Smol change

Begin and end were landing in different scopes: the clear and draw
texture paths opened it with no pass bound, the compare helpers reopened
it after their resolve pass had closed one, and the pass teardown could
return before pausing, leaving it active at submit.

.github/PULL_REQUEST_TEMPLATE.md

@@ -1,5 +0,0 @@
-- [ ] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions).
-- [ ] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md)
-- [ ] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability.
-
--------------------

src/android/app/build.gradle.kts

@@ -65,7 +65,7 @@ android {
 
     defaultConfig {
         applicationId = "dev.eden.eden_emulator"
-        minSdk = 24
+        minSdk = 33
         targetSdk = 36
         versionName = getGitVersion()
         versionCode = autoVersion

src/android/app/src/main/java/org/yuzu/yuzu_emu/NativeLibrary.kt

@@ -218,6 +218,8 @@ object NativeLibrary {
 
     external fun logSettings()
 
+    external fun refreshThreadPolicies()
+
     external fun getDebugKnobAt(index: Int): Boolean
 
     /**

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt

@@ -27,6 +27,7 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting {
     RENDERER_ASYNCHRONOUS_GPU_EMULATION("use_asynchronous_gpu_emulation"),
     RENDERER_ASYNC_PRESENTATION("async_presentation"),
     RENDERER_ASYNCHRONOUS_SHADERS("use_asynchronous_shaders"),
+    RENDERER_UNIFIED_MEMORY("use_unified_memory"),
     RENDERER_REACTIVE_FLUSHING("use_reactive_flushing"),
     ENABLE_BUFFER_HISTORY("enable_buffer_history"),
     USE_OPTIMIZED_VERTEX_BUFFERS("use_optimized_vertex_buffers"),
@@ -36,6 +37,8 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting {
     RENDERER_DEBUG("debug"),
     RENDERER_PATCH_OLD_QCOM_DRIVERS("patch_old_qcom_drivers"),
     RENDERER_VERTEX_INPUT_DYNAMIC_STATE("vertex_input_dynamic_state"),
+    RENDERER_DYNAMIC_RENDERING("dynamic_rendering"),
+    RENDERER_WORKGROUP_MEMORY_EXPLICIT_LAYOUT("workgroup_memory_explicit_layout"),
     RENDERER_SAMPLE_SHADING("sample_shading"),
     GPU_UNSWIZZLE_ENABLED("gpu_unswizzle_enabled"),
     PICTURE_IN_PICTURE("picture_in_picture"),

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt

@@ -155,6 +155,20 @@ abstract class SettingsItem(
                     descriptionId = R.string.vertex_input_dynamic_state_description
                 )
             )
+            put(
+                SwitchSetting(
+                    BooleanSetting.RENDERER_DYNAMIC_RENDERING,
+                    titleId = R.string.dynamic_rendering,
+                    descriptionId = R.string.dynamic_rendering_description
+                )
+            )
+            put(
+                SwitchSetting(
+                    BooleanSetting.RENDERER_WORKGROUP_MEMORY_EXPLICIT_LAYOUT,
+                    titleId = R.string.workgroup_memory_explicit_layout,
+                    descriptionId = R.string.workgroup_memory_explicit_layout_description
+                )
+            )
             put(
                 SliderSetting(
                     IntSetting.RENDERER_SAMPLE_SHADING,
@@ -594,7 +608,7 @@ abstract class SettingsItem(
                     IntSetting.ANDROID_PIPELINE_WORKERS,
                     titleId = R.string.pipeline_worker_cores,
                     descriptionId = R.string.pipeline_worker_cores_description,
-                    min = 4,
+                    min = 2,
                     max = 8,
                     units = "cores"
                 )
@@ -685,6 +699,13 @@ abstract class SettingsItem(
                     descriptionId = R.string.renderer_asynchronous_shaders_description
                 )
             )
+            put(
+                SwitchSetting(
+                    BooleanSetting.RENDERER_UNIFIED_MEMORY,
+                    titleId = R.string.renderer_unified_memory,
+                    descriptionId = R.string.renderer_unified_memory_description
+                )
+            )
             put(
                 SingleChoiceSetting(
                     IntSetting.FAST_GPU_TIME,

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt

@@ -304,6 +304,7 @@ class SettingsFragmentPresenter(
             add(BooleanSetting.EMULATE_BGR565.key)
             add(BooleanSetting.RESCALE_HACK.key)
             add(BooleanSetting.RENDERER_ASYNCHRONOUS_SHADERS.key)
+            add(BooleanSetting.RENDERER_UNIFIED_MEMORY.key)
             add(IntSetting.ANDROID_PIPELINE_WORKERS.key)
             add(BooleanSetting.RENDERER_ASYNCHRONOUS_GPU_EMULATION.key)
             add(BooleanSetting.RENDERER_ASYNC_PRESENTATION.key)
@@ -313,6 +314,8 @@ class SettingsFragmentPresenter(
 
             add(IntSetting.RENDERER_DYNA_STATE.key)
             add(BooleanSetting.RENDERER_VERTEX_INPUT_DYNAMIC_STATE.key)
+            add(BooleanSetting.RENDERER_DYNAMIC_RENDERING.key)
+            add(BooleanSetting.RENDERER_WORKGROUP_MEMORY_EXPLICIT_LAYOUT.key)
             add(IntSetting.RENDERER_SAMPLE_SHADING.key)
 
             add(HeaderSetting(R.string.display))

src/android/app/src/main/java/org/yuzu/yuzu_emu/fragments/EmulationFragment.kt

@@ -1451,6 +1451,7 @@ class EmulationFragment : Fragment(), SurfaceHolder.Callback {
 
     override fun onResume() {
         super.onResume()
+        NativeLibrary.refreshThreadPolicies()
         val b = _binding ?: return
         updateStatsPosition(IntSetting.PERF_OVERLAY_POSITION.getInt())
         updateSocPosition(IntSetting.SOC_OVERLAY_POSITION.getInt())

src/android/app/src/main/jni/android_settings.h

@@ -147,13 +147,6 @@ namespace AndroidSettings {
                                                       &show_performance_overlay};
 
 
-        Settings::Setting<s32> pipeline_worker_count{linkage, 4, "pipeline_worker_count",
-                                 Settings::Category::Android,
-                                 Settings::Specialization::Default,
-                                 true,
-                                 true};
-
-
         Settings::Setting<bool> show_input_overlay{linkage, true, "show_input_overlay",
                                                    Settings::Category::Overlay};
         Settings::Setting<bool> overlay_snap_to_grid{linkage, false, "overlay_snap_to_grid",

src/android/app/src/main/jni/native.cpp

@@ -50,6 +50,7 @@ extern "C" {
 #include "common/scope_exit.h"
 #include "common/settings.h"
 #include "common/string_util.h"
+#include "common/thread.h"
 #include "frontend_common/play_time_manager.h"
 #include "core/constants.h"
 #include "core/core.h"
@@ -1182,6 +1183,10 @@ void Java_org_yuzu_yuzu_1emu_NativeLibrary_logSettings(JNIEnv* env, jobject jobj
     Settings::LogSettings();
 }
 
+void Java_org_yuzu_yuzu_1emu_NativeLibrary_refreshThreadPolicies(JNIEnv* env, jobject jobj) {
+    Common::RefreshThreadPolicies();
+}
+
 jboolean Java_org_yuzu_yuzu_1emu_NativeLibrary_getDebugKnobAt(JNIEnv* env, jobject jobj, jint index) {
     return static_cast<jboolean>(Settings::getDebugKnobAt(static_cast<u8>(index)));
 }

src/android/app/src/main/res/values-ar/strings.xml

@@ -1000,6 +1000,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">سريع</string>
+    <string name="renderer_accuracy_medium">متوازن</string>
     <string name="renderer_accuracy_high">دقيق</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-cs/strings.xml

@@ -707,6 +707,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">Rychlý</string>
+    <string name="renderer_accuracy_medium">Vyvážený</string>
     <string name="renderer_accuracy_high">Přesný</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-de/strings.xml

@@ -927,6 +927,7 @@ Wirklich fortfahren?</string>
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">Schnell</string>
+    <string name="renderer_accuracy_medium">Ausgeglichen</string>
     <string name="renderer_accuracy_high">Genau</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-es/strings.xml

@@ -993,6 +993,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">Rápido</string>
+    <string name="renderer_accuracy_medium">Equilibrado</string>
     <string name="renderer_accuracy_high">Preciso</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-fr/strings.xml

@@ -939,6 +939,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">Rapide</string>
+    <string name="renderer_accuracy_medium">Moyen</string>
     <string name="renderer_accuracy_high">Précis</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-pl/strings.xml

@@ -893,6 +893,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">Szybkie</string>
+    <string name="renderer_accuracy_medium">Zrównoważony</string>
     <string name="renderer_accuracy_high">Dokładny</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-pt-rBR/strings.xml

@@ -840,6 +840,7 @@
 
     <string name="renderer_none">Nenhum</string>
 
+    <string name="renderer_accuracy_medium">Média</string>
     <string name="renderer_accuracy_high">Alta</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-ru/strings.xml

@@ -983,6 +983,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">Быстрый</string>
+    <string name="renderer_accuracy_medium">Сбалансированный</string>
     <string name="renderer_accuracy_high">Точный</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-uk/strings.xml

@@ -986,6 +986,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">Швидко</string>
+    <string name="renderer_accuracy_medium">Збалансовано</string>
     <string name="renderer_accuracy_high">Точно</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-zh-rCN/strings.xml

@@ -990,6 +990,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">快速</string>
+    <string name="renderer_accuracy_medium">均衡</string>
     <string name="renderer_accuracy_high">精确</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values-zh-rTW/strings.xml

@@ -843,6 +843,7 @@
 
     <string name="renderer_none">無</string>
 
+    <string name="renderer_accuracy_medium">平衡</string>
     <string name="renderer_accuracy_high">準確</string>
 
     <!-- DMA Accuracy -->

src/android/app/src/main/res/values/arrays.xml

@@ -103,12 +103,14 @@
 
     <string-array name="rendererAccuracyNames">
         <item>@string/renderer_accuracy_low</item>
+        <item>@string/renderer_accuracy_medium</item>
         <item>@string/renderer_accuracy_high</item>
     </string-array>
 
     <integer-array name="rendererAccuracyValues">
         <item>0</item>
         <item>1</item>
+        <item>2</item>
     </integer-array>
 
     <!-- VRAM USAGE MODE CHOICES -->

src/android/app/src/main/res/values/strings.xml

@@ -479,7 +479,7 @@
     <string name="advanced">Advanced</string>
 
     <string name="renderer_accuracy">GPU Mode</string>
-    <string name="renderer_accuracy_description">Controls the GPU emulation mode. Most games render fine with Fast, but Accurate is still required for some. Particles tend to only render correctly with Accurate mode.</string>
+    <string name="renderer_accuracy_description">Controls the GPU emulation mode. Most games render fine with Fast or Balanced modes, but Accurate is still required for some. Particles tend to only render correctly with Accurate mode.</string>
     <string name="dma_accuracy">DMA Accuracy</string>
     <string name="dma_accuracy_description">Controls the DMA precision accuracy. Safe precision can fix issues in some games, but it can also impact performance in some cases. If unsure, leave this on Default.</string>
     <string name="gpu_fence_behavior">GPU Fence Behavior</string>
@@ -524,6 +524,8 @@
     <string name="rescale_hack_description">Enables a legacy handling for the rescale configuration pass for games by using a quick rescale path</string>
     <string name="renderer_asynchronous_shaders">Use asynchronous shaders</string>
     <string name="renderer_asynchronous_shaders_description">Compiles shaders asynchronously. This may reduce stutters but may also introduce glitches.</string>
+    <string name="renderer_unified_memory">Unified memory access (UMA)</string>
+    <string name="renderer_unified_memory_description">Allows GPU write buffer readbacks directly into guest memory, skipping the CPU staging copy.</string>
     <string name="gpu_unswizzle_settings">GPU Unswizzle Settings</string>
     <string name="gpu_unswizzle_settings_description">Configure GPU-based texture unswizzling parameters or disable it entirely. Adjust these settings to balance performance and texture loading quality.</string>
     <string name="gpu_unswizzle_enable">Enable GPU Unswizzle</string>
@@ -544,6 +546,10 @@
     <string name="disabled">Disabled</string>
     <string name="vertex_input_dynamic_state">Vertex Input Dynamic State</string>
     <string name="vertex_input_dynamic_state_description">Enabling this feature allows for more flexible vertex input handling, potentially reducing pipeline compilation time in vertex/buffer.</string>
+    <string name="dynamic_rendering">Dynamic Rendering</string>
+    <string name="dynamic_rendering_description">Render without render pass and framebuffer objects. Results vary by driver: some gain performance, others lose it.</string>
+    <string name="workgroup_memory_explicit_layout">Workgroup Memory Explicit Layout</string>
+    <string name="workgroup_memory_explicit_layout_description">Let shaders declare explicit layouts for workgroup memory. Disabled by default: some Qualcomm drivers are unstable with it.</string>
     <string name="sample_shading_fraction">Sample Shading</string>
     <string name="sample_shading_fraction_description">Allows the fragment shader to execute per sample in a multi-sampled fragment instead once per fragment. Improves graphics quality at the cost of some performance.</string>
 
@@ -1039,6 +1045,7 @@
 
     <!-- Renderer Accuracy -->
     <string name="renderer_accuracy_low">Fast</string>
+    <string name="renderer_accuracy_medium">Balanced</string>
     <string name="renderer_accuracy_high">Accurate</string>
 
     <!-- DMA Accuracy -->

src/common/fs/path_util.cpp

@@ -476,29 +476,6 @@ std::string SanitizePath(std::string_view path_, DirectorySeparator directory_se
     path.erase(std::unique(start, path.end(),
                            [type2](char c1, char c2) { return c1 == type2 && c2 == type2; }),
                path.end());
-
-    const bool absolute = !path.empty() && path[0] == type2;
-    std::vector<std::string_view> parts;
-
-    for (const auto part : SplitPathComponents(path))
-    {
-        if (part.empty() || part == ".")
-            continue;
-        if (part == ".." && !parts.empty() && parts.back() != "..")
-            parts.pop_back();
-        else if (part != "..") parts.push_back(part);
-    }
-
-    std::string resolved = absolute ? std::string(1, type2) : std::string{};
-    for (std::size_t i = 0; i < parts.size(); ++i)
-    {
-        if (i != 0)
-            resolved += type2;
-        resolved.append(parts[i].data(), parts[i].size());
-    }
-
-    path = std::move(resolved);
-
     return std::string(RemoveTrailingSlash(path));
 }
 

src/common/host_memory.cpp

@@ -51,14 +51,45 @@
 
 #endif // ^^^ POSIX ^^^
 
+#include <atomic>
 #include <mutex>
 #include <random>
+#include <vector>
 
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/free_region_manager.h"
 #include "common/host_memory.h"
 #include "common/logging.h"
+#include "common/memory_detect.h"
+#include "common/settings.h"
+
+#ifdef __ANDROID__
+#include <dlfcn.h>
+#include <android/hardware_buffer.h>
+
+namespace {
+
+struct NativeHandle {
+    int version;
+    int numFds;
+    int numInts;
+    int data[1];
+};
+
+using PFN_AHardwareBuffer_getNativeHandle = const NativeHandle* (*)(const AHardwareBuffer*);
+
+PFN_AHardwareBuffer_getNativeHandle ResolveGetNativeHandle() {
+    void* const lib = dlopen("libnativewindow.so", RTLD_NOW);
+    if (lib == nullptr) {
+        return nullptr;
+    }
+    return reinterpret_cast<PFN_AHardwareBuffer_getNativeHandle>(
+        dlsym(lib, "AHardwareBuffer_getNativeHandle"));
+}
+
+} // namespace
+#endif
 
 #if defined(__ANDROID__) && __ANDROID_API__ < 30
 #include <sys/syscall.h>
@@ -75,6 +106,12 @@ namespace Common {
 [[maybe_unused]] constexpr size_t PageAlignment = 0x1000;
 [[maybe_unused]] constexpr size_t HugePageSize = 0x200000;
 
+static std::atomic<u64> committed_backing_size{};
+
+u64 GetCommittedBackingSize() noexcept {
+    return committed_backing_size.load(std::memory_order_relaxed);
+}
+
 #ifdef _WIN32
 
 // Manually imported for MinGW compatibility
@@ -123,7 +160,7 @@ static void GetFuncAddress(Common::DynamicLibrary& dll, const char* name, T& pfn
 
 class HostMemory::Impl {
 public:
-    explicit Impl(size_t backing_size_, size_t virtual_size_)
+    explicit Impl(size_t backing_size_, size_t virtual_size_, size_t)
         : backing_size{backing_size_}
         , virtual_size{virtual_size_}
         , process{GetCurrentProcess()}
@@ -229,6 +266,10 @@ public:
         UNREACHABLE();
     }
 
+    bool IsBackingShared() const noexcept {
+        return true;
+    }
+
     const size_t backing_size; ///< Size of the backing memory in bytes
     const size_t virtual_size; ///< Size of the virtual address placeholder in bytes
 
@@ -501,9 +542,10 @@ static int shm_open_anon(int flags, mode_t mode) {
 
 class HostMemory::Impl {
 public:
-    explicit Impl(size_t backing_size_, size_t virtual_size_)
+    explicit Impl(size_t backing_size_, size_t virtual_size_, size_t preferred_offset_)
         : backing_size{backing_size_}
         , virtual_size{virtual_size_}
+        , preferred_offset{preferred_offset_}
     {}
 
     bool Init() {
@@ -543,10 +585,15 @@ public:
             LOG_WARNING(Common_Memory, "Using private mappings instead of shared ones");
             backing_base = static_cast<u8*>(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0));
             if (fd > 0) {
-                fd = -1;
                 close(fd);
             }
+            fd = -1;
         } else {
+#ifdef __ANDROID__
+            if (InitAhbBacking()) {
+                return InitVirtual();
+            }
+#endif
             backing_base = static_cast<u8*>(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0));
         }
         if (backing_base == MAP_FAILED) {
@@ -554,7 +601,10 @@ public:
             return false;
         }
 
-        // Virtual memory initialization
+        return InitVirtual();
+    }
+
+    bool InitVirtual() {
         virtual_base = virtual_map_base = static_cast<u8*>(ChooseVirtualBase(virtual_size));
         if (virtual_base == MAP_FAILED) {
             LOG_CRITICAL(HW_Memory, "mmap failed: {}", strerror(errno));
@@ -567,6 +617,248 @@ public:
         return true;
     }
 
+#ifdef __ANDROID__
+    static AHardwareBuffer_Desc MakeBlobDesc(size_t len) {
+        return AHardwareBuffer_Desc{
+            .width = static_cast<u32>(len),
+            .height = 1,
+            .layers = 1,
+            .format = AHARDWAREBUFFER_FORMAT_BLOB,
+            .usage = AHARDWAREBUFFER_USAGE_CPU_READ_OFTEN |
+                     AHARDWAREBUFFER_USAGE_CPU_WRITE_OFTEN |
+                     AHARDWAREBUFFER_USAGE_GPU_DATA_BUFFER,
+            .stride = 0,
+            .rfu0 = 0,
+            .rfu1 = 0,
+        };
+    }
+
+    static bool ProbeAhbBacking(PFN_AHardwareBuffer_getNativeHandle get_native_handle) {
+        const AHardwareBuffer_Desc desc = MakeBlobDesc(PageAlignment * 2);
+        AHardwareBuffer* buffer{};
+        if (AHardwareBuffer_allocate(&desc, &buffer) != 0 || buffer == nullptr) {
+            LOG_WARNING(HW_Memory, "Hardware buffer probe allocation failed");
+            return false;
+        }
+        const NativeHandle* const handle = get_native_handle(buffer);
+        if (handle == nullptr || handle->numFds < 1) {
+            LOG_WARNING(HW_Memory, "Hardware buffer has no mappable file descriptor");
+            AHardwareBuffer_release(buffer);
+            return false;
+        }
+        const int probe_fd = handle->data[0];
+        bool ok = true;
+        const auto try_map = [&](int prot, off_t offset, const char* what) {
+            if (!ok) {
+                return;
+            }
+            void* const ptr = mmap(nullptr, PageAlignment, prot, MAP_SHARED, probe_fd, offset);
+            if (ptr == MAP_FAILED) {
+                LOG_WARNING(HW_Memory, "Hardware buffer backing rejects {}: {}", what,
+                            strerror(errno));
+                ok = false;
+                return;
+            }
+            munmap(ptr, PageAlignment);
+        };
+        try_map(PROT_READ | PROT_WRITE, 0, "shared mappings");
+        try_map(PROT_READ | PROT_WRITE, static_cast<off_t>(PageAlignment), "mappings at an offset");
+#ifdef ARCHITECTURE_arm64
+        try_map(PROT_READ | PROT_EXEC, 0, "executable mappings");
+#endif
+        AHardwareBuffer_release(buffer);
+        return ok;
+    }
+
+    size_t ComputeAhbBudget(size_t window_size) const {
+        const u64 total_physical = Common::GetMemInfo().TotalPhysicalMemory;
+        if (total_physical == 0) {
+            LOG_WARNING(HW_Memory, "Host memory size is unknown, not committing hardware buffers");
+            return 0;
+        }
+        constexpr u64 MinimumTotalPhysical = 7ULL << 30;
+        if (total_physical < MinimumTotalPhysical) {
+            LOG_INFO(HW_Memory,
+                     "Skipping hardware buffer backing, {} MiB of RAM is below the {} MiB minimum",
+                     total_physical >> 20, MinimumTotalPhysical >> 20);
+            return 0;
+        }
+        const u64 max_map_count = Common::GetMaxMapCount();
+        constexpr u64 ReservedMaps = 24576;
+        if (max_map_count == 0 || max_map_count <= ReservedMaps) {
+            LOG_WARNING(HW_Memory,
+                        "Skipping hardware buffer backing, vm.max_map_count is unknown or too low");
+            return 0;
+        }
+        u64 budget = total_physical / 6;
+        budget = (std::min)(budget, (max_map_count - ReservedMaps) * PageAlignment);
+        const u64 available = Common::GetAvailablePhysicalMemory();
+        if (available != 0) {
+            constexpr u64 Headroom = 2ULL << 30;
+            budget = (std::min)(budget, available > Headroom ? available - Headroom : 0);
+        }
+        budget = (std::min)(budget, static_cast<u64>(backing_size));
+        budget = Common::AlignDown(budget, window_size);
+        constexpr u64 MinimumBudget = 256ULL << 20;
+        if (budget < MinimumBudget) {
+            LOG_INFO(HW_Memory,
+                     "Skipping hardware buffer backing, only {} MiB could be committed on a {} MiB "
+                     "system with {} MiB available and vm.max_map_count {}",
+                     budget >> 20, total_physical >> 20, available >> 20, max_map_count);
+            return 0;
+        }
+        return static_cast<size_t>(budget);
+    }
+
+    bool InitAhbBacking() {
+        if (!Settings::values.use_unified_memory.GetValue()) {
+            return false;
+        }
+        static const PFN_AHardwareBuffer_getNativeHandle get_native_handle =
+            ResolveGetNativeHandle();
+        if (get_native_handle == nullptr) {
+            LOG_WARNING(HW_Memory, "AHardwareBuffer_getNativeHandle is not available");
+            return false;
+        }
+        constexpr size_t window_size = 64ULL << 20;
+        const AHardwareBuffer_Desc window_desc = MakeBlobDesc(window_size);
+        if (AHardwareBuffer_isSupported(&window_desc) == 0) {
+            LOG_WARNING(HW_Memory, "Allocator rejects {} MiB hardware buffer windows",
+                        window_size >> 20);
+            return false;
+        }
+        const size_t budget = ComputeAhbBudget(window_size);
+        if (budget == 0) {
+            return false;
+        }
+        if (!ProbeAhbBacking(get_native_handle)) {
+            return false;
+        }
+        const size_t aligned_backing = Common::AlignDown(backing_size, window_size);
+        const size_t region_size = (std::min)(budget, aligned_backing);
+        const size_t region_base = Common::AlignDown(
+            (std::min)(preferred_offset, aligned_backing - region_size), window_size);
+        const size_t num_windows = region_size / window_size;
+
+        std::vector<AHardwareBuffer*> buffers;
+        std::vector<int> buffer_fds;
+        const auto cleanup = [&] {
+            for (AHardwareBuffer* buffer : buffers) {
+                AHardwareBuffer_release(buffer);
+            }
+            buffers.clear();
+            buffer_fds.clear();
+        };
+        for (size_t i = 0; i < num_windows; ++i) {
+            const AHardwareBuffer_Desc desc = MakeBlobDesc(window_size);
+            AHardwareBuffer* buffer{};
+            if (AHardwareBuffer_allocate(&desc, &buffer) != 0 || buffer == nullptr) {
+                LOG_WARNING(HW_Memory, "Hardware buffer allocation failed for window {} of {}", i,
+                            num_windows);
+                cleanup();
+                return false;
+            }
+            buffers.push_back(buffer);
+            const NativeHandle* const handle = get_native_handle(buffer);
+            if (handle == nullptr || handle->numFds < 1) {
+                LOG_WARNING(HW_Memory, "Hardware buffer has no mappable file descriptor");
+                cleanup();
+                return false;
+            }
+            const int buffer_fd = handle->data[0];
+            const off_t buffer_len = lseek(buffer_fd, 0, SEEK_END);
+            if (buffer_len < static_cast<off_t>(window_size)) {
+                LOG_WARNING(HW_Memory, "Hardware buffer descriptor smaller than requested");
+                cleanup();
+                return false;
+            }
+            buffer_fds.push_back(buffer_fd);
+        }
+        u8* const base = static_cast<u8*>(mmap(nullptr, backing_size, PROT_NONE,
+                                               MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0));
+        if (base == MAP_FAILED) {
+            LOG_WARNING(HW_Memory, "Failed to reserve backing address space: {}", strerror(errno));
+            cleanup();
+            return false;
+        }
+        const auto map_over_reservation = [&](size_t offset, size_t len, int map_fd,
+                                              off_t map_offset) {
+            if (len == 0) {
+                return true;
+            }
+            if (mmap(base + offset, len, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, map_fd,
+                     map_offset) == MAP_FAILED) {
+                LOG_WARNING(HW_Memory, "Backing mmap failed: {}", strerror(errno));
+                munmap(base, backing_size);
+                cleanup();
+                return false;
+            }
+            return true;
+        };
+        if (!map_over_reservation(0, region_base, fd, 0)) {
+            return false;
+        }
+        for (size_t i = 0; i < num_windows; ++i) {
+            if (!map_over_reservation(region_base + i * window_size, window_size, buffer_fds[i],
+                                      0)) {
+                return false;
+            }
+        }
+        const size_t tail_offset = region_base + region_size;
+        if (!map_over_reservation(tail_offset, backing_size - tail_offset, fd,
+                                  static_cast<off_t>(tail_offset))) {
+            return false;
+        }
+        backing_base = base;
+        ahb_windows = std::move(buffers);
+        ahb_fds = std::move(buffer_fds);
+        ahb_window_size = window_size;
+        ahb_base = region_base;
+        ahb_bytes = region_size;
+        committed_backing_size.store(region_size, std::memory_order_relaxed);
+        LOG_INFO(HW_Memory,
+                 "Guest memory {:#x}-{:#x} backed by {} hardware buffer windows, {} MiB committed",
+                 region_base, region_base + region_size, ahb_windows.size(), region_size >> 20);
+        return true;
+    }
+
+    void MapBackingRange(size_t virtual_offset, size_t host_offset, size_t length, int prot_flags) {
+        while (length > 0) {
+            int map_fd = fd;
+            off_t map_offset = static_cast<off_t>(host_offset);
+            size_t chunk = length;
+            if (host_offset < ahb_base) {
+                chunk = (std::min)(chunk, ahb_base - host_offset);
+            } else if (host_offset < ahb_base + ahb_bytes) {
+                const size_t relative = host_offset - ahb_base;
+                const size_t window = relative / ahb_window_size;
+                const size_t local = relative % ahb_window_size;
+                map_fd = ahb_fds[window];
+                map_offset = static_cast<off_t>(local);
+                chunk = (std::min)(chunk, ahb_window_size - local);
+            }
+            void* const ret = mmap(virtual_base + virtual_offset, chunk, prot_flags,
+                                   MAP_SHARED | MAP_FIXED, map_fd, map_offset);
+            ASSERT_MSG(ret != MAP_FAILED, "mmap: {}", strerror(errno));
+            virtual_offset += chunk;
+            host_offset += chunk;
+            length -= chunk;
+        }
+    }
+
+    std::span<AHardwareBuffer* const> AhbWindows() const noexcept {
+        return ahb_windows;
+    }
+
+    size_t AhbWindowSize() const noexcept {
+        return ahb_bytes != 0 ? ahb_window_size : 0;
+    }
+
+    size_t AhbBase() const noexcept {
+        return ahb_base;
+    }
+#endif
+
     ~Impl() {
         Release();
     }
@@ -587,6 +879,12 @@ public:
 #ifdef ARCHITECTURE_arm64
         if (True(perms & MemoryPermission::Execute))
             prot_flags |= PROT_EXEC;
+#endif
+#ifdef __ANDROID__
+        if (ahb_bytes != 0) {
+            MapBackingRange(virtual_offset, host_offset, length, prot_flags);
+            return;
+        }
 #endif
         int flags = (fd >= 0 ? MAP_SHARED : MAP_PRIVATE) | MAP_FIXED;
         void* ret = mmap(virtual_base + virtual_offset, length, prot_flags, flags, fd, host_offset);
@@ -632,8 +930,18 @@ public:
         virtual_base = nullptr;
     }
 
