Merge pull request #10457 from Kelebek1/optimise

Remove memory allocations in some hot paths

31 files changed, 253 insertions, 259 deletions

diff --git a/src/core/hle/service/audio/audin_u.cpp b/src/core/hle/service/audio/audin_u.cpp
index f0640c64f..c8d574993 100644
--- a/src/core/hle/service/audio/audin_u.cpp
+++ b/src/core/hle/service/audio/audin_u.cpp
@@ -5,6 +5,7 @@
 #include "audio_core/renderer/audio_device.h"
 #include "common/common_funcs.h"
 #include "common/logging/log.h"
+#include "common/settings.h"
 #include "common/string_util.h"
 #include "core/core.h"
 #include "core/hle/kernel/k_event.h"
@@ -123,19 +124,13 @@ private:
 
     void GetReleasedAudioInBuffer(HLERequestContext& ctx) {
         const auto write_buffer_size = ctx.GetWriteBufferNumElements<u64>();
-        std::vector<u64> released_buffers(write_buffer_size);
+        tmp_buffer.resize_destructive(write_buffer_size);
+        tmp_buffer[0] = 0;
 
-        const auto count = impl->GetReleasedBuffers(released_buffers);
+        const auto count = impl->GetReleasedBuffers(tmp_buffer);
 
-        [[maybe_unused]] std::string tags{};
-        for (u32 i = 0; i < count; i++) {
-            tags += fmt::format("{:08X}, ", released_buffers[i]);
-        }
-        [[maybe_unused]] auto sessionid{impl->GetSystem().GetSessionId()};
-        LOG_TRACE(Service_Audio, "called. Session {} released {} buffers: {}", sessionid, count,
-                  tags);
+        ctx.WriteBuffer(tmp_buffer);
 
-        ctx.WriteBuffer(released_buffers);
         IPC::ResponseBuilder rb{ctx, 3};
         rb.Push(ResultSuccess);
         rb.Push(count);
@@ -200,6 +195,7 @@ private:
     KernelHelpers::ServiceContext service_context;
     Kernel::KEvent* event;
     std::shared_ptr<AudioCore::AudioIn::In> impl;
+    Common::ScratchBuffer<u64> tmp_buffer;
 };
 
 AudInU::AudInU(Core::System& system_)
diff --git a/src/core/hle/service/audio/audout_u.cpp b/src/core/hle/service/audio/audout_u.cpp
index 3e62fa4fc..032c8c11f 100644
--- a/src/core/hle/service/audio/audout_u.cpp
+++ b/src/core/hle/service/audio/audout_u.cpp
@@ -123,19 +123,13 @@ private:
 
     void GetReleasedAudioOutBuffers(HLERequestContext& ctx) {
         const auto write_buffer_size = ctx.GetWriteBufferNumElements<u64>();
-        std::vector<u64> released_buffers(write_buffer_size);
+        tmp_buffer.resize_destructive(write_buffer_size);
+        tmp_buffer[0] = 0;
 
-        const auto count = impl->GetReleasedBuffers(released_buffers);
+        const auto count = impl->GetReleasedBuffers(tmp_buffer);
 
-        [[maybe_unused]] std::string tags{};
-        for (u32 i = 0; i < count; i++) {
-            tags += fmt::format("{:08X}, ", released_buffers[i]);
-        }
-        [[maybe_unused]] const auto sessionid{impl->GetSystem().GetSessionId()};
-        LOG_TRACE(Service_Audio, "called. Session {} released {} buffers: {}", sessionid, count,
-                  tags);
+        ctx.WriteBuffer(tmp_buffer);
 
-        ctx.WriteBuffer(released_buffers);
         IPC::ResponseBuilder rb{ctx, 3};
         rb.Push(ResultSuccess);
         rb.Push(count);
@@ -211,6 +205,7 @@ private:
     KernelHelpers::ServiceContext service_context;
     Kernel::KEvent* event;
     std::shared_ptr<AudioCore::AudioOut::Out> impl;
+    Common::ScratchBuffer<u64> tmp_buffer;
 };
 
 AudOutU::AudOutU(Core::System& system_)
diff --git a/src/core/hle/service/audio/audren_u.cpp b/src/core/hle/service/audio/audren_u.cpp
index 7086d4750..12845c23a 100644
--- a/src/core/hle/service/audio/audren_u.cpp
+++ b/src/core/hle/service/audio/audren_u.cpp
@@ -116,28 +116,26 @@ private:
         // These buffers are written manually to avoid an issue with WriteBuffer throwing errors for
         // checking size 0. Performance size is 0 for most games.
 
-        std::vector<u8> output{};
-        std::vector<u8> performance{};
         auto is_buffer_b{ctx.BufferDescriptorB()[0].Size() != 0};
         if (is_buffer_b) {
             const auto buffersB{ctx.BufferDescriptorB()};
-            output.resize(buffersB[0].Size(), 0);
-            performance.resize(buffersB[1].Size(), 0);
+            tmp_output.resize_destructive(buffersB[0].Size());
+            tmp_performance.resize_destructive(buffersB[1].Size());
         } else {
             const auto buffersC{ctx.BufferDescriptorC()};
-            output.resize(buffersC[0].Size(), 0);
-            performance.resize(buffersC[1].Size(), 0);
+            tmp_output.resize_destructive(buffersC[0].Size());
+            tmp_performance.resize_destructive(buffersC[1].Size());
         }
 
-        auto result = impl->RequestUpdate(input, performance, output);
+        auto result = impl->RequestUpdate(input, tmp_performance, tmp_output);
 
         if (result.IsSuccess()) {
             if (is_buffer_b) {
-                ctx.WriteBufferB(output.data(), output.size(), 0);
-                ctx.WriteBufferB(performance.data(), performance.size(), 1);
+                ctx.WriteBufferB(tmp_output.data(), tmp_output.size(), 0);
+                ctx.WriteBufferB(tmp_performance.data(), tmp_performance.size(), 1);
             } else {
-                ctx.WriteBufferC(output.data(), output.size(), 0);
-                ctx.WriteBufferC(performance.data(), performance.size(), 1);
+                ctx.WriteBufferC(tmp_output.data(), tmp_output.size(), 0);
+                ctx.WriteBufferC(tmp_performance.data(), tmp_performance.size(), 1);
             }
         } else {
             LOG_ERROR(Service_Audio, "RequestUpdate failed error 0x{:02X}!", result.description);
@@ -235,6 +233,8 @@ private:
     Kernel::KEvent* rendered_event;
     Manager& manager;
     std::unique_ptr<Renderer> impl;
+    Common::ScratchBuffer<u8> tmp_output;
+    Common::ScratchBuffer<u8> tmp_performance;
 };
 
 class IAudioDevice final : public ServiceFramework<IAudioDevice> {
diff --git a/src/core/hle/service/audio/audren_u.h b/src/core/hle/service/audio/audren_u.h
index 24ce37e87..d8e9c8719 100644
--- a/src/core/hle/service/audio/audren_u.h
+++ b/src/core/hle/service/audio/audren_u.h
@@ -4,6 +4,7 @@
 #pragma once
 
 #include "audio_core/audio_render_manager.h"
+#include "common/scratch_buffer.h"
 #include "core/hle/service/kernel_helpers.h"
 #include "core/hle/service/service.h"
 
diff --git a/src/core/hle/service/audio/hwopus.cpp b/src/core/hle/service/audio/hwopus.cpp
index 451ac224a..c835f6cb7 100644
--- a/src/core/hle/service/audio/hwopus.cpp
+++ b/src/core/hle/service/audio/hwopus.cpp
@@ -68,13 +68,13 @@ private:
                                  ExtraBehavior extra_behavior) {
         u32 consumed = 0;
         u32 sample_count = 0;
-        std::vector<opus_int16> samples(ctx.GetWriteBufferNumElements<opus_int16>());
+        tmp_samples.resize_destructive(ctx.GetWriteBufferNumElements<opus_int16>());
 
         if (extra_behavior == ExtraBehavior::ResetContext) {
             ResetDecoderContext();
         }
 
