early-access version 3381

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 3380.
+This is the source code for early-access 3381.
 
 ## Legal Notice
 

src/audio_core/device/audio_buffers.h

@@ -48,7 +48,7 @@ public:
      *
      * @param out_buffers - The buffers which were registered.
      */
-    void RegisterBuffers(std::vector<AudioBuffer>& out_buffers) {
+    void RegisterBuffers(std::span<AudioBuffer> out_buffers, u32& out_size) {
         std::scoped_lock l{lock};
         const s32 to_register{std::min(std::min(appended_count, BufferAppendLimit),
                                        BufferAppendLimit - registered_count)};
@@ -59,7 +59,7 @@ public:
                 index += N;
             }
 
-            out_buffers.push_back(buffers[index]);
+            out_buffers[out_size++] = buffers[index];
             registered_count++;
             registered_index = (registered_index + 1) % append_limit;
 
@@ -162,7 +162,7 @@ public:
      * @param max_buffers     - Maximum number of buffers to released.
      * @return The number of buffers released.
      */
-    u32 GetRegisteredAppendedBuffers(std::vector<AudioBuffer>& buffers_flushed, u32 max_buffers) {
+    u32 GetRegisteredAppendedBuffers(std::span<AudioBuffer> out_buffers, u32 max_buffers) {
         std::scoped_lock l{lock};
         if (registered_count + appended_count == 0) {
             return 0;
@@ -174,19 +174,20 @@ public:
             return 0;
         }
 
+        u32 buffers_flushed{0};
         while (registered_count > 0) {
             auto index{registered_index - registered_count};
             if (index < 0) {
                 index += N;
             }
 
-            buffers_flushed.push_back(buffers[index]);
+            out_buffers[buffers_flushed++] = buffers[index];
 
             registered_count--;
             released_count++;
             released_index = (released_index + 1) % append_limit;
 
-            if (buffers_flushed.size() >= buffers_to_flush) {
+            if (buffers_flushed >= buffers_to_flush) {
                 break;
             }
         }
@@ -197,18 +198,18 @@ public:
                 index += N;
             }
 
-            buffers_flushed.push_back(buffers[index]);
+            out_buffers[buffers_flushed++] = buffers[index];
 
             appended_count--;
             released_count++;
             released_index = (released_index + 1) % append_limit;
 
-            if (buffers_flushed.size() >= buffers_to_flush) {
+            if (buffers_flushed >= buffers_to_flush) {
                 break;
             }
         }
 
-        return static_cast<u32>(buffers_flushed.size());
+        return buffers_flushed;
     }
 
     /**
@@ -270,8 +271,9 @@ public:
      */
     bool FlushBuffers(u32& buffers_released) {
         std::scoped_lock l{lock};
-        std::vector<AudioBuffer> buffers_flushed{};
 
+        static Common::ScratchBuffer<AudioBuffer> buffers_flushed{};
+        buffers_flushed.resize_destructive(append_limit);
         buffers_released = GetRegisteredAppendedBuffers(buffers_flushed, append_limit);
 
         if (registered_count > 0) {

src/audio_core/device/device_session.cpp

@@ -6,6 +6,7 @@
 #include "audio_core/device/audio_buffer.h"
 #include "audio_core/device/device_session.h"
 #include "audio_core/sink/sink_stream.h"
+#include "common/scratch_buffer.h"
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "core/memory.h"
@@ -79,21 +80,22 @@ void DeviceSession::ClearBuffers() {
     }
 }
 
