"Merge Tagged PR 6598"

.ci/templates/build-msvc.yml

@@ -8,7 +8,7 @@ steps:
   displayName: 'Install vulkan-sdk'
 - script: python -m pip install --upgrade pip conan
   displayName: 'Install conan'
-- script: refreshenv && mkdir build && cd build && cmake -G "Visual Studio 16 2019" -A x64 -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DUSE_DISCORD_PRESENCE=ON -DENABLE_QT_TRANSLATION=ON -DDISPLAY_VERSION=${{ parameters['version'] }} -DCMAKE_BUILD_TYPE=Release .. && cd ..
+- script: refreshenv && mkdir build && cd build && cmake -E env CXXFLAGS="/Gw /GA /Gr /Ob2" cmake -G "Visual Studio 16 2019" -A x64 -DCMAKE_INTERPROCEDURAL_OPTIMIZATION=ON -DCMAKE_POLICY_DEFAULT_CMP0069=NEW -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DUSE_DISCORD_PRESENCE=ON -DENABLE_QT_TRANSLATION=ON -DDISPLAY_VERSION=${{ parameters['version'] }} -DCMAKE_BUILD_TYPE=Release .. && cd ..
   displayName: 'Configure CMake'
 - task: MSBuild@1
   displayName: 'Build'

src/CMakeLists.txt

@@ -37,7 +37,8 @@ if (MSVC)
     add_compile_options(
         /MP
         /Zi
-        /Zm200
+        /Zm300
+        /Zf
         /Zo
         /permissive-
         /EHsc
@@ -81,7 +82,7 @@ if (MSVC)
     add_compile_options("$<$<CONFIG:Release>:/GS->")
 
     set(CMAKE_EXE_LINKER_FLAGS_DEBUG   "/DEBUG /MANIFEST:NO" CACHE STRING "" FORCE)
-    set(CMAKE_EXE_LINKER_FLAGS_RELEASE "/DEBUG /MANIFEST:NO /INCREMENTAL:NO /OPT:REF,ICF" CACHE STRING "" FORCE)
+    set(CMAKE_EXE_LINKER_FLAGS_RELEASE "${CMAKE_EXE_LINKER_FLAGS_RELEASE} /DEBUG /MANIFEST:NO /INCREMENTAL:NO /OPT:REF,ICF" CACHE STRING "" FORCE)
 else()
     add_compile_options(
         -Wall

src/audio_core/stream.cpp

@@ -87,6 +87,14 @@ static void VolumeAdjustSamples(std::vector<s16>& samples, float game_volume) {
 }
 
 void Stream::PlayNextBuffer(std::chrono::nanoseconds ns_late) {
+    auto now = std::chrono::steady_clock::now();
+    auto duration = now.time_since_epoch();
+    auto nanoseconds = std::chrono::duration_cast<std::chrono::nanoseconds>(duration);
+
+    if (nanoseconds > expected_cb_time) {
+        ns_late = nanoseconds - expected_cb_time;
+    }
+
     if (!IsPlaying()) {
         // Ensure we are in playing state before playing the next buffer
         sink_stream.Flush();
@@ -121,6 +129,7 @@ void Stream::PlayNextBuffer(std::chrono::nanoseconds ns_late) {
         ns_late = {};
     }
 
+    expected_cb_time = nanoseconds + (buffer_release_ns - ns_late);
     core_timing.ScheduleEvent(buffer_release_ns - ns_late, release_event, {});
 }
 