-        if (!DecodeOpusData(consumed, sample_count, ctx.ReadBuffer(), samples, performance)) {
+        if (!DecodeOpusData(consumed, sample_count, ctx.ReadBuffer(), tmp_samples, performance)) {
             LOG_ERROR(Audio, "Failed to decode opus data");
             IPC::ResponseBuilder rb{ctx, 2};
             // TODO(ogniK): Use correct error code
@@ -90,11 +90,11 @@ private:
         if (performance) {
             rb.Push<u64>(*performance);
         }
-        ctx.WriteBuffer(samples);
+        ctx.WriteBuffer(tmp_samples);
     }
 
     bool DecodeOpusData(u32& consumed, u32& sample_count, std::span<const u8> input,
-                        std::vector<opus_int16>& output, u64* out_performance_time) const {
+                        std::span<opus_int16> output, u64* out_performance_time) const {
         const auto start_time = std::chrono::steady_clock::now();
         const std::size_t raw_output_sz = output.size() * sizeof(opus_int16);
         if (sizeof(OpusPacketHeader) > input.size()) {
@@ -154,6 +154,7 @@ private:
     OpusDecoderPtr decoder;
     u32 sample_rate;
     u32 channel_count;
+    Common::ScratchBuffer<opus_int16> tmp_samples;
 };
 
 class IHardwareOpusDecoderManager final : public ServiceFramework<IHardwareOpusDecoderManager> {
diff --git a/src/core/hle/service/nvdrv/devices/nvdevice.h b/src/core/hle/service/nvdrv/devices/nvdevice.h
index ab1f30f9e..a04538d5d 100644
--- a/src/core/hle/service/nvdrv/devices/nvdevice.h
+++ b/src/core/hle/service/nvdrv/devices/nvdevice.h
@@ -34,7 +34,7 @@ public:
      * @returns The result code of the ioctl.
      */
     virtual NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::vector<u8>& output) = 0;
+                            std::span<u8> output) = 0;
 
     /**
      * Handles an ioctl2 request.
@@ -45,7 +45,7 @@ public:
      * @returns The result code of the ioctl.
      */
     virtual NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::span<const u8> inline_input, std::vector<u8>& output) = 0;
+                            std::span<const u8> inline_input, std::span<u8> output) = 0;
 
     /**
      * Handles an ioctl3 request.
@@ -56,7 +56,7 @@ public:
      * @returns The result code of the ioctl.
      */
     virtual NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::vector<u8>& output, std::vector<u8>& inline_output) = 0;
+                            std::span<u8> output, std::span<u8> inline_output) = 0;
 
     /**
      * Called once a device is opened
diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
index 0fe242e9d..05a43d8dc 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
@@ -18,19 +18,19 @@ nvdisp_disp0::nvdisp_disp0(Core::System& system_, NvCore::Container& core)
 nvdisp_disp0::~nvdisp_disp0() = default;
 
 NvResult nvdisp_disp0::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::vector<u8>& output) {
+                              std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvdisp_disp0::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::span<const u8> inline_input, std::vector<u8>& output) {
+                              std::span<const u8> inline_input, std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvdisp_disp0::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::vector<u8>& output, std::vector<u8>& inline_output) {
+                              std::span<u8> output, std::span<u8> inline_output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
index bcd0e3ed5..daee05fe8 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
@@ -26,11 +26,11 @@ public:
     ~nvdisp_disp0() override;
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
index 681bd0867..07e570a9f 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
@@ -28,7 +28,7 @@ nvhost_as_gpu::nvhost_as_gpu(Core::System& system_, Module& module_, NvCore::Con
 nvhost_as_gpu::~nvhost_as_gpu() = default;
 
 NvResult nvhost_as_gpu::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                               std::vector<u8>& output) {
+                               std::span<u8> output) {
     switch (command.group) {
     case 'A':
         switch (command.cmd) {
@@ -61,13 +61,13 @@ NvResult nvhost_as_gpu::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> i
 }
 
 NvResult nvhost_as_gpu::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                               std::span<const u8> inline_input, std::vector<u8>& output) {
+                               std::span<const u8> inline_input, std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_as_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                               std::vector<u8>& output, std::vector<u8>& inline_output) {
+                               std::span<u8> output, std::span<u8> inline_output) {
     switch (command.group) {
     case 'A':
         switch (command.cmd) {
@@ -87,7 +87,7 @@ NvResult nvhost_as_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> i
 void nvhost_as_gpu::OnOpen(DeviceFD fd) {}
 void nvhost_as_gpu::OnClose(DeviceFD fd) {}
 
-NvResult nvhost_as_gpu::AllocAsEx(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_as_gpu::AllocAsEx(std::span<const u8> input, std::span<u8> output) {
     IoctlAllocAsEx params{};
     std::memcpy(&params, input.data(), input.size());
 
@@ -141,7 +141,7 @@ NvResult nvhost_as_gpu::AllocAsEx(std::span<const u8> input, std::vector<u8>& ou
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::AllocateSpace(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_as_gpu::AllocateSpace(std::span<const u8> input, std::span<u8> output) {
     IoctlAllocSpace params{};
     std::memcpy(&params, input.data(), input.size());
 
@@ -220,7 +220,7 @@ void nvhost_as_gpu::FreeMappingLocked(u64 offset) {
     mapping_map.erase(offset);
 }
 
-NvResult nvhost_as_gpu::FreeSpace(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_as_gpu::FreeSpace(std::span<const u8> input, std::span<u8> output) {
     IoctlFreeSpace params{};
     std::memcpy(&params, input.data(), input.size());
 
@@ -266,15 +266,14 @@ NvResult nvhost_as_gpu::FreeSpace(std::span<const u8> input, std::vector<u8>& ou
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::Remap(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_as_gpu::Remap(std::span<const u8> input, std::span<u8> output) {
     const auto num_entries = input.size() / sizeof(IoctlRemapEntry);
 
     LOG_DEBUG(Service_NVDRV, "called, num_entries=0x{:X}", num_entries);
 
-    std::vector<IoctlRemapEntry> entries(num_entries);
-    std::memcpy(entries.data(), input.data(), input.size());
-
     std::scoped_lock lock(mutex);
+    entries.resize_destructive(num_entries);
+    std::memcpy(entries.data(), input.data(), input.size());
 
     if (!vm.initialised) {
         return NvResult::BadValue;
@@ -320,7 +319,7 @@ NvResult nvhost_as_gpu::Remap(std::span<const u8> input, std::vector<u8>& output
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::MapBufferEx(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_as_gpu::MapBufferEx(std::span<const u8> input, std::span<u8> output) {
     IoctlMapBufferEx params{};
     std::memcpy(&params, input.data(), input.size());
 
@@ -424,7 +423,7 @@ NvResult nvhost_as_gpu::MapBufferEx(std::span<const u8> input, std::vector<u8>&
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::UnmapBuffer(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_as_gpu::UnmapBuffer(std::span<const u8> input, std::span<u8> output) {
     IoctlUnmapBuffer params{};
     std::memcpy(&params, input.data(), input.size());
 
@@ -463,7 +462,7 @@ NvResult nvhost_as_gpu::UnmapBuffer(std::span<const u8> input, std::vector<u8>&
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::BindChannel(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_as_gpu::BindChannel(std::span<const u8> input, std::span<u8> output) {
     IoctlBindChannel params{};
     std::memcpy(&params, input.data(), input.size());
     LOG_DEBUG(Service_NVDRV, "called, fd={:X}", params.fd);
@@ -492,7 +491,7 @@ void nvhost_as_gpu::GetVARegionsImpl(IoctlGetVaRegions& params) {
     };
 }
 
-NvResult nvhost_as_gpu::GetVARegions(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_as_gpu::GetVARegions(std::span<const u8> input, std::span<u8> output) {
     IoctlGetVaRegions params{};
     std::memcpy(&params, input.data(), input.size());
 
@@ -511,8 +510,8 @@ NvResult nvhost_as_gpu::GetVARegions(std::span<const u8> input, std::vector<u8>&
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::GetVARegions(std::span<const u8> input, std::vector<u8>& output,
-                                     std::vector<u8>& inline_output) {
+NvResult nvhost_as_gpu::GetVARegions(std::span<const u8> input, std::span<u8> output,
+                                     std::span<u8> inline_output) {
     IoctlGetVaRegions params{};
     std::memcpy(&params, input.data(), input.size());
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
index 1aba8d579..2af3e1260 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
@@ -15,6 +15,7 @@
 #include "common/address_space.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "common/scratch_buffer.h"
 #include "common/swap.h"
 #include "core/hle/service/nvdrv/core/nvmap.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
@@ -48,11 +49,11 @@ public:
     ~nvhost_as_gpu() override;
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
@@ -138,18 +139,18 @@ private:
     static_assert(sizeof(IoctlGetVaRegions) == 16 + sizeof(VaRegion) * 2,
                   "IoctlGetVaRegions is incorrect size");
 