-void DeviceSession::AppendBuffers(std::span<const AudioBuffer> buffers) const {
+void DeviceSession::AppendBuffers(std::span<const AudioBuffer> buffers, u32 out_size) const {
-    for (const auto& buffer : buffers) {
+    static Common::ScratchBuffer<s16> samples{};
+
+    for (u32 i = 0; i < out_size; i++) {
         Sink::SinkBuffer new_buffer{
-            .frames = buffer.size / (channel_count * sizeof(s16)),
+            .frames = buffers[i].size / (channel_count * sizeof(s16)),
             .frames_played = 0,
-            .tag = buffer.tag,
+            .tag = buffers[i].tag,
             .consumed = false,
         };
 
         if (type == Sink::StreamType::In) {
-            std::vector<s16> samples{};
             stream->AppendBuffer(new_buffer, samples);
         } else {
-            std::vector<s16> samples(buffer.size / sizeof(s16));
+            samples.resize_destructive(buffers[i].size / sizeof(s16));
-            system.Memory().ReadBlockUnsafe(buffer.samples, samples.data(), buffer.size);
+            system.Memory().ReadBlockUnsafe(buffers[i].samples, samples.data(), buffers[i].size);
             stream->AppendBuffer(new_buffer, samples);
         }
     }

src/audio_core/device/device_session.h

@@ -62,7 +62,7 @@ public:
      *
      * @param buffers - The buffers to play.
      */
-    void AppendBuffers(std::span<const AudioBuffer> buffers) const;
+    void AppendBuffers(std::span<const AudioBuffer> buffers, u32 out_size) const;
 
     /**
      * (Audio In only) Pop samples from the backend, and write them back to this buffer's address.

src/audio_core/in/audio_in_system.cpp

@@ -89,9 +89,10 @@ Result System::Start() {
     session->Start();
     state = State::Started;
 
-    std::vector<AudioBuffer> buffers_to_flush{};
+    std::array<AudioBuffer, BufferAppendLimit> buffers_to_flush{};
-    buffers.RegisterBuffers(buffers_to_flush);
+    u32 out_size{0};
-    session->AppendBuffers(buffers_to_flush);
+    buffers.RegisterBuffers(buffers_to_flush, out_size);
+    session->AppendBuffers(buffers_to_flush, out_size);
     session->SetRingSize(static_cast<u32>(buffers_to_flush.size()));
 
     return ResultSuccess;
@@ -134,9 +135,10 @@ bool System::AppendBuffer(const AudioInBuffer& buffer, const u64 tag) {
 
 void System::RegisterBuffers() {
     if (state == State::Started) {
-        std::vector<AudioBuffer> registered_buffers{};
+        std::array<AudioBuffer, BufferAppendLimit> registered_buffers{};
-        buffers.RegisterBuffers(registered_buffers);
+        u32 out_size{0};
-        session->AppendBuffers(registered_buffers);
+        buffers.RegisterBuffers(registered_buffers, out_size);
+        session->AppendBuffers(registered_buffers, out_size);
     }
 }
 

src/audio_core/out/audio_out_system.cpp

@@ -89,9 +89,10 @@ Result System::Start() {
     session->Start();
     state = State::Started;
 
-    std::vector<AudioBuffer> buffers_to_flush{};
+    std::array<AudioBuffer, BufferAppendLimit> buffers_to_flush{};
-    buffers.RegisterBuffers(buffers_to_flush);
+    u32 out_size{0};
-    session->AppendBuffers(buffers_to_flush);
+    buffers.RegisterBuffers(buffers_to_flush, out_size);
+    session->AppendBuffers(buffers_to_flush, out_size);
     session->SetRingSize(static_cast<u32>(buffers_to_flush.size()));
 
     return ResultSuccess;
@@ -134,9 +135,10 @@ bool System::AppendBuffer(const AudioOutBuffer& buffer, u64 tag) {
 
 void System::RegisterBuffers() {
     if (state == State::Started) {
-        std::vector<AudioBuffer> registered_buffers{};
+        std::array<AudioBuffer, BufferAppendLimit> registered_buffers{};
-        buffers.RegisterBuffers(registered_buffers);
+        u32 out_size{0};
-        session->AppendBuffers(registered_buffers);
+        buffers.RegisterBuffers(registered_buffers, out_size);
+        session->AppendBuffers(registered_buffers, out_size);
     }
 }
 