+    bool IsBackingShared() const noexcept {
+#ifdef __ANDROID__
+        if (ahb_bytes != 0) {
+            return true;
+        }
+#endif
+        return fd >= 0;
+    }
+
     const size_t backing_size; ///< Size of the backing memory in bytes
     const size_t virtual_size; ///< Size of the virtual address placeholder in bytes
+    const size_t preferred_offset;
 
     u8* backing_base{reinterpret_cast<u8*>(MAP_FAILED)};
     u8* virtual_base{reinterpret_cast<u8*>(MAP_FAILED)};
@@ -656,6 +964,18 @@ private:
             int ret = close(fd);
             ASSERT_MSG(ret == 0, "close failed: {}", strerror(errno));
         }
+
+#ifdef __ANDROID__
+        for (AHardwareBuffer* buffer : ahb_windows) {
+            AHardwareBuffer_release(buffer);
+        }
+        ahb_windows.clear();
+        ahb_fds.clear();
+        if (ahb_bytes != 0) {
+            committed_backing_size.store(0, std::memory_order_relaxed);
+            ahb_bytes = 0;
+        }
+#endif
     }
 
     void AdjustMap(size_t* virtual_offset, size_t* length) {
@@ -681,11 +1001,19 @@ private:
 
     int fd{-1}; // memfd file descriptor, -1 is the error value of memfd_create
     FreeRegionManager free_manager{};
+
+#ifdef __ANDROID__
+    std::vector<AHardwareBuffer*> ahb_windows;
+    std::vector<int> ahb_fds;
+    size_t ahb_window_size{};
+    size_t ahb_base{};
+    size_t ahb_bytes{};
+#endif
 };
 
 #endif // ^^^ POSIX ^^^
 
-HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_)
+HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_, size_t preferred_offset_)
     : backing_size(backing_size_)
     , virtual_size(virtual_size_)
 {
@@ -697,7 +1025,7 @@ HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_)
 #else
     // Try to allocate a fastmem arena.
     // The implementation will fail with std::bad_alloc on errors.
-    impl = std::make_unique<HostMemory::Impl>(AlignUp(backing_size, PageAlignment), AlignUp(virtual_size, PageAlignment) + HugePageSize);
+    impl = std::make_unique<HostMemory::Impl>(AlignUp(backing_size, PageAlignment), AlignUp(virtual_size, PageAlignment) + HugePageSize, preferred_offset_);
     if (impl->Init()) {
         backing_base = impl->backing_base;
         virtual_base = impl->virtual_base;
@@ -767,6 +1095,38 @@ void HostMemory::ClearBackingRegion(size_t physical_offset, size_t length, u32 f
     std::memset(backing_base + physical_offset, fill_value, length);
 }
 
+std::span<AHardwareBuffer* const> HostMemory::BackingHardwareBuffers() const noexcept {
+#ifdef __ANDROID__
+    return impl ? impl->AhbWindows() : std::span<AHardwareBuffer* const>{};
+#else
+    return {};
+#endif
+}
+
+size_t HostMemory::BackingHardwareBufferWindowSize() const noexcept {
+#ifdef __ANDROID__
+    return impl ? impl->AhbWindowSize() : 0;
+#else
+    return 0;
+#endif
+}
+
+bool HostMemory::IsBackingShared() const noexcept {
+#if defined(__OPENORBIS__) || defined(__managarm__)
+    return false;
+#else
+    return impl && impl->IsBackingShared();
+#endif
+}
+
+size_t HostMemory::BackingHardwareBufferBase() const noexcept {
+#ifdef __ANDROID__
+    return impl ? impl->AhbBase() : 0;
+#else
+    return 0;
+#endif
+}
+
 void HostMemory::EnableDirectMappedAddress() {
 #if !(defined(__OPENORBIS__) || defined(__managarm__))
     if (impl) {

src/common/host_memory.h

@@ -8,12 +8,17 @@
 
 #include <memory>
 #include <optional>
+#include <span>
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/virtual_buffer.h"
 
+struct AHardwareBuffer;
+
 namespace Common {
 
+[[nodiscard]] u64 GetCommittedBackingSize() noexcept;
+
 enum class MemoryPermission : u32 {
     Read = 1 << 0,
     Write = 1 << 1,
@@ -28,7 +33,7 @@ DECLARE_ENUM_FLAG_OPERATORS(MemoryPermission)
  */
 class HostMemory {
 public:
-    explicit HostMemory(size_t backing_size_, size_t virtual_size_);
+    explicit HostMemory(size_t backing_size_, size_t virtual_size_, size_t preferred_offset_ = 0);
     ~HostMemory();
 
     /**
@@ -62,6 +67,18 @@ public:
         return backing_base;
     }
 
+    [[nodiscard]] size_t BackingSize() const noexcept {
+        return backing_size;
+    }
+
+    [[nodiscard]] std::span<AHardwareBuffer* const> BackingHardwareBuffers() const noexcept;
+
+    [[nodiscard]] size_t BackingHardwareBufferWindowSize() const noexcept;
+
+    [[nodiscard]] size_t BackingHardwareBufferBase() const noexcept;
+
+    [[nodiscard]] bool IsBackingShared() const noexcept;
+
     [[nodiscard]] u8* VirtualBasePointer() noexcept {
         return virtual_base;
     }

src/common/memory_detect.cpp

@@ -17,6 +17,10 @@
 #endif
 #endif
 
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+
 #include "common/memory_detect.h"
 
 namespace Common {
@@ -69,4 +73,55 @@ const MemoryInfo& GetMemInfo() {
     return mem_info;
 }
 
+u64 GetAvailablePhysicalMemory() {
+#ifdef _WIN32
+    MEMORYSTATUSEX memorystatus;
+    memorystatus.dwLength = sizeof(memorystatus);
+    if (GlobalMemoryStatusEx(&memorystatus)) {
+        return memorystatus.ullAvailPhys;
+    }
+    return 0;
+#elif defined(__linux__)
+    if (std::FILE* const file = std::fopen("/proc/meminfo", "re")) {
+        char line[256];
+        u64 available = 0;
+        while (std::fgets(line, sizeof(line), file) != nullptr) {
+            if (std::strncmp(line, "MemAvailable:", 13) == 0) {
+                available = std::strtoull(line + 13, nullptr, 10) * 1024ULL;
+                break;
+            }
+        }
+        std::fclose(file);
+        if (available != 0) {
+            return available;
+        }
+    }
+    struct sysinfo info;
+    if (sysinfo(&info) == 0) {
+        const u64 unit = info.mem_unit != 0 ? info.mem_unit : 1ULL;
+        return (static_cast<u64>(info.freeram) + static_cast<u64>(info.bufferram)) * unit;
+    }
+    return 0;
+#else
+    return 0;
+#endif
+}
+
+u64 GetMaxMapCount() {
+#ifdef __linux__
+    if (std::FILE* const file = std::fopen("/proc/sys/vm/max_map_count", "re")) {
+        char line[32];
+        u64 count = 0;
+        if (std::fgets(line, sizeof(line), file) != nullptr) {
+            count = std::strtoull(line, nullptr, 10);
+        }
+        std::fclose(file);
+        return count;
+    }
+    return 0;
+#else
+    return 0;
+#endif
+}
+
 } // namespace Common

src/common/memory_detect.h

@@ -18,4 +18,8 @@ struct MemoryInfo {
  */
 [[nodiscard]] const MemoryInfo& GetMemInfo();
 
+[[nodiscard]] u64 GetAvailablePhysicalMemory();
+
+[[nodiscard]] u64 GetMaxMapCount();
+
 } // namespace Common

src/common/settings.h

@@ -587,6 +587,9 @@ struct Values {
     SwitchableSetting<bool> use_asynchronous_shaders{linkage, false, "use_asynchronous_shaders",
                                                      Category::RendererHacks};
 
+    SwitchableSetting<bool> use_unified_memory{linkage, false, "use_unified_memory",
+                                               Category::RendererHacks};
+
     SwitchableSetting<GpuUnswizzleSize> gpu_unswizzle_texture_size{linkage,
                                                   GpuUnswizzleSize::Large,
                                                   "gpu_unswizzle_texture_size",
@@ -635,6 +638,16 @@ struct Values {
 #endif
                                                        "vertex_input_dynamic_state", Category::RendererExtensions};
 
+    SwitchableSetting<bool> dynamic_rendering{linkage, true, "dynamic_rendering",
+                                              Category::RendererExtensions};
+
+    SwitchableSetting<bool> workgroup_memory_explicit_layout{
+        linkage, false, "workgroup_memory_explicit_layout", Category::RendererExtensions};
+
+    SwitchableSetting<s32, true> pipeline_worker_count{
+        linkage, 2, 2, 8, "pipeline_worker_count", Category::RendererAdvanced,
+        Specialization::Scalar};
+
     Setting<bool> renderer_debug{linkage, false, "debug", Category::RendererDebug};
     Setting<bool> renderer_shader_feedback{linkage, false, "shader_feedback",
                                            Category::RendererDebug};

src/common/thread.cpp

@@ -1,5 +1,6 @@
 // SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: 2013 Dolphin Emulator Project
 // SPDX-FileCopyrightText: 2014 Citra Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
@@ -39,6 +40,258 @@
 #include <unistd.h>
 #endif
 
+#ifdef __ANDROID__
+#include <sys/resource.h>
+#include <algorithm>
+#include <cstdlib>
+#include <cstring>
+#include <fstream>
+#include <mutex>
+#include <utility>
+#include <vector>
+
+namespace {
+constexpr int ANDROID_THREAD_PRIORITY_AUDIO = -16;
+constexpr int ANDROID_THREAD_PRIORITY_URGENT_DISPLAY = -8;
+constexpr int ANDROID_THREAD_PRIORITY_DISPLAY = -4;
+constexpr int ANDROID_THREAD_PRIORITY_DEFAULT = 0;
+constexpr int ANDROID_THREAD_PRIORITY_BACKGROUND = 10;
+
+constexpr size_t ANDROID_MINIMUM_PERFORMANCE_CORES = 4;
+
+enum class CoreGroup {
+    Unrestricted,
+    Performance,
+    Efficiency,
+};
+
+struct CoreTopology {
+    cpu_set_t allowed;
+    cpu_set_t performance;
+    cpu_set_t efficiency;
+    bool separated;
+    bool initialized;
+};
+
+struct ThreadPolicy {
+    pid_t tid;
+    CoreGroup group;
+    int nice_value;
+    bool has_nice;
+};
+
+std::mutex g_topology_mutex;
+CoreTopology g_topology{};
+
+std::mutex g_policy_mutex;
+
+std::vector<ThreadPolicy>& Policies() {
+    static auto* const policies = new std::vector<ThreadPolicy>();
+    return *policies;
+}
+
+struct PolicyRegistration {
+    ~PolicyRegistration() {
+        const pid_t tid = gettid();
+        std::scoped_lock lock{g_policy_mutex};
+        std::erase_if(Policies(), [tid](const ThreadPolicy& policy) { return policy.tid == tid; });
+    }
+};
+
+thread_local PolicyRegistration t_policy_registration;
+
+int PossibleCpuCount() {
+    std::ifstream file("/sys/devices/system/cpu/possible");
+    std::string list;
+    if (file && std::getline(file, list) && !list.empty()) {
+        int highest = -1;
+        const char* cursor = list.c_str();
+        while (*cursor != '\0') {
+            char* end = nullptr;
+            const long value = std::strtol(cursor, &end, 10);
+            if (end == cursor) {
+                break;
+            }
+            highest = (std::max)(highest, static_cast<int>(value));
+            cursor = end;
+            while (*cursor == '-' || *cursor == ',') {
+                ++cursor;
+            }
+        }
+        if (highest >= 0) {
+            return (std::min)(highest + 1, CPU_SETSIZE);
+        }
+    }
+    const long configured = sysconf(_SC_NPROCESSORS_CONF);
+    if (configured > 0) {
+        return static_cast<int>((std::min<long>)(configured, CPU_SETSIZE));
+    }
+    return static_cast<int>((std::min<unsigned>)(std::thread::hardware_concurrency(), CPU_SETSIZE));
+}
+
+long ReadCpuScalar(int cpu, const char* node) {
+    long value = 0;
+    std::ifstream file("/sys/devices/system/cpu/cpu" + std::to_string(cpu) + "/" + node);
+    if (!file || !(file >> value) || value <= 0) {
+        return 0;
+    }
+    return value;
+}
+
+std::vector<std::pair<long, int>> CollectCoreWeights(const cpu_set_t& allowed, int total,
+                                                     const char* node, bool require_all) {
+    std::vector<std::pair<long, int>> cores;
+    for (int cpu = 0; cpu < total; ++cpu) {
+        if (!CPU_ISSET(cpu, &allowed)) {
+            continue;
+        }
+        const long weight = ReadCpuScalar(cpu, node);
+        if (weight <= 0) {
+            if (require_all) {
+                return {};
+            }
+            LOG_WARNING(Common, "Could not read {} for CPU {}, treating it as an efficiency core",
+                        node, cpu);
+            continue;
+        }
+        cores.emplace_back(weight, cpu);
+    }
+    return cores;
+}
+
+void ComputeTopologyLocked() {
+    g_topology.initialized = true;
+    g_topology.separated = false;
+    CPU_ZERO(&g_topology.allowed);
+    CPU_ZERO(&g_topology.performance);
+    CPU_ZERO(&g_topology.efficiency);
+
+    if (sched_getaffinity(getpid(), sizeof(g_topology.allowed), &g_topology.allowed) != 0) {
+        LOG_WARNING(Common, "Could not query process CPU affinity: {}",
+                    ::Common::GetLastErrorMsg());
+        return;
+    }
+
+    const int total = PossibleCpuCount();
+    auto cores = CollectCoreWeights(g_topology.allowed, total, "cpu_capacity", true);
+    if (cores.empty()) {
+        cores = CollectCoreWeights(g_topology.allowed, total, "cpufreq/cpuinfo_max_freq", false);
+    }
+    if (cores.empty()) {
+        LOG_WARNING(Common, "Could not determine CPU topology, thread placement is disabled");
+        return;
+    }
+
+    std::sort(cores.begin(), cores.end(),
+              [](const auto& lhs, const auto& rhs) { return lhs.first > rhs.first; });
+
+    const size_t allowed_count = static_cast<size_t>(CPU_COUNT(&g_topology.allowed));
+    const size_t maximum =
+        allowed_count > 2 * ANDROID_MINIMUM_PERFORMANCE_CORES
+            ? allowed_count - ANDROID_MINIMUM_PERFORMANCE_CORES
+            : ANDROID_MINIMUM_PERFORMANCE_CORES;
+
+    size_t taken = 0;
+    long cluster_weight = cores.front().first;
+    for (const auto& [weight, cpu] : cores) {
+        if (weight != cluster_weight) {
+            if (taken >= ANDROID_MINIMUM_PERFORMANCE_CORES) {
+                break;
+            }
+            cluster_weight = weight;
+        }
+        if (taken >= maximum) {
+            break;
+        }
+        CPU_SET(cpu, &g_topology.performance);
+        ++taken;
+    }
+    if (taken == 0) {
+        return;
+    }
+
+    for (int cpu = 0; cpu < total; ++cpu) {
+        if (CPU_ISSET(cpu, &g_topology.allowed) && !CPU_ISSET(cpu, &g_topology.performance)) {
+            CPU_SET(cpu, &g_topology.efficiency);
+        }
+    }
+
+    g_topology.separated = CPU_COUNT(&g_topology.efficiency) > 0;
+    LOG_INFO(Common, "CPU topology: {} performance cores, {} efficiency cores, separation {}",
+             CPU_COUNT(&g_topology.performance), CPU_COUNT(&g_topology.efficiency),
+             g_topology.separated ? "enabled" : "unavailable");
+}
+
+void EnsureTopologyLocked() {
+    if (!g_topology.initialized) {
+        ComputeTopologyLocked();
+    }
+}
+
+void RefreshTopologyLocked() {
+    if (!g_topology.initialized) {
+        ComputeTopologyLocked();
+        return;
+    }
+    cpu_set_t current;
+    CPU_ZERO(&current);
+    if (sched_getaffinity(getpid(), sizeof(current), &current) != 0) {
+        return;
+    }
+    if (std::memcmp(&current, &g_topology.allowed, sizeof(current)) != 0) {
+        ComputeTopologyLocked();
+    }
+}
+
+bool ApplyCoreGroupLocked(pid_t tid, CoreGroup group) {
+    if (!g_topology.separated || group == CoreGroup::Unrestricted) {
+        return false;
+    }
+    const cpu_set_t& mask =
+        group == CoreGroup::Performance ? g_topology.performance : g_topology.efficiency;
+    if (CPU_COUNT(&mask) == 0) {
+        return false;
+    }
+    if (sched_setaffinity(tid, sizeof(mask), &mask) != 0) {
+        LOG_WARNING(Common, "Could not restrict thread {} to its core group: {}", tid,
+                    ::Common::GetLastErrorMsg());
+        return false;
+    }
+    return true;
+}
+
+ThreadPolicy& AcquirePolicyLocked(pid_t tid) {
+    auto& policies = Policies();
+    for (auto& policy : policies) {
+        if (policy.tid == tid) {
+            return policy;
+        }
+    }
+    return policies.emplace_back(ThreadPolicy{tid, CoreGroup::Unrestricted, 0, false});
+}
+
+void SetCurrentThreadCoreGroup(CoreGroup group) {
+    const pid_t tid = gettid();
+    {
+        std::scoped_lock lock{g_topology_mutex};
+        EnsureTopologyLocked();
+        ApplyCoreGroupLocked(tid, group);
+    }
+    (void)&t_policy_registration;
+    std::scoped_lock lock{g_policy_mutex};
+    AcquirePolicyLocked(tid).group = group;
+}
+
+void RememberCurrentThreadNice(pid_t tid, int nice_value) {
+    (void)&t_policy_registration;
+    std::scoped_lock lock{g_policy_mutex};
+    ThreadPolicy& policy = AcquirePolicyLocked(tid);
+    policy.nice_value = nice_value;
+    policy.has_nice = true;
+}
+} // Anonymous namespace
+#endif
+
 #include "common/cpu_features.h"
 #ifdef ARCHITECTURE_x86_64
 #ifdef _MSC_VER
@@ -48,7 +301,6 @@
 #endif
 #include "common/x64/rdtsc.h"
 #endif
-#include "core/core_timing.h"
 
 namespace Common {
 
@@ -78,6 +330,24 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
         }
     }();
     set_thread_priority(find_thread(NULL), priority);
+#elif defined(__ANDROID__)
+    const int nice_value = [&]() {
+        switch (new_priority) {
+        case ThreadPriority::Low: return ANDROID_THREAD_PRIORITY_BACKGROUND;
+        case ThreadPriority::Normal: return ANDROID_THREAD_PRIORITY_DEFAULT;
+        case ThreadPriority::High: return ANDROID_THREAD_PRIORITY_DISPLAY;
+        case ThreadPriority::VeryHigh: return ANDROID_THREAD_PRIORITY_URGENT_DISPLAY;
+        case ThreadPriority::Critical: return ANDROID_THREAD_PRIORITY_AUDIO;
+        default: return ANDROID_THREAD_PRIORITY_DEFAULT;
+        }
+    }();
+    const pid_t tid = gettid();
+    if (setpriority(PRIO_PROCESS, static_cast<id_t>(tid), nice_value) != 0) {
+        LOG_WARNING(Common, "Could not set thread nice value to {}: {}", nice_value,
+                    GetLastErrorMsg());
+        return;
+    }
+    RememberCurrentThreadNice(tid, nice_value);
 #else
     pthread_t this_thread = pthread_self();
     const auto scheduling_type = SCHED_OTHER;
@@ -132,29 +402,31 @@ void SetCurrentThreadName(const char* name) {
 #endif
 }
 
