Merge pull request #10457 from Kelebek1/optimise

Remove memory allocations in some hot paths

12 files changed, 47 insertions, 48 deletions

diff --git a/src/audio_core/renderer/command/data_source/decode.cpp b/src/audio_core/renderer/command/data_source/decode.cpp
index ff5d31bd6..f45933203 100644
--- a/src/audio_core/renderer/command/data_source/decode.cpp
+++ b/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 {
@@ -27,6 +28,7 @@ constexpr std::array<u8, 3> PitchBySrcQuality = {4, 8, 4};
 template <typename T>
 static u32 DecodePcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
                      const DecodeArg& req) {
+    std::array<T, TempBufferSize> tmp_samples{};
     constexpr s32 min{std::numeric_limits<s16>::min()};
     constexpr s32 max{std::numeric_limits<s16>::max()};
 
@@ -49,18 +51,17 @@ 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);
-        memory.ReadBlockUnsafe(source, samples.data(), size_bytes);
+        memory.ReadBlockUnsafe(source, tmp_samples.data(), size_bytes);
 
         if constexpr (std::is_floating_point_v<T>) {
             for (u32 i = 0; i < samples_to_decode; i++) {
-                auto sample{static_cast<s32>(samples[i * channel_count + req.target_channel] *
+                auto sample{static_cast<s32>(tmp_samples[i * channel_count + req.target_channel] *
                                              std::numeric_limits<s16>::max())};
                 out_buffer[i] = static_cast<s16>(std::clamp(sample, min, max));
             }
         } else {
             for (u32 i = 0; i < samples_to_decode; i++) {
-                out_buffer[i] = samples[i * channel_count + req.target_channel];
+                out_buffer[i] = tmp_samples[i * channel_count + req.target_channel];
             }
         }
     } break;
@@ -73,17 +74,16 @@ 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);
-        memory.ReadBlockUnsafe(source, samples.data(), samples_to_decode * sizeof(T));
+        memory.ReadBlockUnsafe(source, tmp_samples.data(), samples_to_decode * sizeof(T));
 
         if constexpr (std::is_floating_point_v<T>) {
             for (u32 i = 0; i < samples_to_decode; i++) {
-                auto sample{static_cast<s32>(samples[i * channel_count + req.target_channel] *
+                auto sample{static_cast<s32>(tmp_samples[i * channel_count + req.target_channel] *
                                              std::numeric_limits<s16>::max())};
                 out_buffer[i] = static_cast<s16>(std::clamp(sample, min, max));
             }
         } else {
-            std::memcpy(out_buffer.data(), samples.data(), samples_to_decode * sizeof(s16));
+            std::memcpy(out_buffer.data(), tmp_samples.data(), samples_to_decode * sizeof(s16));
         }
         break;
     }
@@ -101,6 +101,7 @@ static u32 DecodePcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
  */
 static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
                        const DecodeArg& req) {
+    std::array<u8, TempBufferSize> wavebuffer{};
     constexpr u32 SamplesPerFrame{14};
     constexpr u32 NibblesPerFrame{16};
 
@@ -138,9 +139,7 @@ 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);
-    memory.ReadBlockUnsafe(req.buffer + position_in_frame / 2, wavebuffer.data(),
-                           wavebuffer.size());
+    memory.ReadBlockUnsafe(req.buffer + position_in_frame / 2, wavebuffer.data(), size);
 
     auto context{req.adpcm_context};
     auto header{context->header};
@@ -258,7 +257,7 @@ void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuf
     u32 offset{voice_state.offset};
 
     auto output_buffer{args.output};
-    std::vector<s16> temp_buffer(TempBufferSize, 0);
+    std::array<s16, TempBufferSize> temp_buffer{};
 
     while (remaining_sample_count > 0) {
         const auto samples_to_write{std::min(remaining_sample_count, max_remaining_sample_count)};
diff --git a/src/audio_core/renderer/command/effect/compressor.cpp b/src/audio_core/renderer/command/effect/compressor.cpp
index 7229618e8..ee9b68d5b 100644
--- a/src/audio_core/renderer/command/effect/compressor.cpp
+++ b/src/audio_core/renderer/command/effect/compressor.cpp
@@ -44,8 +44,8 @@ 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::vector<std::span<s32>> output_buffers, u32 sample_count) {
+                                  std::span<std::span<const s32>> input_buffers,
+                                  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 +124,8 @@ 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);
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    std::array<std::span<const s32>, MaxChannels> input_buffers{};
+    std::array<std::span<s32>, MaxChannels> output_buffers{};
 
