[scarletnx] Add CPU backend

Signed-off-by: lizzie <lizzie@eden-emu.dev>
Assisted-by: Claude Opus 5.5 <claude@anthropic.com>

.gitignore

@@ -7,7 +7,6 @@
 # Build directory
 /[Bb]uild*/
 doc-build/
-out/
 AppDir/
 uruntime
 

CMakeLists.txt

@@ -304,6 +304,16 @@ if (ARCHITECTURE_arm64 AND (ANDROID OR LINUX))
     add_compile_definitions(HAS_NCE=1)
 endif()
 
+set(ENABLE_SCARLETNX_DEFAULT ON)
+if ((NOT LINUX AND NOT FREEBSD) OR (NOT CMAKE_C_COMPILER_ID MATCHES "Clang" OR NOT CMAKE_CXX_COMPILER_ID MATCHES "Clang"))
+    set(ENABLE_SCARLETNX_DEFAULT OFF)
+endif()
+option(ENABLE_SCARLETNX "Enable the ScarletNX recompiler CPU backend (requires Clang and LLVM)" "${ENABLE_SCARLETNX_DEFAULT}")
+if (ENABLE_SCARLETNX)
+    set(HAS_SCARLETNX 1)
+    add_compile_definitions(HAS_SCARLETNX=1)
+endif()
+
 if (YUZU_ROOM)
     add_compile_definitions(YUZU_ROOM)
 endif()

cpmfile.json

@@ -243,6 +243,27 @@
         "repo": "crosire/reshade",
         "version": "v6.8.0"
     },
+    "scarletnx": {
+        "bundled": true,
+        "hash": "93f7e6810644253d6396239092d5769c44b005fa7a17cf39c6d773050a19de74fca1f96bb6eec7cb40384eb158b2595c9e1171ad592d8eb8326fd31f5632f505",
+        "options": [
+            "SNX_BUILD_NX OFF",
+            "SNX_BUILD_GNUABI OFF",
+            "SNX_BUILD_3DS OFF",
+            "SNX_BUILD_VITA OFF",
+            "SNX_BUILD_AGB OFF",
+            "SNX_DEPLOY OFF",
+            "SNX_BUILD_TESTING OFF",
+            "SNX_BUILD_GUI OFF",
+            "SNX_MULTICALL OFF",
+            "SNX_TRACY OFF",
+            "SNX_RECOMPILER_SHARED OFF",
+            "SNX_LLVM_SHARED OFF"
+        ],
+        "package": "ScarletNX",
+        "repo": "scarletnx/ScarletNX",
+        "version": "f16810ce9c"
+    },
     "sdl3": {
         "hash": "df5a323af7ac366661a3c0e887969c72584d232f3cc211419d59b0487b620b6b2859d4549c9e8df002ee489290062e466fcfddf7edc0872a37b1f2845e81c0f3",
         "min_version": "3.2.10",
@@ -351,13 +372,13 @@
     },
     "zstd": {
         "find_args": "MODULE",
-        "hash": "2b006c3c110c38bd2f5896b42d36a38931c8eae261f30552f9a9fd04458d218d8f1a1425f117fd12af545ed77042e4b2d1a1a9b6286c4d0b9d74ef79f20a8f08",
+        "hash": "cc5ad4b119a9c2ea57f0b71eeff01113bb506e0d17000159c5409cb8236d22e38c52d5e9e97e7947a4bf1b2dfc44b6c503ab2d9aedbd59458435c6a2849cb029",
         "min_version": "1.5",
         "options": [
             "ZSTD_BUILD_SHARED OFF"
         ],
         "repo": "facebook/zstd",
         "source_subdir": "build/cmake",
-        "version": "33618c8"
+        "version": "b8d6101fba"
     }
 }

docs/Options.md

@@ -78,6 +78,9 @@ The following options are desktop only.
 - `ENABLE_OPENGL` (ON) Enable the OpenGL graphics backend
   - Unavailable on Windows/ARM64 and on Android
   - You probably shouldn't turn this off.
+- `ENABLE_SCARLETNX` (OFF) Enable the ScarletNX recompiler
+  - Requires Clang and LLVM's development files
+  - Needs multicore and fastmem
 