src/audio_core/stream.h

@@ -117,13 +117,14 @@ private:
     ReleaseCallback release_callback; ///< Buffer release callback for the stream
     State state{State::Stopped};      ///< Playback state of the stream
     std::shared_ptr<Core::Timing::EventType>
-        release_event;                      ///< Core timing release event for the stream
-    BufferPtr active_buffer;                ///< Actively playing buffer in the stream
-    std::queue<BufferPtr> queued_buffers;   ///< Buffers queued to be played in the stream
-    std::queue<BufferPtr> released_buffers; ///< Buffers recently released from the stream
-    SinkStream& sink_stream;                ///< Output sink for the stream
-    Core::Timing::CoreTiming& core_timing;  ///< Core timing instance.
-    std::string name;                       ///< Name of the stream, must be unique
+        release_event;                              ///< Core timing release event for the stream
+    BufferPtr active_buffer;                        ///< Actively playing buffer in the stream
+    std::queue<BufferPtr> queued_buffers;           ///< Buffers queued to be played in the stream
+    std::queue<BufferPtr> released_buffers;         ///< Buffers recently released from the stream
+    SinkStream& sink_stream;                        ///< Output sink for the stream
+    Core::Timing::CoreTiming& core_timing;          ///< Core timing instance.
+    std::string name;                               ///< Name of the stream, must be unique
+    std::chrono::nanoseconds expected_cb_time = {}; ///< Estimated time of next callback
 };
 
 using StreamPtr = std::shared_ptr<Stream>;

src/common/assert.h

@@ -32,14 +32,14 @@ assert_noinline_call(const Fn& fn) {
 
 #define ASSERT(_a_)                                                                                \
     do                                                                                             \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             assert_noinline_call([] { LOG_CRITICAL(Debug, "Assertion Failed!"); });                \
         }                                                                                          \
     while (0)
 
 #define ASSERT_MSG(_a_, ...)                                                                       \
     do                                                                                             \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             assert_noinline_call([&] { LOG_CRITICAL(Debug, "Assertion Failed!\n" __VA_ARGS__); }); \
         }                                                                                          \
     while (0)
@@ -70,7 +70,7 @@ assert_noinline_call(const Fn& fn) {
 #define ASSERT_OR_EXECUTE(_a_, _b_)                                                                \
     do {                                                                                           \
         ASSERT(_a_);                                                                               \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             _b_                                                                                    \
         }                                                                                          \
     } while (0)
@@ -79,7 +79,7 @@ assert_noinline_call(const Fn& fn) {
 #define ASSERT_OR_EXECUTE_MSG(_a_, _b_, ...)                                                       \
     do {                                                                                           \
         ASSERT_MSG(_a_, __VA_ARGS__);                                                              \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             _b_                                                                                    \
         }                                                                                          \
     } while (0)

src/video_core/buffer_cache/buffer_cache.h

@@ -131,6 +131,8 @@ public:
 
     void DownloadMemory(VAddr cpu_addr, u64 size);
 
+    bool InlineMemory(VAddr dest_address, size_t copy_size, std::span<u8> inlined_buffer);
+
     void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size);
 
     void DisableGraphicsUniformBuffer(size_t stage, u32 index);
@@ -808,6 +810,8 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
         return;
     }
     MICROPROFILE_SCOPE(GPU_DownloadMemory);
+    const bool is_accuracy_normal =
+        Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::Normal;
 
     boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads;
     u64 total_size_bytes = 0;
@@ -819,6 +823,9 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
             ForEachBufferInRange(cpu_addr, size, [&](BufferId buffer_id, Buffer& buffer) {
                 buffer.ForEachDownloadRangeAndClear(
                     cpu_addr, size, [&](u64 range_offset, u64 range_size) {
+                        if (is_accuracy_normal) {
+                            return;
+                        }
                         const VAddr buffer_addr = buffer.CpuAddr();
                         const auto add_download = [&](VAddr start, VAddr end) {
                             const u64 new_offset = start - buffer_addr;
@@ -1417,10 +1424,8 @@ void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 s
     const IntervalType base_interval{cpu_addr, cpu_addr + size};
     common_ranges.add(base_interval);
 
-    const bool is_accuracy_high =
-        Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::High;
     const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue();
-    if (!is_async && !is_accuracy_high) {
+    if (!is_async) {
         return;
     }
     uncommitted_ranges.add(base_interval);
@@ -1643,6 +1648,41 @@ void BufferCache<P>::MappedUploadMemory(Buffer& buffer, u64 total_size_bytes,
     runtime.CopyBuffer(buffer, upload_staging.buffer, copies);
 }
 