src/audio_core/renderer/command/command_generator.cpp

@@ -46,7 +46,7 @@ void CommandGenerator::GenerateDataSourceCommand(VoiceInfo& voice_info,
             while (destination != nullptr) {
                 if (destination->IsConfigured()) {
                     auto mix_id{destination->GetMixId()};
-                    if (mix_id < mix_context.GetCount()) {
+                    if (mix_id < mix_context.GetCount() && mix_id != -1) {
                         auto mix_info{mix_context.GetInfo(mix_id)};
                         command_buffer.GenerateDepopPrepareCommand(
                             voice_info.node_id, voice_state, render_context.depop_buffer,

src/audio_core/renderer/command/data_source/decode.cpp

@@ -8,6 +8,7 @@
 #include "audio_core/renderer/command/resample/resample.h"
 #include "common/fixed_point.h"
 #include "common/logging/log.h"
+#include "common/scratch_buffer.h"
 #include "core/memory.h"
 
 namespace AudioCore::AudioRenderer {
@@ -29,6 +30,7 @@ static u32 DecodePcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
                      const DecodeArg& req) {
     constexpr s32 min{std::numeric_limits<s16>::min()};
     constexpr s32 max{std::numeric_limits<s16>::max()};
+    static Common::ScratchBuffer<T> samples;
 
     if (req.buffer == 0 || req.buffer_size == 0) {
         return 0;
@@ -49,7 +51,7 @@ static u32 DecodePcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
         const u64 size{channel_count * samples_to_decode};
         const u64 size_bytes{size * sizeof(T)};
 
-        std::vector<T> samples(size);
+        samples.resize_destructive(size);
         memory.ReadBlockUnsafe(source, samples.data(), size_bytes);
 
         if constexpr (std::is_floating_point_v<T>) {
@@ -73,7 +75,7 @@ static u32 DecodePcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
         }
 
         const VAddr source{req.buffer + ((req.start_offset + req.offset) * sizeof(T))};
-        std::vector<T> samples(samples_to_decode);
+        samples.resize_destructive(samples_to_decode);
         memory.ReadBlockUnsafe(source, samples.data(), samples_to_decode * sizeof(T));
 
         if constexpr (std::is_floating_point_v<T>) {
@@ -103,6 +105,7 @@ static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
                        const DecodeArg& req) {
     constexpr u32 SamplesPerFrame{14};
     constexpr u32 NibblesPerFrame{16};
+    static Common::ScratchBuffer<u8> wavebuffer;
 
     if (req.buffer == 0 || req.buffer_size == 0) {
         return 0;
@@ -138,7 +141,7 @@ static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
     }
 
     const auto size{std::max((samples_to_process / 8U) * SamplesPerFrame, 8U)};
-    std::vector<u8> wavebuffer(size);
+    wavebuffer.resize_destructive(size);
     memory.ReadBlockUnsafe(req.buffer + position_in_frame / 2, wavebuffer.data(),
                            wavebuffer.size());
 
@@ -227,6 +230,8 @@ static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
  * @param args   - The wavebuffer data, and information for how to decode it.
  */
 void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuffersArgs& args) {
+    Common::ScratchBuffer<s16> temp_buffer(TempBufferSize);
+
     auto& voice_state{*args.voice_state};
     auto remaining_sample_count{args.sample_count};
     auto fraction{voice_state.fraction};
@@ -256,9 +261,8 @@ void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuf
 
     bool is_buffer_starved{false};
     u32 offset{voice_state.offset};
-
+    std::memset(temp_buffer.data(), 0, temp_buffer.size() * sizeof(s16));
     auto output_buffer{args.output};
-    std::vector<s16> temp_buffer(TempBufferSize, 0);
 
     while (remaining_sample_count > 0) {
         const auto samples_to_write{std::min(remaining_sample_count, max_remaining_sample_count)};

src/audio_core/renderer/command/effect/aux_.cpp

@@ -4,6 +4,7 @@
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "audio_core/renderer/command/effect/aux_.h"
 #include "audio_core/renderer/effect/aux_.h"
+#include "core/core.h"
 #include "core/memory.h"
 
 namespace AudioCore::AudioRenderer {
@@ -174,6 +175,19 @@ void AuxCommand::Dump([[maybe_unused]] const ADSP::CommandListProcessor& process
 }
 
