4 files changed, 50 insertions, 52 deletions

diff --git a/src/audio_core/sink/cubeb_sink.cpp b/src/audio_core/sink/cubeb_sink.cpp
index 9133f5388..9a0801888 100644
--- a/src/audio_core/sink/cubeb_sink.cpp
+++ b/src/audio_core/sink/cubeb_sink.cpp
@@ -101,8 +101,6 @@ public:
     ~CubebSinkStream() override {
         LOG_DEBUG(Service_Audio, "Destructing cubeb stream {}", name);
 
-        Unstall();
-
         if (!ctx) {
             return;
         }
@@ -143,8 +141,6 @@ public:
      * Stop the sink stream.
      */
     void Stop() override {
-        Unstall();
-
         if (!ctx || paused) {
             return;
         }
diff --git a/src/audio_core/sink/sdl2_sink.cpp b/src/audio_core/sink/sdl2_sink.cpp
index c138dc628..ee1a0652f 100644
--- a/src/audio_core/sink/sdl2_sink.cpp
+++ b/src/audio_core/sink/sdl2_sink.cpp
@@ -88,7 +88,6 @@ public:
      * Finalize the sink stream.
      */
     void Finalize() override {
-        Unstall();
         if (device == 0) {
             return;
         }
@@ -116,7 +115,6 @@ public:
      * Stop the sink stream.
      */
     void Stop() override {
-        Unstall();
         if (device == 0 || paused) {
             return;
         }
diff --git a/src/audio_core/sink/sink_stream.cpp b/src/audio_core/sink/sink_stream.cpp
index 39a21b0f0..f99dbd8ec 100644
--- a/src/audio_core/sink/sink_stream.cpp
+++ b/src/audio_core/sink/sink_stream.cpp
@@ -14,6 +14,8 @@
 #include "common/fixed_point.h"
 #include "common/settings.h"
 #include "core/core.h"
+#include "core/core_timing.h"
+#include "core/core_timing_util.h"
 
 namespace AudioCore::Sink {
 
@@ -149,10 +151,6 @@ void SinkStream::ProcessAudioIn(std::span<const s16> input_buffer, std::size_t n
         return;
     }
 
-    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) {
@@ -187,10 +185,6 @@ void SinkStream::ProcessAudioIn(std::span<const s16> input_buffer, std::size_t n
     }
 
     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) {
@@ -198,10 +192,15 @@ void SinkStream::ProcessAudioOutAndRender(std::span<s16> output_buffer, std::siz
     const std::size_t frame_size = num_channels;
     const std::size_t frame_size_bytes = frame_size * sizeof(s16);
     size_t frames_written{0};
+    size_t actual_frames_written{0};
 
     // If we're paused or going to shut down, we don't want to consume buffers as coretiming is
     // paused and we'll desync, so just play silence.
     if (system.IsPaused() || system.IsShuttingDown()) {
+        if (system.IsShuttingDown()) {
+            release_cv.notify_one();
+        }
+
         static constexpr std::array<s16, 6> silence{};
         for (size_t i = frames_written; i < num_frames; i++) {
             std::memcpy(&output_buffer[i * frame_size], &silence[0], frame_size_bytes);
@@ -209,20 +208,6 @@ void SinkStream::ProcessAudioOutAndRender(std::span<s16> output_buffer, std::siz
         return;
     }
 
-    // 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()};
-
-    // Core timing cannot be paused in single-core mode, so Stall ends up being called over and over
-    // and never recovers to a normal state, so just skip attempting to sync things on single-core.
-    if (system.IsMulticore() && !nvdec_active && queued_buffers > max_queue_size) {
-        Stall();
-    } else if (system.IsMulticore() && queued_buffers <= max_queue_size) {
-        Unstall();
-    }
-
     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) {
@@ -237,6 +222,10 @@ void SinkStream::ProcessAudioOutAndRender(std::span<s16> output_buffer, std::siz
             }
             // Successfully dequeued a new buffer.
             queued_buffers--;
+
+            { std::unique_lock lk{release_mutex}; }
+
+            release_cv.notify_one();
         }
 
         // Get the minimum frames available between the currently playing buffer, and the
@@ -248,6 +237,7 @@ void SinkStream::ProcessAudioOutAndRender(std::span<s16> output_buffer, std::siz
                            frames_available * frame_size);
 
         frames_written += frames_available;
+        actual_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
@@ -260,26 +250,29 @@ void SinkStream::ProcessAudioOutAndRender(std::span<s16> output_buffer, std::siz
     std::memcpy(&last_frame[0], &output_buffer[(frames_written - 1) * frame_size],
                 frame_size_bytes);
 