 ### Qt
 

externals/CMakeLists.txt

@@ -249,6 +249,14 @@ if (YUZU_TESTS OR DYNARMIC_TESTS)
     AddJsonPackage(catch2)
 endif()
 
+# ScarletNX
+if (ENABLE_SCARLETNX)
+    AddJsonPackage(scarletnx)
+    if (TARGET scarletnx-recompiler)
+        set_target_properties(scarletnx-recompiler PROPERTIES SYSTEM ON)
+    endif()
+endif()
+
 # getopt
 if (MSVC)
     add_subdirectory(getopt)

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

@@ -16,6 +16,7 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting {
     RENDERER_USE_SPEED_LIMIT("use_speed_limit"),
     USE_CUSTOM_CPU_TICKS("use_custom_cpu_ticks"),
     SKIP_CPU_INNER_INVALIDATION("skip_cpu_inner_invalidation"),
+    NCE_INVALIDATION_GPU_READBACK("nce_invalidation_gpu_readback"),
     FIX_BLOOM_EFFECTS("fix_bloom_effects"),
     EMULATE_BGR565("emulate_bgr565"),
     RESCALE_HACK("rescale_hack"),

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

@@ -859,6 +859,13 @@ abstract class SettingsItem(
                     descriptionId = R.string.skip_cpu_inner_invalidation_description
                 )
             )
+            put(
+                SwitchSetting(
+                    BooleanSetting.NCE_INVALIDATION_GPU_READBACK,
+                    titleId = R.string.nce_invalidation_gpu_readback,
+                    descriptionId = R.string.nce_invalidation_gpu_readback_description
+                )
+            )
             put(
                 SwitchSetting(
                     BooleanSetting.FIX_BLOOM_EFFECTS,

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

@@ -554,6 +554,7 @@ class SettingsFragmentPresenter(
             add(HeaderSetting(R.string.hacks))
 
             add(BooleanSetting.SKIP_CPU_INNER_INVALIDATION.key)
+            add(BooleanSetting.NCE_INVALIDATION_GPU_READBACK.key)
             add(BooleanSetting.FIX_BLOOM_EFFECTS.key)
             add(BooleanSetting.EMULATE_BGR565.key)
             add(BooleanSetting.RENDERER_ASYNCHRONOUS_SHADERS.key)

src/android/app/src/main/java/org/yuzu/yuzu_emu/model/Game.kt

@@ -100,7 +100,7 @@ class Game(
 
     companion object {
         val extensions: Set<String> = HashSet(
-            listOf("xci", "nsp", "nca", "nro", "nsz", "xcz")
+            listOf("xci", "nsp", "nca", "nro")
         )
     }
 }

src/android/app/src/main/java/org/yuzu/yuzu_emu/utils/GameHelper.kt

@@ -109,7 +109,7 @@ object GameHelper {
 
     // File extensions considered as external content, buuut should
     // be done better imo.
-    private val externalContentExtensions = setOf("nsp", "xci", "nsz", "xcz")
+    private val externalContentExtensions = setOf("nsp", "xci")
 
     private fun scanContentContainersRecursive(
         files: Array<MinimalDocumentFile>,

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

@@ -76,9 +76,7 @@ jboolean Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getIsValid(JNIEnv* env, jobj
             auto const file_type = loader->GetFileType();
             if (file_type == Loader::FileType::Unknown || file_type == Loader::FileType::Error)
                 return false;
-            if ((file_type == Loader::FileType::NSP || file_type == Loader::FileType::XCI ||
+            if ((file_type == Loader::FileType::NSP || file_type == Loader::FileType::XCI) && !Loader::IsBootableGameContainer(file, file_type))
-                 file_type == Loader::FileType::NSZ || file_type == Loader::FileType::XCZ) &&
-                !Loader::IsBootableGameContainer(file, file_type))
                 return false;
             u64 program_id = 0;
             return loader->ReadProgramId(program_id) == Loader::ResultStatus::Success;