 void AuxCommand::Process(const ADSP::CommandListProcessor& processor) {
+    // HACK!
+    // Ignore aux for Super Mario Odyssey and Metroid Prime Remastered.
+    // For some reason these games receive output samples, and then send them back in as input
+    // again. Problem is the data being sent back in is slightly offset from the current output by
+    // 240 or 480 samples, leading to a very fast echoing effect, which should not be there.
+    // Timing issue or some bug in the code?
+    // We can't disable this unconditionally as some games rely on it for synchronisation and will
+    // softlock without it (Age of Calamity).
+    const auto program_id = processor.system->GetCurrentProcessProgramID();
+    if (program_id == 0x0100000000010000ull || program_id == 0x010012101468C000ull) {
+        return;
+    }
+
     auto input_buffer{
         processor.mix_buffers.subspan(input * processor.sample_count, processor.sample_count)};
     auto output_buffer{

src/audio_core/renderer/command/effect/biquad_filter.cpp

@@ -4,6 +4,7 @@
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "audio_core/renderer/command/effect/biquad_filter.h"
 #include "audio_core/renderer/voice/voice_state.h"
+#include "common/bit_cast.h"
 
 namespace AudioCore::AudioRenderer {
 /**
@@ -19,21 +20,21 @@ namespace AudioCore::AudioRenderer {
 void ApplyBiquadFilterFloat(std::span<s32> output, std::span<const s32> input,
                             std::array<s16, 3>& b_, std::array<s16, 2>& a_,
                             VoiceState::BiquadFilterState& state, const u32 sample_count) {
-    constexpr s64 min{std::numeric_limits<s32>::min()};
+    constexpr f64 min{std::numeric_limits<s32>::min()};
-    constexpr s64 max{std::numeric_limits<s32>::max()};
+    constexpr f64 max{std::numeric_limits<s32>::max()};
     std::array<f64, 3> b{Common::FixedPoint<50, 14>::from_base(b_[0]).to_double(),
                          Common::FixedPoint<50, 14>::from_base(b_[1]).to_double(),
                          Common::FixedPoint<50, 14>::from_base(b_[2]).to_double()};
     std::array<f64, 2> a{Common::FixedPoint<50, 14>::from_base(a_[0]).to_double(),
                          Common::FixedPoint<50, 14>::from_base(a_[1]).to_double()};
-    std::array<f64, 4> s{state.s0.to_double(), state.s1.to_double(), state.s2.to_double(),
+    std::array<f64, 4> s{Common::BitCast<f64>(state.s0), Common::BitCast<f64>(state.s1),
-                         state.s3.to_double()};
+                         Common::BitCast<f64>(state.s2), Common::BitCast<f64>(state.s3)};
 
     for (u32 i = 0; i < sample_count; i++) {
         f64 in_sample{static_cast<f64>(input[i])};
         auto sample{in_sample * b[0] + s[0] * b[1] + s[1] * b[2] + s[2] * a[0] + s[3] * a[1]};
 
-        output[i] = static_cast<s32>(std::clamp(static_cast<s64>(sample), min, max));
+        output[i] = static_cast<s32>(std::clamp(sample, min, max));
 
         s[1] = s[0];
         s[0] = in_sample;
@@ -41,10 +42,10 @@ void ApplyBiquadFilterFloat(std::span<s32> output, std::span<const s32> input,
         s[2] = sample;
     }
 
-    state.s0 = s[0];
+    state.s0 = Common::BitCast<s64>(s[0]);
-    state.s1 = s[1];
+    state.s1 = Common::BitCast<s64>(s[1]);
-    state.s2 = s[2];
+    state.s2 = Common::BitCast<s64>(s[2]);
-    state.s3 = s[3];
+    state.s3 = Common::BitCast<s64>(s[3]);
 }
 