-void PinCurrentThreadToPerformanceCore(size_t core_id) {
-    ASSERT(core_id < 4);
-    // If we set a flag for a CPU that doesn't exist, the thread may not be allowed to
-    // run in ANY processor!
-    auto const total_cores = std::thread::hardware_concurrency();
-    if (core_id < total_cores) {
+void SetCurrentThreadToPerformanceCores() {
 #if defined(__ANDROID__)
-        cpu_set_t set;
-        CPU_ZERO(&set);
-        CPU_SET(core_id, &set);
-        sched_setaffinity(pthread_self(), sizeof(set), &set);
-#elif defined(__linux__) || defined(__FreeBSD__)
-        cpu_set_t set;
-        CPU_ZERO(&set);
-        CPU_SET(core_id, &set);
-        pthread_setaffinity_np(pthread_self(), sizeof(set), &set);
-#elif defined(_WIN32)
-        DWORD set = 1UL << core_id;
-        SetThreadAffinityMask(GetCurrentThread(), set);
-#else
-        // No pin functionality implemented
+    SetCurrentThreadCoreGroup(CoreGroup::Performance);
 #endif
+}
+
+void SetCurrentThreadToEfficiencyCores() {
+#if defined(__ANDROID__)
+    SetCurrentThreadCoreGroup(CoreGroup::Efficiency);
+#endif
+}
+
+void RefreshThreadPolicies() {
+#if defined(__ANDROID__)
+    std::scoped_lock topology_lock{g_topology_mutex};
+    RefreshTopologyLocked();
+
+    std::scoped_lock policy_lock{g_policy_mutex};
+    for (const auto& policy : Policies()) {
+        if (policy.has_nice) {
+            setpriority(PRIO_PROCESS, static_cast<id_t>(policy.tid), policy.nice_value);
+        }
+        ApplyCoreGroupLocked(policy.tid, policy.group);
     }
+#endif
 }
 
 #ifdef ARCHITECTURE_x86_64

src/common/thread.h

@@ -99,8 +99,16 @@ enum class ThreadPriority : u32 {
     Critical = 4,
 };
 
+enum class ThreadPlacement : u32 {
+    Default = 0,
+    Background = 1,
+    Efficiency = 2,
+};
+
 void SetCurrentThreadPriority(ThreadPriority new_priority);
 void SetCurrentThreadName(const char* name);
-void PinCurrentThreadToPerformanceCore(size_t core_id);
+void SetCurrentThreadToPerformanceCores();
+void SetCurrentThreadToEfficiencyCores();
+void RefreshThreadPolicies();
 
 } // namespace Common

src/common/thread_worker.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
@@ -37,10 +37,17 @@ class StatefulThreadWorker {
     using StateMaker = std::conditional_t<with_state, std::function<StateType()>, DummyCallable>;
 
 public:
-    explicit StatefulThreadWorker(size_t num_workers, std::string name, StateMaker func = {})
+    explicit StatefulThreadWorker(size_t num_workers, std::string name, StateMaker func = {},
+                                  ThreadPlacement placement = ThreadPlacement::Default)
         : workers_queued{num_workers}, thread_name{std::move(name)} {
-        const auto lambda = [this, func](std::stop_token stop_token) {
+        const auto lambda = [this, func, placement](std::stop_token stop_token) {
             Common::SetCurrentThreadName(thread_name.c_str());
+            if (placement != ThreadPlacement::Default) {
+                Common::SetCurrentThreadPriority(ThreadPriority::Low);
+            }
+            if (placement == ThreadPlacement::Efficiency) {
+                Common::SetCurrentThreadToEfficiencyCores();
+            }
             {
                 [[maybe_unused]] std::conditional_t<with_state, StateType, int> state{func()};
                 while (!stop_token.stop_requested()) {

src/core/arm/nce/arm_nce.cpp

@@ -157,6 +157,8 @@ bool ArmNce::HandleGuestAlignmentFault(GuestContext* guest_ctx, void* raw_info,
     return HandleFailedGuestFault(guest_ctx, raw_info, raw_context);
 }
 
+constexpr size_t NCE_WRITE_FAULT_CLUSTER_PAGES = 4;
+
 bool ArmNce::HandleGuestAccessFault(GuestContext* guest_ctx, void* raw_info, void* raw_context) {
     auto* info = static_cast<siginfo_t*>(raw_info);
 
@@ -165,7 +167,7 @@ bool ArmNce::HandleGuestAccessFault(GuestContext* guest_ctx, void* raw_info, voi
     const Common::ProcessAddress addr =
         (reinterpret_cast<u64>(info->si_addr) & ~Memory::YUZU_PAGEMASK);
     auto& memory = guest_ctx->parent->m_running_thread->GetOwnerProcess()->GetMemory();
-    if (memory.InvalidateNCE(addr, Memory::YUZU_PAGESIZE)) {
+    if (memory.InvalidateNCE(addr, Memory::YUZU_PAGESIZE * NCE_WRITE_FAULT_CLUSTER_PAGES)) {
         // We handled the access successfully and are returning to guest code.
         return true;
     }

src/core/arm/nce/patcher.cpp

@@ -145,14 +145,7 @@ bool Patcher::PatchText(std::span<const u8> program_image, const Kernel::CodeSet
 
         // MRS Xn, CNTFRQ_EL0
         if (auto mrs = MRS{inst}; mrs.Verify() && mrs.GetSystemReg() == CntfrqEl0) {
-            bool pre_buffer = false;
-            auto ret = AddRelocations(pre_buffer);
-            if (pre_buffer) {
-                WriteCntfrqHandler(ret, oaknut::XReg{static_cast<int>(mrs.GetRt())}, c_pre);
-            } else {
-                WriteCntfrqHandler(ret, oaknut::XReg{static_cast<int>(mrs.GetRt())}, c);
-            }
-            continue;
+            UNREACHABLE();
         }
 
         // MSR TPIDR_EL0, Xn
@@ -584,16 +577,6 @@ void Patcher::WriteMsrHandler(ModuleDestLabel module_dest, oaknut::XReg src_reg,
         this->BranchToModule(module_dest);
 }
 
-void Patcher::WriteCntfrqHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg, oaknut::VectorCodeGenerator& cg) {
-    cg.MOV(dest_reg, Common::WallClock::CNTFRQ);
-
-    // Jump back to the instruction after the emulated MRS.
-    if (&cg == &c_pre)
-        this->BranchToModulePre(module_dest);
-    else
-        this->BranchToModule(module_dest);
-}
-
 void Patcher::WriteCntpctHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg, oaknut::VectorCodeGenerator& cg) {
 #if defined(HAS_NCE)
     static Common::WallClock clock(false, 1);

src/core/arm/nce/patcher.h

@@ -80,7 +80,6 @@ private:
     void WriteSvcTrampoline(ModuleDestLabel module_dest, u32 svc_id, oaknut::VectorCodeGenerator& code, oaknut::Label& save_ctx, oaknut::Label& load_ctx);
     void WriteMrsHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg, oaknut::SystemReg src_reg, oaknut::VectorCodeGenerator& code);
     void WriteMsrHandler(ModuleDestLabel module_dest, oaknut::XReg src_reg, oaknut::VectorCodeGenerator& code);
-    void WriteCntfrqHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg, oaknut::VectorCodeGenerator& code);
     void WriteCntpctHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg, oaknut::VectorCodeGenerator& code);
 
     // Convenience wrappers using default code generator
@@ -91,7 +90,6 @@ private:
     void WriteSvcTrampoline(ModuleDestLabel module_dest, u32 svc_id) { WriteSvcTrampoline(module_dest, svc_id, c, m_save_context, m_load_context); }
     void WriteMrsHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg, oaknut::SystemReg src_reg) { WriteMrsHandler(module_dest, dest_reg, src_reg, c); }
     void WriteMsrHandler(ModuleDestLabel module_dest, oaknut::XReg src_reg) { WriteMsrHandler(module_dest, src_reg, c); }
-    void WriteCntfrqHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg) { WriteCntfrqHandler(module_dest, dest_reg, c); }
     void WriteCntpctHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg) { WriteCntpctHandler(module_dest, dest_reg, c); }
 
 private:

src/core/core.cpp

@@ -118,6 +118,7 @@ struct System::Impl {
 
         is_multicore = Settings::values.use_multi_core.GetValue();
         extended_memory_layout = Settings::values.memory_layout_mode.GetValue() != Settings::MemoryLayout::Memory_4Gb;
+        unified_memory = Settings::values.use_unified_memory.GetValue();
 
         core_timing.SetMulticore(is_multicore);
         core_timing.Initialize([&system]() { system.RegisterHostThread(); });
@@ -145,7 +146,8 @@ struct System::Impl {
             !device_memory.has_value() ||
             is_multicore != Settings::values.use_multi_core.GetValue() ||
             extended_memory_layout != (Settings::values.memory_layout_mode.GetValue() !=
-                                       Settings::MemoryLayout::Memory_4Gb);
+                                       Settings::MemoryLayout::Memory_4Gb) ||
+            unified_memory != Settings::values.use_unified_memory.GetValue();
 
         if (!must_reinitialize) {
             return;
@@ -156,6 +158,7 @@ struct System::Impl {
         is_multicore = Settings::values.use_multi_core.GetValue();
         extended_memory_layout =
             Settings::values.memory_layout_mode.GetValue() != Settings::MemoryLayout::Memory_4Gb;
+        unified_memory = Settings::values.use_unified_memory.GetValue();
 
         Initialize(system);
     }
@@ -503,6 +506,7 @@ struct System::Impl {
     std::atomic_bool is_powered_on{};
     bool is_multicore : 1 = false;
     bool extended_memory_layout : 1 = false;
+    bool unified_memory : 1 = false;
     bool exit_locked : 1 = false;
     bool exit_requested : 1 = false;
     bool nvdec_active : 1 = false;

src/core/core_timing.cpp

@@ -58,7 +58,8 @@ void CoreTiming::Initialize(std::function<void()>&& on_thread_init_) {
     if (is_multicore) {
         timer_thread = std::jthread([this](std::stop_token stop_token) {
             Common::SetCurrentThreadName("HostTiming");
-            Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
+            Common::SetCurrentThreadPriority(Common::ThreadPriority::VeryHigh);
+            Common::SetCurrentThreadToPerformanceCores();
             on_thread_init();
             has_started = true;
 

src/core/cpu_manager.cpp

@@ -174,12 +174,7 @@ void CpuManager::RunThread(std::stop_token token, std::size_t core) {
     std::string name = is_multicore ? ("CPUCore_" + std::to_string(core)) : std::string{"CPUThread"};
     Common::SetCurrentThreadName(name.c_str());
     Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical);
-#ifdef __ANDROID__
-    // Aimed specifically for Snapdragon 8 Elite devices
-    // This kills performance on desktop, but boosts perf for UMA devices
-    // like the S8E. Mediatek and Mali likely won't suffer.
-    Common::PinCurrentThreadToPerformanceCore(core);
-#endif
+    Common::SetCurrentThreadToPerformanceCores();
     auto& data = core_data[core];
     data.host_context = Common::Fiber::ThreadToFiber();
 

src/core/device_memory.cpp

@@ -12,9 +12,18 @@ constexpr size_t VirtualReserveSize = 1ULL << 38;
 constexpr size_t VirtualReserveSize = 1ULL << 39;
 #endif
 
+namespace {
+size_t ApplicationPoolOffset() {
+    using Init = Kernel::Board::Nintendo::Nx::KSystemControl::Init;
+    const size_t dram_size = Init::GetIntendedMemorySize();
+    const size_t application_pool_size = Init::GetApplicationPoolSize();
+    return dram_size > application_pool_size ? dram_size - application_pool_size : 0;
+}
+}
+
 DeviceMemory::DeviceMemory()
     : buffer{Kernel::Board::Nintendo::Nx::KSystemControl::Init::GetIntendedMemorySize(),
-             VirtualReserveSize} {}
+             VirtualReserveSize, ApplicationPoolOffset()} {}
 
 DeviceMemory::~DeviceMemory() = default;
 

src/core/device_memory_manager.h

@@ -20,6 +20,8 @@
 #include "common/scratch_buffer.h"
 #include "common/virtual_buffer.h"
 
+struct AHardwareBuffer;
+
 namespace Core {
 
 constexpr size_t DEVICE_PAGEBITS = 12ULL;
@@ -95,6 +97,34 @@ public:
         ApplyOpOnPAddr(address, buffer, operation);
     }
 
+    u8* GetPhysicalBase() noexcept {
+        return reinterpret_cast<u8*>(physical_base);
+    }
+
+    const u8* GetPhysicalBase() const noexcept {
+        return reinterpret_cast<const u8*>(physical_base);
+    }
+
+    size_t GetPhysicalSize() const noexcept {
+        return physical_size;
+    }
+
+    std::span<AHardwareBuffer* const> GetBackingHardwareBuffers() const noexcept {
+        return ahb_windows;
+    }
+
+    size_t GetBackingHardwareBufferWindowSize() const noexcept {
+        return ahb_window_size;
+    }
+
+    size_t GetBackingHardwareBufferBase() const noexcept {
+        return ahb_base;
+    }
+
+    bool IsBackingShared() const noexcept {
+        return backing_is_shared;
+    }
+
     PAddr GetPhysicalRawAddressFromDAddr(DAddr address) const {
         PAddr subbits = PAddr(address & page_mask);
         auto paddr = tracked_entries[(address >> page_bits)].compressed_physical_ptr;
@@ -126,6 +156,10 @@ public:
     // New batch API to update multiple ranges with a single lock acquisition.
     void UpdatePagesCachedBatch(std::span<const std::pair<DAddr, size_t>> ranges, s32 delta);
 
+    void UpdateTexturePagesCount(DAddr addr, size_t size, s32 delta);
+
+    [[nodiscard]] bool IsRegionTextureCached(DAddr addr, size_t size) const noexcept;
+
 private:
     struct TranslationEntry {
         DAddr guest_page{};
@@ -171,6 +205,11 @@ private:
     std::unique_ptr<DeviceMemoryManagerAllocator<Traits>> impl;
 
     const uintptr_t physical_base;
+    const size_t physical_size;
+    const std::span<AHardwareBuffer* const> ahb_windows;
+    const size_t ahb_window_size;
+    const size_t ahb_base;
+    const bool backing_is_shared;
     DeviceInterface* device_inter;
 
     struct TrackedEntry {
@@ -234,6 +273,7 @@ private:
         (1ULL << (device_virtual_bits - page_bits)) / subentries;
     using CachedPages = std::array<CounterEntry, num_counter_entries>;
     std::unique_ptr<CachedPages> cached_pages;
+    std::unique_ptr<CachedPages> texture_cached_pages;
     Common::RangeMutex counter_guard;
     std::mutex mapping_guard;
 

src/core/device_memory_manager.inc

@@ -171,12 +171,18 @@ struct DeviceMemoryManagerAllocator {
 template <typename Traits>
 DeviceMemoryManager<Traits>::DeviceMemoryManager(const DeviceMemory& device_memory_)
     : physical_base{uintptr_t(device_memory_.buffer.BackingBasePointer())}
+    , physical_size{device_memory_.buffer.BackingSize()}
+    , ahb_windows{device_memory_.buffer.BackingHardwareBuffers()}
+    , ahb_window_size{device_memory_.buffer.BackingHardwareBufferWindowSize()}
+    , ahb_base{device_memory_.buffer.BackingHardwareBufferBase()}
+    , backing_is_shared{device_memory_.buffer.IsBackingShared()}
     , device_inter{nullptr}
     , compressed_device_addr(1ULL << ((Settings::values.memory_layout_mode.GetValue() == Settings::MemoryLayout::Memory_4Gb ? physical_min_bits : physical_max_bits) - Memory::YUZU_PAGEBITS))
     , tracked_entries(device_as_size >> Memory::YUZU_PAGEBITS)
 {
     impl = std::make_unique<DeviceMemoryManagerAllocator<Traits>>();
     cached_pages = std::make_unique<CachedPages>();
+    texture_cached_pages = std::make_unique<CachedPages>();
 
     const size_t total_virtual = device_as_size >> Memory::YUZU_PAGEBITS;
     for (size_t i = 0; i < total_virtual; i++) {
@@ -625,6 +631,28 @@ void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size
     UpdatePagesCachedCountNoLock(addr, size, delta);
 }
 
+template <typename Traits>
+void DeviceMemoryManager<Traits>::UpdateTexturePagesCount(DAddr addr, size_t size, s32 delta) {
+    Common::ScopedRangeLock lk(counter_guard, addr, size);
+    const size_t page_end = Common::DivCeil(addr + size, Memory::YUZU_PAGESIZE);
+    for (size_t page = addr >> Memory::YUZU_PAGEBITS; page != page_end; ++page) {
+        CounterAtomicType& count = texture_cached_pages->at(page >> subentries_shift).Count(page);
+        count.fetch_add(static_cast<CounterType>(delta), std::memory_order_release);
+    }
+}
+
+template <typename Traits>
+bool DeviceMemoryManager<Traits>::IsRegionTextureCached(DAddr addr, size_t size) const noexcept {
+    const size_t page_end = Common::DivCeil(addr + size, Memory::YUZU_PAGESIZE);
+    for (size_t page = addr >> Memory::YUZU_PAGEBITS; page != page_end; ++page) {
+        if (texture_cached_pages->at(page >> subentries_shift).Count(page).load(
+                std::memory_order_acquire) != 0) {
+            return true;
+        }
+    }
+    return false;
+}
+
 template <typename Traits>
 void DeviceMemoryManager<Traits>::UpdatePagesCachedBatch(std::span<const std::pair<DAddr, size_t>> ranges, s32 delta) {
     if (ranges.empty()) {

src/core/file_sys/registered_cache.cpp

@@ -984,22 +984,6 @@ bool RegisteredCache::RemoveExistingEntry(u64 title_id) const {
     return removed_data;
 }
 
-bool RegisteredCache::Delete(const NcaID& id) const {
-    const auto path = GetRelativePathFromNcaID(id, false, true, false);
-
-    const bool is_file = dir->GetFileRelative(path) != nullptr;
-    const bool is_dir = dir->GetDirectoryRelative(path) != nullptr;
-
-    if (is_file) {
-        return dir->DeleteFile(path);
-    }
-    if (is_dir) {
-        return dir->DeleteSubdirectoryRecursive(path);
-    }
-
-    return true;
-}
-
 InstallResult RegisteredCache::RawInstallNCA(const NCA& nca, const VfsCopyFunction& copy,
                                              bool overwrite_if_exists,
                                              std::optional<NcaID> override_id) {

src/core/file_sys/registered_cache.h

@@ -188,7 +188,6 @@ public:
 
     // Removes an existing entry based on title id
     bool RemoveExistingEntry(u64 title_id) const;
-    bool Delete(const NcaID& id) const;
 
 private:
     template <typename T>

src/core/file_sys/vfs/vfs_real.cpp

@@ -35,16 +35,6 @@ namespace {
 
 constexpr size_t MaxOpenFiles = 8192;
 
-bool IsWithinRoot(std::string_view root, std::string_view full_path) {
-    if (root.empty())
-        return true;
-
-    if (full_path.size() < root.size() || full_path.substr(0, root.size()) != root)
-        return false;
-
-    return full_path.size() == root.size() || full_path[root.size()] == '/' || full_path[root.size()] == '\\';
-}
-
 constexpr FS::FileAccessMode ModeFlagsToFileAccessMode(OpenMode mode) {
     switch (mode) {
     case OpenMode::Read:
@@ -413,8 +403,7 @@ RealVfsDirectory::~RealVfsDirectory() = default;
 
 VirtualFile RealVfsDirectory::GetFileRelative(std::string_view relative_path) const {
     const auto full_path = FS::SanitizePath(path + '/' + std::string(relative_path));
-    if (!FS::Exists(full_path) || FS::IsDir(full_path)
-        || !IsWithinRoot(FS::SanitizePath(path), full_path)) {
+    if (!FS::Exists(full_path) || FS::IsDir(full_path)) {
         return nullptr;
     }
     return base.OpenFile(full_path, perms);
@@ -422,8 +411,7 @@ VirtualFile RealVfsDirectory::GetFileRelative(std::string_view relative_path) co
 
 VirtualDir RealVfsDirectory::GetDirectoryRelative(std::string_view relative_path) const {
     const auto full_path = FS::SanitizePath(path + '/' + std::string(relative_path));
-    if (!FS::Exists(full_path) || !FS::IsDir(full_path)
-        || !IsWithinRoot(FS::SanitizePath(path), full_path)) {
+    if (!FS::Exists(full_path) || !FS::IsDir(full_path)) {
         return nullptr;
     }
     return base.OpenDirectory(full_path, perms);
@@ -439,7 +427,7 @@ VirtualDir RealVfsDirectory::GetSubdirectory(std::string_view name) const {
 
 VirtualFile RealVfsDirectory::CreateFileRelative(std::string_view relative_path) {
     const auto full_path = FS::SanitizePath(path + '/' + std::string(relative_path));
-    if (!FS::CreateParentDirs(full_path) || !IsWithinRoot(FS::SanitizePath(path), full_path)) {
+    if (!FS::CreateParentDirs(full_path)) {
         return nullptr;
     }
     return base.CreateFile(full_path, perms);
@@ -452,7 +440,7 @@ VirtualDir RealVfsDirectory::CreateDirectoryRelative(std::string_view relative_p
 
 bool RealVfsDirectory::DeleteSubdirectoryRecursive(std::string_view name) {
     const auto full_path = FS::SanitizePath(this->path + '/' + std::string(name));
-    return FS::RemoveDirRecursively(full_path);
+    return base.DeleteDirectory(full_path);
 }
 
 std::vector<VirtualFile> RealVfsDirectory::GetFiles() const {
@@ -518,7 +506,7 @@ VirtualFile RealVfsDirectory::CreateFile(std::string_view name) {
 
 bool RealVfsDirectory::DeleteSubdirectory(std::string_view name) {
     const std::string subdir_path = (path + '/').append(name);
-    return FS::RemoveDir(subdir_path);
+    return base.DeleteDirectory(subdir_path);
 }
 
 bool RealVfsDirectory::DeleteFile(std::string_view name) {

src/core/hle/service/am/service/self_controller.cpp

@@ -57,7 +57,7 @@ ISelfController::ISelfController(Core::System& system_, std::shared_ptr<Applet>
         {64, nullptr, "SetInputDetectionSourceSet"},
         {65, D<&ISelfController::ReportUserIsActive>, "ReportUserIsActive"},
         {66, nullptr, "GetCurrentIlluminance"},
-        {67, D<&ISelfController::IsIlluminanceAvailable>, "IsIlluminanceAvailable"},
+        {67, nullptr, "IsIlluminanceAvailable"},
         {68, D<&ISelfController::SetAutoSleepDisabled>, "SetAutoSleepDisabled"},
         {69, D<&ISelfController::IsAutoSleepDisabled>, "IsAutoSleepDisabled"},
         {70, nullptr, "ReportMultimediaError"},
@@ -347,12 +347,6 @@ Result ISelfController::IsAutoSleepDisabled(Out<bool> out_is_auto_sleep_disabled
     R_SUCCEED();
 }
 
-Result ISelfController::IsIlluminanceAvailable(Out<bool> out_is_illuminance_available) {
-    LOG_WARNING(Service_AM, "(stubbed)");
-    *out_is_illuminance_available = false;
-    R_SUCCEED();
-}
-
 Result ISelfController::SetInputDetectionPolicy(InputDetectionPolicy input_detection_policy) {
     LOG_WARNING(Service_AM, "(STUBBED) called");
     R_SUCCEED();

src/core/hle/service/am/service/self_controller.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2024 yuzu Emulator Project
@@ -60,7 +60,6 @@ private:
     Result ReportUserIsActive();
     Result SetAutoSleepDisabled(bool is_auto_sleep_disabled);
     Result IsAutoSleepDisabled(Out<bool> out_is_auto_sleep_disabled);
-    Result IsIlluminanceAvailable(Out<bool> out_is_illuminance_available);
     Result SetInputDetectionPolicy(InputDetectionPolicy input_detection_policy);
     Result GetAccumulatedSuspendedTickValue(Out<u64> out_accumulated_suspended_tick_value);
     Result GetAccumulatedSuspendedTickChangedEvent(OutCopyHandle<Kernel::KReadableEvent> out_event);

src/core/hle/service/filesystem/filesystem.cpp

@@ -43,16 +43,6 @@ static FileSys::VirtualDir GetDirectoryRelativeWrapped(FileSys::VirtualDir base,
     return base->GetDirectoryRelative(dir_name);
 }
 