-    NvResult AllocAsEx(std::span<const u8> input, std::vector<u8>& output);
-    NvResult AllocateSpace(std::span<const u8> input, std::vector<u8>& output);
-    NvResult Remap(std::span<const u8> input, std::vector<u8>& output);
-    NvResult MapBufferEx(std::span<const u8> input, std::vector<u8>& output);
-    NvResult UnmapBuffer(std::span<const u8> input, std::vector<u8>& output);
-    NvResult FreeSpace(std::span<const u8> input, std::vector<u8>& output);
-    NvResult BindChannel(std::span<const u8> input, std::vector<u8>& output);
+    NvResult AllocAsEx(std::span<const u8> input, std::span<u8> output);
+    NvResult AllocateSpace(std::span<const u8> input, std::span<u8> output);
+    NvResult Remap(std::span<const u8> input, std::span<u8> output);
+    NvResult MapBufferEx(std::span<const u8> input, std::span<u8> output);
+    NvResult UnmapBuffer(std::span<const u8> input, std::span<u8> output);
+    NvResult FreeSpace(std::span<const u8> input, std::span<u8> output);
+    NvResult BindChannel(std::span<const u8> input, std::span<u8> output);
 
     void GetVARegionsImpl(IoctlGetVaRegions& params);
-    NvResult GetVARegions(std::span<const u8> input, std::vector<u8>& output);
-    NvResult GetVARegions(std::span<const u8> input, std::vector<u8>& output,
-                          std::vector<u8>& inline_output);
+    NvResult GetVARegions(std::span<const u8> input, std::span<u8> output);
+    NvResult GetVARegions(std::span<const u8> input, std::span<u8> output,
+                          std::span<u8> inline_output);
 
     void FreeMappingLocked(u64 offset);
 
@@ -212,6 +213,7 @@ private:
         bool initialised{};
     } vm;
     std::shared_ptr<Tegra::MemoryManager> gmmu;
+    Common::ScratchBuffer<IoctlRemapEntry> entries;
 
     // s32 channel{};
     // u32 big_page_size{VM::DEFAULT_BIG_PAGE_SIZE};
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
index e12025560..4d55554b4 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
@@ -35,7 +35,7 @@ nvhost_ctrl::~nvhost_ctrl() {
 }
 
 NvResult nvhost_ctrl::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                             std::vector<u8>& output) {
+                             std::span<u8> output) {
     switch (command.group) {
     case 0x0:
         switch (command.cmd) {
@@ -64,13 +64,13 @@ NvResult nvhost_ctrl::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> inp
 }
 
 NvResult nvhost_ctrl::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                             std::span<const u8> inline_input, std::vector<u8>& output) {
+                             std::span<const u8> inline_input, std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_ctrl::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                             std::vector<u8>& output, std::vector<u8>& inline_outpu) {
+                             std::span<u8> output, std::span<u8> inline_outpu) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
@@ -79,7 +79,7 @@ void nvhost_ctrl::OnOpen(DeviceFD fd) {}
 
 void nvhost_ctrl::OnClose(DeviceFD fd) {}
 
-NvResult nvhost_ctrl::NvOsGetConfigU32(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl::NvOsGetConfigU32(std::span<const u8> input, std::span<u8> output) {
     IocGetConfigParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
     LOG_TRACE(Service_NVDRV, "called, setting={}!{}", params.domain_str.data(),
@@ -87,7 +87,7 @@ NvResult nvhost_ctrl::NvOsGetConfigU32(std::span<const u8> input, std::vector<u8
     return NvResult::ConfigVarNotFound; // Returns error on production mode
 }
 
-NvResult nvhost_ctrl::IocCtrlEventWait(std::span<const u8> input, std::vector<u8>& output,
+NvResult nvhost_ctrl::IocCtrlEventWait(std::span<const u8> input, std::span<u8> output,
                                        bool is_allocation) {
     IocCtrlEventWaitParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
@@ -231,7 +231,7 @@ NvResult nvhost_ctrl::FreeEvent(u32 slot) {
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl::IocCtrlEventRegister(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl::IocCtrlEventRegister(std::span<const u8> input, std::span<u8> output) {
     IocCtrlEventRegisterParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
     const u32 event_id = params.user_event_id;
@@ -252,7 +252,7 @@ NvResult nvhost_ctrl::IocCtrlEventRegister(std::span<const u8> input, std::vecto
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl::IocCtrlEventUnregister(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl::IocCtrlEventUnregister(std::span<const u8> input, std::span<u8> output) {
     IocCtrlEventUnregisterParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
     const u32 event_id = params.user_event_id & 0x00FF;
@@ -262,8 +262,7 @@ NvResult nvhost_ctrl::IocCtrlEventUnregister(std::span<const u8> input, std::vec
     return FreeEvent(event_id);
 }
 
-NvResult nvhost_ctrl::IocCtrlEventUnregisterBatch(std::span<const u8> input,
-                                                  std::vector<u8>& output) {
+NvResult nvhost_ctrl::IocCtrlEventUnregisterBatch(std::span<const u8> input, std::span<u8> output) {
     IocCtrlEventUnregisterBatchParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
     u64 event_mask = params.user_events;
@@ -281,7 +280,7 @@ NvResult nvhost_ctrl::IocCtrlEventUnregisterBatch(std::span<const u8> input,
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl::IocCtrlClearEventWait(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl::IocCtrlClearEventWait(std::span<const u8> input, std::span<u8> output) {
     IocCtrlEventClearParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
index dd2e7888a..2efed4862 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
@@ -26,11 +26,11 @@ public:
     ~nvhost_ctrl() override;
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
@@ -186,13 +186,12 @@ private:
     static_assert(sizeof(IocCtrlEventUnregisterBatchParams) == 8,
                   "IocCtrlEventKill is incorrect size");
 
-    NvResult NvOsGetConfigU32(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocCtrlEventWait(std::span<const u8> input, std::vector<u8>& output,
-                              bool is_allocation);
-    NvResult IocCtrlEventRegister(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocCtrlEventUnregister(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocCtrlEventUnregisterBatch(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocCtrlClearEventWait(std::span<const u8> input, std::vector<u8>& output);
+    NvResult NvOsGetConfigU32(std::span<const u8> input, std::span<u8> output);
+    NvResult IocCtrlEventWait(std::span<const u8> input, std::span<u8> output, bool is_allocation);
+    NvResult IocCtrlEventRegister(std::span<const u8> input, std::span<u8> output);
+    NvResult IocCtrlEventUnregister(std::span<const u8> input, std::span<u8> output);
+    NvResult IocCtrlEventUnregisterBatch(std::span<const u8> input, std::span<u8> output);
+    NvResult IocCtrlClearEventWait(std::span<const u8> input, std::span<u8> output);
 
     NvResult FreeEvent(u32 slot);
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp
index be3c083db..6081d92e9 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp
@@ -22,7 +22,7 @@ nvhost_ctrl_gpu::~nvhost_ctrl_gpu() {
 }
 
 NvResult nvhost_ctrl_gpu::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                                 std::vector<u8>& output) {
+                                 std::span<u8> output) {
     switch (command.group) {
     case 'G':
         switch (command.cmd) {
@@ -54,13 +54,13 @@ NvResult nvhost_ctrl_gpu::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8>
 }
 
 NvResult nvhost_ctrl_gpu::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                                 std::span<const u8> inline_input, std::vector<u8>& output) {
+                                 std::span<const u8> inline_input, std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_ctrl_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                                 std::vector<u8>& output, std::vector<u8>& inline_output) {
+                                 std::span<u8> output, std::span<u8> inline_output) {
     switch (command.group) {
     case 'G':
         switch (command.cmd) {
@@ -82,7 +82,7 @@ NvResult nvhost_ctrl_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8>
 void nvhost_ctrl_gpu::OnOpen(DeviceFD fd) {}
 void nvhost_ctrl_gpu::OnClose(DeviceFD fd) {}
 