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

@@ -587,6 +587,8 @@
     <string name="fast_gpu_time_description">يجعل اللعبة تعتقد أن عمل وحدة معالجة الرسومات ينتهي بشكل أسرع مما هو عليه في الواقع، لذلك تتوقف عن خفض الدقة ومسافة العرض لتتناسب مع سرعات جهاز Switch.</string>
     <string name="skip_cpu_inner_invalidation">تخطي إبطال صلاحية وحدة المعالجة المركزية الداخلية</string>
     <string name="skip_cpu_inner_invalidation_description">يتخطى بعض عمليات إبطال ذاكرة التخزين المؤقتة من جانب وحدة المعالجة المركزية أثناء تحديثات الذاكرة، مما يقلل من استخدام وحدة المعالجة المركزية ويحسن أداءها. قد يتسبب ذلك في حدوث أعطال أو تعطل في بعض الألعاب.</string>
+    <string name="nce_invalidation_gpu_readback">إعادة قراءة وحدة معالجة الرسومات قبل إبطال صلاحية NCE</string>
+    <string name="nce_invalidation_gpu_readback_description">إصلاح مؤقت. يعالج مشكلة تعطل لعبة Ender Magnolia (وغيرها من ألعاب UE). قد يؤدي إلى انخفاض في الأداء. يجري العمل على تحسينه.</string>
     <string name="fix_bloom_effects">إصلاح تأثيرات التوهج</string>
     <string name="fix_bloom_effects_description">يقلل من ضبابية التوهج في LA/EOW  (Adreno A6XX - A7XX/ Turnip)، ويزيل التوهج في Burnout. تحذير: قد يسبب تشوهات رسومية في ألعاب أخرى.</string>
     <string name="emulate_bgr565">محاكاة BGR565</string>

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

@@ -577,6 +577,8 @@
     <string name="fast_gpu_time_description">讓遊戲誤以為 GPU 工作能更快完成，避免遊戲為了配合 Switch 的時脈而降低解析度與渲染距離</string>
     <string name="skip_cpu_inner_invalidation">跳過 CPU 內部失效處理</string>
     <string name="skip_cpu_inner_invalidation_description">在記憶體更新期間跳過某些 CPU 端快取的失效處理，減少 CPU 使用率並提高其性能。可能會導致某些遊戲出現故障或當機</string>
+    <string name="nce_invalidation_gpu_readback">在 NCE 無效化前進行 GPU 回讀</string>
+    <string name="nce_invalidation_gpu_readback_description">（暫時性功能）修復 \"終結者瑪格諾莉亞：霧中綻放\" 及其它使用虛幻引擎之遊戲中的當機問題，可能會降低效能。此功能正在改善中</string>
     <string name="fix_bloom_effects">修正光暈特效</string>
     <string name="fix_bloom_effects_description">在 \"薩爾達傳說：織夢島\" 和 \"薩爾達傳說：智慧的再現\" 中減少光暈模糊的問題並移除 \"橫衝直撞：狂飆樂園重製版\" 中的光暈效果（適用於 Adreno A6XX - A7XX/ Turnip）。警告：可能會在其它遊戲中導致圖形異常</string>
     <string name="emulate_bgr565">模擬 BGR565</string>

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

@@ -593,6 +593,8 @@
     <string name="fast_gpu_time_description">Makes the game believe GPU work finishes faster than it does, so it stops lowering resolution and render distance to fit the Switch\'s clocks.</string>
     <string name="skip_cpu_inner_invalidation">Skip CPU Inner Invalidation</string>
     <string name="skip_cpu_inner_invalidation_description">Skips certain CPU-side cache invalidations during memory updates, reducing CPU usage and improving it\'s performance. This may cause glitches or crashes on some games.</string>
+    <string name="nce_invalidation_gpu_readback">GPU Readback Before NCE Invalidation</string>
+    <string name="nce_invalidation_gpu_readback_description">Temporary. Fix Ender Magnolia (and other UE titles) crashes. May reduce performance. Improvement in progress.</string>
     <string name="fix_bloom_effects">Fix Bloom Effects</string>
     <string name="fix_bloom_effects_description">Reduces bloom blur in LA/EOW (Adreno A6XX - A7XX/ Turnip), removes bloom in Burnout. Warning: may cause graphical artifacts in other games.</string>
     <string name="emulate_bgr565">Emulate BGR565</string>