 /**
@@ -58,29 +59,20 @@ void ApplyBiquadFilterFloat(std::span<s32> output, std::span<const s32> input,
  * @param sample_count - Number of samples to process.
  */
 static void ApplyBiquadFilterInt(std::span<s32> output, std::span<const s32> input,
-                                 std::array<s16, 3>& b_, std::array<s16, 2>& a_,
+                                 std::array<s16, 3>& b, std::array<s16, 2>& a,
                                  VoiceState::BiquadFilterState& state, const u32 sample_count) {
     constexpr s64 min{std::numeric_limits<s32>::min()};
     constexpr s64 max{std::numeric_limits<s32>::max()};
-    std::array<Common::FixedPoint<50, 14>, 3> b{
-        Common::FixedPoint<50, 14>::from_base(b_[0]),
-        Common::FixedPoint<50, 14>::from_base(b_[1]),
-        Common::FixedPoint<50, 14>::from_base(b_[2]),
-    };
-    std::array<Common::FixedPoint<50, 14>, 3> a{
-        Common::FixedPoint<50, 14>::from_base(a_[0]),
-        Common::FixedPoint<50, 14>::from_base(a_[1]),
-    };
 
     for (u32 i = 0; i < sample_count; i++) {
-        s64 in_sample{input[i]};
+        const s64 in_sample{input[i]};
-        auto sample{in_sample * b[0] + state.s0};
+        const s64 sample{in_sample * b[0] + state.s0};
-        const auto out_sample{std::clamp(sample.to_long(), min, max)};
+        const s64 out_sample{std::clamp<s64>((sample + (1 << 13)) >> 14, min, max)};
 
         output[i] = static_cast<s32>(out_sample);
 
         state.s0 = state.s1 + b[1] * in_sample + a[0] * out_sample;
-        state.s1 = 0 + b[2] * in_sample + a[1] * out_sample;
+        state.s1 = b[2] * in_sample + a[1] * out_sample;
     }
 }
 

src/audio_core/renderer/command/effect/compressor.cpp

@@ -8,6 +8,7 @@
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "audio_core/renderer/command/effect/compressor.h"
 #include "audio_core/renderer/effect/compressor.h"
+#include "common/scratch_buffer.h"
 
 namespace AudioCore::AudioRenderer {
 
@@ -44,8 +45,8 @@ static void InitializeCompressorEffect(const CompressorInfo::ParameterVersion2&
 
 static void ApplyCompressorEffect(const CompressorInfo::ParameterVersion2& params,
                                   CompressorInfo::State& state, bool enabled,
-                                  std::vector<std::span<const s32>> input_buffers,
+                                  std::span<std::span<const s32>> input_buffers,
-                                  std::vector<std::span<s32>> output_buffers, u32 sample_count) {
+                                  std::span<std::span<s32>> output_buffers, u32 sample_count) {
     if (enabled) {
         auto state_00{state.unk_00};
         auto state_04{state.unk_04};
@@ -124,8 +125,10 @@ void CompressorCommand::Dump([[maybe_unused]] const ADSP::CommandListProcessor&
 }
 
 void CompressorCommand::Process(const ADSP::CommandListProcessor& processor) {
-    std::vector<std::span<const s32>> input_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<const s32>> input_buffers{};
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<s32>> output_buffers{};
+    input_buffers.resize_destructive(parameter.channel_count);
+    output_buffers.resize_destructive(parameter.channel_count);
 
     for (s16 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,

src/audio_core/renderer/command/effect/delay.cpp

@@ -3,6 +3,7 @@
 
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "audio_core/renderer/command/effect/delay.h"
+#include "common/scratch_buffer.h"
 