-static std::string_view GetGuestParentPath(std::string_view path) {
-    const auto name_index = path.find_last_of("\\/");
-    return name_index == std::string_view::npos ? std::string_view{} : path.substr(0, name_index);
-}
-
-static std::string_view GetGuestFilename(std::string_view path) {
-    const auto name_index = path.find_last_of("\\/");
-    return name_index == std::string_view::npos ? path : path.substr(name_index + 1);
-}
-
 VfsDirectoryServiceWrapper::VfsDirectoryServiceWrapper(FileSys::VirtualDir backing_)
     : backing(std::move(backing_)) {}
 
@@ -93,12 +83,11 @@ Result VfsDirectoryServiceWrapper::DeleteFile(const std::string& path_) const {
         return ResultSuccess;
     }
 
-    const auto filename = GetGuestFilename(path);
-    auto dir = GetDirectoryRelativeWrapped(backing, GetGuestParentPath(path));
-    if (filename.empty() || dir == nullptr || dir->GetFile(filename) == nullptr) {
+    auto dir = GetDirectoryRelativeWrapped(backing, Common::FS::GetParentPath(path));
+    if (dir == nullptr || dir->GetFile(Common::FS::GetFilename(path)) == nullptr) {
         return FileSys::ResultPathNotFound;
     }
-    if (!dir->DeleteFile(filename)) {
+    if (!dir->DeleteFile(Common::FS::GetFilename(path))) {
         // TODO(DarkLordZach): Find a better error code for this
         return ResultUnknown;
     }
@@ -108,9 +97,7 @@ Result VfsDirectoryServiceWrapper::DeleteFile(const std::string& path_) const {
 
 Result VfsDirectoryServiceWrapper::CreateDirectory(const std::string& path_) const {
     std::string path(Common::FS::SanitizePath(path_));
-    if (GetDirectoryRelativeWrapped(backing, path) != nullptr) {
-        return FileSys::ResultPathAlreadyExists;
-    }
+
     // NOTE: This is inaccurate behavior. CreateDirectory is not recursive.
     // CreateDirectory should return PathNotFound if the parent directory does not exist.
     // This is here temporarily in order to have UMM "work" in the meantime.
@@ -130,19 +117,8 @@ Result VfsDirectoryServiceWrapper::CreateDirectory(const std::string& path_) con
 
 Result VfsDirectoryServiceWrapper::DeleteDirectory(const std::string& path_) const {
     std::string path(Common::FS::SanitizePath(path_));
-    const auto dirname = GetGuestFilename(path);
-    auto dir = GetDirectoryRelativeWrapped(backing, GetGuestParentPath(path));
-    FileSys::VirtualDir target{};
-    if (!dirname.empty() && dir != nullptr) {
-        target = dir->GetSubdirectory(dirname);
-    }
-    if (target == nullptr) {
-        return FileSys::ResultPathNotFound;
-    }
-    if (!target->GetFiles().empty() || !target->GetSubdirectories().empty()) {
-        return ResultUnknown;
-    }
-    if (!dir->DeleteSubdirectory(dirname)) {
+    auto dir = GetDirectoryRelativeWrapped(backing, Common::FS::GetParentPath(path));
+    if (!dir->DeleteSubdirectory(Common::FS::GetFilename(path))) {
         // TODO(DarkLordZach): Find a better error code for this
         return ResultUnknown;
     }
@@ -151,12 +127,8 @@ Result VfsDirectoryServiceWrapper::DeleteDirectory(const std::string& path_) con
 
 Result VfsDirectoryServiceWrapper::DeleteDirectoryRecursively(const std::string& path_) const {
     std::string path(Common::FS::SanitizePath(path_));
-    const auto dirname = GetGuestFilename(path);
-    auto dir = GetDirectoryRelativeWrapped(backing, GetGuestParentPath(path));
-    if (dirname.empty() || dir == nullptr || dir->GetSubdirectory(dirname) == nullptr) {
-        return FileSys::ResultPathNotFound;
-    }
-    if (!dir->DeleteSubdirectoryRecursive(dirname)) {
+    auto dir = GetDirectoryRelativeWrapped(backing, Common::FS::GetParentPath(path));
+    if (!dir->DeleteSubdirectoryRecursive(Common::FS::GetFilename(path))) {
         // TODO(DarkLordZach): Find a better error code for this
         return ResultUnknown;
     }
@@ -165,13 +137,9 @@ Result VfsDirectoryServiceWrapper::DeleteDirectoryRecursively(const std::string&
 
 Result VfsDirectoryServiceWrapper::CleanDirectoryRecursively(const std::string& path) const {
     const std::string sanitized_path(Common::FS::SanitizePath(path));
-    const auto dirname = GetGuestFilename(sanitized_path);
-    auto dir = GetDirectoryRelativeWrapped(backing, GetGuestParentPath(sanitized_path));
+    auto dir = GetDirectoryRelativeWrapped(backing, Common::FS::GetParentPath(sanitized_path));
 
-    if (dirname.empty() || dir == nullptr || dir->GetSubdirectory(dirname) == nullptr) {
-        return FileSys::ResultPathNotFound;
-    }
-    if (!dir->CleanSubdirectoryRecursive(dirname)) {
+    if (!dir->CleanSubdirectoryRecursive(Common::FS::GetFilename(sanitized_path))) {
         // TODO(DarkLordZach): Find a better error code for this
         return ResultUnknown;
     }

src/core/hle/service/hid/hid_server.cpp

@@ -237,7 +237,7 @@ IHidServer::IHidServer(Core::System& system_, std::shared_ptr<ResourceManager> r
         {3013, nullptr, "SetDebugPadGenericPadMap"}, //21.0.0+
         {3014, nullptr, "GetDebugPadKeyboardMap"}, //21.0.0+
         {3015, nullptr, "SetDebugPadKeyboardMap"}, //21.0.0+
-        {3150, C<&IHidServer::SetMouseLibraryVersion>, "SetMouseLibraryVersion"}, //21.0.0+
+        {3150, nullptr, "SetMouseLibraryVersion"}, //21.0.0+
         // What? -- {12010, nullptr, "SetButtonConfigLeft"},
     };
     // clang-format on
@@ -1471,12 +1471,6 @@ Result IHidServer::SetTouchScreenResolution(u32 width, u32 height,
     R_SUCCEED();
 }
 
-Result IHidServer::SetMouseLibraryVersion(ClientAppletResourceUserId aruid) {
-    LOG_INFO(Service_HID, "(STUBBED) called, applet_resource_user_id={}", aruid.pid);
-
-    R_SUCCEED();
-}
-
 std::shared_ptr<ResourceManager> IHidServer::GetResourceManager() {
     resource_manager->Initialize();
     return resource_manager;

src/core/hle/service/hid/hid_server.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
@@ -263,7 +263,6 @@ private:
     Result IsFirmwareUpdateNeededForNotification(Out<bool> out_is_firmware_update_needed,
                                                  s32 unknown, ClientAppletResourceUserId aruid);
     Result SetTouchScreenResolution(u32 width, u32 height, ClientAppletResourceUserId aruid);
-    Result SetMouseLibraryVersion(ClientAppletResourceUserId aruid);
 
     std::shared_ptr<ResourceManager> resource_manager;
     std::shared_ptr<HidFirmwareSettings> firmware_settings;

src/core/hle/service/ncm/ncm.cpp

@@ -1,19 +1,9 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <algorithm>
-#include <array>
 #include <memory>
-#include <vector>
 
-#include "core/core.h"
-#include "core/file_sys/registered_cache.h"
 #include "core/file_sys/romfs_factory.h"
-#include "core/hle/api_version.h"
-#include "core/hle/service/filesystem/filesystem.h"
 #include "core/hle/service/ipc_helpers.h"
 #include "core/hle/service/ncm/ncm.h"
 #include "core/hle/service/server_manager.h"
@@ -98,268 +88,6 @@ public:
     }
 };
 
-class IContentStorage final : public ServiceFramework<IContentStorage> {
-public:
-    explicit IContentStorage(Core::System& system_, FileSys::StorageId id)
-        : ServiceFramework{system_, "IContentStorage"}, storage{id} {
-        // clang-format off
-        static const FunctionInfo functions[] = {
-            {0, &IContentStorage::GeneratePlaceHolderId, "GeneratePlaceHolderId"},
-            {1, &IContentStorage::CreatePlaceHolder, "CreatePlaceHolder"},
-            {2, &IContentStorage::DeletePlaceHolder, "DeletePlaceHolder"},
-            {4, &IContentStorage::WritePlaceHolder, "WritePlaceHolder"},
-            {5, &IContentStorage::Register, "Register"},
-            {6, &IContentStorage::Delete, "Delete"},
-        };
-        // clang-format on
-
-        RegisterHandlers(functions);
-    }
-
-private:
-    void GeneratePlaceHolderId(HLERequestContext& ctx) {
-        LOG_DEBUG(Service_NCM, "called");
-
-        IPC::ResponseBuilder rb{ctx, 6};
-        rb.Push(ResultSuccess);
-        rb.PushRaw(FileSys::PlaceholderCache::Generate());
-    }
-
-    void CreatePlaceHolder(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        [[maybe_unused]] FileSys::NcaID content_id{};
-        FileSys::NcaID placeholder_id{};
-        if constexpr (HLE::ApiVersion::HOS_VERSION_MAJOR >= 16) {
-            placeholder_id = rp.PopRaw<FileSys::NcaID>();
-            content_id = rp.PopRaw<FileSys::NcaID>();
-        } else {
-            content_id = rp.PopRaw<FileSys::NcaID>();
-            placeholder_id = rp.PopRaw<FileSys::NcaID>();
-        }
-        const auto size = rp.Pop<s64>();
-
-        auto* const placeholder_cache =
-            system.GetFileSystemController().GetPlaceholderCacheForStorage(storage);
-        const bool succeeded =
-            placeholder_cache != nullptr && size >= 0 &&
-            (placeholder_cache->Exists(placeholder_id) ||
-             placeholder_cache->Create(placeholder_id, static_cast<u64>(size)));
-
-        if (succeeded) {
-            LOG_DEBUG(Service_NCM, "called, storage_id={}, size={}", static_cast<u32>(storage),
-                      size);
-        } else {
-            LOG_WARNING(Service_NCM, "failed, storage_id={}, size={}", static_cast<u32>(storage),
-                        size);
-        }
-
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(succeeded ? ResultSuccess : ResultUnknown);
-    }
-
-    void DeletePlaceHolder(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const auto placeholder_id = rp.PopRaw<FileSys::NcaID>();
-
-        auto* const placeholder_cache =
-            system.GetFileSystemController().GetPlaceholderCacheForStorage(storage);
-        const bool succeeded =
-            placeholder_cache != nullptr &&
-            (!placeholder_cache->Exists(placeholder_id) || placeholder_cache->Delete(placeholder_id));
-
-        if (succeeded) {
-            LOG_DEBUG(Service_NCM, "called, storage_id={}", static_cast<u32>(storage));
-        } else {
-            LOG_WARNING(Service_NCM, "failed, storage_id={}", static_cast<u32>(storage));
-        }
-
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(succeeded ? ResultSuccess : ResultUnknown);
-    }
-
-    void WritePlaceHolder(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const auto placeholder_id = rp.PopRaw<FileSys::NcaID>();
-        const auto offset = rp.Pop<u64>();
-        const auto data = ctx.ReadBuffer();
-
-        auto* const placeholder_cache =
-            system.GetFileSystemController().GetPlaceholderCacheForStorage(storage);
-        const std::vector<u8> write_data{data.begin(), data.end()};
-        const bool succeeded =
-            placeholder_cache != nullptr &&
-            placeholder_cache->Write(placeholder_id, offset, write_data);
-
-        if (succeeded) {
-            LOG_DEBUG(Service_NCM, "called, storage_id={}, offset={}, size={}",
-                      static_cast<u32>(storage), offset, data.size());
-        } else {
-            LOG_WARNING(Service_NCM, "failed, storage_id={}, offset={}, size={}",
-                        static_cast<u32>(storage), offset, data.size());
-        }
-
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(succeeded ? ResultSuccess : ResultUnknown);
-    }
-
-    void Register(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        FileSys::NcaID content_id{};
-        FileSys::NcaID placeholder_id{};
-        if constexpr (HLE::ApiVersion::HOS_VERSION_MAJOR >= 16) {
-            placeholder_id = rp.PopRaw<FileSys::NcaID>();
-            content_id = rp.PopRaw<FileSys::NcaID>();
-        } else {
-            content_id = rp.PopRaw<FileSys::NcaID>();
-            placeholder_id = rp.PopRaw<FileSys::NcaID>();
-        }
-
-        auto& fsc = system.GetFileSystemController();
-        auto* const placeholder_cache = fsc.GetPlaceholderCacheForStorage(storage);
-        auto* const registered_cache = fsc.GetRegisteredCacheForStorage(storage);
-        const bool succeeded =
-            placeholder_cache != nullptr && registered_cache != nullptr &&
-            placeholder_cache->Register(registered_cache, placeholder_id, content_id);
-
-        if (succeeded) {
-            LOG_DEBUG(Service_NCM, "called, storage_id={}", static_cast<u32>(storage));
-        } else {
-            LOG_WARNING(Service_NCM, "failed, storage_id={}", static_cast<u32>(storage));
-        }
-
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(succeeded ? ResultSuccess : ResultUnknown);
-    }
-
-    void Delete(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const auto content_id = rp.PopRaw<FileSys::NcaID>();
-
-        auto* const registered_cache =
-            system.GetFileSystemController().GetRegisteredCacheForStorage(storage);
-        const bool succeeded = registered_cache != nullptr && registered_cache->Delete(content_id);
-        if (succeeded) {
-            registered_cache->Refresh();
-        }
-
-        if (succeeded) {
-            LOG_DEBUG(Service_NCM, "called, storage_id={}", static_cast<u32>(storage));
-        } else {
-            LOG_WARNING(Service_NCM, "failed, storage_id={}", static_cast<u32>(storage));
-        }
-
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(succeeded ? ResultSuccess : ResultUnknown);
-    }
-
-    FileSys::StorageId storage;
-};
-
-class IContentMetaDatabase final : public ServiceFramework<IContentMetaDatabase> {
-public:
-    explicit IContentMetaDatabase(Core::System& system_, FileSys::StorageId id)
-        : ServiceFramework{system_, "IContentMetaDatabase"}, storage{id} {
-        // clang-format off
-        static const FunctionInfo functions[] = {
-            {0, &IContentMetaDatabase::Set, "Set"},
-            {2, &IContentMetaDatabase::Remove, "Remove"},
-            {8, &IContentMetaDatabase::Has, "Has"},
-            {15, &IContentMetaDatabase::Commit, "Commit"},
-        };
-        // clang-format on
-
-        RegisterHandlers(functions);
-    }
-
-private:
-    struct ContentMetaKey {
-        u64 id;
-        u32 version;
-        FileSys::TitleType type;
-        u8 install_type;
-        std::array<u8, 2> padding;
-    };
-    static_assert(sizeof(ContentMetaKey) == 0x10);
-
-    void Set(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const auto key = rp.PopRaw<ContentMetaKey>();
-
-        const auto entry_matches = [&key](const ContentMetaKey& entry) {
-            return entry.id == key.id && entry.version == key.version && entry.type == key.type &&
-                   entry.install_type == key.install_type;
-        };
-        if (std::find_if(entries.begin(), entries.end(), entry_matches) == entries.end()) {
-            entries.push_back(key);
-        }
-
-        LOG_DEBUG(Service_NCM,
-                  "called, storage_id={}, title_id={:016X}, version={}, type={}, size={}",
-                  static_cast<u32>(storage), key.id, key.version, static_cast<u8>(key.type),
-                  ctx.GetReadBufferSize());
-
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(ResultSuccess);
-    }
-
-    void Remove(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const auto key = rp.PopRaw<ContentMetaKey>();
-
-        std::erase_if(entries, [&key](const ContentMetaKey& entry) {
-            return entry.id == key.id && entry.version == key.version && entry.type == key.type &&
-                   entry.install_type == key.install_type;
-        });
-
-        LOG_DEBUG(Service_NCM, "called, storage_id={}, title_id={:016X}, version={}, type={}",
-                  static_cast<u32>(storage), key.id, key.version, static_cast<u8>(key.type));
-
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(ResultSuccess);
-    }
-
-    void Has(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const auto key = rp.PopRaw<ContentMetaKey>();
-
-        const bool has_pending =
-            std::find_if(entries.begin(), entries.end(), [&key](const ContentMetaKey& entry) {
-                return entry.id == key.id && entry.version == key.version &&
-                       entry.type == key.type && entry.install_type == key.install_type;
-            }) != entries.end();
-
-        auto* const registered_cache =
-            system.GetFileSystemController().GetRegisteredCacheForStorage(storage);
-        const bool has_registered =
-            registered_cache != nullptr &&
-            registered_cache->HasEntry(key.id, FileSys::ContentRecordType::Meta);
-
-        LOG_DEBUG(Service_NCM, "called, storage_id={}, title_id={:016X}, version={}, type={}, has={}",
-                  static_cast<u32>(storage), key.id, key.version, static_cast<u8>(key.type),
-                  has_pending || has_registered);
-
-        IPC::ResponseBuilder rb{ctx, 3};
-        rb.Push(ResultSuccess);
-        rb.Push(has_pending || has_registered);
-    }
-
-    void Commit(HLERequestContext& ctx) {
-        auto* const registered_cache =
-            system.GetFileSystemController().GetRegisteredCacheForStorage(storage);
-        if (registered_cache != nullptr) {
-            registered_cache->Refresh();
-        }
-
-        LOG_DEBUG(Service_NCM, "called, storage_id={}", static_cast<u32>(storage));
-
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(ResultSuccess);
-    }
-
-    FileSys::StorageId storage;
-    std::vector<ContentMetaKey> entries;
-};
-
 class LR final : public ServiceFramework<LR> {
 public:
     explicit LR(Core::System& system_) : ServiceFramework{system_, "lr"} {
@@ -385,8 +113,8 @@ public:
             {1, nullptr, "CreateContentMetaDatabase"},
             {2, nullptr, "VerifyContentStorage"},
             {3, nullptr, "VerifyContentMetaDatabase"},
-            {4, &NCM::OpenContentStorage, "OpenContentStorage"},
-            {5, &NCM::OpenContentMetaDatabase, "OpenContentMetaDatabase"},
+            {4, nullptr, "OpenContentStorage"},
+            {5, nullptr, "OpenContentMetaDatabase"},
             {6, nullptr, "CloseContentStorageForcibly"},
             {7, nullptr, "CloseContentMetaDatabaseForcibly"},
             {8, nullptr, "CleanupContentMetaDatabase"},
@@ -402,29 +130,6 @@ public:
 
         RegisterHandlers(functions);
     }
-
-private:
-    void OpenContentStorage(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const auto storage_id = rp.PopEnum<FileSys::StorageId>();
-
-        LOG_DEBUG(Service_NCM, "called, storage_id={}", static_cast<u32>(storage_id));
-
-        IPC::ResponseBuilder rb{ctx, 2, 0, 1};
-        rb.Push(ResultSuccess);
-        rb.PushIpcInterface<IContentStorage>(ctx, system, storage_id);
-    }
-
-    void OpenContentMetaDatabase(HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const auto storage_id = rp.PopEnum<FileSys::StorageId>();
-
-        LOG_DEBUG(Service_NCM, "called, storage_id={}", static_cast<u32>(storage_id));
-
-        IPC::ResponseBuilder rb{ctx, 2, 0, 1};
-        rb.Push(ResultSuccess);
-        rb.PushIpcInterface<IContentMetaDatabase>(ctx, system, storage_id);
-    }
 };
 
 void LoopProcess(Core::System& system) {

src/core/hle/service/ns/application_manager_interface.cpp

@@ -9,7 +9,6 @@
 #include "core/file_sys/registered_cache.h"
 #include "core/hle/service/cmif_serialization.h"
 #include "core/hle/service/filesystem/filesystem.h"
-#include "core/hle/service/ipc_helpers.h"
 #include "core/hle/service/ns/application_manager_interface.h"
 
 #include "core/file_sys/content_archive.h"
@@ -20,7 +19,6 @@
 #include "core/launch_timestamp_cache.h"
 
 #include <algorithm>
-#include <cstring>
 #include <vector>
 
 namespace Service::NS {
@@ -38,14 +36,14 @@ IApplicationManagerInterface::IApplicationManagerInterface(Core::System& system_
         {1, nullptr, "GenerateApplicationRecordCount"},
         {2, D<&IApplicationManagerInterface::GetApplicationRecordUpdateSystemEvent>, "GetApplicationRecordUpdateSystemEvent"},
         {3, nullptr, "GetApplicationViewDeprecated"},
-        {4, D<&IApplicationManagerInterface::DeleteApplicationEntity>, "DeleteApplicationEntity"},
-        {5, D<&IApplicationManagerInterface::DeleteApplicationCompletely>, "DeleteApplicationCompletely"},
+        {4, nullptr, "DeleteApplicationEntity"},
+        {5, nullptr, "DeleteApplicationCompletely"},
         {6, nullptr, "IsAnyApplicationEntityRedundant"},
         {7, nullptr, "DeleteRedundantApplicationEntity"},
         {8, nullptr, "IsApplicationEntityMovable"},
         {9, nullptr, "MoveApplicationEntity"},
         {11, nullptr, "CalculateApplicationOccupiedSize"},
-        {16, &IApplicationManagerInterface::PushApplicationRecord, "PushApplicationRecord"},
+        {16, nullptr, "PushApplicationRecord"},
         {17, nullptr, "ListApplicationRecordContentMeta"},
         {19, nullptr, "LaunchApplicationOld"},
         {21, nullptr, "GetApplicationContentPath"},
@@ -645,27 +643,6 @@ Result IApplicationManagerInterface::IsAnyApplicationEntityInstalled(
     R_SUCCEED();
 }
 
-Result IApplicationManagerInterface::DeleteApplicationEntity(u64 application_id) {
-    LOG_DEBUG(Service_NS, "called, application_id={:016X}", application_id);
-
-    auto& fsc = system.GetFileSystemController();
-    if (auto* const user_cache = fsc.GetUserNANDContents(); user_cache != nullptr) {
-        user_cache->RemoveExistingEntry(application_id);
-        user_cache->Refresh();
-    }
-    if (auto* const sdmc_cache = fsc.GetSDMCContents(); sdmc_cache != nullptr) {
-        sdmc_cache->RemoveExistingEntry(application_id);
-        sdmc_cache->Refresh();
-    }
-
-    record_update_system_event.Signal(system.Kernel());
-    R_SUCCEED();
-}
-
-Result IApplicationManagerInterface::DeleteApplicationCompletely(u64 application_id) {
-    R_RETURN(DeleteApplicationEntity(application_id));
-}
-
 Result IApplicationManagerInterface::GetApplicationViewDeprecated(
     OutArray<ApplicationViewV19, BufferAttr_HipcMapAlias> out_application_views,
     InArray<u64, BufferAttr_HipcMapAlias> application_ids) {
@@ -866,29 +843,6 @@ Result IApplicationManagerInterface::Unknown4053() {
     R_SUCCEED();
 }
 
-void IApplicationManagerInterface::PushApplicationRecord(HLERequestContext& ctx) {
-    const auto record = ctx.ReadBuffer();
-    u64 application_id{};
-    if (record.size() >= sizeof(application_id)) {
-        std::memcpy(&application_id, record.data(), sizeof(application_id));
-    }
-
-    LOG_DEBUG(Service_NS, "called, application_id={:016X}, size={}", application_id, record.size());
-
-    auto& fsc = system.GetFileSystemController();
-    if (auto* const user_cache = fsc.GetUserNANDContents(); user_cache != nullptr) {
-        user_cache->Refresh();
-    }
-    if (auto* const sdmc_cache = fsc.GetSDMCContents(); sdmc_cache != nullptr) {
-        sdmc_cache->Refresh();
-    }
-
-    record_update_system_event.Signal(system.Kernel());
-
-    IPC::ResponseBuilder rb{ctx, 2};
-    rb.Push(ResultSuccess);
-}
-
 void IApplicationManagerInterface::ListApplicationTitle(HLERequestContext& ctx) {
     LOG_DEBUG(Service_NS, "called");
     IReadOnlyApplicationControlDataInterface(system).ListApplicationTitle(ctx);

src/core/hle/service/ns/application_manager_interface.h

@@ -56,8 +56,6 @@ public:
     Result ResumeAll();
     Result IsQualificationTransitionSupportedByProcessId(Out<bool> out_is_supported,
                                                          u64 process_id);
-    Result DeleteApplicationEntity(u64 application_id);
-    Result DeleteApplicationCompletely(u64 application_id);
     Result GetStorageSize(Out<s64> out_total_space_size, Out<s64> out_free_space_size,
                           FileSys::StorageId storage_id);
     Result TouchApplication(u64 application_id);
@@ -76,7 +74,6 @@ public:
     Result RequestDownloadApplicationControlDataInBackground(u64 control_source,
                                                              u64 application_id);
 