     for (s16 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,
diff --git a/src/audio_core/renderer/command/effect/delay.cpp b/src/audio_core/renderer/command/effect/delay.cpp
index a4e408d40..e536cbb1e 100644
--- a/src/audio_core/renderer/command/effect/delay.cpp
+++ b/src/audio_core/renderer/command/effect/delay.cpp
@@ -51,7 +51,7 @@ static void InitializeDelayEffect(const DelayInfo::ParameterVersion1& params,
         state.delay_lines[channel].sample_count_max = sample_count_max.to_int_floor();
         state.delay_lines[channel].sample_count = sample_count.to_int_floor();
         state.delay_lines[channel].buffer.resize(state.delay_lines[channel].sample_count, 0);
-        if (state.delay_lines[channel].buffer.size() == 0) {
+        if (state.delay_lines[channel].sample_count == 0) {
             state.delay_lines[channel].buffer.push_back(0);
         }
         state.delay_lines[channel].buffer_pos = 0;
@@ -74,8 +74,8 @@ 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::vector<std::span<s32>>& outputs, const u32 sample_count) {
+                       std::span<std::span<const s32>> inputs, std::span<std::span<s32>> outputs,
+                       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 +153,8 @@ 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,
-                             std::vector<std::span<s32>>& outputs, const u32 sample_count) {
+                             const bool enabled, std::span<std::span<const s32>> inputs,
+                             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 +208,8 @@ 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);
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    std::array<std::span<const s32>, MaxChannels> input_buffers{};
+    std::array<std::span<s32>, MaxChannels> output_buffers{};
 
     for (s16 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,
diff --git a/src/audio_core/renderer/command/effect/i3dl2_reverb.cpp b/src/audio_core/renderer/command/effect/i3dl2_reverb.cpp
index 27d8b9844..d2bfb67cc 100644
--- a/src/audio_core/renderer/command/effect/i3dl2_reverb.cpp
+++ b/src/audio_core/renderer/command/effect/i3dl2_reverb.cpp
@@ -408,8 +408,8 @@ 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);
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    std::array<std::span<const s32>, MaxChannels> input_buffers{};
+    std::array<std::span<s32>, MaxChannels> output_buffers{};
 
     for (u32 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,
diff --git a/src/audio_core/renderer/command/effect/light_limiter.cpp b/src/audio_core/renderer/command/effect/light_limiter.cpp
index e8fb0e2fc..4161a9821 100644
--- a/src/audio_core/renderer/command/effect/light_limiter.cpp
+++ b/src/audio_core/renderer/command/effect/light_limiter.cpp
@@ -47,8 +47,8 @@ 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::vector<std::span<s32>>& outputs, const u32 sample_count,
+                                    std::span<std::span<const s32>> inputs,
+                                    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 +147,8 @@ 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);
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    std::array<std::span<const s32>, MaxChannels> input_buffers{};
+    std::array<std::span<s32>, MaxChannels> output_buffers{};
 
     for (u32 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,
@@ -190,8 +190,8 @@ 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);
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    std::array<std::span<const s32>, MaxChannels> input_buffers{};
+    std::array<std::span<s32>, MaxChannels> output_buffers{};
 
     for (u32 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,
diff --git a/src/audio_core/renderer/command/effect/reverb.cpp b/src/audio_core/renderer/command/effect/reverb.cpp
index 8b9b65214..fc2f15a5e 100644
--- a/src/audio_core/renderer/command/effect/reverb.cpp
+++ b/src/audio_core/renderer/command/effect/reverb.cpp
@@ -250,8 +250,8 @@ 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::vector<std::span<s32>>& outputs, const u32 sample_count) {
+                              std::span<std::span<const s32>> inputs,
+                              std::span<std::span<s32>> outputs, const u32 sample_count) {
     static constexpr std::array<u8, ReverbInfo::MaxDelayTaps> OutTapIndexes1Ch{
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
     };
@@ -369,8 +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,
-                              std::vector<std::span<s32>>& outputs, const u32 sample_count) {
+                              const bool enabled, std::span<std::span<const s32>> inputs,
+                              std::span<std::span<s32>> outputs, const u32 sample_count) {
     if (enabled) {
         switch (params.channel_count) {
         case 0:
@@ -412,8 +412,8 @@ 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);
-    std::vector<std::span<s32>> output_buffers(parameter.channel_count);
+    std::array<std::span<const s32>, MaxChannels> input_buffers{};
+    std::array<std::span<s32>, MaxChannels> output_buffers{};
 
     for (u32 i = 0; i < parameter.channel_count; i++) {
         input_buffers[i] = processor.mix_buffers.subspan(inputs[i] * processor.sample_count,
diff --git a/src/audio_core/renderer/command/sink/circular_buffer.cpp b/src/audio_core/renderer/command/sink/circular_buffer.cpp
index ded5afc94..e2ce59792 100644
--- a/src/audio_core/renderer/command/sink/circular_buffer.cpp
+++ b/src/audio_core/renderer/command/sink/circular_buffer.cpp
@@ -24,7 +24,7 @@ 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);
+    std::array<s16, TargetSampleCount * MaxChannels> output{};
     for (u32 channel = 0; channel < input_count; channel++) {
         auto input{processor.mix_buffers.subspan(inputs[channel] * processor.sample_count,
                                                  processor.sample_count)};
@@ -33,7 +33,7 @@ void CircularBufferSinkCommand::Process(const ADSP::CommandListProcessor& proces
         }
 