 namespace AudioCore::AudioRenderer {
 /**
@@ -74,8 +75,8 @@ static void InitializeDelayEffect(const DelayInfo::ParameterVersion1& params,
  */
 template <size_t NumChannels>
 static void ApplyDelay(const DelayInfo::ParameterVersion1& params, DelayInfo::State& state,
-                       std::vector<std::span<const s32>>& inputs,
+                       std::span<std::span<const s32>> inputs, std::span<std::span<s32>> outputs,
-                       std::vector<std::span<s32>>& outputs, const u32 sample_count) {
+                       const u32 sample_count) {
     for (u32 sample_index = 0; sample_index < sample_count; sample_index++) {
         std::array<Common::FixedPoint<50, 14>, NumChannels> input_samples{};
         for (u32 channel = 0; channel < NumChannels; channel++) {
@@ -153,8 +154,8 @@ static void ApplyDelay(const DelayInfo::ParameterVersion1& params, DelayInfo::St
  * @param sample_count - Number of samples to process.
  */
 static void ApplyDelayEffect(const DelayInfo::ParameterVersion1& params, DelayInfo::State& state,
-                             const bool enabled, std::vector<std::span<const s32>>& inputs,
+                             const bool enabled, std::span<std::span<const s32>> inputs,
-                             std::vector<std::span<s32>>& outputs, const u32 sample_count) {
+                             std::span<std::span<s32>> outputs, const u32 sample_count) {
 
     if (!IsChannelCountValid(params.channel_count)) {
         LOG_ERROR(Service_Audio, "Invalid delay channels {}", params.channel_count);
@@ -208,8 +209,10 @@ void DelayCommand::Dump([[maybe_unused]] const ADSP::CommandListProcessor& proce
 }
 
 void DelayCommand::Process(const ADSP::CommandListProcessor& processor) {
-    std::vector<std::span<const s32>> input_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<const s32>> input_buffers{};
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<s32>> output_buffers{};
+    input_buffers.resize_destructive(parameter.channel_count);
+    output_buffers.resize_destructive(parameter.channel_count);
 
     for (s16 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,

src/audio_core/renderer/command/effect/i3dl2_reverb.cpp

@@ -6,6 +6,7 @@
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "audio_core/renderer/command/effect/i3dl2_reverb.h"
 #include "common/polyfill_ranges.h"
+#include "common/scratch_buffer.h"
 
 namespace AudioCore::AudioRenderer {
 
@@ -408,8 +409,10 @@ void I3dl2ReverbCommand::Dump([[maybe_unused]] const ADSP::CommandListProcessor&
 }
 
 void I3dl2ReverbCommand::Process(const ADSP::CommandListProcessor& processor) {
-    std::vector<std::span<const s32>> input_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<const s32>> input_buffers{};
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<s32>> output_buffers{};
+    input_buffers.resize_destructive(parameter.channel_count);
+    output_buffers.resize_destructive(parameter.channel_count);
 
     for (u32 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,

src/audio_core/renderer/command/effect/light_limiter.cpp

@@ -3,6 +3,7 @@
 
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "audio_core/renderer/command/effect/light_limiter.h"
+#include "common/scratch_buffer.h"
 
 namespace AudioCore::AudioRenderer {
 /**
@@ -47,8 +48,8 @@ static void InitializeLightLimiterEffect(const LightLimiterInfo::ParameterVersio
  */
 static void ApplyLightLimiterEffect(const LightLimiterInfo::ParameterVersion2& params,
                                     LightLimiterInfo::State& state, const bool enabled,
-                                    std::vector<std::span<const s32>>& inputs,
+                                    std::span<std::span<const s32>> inputs,
-                                    std::vector<std::span<s32>>& outputs, const u32 sample_count,
+                                    std::span<std::span<s32>> outputs, const u32 sample_count,
                                     LightLimiterInfo::StatisticsInternal* statistics) {
     constexpr s64 min{std::numeric_limits<s32>::min()};
     constexpr s64 max{std::numeric_limits<s32>::max()};
@@ -147,8 +148,10 @@ void LightLimiterVersion1Command::Dump([[maybe_unused]] const ADSP::CommandListP
 }
 
 void LightLimiterVersion1Command::Process(const ADSP::CommandListProcessor& processor) {
-    std::vector<std::span<const s32>> input_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<const s32>> input_buffers{};
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<s32>> output_buffers{};
+    input_buffers.resize_destructive(parameter.channel_count);
+    output_buffers.resize_destructive(parameter.channel_count);
 
     for (u32 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,
@@ -190,8 +193,10 @@ void LightLimiterVersion2Command::Dump([[maybe_unused]] const ADSP::CommandListP
 }
 