src/common/settings.h

@@ -256,7 +256,9 @@ struct Values {
                                                     CpuBackend::Dynarmic,
 #endif
                                                     CpuBackend::Dynarmic,
-#ifdef HAS_NCE
+#if defined(HAS_SCARLETNX)
+                                                    CpuBackend::ScarletNx,
+#elif defined(HAS_NCE)
                                                     CpuBackend::Nce,
 #else
                                                     CpuBackend::Dynarmic,
@@ -639,6 +641,8 @@ struct Values {
                                                         Specialization::Default,
                                                         true,
                                                         true};
+    SwitchableSetting<bool> nce_invalidation_gpu_readback{
+        linkage, false, "nce_invalidation_gpu_readback", Category::RendererHacks};
     SwitchableSetting<bool> async_presentation{linkage,
 #ifdef __ANDROID__
                                                true,

src/common/settings_enums.h

@@ -138,7 +138,7 @@ ENUM(RendererBackend, OpenGL_GLSL, Vulkan, Null, OpenGL_GLASM, OpenGL_SPIRV);
 ENUM(GpuAccuracy, Low, High);
 ENUM(DmaAccuracy, Default, Unsafe, Safe);
 ENUM(GpuFenceBehavior, Default, Immediate, Balanced, Accurate);
-ENUM(CpuBackend, Dynarmic, Nce);
+ENUM(CpuBackend, Dynarmic, Nce, ScarletNx);
 ENUM(CpuAccuracy, Auto, Accurate, Unsafe, Paranoid, Debugging);
 ENUM(CpuClock, Normal, Boost, Overclock)
 ENUM(MemoryLayout, Memory_4Gb, Memory_6Gb, Memory_8Gb, Memory_10Gb, Memory_12Gb);

src/common/zstd_compression.cpp

@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -49,10 +46,4 @@ std::vector<u8> DecompressDataZSTD(std::span<const u8> compressed) {
     return decompressed;
 }
 
-std::size_t DecompressDataZSTD(std::span<const u8> compressed, std::span<u8> output) {
-    const std::size_t result =
-        ZSTD_decompress(output.data(), output.size(), compressed.data(), compressed.size());
-    return ZSTD_isError(result) ? 0 : result;
-}
-
 } // namespace Common::Compression

src/common/zstd_compression.h

@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -45,15 +42,4 @@ namespace Common::Compression {
  */
 [[nodiscard]] std::vector<u8> DecompressDataZSTD(std::span<const u8> compressed);
 
-/**
- * Decompresses a single Zstandard frame into a caller-provided buffer of known size.
- *
- * @param compressed the compressed source memory region.
- * @param output     the destination buffer. Its size is the maximum number of bytes to write.
- *
- * @return the number of bytes written, or 0 if the frame is invalid or doesn't fit. An empty
- *         frame also returns 0.
- */
-[[nodiscard]] std::size_t DecompressDataZSTD(std::span<const u8> compressed, std::span<u8> output);
-
 } // namespace Common::Compression

src/core/CMakeLists.txt

@@ -33,8 +33,6 @@ add_library(core STATIC
     crypto/encryption_layer.h
     crypto/key_manager.cpp
     crypto/key_manager.h
-    crypto/ncz_decompressor.cpp
-    crypto/ncz_decompressor.h
     crypto/partition_data_manager.cpp
     crypto/partition_data_manager.h
     crypto/xts_encryption_layer.cpp
@@ -1250,6 +1248,18 @@ if (HAS_NCE)
     target_link_libraries(core PRIVATE merry::oaknut)
 endif()
 