         processor.memory->WriteBlockUnsafe(address + pos, output.data(),
-                                           output.size() * sizeof(s16));
+                                           processor.sample_count * sizeof(s16));
         pos += static_cast<u32>(processor.sample_count * sizeof(s16));
         if (pos >= size) {
             pos = 0;
diff --git a/src/audio_core/renderer/command/sink/device.cpp b/src/audio_core/renderer/command/sink/device.cpp
index e88372a75..5f74dd7ad 100644
--- a/src/audio_core/renderer/command/sink/device.cpp
+++ b/src/audio_core/renderer/command/sink/device.cpp
@@ -33,8 +33,7 @@ void DeviceSinkCommand::Process(const ADSP::CommandListProcessor& processor) {
         .consumed{false},
     };
 
-    std::vector<s16> samples(out_buffer.frames * input_count);
-
+    std::array<s16, TargetSampleCount * MaxChannels> samples{};
     for (u32 channel = 0; channel < input_count; channel++) {
         const auto offset{inputs[channel] * out_buffer.frames};
 
@@ -45,7 +44,7 @@ void DeviceSinkCommand::Process(const ADSP::CommandListProcessor& processor) {
     }
 
     out_buffer.tag = reinterpret_cast<u64>(samples.data());
-    stream->AppendBuffer(out_buffer, samples);
+    stream->AppendBuffer(out_buffer, {samples.data(), out_buffer.frames * input_count});
 
     if (stream->IsPaused()) {
         stream->Start();
diff --git a/src/audio_core/renderer/mix/mix_context.cpp b/src/audio_core/renderer/mix/mix_context.cpp
index 35b748ede..3a18ae7c2 100644
--- a/src/audio_core/renderer/mix/mix_context.cpp
+++ b/src/audio_core/renderer/mix/mix_context.cpp
@@ -125,10 +125,10 @@ bool MixContext::TSortInfo(const SplitterContext& splitter_context) {
         return false;
     }
 
-    std::vector<s32> sorted_results{node_states.GetSortedResuls()};
-    const auto result_size{std::min(count, static_cast<s32>(sorted_results.size()))};
+    auto sorted_results{node_states.GetSortedResuls()};
+    const auto result_size{std::min(count, static_cast<s32>(sorted_results.second))};
     for (s32 i = 0; i < result_size; i++) {
-        sorted_mix_infos[i] = &mix_infos[sorted_results[i]];
+        sorted_mix_infos[i] = &mix_infos[sorted_results.first[i]];
     }
 
     CalcMixBufferOffset();
diff --git a/src/audio_core/renderer/nodes/node_states.cpp b/src/audio_core/renderer/nodes/node_states.cpp
index 1821a51e6..b7a44a54c 100644
--- a/src/audio_core/renderer/nodes/node_states.cpp
+++ b/src/audio_core/renderer/nodes/node_states.cpp
@@ -134,8 +134,8 @@ u32 NodeStates::GetNodeCount() const {
     return node_count;
 }
 
-std::vector<s32> NodeStates::GetSortedResuls() const {
-    return {results.rbegin(), results.rbegin() + result_pos};
+std::pair<std::span<u32>::reverse_iterator, size_t> NodeStates::GetSortedResuls() const {
+    return {results.rbegin(), result_pos};
 }
 
 } // namespace AudioCore::AudioRenderer
diff --git a/src/audio_core/renderer/nodes/node_states.h b/src/audio_core/renderer/nodes/node_states.h
index 94b1d1254..e768cd4b5 100644
--- a/src/audio_core/renderer/nodes/node_states.h
+++ b/src/audio_core/renderer/nodes/node_states.h
@@ -175,7 +175,7 @@ public:
      *
      * @return Vector of nodes in reverse order.
      */
-    std::vector<s32> GetSortedResuls() const;
+    std::pair<std::span<u32>::reverse_iterator, size_t> GetSortedResuls() const;
 
 private:
     /// Number of nodes in the graph
diff --git a/src/audio_core/renderer/system.cpp b/src/audio_core/renderer/system.cpp
index 53b258c4f..a23627472 100644
--- a/src/audio_core/renderer/system.cpp
+++ b/src/audio_core/renderer/system.cpp
@@ -444,6 +444,7 @@ Result System::Update(std::span<const u8> input, std::span<u8> performance, std:
     std::scoped_lock l{lock};
 
     const auto start_time{core.CoreTiming().GetClockTicks()};
+    std::memset(output.data(), 0, output.size());
 
     InfoUpdater info_updater(input, output, process_handle, behavior);