-    if (system.IsMulticore() && queued_buffers <= max_queue_size) {
-        Unstall();
+    {
+        std::scoped_lock lk{sample_count_lock};
+        last_sample_count_update_time =
+            Core::Timing::CyclesToUs(system.CoreTiming().GetClockTicks());
+        min_played_sample_count = max_played_sample_count;
+        max_played_sample_count += actual_frames_written;
     }
 }
 
-void SinkStream::Stall() {
-    std::scoped_lock lk{stall_guard};
-    if (stalled_lock) {
-        return;
-    }
-    stalled_lock = system.StallApplication();
+u64 SinkStream::GetExpectedPlayedSampleCount() {
+    std::scoped_lock lk{sample_count_lock};
+    auto cur_time{Core::Timing::CyclesToUs(system.CoreTiming().GetClockTicks())};
+    auto time_delta{cur_time - last_sample_count_update_time};
+    auto exp_played_sample_count{min_played_sample_count +
+                                 (TargetSampleRate * time_delta) / std::chrono::seconds{1}};
+
+    return std::min<u64>(exp_played_sample_count, max_played_sample_count);
 }
 
-void SinkStream::Unstall() {
-    std::scoped_lock lk{stall_guard};
-    if (!stalled_lock) {
-        return;
-    }
-    system.UnstallApplication();
-    stalled_lock.unlock();
+void SinkStream::WaitFreeSpace() {
+    std::unique_lock lk{release_mutex};
+    release_cv.wait(
+        lk, [this]() { return queued_buffers < max_queue_size || system.IsShuttingDown(); });
 }
 
 } // namespace AudioCore::Sink
diff --git a/src/audio_core/sink/sink_stream.h b/src/audio_core/sink/sink_stream.h
index 5fea72ab7..23e289c7b 100644
--- a/src/audio_core/sink/sink_stream.h
+++ b/src/audio_core/sink/sink_stream.h
@@ -5,6 +5,7 @@
 
 #include <array>
 #include <atomic>
+#include <chrono>
 #include <memory>
 #include <mutex>
 #include <span>
@@ -14,6 +15,7 @@
 #include "common/common_types.h"
 #include "common/reader_writer_queue.h"
 #include "common/ring_buffer.h"
+#include "common/thread.h"
 
 namespace Core {
 class System;
@@ -53,9 +55,7 @@ struct SinkBuffer {
 class SinkStream {
 public:
     explicit SinkStream(Core::System& system_, StreamType type_) : system{system_}, type{type_} {}
-    virtual ~SinkStream() {
-        Unstall();
-    }
+    virtual ~SinkStream() {}
 
     /**
      * Finalize the sink stream.
@@ -201,14 +201,16 @@ public:
     void ProcessAudioOutAndRender(std::span<s16> output_buffer, std::size_t num_frames);
 
     /**
-     * Stall core processes if the audio thread falls too far behind.
+     * Get the total number of samples expected to have been played by this stream.
+     *
+     * @return The number of samples.
      */
-    void Stall();
+    u64 GetExpectedPlayedSampleCount();
 
     /**
-     * Unstall core processes.
+     * Waits for free space in the sample ring buffer
      */
-    void Unstall();
+    void WaitFreeSpace();
 
 protected:
     /// Core system
@@ -237,12 +239,21 @@ private:
     std::atomic<u32> queued_buffers{};
     /// The ring size for audio out buffers (usually 4, rarely 2 or 8)
     u32 max_queue_size{};
+    /// Locks access to sample count tracking info
+    std::mutex sample_count_lock;
+    /// Minimum number of total samples that have been played since the last callback
+    u64 min_played_sample_count{};
+    /// Maximum number of total samples that can be played since the last callback
+    u64 max_played_sample_count{};
+    /// The time the two above tracking variables were last written to
+    std::chrono::microseconds last_sample_count_update_time{};
     /// 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};
-    std::mutex stall_guard;
-    std::unique_lock<std::mutex> stalled_lock;
+    /// Signalled when ring buffer entries are consumed
+    std::condition_variable release_cv;
+    std::mutex release_mutex;
 };
 
 using SinkStreamPtr = std::unique_ptr<SinkStream>;