-    void PushApplicationRecord(HLERequestContext& ctx);
     void ListApplicationTitle(HLERequestContext& ctx);
 
 private:

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp

@@ -375,39 +375,42 @@ NvResult nvhost_as_gpu::MapBufferEx(IoctlMapBufferEx& params) {
         mapping_map.insert_or_assign(params.offset, Mapping(params.handle, device_address, params.offset, size, false, big_page, false));
     }
 
-    map_buffer_offsets.insert(params.offset);
-
     return NvResult::Success;
 }
 
 NvResult nvhost_as_gpu::UnmapBuffer(IoctlUnmapBuffer& params) {
+    LOG_DEBUG(Service_NVDRV, "called, offset={:#X}", params.offset);
+
     std::scoped_lock lock(mutex);
-    if (auto const offset_it = map_buffer_offsets.find(params.offset); offset_it != map_buffer_offsets.end()) {
-        LOG_DEBUG(Service_NVDRV, "called, offset={:#X}", params.offset);
-        if (!vm.initialised) {
-            return NvResult::BadValue;
-        }
 
-        auto const it = mapping_map.find(params.offset);
-        auto const mapping = it->second;
-        if (!mapping.fixed) {
-            auto& allocator{mapping.big_page ? *vm.big_page_allocator : *vm.small_page_allocator};
-            u32 page_size_bits{mapping.big_page ? vm.big_page_size_bits : VM::PAGE_SIZE_BITS};
-            allocator.Free(u32(mapping.offset >> page_size_bits), u32(mapping.size >> page_size_bits));
-        }
-
-        // Sparse mappings shouldn't be fully unmapped, just returned to their sparse state
-        // Only FreeSpace can unmap them fully
-        if (mapping.sparse_alloc) {
-            gmmu->MapSparse(params.offset, mapping.size, mapping.big_page);
-        } else {
-            gmmu->Unmap(params.offset, mapping.size);
-        }
-
-        nvmap.UnpinHandle(mapping.handle);
-        mapping_map.erase(params.offset);
-        map_buffer_offsets.erase(params.offset);
+    if (!vm.initialised) {
+        return NvResult::BadValue;
     }
+
+    auto const it = mapping_map.find(params.offset);
+    if (it == mapping_map.end()) {
+        LOG_WARNING(Service_NVDRV, "Couldn't find region to unmap at {:#X}", params.offset);
+        return NvResult::Success;
+    }
+
+    auto const mapping = it->second;
+    if (!mapping.fixed) {
+        auto& allocator{mapping.big_page ? *vm.big_page_allocator : *vm.small_page_allocator};
+        u32 page_size_bits{mapping.big_page ? vm.big_page_size_bits : VM::PAGE_SIZE_BITS};
+        allocator.Free(u32(mapping.offset >> page_size_bits), u32(mapping.size >> page_size_bits));
+    }
+
+    // Sparse mappings shouldn't be fully unmapped, just returned to their sparse state
+    // Only FreeSpace can unmap them fully
+    if (mapping.sparse_alloc) {
+        gmmu->MapSparse(params.offset, mapping.size, mapping.big_page);
+    } else {
+        gmmu->Unmap(params.offset, mapping.size);
+    }
+
+    nvmap.UnpinHandle(mapping.handle);
+    mapping_map.erase(it);
+
     return NvResult::Success;
 }
 

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h

@@ -13,7 +13,6 @@
 #include <memory>
 #include <mutex>
 #include <optional>
-#include <ankerl/unordered_dense.h>
 #include <vector>
 
 #include "common/address_space.h"
@@ -113,8 +112,6 @@ private:
     };
     static_assert(sizeof(IoctlRemapEntry) == 20, "IoctlRemapEntry is incorrect size");
 
-    ankerl::unordered_dense::set<s64_le> map_buffer_offsets{};
-
     struct IoctlMapBufferEx {
         MappingFlags flags{}; // bit0: fixed_offset, bit2: cacheable
         u32_le kind{};        // -1 is default

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp

@@ -4,6 +4,7 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <algorithm>
 #include <cstring>
 #include "common/assert.h"
 #include "common/logging.h"
@@ -264,7 +265,7 @@ NvResult nvhost_ctrl_gpu::ZCullGetInfo(IoctlNvgpuGpuZcullGetInfoArgs& params) {
 }
 
 NvResult nvhost_ctrl_gpu::ZBCSetTable(IoctlZbcSetTable& params) {
-    if (params.type > supported_types) {
+    if (params.type == 0 || params.type > supported_types) {
         LOG_ERROR(Service_NVDRV, "ZBCSetTable: invalid type {:#X}", params.type);
         return NvResult::BadParameter;
     }
@@ -279,42 +280,61 @@ NvResult nvhost_ctrl_gpu::ZBCSetTable(IoctlZbcSetTable& params) {
             color_entry.format = params.format;
             color_entry.ref_cnt = 1u;
 
-            auto color_it = std::ranges::find_if(zbc_colors,
-                                                 [&](const ZbcColorEntry& color_in_question) {
-                                                     return color_entry.format == color_in_question.format &&
-                                                            color_entry.color_ds == color_in_question.color_ds &&
-                                                            color_entry.color_l2 == color_in_question.color_l2;
-                                                 });
+            const auto color_end = zbc_colors.begin() + zbc_used_color_entries;
+            auto color_it = std::find_if(zbc_colors.begin(), color_end,
+                                         [&](const ZbcColorEntry& color_in_question) {
+                                             return color_entry.format == color_in_question.format &&
+                                                    color_entry.color_ds == color_in_question.color_ds &&
+                                                    color_entry.color_l2 == color_in_question.color_l2;
+                                         });
 
-            if (color_it != zbc_colors.end()) {
+            if (color_it != color_end) {
                 ++color_it->ref_cnt;
                 LOG_DEBUG(Service_NVDRV, "ZBCSetTable: reused color entry fmt={:#X}, ref_cnt={:#X}",
                           params.format, color_it->ref_cnt);
-            } else {
-                zbc_colors.push_back(color_entry);
-                LOG_DEBUG(Service_NVDRV, "ZBCSetTable: added color entry fmt={:#X}, index={:#X}",
-                          params.format, zbc_colors.size() - 1);
+                break;
             }
+
+            if (zbc_used_color_entries >= zbc_table_size) {
+                LOG_WARNING(Service_NVDRV, "ZBCSetTable: color table is full, fmt={:#X}",
+                            params.format);
+                return NvResult::InsufficientMemory;
+            }
+
+            zbc_colors[zbc_used_color_entries] = color_entry;
+            LOG_DEBUG(Service_NVDRV, "ZBCSetTable: added color entry fmt={:#X}, index={:#X}",
+                      params.format, zbc_used_color_entries);
+            ++zbc_used_color_entries;
             break;
         }
         case ZBCTypes::depth: {
             ZbcDepthEntry depth_entry{params.depth, params.format, 1u};
 
-            auto depth_it = std::ranges::find_if(zbc_depths,
-                                                 [&](const ZbcDepthEntry& depth_entry_in_question) {
-                                                     return depth_entry.format == depth_entry_in_question.format &&
-                                                            depth_entry.depth == depth_entry_in_question.depth;
-                                                 });
+            const auto depth_end = zbc_depths.begin() + zbc_used_depth_entries;
+            auto depth_it = std::find_if(zbc_depths.begin(), depth_end,
+                                         [&](const ZbcDepthEntry& depth_entry_in_question) {
+                                             return depth_entry.format == depth_entry_in_question.format &&
+                                                    depth_entry.depth == depth_entry_in_question.depth;
+                                         });
 
-            if (depth_it != zbc_depths.end()) {
+            if (depth_it != depth_end) {
                 ++depth_it->ref_cnt;
                 LOG_DEBUG(Service_NVDRV, "ZBCSetTable: reused depth entry fmt={:#X}, ref_cnt={:#X}",
                           depth_entry.format, depth_it->ref_cnt);
-            } else {
-                zbc_depths.push_back(depth_entry);
-                LOG_DEBUG(Service_NVDRV, "ZBCSetTable: added depth entry fmt={:#X}, index={:#X}",
-                          depth_entry.format, zbc_depths.size() - 1);
+                break;
             }
+
+            if (zbc_used_depth_entries >= zbc_table_size) {
+                LOG_WARNING(Service_NVDRV, "ZBCSetTable: depth table is full, fmt={:#X}",
+                            depth_entry.format);
+                return NvResult::InsufficientMemory;
+            }
+
+            zbc_depths[zbc_used_depth_entries] = depth_entry;
+            LOG_DEBUG(Service_NVDRV, "ZBCSetTable: added depth entry fmt={:#X}, index={:#X}",
+                      depth_entry.format, zbc_used_depth_entries);
+            ++zbc_used_depth_entries;
+            break;
         }
     }
 
@@ -329,35 +349,34 @@ NvResult nvhost_ctrl_gpu::ZBCQueryTable(IoctlZbcQueryTable& params) {
 
     std::scoped_lock lk(zbc_mutex);
 
+    if (params.type == 0) {
+        params.index_size = zbc_table_size;
+        return NvResult::Success;
+    }
+
+    if (params.index_size >= zbc_table_size) {
+        LOG_ERROR(Service_NVDRV, "ZBCQueryTable: invalid index {:#X}", params.index_size);
+        return NvResult::BadParameter;
+    }
+
     switch (static_cast<ZBCTypes>(params.type)) {
         case ZBCTypes::color: {
-            if (params.index_size >= zbc_colors.size()) {
-                LOG_ERROR(Service_NVDRV, "ZBCQueryTable: invalid color index {:#X}", params.index_size);
-                return NvResult::BadParameter;
-            }
-
             const auto& colors = zbc_colors[params.index_size];
             std::copy_n(colors.color_ds.begin(), colors.color_ds.size(), std::begin(params.color_ds));
             std::copy_n(colors.color_l2.begin(), colors.color_l2.size(), std::begin(params.color_l2));
             params.depth = 0;
             params.ref_cnt = colors.ref_cnt;
             params.format = colors.format;
-            params.index_size = static_cast<u32>(zbc_colors.size());
             break;
         }
         case ZBCTypes::depth: {
-            if (params.index_size >= zbc_depths.size()) {
-                LOG_ERROR(Service_NVDRV, "ZBCQueryTable: invalid depth index {:#X}", params.index_size);
-                return NvResult::BadParameter;
-            }
-
             const auto& depth_entry = zbc_depths[params.index_size];
             std::fill(std::begin(params.color_ds), std::end(params.color_ds), 0);
             std::fill(std::begin(params.color_l2), std::end(params.color_l2), 0);
             params.depth = depth_entry.depth;
             params.ref_cnt = depth_entry.ref_cnt;
             params.format = depth_entry.format;
-            params.index_size = static_cast<u32>(zbc_depths.size());
+            break;
         }
     }
 

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h

@@ -6,7 +6,7 @@
 
 #pragma once
 
-#include <vector>
+#include <array>
 
 #include "common/common_funcs.h"
 #include "common/common_types.h"
@@ -212,9 +212,13 @@ private:
     Kernel::KEvent* unknown_event;
 
     // ZBC Tables
+    static constexpr u32 zbc_table_size = 15u;
+
     std::mutex zbc_mutex{};
-    std::vector<ZbcColorEntry> zbc_colors{};
-    std::vector<ZbcDepthEntry> zbc_depths{};
+    std::array<ZbcColorEntry, zbc_table_size> zbc_colors{};
+    std::array<ZbcDepthEntry, zbc_table_size> zbc_depths{};
+    u32 zbc_used_color_entries{};
+    u32 zbc_used_depth_entries{};
     const u32 supported_types = 2u;
 };
 

src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp

@@ -174,7 +174,9 @@ NvResult nvhost_gpu::SetChannelPriority(IoctlChannelSetPriority& params) {
     case ChannelPriority::Low: channel_timeslice = 1300; break;
     case ChannelPriority::Medium: channel_timeslice = 2600; break;
     case ChannelPriority::High: channel_timeslice = 5200; break;
-    default : return NvResult::BadParameter;
+    default:
+        LOG_WARNING(Service_NVDRV, "unknown channel priority {:#X}", channel_priority);
+        break;
     }
 
     return NvResult::Success;
@@ -278,18 +280,20 @@ NvResult nvhost_gpu::AllocateObjectContext(IoctlAllocObjCtx& params) {
         params.flags = allowed_mask;
     }
 
-    s32_le ctx_class_number_index = 
+    params.obj_id = 0;
+
+    s32_le ctx_class_number_index =
         GetObjectContextClassNumberIndex(static_cast<CtxClasses>(params.class_num));
     if (ctx_class_number_index < 0) {
-        LOG_ERROR(Service_NVDRV, "Invalid class number for object context: {:#X}",
-                  params.class_num);
-        return NvResult::BadParameter;
+        LOG_WARNING(Service_NVDRV, "Untracked class number for object context: {:#X}",
+                    params.class_num);
+        return NvResult::Success;
     }
 
     if (ctxObjs[ctx_class_number_index].has_value()) {
-        LOG_WARNING(Service_NVDRV, "Object context for class {:#X} already allocated on this channel",
-                    params.class_num);
-        return NvResult::AlreadyAllocated;
+        LOG_DEBUG(Service_NVDRV, "Object context for class {:#X} already allocated on this channel",
+                  params.class_num);
+        return NvResult::Success;
     }
 
     // Defer actual hardware context binding until channel is initialized.
@@ -435,10 +439,6 @@ NvResult nvhost_gpu::ChannelSetTimeout(IoctlChannelSetTimeout& params) {
 NvResult nvhost_gpu::ChannelSetTimeslice(IoctlSetTimeslice& params) {
     LOG_INFO(Service_NVDRV, "called, timeslice={:#X}", params.timeslice);
 
-    if (params.timeslice < 1000 || params.timeslice > 5000) {
-        return NvResult::BadParameter;
-    }
-
     channel_timeslice = params.timeslice;
 
     return NvResult::Success;

src/core/hle/service/nvnflinger/buffer_queue_core.cpp

@@ -20,33 +20,23 @@ BufferQueueCore::~BufferQueueCore() = default;
 void BufferQueueCore::PushHistory(u64 frame_number, s64 queue_time, s64 presentation_time, BufferState state) {
     std::lock_guard lk(buffer_history_mutex);
 
-    auto it = buffer_history_map.find(frame_number);
-    if (it != buffer_history_map.end()) {
-        it->second.state = state;
-        return;
-    }
-
-    buffer_history_map.emplace(frame_number, BufferHistoryInfo{
+    buffer_history_pos = (buffer_history_pos + 1) % BUFFER_HISTORY_SIZE;
+    buffer_history[buffer_history_pos] = BufferHistoryInfo{
         frame_number,
         queue_time,
         presentation_time,
         state
-    });
-    buffer_history_order.push_back(frame_number);
-
-    if (buffer_history_order.size() > BUFFER_HISTORY_SIZE) {
-        u64 oldest_frame = buffer_history_order.front();
-        buffer_history_order.pop_front();
-        buffer_history_map.erase(oldest_frame);
-    }
+    };
 }
 
 void BufferQueueCore::UpdateHistory(u64 frame_number, BufferState state) {
     std::lock_guard lk(buffer_history_mutex);
 
-    auto it = buffer_history_map.find(frame_number);
-    if (it != buffer_history_map.end()) {
-        it->second.state = state;
+    for (auto& entry : buffer_history) {
+        if (entry.frame_number == frame_number) {
+            entry.state = state;
+            return;
+        }
     }
 }
 

src/core/hle/service/nvnflinger/buffer_queue_core.h

@@ -9,14 +9,13 @@
 
 #pragma once
 
+#include <array>
 #include <condition_variable>
-#include <deque>
 #include <list>
 #include <memory>
 #include <mutex>
 #include <set>
 #include <vector>
-#include <unordered_map>
 #include <algorithm>
 
 #include "core/hle/service/nvnflinger/buffer_item.h"
@@ -28,12 +27,15 @@
 
 namespace Service::android {
 
+#pragma pack(push, 1)
 struct BufferHistoryInfo {
-    u64 frame_number{};
-    s64 queue_time{};
-    s64 presentation_time{};
-    BufferState state{};
+    u64 frame_number;
+    s64 queue_time;
+    s64 presentation_time;
+    BufferState state;
 };
+#pragma pack(pop)
+static_assert(sizeof(BufferHistoryInfo) == 0x1C, "BufferHistoryInfo must be 28 bytes");
 
 class IConsumerListener;
 class IProducerListener;
@@ -88,9 +90,9 @@ private:
     bool buffer_has_been_queued{};
     u64 frame_counter{};
 
-    std::unordered_map<u64, BufferHistoryInfo> buffer_history_map{};
+    std::array<BufferHistoryInfo, BUFFER_HISTORY_SIZE> buffer_history{};
+    u32 buffer_history_pos{BUFFER_HISTORY_SIZE - 1};
     mutable std::mutex buffer_history_mutex{};
-    std::deque<u64> buffer_history_order;
 
     u32 transform_hint{};
     bool is_allocating{};

src/core/hle/service/nvnflinger/buffer_queue_producer.cpp

@@ -507,6 +507,8 @@ Status BufferQueueProducer::QueueBuffer(s32 slot, const QueueBufferInput& input,
 
         sticky_transform = sticky_transform_;
 
+        const bool track_history = Settings::values.enable_buffer_history.GetValue();
+
         if (core->queue.empty()) {
             core->queue.push_back(item);
             listener_available = core->consumer_listener;
@@ -514,7 +516,7 @@ Status BufferQueueProducer::QueueBuffer(s32 slot, const QueueBufferInput& input,
             auto front = core->queue.begin();
             if (front->is_droppable && core->StillTracking(*front)) {
                 slots[front->slot].buffer_state = BufferState::Free;
-                if (Settings::values.enable_buffer_history.GetValue()) {
+                if (track_history) {
                     core->UpdateHistory(front->frame_number, BufferState::Free);
                 }
                 slots[front->slot].frame_number = 0;
@@ -529,7 +531,7 @@ Status BufferQueueProducer::QueueBuffer(s32 slot, const QueueBufferInput& input,
             }
         }
 
-        if (Settings::values.enable_buffer_history.GetValue()) {
+        if (track_history) {
             core->PushHistory(core->frame_counter, slots[slot].queue_time, slots[slot].presentation_time, BufferState::Queued);
         }
 
@@ -902,26 +904,31 @@ void BufferQueueProducer::Transact(u32 code, std::span<const u8> parcel_data,
 
         const s32 request = parcel_in.Read<s32>();
         if (request <= 0) {
-            parcel_out.Write(Status::BadValue);
+            status = Status::BadValue;
             parcel_out.Write<s32>(0);
             break;
         }
 
-        std::vector<BufferHistoryInfo> snapshot;
+        constexpr u32 history_size = BufferQueueCore::BUFFER_HISTORY_SIZE;
+        std::array<BufferHistoryInfo, history_size> snapshot{};
+        s32 count{};
 
         {
             std::scoped_lock lk(core->buffer_history_mutex);
-            for (auto& [frame, info] : core->buffer_history_map) {
-                snapshot.push_back(info);
+
+            const u32 newest = core->buffer_history_pos;
+            for (u32 i = 0; i < history_size; ++i) {
+                const auto& entry = core->buffer_history[(newest + history_size - i) % history_size];
+                if (entry.frame_number == 0) {
+                    break;
+                }
+
+                snapshot[count] = entry;
+                ++count;
             }
         }
 
-        std::sort(snapshot.begin(), snapshot.end(), [](auto& a, auto& b){
-            return a.frame_number > b.frame_number;
-        });
-
-        const s32 limit = std::min(request, (s32)snapshot.size());
-        parcel_out.Write(Status::NoError);
+        const s32 limit = (std::min)(request, count);
         parcel_out.Write<s32>(limit);
         for (s32 i = 0; i < limit; ++i) {
             parcel_out.Write(snapshot[i]);

src/core/hle/service/spl/spl.cpp

@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -31,9 +28,9 @@ SPL_MIG::SPL_MIG(Core::System& system_, std::shared_ptr<Module> module_)
     static const FunctionInfo functions[] = {
         {0, &SPL::GetConfig, "GetConfig"},
         {1, &SPL::ModularExponentiate, "ModularExponentiate"},
-        {2, &SPL::GenerateAesKek, "GenerateAesKek"},
+        {2, nullptr, "GenerateAesKek"},
         {3, nullptr, "LoadAesKey"},
-        {4, &SPL::GenerateAesKey, "GenerateAesKey"},
+        {4, nullptr, "GenerateAesKey"},
         {5, &SPL::SetConfig, "SetConfig"},
         {7, &SPL::GenerateRandomBytes, "GenerateRandomBytes"},
         {11, &SPL::IsDevelopment, "IsDevelopment"},

src/core/hle/service/spl/spl_module.cpp

@@ -59,36 +59,6 @@ void Module::Interface::ModularExponentiate(HLERequestContext& ctx) {
     rb.Push(ResultSecureMonitorNotImplemented);
 }
 
-void Module::Interface::GenerateAesKek(HLERequestContext& ctx) {
-    IPC::RequestParser rp{ctx};
-    [[maybe_unused]] const auto key_source = rp.PopRaw<KeySource>();
-    const auto generation = rp.Pop<u32>();
-    const auto option = rp.Pop<u32>();
-
-    LOG_WARNING(Service_SPL, "(STUBBED) called, generation={:#x}, option={:#x}", generation,
-                option);
-
-    AccessKey access_key{};
-
-    IPC::ResponseBuilder rb{ctx, 6};
-    rb.Push(ResultSuccess);
-    rb.PushRaw(access_key);
-}
-
-void Module::Interface::GenerateAesKey(HLERequestContext& ctx) {
-    IPC::RequestParser rp{ctx};
-    [[maybe_unused]] const auto access_key = rp.PopRaw<AccessKey>();
-    [[maybe_unused]] const auto key_source = rp.PopRaw<KeySource>();
-
-    LOG_WARNING(Service_SPL, "(STUBBED) called");
-
-    AesKey aes_key{};
-
-    IPC::ResponseBuilder rb{ctx, 6};
-    rb.Push(ResultSuccess);
-    rb.PushRaw(aes_key);
-}
-
 void Module::Interface::SetConfig(HLERequestContext& ctx) {
     UNIMPLEMENTED_MSG("SetConfig is not implemented!");
 

src/core/hle/service/spl/spl_module.h

@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -28,8 +25,6 @@ public:
         // General
         void GetConfig(HLERequestContext& ctx);
         void ModularExponentiate(HLERequestContext& ctx);
-        void GenerateAesKek(HLERequestContext& ctx);
-        void GenerateAesKey(HLERequestContext& ctx);
         void SetConfig(HLERequestContext& ctx);
         void GenerateRandomBytes(HLERequestContext& ctx);
         void IsDevelopment(HLERequestContext& ctx);

src/core/hle/service/vi/conductor.cpp

@@ -5,6 +5,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include "common/settings.h"
+#include "common/thread.h"
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "core/hle/service/vi/conductor.h"
@@ -76,6 +77,7 @@ void Conductor::ProcessVsync() {
 
 void Conductor::VsyncThread(std::stop_token token) {
     Common::SetCurrentThreadName("VSyncThread");
+    Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
 
     while (!token.stop_requested()) {
         m_signal.Wait();

src/qt_common/config/shared_translation.cpp

@@ -230,6 +230,11 @@ std::unique_ptr<TranslationMap> InitializeTranslations(QObject* parent) {
            tr("Preserves GPU-modified data by reading it back before uploading.\nSome games require this to render certain effects properly."));
     INSERT(Settings, use_asynchronous_shaders, tr("Enable asynchronous shader compilation"),
            tr("May reduce shader stutter."));
+    INSERT(Settings, use_unified_memory, tr("Enable unified memory access (UMA)"),
+           tr("Lets the GPU write buffer readbacks directly into guest memory."));
+    INSERT(Settings, pipeline_worker_count, tr("Pipeline Worker Threads"),
+           tr("Number of threads used to build Vulkan pipelines.\n"
+              "Higher values speed up compilation at the cost of heat and power."));
     INSERT(Settings, fast_gpu_time, tr("Fast GPU Time"),
            tr("Overclocks the emulated GPU to increase dynamic resolution and render "
               "distance.\nUse 256 for maximal performance and 512 for maximal graphics fidelity."));
@@ -287,6 +292,12 @@ std::unique_ptr<TranslationMap> InitializeTranslations(QObject* parent) {
     INSERT(Settings, vertex_input_dynamic_state, tr("Vertex Input Dynamic State"),
            tr("Enables vertex input dynamic state feature for better quality and performance."));
 