-NvResult nvhost_ctrl_gpu::GetCharacteristics(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::GetCharacteristics(std::span<const u8> input, std::span<u8> output) {
     LOG_DEBUG(Service_NVDRV, "called");
     IoctlCharacteristics params{};
     std::memcpy(&params, input.data(), input.size());
@@ -127,8 +127,8 @@ NvResult nvhost_ctrl_gpu::GetCharacteristics(std::span<const u8> input, std::vec
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::GetCharacteristics(std::span<const u8> input, std::vector<u8>& output,
-                                             std::vector<u8>& inline_output) {
+NvResult nvhost_ctrl_gpu::GetCharacteristics(std::span<const u8> input, std::span<u8> output,
+                                             std::span<u8> inline_output) {
     LOG_DEBUG(Service_NVDRV, "called");
     IoctlCharacteristics params{};
     std::memcpy(&params, input.data(), input.size());
@@ -175,7 +175,7 @@ NvResult nvhost_ctrl_gpu::GetCharacteristics(std::span<const u8> input, std::vec
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::GetTPCMasks(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::GetTPCMasks(std::span<const u8> input, std::span<u8> output) {
     IoctlGpuGetTpcMasksArgs params{};
     std::memcpy(&params, input.data(), input.size());
     LOG_DEBUG(Service_NVDRV, "called, mask_buffer_size=0x{:X}", params.mask_buffer_size);
@@ -186,8 +186,8 @@ NvResult nvhost_ctrl_gpu::GetTPCMasks(std::span<const u8> input, std::vector<u8>
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::GetTPCMasks(std::span<const u8> input, std::vector<u8>& output,
-                                      std::vector<u8>& inline_output) {
+NvResult nvhost_ctrl_gpu::GetTPCMasks(std::span<const u8> input, std::span<u8> output,
+                                      std::span<u8> inline_output) {
     IoctlGpuGetTpcMasksArgs params{};
     std::memcpy(&params, input.data(), input.size());
     LOG_DEBUG(Service_NVDRV, "called, mask_buffer_size=0x{:X}", params.mask_buffer_size);
@@ -199,7 +199,7 @@ NvResult nvhost_ctrl_gpu::GetTPCMasks(std::span<const u8> input, std::vector<u8>
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::GetActiveSlotMask(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::GetActiveSlotMask(std::span<const u8> input, std::span<u8> output) {
     LOG_DEBUG(Service_NVDRV, "called");
 
     IoctlActiveSlotMask params{};
@@ -212,7 +212,7 @@ NvResult nvhost_ctrl_gpu::GetActiveSlotMask(std::span<const u8> input, std::vect
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::ZCullGetCtxSize(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::ZCullGetCtxSize(std::span<const u8> input, std::span<u8> output) {
     LOG_DEBUG(Service_NVDRV, "called");
 
     IoctlZcullGetCtxSize params{};
@@ -224,7 +224,7 @@ NvResult nvhost_ctrl_gpu::ZCullGetCtxSize(std::span<const u8> input, std::vector
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::ZCullGetInfo(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::ZCullGetInfo(std::span<const u8> input, std::span<u8> output) {
     LOG_DEBUG(Service_NVDRV, "called");
 
     IoctlNvgpuGpuZcullGetInfoArgs params{};
@@ -247,7 +247,7 @@ NvResult nvhost_ctrl_gpu::ZCullGetInfo(std::span<const u8> input, std::vector<u8
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::ZBCSetTable(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::ZBCSetTable(std::span<const u8> input, std::span<u8> output) {
     LOG_WARNING(Service_NVDRV, "(STUBBED) called");
 
     IoctlZbcSetTable params{};
@@ -263,7 +263,7 @@ NvResult nvhost_ctrl_gpu::ZBCSetTable(std::span<const u8> input, std::vector<u8>
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::ZBCQueryTable(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::ZBCQueryTable(std::span<const u8> input, std::span<u8> output) {
     LOG_WARNING(Service_NVDRV, "(STUBBED) called");
 
     IoctlZbcQueryTable params{};
@@ -273,7 +273,7 @@ NvResult nvhost_ctrl_gpu::ZBCQueryTable(std::span<const u8> input, std::vector<u
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::FlushL2(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::FlushL2(std::span<const u8> input, std::span<u8> output) {
     LOG_WARNING(Service_NVDRV, "(STUBBED) called");
 
     IoctlFlushL2 params{};
@@ -283,7 +283,7 @@ NvResult nvhost_ctrl_gpu::FlushL2(std::span<const u8> input, std::vector<u8>& ou
     return NvResult::Success;
 }
 
-NvResult nvhost_ctrl_gpu::GetGpuTime(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_ctrl_gpu::GetGpuTime(std::span<const u8> input, std::span<u8> output) {
     LOG_DEBUG(Service_NVDRV, "called");
 
     IoctlGetGpuTime params{};
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h
index b9333d9d3..97995551c 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h
@@ -22,11 +22,11 @@ public:
     ~nvhost_ctrl_gpu() override;
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
@@ -151,21 +151,21 @@ private:
     };
     static_assert(sizeof(IoctlGetGpuTime) == 0x10, "IoctlGetGpuTime is incorrect size");
 
-    NvResult GetCharacteristics(std::span<const u8> input, std::vector<u8>& output);
-    NvResult GetCharacteristics(std::span<const u8> input, std::vector<u8>& output,
-                                std::vector<u8>& inline_output);
-
-    NvResult GetTPCMasks(std::span<const u8> input, std::vector<u8>& output);
-    NvResult GetTPCMasks(std::span<const u8> input, std::vector<u8>& output,
-                         std::vector<u8>& inline_output);
-
-    NvResult GetActiveSlotMask(std::span<const u8> input, std::vector<u8>& output);
-    NvResult ZCullGetCtxSize(std::span<const u8> input, std::vector<u8>& output);
-    NvResult ZCullGetInfo(std::span<const u8> input, std::vector<u8>& output);
-    NvResult ZBCSetTable(std::span<const u8> input, std::vector<u8>& output);
-    NvResult ZBCQueryTable(std::span<const u8> input, std::vector<u8>& output);
-    NvResult FlushL2(std::span<const u8> input, std::vector<u8>& output);
-    NvResult GetGpuTime(std::span<const u8> input, std::vector<u8>& output);
+    NvResult GetCharacteristics(std::span<const u8> input, std::span<u8> output);
+    NvResult GetCharacteristics(std::span<const u8> input, std::span<u8> output,
+                                std::span<u8> inline_output);
+
+    NvResult GetTPCMasks(std::span<const u8> input, std::span<u8> output);
+    NvResult GetTPCMasks(std::span<const u8> input, std::span<u8> output,
+                         std::span<u8> inline_output);
+
+    NvResult GetActiveSlotMask(std::span<const u8> input, std::span<u8> output);
+    NvResult ZCullGetCtxSize(std::span<const u8> input, std::span<u8> output);
+    NvResult ZCullGetInfo(std::span<const u8> input, std::span<u8> output);
+    NvResult ZBCSetTable(std::span<const u8> input, std::span<u8> output);
+    NvResult ZBCQueryTable(std::span<const u8> input, std::span<u8> output);
+    NvResult FlushL2(std::span<const u8> input, std::span<u8> output);
+    NvResult GetGpuTime(std::span<const u8> input, std::span<u8> output);
 
     EventInterface& events_interface;
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
index 453a965dc..46a25fcab 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
@@ -47,7 +47,7 @@ nvhost_gpu::~nvhost_gpu() {
 }
 
 NvResult nvhost_gpu::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::vector<u8>& output) {
+                            std::span<u8> output) {
     switch (command.group) {
     case 0x0:
         switch (command.cmd) {
@@ -99,7 +99,7 @@ NvResult nvhost_gpu::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> inpu
 };
 
 NvResult nvhost_gpu::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::span<const u8> inline_input, std::vector<u8>& output) {
+                            std::span<const u8> inline_input, std::span<u8> output) {
     switch (command.group) {
     case 'H':
         switch (command.cmd) {
@@ -113,7 +113,7 @@ NvResult nvhost_gpu::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> inpu
 }
 
 NvResult nvhost_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::vector<u8>& output, std::vector<u8>& inline_output) {
+                            std::span<u8> output, std::span<u8> inline_output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
@@ -121,7 +121,7 @@ NvResult nvhost_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> inpu
 void nvhost_gpu::OnOpen(DeviceFD fd) {}
 void nvhost_gpu::OnClose(DeviceFD fd) {}
 
