audio_core: No longer stall when sink queue is full

Now the audout and audren update rates are tied to the sink status stalling is no longer necessary.

CMakeLists.txt

@@ -210,7 +210,7 @@ set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/bin)
 # =======================================================================
 
 # Enforce the search mode of non-required packages for better and shorter failure messages
-find_package(Boost 1.81.0 REQUIRED context)
+find_package(Boost 1.79.0 REQUIRED context)
 find_package(enet 1.3 MODULE)
 find_package(fmt 9 REQUIRED)
 find_package(inih 52 MODULE COMPONENTS INIReader)

src/audio_core/device/device_session.cpp

@@ -121,8 +121,7 @@ u64 DeviceSession::GetPlayedSampleCount() const {
 }
 
 std::optional<std::chrono::nanoseconds> DeviceSession::ThreadFunc() {
-    // Add 5ms of samples at a 48K sample rate.
-    played_sample_count += 48'000 * INCREMENT_TIME / 1s;
+    played_sample_count = stream->GetExpectedPlayedSampleCount();
     if (type == Sink::StreamType::Out) {
         system.AudioCore().GetAudioManager().SetEvent(Event::Type::AudioOutManager, true);
     } else {

src/audio_core/renderer/adsp/audio_renderer.cpp

@@ -189,6 +189,8 @@ void AudioRenderer::ThreadFunc() {
                     max_time = std::min(command_buffer.time_limit, max_time);
                     command_list_processor.SetProcessTimeMax(max_time);
 
+                    streams[index]->WaitFreeSpace();
+
                     // Process the command list
                     {
                         MICROPROFILE_SCOPE(Audio_Renderer);

src/audio_core/renderer/adsp/audio_renderer.h

@@ -10,6 +10,7 @@
 #include "audio_core/renderer/adsp/command_buffer.h"
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "common/common_types.h"
+#include "common/polyfill_thread.h"
 #include "common/reader_writer_queue.h"
 #include "common/thread.h"
 

src/audio_core/renderer/system_manager.cpp

@@ -15,14 +15,9 @@ MICROPROFILE_DEFINE(Audio_RenderSystemManager, "Audio", "Render System Manager",
                     MP_RGB(60, 19, 97));
 
 namespace AudioCore::AudioRenderer {
-constexpr std::chrono::nanoseconds RENDER_TIME{5'000'000UL};
 
 SystemManager::SystemManager(Core::System& core_)
-    : core{core_}, adsp{core.AudioCore().GetADSP()}, mailbox{adsp.GetRenderMailbox()},
-      thread_event{Core::Timing::CreateEvent(
-          "AudioRendererSystemManager", [this](std::uintptr_t, s64 time, std::chrono::nanoseconds) {
-              return ThreadFunc2(time);
-          })} {}
+    : core{core_}, adsp{core.AudioCore().GetADSP()}, mailbox{adsp.GetRenderMailbox()} {}
 
 SystemManager::~SystemManager() {
     Stop();
@@ -33,8 +28,6 @@ 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), RENDER_TIME,
-                                                   thread_event);
         }
     }
 
@@ -45,7 +38,6 @@ void SystemManager::Stop() {
     if (!active) {
         return;
     }
-    core.CoreTiming().UnscheduleEvent(thread_event, {});
     active = false;
     update.store(true);
     update.notify_all();
@@ -111,16 +103,7 @@ void SystemManager::ThreadFunc() {
 
         adsp.Signal();
         adsp.Wait();
-
-        update.wait(false);
-        update.store(false);
     }
 }
 
-std::optional<std::chrono::nanoseconds> SystemManager::ThreadFunc2(s64 time) {
-    update.store(true);
-    update.notify_all();
-    return std::nullopt;
-}
-
 } // namespace AudioCore::AudioRenderer

src/audio_core/renderer/system_manager.h

@@ -68,11 +68,6 @@ private:
      */
     void ThreadFunc();
 