+    INSERT(Settings, dynamic_rendering, tr("Dynamic Rendering"),
+           tr("Renders without render pass and framebuffer objects.\n"
+              "Results vary by driver: some gain performance, others lose it."));
+
+    INSERT(Settings, workgroup_memory_explicit_layout, QString(), QString());
+
     INSERT(
         Settings, sample_shading, tr("Sample Shading"),
         tr("Allows the fragment shader to execute per sample in a multi-sampled fragment "

src/shader_recompiler/backend/glasm/emit_glasm_instructions.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -665,6 +665,8 @@ void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU3
                      const IR::Value& clamp, const IR::Value& segmentation_mask);
 void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 index,
                           const IR::Value& clamp, const IR::Value& segmentation_mask);
+void EmitQuadBroadcast(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 lane);
+void EmitQuadSwap(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 direction);
 void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, ScalarF32 op_a, ScalarF32 op_b,
                      ScalarU32 swizzle);
 void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, ScalarF32 op_a);

src/shader_recompiler/backend/glasm/emit_glasm_warp.cpp

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -97,6 +100,24 @@ void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, Sca
     Shuffle(ctx, inst, value, index, clamp, segmentation_mask, "XOR");
 }
 
+void EmitQuadBroadcast(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 lane) {
+    const Register ret{ctx.reg_alloc.Define(inst)};
+    ctx.Add("AND.U RC.x,{}.threadid,~3;"
+            "AND.U RC.y,{},3;"
+            "OR.U RC.x,RC.x,RC.y;"
+            "SHFIDX.U {},{},RC.x,0x1C03;"
+            "MOV.U {}.x,{}.y;",
+            ctx.stage_name, lane, ret, value, ret, ret);
+}
+
+void EmitQuadSwap(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 direction) {
+    const Register ret{ctx.reg_alloc.Define(inst)};
+    ctx.Add("ADD.U RC.x,{},1;"
+            "SHFXOR.U {},{},RC.x,0x1C03;"
+            "MOV.U {}.x,{}.y;",
+            direction, ret, value, ret, ret);
+}
+
 void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, ScalarF32 op_a, ScalarF32 op_b,
                      ScalarU32 swizzle) {
     const auto ret{ctx.reg_alloc.Define(inst)};

src/shader_recompiler/backend/glsl/emit_glsl_instructions.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -743,6 +743,10 @@ void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
 void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
                           std::string_view index, std::string_view clamp,
                           std::string_view segmentation_mask);
+void EmitQuadBroadcast(EmitContext& ctx, IR::Inst& inst, std::string_view value,
+                       std::string_view lane);
+void EmitQuadSwap(EmitContext& ctx, IR::Inst& inst, std::string_view value,
+                  std::string_view direction);
 void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
                      std::string_view swizzle);
 void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);

src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -200,6 +203,18 @@ void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view val
     ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value);
 }
 
+void EmitQuadBroadcast(EmitContext& ctx, IR::Inst& inst, std::string_view value,
+                       std::string_view lane) {
+    const auto src_thread_id{fmt::format("(({}&~3)|({}& 3))", THREAD_ID, lane)};
+    ctx.AddU32("{}=readInvocationARB({},{});", inst, value, src_thread_id);
+}
+
+void EmitQuadSwap(EmitContext& ctx, IR::Inst& inst, std::string_view value,
+                  std::string_view direction) {
+    const auto src_thread_id{fmt::format("({}^({}+1))", THREAD_ID, direction)};
+    ctx.AddU32("{}=readInvocationARB({},{});", inst, value, src_thread_id);
+}
+
 void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
                      std::string_view swizzle) {
     const auto mask{fmt::format("({}>>((gl_SubGroupInvocationARB&3)<<1))&3", swizzle)};

src/shader_recompiler/backend/spirv/emit_spirv.cpp

@@ -322,6 +322,11 @@ void DefineEntryPoint(const IR::Program& program, EmitContext& ctx, Id main) {
         if (ctx.runtime_info.force_early_z) {
             ctx.AddExecutionMode(main, spv::ExecutionMode::EarlyFragmentTests);
         }
+        if (ctx.profile.support_shader_quad_control && program.info.uses_quad_shuffles) {
+            ctx.AddExtension("SPV_KHR_quad_control");
+            ctx.AddCapability(spv::Capability::QuadControlKHR);
+            ctx.AddExecutionMode(main, spv::ExecutionMode::RequireFullQuadsKHR);
+        }
         break;
     default:
         throw NotImplementedException("Stage {}", program.stage);
@@ -443,6 +448,12 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
             ctx.AddCapability(spv::Capability::GroupNonUniformVote);
         }
     }
+    if (info.uses_quad_shuffles) {
+        if (profile.support_quad_shuffles) {
+            ctx.AddCapability(spv::Capability::GroupNonUniformQuad);
+        }
+        ctx.AddCapability(spv::Capability::GroupNonUniformShuffle);
+    }
     if (info.uses_int64_bit_atomics && profile.support_int64_atomics) {
         ctx.AddCapability(spv::Capability::Int64Atomics);
     }

src/shader_recompiler/backend/spirv/emit_spirv_instructions.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -622,6 +622,8 @@ Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clam
                    Id segmentation_mask);
 Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                         Id segmentation_mask);
+Id EmitQuadBroadcast(EmitContext& ctx, Id value, Id lane);
+Id EmitQuadSwap(EmitContext& ctx, Id value, Id direction);
 Id EmitFSwizzleAdd(EmitContext& ctx, Id op_a, Id op_b, Id swizzle);
 Id EmitDPdxFine(EmitContext& ctx, Id op_a);
 Id EmitDPdyFine(EmitContext& ctx, Id op_a);

src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp

@@ -260,6 +260,21 @@ Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id
     return SelectValue(ctx, in_range, value, src_thread_id);
 }
 
+Id EmitQuadBroadcast(EmitContext& ctx, Id value, Id lane) {
+    if (ctx.profile.support_quad_shuffles) {
+        return ctx.OpGroupNonUniformQuadBroadcast(ctx.U32[1], SubgroupScope(ctx), value, lane);
+    }
+    const Id base{ctx.OpBitwiseAnd(ctx.U32[1], GetThreadId(ctx), ctx.Const(~3u))};
+    const Id local_lane{ctx.OpBitwiseAnd(ctx.U32[1], lane, ctx.Const(3u))};
+    const Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], base, local_lane)};
+    return ctx.OpGroupNonUniformShuffle(ctx.U32[1], SubgroupScope(ctx), value, src_thread_id);
+}
+
+Id EmitQuadSwap(EmitContext& ctx, Id value, Id direction) {
+    const Id xor_mask{ctx.OpIAdd(ctx.U32[1], direction, ctx.Const(1u))};
+    return ctx.OpGroupNonUniformShuffleXor(ctx.U32[1], SubgroupScope(ctx), value, xor_mask);
+}
+
 Id EmitFSwizzleAdd(EmitContext& ctx, Id op_a, Id op_b, Id swizzle) {
     const Id three{ctx.Const(3U)};
     Id mask{GetThreadId(ctx)};

src/shader_recompiler/frontend/ir/ir_emitter.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -2100,6 +2100,14 @@ U32 IREmitter::ShuffleButterfly(const IR::U32& value, const IR::U32& index, cons
     return Inst<U32>(Opcode::ShuffleButterfly, value, index, clamp, seg_mask);
 }
 
+U32 IREmitter::QuadBroadcast(const IR::U32& value, const IR::U32& lane) {
+    return Inst<U32>(Opcode::QuadBroadcast, value, lane);
+}
+
+U32 IREmitter::QuadSwap(const IR::U32& value, const IR::U32& direction) {
+    return Inst<U32>(Opcode::QuadSwap, value, direction);
+}
+
 F32 IREmitter::FSwizzleAdd(const F32& a, const F32& b, const U32& swizzle, FpControl control) {
     return Inst<F32>(Opcode::FSwizzleAdd, Flags{control}, a, b, swizzle);
 }

src/shader_recompiler/frontend/ir/ir_emitter.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -394,6 +394,8 @@ public:
                                   const IR::U32& seg_mask);
     [[nodiscard]] U32 ShuffleButterfly(const IR::U32& value, const IR::U32& index,
                                        const IR::U32& clamp, const IR::U32& seg_mask);
+    [[nodiscard]] U32 QuadBroadcast(const IR::U32& value, const IR::U32& lane);
+    [[nodiscard]] U32 QuadSwap(const IR::U32& value, const IR::U32& direction);
     [[nodiscard]] F32 FSwizzleAdd(const F32& a, const F32& b, const U32& swizzle,
                                   FpControl control = {});
 

src/shader_recompiler/frontend/ir/opcodes.cpp

@@ -10,7 +10,7 @@ namespace Shader::IR {
 
 namespace Detail {
 
-OpcodeMeta META_TABLE[532] = {
+OpcodeMeta META_TABLE[] = {
 #define OPCODE(name_token, type_token, ...)                                                        \
     {                                                                                              \
         .name{#name_token},                                                                        \
@@ -21,7 +21,7 @@ OpcodeMeta META_TABLE[532] = {
 #undef OPCODE
 };
 
-u8 NUM_ARGS[532] = {
+u8 NUM_ARGS[] = {
 #define OPCODE(name_token, type_token, ...) u8(CalculateNumArgsOf(Opcode::name_token)),
 #include "opcodes.inc"
 #undef OPCODE

src/shader_recompiler/frontend/ir/opcodes.h

@@ -57,12 +57,12 @@ static constexpr Type F64x2{Type::F64x2};
 static constexpr Type F64x3{Type::F64x3};
 static constexpr Type F64x4{Type::F64x4};
 
-extern OpcodeMeta META_TABLE[532];
+extern OpcodeMeta META_TABLE[];
 constexpr size_t CalculateNumArgsOf(Opcode op) noexcept {
     const auto& arg_types = META_TABLE[size_t(op)].arg_types;
     return size_t(std::distance(arg_types.begin(), std::ranges::find(arg_types, Type::Void)));
 }
-extern u8 NUM_ARGS[532];
+extern u8 NUM_ARGS[];
 } // namespace Detail
 
 /// Get return type of an opcode

src/shader_recompiler/frontend/ir/opcodes.inc

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -579,6 +582,8 @@ OPCODE(ShuffleIndex,                                        U32,            U32,
 OPCODE(ShuffleUp,                                           U32,            U32,            U32,            U32,            U32,                            )
 OPCODE(ShuffleDown,                                         U32,            U32,            U32,            U32,            U32,                            )
 OPCODE(ShuffleButterfly,                                    U32,            U32,            U32,            U32,            U32,                            )
+OPCODE(QuadBroadcast,                                       U32,            U32,            U32,                                                                    )
+OPCODE(QuadSwap,                                            U32,            U32,            U32,                                                                    )
 OPCODE(FSwizzleAdd,                                         F32,            F32,            F32,            U32,                                            )
 OPCODE(DPdxFine,                                            F32,            F32,                                                                            )
 OPCODE(DPdyFine,                                            F32,            F32,                                                                            )

src/shader_recompiler/frontend/maxwell/translate/impl/internal_stage_buffer_entry_read.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -17,39 +17,12 @@ enum class Mode : u64 {
     Attr,
 };
 
-enum class SZ : u64 {
-    U8,
-    U16,
-    U32,
-    F32
-};
-
 enum class Shift : u64 {
     Default,
     U16,
     B32,
 };
 
-IR::U32 scaleIndex(IR::IREmitter& ir, IR::U32 index, Shift shift) {
-    switch (shift) {
-        case Shift::Default: return index;
-        case Shift::U16: return ir.ShiftLeftLogical(index, ir.Imm32(1));
-        case Shift::B32: return ir.ShiftLeftLogical(index, ir.Imm32(2));
-        default: UNREACHABLE();
-    }
-}
-
-IR::U32 skewBytes(IR::IREmitter& ir, SZ sizeRead) {
-    const IR::U32 lane = ir.LaneId();
-    switch (sizeRead) {
-        case SZ::U8:  return lane;
-        case SZ::U16: return ir.ShiftLeftLogical(lane, ir.Imm32(1));
-        case SZ::U32:
-        case SZ::F32: return ir.ShiftLeftLogical(lane, ir.Imm32(2));
-        default: UNREACHABLE();
-    }
-}
-
 } // Anonymous namespace
 
 void TranslatorVisitor::ISBERD(u64 insn) {
@@ -64,65 +37,28 @@ void TranslatorVisitor::ISBERD(u64 insn) {
         BitField<31, 1, u64> skew;
         BitField<32, 1, u64> o;
         BitField<33, 2, Mode> mode;
-        BitField<36, 4, SZ> sz;
         BitField<47, 2, Shift> shift;
     } const isberd{insn};
 
-    IR::U32 index{};
-    if (isberd.src_reg_num.Value() == 0xFF) {
-        index = ir.Imm32(isberd.imm.Value());
-    } else {
-        const IR::U32 scaledIndex = scaleIndex(ir, X(isberd.src_reg.Value()), isberd.shift.Value());
-        index = ir.IAdd(scaledIndex, ir.Imm32(isberd.imm.Value()));
+    if (isberd.skew != 0) {
+        throw NotImplementedException("ISBERD SKEW");
+    }
+    if (isberd.o != 0) {
+        throw NotImplementedException("ISBERD O");
     }
 
-    if (isberd.o.Value()) {
-        if (isberd.skew.Value()) {
-            index = ir.IAdd(index, skewBytes(ir, isberd.sz.Value()));
-        }
-
-        const IR::U64 index64 = ir.UConvert(64, index);
-        IR::U32 globalLoaded{};
-        switch (isberd.sz.Value()) {
-        case SZ::U8:  globalLoaded = ir.LoadGlobalU8 (index64); break;
-        case SZ::U16: globalLoaded = ir.LoadGlobalU16(index64); break;
-        case SZ::U32:
-        case SZ::F32: globalLoaded = ir.LoadGlobal32(index64);  break;
-        default: UNREACHABLE();
-        }
-        X(isberd.dest_reg.Value(), globalLoaded);
-
+    switch (isberd.mode.Value()) {
+    case Mode::Default:
+        X(isberd.dest_reg.Value(), X(isberd.src_reg.Value()));
         return;
-    }
-
-    if (isberd.mode.Value() != Mode::Default) {
-        if (isberd.skew.Value()) {
-            index = ir.IAdd(index, skewBytes(ir, SZ::U32));
-        }
-
-        IR::F32 float_index{};
-        switch (isberd.mode.Value()) {
-        case Mode::Patch: float_index = ir.GetPatch(index.Patch());
-            break;
-        case Mode::Prim:  float_index = ir.GetAttribute(index.Attribute());
-            break;
-        case Mode::Attr:  float_index = ir.GetAttributeIndexed(index);
-            break;
-        default: UNREACHABLE();
-        }
-        X(isberd.dest_reg.Value(), ir.BitCast<IR::U32>(float_index));
-
+    case Mode::Attr:
+        LOG_DEBUG(Shader, "(STUBBED) ISBERD Mode Attr");
+        X(isberd.dest_reg.Value(), X(isberd.src_reg.Value()));
         return;
+    default:
+        throw NotImplementedException("ISBERD Mode {}",
+                                      static_cast<u64>(isberd.mode.Value()));
     }
-
-    if (isberd.skew.Value()) {
-        X(isberd.dest_reg.Value(), ir.IAdd(X(isberd.src_reg.Value()), ir.LaneId()));
-
-        return;
-    }
-
-    // Fallback copy
-    X(isberd.dest_reg.Value(), X(isberd.src_reg.Value()));
 }
 
 } // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/warp_shuffle.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -36,7 +36,10 @@ enum class ShuffleMode : u64 {
     }
 }
 
-void Shuffle(TranslatorVisitor& v, u64 insn, const IR::U32& index, const IR::U32& mask) {
+constexpr u32 QUAD_MASK = (28u << 8) | 3u;
+
+void Shuffle(TranslatorVisitor& v, u64 insn, const IR::U32& index, const IR::U32& mask,
+             bool index_is_imm, u32 index_imm, bool mask_is_imm, u32 mask_imm) {
     union {
         u64 insn;
         BitField<0, 8, IR::Reg> dest_reg;
@@ -45,6 +48,21 @@ void Shuffle(TranslatorVisitor& v, u64 insn, const IR::U32& index, const IR::U32
         BitField<48, 3, IR::Pred> pred;
     } const shfl{insn};
 
+    const bool is_quad_candidate{mask_is_imm && mask_imm == QUAD_MASK && index_is_imm &&
+                                  v.env.ShaderStage() == Stage::Fragment};
+    if (is_quad_candidate) {
+        if (shfl.mode == ShuffleMode::IDX && index_imm <= 3) {
+            v.X(shfl.dest_reg, v.ir.QuadBroadcast(v.X(shfl.src_reg), v.ir.Imm32(index_imm)));
+            v.ir.SetPred(shfl.pred, v.ir.Imm1(true));
+            return;
+        }
+        if (shfl.mode == ShuffleMode::BFLY && index_imm >= 1 && index_imm <= 3) {
+            v.X(shfl.dest_reg, v.ir.QuadSwap(v.X(shfl.src_reg), v.ir.Imm32(index_imm - 1)));
+            v.ir.SetPred(shfl.pred, v.ir.Imm1(true));
+            return;
+        }
+    }
+
     const IR::U32 result{ShuffleOperation(v.ir, v.X(shfl.src_reg), index, mask, shfl.mode)};
     v.ir.SetPred(shfl.pred, v.ir.GetInBoundsFromOp(result));
     v.X(shfl.dest_reg, result);
@@ -59,11 +77,14 @@ void TranslatorVisitor::SHFL(u64 insn) {
         BitField<29, 1, u64> src_b_flag;
         BitField<34, 13, u64> src_b_imm;
     } const flags{insn};
-    const IR::U32 src_a{flags.src_a_flag != 0 ? ir.Imm32(static_cast<u32>(flags.src_a_imm))
-                                              : GetReg20(insn)};
-    const IR::U32 src_b{flags.src_b_flag != 0 ? ir.Imm32(static_cast<u32>(flags.src_b_imm))
-                                              : GetReg39(insn)};
-    Shuffle(*this, insn, src_a, src_b);
+    const bool index_is_imm{flags.src_a_flag != 0};
+    const bool mask_is_imm{flags.src_b_flag != 0};
+    const IR::U32 src_a{index_is_imm ? ir.Imm32(static_cast<u32>(flags.src_a_imm))
+                                     : GetReg20(insn)};
+    const IR::U32 src_b{mask_is_imm ? ir.Imm32(static_cast<u32>(flags.src_b_imm))
+                                    : GetReg39(insn)};
+    Shuffle(*this, insn, src_a, src_b, index_is_imm, static_cast<u32>(flags.src_a_imm),
+            mask_is_imm, static_cast<u32>(flags.src_b_imm));
 }
 
 } // namespace Shader::Maxwell

src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp

@@ -498,6 +498,10 @@ void VisitUsages(Info& info, IR::Inst& inst) {
     case IR::Opcode::ShuffleButterfly:
         info.uses_subgroup_shuffles = true;
         break;
+    case IR::Opcode::QuadBroadcast:
+    case IR::Opcode::QuadSwap:
+        info.uses_quad_shuffles = true;
+        break;
     case IR::Opcode::GetCbufU8:
     case IR::Opcode::GetCbufS8:
     case IR::Opcode::GetCbufU16:

src/shader_recompiler/profile.h

@@ -37,6 +37,8 @@ struct Profile {
     bool support_explicit_workgroup_layout{};
     bool support_workgroup_layout_8bit_access{};
     bool support_workgroup_layout_16bit_access{};
+    bool support_shader_quad_control{};
+    bool support_quad_shuffles{};
     bool support_vote{};
     u32 supported_subgroup_stages{0x7F};
     bool support_viewport_index_layer_non_geometry{};

src/shader_recompiler/shader_info.h

@@ -252,6 +252,7 @@ struct Info {
     bool uses_is_helper_invocation{};
     bool uses_subgroup_invocation_id{};
     bool uses_subgroup_shuffles{};
+    bool uses_quad_shuffles{};
     std::array<bool, 30> uses_patches{};
 
     std::array<Interpolation, 32> interpolation{};

src/video_core/CMakeLists.txt

@@ -33,7 +33,7 @@ add_library(video_core STATIC
     control/channel_state_cache.h
     control/scheduler.cpp
     control/scheduler.h
-    delayed_destruction_ring.h
+    deferred_destruction_queue.h
     dirty_flags.cpp
     dirty_flags.h
     dma_pusher.cpp
@@ -158,6 +158,8 @@ add_library(video_core STATIC
     renderer_vulkan/vk_compute_pass.h
     renderer_vulkan/vk_compute_pipeline.cpp
     renderer_vulkan/vk_compute_pipeline.h
+    renderer_vulkan/vk_descriptor_buffer.cpp
+    renderer_vulkan/vk_descriptor_buffer.h
     renderer_vulkan/vk_descriptor_pool.cpp
     renderer_vulkan/vk_descriptor_pool.h
     renderer_vulkan/vk_fence_manager.cpp

src/video_core/buffer_cache/buffer_cache.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #include <algorithm>
+#include <bit>
 #include <memory>
 #include <numeric>
 
@@ -31,44 +32,89 @@ BufferCache<P>::BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, R
     immediately_free = (Settings::values.vram_usage_mode.GetValue() == Settings::VramUsageMode::Aggressive);
 #endif
     if (!runtime.CanReportMemoryUsage()) {
-        minimum_memory = DEFAULT_EXPECTED_MEMORY;
-        critical_memory = DEFAULT_CRITICAL_MEMORY;
+        memory_budget = FALLBACK_MEMORY_BUDGET;
         return;
     }
 
-    const s64 device_local_memory = static_cast<s64>(runtime.GetDeviceLocalMemory());
-    const s64 min_spacing_expected = device_local_memory - 1_GiB;
-    const s64 min_spacing_critical = device_local_memory - 512_MiB;
-    const s64 mem_threshold = (std::min)(device_local_memory, TARGET_THRESHOLD);
-    const s64 min_vacancy_expected = (6 * mem_threshold) / 10;
-    const s64 min_vacancy_critical = (2 * mem_threshold) / 10;
-    minimum_memory = static_cast<u64>(
-        (std::max)((std::min)(device_local_memory - min_vacancy_expected, min_spacing_expected),
-                 DEFAULT_EXPECTED_MEMORY));
-    critical_memory = static_cast<u64>(
-        (std::max)((std::min)(device_local_memory - min_vacancy_critical, min_spacing_critical),
-                 DEFAULT_CRITICAL_MEMORY));
+    memory_budget = runtime.GetDeviceLocalMemory();
 }
 
 template <class P>
 BufferCache<P>::~BufferCache() = default;
 
 template <class P>
-void BufferCache<P>::RunGarbageCollector() {
-    const bool aggressive_gc = total_used_memory >= critical_memory;
-    const u64 ticks_to_destroy = aggressive_gc ? 60 : 120;
-    int num_iterations = aggressive_gc ? 64 : 32;
-    const auto clean_up = [this, &num_iterations](BufferId buffer_id) {
-        if (num_iterations == 0) {
+u64 BufferCache<P>::DeviceUsage(bool force_refresh) {
+    if (!runtime.CanReportMemoryUsage()) {
+        return total_used_memory;
+    }
+    if (force_refresh || usage_refresh_countdown == 0) {
+        cached_device_usage = runtime.GetDeviceAllocationUsage();
+        usage_refresh_countdown = USAGE_REFRESH_INTERVAL;
+    } else {
+        --usage_refresh_countdown;
+    }
+    return cached_device_usage;
+}
+
+template <class P>
+u64 BufferCache<P>::ReclaimMemory(u64 target_bytes, bool allow_download) {
+    if (target_bytes == 0 || in_reclaim) {
+        return 0;
+    }
+    in_reclaim = true;
+    u64 freed = 0;
+    const auto clean_up = [&](BufferId buffer_id) {
+        if (freed >= target_bytes) {
             return true;
         }
-        --num_iterations;
         auto& buffer = slot_buffers[buffer_id];
+        if (!allow_download && IsRegionGpuModified(buffer.CpuAddr(), buffer.SizeBytes())) {
+            return false;
+        }
+        const u64 buffer_bytes = Common::AlignUp(buffer.SizeBytes(), 1024);
         DownloadBufferMemory(buffer);
         DeleteBuffer(buffer_id);
+        freed += buffer_bytes;
         return false;
     };
-    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up);
+    const u64 cold_tick =
+        frame_tick > RECLAIM_GUARD_FRAMES ? frame_tick - RECLAIM_GUARD_FRAMES : 0;
+    lru_cache.ForEachItemBelow(cold_tick, clean_up);
+    if (freed == 0) {
+        lru_cache.ForEachItemBelow(frame_tick > 0 ? frame_tick - 1 : 0, clean_up);
+    }
+    in_reclaim = false;
+    usage_refresh_countdown = 0;
+    reclaim_stalled = freed == 0;
+    if (freed > 0) {
+        reclaim_wait_sync_point = runtime.CurrentSyncPoint();
+    }
+    return freed;
+}
+
+template <class P>
+void BufferCache<P>::ReclaimDeferredResources(u64 completed_sync_point) {
+    sentenced_buffers.Reclaim(completed_sync_point);
+}
+
+template <class P>
+void BufferCache<P>::EnsureHeadroom(bool allow_download) {
+    if (reclaim_stalled) {
+        return;
+    }
+    if (runtime.CompletedSyncPoint() < reclaim_wait_sync_point) {
+        return;
+    }
+    const u64 limit = memory_budget > RECLAIM_HEADROOM ? memory_budget - RECLAIM_HEADROOM : 0;
+    const u64 usage = DeviceUsage(false);
+    if (usage <= limit) {
+        return;
+    }
+    const u64 target = (limit / 100) * RECLAIM_TARGET_PERCENT;
+    const u64 excess = usage - target;
+    const u64 usage_mib = (std::max)(usage >> 20, u64{1});
+    const u64 share = (((excess >> 20) * (total_used_memory >> 20)) / usage_mib) << 20;
+    ReclaimMemory((std::min)(share, total_used_memory), allow_download);
 }
 
 template <class P>
@@ -96,15 +142,11 @@ void BufferCache<P>::TickFrame() {
     const bool skip_preferred = hits * 256 < shots * 251;
     channel_state->uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0;
 