-NvResult nvhost_gpu::SetNVMAPfd(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::SetNVMAPfd(std::span<const u8> input, std::span<u8> output) {
     IoctlSetNvmapFD params{};
     std::memcpy(&params, input.data(), input.size());
     LOG_DEBUG(Service_NVDRV, "called, fd={}", params.nvmap_fd);
@@ -130,7 +130,7 @@ NvResult nvhost_gpu::SetNVMAPfd(std::span<const u8> input, std::vector<u8>& outp
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::SetClientData(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::SetClientData(std::span<const u8> input, std::span<u8> output) {
     LOG_DEBUG(Service_NVDRV, "called");
 
     IoctlClientData params{};
@@ -139,7 +139,7 @@ NvResult nvhost_gpu::SetClientData(std::span<const u8> input, std::vector<u8>& o
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::GetClientData(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::GetClientData(std::span<const u8> input, std::span<u8> output) {
     LOG_DEBUG(Service_NVDRV, "called");
 
     IoctlClientData params{};
@@ -149,7 +149,7 @@ NvResult nvhost_gpu::GetClientData(std::span<const u8> input, std::vector<u8>& o
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::ZCullBind(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::ZCullBind(std::span<const u8> input, std::span<u8> output) {
     std::memcpy(&zcull_params, input.data(), input.size());
     LOG_DEBUG(Service_NVDRV, "called, gpu_va={:X}, mode={:X}", zcull_params.gpu_va,
               zcull_params.mode);
@@ -158,7 +158,7 @@ NvResult nvhost_gpu::ZCullBind(std::span<const u8> input, std::vector<u8>& outpu
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::SetErrorNotifier(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::SetErrorNotifier(std::span<const u8> input, std::span<u8> output) {
     IoctlSetErrorNotifier params{};
     std::memcpy(&params, input.data(), input.size());
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, offset={:X}, size={:X}, mem={:X}", params.offset,
@@ -168,14 +168,14 @@ NvResult nvhost_gpu::SetErrorNotifier(std::span<const u8> input, std::vector<u8>
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::SetChannelPriority(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::SetChannelPriority(std::span<const u8> input, std::span<u8> output) {
     std::memcpy(&channel_priority, input.data(), input.size());
     LOG_DEBUG(Service_NVDRV, "(STUBBED) called, priority={:X}", channel_priority);
 
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::AllocGPFIFOEx2(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::AllocGPFIFOEx2(std::span<const u8> input, std::span<u8> output) {
     IoctlAllocGpfifoEx2 params{};
     std::memcpy(&params, input.data(), input.size());
     LOG_WARNING(Service_NVDRV,
@@ -197,7 +197,7 @@ NvResult nvhost_gpu::AllocGPFIFOEx2(std::span<const u8> input, std::vector<u8>&
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::AllocateObjectContext(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::AllocateObjectContext(std::span<const u8> input, std::span<u8> output) {
     IoctlAllocObjCtx params{};
     std::memcpy(&params, input.data(), input.size());
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, class_num={:X}, flags={:X}", params.class_num,
@@ -208,7 +208,8 @@ NvResult nvhost_gpu::AllocateObjectContext(std::span<const u8> input, std::vecto
     return NvResult::Success;
 }
 
-static std::vector<Tegra::CommandHeader> BuildWaitCommandList(NvFence fence) {
+static boost::container::small_vector<Tegra::CommandHeader, 512> BuildWaitCommandList(
+    NvFence fence) {
     return {
         Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointPayload, 1,
                                   Tegra::SubmissionMode::Increasing),
@@ -219,35 +220,35 @@ static std::vector<Tegra::CommandHeader> BuildWaitCommandList(NvFence fence) {
     };
 }
 
-static std::vector<Tegra::CommandHeader> BuildIncrementCommandList(NvFence fence) {
-    std::vector<Tegra::CommandHeader> result{
+static boost::container::small_vector<Tegra::CommandHeader, 512> BuildIncrementCommandList(
+    NvFence fence) {
+    boost::container::small_vector<Tegra::CommandHeader, 512> result{
         Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointPayload, 1,
                                   Tegra::SubmissionMode::Increasing),
         {}};
 
     for (u32 count = 0; count < 2; ++count) {
-        result.emplace_back(Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointOperation, 1,
-                                                      Tegra::SubmissionMode::Increasing));
-        result.emplace_back(
+        result.push_back(Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointOperation, 1,
+                                                   Tegra::SubmissionMode::Increasing));
+        result.push_back(
             BuildFenceAction(Tegra::Engines::Puller::FenceOperation::Increment, fence.id));
     }
 
     return result;
 }
 
-static std::vector<Tegra::CommandHeader> BuildIncrementWithWfiCommandList(NvFence fence) {
-    std::vector<Tegra::CommandHeader> result{
+static boost::container::small_vector<Tegra::CommandHeader, 512> BuildIncrementWithWfiCommandList(
+    NvFence fence) {
+    boost::container::small_vector<Tegra::CommandHeader, 512> result{
         Tegra::BuildCommandHeader(Tegra::BufferMethods::WaitForIdle, 1,
                                   Tegra::SubmissionMode::Increasing),
         {}};
-    const std::vector<Tegra::CommandHeader> increment{BuildIncrementCommandList(fence)};
-
-    result.insert(result.end(), increment.begin(), increment.end());
-
+    auto increment_list{BuildIncrementCommandList(fence)};
+    result.insert(result.end(), increment_list.begin(), increment_list.end());
     return result;
 }
 
-NvResult nvhost_gpu::SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::vector<u8>& output,
+NvResult nvhost_gpu::SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::span<u8> output,
                                       Tegra::CommandList&& entries) {
     LOG_TRACE(Service_NVDRV, "called, gpfifo={:X}, num_entries={:X}, flags={:X}", params.address,
               params.num_entries, params.flags.raw);
@@ -293,7 +294,7 @@ NvResult nvhost_gpu::SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::vector<u8>
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::SubmitGPFIFOBase(std::span<const u8> input, std::vector<u8>& output,
+NvResult nvhost_gpu::SubmitGPFIFOBase(std::span<const u8> input, std::span<u8> output,
                                       bool kickoff) {
     if (input.size() < sizeof(IoctlSubmitGpfifo)) {
         UNIMPLEMENTED();
@@ -315,7 +316,7 @@ NvResult nvhost_gpu::SubmitGPFIFOBase(std::span<const u8> input, std::vector<u8>
 }
 
 NvResult nvhost_gpu::SubmitGPFIFOBase(std::span<const u8> input, std::span<const u8> input_inline,
-                                      std::vector<u8>& output) {
+                                      std::span<u8> output) {
     if (input.size() < sizeof(IoctlSubmitGpfifo)) {
         UNIMPLEMENTED();
         return NvResult::InvalidSize;
@@ -327,7 +328,7 @@ NvResult nvhost_gpu::SubmitGPFIFOBase(std::span<const u8> input, std::span<const
     return SubmitGPFIFOImpl(params, output, std::move(entries));
 }
 
-NvResult nvhost_gpu::GetWaitbase(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::GetWaitbase(std::span<const u8> input, std::span<u8> output) {
     IoctlGetWaitbase params{};
     std::memcpy(&params, input.data(), sizeof(IoctlGetWaitbase));
     LOG_INFO(Service_NVDRV, "called, unknown=0x{:X}", params.unknown);
@@ -337,7 +338,7 @@ NvResult nvhost_gpu::GetWaitbase(std::span<const u8> input, std::vector<u8>& out
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::ChannelSetTimeout(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::ChannelSetTimeout(std::span<const u8> input, std::span<u8> output) {
     IoctlChannelSetTimeout params{};
     std::memcpy(&params, input.data(), sizeof(IoctlChannelSetTimeout));
     LOG_INFO(Service_NVDRV, "called, timeout=0x{:X}", params.timeout);
@@ -345,7 +346,7 @@ NvResult nvhost_gpu::ChannelSetTimeout(std::span<const u8> input, std::vector<u8
     return NvResult::Success;
 }
 