-    /**
-     * Signalling core timing thread to run ThreadFunc.
-     */
-    std::optional<std::chrono::nanoseconds> ThreadFunc2(s64 time);
-
     enum class StreamState {
         Filling,
         Steady,
@@ -95,8 +90,6 @@ private:
     ADSP::ADSP& adsp;
     /// AudioRenderer mailbox for communication
     ADSP::AudioRenderer_Mailbox* mailbox{};
-    /// Core timing event to signal main thread
-    std::shared_ptr<Core::Timing::EventType> thread_event;
     /// Atomic for main thread to wait on
     std::atomic<bool> update{};
 };

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;
         }

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;
         }

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

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>;

src/common/CMakeLists.txt

@@ -38,6 +38,7 @@ add_library(common STATIC
     common_precompiled_headers.h
     common_types.h
     concepts.h
+    container_hash.h
     demangle.cpp
     demangle.h
     div_ceil.h

src/common/container_hash.h

@@ -0,0 +1,92 @@
+// SPDX-FileCopyrightText: 2005-2014 Daniel James
+// SPDX-FileCopyrightText: 2016 Austin Appleby
+// SPDX-License-Identifier: BSL-1.0
+
+#include <array>
+#include <climits>
+#include <cstdint>
+#include <limits>
+#include <type_traits>
+#include <vector>
+
+namespace Common {
+
+namespace detail {
+
+template <typename T>
+    requires std::is_unsigned_v<T>
+inline std::size_t HashValue(T val) {
+    const unsigned int size_t_bits = std::numeric_limits<std::size_t>::digits;
+    const unsigned int length =
+        (std::numeric_limits<T>::digits - 1) / static_cast<unsigned int>(size_t_bits);
+
+    std::size_t seed = 0;
+
+    for (unsigned int i = length * size_t_bits; i > 0; i -= size_t_bits) {
+        seed ^= static_cast<size_t>(val >> i) + (seed << 6) + (seed >> 2);
+    }
+
+    seed ^= static_cast<size_t>(val) + (seed << 6) + (seed >> 2);
+
+    return seed;
+}
+
+template <size_t Bits>
+struct HashCombineImpl {
+    template <typename T>
+    static inline T fn(T seed, T value) {
+        seed ^= value + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+        return seed;
+    }
+};
+
+template <>
+struct HashCombineImpl<64> {
+    static inline std::uint64_t fn(std::uint64_t h, std::uint64_t k) {
+        const std::uint64_t m = (std::uint64_t(0xc6a4a793) << 32) + 0x5bd1e995;
+        const int r = 47;
+
+        k *= m;
+        k ^= k >> r;
+        k *= m;
+
+        h ^= k;
+        h *= m;
+
+        // Completely arbitrary number, to prevent 0's
+        // from hashing to 0.
+        h += 0xe6546b64;
+
+        return h;
+    }
+};
+
+} // namespace detail
+
+template <typename T>
+inline void HashCombine(std::size_t& seed, const T& v) {
+    seed = detail::HashCombineImpl<sizeof(std::size_t) * CHAR_BIT>::fn(seed, detail::HashValue(v));
+}
+
+template <typename It>
+inline std::size_t HashRange(It first, It last) {
+    std::size_t seed = 0;
+
+    for (; first != last; ++first) {
+        HashCombine<typename std::iterator_traits<It>::value_type>(seed, *first);
+    }
+
+    return seed;
+}
+
+template <typename T, size_t Size>
+std::size_t HashValue(const std::array<T, Size>& v) {
+    return HashRange(v.cbegin(), v.cend());
+}
+
+template <typename T, typename Allocator>
+std::size_t HashValue(const std::vector<T, Allocator>& v) {
+    return HashRange(v.cbegin(), v.cend());
+}
+
+} // namespace Common

src/tests/CMakeLists.txt

@@ -4,6 +4,7 @@
 add_executable(tests
     common/bit_field.cpp
     common/cityhash.cpp
+    common/container_hash.cpp
     common/fibers.cpp
     common/host_memory.cpp
     common/param_package.cpp