-    // If we can obtain the memory info, use it instead of the estimate.
-    if (runtime.CanReportMemoryUsage()) {
-        total_used_memory = runtime.GetDeviceMemoryUsage();
-    }
-    if (total_used_memory >= minimum_memory) {
-        RunGarbageCollector();
-    }
+    usage_refresh_countdown = 0;
+    reclaim_stalled = false;
+    ReclaimDeferredResources(runtime.CompletedSyncPoint());
+    EnsureHeadroom(true);
     ++frame_tick;
-    delayed_destruction_ring.Tick();
 
     for (auto& buffer : async_buffers_death_ring) {
         runtime.FreeDeferredStagingBuffer(buffer);
@@ -175,9 +217,71 @@ std::optional<VideoCore::RasterizerDownloadArea> BufferCache<P>::GetFlushArea(DA
 
 template <class P>
 void BufferCache<P>::DownloadMemory(DAddr device_addr, u64 size) {
-    ForEachBufferInRange(device_addr, size, [&](BufferId, Buffer& buffer) {
-        DownloadBufferMemory(buffer, device_addr, size);
+    if constexpr (!USE_MEMORY_MAPS) {
+        std::scoped_lock lock{mutex};
+        ForEachBufferInRange(device_addr, size, [&](BufferId, Buffer& buffer) {
+            DownloadBufferMemory(buffer, device_addr, size);
+        });
+        return;
+    }
+
+    boost::container::small_vector<std::pair<BufferCopy, BufferId>, 8> downloads;
+    u64 total_size_bytes = 0;
+    u64 largest_copy = 0;
+
+    std::unique_lock lock{mutex};
+    ForEachBufferInRange(device_addr, size, [&](BufferId buffer_id, Buffer& buffer) {
+        memory_tracker.ForEachDownloadRangeAndClear(
+            device_addr, size, [&](u64 device_addr_out, u64 range_size) {
+                const DAddr buffer_addr = buffer.CpuAddr();
+                const auto add_download = [&](DAddr start, DAddr end) {
+                    const u64 new_offset = start - buffer_addr;
+                    const u64 new_size = end - start;
+                    downloads.push_back({
+                        BufferCopy{
+                            .src_offset = new_offset,
+                            .dst_offset = total_size_bytes,
+                            .size = new_size,
+                        },
+                        buffer_id,
+                    });
+                    constexpr u64 align = 64ULL;
+                    constexpr u64 mask = ~(align - 1ULL);
+                    total_size_bytes += (new_size + align - 1) & mask;
+                    largest_copy = (std::max)(largest_copy, new_size);
+                };
+                gpu_modified_ranges.ForEachInRange(device_addr_out, range_size, add_download);
+                ClearDownload(device_addr_out, range_size);
+                gpu_modified_ranges.Subtract(device_addr_out, range_size);
+            });
     });
+    if (total_size_bytes == 0) {
+        return;
+    }
+
+    auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
+    boost::container::small_vector<BufferCopy, 8> writebacks;
+    runtime.PreCopyBarrier();
+    for (auto& [copy, buffer_id] : downloads) {
+        copy.dst_offset += download_staging.offset;
+        Buffer& buffer = slot_buffers[buffer_id];
+        buffer.MarkUsage(copy.src_offset, copy.size);
+        const std::array copies{copy};
+        runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
+        BufferCopy writeback{copy};
+        writeback.src_offset = static_cast<u64>(buffer.CpuAddr()) + copy.src_offset;
+        writebacks.push_back(writeback);
+    }
+    runtime.PostCopyBarrier();
+    lock.unlock();
+
+    runtime.Finish();
+    const u8* const base = download_staging.mapped_span.data();
+    for (const BufferCopy& writeback : writebacks) {
+        const u64 staging_offset = writeback.dst_offset - download_staging.offset;
+        device_memory.WriteBlockUnsafe(static_cast<DAddr>(writeback.src_offset),
+                                       base + staging_offset, writeback.size);
+    }
 }
 
 template <class P>
@@ -214,7 +318,7 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
     auto& src_buffer = slot_buffers[buffer_a];
     auto& dest_buffer = slot_buffers[buffer_b];
     SynchronizeBuffer(src_buffer, *cpu_src_address, static_cast<u32>(amount));
-    SynchronizeBuffer(dest_buffer, *cpu_dest_address, static_cast<u32>(amount));
+    memory_tracker.UnmarkRegionAsCpuModified(*cpu_dest_address, static_cast<u32>(amount));
     std::array copies{BufferCopy{
         .src_offset = src_buffer.Offset(*cpu_src_address),
         .dst_offset = dest_buffer.Offset(*cpu_dest_address),
@@ -571,7 +675,11 @@ void BufferCache<P>::AccumulateFlushes() {
 
 template <class P>
 bool BufferCache<P>::ShouldWaitAsyncFlushes() const noexcept {
-    return (!async_buffers.empty() && async_buffers.front().has_value());
+    if (async_buffers.empty()) {
+        return false;
+    }
+    return async_buffers.front().has_value() ||
+           !pending_downloads.front().unified_copies.empty();
 }
 
 template <class P>
@@ -579,6 +687,7 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
     AccumulateFlushes();
 
     if (committed_gpu_modified_ranges.empty()) {
+        pending_downloads.emplace_back();
         async_buffers.emplace_back(std::optional<Async_Buffer>{});
         return;
     }
@@ -638,27 +747,83 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
     }
     committed_gpu_modified_ranges.clear();
     if (downloads.empty()) {
+        pending_downloads.emplace_back();
         async_buffers.emplace_back(std::optional<Async_Buffer>{});
         return;
     }
-    auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes, true);
-    boost::container::small_vector<BufferCopy, 4> normalized_copies;
-    runtime.PreCopyBarrier();
+
+    struct QueuedUnifiedCopy {
+        u64 window;
+        BufferId buffer_id;
+        boost::container::small_vector<BufferCopy, 16> copies;
+    };
+
+    AsyncDownloadBatch batch;
+    boost::container::small_vector<std::pair<BufferCopy, BufferId>, 16> staging_downloads;
+    boost::container::small_vector<QueuedUnifiedCopy, 4> unified_copy_queue;
+    boost::container::small_vector<u64, 4> window_ids;
+    UnifiedWindowGroups groups;
+    u64 staging_size_bytes = 0;
     for (auto& [copy, buffer_id] : downloads) {
-        copy.dst_offset += download_staging.offset;
-        const std::array copies{copy};
-        BufferCopy second_copy{copy};
         Buffer& buffer = slot_buffers[buffer_id];
-        second_copy.src_offset = static_cast<size_t>(buffer.CpuAddr()) + copy.src_offset;
-        const DAddr orig_device_addr = static_cast<DAddr>(second_copy.src_offset);
+        const DAddr orig_device_addr = buffer.CpuAddr() + copy.src_offset;
+        bool unified = false;
+        if constexpr (USE_UNIFIED_MEMORY) {
+            if (runtime.HasUnifiedMemory()) {
+                window_ids.clear();
+                groups.clear();
+                unified = ResolveUnifiedWindows(orig_device_addr, copy.src_offset, copy.size,
+                                                window_ids, groups);
+            }
+        }
+        BufferCopy record{copy};
+        record.src_offset = static_cast<size_t>(orig_device_addr);
+        if (unified) {
+            async_downloads.Add(orig_device_addr, copy.size);
+            buffer.MarkUsage(copy.src_offset, copy.size);
+            for (size_t i = 0; i < window_ids.size(); ++i) {
+                unified_copy_queue.push_back(
+                    QueuedUnifiedCopy{window_ids[i], buffer_id, std::move(groups[i])});
+            }
+            batch.unified_copies.push_back(record);
+            continue;
+        }
+        copy.dst_offset = staging_size_bytes;
+        constexpr u64 align = 64ULL;
+        staging_size_bytes += (copy.size + align - 1) & ~(align - 1ULL);
+        staging_downloads.push_back({copy, buffer_id});
+    }
+
+    std::optional<Async_Buffer> download_staging;
+    if (!staging_downloads.empty()) {
+        download_staging = runtime.DownloadStagingBuffer(staging_size_bytes, true);
+    }
+    runtime.PreCopyBarrier();
+    for (auto& [copy, buffer_id] : staging_downloads) {
+        copy.dst_offset += download_staging->offset;
+        const std::array copies{copy};
+        Buffer& buffer = slot_buffers[buffer_id];
+        BufferCopy record{copy};
+        record.src_offset = static_cast<size_t>(buffer.CpuAddr()) + copy.src_offset;
+        const DAddr orig_device_addr = static_cast<DAddr>(record.src_offset);
         async_downloads.Add(orig_device_addr, copy.size);
         buffer.MarkUsage(copy.src_offset, copy.size);
-        runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
-        normalized_copies.push_back(second_copy);
+        runtime.CopyBuffer(download_staging->buffer, buffer, copies, false);
+        batch.staging_copies.push_back(record);
+    }
+    if constexpr (USE_UNIFIED_MEMORY) {
+        for (const auto& queued : unified_copy_queue) {
+            const std::span<const BufferCopy> group_span(queued.copies.data(),
+                                                         queued.copies.size());
+            runtime.CopyToUnifiedMemory(queued.window, slot_buffers[queued.buffer_id], group_span);
+        }
+        if (!unified_copy_queue.empty()) {
+            runtime.UnifiedMemoryHostBarrier();
+        }
     }
     runtime.PostCopyBarrier();
-    pending_downloads.emplace_back(std::move(normalized_copies));
-    async_buffers.emplace_back(download_staging);
+    pending_downloads.emplace_back(std::move(batch));
+    async_buffers.emplace_back(std::move(download_staging));
 }
 
 template <class P>
@@ -673,32 +838,49 @@ void BufferCache<P>::PopAsyncFlushes() {
 
 template <class P>
 void BufferCache<P>::PopAsyncBuffers() {
-    if (async_buffers.empty()) {
-        return;
-    }
-    if (!async_buffers.front().has_value()) {
+    struct Writeback {
+        DAddr addr;
+        const u8* src;
+        u64 size;
+    };
+    boost::container::small_vector<Writeback, 8> writebacks;
+    {
+        std::scoped_lock lock{mutex};
+        if (async_buffers.empty()) {
+            return;
+        }
+        auto& batch = pending_downloads.front();
+        auto& async_buffer = async_buffers.front();
+        if (async_buffer.has_value()) {
+            const u8* base = async_buffer->mapped_span.data();
+            const size_t base_offset = async_buffer->offset;
+            for (const auto& copy : batch.staging_copies) {
+                const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
+                const u64 dst_offset = copy.dst_offset - base_offset;
+                const u8* read_mapped_memory = base + dst_offset;
+                async_downloads.ForEachInRange(
+                    device_addr, copy.size, [&](DAddr start, DAddr end, s32) {
+                        writebacks.push_back(
+                            {start, &read_mapped_memory[start - device_addr], end - start});
+                    });
+                async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
+                    gpu_modified_ranges.Subtract(start, end - start);
+                });
+            }
+            async_buffers_death_ring.emplace_back(*async_buffer);
+        }
+        for (const auto& copy : batch.unified_copies) {
+            const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
+            async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
+                gpu_modified_ranges.Subtract(start, end - start);
+            });
+        }
         async_buffers.pop_front();
-        return;
+        pending_downloads.pop_front();
     }
-    auto& downloads = pending_downloads.front();
-    auto& async_buffer = async_buffers.front();
-    u8* base = async_buffer->mapped_span.data();
-    const size_t base_offset = async_buffer->offset;
-    for (const auto& copy : downloads) {
-        const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
-        const u64 dst_offset = copy.dst_offset - base_offset;
-        const u8* read_mapped_memory = base + dst_offset;
-        async_downloads.ForEachInRange(device_addr, copy.size, [&](DAddr start, DAddr end, s32) {
-            device_memory.WriteBlockUnsafe(start, &read_mapped_memory[start - device_addr],
-                                           end - start);
-        });
-        async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
-            gpu_modified_ranges.Subtract(start, end - start);
-        });
+    for (const auto& wb : writebacks) {
+        device_memory.WriteBlockUnsafe(wb.addr, wb.src, wb.size);
     }
-    async_buffers_death_ring.emplace_back(*async_buffer);
-    async_buffers.pop_front();
-    pending_downloads.pop_front();
 }
 
 template <class P>
@@ -809,46 +991,46 @@ void BufferCache<P>::BindHostVertexBuffers() {
 
     if (use_optimized_vertex_buffers) {
         auto& flags = maxwell3d->dirty.flags;
-        u32 enabled_mask = enabled_vertex_buffers_mask;
-        HostBindings<Buffer> bindings{};
-        u32 last_index = (std::numeric_limits<u32>::max)();
-        const auto flush_bindings = [&]() {
-            if (bindings.buffers.empty()) {
-                return;
-            }
-            bindings.max_index = bindings.min_index + static_cast<u32>(bindings.buffers.size());
-            runtime.BindVertexBuffers(bindings);
-            bindings = HostBindings<Buffer>{};
-            last_index = (std::numeric_limits<u32>::max)();
-        };
-        while (enabled_mask != 0) {
-            const u32 index = std::countr_zero(enabled_mask);
-            enabled_mask &= (enabled_mask - 1);
+        const u32 enabled_mask = enabled_vertex_buffers_mask;
+        bool any_dirty = false;
+        u32 pending_mask = enabled_mask;
+        while (pending_mask != 0) {
+            const u32 index = std::countr_zero(pending_mask);
+            pending_mask &= (pending_mask - 1);
             const Binding& binding = VertexBufferSlot(index);
             Buffer& buffer = slot_buffers[binding.buffer_id];
             TouchBuffer(buffer, binding.buffer_id);
             SynchronizeBuffer(buffer, binding.device_addr, binding.size);
-            if (!flags[Dirty::VertexBuffer0 + index]) {
-                flush_bindings();
-                continue;
-            }
+            any_dirty |= flags[Dirty::VertexBuffer0 + index];
+        }
+        if (enabled_mask == 0 || !any_dirty) {
+            return;
+        }
+        const u32 min_index = static_cast<u32>(std::countr_zero(enabled_mask));
+        const u32 max_index = 32u - static_cast<u32>(std::countl_zero(enabled_mask));
+        HostBindings<Buffer> bindings{};
+        bindings.min_index = min_index;
+        bindings.max_index = max_index;
+        for (u32 index = min_index; index < max_index; ++index) {
             flags[Dirty::VertexBuffer0 + index] = false;
             const u32 stride = maxwell3d->regs.vertex_streams[index].stride;
+            if ((enabled_mask & (1u << index)) == 0) {
+                bindings.buffers.push_back(&slot_buffers[NULL_BUFFER_ID]);
+                bindings.offsets.push_back(0);
+                bindings.sizes.push_back(0);
+                bindings.strides.push_back(stride);
+                continue;
+            }
+            const Binding& binding = VertexBufferSlot(index);
+            Buffer& buffer = slot_buffers[binding.buffer_id];
             const u32 offset = buffer.Offset(binding.device_addr);
             buffer.MarkUsage(offset, binding.size);
-            if (!bindings.buffers.empty() && index != last_index + 1) {
-                flush_bindings();
-            }
-            if (bindings.buffers.empty()) {
-                bindings.min_index = index;
-            }
             bindings.buffers.push_back(&buffer);
             bindings.offsets.push_back(offset);
             bindings.sizes.push_back(binding.size);
             bindings.strides.push_back(stride);
-            last_index = index;
         }
-        flush_bindings();
+        runtime.BindVertexBuffers(bindings);
     } else {
         HostBindings<typename P::Buffer> host_bindings;
         bool any_valid{false};
@@ -921,7 +1103,6 @@ void BufferCache<P>::BindHostGraphicsUniformBuffers(size_t stage) {
 
 template <class P>
 void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 binding_index, bool needs_bind) {
-    ++channel_state->uniform_cache_shots[0];
     const Binding& binding = channel_state->uniform_buffers[stage][index];
     const DAddr device_addr = binding.device_addr;
     const u32 size = (std::min)(binding.size, (*channel_state->uniform_buffer_sizes)[stage][index]);
@@ -940,8 +1121,12 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
             return alignment > 1 && (offset % alignment) != 0;
         }
     }();
+
+    const bool cached_buffer_is_current =
+        has_host_buffer && !memory_tracker.IsRegionCpuModified(device_addr, size);
     const bool use_fast_buffer = needs_alignment_stream
-        || (has_host_buffer && size <= channel_state->uniform_buffer_skip_cache_size
+        || (has_host_buffer && !cached_buffer_is_current
+            && size <= channel_state->uniform_buffer_skip_cache_size
             && !memory_tracker.IsRegionGpuModified(device_addr, size));
     if (use_fast_buffer) {
         if constexpr (IS_OPENGL) {
@@ -968,7 +1153,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
         device_memory.ReadBlockUnsafe(device_addr, span.data(), size);
         return;
     }
-    // Classic cached path
+    ++channel_state->uniform_cache_shots[0];
     if (SynchronizeBuffer(buffer, device_addr, size)) {
         ++channel_state->uniform_cache_hits[0];
     }
@@ -1576,6 +1761,7 @@ void BufferCache<P>::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id,
 
 template <class P>
 BufferId BufferCache<P>::CreateBuffer(DAddr device_addr, u32 wanted_size) {
+    EnsureHeadroom(false);
     DAddr device_addr_end = Common::AlignUp(device_addr + wanted_size, CACHING_PAGESIZE);
     device_addr = Common::AlignDown(device_addr, CACHING_PAGESIZE);
     wanted_size = static_cast<u32>(device_addr_end - device_addr);
@@ -1613,7 +1799,7 @@ void BufferCache<P>::ChangeRegister(BufferId buffer_id) {
         total_used_memory += Common::AlignUp(size, 1024);
         buffer.setLRUID(lru_cache.Insert(buffer_id, frame_tick));
     } else {
-        total_used_memory -= Common::AlignUp(size, 1024);
+        total_used_memory -= std::min<u64>(total_used_memory, Common::AlignUp(size, 1024));
         lru_cache.Free(buffer.getLRUID());
     }
     const DAddr device_addr_begin = buffer.CpuAddr();
@@ -1699,6 +1885,98 @@ void BufferCache<P>::ImmediateUploadMemory([[maybe_unused]] Buffer& buffer,
     }
 }
 
+template <class P>
+bool BufferCache<P>::ResolveUnifiedWindows(
+    [[maybe_unused]] DAddr device_addr, [[maybe_unused]] u64 buffer_offset,
+    [[maybe_unused]] u64 size, [[maybe_unused]] boost::container::small_vector<u64, 4>& window_ids,
+    [[maybe_unused]] UnifiedWindowGroups& groups) {
+    if constexpr (USE_UNIFIED_MEMORY) {
+        const u8* const physical_base = device_memory.GetPhysicalBase();
+        const u64 unified_base = runtime.UnifiedMemoryBase();
+        const u64 unified_size = runtime.UnifiedMemorySize();
+        const u64 window_size = runtime.UnifiedMemoryWindowSize();
+        if (window_size == 0) {
+            return false;
+        }
+        const auto group_for = [&](u64 window) -> boost::container::small_vector<BufferCopy, 16>& {
+            for (size_t i = 0; i < window_ids.size(); ++i) {
+                if (window_ids[i] == window) {
+                    return groups[i];
+                }
+            }
+            window_ids.push_back(window);
+            groups.emplace_back();
+            return groups.back();
+        };
+        u64 downloaded = 0;
+        while (downloaded < size) {
+            const DAddr page_addr = device_addr + downloaded;
+            const u8* const ptr = device_memory.GetPointer<u8>(page_addr);
+            if (ptr == nullptr) {
+                return false;
+            }
+            const u64 page_offset = page_addr & Core::DEVICE_PAGEMASK;
+            u64 chunk = (std::min)(size - downloaded,
+                                   static_cast<u64>(Core::DEVICE_PAGESIZE) - page_offset);
+            const u64 phys_offset = static_cast<u64>(ptr - physical_base);
+            if (phys_offset < unified_base || phys_offset - unified_base + chunk > unified_size) {
+                return false;
+            }
+            const u64 relative = phys_offset - unified_base;
+            const u64 window = relative / window_size;
+            const u64 local_offset = relative % window_size;
+            chunk = (std::min)(chunk, window_size - local_offset);
+            auto& group = group_for(window);
+            if (!group.empty()) {
+                BufferCopy& last = group.back();
+                if (last.src_offset + last.size == buffer_offset + downloaded &&
+                    last.dst_offset + last.size == local_offset) {
+                    last.size += chunk;
+                    downloaded += chunk;
+                    continue;
+                }
+            }
+            group.push_back(BufferCopy{
+                .src_offset = buffer_offset + downloaded,
+                .dst_offset = local_offset,
+                .size = chunk,
+            });
+            downloaded += chunk;
+        }
+        return true;
+    } else {
+        return false;
+    }
+}
+
+template <class P>
+bool BufferCache<P>::TryUnifiedDownloadMemory([[maybe_unused]] Buffer& buffer,
+                                              [[maybe_unused]] std::span<BufferCopy> copies) {
+    if constexpr (USE_UNIFIED_MEMORY) {
+        boost::container::small_vector<u64, 4> window_ids;
+        UnifiedWindowGroups groups;
+        for (const BufferCopy& copy : copies) {
+            if (!ResolveUnifiedWindows(buffer.CpuAddr() + copy.src_offset, copy.src_offset,
+                                       copy.size, window_ids, groups)) {
+                return false;
+            }
+        }
+        for (const BufferCopy& copy : copies) {
+            buffer.MarkUsage(copy.src_offset, copy.size);
+        }
+        runtime.PreCopyBarrier();
+        for (size_t i = 0; i < window_ids.size(); ++i) {
+            const std::span<const BufferCopy> group_span(groups[i].data(), groups[i].size());
+            runtime.CopyToUnifiedMemory(window_ids[i], buffer, group_span);
+        }
+        runtime.UnifiedMemoryHostBarrier();
+        runtime.Finish();
+        return true;
+    } else {
+        return false;
+    }
+}
+
 template <class P>
 void BufferCache<P>::MappedUploadMemory([[maybe_unused]] Buffer& buffer,
                                         [[maybe_unused]] u64 total_size_bytes,
@@ -1802,6 +2080,12 @@ void BufferCache<P>::DownloadBufferMemory(Buffer& buffer, DAddr device_addr, u64
     }
 
     if constexpr (USE_MEMORY_MAPS) {
+        if constexpr (USE_UNIFIED_MEMORY) {
+            if (runtime.HasUnifiedMemory() &&
+                TryUnifiedDownloadMemory(buffer, std::span(copies.data(), copies.size()))) {
+                return;
+            }
+        }
         auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
         const u8* const mapped_memory = download_staging.mapped_span.data();
         const std::span<BufferCopy> copies_span(copies.data(), copies.data() + copies.size());
@@ -1872,7 +2156,7 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
 #ifdef YUZU_LEGACY
     if (!do_not_mark || !immediately_free)
 #endif
-        delayed_destruction_ring.Push(std::move(slot_buffers[buffer_id]));
+        sentenced_buffers.Push(std::move(slot_buffers[buffer_id]), runtime.CurrentSyncPoint());
 
     slot_buffers.erase(buffer_id);
 

src/video_core/buffer_cache/buffer_cache_base.h

@@ -9,6 +9,7 @@
 #include <algorithm>
 #include <array>
 #include <bit>
+#include <deque>
 #include <functional>
 #include <memory>
 #include <mutex>
@@ -30,7 +31,7 @@
 #include "common/slot_vector.h"
 #include "video_core/buffer_cache/buffer_base.h"
 #include "video_core/control/channel_state_cache.h"
-#include "video_core/delayed_destruction_ring.h"
+#include "video_core/deferred_destruction_queue.h"
 #include "video_core/dirty_flags.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/kepler_compute.h"
@@ -180,15 +181,18 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
     static constexpr bool USE_MEMORY_MAPS = P::USE_MEMORY_MAPS;
     static constexpr bool SEPARATE_IMAGE_BUFFERS_BINDINGS = P::SEPARATE_IMAGE_BUFFER_BINDINGS;
     static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = P::USE_MEMORY_MAPS_FOR_UPLOADS;
+    static constexpr bool USE_UNIFIED_MEMORY = P::USE_UNIFIED_MEMORY;
 
 #ifdef YUZU_LEGACY
-    static constexpr s64 TARGET_THRESHOLD = 3_GiB;
+    static constexpr u64 RECLAIM_HEADROOM = 384_MiB;
 #else
-    static constexpr s64 TARGET_THRESHOLD = 4_GiB;
+    static constexpr u64 RECLAIM_HEADROOM = 512_MiB;
 #endif
 
-    static constexpr s64 DEFAULT_EXPECTED_MEMORY = 512_MiB;
-    static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB;
+    static constexpr u64 FALLBACK_MEMORY_BUDGET = 2_GiB;
+    static constexpr u32 USAGE_REFRESH_INTERVAL = 16;
+    static constexpr u64 RECLAIM_GUARD_FRAMES = 8;
+    static constexpr u64 RECLAIM_TARGET_PERCENT = 95;
 
     // Debug Flags.
 