-NvResult nvhost_gpu::ChannelSetTimeslice(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_gpu::ChannelSetTimeslice(std::span<const u8> input, std::span<u8> output) {
     IoctlSetTimeslice params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSetTimeslice));
     LOG_INFO(Service_NVDRV, "called, timeslice=0x{:X}", params.timeslice);
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
index 3ca58202d..529c20526 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
@@ -41,11 +41,11 @@ public:
     ~nvhost_gpu() override;
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
@@ -186,23 +186,23 @@ private:
     u32_le channel_priority{};
     u32_le channel_timeslice{};
 
-    NvResult SetNVMAPfd(std::span<const u8> input, std::vector<u8>& output);
-    NvResult SetClientData(std::span<const u8> input, std::vector<u8>& output);
-    NvResult GetClientData(std::span<const u8> input, std::vector<u8>& output);
-    NvResult ZCullBind(std::span<const u8> input, std::vector<u8>& output);
-    NvResult SetErrorNotifier(std::span<const u8> input, std::vector<u8>& output);
-    NvResult SetChannelPriority(std::span<const u8> input, std::vector<u8>& output);
-    NvResult AllocGPFIFOEx2(std::span<const u8> input, std::vector<u8>& output);
-    NvResult AllocateObjectContext(std::span<const u8> input, std::vector<u8>& output);
-    NvResult SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::vector<u8>& output,
+    NvResult SetNVMAPfd(std::span<const u8> input, std::span<u8> output);
+    NvResult SetClientData(std::span<const u8> input, std::span<u8> output);
+    NvResult GetClientData(std::span<const u8> input, std::span<u8> output);
+    NvResult ZCullBind(std::span<const u8> input, std::span<u8> output);
+    NvResult SetErrorNotifier(std::span<const u8> input, std::span<u8> output);
+    NvResult SetChannelPriority(std::span<const u8> input, std::span<u8> output);
+    NvResult AllocGPFIFOEx2(std::span<const u8> input, std::span<u8> output);
+    NvResult AllocateObjectContext(std::span<const u8> input, std::span<u8> output);
+    NvResult SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::span<u8> output,
                               Tegra::CommandList&& entries);
-    NvResult SubmitGPFIFOBase(std::span<const u8> input, std::vector<u8>& output,
+    NvResult SubmitGPFIFOBase(std::span<const u8> input, std::span<u8> output,
                               bool kickoff = false);
     NvResult SubmitGPFIFOBase(std::span<const u8> input, std::span<const u8> input_inline,
-                              std::vector<u8>& output);
-    NvResult GetWaitbase(std::span<const u8> input, std::vector<u8>& output);
-    NvResult ChannelSetTimeout(std::span<const u8> input, std::vector<u8>& output);
-    NvResult ChannelSetTimeslice(std::span<const u8> input, std::vector<u8>& output);
+                              std::span<u8> output);
+    NvResult GetWaitbase(std::span<const u8> input, std::span<u8> output);
+    NvResult ChannelSetTimeout(std::span<const u8> input, std::span<u8> output);
+    NvResult ChannelSetTimeslice(std::span<const u8> input, std::span<u8> output);
 
     EventInterface& events_interface;
     NvCore::Container& core;
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
index dc45169ad..a174442a6 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
@@ -16,7 +16,7 @@ nvhost_nvdec::nvhost_nvdec(Core::System& system_, NvCore::Container& core_)
 nvhost_nvdec::~nvhost_nvdec() = default;
 
 NvResult nvhost_nvdec::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::vector<u8>& output) {
+                              std::span<u8> output) {
     switch (command.group) {
     case 0x0:
         switch (command.cmd) {
@@ -56,13 +56,13 @@ NvResult nvhost_nvdec::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> in
 }
 
 NvResult nvhost_nvdec::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::span<const u8> inline_input, std::vector<u8>& output) {
+                              std::span<const u8> inline_input, std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_nvdec::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::vector<u8>& output, std::vector<u8>& inline_output) {
+                              std::span<u8> output, std::span<u8> inline_output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
index 0d615bbcb..ad2233c49 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
@@ -14,11 +14,11 @@ public:
     ~nvhost_nvdec() override;
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
index 1ab51f10b..61649aa4a 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
@@ -36,7 +36,7 @@ std::size_t SliceVectors(std::span<const u8> input, std::vector<T>& dst, std::si
 // Writes the data in src to an offset into the dst vector. The offset is specified in bytes
 // Returns the number of bytes written into dst.
 template <typename T>
-std::size_t WriteVectors(std::vector<u8>& dst, const std::vector<T>& src, std::size_t offset) {
+std::size_t WriteVectors(std::span<u8> dst, const std::vector<T>& src, std::size_t offset) {
     if (src.empty()) {
         return 0;
     }
@@ -72,8 +72,7 @@ NvResult nvhost_nvdec_common::SetNVMAPfd(std::span<const u8> input) {
     return NvResult::Success;
 }
 
-NvResult nvhost_nvdec_common::Submit(DeviceFD fd, std::span<const u8> input,
-                                     std::vector<u8>& output) {
+NvResult nvhost_nvdec_common::Submit(DeviceFD fd, std::span<const u8> input, std::span<u8> output) {
     IoctlSubmit params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmit));
     LOG_DEBUG(Service_NVDRV, "called NVDEC Submit, cmd_buffer_count={}", params.cmd_buffer_count);
@@ -121,7 +120,7 @@ NvResult nvhost_nvdec_common::Submit(DeviceFD fd, std::span<const u8> input,
     return NvResult::Success;
 }
 
-NvResult nvhost_nvdec_common::GetSyncpoint(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_nvdec_common::GetSyncpoint(std::span<const u8> input, std::span<u8> output) {
     IoctlGetSyncpoint params{};
     std::memcpy(&params, input.data(), sizeof(IoctlGetSyncpoint));
     LOG_DEBUG(Service_NVDRV, "called GetSyncpoint, id={}", params.param);
@@ -133,7 +132,7 @@ NvResult nvhost_nvdec_common::GetSyncpoint(std::span<const u8> input, std::vecto
     return NvResult::Success;
 }
 
-NvResult nvhost_nvdec_common::GetWaitbase(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_nvdec_common::GetWaitbase(std::span<const u8> input, std::span<u8> output) {
     IoctlGetWaitbase params{};
     LOG_CRITICAL(Service_NVDRV, "called WAITBASE");
     std::memcpy(&params, input.data(), sizeof(IoctlGetWaitbase));
@@ -142,7 +141,7 @@ NvResult nvhost_nvdec_common::GetWaitbase(std::span<const u8> input, std::vector
     return NvResult::Success;
 }
 
-NvResult nvhost_nvdec_common::MapBuffer(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_nvdec_common::MapBuffer(std::span<const u8> input, std::span<u8> output) {
     IoctlMapBuffer params{};
     std::memcpy(&params, input.data(), sizeof(IoctlMapBuffer));
     std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
@@ -159,7 +158,7 @@ NvResult nvhost_nvdec_common::MapBuffer(std::span<const u8> input, std::vector<u
     return NvResult::Success;
 }
 
-NvResult nvhost_nvdec_common::UnmapBuffer(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_nvdec_common::UnmapBuffer(std::span<const u8> input, std::span<u8> output) {
     IoctlMapBuffer params{};
     std::memcpy(&params, input.data(), sizeof(IoctlMapBuffer));
     std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
@@ -173,7 +172,7 @@ NvResult nvhost_nvdec_common::UnmapBuffer(std::span<const u8> input, std::vector
     return NvResult::Success;
 }
 
-NvResult nvhost_nvdec_common::SetSubmitTimeout(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_nvdec_common::SetSubmitTimeout(std::span<const u8> input, std::span<u8> output) {
     std::memcpy(&submit_timeout, input.data(), input.size());
     LOG_WARNING(Service_NVDRV, "(STUBBED) called");
     return NvResult::Success;
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h
index 5af26a26f..9bb573bfe 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h
@@ -108,12 +108,12 @@ protected:
 
     /// Ioctl command implementations
     NvResult SetNVMAPfd(std::span<const u8> input);
-    NvResult Submit(DeviceFD fd, std::span<const u8> input, std::vector<u8>& output);
-    NvResult GetSyncpoint(std::span<const u8> input, std::vector<u8>& output);
-    NvResult GetWaitbase(std::span<const u8> input, std::vector<u8>& output);
-    NvResult MapBuffer(std::span<const u8> input, std::vector<u8>& output);
-    NvResult UnmapBuffer(std::span<const u8> input, std::vector<u8>& output);
-    NvResult SetSubmitTimeout(std::span<const u8> input, std::vector<u8>& output);
+    NvResult Submit(DeviceFD fd, std::span<const u8> input, std::span<u8> output);
+    NvResult GetSyncpoint(std::span<const u8> input, std::span<u8> output);
+    NvResult GetWaitbase(std::span<const u8> input, std::span<u8> output);
+    NvResult MapBuffer(std::span<const u8> input, std::span<u8> output);
+    NvResult UnmapBuffer(std::span<const u8> input, std::span<u8> output);
+    NvResult SetSubmitTimeout(std::span<const u8> input, std::span<u8> output);
 