src/tests/common/container_hash.cpp

@@ -0,0 +1,44 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <catch2/catch_test_macros.hpp>
+
+#include "common/common_types.h"
+#include "common/container_hash.h"
+
+TEST_CASE("ContainerHash", "[common]") {
+    constexpr std::array<u8, 32> U8Values{
+        114, 10, 238, 189, 199, 242, 86, 96, 53,  193, 195, 247, 249, 56, 253, 61,
+        205, 3,  172, 4,   210, 197, 43, 72, 103, 8,   99,  89,  5,   97, 68,  196,
+    };
+    constexpr std::array<u16, 32> U16Values{
+        61586, 49151, 3313,  11641, 31695, 54795, 46764, 20965, 23287, 14039, 19265,
+        49093, 58932, 22518, 27139, 42825, 57417, 54237, 48057, 14586, 42813, 32994,
+        33970, 45501, 5619,  15895, 33227, 27509, 25391, 37275, 60218, 17599,
+    };
+    constexpr std::array<u32, 32> U32Values{
+        3838402410U, 2029146863U, 1730869921U, 985528872U,  186773874U,  2094639868U, 3324775932U,
+        1795512424U, 2571165571U, 3256934519U, 2358691590U, 2752682538U, 1484336451U, 378124520U,
+        3463015699U, 3395942161U, 1263211979U, 3473632889U, 3039822212U, 2068707357U, 2223837919U,
+        1823232191U, 1583884041U, 1264393380U, 4087566993U, 3188607101U, 3933680362U, 1464520765U,
+        1786838406U, 1311734848U, 2773642241U, 3993641692U,
+    };
+    constexpr std::array<u64, 32> U64Values{
+        5908025796157537817ULL, 10947547850358315100ULL, 844798943576724669ULL,
+        7999662937458523703ULL, 4006550374705895164ULL,  1832550525423503632ULL,
+        9323088254855830976ULL, 12028890075598379412ULL, 6021511300787826236ULL,
+        7864675007938747948ULL, 18099387408859708806ULL, 6438638299316820708ULL,
+        9029399285648501543ULL, 18195459433089960253ULL, 17214335092761966083ULL,
+        5549347964591337833ULL, 14899526073304962015ULL, 5058883181561464475ULL,
+        7436311795731206973ULL, 7535129567768649864ULL,  1287169596809258072ULL,
+        8237671246353565927ULL, 1715230541978016153ULL,  8443157615068813300ULL,
+        6098675262328527839ULL, 704652094100376853ULL,   1303411723202926503ULL,
+        7808312933946424854ULL, 6863726670433556594ULL,  9870361541383217495ULL,
+        9273671094091079488ULL, 17541434976160119010ULL,
+    };
+
+    REQUIRE(Common::HashValue(U8Values) == 5867183267093890552ULL);
+    REQUIRE(Common::HashValue(U16Values) == 9594135570564347135ULL);
+    REQUIRE(Common::HashValue(U32Values) == 13123757214696618460ULL);
+    REQUIRE(Common::HashValue(U64Values) == 7296500016546938380ULL);
+}

src/video_core/macro/macro.cpp

@@ -6,7 +6,7 @@
 #include <optional>
 #include <span>
 
-#include <boost/container_hash/hash.hpp>
+#include "common/container_hash.h"
 
 #include <fstream>
 #include "common/assert.h"
@@ -89,7 +89,7 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
 
         if (!mid_method.has_value()) {
             cache_info.lle_program = Compile(macro_code->second);
-            cache_info.hash = boost::hash_value(macro_code->second);
+            cache_info.hash = Common::HashValue(macro_code->second);
             if (Settings::values.dump_macros) {
                 Dump(cache_info.hash, macro_code->second);
             }
@@ -100,7 +100,7 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
             code.resize(macro_cached.size() - rebased_method);
             std::memcpy(code.data(), macro_cached.data() + rebased_method,
                         code.size() * sizeof(u32));
-            cache_info.hash = boost::hash_value(code);
+            cache_info.hash = Common::HashValue(code);
             cache_info.lle_program = Compile(code);
             if (Settings::values.dump_macros) {
                 Dump(cache_info.hash, code);

vcpkg.json

@@ -1,7 +1,7 @@
 {
     "$schema": "https://raw.githubusercontent.com/microsoft/vcpkg-tool/main/docs/vcpkg.schema.json",
     "name": "yuzu",
-    "builtin-baseline": "a7b6122f6b6504d16d96117336a0562693579933",
+    "builtin-baseline": "acc3bcf76b84ae5041c86ab55fe138ae7b8255c7",
     "version": "1.0",
     "dependencies": [
         "boost-algorithm",