QObject::tr("%1%2Axis %3").arg(toggle, invert, axis);
         }
         if (param.Has("axis_x") && param.Has("axis_y") && param.Has("axis_z")) {
             const QString axis_x = QString::fromStdString(param.Get("axis_x", ""));
@@ -362,18 +363,18 @@ ConfigureInputPlayer::ConfigureInputPlayer(QWidget* parent, std::size_t player_i
                         button_map[button_id]->setText(tr("[not set]"));
                     });
                     if (param.Has("code") || param.Has("button") || param.Has("hat")) {
-                        context_menu.addAction(tr("Toggle button"), [&] {
-                            const bool toggle_value = !param.Get("toggle", false);
-                            param.Set("toggle", toggle_value);
-                            button_map[button_id]->setText(ButtonToText(param));
-                            emulated_controller->SetButtonParam(button_id, param);
-                        });
                         context_menu.addAction(tr("Invert button"), [&] {
                             const bool invert_value = !param.Get("inverted", false);
                             param.Set("inverted", invert_value);
                             button_map[button_id]->setText(ButtonToText(param));
                             emulated_controller->SetButtonParam(button_id, param);
                         });
+                        context_menu.addAction(tr("Toggle button"), [&] {
+                            const bool toggle_value = !param.Get("toggle", false);
+                            param.Set("toggle", toggle_value);
+                            button_map[button_id]->setText(ButtonToText(param));
+                            emulated_controller->SetButtonParam(button_id, param);
+                        });
                     }
                     if (param.Has("axis")) {
                         context_menu.addAction(tr("Invert axis"), [&] {
@@ -398,6 +399,12 @@ ConfigureInputPlayer::ConfigureInputPlayer(QWidget* parent, std::size_t player_i
                             }
                             emulated_controller->SetButtonParam(button_id, param);
                         });
+                        context_menu.addAction(tr("Toggle axis"), [&] {
+                            const bool toggle_value = !param.Get("toggle", false);
+                            param.Set("toggle", toggle_value);
+                            button_map[button_id]->setText(ButtonToText(param));
+                            emulated_controller->SetButtonParam(button_id, param);
+                        });
                     }
                     context_menu.exec(button_map[button_id]->mapToGlobal(menu_location));
                 });

src/yuzu_cmd/config.cpp

@@ -314,6 +314,7 @@ void Config::ReadValues() {
     ReadSetting("Renderer", Settings::values.nvdec_emulation);
     ReadSetting("Renderer", Settings::values.accelerate_astc);
     ReadSetting("Renderer", Settings::values.use_fast_gpu_time);
+    ReadSetting("Renderer", Settings::values.use_pessimistic_flushes);
 
     ReadSetting("Renderer", Settings::values.bg_red);
     ReadSetting("Renderer", Settings::values.bg_green);

src/yuzu_cmd/default_ini.h

@@ -319,6 +319,10 @@ use_asynchronous_gpu_emulation =
 # 0: Off, 1 (default): On
 use_fast_gpu_time =
 
+# Force unmodified buffers to be flushed, which can cost performance.
+# 0: Off (default), 1: On
+use_pessimistic_flushes =
+
 # Whether to use garbage collection or not for GPU caches.
 # 0 (default): Off, 1: On
 use_caches_gc =