@@ -215,6 +219,10 @@ public:
 
     void TickFrame();
 
+    u64 ReclaimMemory(u64 target_bytes, bool allow_download);
+
+    void ReclaimDeferredResources(u64 completed_sync_point);
+
     void WriteMemory(DAddr device_addr, u64 size);
 
     void CachedWriteMemory(DAddr device_addr, u64 size);
@@ -358,7 +366,9 @@ private:
                ((device_addr + size) & ~Core::DEVICE_PAGEMASK);
     }
 
-    void RunGarbageCollector();
+    u64 DeviceUsage(bool force_refresh);
+
+    void EnsureHeadroom(bool allow_download);
 
     void BindHostIndexBuffer();
 
@@ -443,6 +453,15 @@ private:
 
     void MappedUploadMemory(Buffer& buffer, u64 total_size_bytes, std::span<BufferCopy> copies);
 
+    bool TryUnifiedDownloadMemory(Buffer& buffer, std::span<BufferCopy> copies);
+
+    using UnifiedWindowGroups =
+        boost::container::small_vector<boost::container::small_vector<BufferCopy, 16>, 4>;
+
+    bool ResolveUnifiedWindows(DAddr device_addr, u64 buffer_offset, u64 size,
+                               boost::container::small_vector<u64, 4>& window_ids,
+                               UnifiedWindowGroups& groups);
+
     void DownloadBufferMemory(Buffer& buffer_id);
 
     void DownloadBufferMemory(Buffer& buffer_id, DAddr device_addr, u64 size);
@@ -475,12 +494,7 @@ private:
     Tegra::MaxwellDeviceMemoryManager& device_memory;
 
     Common::SlotVector<Buffer> slot_buffers;
-#ifdef YUZU_LEGACY
-    static constexpr size_t TICKS_TO_DESTROY = 6;
-#else
-    static constexpr size_t TICKS_TO_DESTROY = 8;
-#endif
-    DelayedDestructionRing<Buffer, TICKS_TO_DESTROY> delayed_destruction_ring;
+    DeferredDestructionQueue<Buffer> sentenced_buffers;
 
     const Tegra::Engines::Maxwell3D::DrawManager::IndirectParams* current_draw_indirect{};
 
@@ -498,9 +512,14 @@ private:
     std::deque<Common::RangeSet<DAddr>> committed_gpu_modified_ranges;
 
     // Async Buffers
+    struct AsyncDownloadBatch {
+        boost::container::small_vector<BufferCopy, 4> staging_copies;
+        boost::container::small_vector<BufferCopy, 4> unified_copies;
+    };
+
     Common::OverlapRangeSet<DAddr> async_downloads;
     std::deque<std::optional<Async_Buffer>> async_buffers;
-    std::deque<boost::container::small_vector<BufferCopy, 4>> pending_downloads;
+    std::deque<AsyncDownloadBatch> pending_downloads;
     std::optional<Async_Buffer> current_buffer;
 
     std::deque<Async_Buffer> async_buffers_death_ring;
@@ -515,8 +534,14 @@ private:
     Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
     u64 frame_tick = 0;
     u64 total_used_memory = 0;
-    u64 minimum_memory = 0;
-    u64 critical_memory = 0;
+    u64 memory_budget = 0;
+    u64 cached_device_usage = 0;
+    /// Sync point the last reclaim's evictions were queued at. Their memory is not back with the
+    /// device until this completes, so reclaiming again before then measures stale usage.
+    u64 reclaim_wait_sync_point = 0;
+    u32 usage_refresh_countdown = 0;
+    bool in_reclaim = false;
+    bool reclaim_stalled = false;
     BufferId inline_buffer_id;
 #ifdef YUZU_LEGACY
     bool immediately_free = false;

src/video_core/control/channel_state.cpp

@@ -16,26 +16,6 @@
 #include "video_core/memory_manager.h"
 
 namespace Tegra::Control {
-namespace {
-
-// Match NVK/Nouveau's initial pushbuffer subchannel layout.
-constexpr u32 Nvk3DSubchannel = 0;
-constexpr u32 NvkComputeSubchannel = 1;
-constexpr u32 Nvk2DSubchannel = 3;
-constexpr u32 NvkCopySubchannel = 4;
-
-void BindNvkDefaultSubchannels(ChannelState::Payload& payload) {
-    auto& dma_pusher = payload.dma_pusher;
-    dma_pusher.BindSubchannel(&payload.maxwell_3d, Nvk3DSubchannel, Engines::EngineTypes::Maxwell3D);
-    dma_pusher.BindSubchannel(&payload.kepler_compute, NvkComputeSubchannel,
-                              Engines::EngineTypes::KeplerCompute);
-    // Subchannel 2 is M2MF there; Eden does not expose a 0x9039 engine yet.
-    dma_pusher.BindSubchannel(&payload.fermi_2d, Nvk2DSubchannel, Engines::EngineTypes::Fermi2D);
-    dma_pusher.BindSubchannel(&payload.maxwell_dma, NvkCopySubchannel,
-                              Engines::EngineTypes::MaxwellDMA);
-}
-
-} // Anonymous namespace
 
 ChannelState::Payload::Payload(Core::System& system, MemoryManager& memory_manager, ChannelState& channel_state)
     : maxwell_3d(memory_manager)
@@ -55,7 +35,6 @@ void ChannelState::Init(Core::System& system, u64 program_id_) {
     ASSERT(memory_manager);
     program_id = program_id_;
     payload.emplace(system, *memory_manager, *this);
-    BindNvkDefaultSubchannels(*payload);
     initialized = true;
 }
 

src/video_core/deferred_destruction_queue.h

@@ -0,0 +1,56 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <cstddef>
+#include <utility>
+
+#include <boost/container/deque.hpp>
+#include <boost/container/options.hpp>
+
+#include "common/common_types.h"
+
+namespace VideoCommon {
+
+template <typename T>
+class DeferredDestructionQueue {
+public:
+    void Push(T&& object, u64 sync_point) {
+        entries.emplace_back(std::move(object), sync_point);
+    }
+
+    void Reclaim(u64 completed_sync_point) {
+        while (!entries.empty() && entries.front().sync_point <= completed_sync_point) {
+            entries.pop_front();
+        }
+    }
+
+    void Clear() {
+        entries.clear();
+    }
+
+    [[nodiscard]] size_t Size() const noexcept {
+        return entries.size();
+    }
+
+    [[nodiscard]] bool Empty() const noexcept {
+        return entries.empty();
+    }
+
+private:
+    struct Entry {
+        Entry(T&& object_, u64 sync_point_) noexcept
+            : object{std::move(object_)}, sync_point{sync_point_} {}
+
+        T object;
+        u64 sync_point;
+    };
+
+    using EntryDequeOptions =
+        boost::container::deque_options<boost::container::block_size<8u>>::type;
+
+    boost::container::deque<Entry, void, EntryDequeOptions> entries;
+};
+
+} // namespace VideoCommon

src/video_core/delayed_destruction_ring.h

@@ -1,34 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#pragma once
-
-#include <array>
-#include <cstddef>
-#include <utility>
-#include <vector>
-
-namespace VideoCommon {
-
-/// Container to push objects to be destroyed a few ticks in the future
-template <typename T, size_t TICKS_TO_DESTROY>
-class DelayedDestructionRing {
-public:
-    void Tick() {
-        index = (index + 1) % TICKS_TO_DESTROY;
-        elements[index].clear();
-    }
-
-    void Push(T&& object) {
-        elements[index].push_back(std::move(object));
-    }
-
-private:
-    size_t index = 0;
-    std::array<std::vector<T>, TICKS_TO_DESTROY> elements;
-};
-
-} // namespace VideoCommon

src/video_core/engines/fermi_2d.cpp

@@ -71,7 +71,8 @@ void Fermi2D::Blit() {
     constexpr s64 null_derivative = 1ULL << 32;
     Surface src = regs.src;
     const auto bytes_per_pixel = BytesPerBlock(PixelFormatFromRenderTargetFormat(src.format));
-    const bool delegate_to_gpu = src.width > 512 && src.height > 512 && bytes_per_pixel <= 8 &&
+    const u64 src_area = static_cast<u64>(src.width) * static_cast<u64>(src.height);
+    const bool delegate_to_gpu = src_area > 512ULL * 512ULL && bytes_per_pixel <= 8 &&
                                  src.format != regs.dst.format;
 
     auto srcX = args.src_x0;

src/video_core/engines/puller.cpp

@@ -22,21 +22,10 @@
 
 namespace Tegra::Engines {
 
-namespace {
-
-constexpr u32 Gf100BindClassMask = 0xffff;
-constexpr u32 Gf100BindValidMask =  0x1f0000 | Gf100BindClassMask;
-
-} // Anonymous namespace
-
 void Puller::ProcessBindMethod(DmaPusher& dma_pusher, const MethodCall& method_call) {
-    LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {:#x}", method_call.subchannel, method_call.argument);
-    u32 engine = method_call.argument;
-    if ((engine & ~Gf100BindClassMask) != 0 && (engine & ~Gf100BindValidMask) == 0) {
-        engine &= Gf100BindClassMask;
-    }
-
-    const auto engine_id = static_cast<EngineID>(engine);
+    // Bind the current subchannel to the desired engine id.
+    LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel, method_call.argument);
+    const auto engine_id = static_cast<EngineID>(method_call.argument);
     bound_engines[method_call.subchannel] = engine_id;
     switch (engine_id) {
     case EngineID::FERMI_TWOD_A:

src/video_core/fence_manager.h

@@ -18,7 +18,7 @@
 #include "common/common_types.h"
 #include "common/settings.h"
 #include "common/thread.h"
-#include "video_core/delayed_destruction_ring.h"
+#include "video_core/deferred_destruction_queue.h"
 #include "video_core/gpu.h"
 #include "video_core/host1x/host1x.h"
 #include "video_core/host1x/syncpoint_manager.h"
@@ -50,7 +50,8 @@ public:
     /// Notify the fence manager about a new frame
     void TickFrame() {
         std::unique_lock lock(ring_guard);
-        delayed_destruction_ring.Tick();
+        ++retire_tick;
+        sentenced_fences.Reclaim(retire_tick > RETIRE_DELAY ? retire_tick - RETIRE_DELAY : 0);
     }
 
     // Unlike other fences, this one doesn't
@@ -91,9 +92,6 @@ public:
             func();
         }
         fences.push(std::move(new_fence));
-        if (should_flush) {
-            rasterizer.FlushCommands();
-        }
         if constexpr (can_async_check) {
             guard.unlock();
             cv.notify_all();
@@ -186,7 +184,7 @@ private:
             }
             {
                 std::unique_lock lock(ring_guard);
-                delayed_destruction_ring.Push(std::move(current_fence));
+                sentenced_fences.Push(std::move(current_fence), retire_tick);
             }
             fences.pop();
         }
@@ -219,7 +217,7 @@ private:
             }
             {
                 std::unique_lock lock(ring_guard);
-                delayed_destruction_ring.Push(std::move(current_fence));
+                sentenced_fences.Push(std::move(current_fence), retire_tick);
             }
         }
     }
@@ -238,10 +236,10 @@ private:
 
     void PopAsyncFlushes() {
         {
-            std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
+            std::scoped_lock lock{texture_cache.mutex};
             texture_cache.PopAsyncFlushes();
-            buffer_cache.PopAsyncFlushes();
         }
+        buffer_cache.PopAsyncFlushes();
         query_cache.PopAsyncFlushes();
     }
 
@@ -264,7 +262,9 @@ private:
 
     std::jthread fence_thread;
 
-    DelayedDestructionRing<TFence, 8> delayed_destruction_ring;
+    static constexpr u64 RETIRE_DELAY = 8;
+    u64 retire_tick = 1;
+    DeferredDestructionQueue<TFence> sentenced_fences;
 };
 
 } // namespace VideoCommon

src/video_core/gpu_thread.cpp

@@ -30,6 +30,7 @@ void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Core::Fronten
     thread = std::jthread([&](std::stop_token stop_token) {
         Common::SetCurrentThreadName("GPU");
         Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical);
+        Common::SetCurrentThreadToPerformanceCores();
         system.RegisterHostThread();
 
         auto current_context = context.Acquire();

src/video_core/host_shaders/CMakeLists.txt

@@ -17,11 +17,13 @@ set(SHADER_FILES
     ${CMAKE_CURRENT_SOURCE_DIR}/astc_decoder.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_color_float.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_2d.comp
+    ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_2d_buffer.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_color_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_stencil_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d_bcn.comp
+    ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d_buffer.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d24s8.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d32f.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_d32f_to_abgr8.frag
@@ -32,6 +34,7 @@ set(SHADER_FILES
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_msaa_to_non_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa.frag
+    ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa_depth.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_s8d24_to_abgr8.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/full_screen_triangle.vert
     ${CMAKE_CURRENT_SOURCE_DIR}/fxaa.frag

src/video_core/host_shaders/astc_decoder.comp

@@ -77,14 +77,8 @@ uvec4 local_buff;
 uvec4 color_endpoint_data;
 int color_bitsread = 0;
 
-// Global "vector" to be pushed into when decoding
-// At most will require BLOCK_WIDTH x BLOCK_HEIGHT in single plane mode
-// At most will require BLOCK_WIDTH x BLOCK_HEIGHT x 2 in dual plane mode
-// So the maximum would be 144 (12 x 12) elements, x 2 for two planes
-#define DIVCEIL(number, divisor) (number + divisor - 1) / divisor
-#define ARRAY_NUM_ELEMENTS 144
-#define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
-uint result_vector[ARRAY_NUM_ELEMENTS * 2];
+#define MAX_WEIGHT_VALUES 64
+uint result_vector[MAX_WEIGHT_VALUES];
 
 int result_index = 0;
 uint result_vector_max_index;
@@ -492,7 +486,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, o
         A = ReplicateBitTo9((bitval & 1));
         switch (encoding) {
         case JUST_BITS:
-            color_values[++out_index] = FastReplicateTo8(bitval, bitlen);
+            color_values[out_index++] = FastReplicateTo8(bitval, bitlen);
             break;
         case TRIT: {
             D = QuintTritValue(val);
@@ -571,7 +565,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, o
             uint T = (D * C) + B;
             T ^= A;
             T = (A & 0x80) | (T >> 2);
-            color_values[++out_index] = T;
+            color_values[out_index++] = T;
         }
     }
 }
@@ -753,12 +747,12 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui
 #define READ_UINT_VALUES(N)                                                                        \
     uvec4 V[2];                                                                                    \
     for (uint i = 0; i < N; i++) {                                                                 \
-        V[i / 4][i % 4] = color_values[++colvals_index];                      \
+        V[i / 4][i % 4] = color_values[colvals_index++];                      \
     }
 #define READ_INT_VALUES(N)                                                                         \
     ivec4 V[2];                                                                                    \
     for (uint i = 0; i < N; i++) {                                                                 \
-        V[i / 4][i % 4] = int(color_values[++colvals_index]);                      \
+        V[i / 4][i % 4] = int(color_values[colvals_index++]);                      \
     }
 
     switch (color_endpoint_mode) {
@@ -1225,6 +1219,10 @@ void DecompressBlock(ivec3 coord) {
         FillError(coord);
         return;
     }
+    if (GetNumWeightValues(size_params, dual_plane) > MAX_WEIGHT_VALUES) {
+        FillError(coord);
+        return;
+    }
     uint partition_index = 1;
     uvec4 color_endpoint_mode = uvec4(0);
     uint ced_pointer = 0;
@@ -1384,11 +1382,7 @@ void DecompressBlock(ivec3 coord) {
                 p = Cf / 65535.0f;
             }
 
-#ifdef VULKAN
-            imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar);
-#else
             imageStore(dest_image, coord + ivec3(i, j, 0), clamp(p, 0.0f, 1.0f).gbar);
-#endif
         }
     }
 }

src/video_core/host_shaders/block_linear_unswizzle_2d_buffer.comp

@@ -0,0 +1,104 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 430
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_BUFFER 1
+
+layout(push_constant) uniform PushConstants {
+    uvec3 dim;
+    uint bytes_per_block_log2;
+
+    uvec3 origin;
+    uint layer_stride;
+
+    uint block_size;
+    uint x_shift;
+    uint block_height;
+    uint block_height_mask;
+} pc;
+
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_BUFFER, std430) writeonly buffer OutputBuffer {
+    uint out_u32[];
+};
+
+layout(local_size_x = 16, local_size_y = 8, local_size_z = 1) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT;
+
+const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1u, GOB_SIZE_Y - 1u);
+
+uint SwizzleTable(uint pos) {
+    const uint t[8] = uint[](
+        0x12100200, 0x13110301, 0x16140604, 0x17150705,
+        0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d
+    );
+    const uint i = pos >> 4;
+    const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff;
+    return (h << 4) | (pos & 0xf);
+}
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return SwizzleTable(pos.y * 64u + pos.x);
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (pc.bytes_per_block_log2) {
+    case 2u:
+        return uvec4(u32data[offset / 4u], 0u, 0u, 0u);
+    case 3u:
+        return uvec4(u64data[offset / 8u], 0u, 0u);
+    case 4u:
+        return u128data[offset / 16u];
+    }
+    return uvec4(0u);
+}
+
+void main() {
+    uvec3 coord = gl_GlobalInvocationID;
+    if (coord.x >= pc.dim.x || coord.y >= pc.dim.y || coord.z >= pc.dim.z) {
+        return;
+    }
+
+    uvec3 pos = coord + pc.origin;
+    pos.x <<= pc.bytes_per_block_log2;
+
+    uint swizzle = SwizzleOffset(pos.xy);
+    uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0u;
+    offset += pos.z * pc.layer_stride;
+    offset += (block_y >> pc.block_height) * pc.block_size;
+    offset += (block_y & pc.block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << pc.x_shift;
+    offset += swizzle;
+
+    uvec4 texel = ReadTexel(offset);
+
+    uint words = 1u << (pc.bytes_per_block_log2 - 2u);
+    uint linear_index = coord.x + coord.y * pc.dim.x + coord.z * pc.dim.x * pc.dim.y;
+    uint out_idx = linear_index * words;
+
+    out_u32[out_idx] = texel.x;
+    if (words > 1u) {
+        out_u32[out_idx + 1u] = texel.y;
+    }
+    if (words > 2u) {
+        out_u32[out_idx + 2u] = texel.z;
+        out_u32[out_idx + 3u] = texel.w;
+    }
+}

src/video_core/host_shaders/block_linear_unswizzle_3d_buffer.comp

@@ -0,0 +1,105 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 430
+
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_BUFFER 1
+
+layout(push_constant) uniform PushConstants {
+    uvec3 dim;
+    uint bytes_per_block_log2;
+
+    uvec3 origin;
+    uint slice_size;
+
+    uint block_size;
+    uint x_shift;
+    uint block_height;
+    uint block_height_mask;
+
+    uint block_depth;
+    uint block_depth_mask;
+} pc;
+
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_BUFFER, std430) writeonly buffer OutputBuffer {
+    uint out_u32[];
+};
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 4) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT;
+
+const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1u, GOB_SIZE_Y - 1u);
+
+uint SwizzleTable(uint pos) {
+    const uint t[8] = uint[](
+        0x12100200, 0x13110301, 0x16140604, 0x17150705,
+        0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d
+    );
+    const uint i = pos >> 4;
+    const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff;
+    return (h << 4) | (pos & 0xf);
+}
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return SwizzleTable(pos.y * 64u + pos.x);
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (pc.bytes_per_block_log2) {
+    case 2u:
+        return uvec4(u32data[offset / 4u], 0u, 0u, 0u);
+    case 3u:
+        return uvec4(u64data[offset / 8u], 0u, 0u);
+    case 4u:
+        return u128data[offset / 16u];
+    }
+    return uvec4(0u);
+}
+
+void main() {
+    uvec3 coord = gl_GlobalInvocationID;
+    if (coord.x >= pc.dim.x || coord.y >= pc.dim.y || coord.z >= pc.dim.z) {
+        return;
+    }
+
+    uvec3 pos = coord + pc.origin;
+    pos.x <<= pc.bytes_per_block_log2;
+
+    uint swizzle = SwizzleOffset(pos.xy);
+    uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0u;
+    offset += (pos.z >> pc.block_depth) * pc.slice_size;
+    offset += (pos.z & pc.block_depth_mask) << (GOB_SIZE_SHIFT + pc.block_height);
+    offset += (block_y >> pc.block_height) * pc.block_size;
+    offset += (block_y & pc.block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << pc.x_shift;
+    offset += swizzle;
+
+    uvec4 texel = ReadTexel(offset);
+
+    uint words = 1u << (pc.bytes_per_block_log2 - 2u);
+    uint linear_index = coord.x + coord.y * pc.dim.x + coord.z * pc.dim.x * pc.dim.y;
+    uint out_idx = linear_index * words;
+
+    out_u32[out_idx] = texel.x;
+    if (words > 1u) {
+        out_u32[out_idx + 1u] = texel.y;
+    }
+    if (words > 2u) {
+        out_u32[out_idx + 2u] = texel.z;
+        out_u32[out_idx + 3u] = texel.w;
+    }
+}

src/video_core/host_shaders/convert_non_msaa_to_msaa_depth.frag

@@ -0,0 +1,19 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 450 core
+
+layout(binding = 0) uniform sampler2D img_in;
+
+layout(push_constant) uniform PushConstants {
+    ivec2 dst_offset;
+    ivec2 src_offset;
+    ivec2 scale;
+};
+
+void main() {
+    const ivec2 msaa_coord = ivec2(gl_FragCoord.xy) - dst_offset;
+    const ivec2 sample_offset = ivec2(gl_SampleID % scale.x, gl_SampleID / scale.x);
+    const ivec2 coord = msaa_coord * scale + sample_offset + src_offset;
+    gl_FragDepth = texelFetch(img_in, coord, 0).r;
+}