     Kernel::KEvent* QueryEvent(u32 event_id) override;
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp
index 39f30e7c8..a05c8cdae 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp
@@ -13,7 +13,7 @@ nvhost_nvjpg::nvhost_nvjpg(Core::System& system_) : nvdevice{system_} {}
 nvhost_nvjpg::~nvhost_nvjpg() = default;
 
 NvResult nvhost_nvjpg::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::vector<u8>& output) {
+                              std::span<u8> output) {
     switch (command.group) {
     case 'H':
         switch (command.cmd) {
@@ -32,13 +32,13 @@ NvResult nvhost_nvjpg::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> in
 }
 
 NvResult nvhost_nvjpg::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::span<const u8> inline_input, std::vector<u8>& output) {
+                              std::span<const u8> inline_input, std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_nvjpg::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                              std::vector<u8>& output, std::vector<u8>& inline_output) {
+                              std::span<u8> output, std::span<u8> inline_output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
@@ -46,7 +46,7 @@ NvResult nvhost_nvjpg::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> in
 void nvhost_nvjpg::OnOpen(DeviceFD fd) {}
 void nvhost_nvjpg::OnClose(DeviceFD fd) {}
 
-NvResult nvhost_nvjpg::SetNVMAPfd(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvhost_nvjpg::SetNVMAPfd(std::span<const u8> input, std::span<u8> output) {
     IoctlSetNvmapFD params{};
     std::memcpy(&params, input.data(), input.size());
     LOG_DEBUG(Service_NVDRV, "called, fd={}", params.nvmap_fd);
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h b/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h
index 41b57e872..5623e0d47 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h
@@ -16,11 +16,11 @@ public:
     ~nvhost_nvjpg() override;
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
@@ -33,7 +33,7 @@ private:
 
     s32_le nvmap_fd{};
 
-    NvResult SetNVMAPfd(std::span<const u8> input, std::vector<u8>& output);
+    NvResult SetNVMAPfd(std::span<const u8> input, std::span<u8> output);
 };
 
 } // namespace Service::Nvidia::Devices
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp b/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
index b0ea402a7..c0b8684c3 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
@@ -16,7 +16,7 @@ nvhost_vic::nvhost_vic(Core::System& system_, NvCore::Container& core_)
 nvhost_vic::~nvhost_vic() = default;
 
 NvResult nvhost_vic::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::vector<u8>& output) {
+                            std::span<u8> output) {
     switch (command.group) {
     case 0x0:
         switch (command.cmd) {
@@ -56,13 +56,13 @@ NvResult nvhost_vic::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> inpu
 }
 
 NvResult nvhost_vic::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::span<const u8> inline_input, std::vector<u8>& output) {
+                            std::span<const u8> inline_input, std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_vic::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                            std::vector<u8>& output, std::vector<u8>& inline_output) {
+                            std::span<u8> output, std::span<u8> inline_output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_vic.h b/src/core/hle/service/nvdrv/devices/nvhost_vic.h
index b5e350a83..cadbcb0a5 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_vic.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_vic.h
@@ -13,11 +13,11 @@ public:
     ~nvhost_vic();
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
diff --git a/src/core/hle/service/nvdrv/devices/nvmap.cpp b/src/core/hle/service/nvdrv/devices/nvmap.cpp
index 07417f045..e7f7e273b 100644
--- a/src/core/hle/service/nvdrv/devices/nvmap.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvmap.cpp
@@ -26,7 +26,7 @@ nvmap::nvmap(Core::System& system_, NvCore::Container& container_)
 nvmap::~nvmap() = default;
 
 NvResult nvmap::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                       std::vector<u8>& output) {
+                       std::span<u8> output) {
     switch (command.group) {
     case 0x1:
         switch (command.cmd) {
@@ -55,13 +55,13 @@ NvResult nvmap::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
 }
 
 NvResult nvmap::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                       std::span<const u8> inline_input, std::vector<u8>& output) {
+                       std::span<const u8> inline_input, std::span<u8> output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
-NvResult nvmap::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                       std::vector<u8>& output, std::vector<u8>& inline_output) {
+NvResult nvmap::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                       std::span<u8> inline_output) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
@@ -69,7 +69,7 @@ NvResult nvmap::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
 void nvmap::OnOpen(DeviceFD fd) {}
 void nvmap::OnClose(DeviceFD fd) {}
 
-NvResult nvmap::IocCreate(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvmap::IocCreate(std::span<const u8> input, std::span<u8> output) {
     IocCreateParams params;
     std::memcpy(&params, input.data(), sizeof(params));
     LOG_DEBUG(Service_NVDRV, "called, size=0x{:08X}", params.size);
@@ -89,7 +89,7 @@ NvResult nvmap::IocCreate(std::span<const u8> input, std::vector<u8>& output) {
     return NvResult::Success;
 }
 
-NvResult nvmap::IocAlloc(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvmap::IocAlloc(std::span<const u8> input, std::span<u8> output) {
     IocAllocParams params;
     std::memcpy(&params, input.data(), sizeof(params));
     LOG_DEBUG(Service_NVDRV, "called, addr={:X}", params.address);
@@ -137,7 +137,7 @@ NvResult nvmap::IocAlloc(std::span<const u8> input, std::vector<u8>& output) {
     return result;
 }
 
-NvResult nvmap::IocGetId(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvmap::IocGetId(std::span<const u8> input, std::span<u8> output) {
     IocGetIdParams params;
     std::memcpy(&params, input.data(), sizeof(params));
 
@@ -161,7 +161,7 @@ NvResult nvmap::IocGetId(std::span<const u8> input, std::vector<u8>& output) {
     return NvResult::Success;
 }
 
-NvResult nvmap::IocFromId(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvmap::IocFromId(std::span<const u8> input, std::span<u8> output) {
     IocFromIdParams params;
     std::memcpy(&params, input.data(), sizeof(params));
 
@@ -192,7 +192,7 @@ NvResult nvmap::IocFromId(std::span<const u8> input, std::vector<u8>& output) {
     return NvResult::Success;
 }
 
-NvResult nvmap::IocParam(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvmap::IocParam(std::span<const u8> input, std::span<u8> output) {
     enum class ParamTypes { Size = 1, Alignment = 2, Base = 3, Heap = 4, Kind = 5, Compr = 6 };
 
     IocParamParams params;
@@ -241,7 +241,7 @@ NvResult nvmap::IocParam(std::span<const u8> input, std::vector<u8>& output) {
     return NvResult::Success;
 }
 
-NvResult nvmap::IocFree(std::span<const u8> input, std::vector<u8>& output) {
+NvResult nvmap::IocFree(std::span<const u8> input, std::span<u8> output) {
     IocFreeParams params;
     std::memcpy(&params, input.data(), sizeof(params));
 
diff --git a/src/core/hle/service/nvdrv/devices/nvmap.h b/src/core/hle/service/nvdrv/devices/nvmap.h
index 82bd3b118..40c65b430 100644
--- a/src/core/hle/service/nvdrv/devices/nvmap.h
+++ b/src/core/hle/service/nvdrv/devices/nvmap.h
@@ -27,11 +27,11 @@ public:
     nvmap& operator=(const nvmap&) = delete;
 
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::vector<u8>& output) override;
+                    std::span<u8> output) override;
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output) override;
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::span<const u8> inline_input, std::span<u8> output) override;
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output) override;
 
     void OnOpen(DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
@@ -106,12 +106,12 @@ private:
     };
     static_assert(sizeof(IocGetIdParams) == 8, "IocGetIdParams has wrong size");
 