+template <class P>
+bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size,
+                                  std::span<u8> inlined_buffer) {
+    const bool is_dirty = IsRegionRegistered(dest_address, copy_size);
+    if (!is_dirty) {
+        return false;
+    }
+    if (!IsRegionGpuModified(dest_address, copy_size)) {
+        return false;
+    }
+
+    const IntervalType subtract_interval{dest_address, dest_address + copy_size};
+    ClearDownload(subtract_interval);
+
+    BufferId buffer_id = FindBuffer(dest_address, static_cast<u32>(copy_size));
+    auto& buffer = slot_buffers[buffer_id];
+    SynchronizeBuffer(buffer, dest_address, static_cast<u32>(copy_size));
+
+    if constexpr (USE_MEMORY_MAPS) {
+        std::array copies{BufferCopy{
+            .src_offset = 0,
+            .dst_offset = buffer.Offset(dest_address),
+            .size = copy_size,
+        }};
+        auto upload_staging = runtime.UploadStagingBuffer(copy_size);
+        u8* const src_pointer = upload_staging.mapped_span.data();
+        std::memcpy(src_pointer, inlined_buffer.data(), copy_size);
+        runtime.CopyBuffer(buffer, upload_staging.buffer, copies);
+    } else {
+        buffer.ImmediateUpload(buffer.Offset(dest_address), inlined_buffer);
+    }
+
+    return true;
+}
+
 template <class P>
 void BufferCache<P>::DownloadBufferMemory(Buffer& buffer) {
     DownloadBufferMemory(buffer, buffer.CpuAddr(), buffer.SizeBytes());

src/video_core/engines/engine_upload.cpp

@@ -7,6 +7,7 @@
 #include "common/assert.h"
 #include "video_core/engines/engine_upload.h"
 #include "video_core/memory_manager.h"
+#include "video_core/rasterizer_interface.h"
 #include "video_core/textures/decoders.h"
 
 namespace Tegra::Engines::Upload {
@@ -16,6 +17,10 @@ State::State(MemoryManager& memory_manager_, Registers& regs_)
 
 State::~State() = default;
 
+void State::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
+    rasterizer = rasterizer_;
+}
+
 void State::ProcessExec(const bool is_linear_) {
     write_offset = 0;
     copy_size = regs.line_length_in * regs.line_count;
@@ -32,7 +37,7 @@ void State::ProcessData(const u32 data, const bool is_last_call) {
     }
     const GPUVAddr address{regs.dest.Address()};
     if (is_linear) {
-        memory_manager.WriteBlock(address, inner_buffer.data(), copy_size);
+        rasterizer->AccelerateInline2Memory(address, copy_size, inner_buffer);
     } else {
         UNIMPLEMENTED_IF(regs.dest.z != 0);
         UNIMPLEMENTED_IF(regs.dest.depth != 1);

src/video_core/engines/engine_upload.h

@@ -12,6 +12,10 @@ namespace Tegra {
 class MemoryManager;
 }
 
+namespace VideoCore {
+class RasterizerInterface;
+}
+
 namespace Tegra::Engines::Upload {
 
 struct Registers {
@@ -60,6 +64,9 @@ public:
     void ProcessExec(bool is_linear_);
     void ProcessData(u32 data, bool is_last_call);
 
+    /// Binds a rasterizer to this engine.
+    void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
+
 private:
     u32 write_offset = 0;
     u32 copy_size = 0;
@@ -68,6 +75,7 @@ private:
     bool is_linear = false;
     Registers& regs;
     MemoryManager& memory_manager;
+    VideoCore::RasterizerInterface* rasterizer = nullptr;
 };
 
 } // namespace Tegra::Engines::Upload

src/video_core/engines/kepler_compute.cpp

@@ -22,6 +22,7 @@ KeplerCompute::~KeplerCompute() = default;
 
 void KeplerCompute::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
     rasterizer = rasterizer_;
+    upload_state.BindRasterizer(rasterizer);
 }
 
 void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_call) {

src/video_core/engines/kepler_memory.cpp

@@ -19,6 +19,10 @@ KeplerMemory::KeplerMemory(Core::System& system_, MemoryManager& memory_manager)
 