 void LightLimiterVersion2Command::Process(const ADSP::CommandListProcessor& processor) {
-    std::vector<std::span<const s32>> input_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<const s32>> input_buffers{};
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<s32>> output_buffers{};
+    input_buffers.resize_destructive(parameter.channel_count);
+    output_buffers.resize_destructive(parameter.channel_count);
 
     for (u32 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,

src/audio_core/renderer/command/effect/reverb.cpp

@@ -7,6 +7,7 @@
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "audio_core/renderer/command/effect/reverb.h"
 #include "common/polyfill_ranges.h"
+#include "common/scratch_buffer.h"
 
 namespace AudioCore::AudioRenderer {
 
@@ -250,8 +251,8 @@ static Common::FixedPoint<50, 14> Axfx2AllPassTick(ReverbInfo::ReverbDelayLine&
  */
 template <size_t NumChannels>
 static void ApplyReverbEffect(const ReverbInfo::ParameterVersion2& params, ReverbInfo::State& state,
-                              std::vector<std::span<const s32>>& inputs,
+                              std::span<std::span<const s32>> inputs,
-                              std::vector<std::span<s32>>& outputs, const u32 sample_count) {
+                              std::span<std::span<s32>> outputs, const u32 sample_count) {
     constexpr std::array<u8, ReverbInfo::MaxDelayTaps> OutTapIndexes1Ch{
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
     };
@@ -368,8 +369,8 @@ static void ApplyReverbEffect(const ReverbInfo::ParameterVersion2& params, Rever
  * @param sample_count - Number of samples to process.
  */
 static void ApplyReverbEffect(const ReverbInfo::ParameterVersion2& params, ReverbInfo::State& state,
-                              const bool enabled, std::vector<std::span<const s32>>& inputs,
+                              const bool enabled, std::span<std::span<const s32>> inputs,
-                              std::vector<std::span<s32>>& outputs, const u32 sample_count) {
+                              std::span<std::span<s32>> outputs, const u32 sample_count) {
     if (enabled) {
         switch (params.channel_count) {
         case 0:
@@ -411,8 +412,10 @@ void ReverbCommand::Dump([[maybe_unused]] const ADSP::CommandListProcessor& proc
 }
 
 void ReverbCommand::Process(const ADSP::CommandListProcessor& processor) {
-    std::vector<std::span<const s32>> input_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<const s32>> input_buffers{};
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    static Common::ScratchBuffer<std::span<s32>> output_buffers{};
+    input_buffers.resize_destructive(parameter.channel_count);
+    output_buffers.resize_destructive(parameter.channel_count);
 
     for (u32 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,

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

@@ -5,6 +5,7 @@
 
 #include "audio_core/renderer/adsp/command_list_processor.h"
 #include "audio_core/renderer/command/sink/circular_buffer.h"
+#include "common/scratch_buffer.h"
 #include "core/memory.h"
 
 namespace AudioCore::AudioRenderer {
@@ -24,7 +25,9 @@ void CircularBufferSinkCommand::Process(const ADSP::CommandListProcessor& proces
     constexpr s32 min{std::numeric_limits<s16>::min()};
     constexpr s32 max{std::numeric_limits<s16>::max()};
 
-    std::vector<s16> output(processor.sample_count);
+    static Common::ScratchBuffer<s16> output{};
+    output.resize_destructive(processor.sample_count);
+
     for (u32 channel = 0; channel < input_count; channel++) {
         auto input{processor.mix_buffers.subspan(inputs[channel] * processor.sample_count,
                                                  processor.sample_count)};

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

@@ -33,7 +33,8 @@ void DeviceSinkCommand::Process(const ADSP::CommandListProcessor& processor) {
         .consumed{false},
     };
 