-    NvResult IocCreate(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocAlloc(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocGetId(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocFromId(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocParam(std::span<const u8> input, std::vector<u8>& output);
-    NvResult IocFree(std::span<const u8> input, std::vector<u8>& output);
+    NvResult IocCreate(std::span<const u8> input, std::span<u8> output);
+    NvResult IocAlloc(std::span<const u8> input, std::span<u8> output);
+    NvResult IocGetId(std::span<const u8> input, std::span<u8> output);
+    NvResult IocFromId(std::span<const u8> input, std::span<u8> output);
+    NvResult IocParam(std::span<const u8> input, std::span<u8> output);
+    NvResult IocFree(std::span<const u8> input, std::span<u8> output);
 
     NvCore::Container& container;
     NvCore::NvMap& file;
diff --git a/src/core/hle/service/nvdrv/nvdrv.cpp b/src/core/hle/service/nvdrv/nvdrv.cpp
index 3d774eec4..9e46ee8dd 100644
--- a/src/core/hle/service/nvdrv/nvdrv.cpp
+++ b/src/core/hle/service/nvdrv/nvdrv.cpp
@@ -130,7 +130,7 @@ DeviceFD Module::Open(const std::string& device_name) {
 }
 
 NvResult Module::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                        std::vector<u8>& output) {
+                        std::span<u8> output) {
     if (fd < 0) {
         LOG_ERROR(Service_NVDRV, "Invalid DeviceFD={}!", fd);
         return NvResult::InvalidState;
@@ -147,7 +147,7 @@ NvResult Module::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
 }
 
 NvResult Module::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                        std::span<const u8> inline_input, std::vector<u8>& output) {
+                        std::span<const u8> inline_input, std::span<u8> output) {
     if (fd < 0) {
         LOG_ERROR(Service_NVDRV, "Invalid DeviceFD={}!", fd);
         return NvResult::InvalidState;
@@ -163,8 +163,8 @@ NvResult Module::Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
     return itr->second->Ioctl2(fd, command, input, inline_input, output);
 }
 
-NvResult Module::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                        std::vector<u8>& output, std::vector<u8>& inline_output) {
+NvResult Module::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                        std::span<u8> inline_output) {
     if (fd < 0) {
         LOG_ERROR(Service_NVDRV, "Invalid DeviceFD={}!", fd);
         return NvResult::InvalidState;
diff --git a/src/core/hle/service/nvdrv/nvdrv.h b/src/core/hle/service/nvdrv/nvdrv.h
index 668be742b..d8622b3ca 100644
--- a/src/core/hle/service/nvdrv/nvdrv.h
+++ b/src/core/hle/service/nvdrv/nvdrv.h
@@ -80,13 +80,13 @@ public:
     DeviceFD Open(const std::string& device_name);
 
     /// Sends an ioctl command to the specified file descriptor.
-    NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output);
+    NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output);
 
     NvResult Ioctl2(DeviceFD fd, Ioctl command, std::span<const u8> input,
-                    std::span<const u8> inline_input, std::vector<u8>& output);
+                    std::span<const u8> inline_input, std::span<u8> output);
 
-    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output);
+    NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
+                    std::span<u8> inline_output);
 
     /// Closes a device file descriptor and returns operation success.
     NvResult Close(DeviceFD fd);
diff --git a/src/core/hle/service/nvdrv/nvdrv_interface.cpp b/src/core/hle/service/nvdrv/nvdrv_interface.cpp
index d010a1e03..348207e25 100644
--- a/src/core/hle/service/nvdrv/nvdrv_interface.cpp
+++ b/src/core/hle/service/nvdrv/nvdrv_interface.cpp
@@ -63,12 +63,12 @@ void NVDRV::Ioctl1(HLERequestContext& ctx) {
     }
 
     // Check device
-    std::vector<u8> output_buffer(ctx.GetWriteBufferSize(0));
+    tmp_output.resize_destructive(ctx.GetWriteBufferSize(0));
     const auto input_buffer = ctx.ReadBuffer(0);
 
-    const auto nv_result = nvdrv->Ioctl1(fd, command, input_buffer, output_buffer);
+    const auto nv_result = nvdrv->Ioctl1(fd, command, input_buffer, tmp_output);
     if (command.is_out != 0) {
-        ctx.WriteBuffer(output_buffer);
+        ctx.WriteBuffer(tmp_output);
     }
 
     IPC::ResponseBuilder rb{ctx, 3};
@@ -90,12 +90,12 @@ void NVDRV::Ioctl2(HLERequestContext& ctx) {
 
     const auto input_buffer = ctx.ReadBuffer(0);
     const auto input_inlined_buffer = ctx.ReadBuffer(1);
-    std::vector<u8> output_buffer(ctx.GetWriteBufferSize(0));
+    tmp_output.resize_destructive(ctx.GetWriteBufferSize(0));
 
     const auto nv_result =
-        nvdrv->Ioctl2(fd, command, input_buffer, input_inlined_buffer, output_buffer);
+        nvdrv->Ioctl2(fd, command, input_buffer, input_inlined_buffer, tmp_output);
     if (command.is_out != 0) {
-        ctx.WriteBuffer(output_buffer);
+        ctx.WriteBuffer(tmp_output);
     }
 
     IPC::ResponseBuilder rb{ctx, 3};
@@ -116,14 +116,12 @@ void NVDRV::Ioctl3(HLERequestContext& ctx) {
     }
 
     const auto input_buffer = ctx.ReadBuffer(0);
-    std::vector<u8> output_buffer(ctx.GetWriteBufferSize(0));
-    std::vector<u8> output_buffer_inline(ctx.GetWriteBufferSize(1));
-
-    const auto nv_result =
-        nvdrv->Ioctl3(fd, command, input_buffer, output_buffer, output_buffer_inline);
+    tmp_output.resize_destructive(ctx.GetWriteBufferSize(0));
+    tmp_output_inline.resize_destructive(ctx.GetWriteBufferSize(1));
+    const auto nv_result = nvdrv->Ioctl3(fd, command, input_buffer, tmp_output, tmp_output_inline);
     if (command.is_out != 0) {
-        ctx.WriteBuffer(output_buffer, 0);
-        ctx.WriteBuffer(output_buffer_inline, 1);
+        ctx.WriteBuffer(tmp_output, 0);
+        ctx.WriteBuffer(tmp_output_inline, 1);
     }
 
     IPC::ResponseBuilder rb{ctx, 3};
diff --git a/src/core/hle/service/nvdrv/nvdrv_interface.h b/src/core/hle/service/nvdrv/nvdrv_interface.h
index 881ea1a6b..4b593ff90 100644
--- a/src/core/hle/service/nvdrv/nvdrv_interface.h
+++ b/src/core/hle/service/nvdrv/nvdrv_interface.h
@@ -4,6 +4,7 @@
 #pragma once
 
 #include <memory>
+#include "common/scratch_buffer.h"
 #include "core/hle/service/nvdrv/nvdrv.h"
 #include "core/hle/service/service.h"
 
@@ -33,6 +34,8 @@ private:
 
     u64 pid{};
     bool is_initialized{};
+    Common::ScratchBuffer<u8> tmp_output;
+    Common::ScratchBuffer<u8> tmp_output_inline;
 };
 
 } // namespace Service::Nvidia
diff --git a/src/core/hle/service/nvnflinger/parcel.h b/src/core/hle/service/nvnflinger/parcel.h
index fb56d75d7..23ba315a0 100644
--- a/src/core/hle/service/nvnflinger/parcel.h
+++ b/src/core/hle/service/nvnflinger/parcel.h
@@ -6,6 +6,7 @@
 #include <memory>
 #include <span>
 #include <vector>
+#include <boost/container/small_vector.hpp>
 
 #include "common/alignment.h"
 #include "common/assert.h"
@@ -167,7 +168,7 @@ public:
 private:
     template <typename T>
         requires(std::is_trivially_copyable_v<T>)
-    void WriteImpl(const T& val, std::vector<u8>& buffer) {
+    void WriteImpl(const T& val, boost::container::small_vector<u8, 0x200>& buffer) {
         const size_t aligned_size = Common::AlignUp(sizeof(T), 4);
         const size_t old_size = buffer.size();
         buffer.resize(old_size + aligned_size);
@@ -176,8 +177,8 @@ private:
     }
 
 private:
-    std::vector<u8> m_data_buffer;
-    std::vector<u8> m_object_buffer;
+    boost::container::small_vector<u8, 0x200> m_data_buffer;
+    boost::container::small_vector<u8, 0x200> m_object_buffer;
 };
 
 } // namespace Service::android