 KeplerMemory::~KeplerMemory() = default;
 
+void KeplerMemory::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
+    upload_state.BindRasterizer(rasterizer_);
+}
+
 void KeplerMemory::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
     ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid KeplerMemory register, increase the size of the Regs structure");

src/video_core/engines/kepler_memory.h

@@ -22,6 +22,10 @@ namespace Tegra {
 class MemoryManager;
 }
 
+namespace VideoCore {
+class RasterizerInterface;
+}
+
 namespace Tegra::Engines {
 
 /**
@@ -38,6 +42,9 @@ public:
     explicit KeplerMemory(Core::System& system_, MemoryManager& memory_manager);
     ~KeplerMemory() override;
 
+    /// Binds a rasterizer to this engine.
+    void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
+
     /// Write the value to the register identified by method.
     void CallMethod(u32 method, u32 method_argument, bool is_last_call) override;
 

src/video_core/engines/maxwell_3d.cpp

@@ -31,6 +31,7 @@ Maxwell3D::~Maxwell3D() = default;
 
 void Maxwell3D::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
     rasterizer = rasterizer_;
+    upload_state.BindRasterizer(rasterizer_);
 }
 
 void Maxwell3D::InitializeRegisterDefaults() {

src/video_core/engines/maxwell_3d.h

@@ -1557,7 +1557,8 @@ private:
 
     static constexpr u32 null_cb_data = 0xFFFFFFFF;
     struct CBDataState {
-        std::array<std::array<u32, 0x4000>, 16> buffer;
+        static constexpr size_t inline_size = 0x8000;
+        std::array<std::array<u32, inline_size>, 16> buffer;
         u32 current{null_cb_data};
         u32 id{null_cb_data};
         u32 start_pos{};

src/video_core/gpu.cpp

@@ -59,6 +59,7 @@ struct GPU::Impl {
         maxwell_3d->BindRasterizer(rasterizer);
         fermi_2d->BindRasterizer(rasterizer);
         kepler_compute->BindRasterizer(rasterizer);
+        kepler_memory->BindRasterizer(rasterizer);
         maxwell_dma->BindRasterizer(rasterizer);
     }
 
@@ -502,8 +503,15 @@ struct GPU::Impl {
         case BufferMethods::SemaphoreAddressHigh:
         case BufferMethods::SemaphoreAddressLow:
         case BufferMethods::SemaphoreSequence:
-        case BufferMethods::UnkCacheFlush:
-        case BufferMethods::WrcacheFlush:
+            break;
+        case BufferMethods::UnkCacheFlush: {
+            rasterizer->SyncGuestHost();
+            break;
+        }
+        case BufferMethods::WrcacheFlush: {
+            rasterizer->SignalReference();
+            break;
+        }
         case BufferMethods::FenceValue:
             break;
         case BufferMethods::RefCnt:
@@ -513,7 +521,7 @@ struct GPU::Impl {
             ProcessFenceActionMethod();
             break;
         case BufferMethods::WaitForInterrupt:
-            ProcessWaitForInterruptMethod();
+            rasterizer->WaitForIdle();
             break;
         case BufferMethods::SemaphoreTrigger: {
             ProcessSemaphoreTriggerMethod();

src/video_core/macro/macro_hle.cpp

@@ -4,6 +4,8 @@
 
 #include <array>
 #include <vector>
+#include "common/scope_exit.h"
+#include "video_core/dirty_flags.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
@@ -59,6 +61,7 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.regs.index_array.first = parameters[3];
     maxwell3d.regs.reg_array[0x446] = element_base; // vertex id base?
     maxwell3d.regs.index_array.count = parameters[1];
+    maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
     maxwell3d.regs.vb_element_base = element_base;
     maxwell3d.regs.vb_base_instance = base_instance;
     maxwell3d.mme_draw.instance_count = instance_count;
@@ -81,10 +84,67 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
 }
 
-constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{
+// Multidraw Indirect
+void HLE_3f5e74b9c9a50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
+    SCOPE_EXIT({
+        // Clean everything.
+        maxwell3d.regs.reg_array[0x446] = 0x0; // vertex id base?
+        maxwell3d.regs.index_array.count = 0;
+        maxwell3d.regs.vb_element_base = 0x0;
+        maxwell3d.regs.vb_base_instance = 0x0;
+        maxwell3d.mme_draw.instance_count = 0;
+        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+        maxwell3d.CallMethodFromMME(0x8e4, 0x0);
+        maxwell3d.CallMethodFromMME(0x8e5, 0x0);
+        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+    });
+    const u32 start_indirect = parameters[0];
+    const u32 end_indirect = parameters[1];
+    if (start_indirect >= end_indirect) {
+        // Nothing to do.
+        return;
+    }
+    const auto topology =
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[2]);
+    maxwell3d.regs.draw.topology.Assign(topology);
+    const u32 padding = parameters[3];
+    const std::size_t max_draws = parameters[4];
+
+    const u32 indirect_words = 5 + padding;
+    const std::size_t first_draw = start_indirect;
+    const std::size_t effective_draws = end_indirect - start_indirect;
+    const std::size_t last_draw = start_indirect + std::min(effective_draws, max_draws);
+
+    for (std::size_t index = first_draw; index < last_draw; index++) {
+        const std::size_t base = index * indirect_words + 5;
+        const u32 num_vertices = parameters[base];
+        const u32 instance_count = parameters[base + 1];
+        const u32 first_index = parameters[base + 2];
+        const u32 base_vertex = parameters[base + 3];
+        const u32 base_instance = parameters[base + 4];
+        maxwell3d.regs.index_array.first = first_index;
+        maxwell3d.regs.reg_array[0x446] = base_vertex;
+        maxwell3d.regs.index_array.count = num_vertices;
+        maxwell3d.regs.vb_element_base = base_vertex;
+        maxwell3d.regs.vb_base_instance = base_instance;
+        maxwell3d.mme_draw.instance_count = instance_count;
+        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+        maxwell3d.CallMethodFromMME(0x8e4, base_vertex);
+        maxwell3d.CallMethodFromMME(0x8e5, base_instance);
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+        if (maxwell3d.ShouldExecute()) {
+            maxwell3d.Rasterizer().Draw(true, true);
+        }
+        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+    }
+}
+
+constexpr std::array<std::pair<u64, HLEFunction>, 4> hle_funcs{{
     {0x771BB18C62444DA0, &HLE_771BB18C62444DA0},
     {0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD},
     {0x0217920100488FF7, &HLE_0217920100488FF7},
+    {0x3f5e74b9c9a50164, &HLE_3f5e74b9c9a50164},
 }};
 
 class HLEMacroImpl final : public CachedMacro {
@@ -100,6 +160,7 @@ private:
     Engines::Maxwell3D& maxwell3d;
     HLEFunction func;
 };
+
 } // Anonymous namespace
 
 HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {}

src/video_core/memory_manager.h

@@ -143,6 +143,8 @@ public:
     [[nodiscard]] GPUVAddr Allocate(std::size_t size, std::size_t align);
     void Unmap(GPUVAddr gpu_addr, std::size_t size);
 
+    void FlushRegion(GPUVAddr gpu_addr, size_t size) const;
+
 private:
     [[nodiscard]] PageEntry GetPageEntry(GPUVAddr gpu_addr) const;
     void SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size = page_size);
@@ -153,8 +155,6 @@ private:
     void TryLockPage(PageEntry page_entry, std::size_t size);
     void TryUnlockPage(PageEntry page_entry, std::size_t size);
 
-    void FlushRegion(GPUVAddr gpu_addr, size_t size) const;
-
     void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size,
                        bool is_safe) const;
     void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size,

src/video_core/rasterizer_interface.h

@@ -123,6 +123,9 @@ public:
 