-    std::vector<s16> samples(out_buffer.frames * input_count);
+    static Common::ScratchBuffer<s16> samples{};
+    samples.resize_destructive(out_buffer.frames * input_count);
 
     for (u32 channel = 0; channel < input_count; channel++) {
         const auto offset{inputs[channel] * out_buffer.frames};

src/audio_core/renderer/voice/voice_state.h

@@ -19,10 +19,10 @@ struct VoiceState {
      * State of the voice's biquad filter.
      */
     struct BiquadFilterState {
-        Common::FixedPoint<50, 14> s0;
+        s64 s0;
-        Common::FixedPoint<50, 14> s1;
+        s64 s1;
-        Common::FixedPoint<50, 14> s2;
+        s64 s2;
-        Common::FixedPoint<50, 14> s3;
+        s64 s3;
     };
 
     /**

src/audio_core/sink/null_sink.h

@@ -9,6 +9,7 @@
 
 #include "audio_core/sink/sink.h"
 #include "audio_core/sink/sink_stream.h"
+#include "common/scratch_buffer.h"
 
 namespace Core {
 class System;
@@ -20,7 +21,7 @@ public:
     explicit NullSinkStreamImpl(Core::System& system_, StreamType type_)
         : SinkStream{system_, type_} {}
     ~NullSinkStreamImpl() override {}
-    void AppendBuffer(SinkBuffer&, std::vector<s16>&) override {}
+    void AppendBuffer(SinkBuffer&, Common::ScratchBuffer<s16>&) override {}
     std::vector<s16> ReleaseBuffer(u64) override {
         return {};
     }

src/audio_core/sink/sink_stream.cpp

@@ -17,7 +17,7 @@
 
 namespace AudioCore::Sink {
 
-void SinkStream::AppendBuffer(SinkBuffer& buffer, std::vector<s16>& samples) {
+void SinkStream::AppendBuffer(SinkBuffer& buffer, Common::ScratchBuffer<s16>& samples) {
     if (type == StreamType::In) {
         queue.enqueue(buffer);
         queued_buffers++;
@@ -71,7 +71,9 @@ void SinkStream::AppendBuffer(SinkBuffer& buffer, std::vector<s16>& samples) {
         // 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);
+        static Common::ScratchBuffer<s16> new_samples{};
+        new_samples.resize_destructive(samples.size() / system_channels * device_channels);
+        std::memset(new_samples.data(), 0, new_samples.size() * sizeof(s16));
 
         for (u32 read_index = 0, write_index = 0; read_index < samples.size();
              read_index += system_channels, write_index += device_channels) {
@@ -100,7 +102,7 @@ void SinkStream::AppendBuffer(SinkBuffer& buffer, std::vector<s16>& samples) {
         }
     }
 
-    samples_buffer.Push(samples);
+    samples_buffer.Push(samples.data(), samples.size());
     queue.enqueue(buffer);
     queued_buffers++;
 }

src/audio_core/sink/sink_stream.h

@@ -14,6 +14,7 @@
 #include "common/common_types.h"
 #include "common/reader_writer_queue.h"
 #include "common/ring_buffer.h"
+#include "common/scratch_buffer.h"
 
 namespace Core {
 class System;
@@ -169,7 +170,7 @@ public:
      * @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);
+    virtual void AppendBuffer(SinkBuffer& buffer, Common::ScratchBuffer<s16>& samples);
 
     /**
      * Release a buffer. Audio In only, will fill a buffer with recorded samples.

src/common/scratch_buffer.h

@@ -24,6 +24,24 @@ public:
 
     ~ScratchBuffer() = default;
 
+    ScratchBuffer(ScratchBuffer&& rhs) {
+        last_requested_size = rhs.last_requested_size;
+        buffer_capacity = rhs.buffer_capacity;
+        buffer = std::move(rhs.buffer);
+
+        rhs.last_requested_size = 0;
+        rhs.buffer_capacity = 0;
+    }
+
+    void operator=(ScratchBuffer&& rhs) {
+        last_requested_size = rhs.last_requested_size;
+        buffer_capacity = rhs.buffer_capacity;
+        buffer = std::move(rhs.buffer);
+
+        rhs.last_requested_size = 0;
+        rhs.buffer_capacity = 0;
+    }
+
     /// This will only grow the buffer's capacity if size is greater than the current capacity.
     /// The previously held data will remain intact.
     void resize(size_t size) {