     [[nodiscard]] virtual Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() = 0;
 
+    virtual void AccelerateInline2Memory(GPUVAddr address, size_t copy_size,
+                                         std::span<u8> memory) = 0;
+
     /// Attempt to use a faster method to display the framebuffer to screen
     [[nodiscard]] virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& config,
                                                  VAddr framebuffer_addr, u32 pixel_stride) {

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -484,6 +484,28 @@ Tegra::Engines::AccelerateDMAInterface& RasterizerOpenGL::AccessAccelerateDMA()
     return accelerate_dma;
 }
 
+void RasterizerOpenGL::AccelerateInline2Memory(GPUVAddr address, size_t copy_size,
+                                               std::span<u8> memory) {
+    auto cpu_addr = gpu_memory.GpuToCpuAddress(address);
+    if (!cpu_addr) [[unlikely]] {
+        gpu_memory.WriteBlock(address, memory.data(), copy_size);
+        return;
+    }
+    gpu_memory.WriteBlockUnsafe(address, memory.data(), copy_size);
+    {
+        std::unique_lock<std::mutex> lock{buffer_cache.mutex};
+        if (!buffer_cache.InlineMemory(*cpu_addr, copy_size, memory)) {
+            buffer_cache.WriteMemory(*cpu_addr, copy_size);
+        }
+    }
+    {
+        std::scoped_lock lock_texture{texture_cache.mutex};
+        texture_cache.WriteMemory(*cpu_addr, copy_size);
+    }
+    shader_cache.InvalidateRegion(*cpu_addr, copy_size);
+    query_cache.InvalidateRegion(*cpu_addr, copy_size);
+}
+
 bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
                                          VAddr framebuffer_addr, u32 pixel_stride) {
     if (framebuffer_addr == 0) {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -106,6 +106,7 @@ public:
                                const Tegra::Engines::Fermi2D::Surface& dst,
                                const Tegra::Engines::Fermi2D::Config& copy_config) override;
     Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override;
+    void AccelerateInline2Memory(GPUVAddr address, size_t copy_size, std::span<u8> memory) override;
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
     void LoadDiskResources(u64 title_id, std::stop_token stop_loading,

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -548,6 +548,28 @@ Tegra::Engines::AccelerateDMAInterface& RasterizerVulkan::AccessAccelerateDMA()
     return accelerate_dma;
 }
 
+void RasterizerVulkan::AccelerateInline2Memory(GPUVAddr address, size_t copy_size,
+                                               std::span<u8> memory) {
+    auto cpu_addr = gpu_memory.GpuToCpuAddress(address);
+    if (!cpu_addr) [[unlikely]] {
+        gpu_memory.WriteBlock(address, memory.data(), copy_size);
+        return;
+    }
+    gpu_memory.WriteBlockUnsafe(address, memory.data(), copy_size);
+    {
+        std::unique_lock<std::mutex> lock{buffer_cache.mutex};
+        if (!buffer_cache.InlineMemory(*cpu_addr, copy_size, memory)) {
+            buffer_cache.WriteMemory(*cpu_addr, copy_size);
+        }
+    }
+    {
+        std::scoped_lock lock_texture{texture_cache.mutex};
+        texture_cache.WriteMemory(*cpu_addr, copy_size);
+    }
+    pipeline_cache.InvalidateRegion(*cpu_addr, copy_size);
+    query_cache.InvalidateRegion(*cpu_addr, copy_size);
+}
+
 bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config,
                                          VAddr framebuffer_addr, u32 pixel_stride) {
     if (!framebuffer_addr) {

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -99,6 +99,7 @@ public:
                                const Tegra::Engines::Fermi2D::Surface& dst,
                                const Tegra::Engines::Fermi2D::Config& copy_config) override;
     Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override;
+    void AccelerateInline2Memory(GPUVAddr address, size_t copy_size, std::span<u8> memory) override;
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
     void LoadDiskResources(u64 title_id, std::stop_token stop_loading,