+if (HAS_SCARLETNX)
+    target_sources(core PRIVATE
+        arm/scarletnx/arm_scarletnx_32.cpp
+        arm/scarletnx/arm_scarletnx_32.h
+        arm/scarletnx/arm_scarletnx_64.cpp
+        arm/scarletnx/arm_scarletnx_64.h
+        arm/scarletnx/scarletnx_api.h
+        arm/scarletnx/scarletnx_process.cpp
+        arm/scarletnx/scarletnx_process.h)
+    target_link_libraries(core PRIVATE ScarletNX::recompiler)
+endif()
+
 if (ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64 OR ARCHITECTURE_riscv64 OR ARCHITECTURE_loongarch64)
     target_sources(core PRIVATE
         arm/dynarmic/arm_dynarmic.h

src/core/arm/nce/arm_nce.cpp

@@ -369,8 +369,8 @@ void ArmNce::SetContext(const Kernel::Svc::ThreadContext& ctx) {
 }
 
 void ArmNce::SignalInterrupt(Kernel::KThread* thread) {
-    constexpr u64 break_loop = static_cast<u64>(HaltReason::BreakLoop);
+    // Add break loop condition.
-    if ((m_guest_ctx.esr_el1.fetch_or(break_loop) & break_loop) != 0) return;
+    m_guest_ctx.esr_el1.fetch_or(static_cast<u64>(HaltReason::BreakLoop));
 
     auto* params = &thread->GetNativeExecutionParameters();
     LockThreadParameters(params);

src/core/arm/scarletnx/arm_scarletnx_32.cpp

@@ -0,0 +1,227 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include <cstring>
+#include <utility>
+#include <atomic>
+
+#include "common/logging.h"
+#include "core/arm/scarletnx/arm_scarletnx_32.h"
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/hle/kernel/k_process.h"
+
+namespace Core {
+
+namespace {
+
+// The exit SignalInterrupt sets, which no block sets itself.
+constexpr u32 InterruptExit = 0x1000;
+constexpr u32 UserMode = 0x10;
+constexpr u32 ModeMask = 0x1F;
+
+// FPSCR splits into FPSR and FPCR as in ArmDynarmic32.
+std::pair<u32, u32> FpscrToFpsrFpcr(u32 fpscr) {
+    const u32 fpsr = (fpscr & 0xf8000000) | (fpscr & 0x80) | (fpscr & 0x1f);
+    const u32 fpcr = (fpscr & 0x7ff8000) | (fpscr & 0x1f00);
+    return {fpsr, fpcr};
+}
+
+u32 FpsrFpcrToFpscr(u64 fpsr, u64 fpcr) {
+    const auto [s, c] = FpscrToFpsrFpcr(u32(fpsr | fpcr));
+    return s | c;
+}
+
+} // namespace
+
+ArmScarletNx32::ArmScarletNx32(std::shared_ptr<ScarletNxProcess> process, std::size_t core_index)
+    : ArmInterface{true}, m_process{std::move(process)}, m_core_index{core_index} {
+    m_rt.coproc = &Coprocessor;
+    m_cpu.parent = this;
+    m_cpu.cpu.mem = m_process->m_arena;
+    m_cpu.cpu.mem_size = 1ULL << 32;
+    m_cpu.cpu.rt = &m_rt;
+
+    m_fault_scope.memory = &m_process->m_process.GetMemory();
+    m_fault_scope.arena = m_process->m_arena;
+    m_fault_scope.arena_size = ScarletNxProcess::ArenaSize;
+}
+
+ArmScarletNx32::~ArmScarletNx32() {
+    snx_recompiled32_free(m_module);
+}
+
+int ArmScarletNx32::Svc(void* context, u32 imm, u64 number, const u64 arguments[6]) {
+    auto& self = *static_cast<ArmScarletNx32*>(context);
+    self.m_svc = imm;
+    self.m_in_svc = true;
+    return 1;
+}
+
+void ArmScarletNx32::Coprocessor(SnxArmCPU* cpu, u32 word) {
+    auto& self = *reinterpret_cast<Cpu32*>(cpu)->parent;
+    const u32 coproc = (word >> 8) & 0xF;
+    const u32 crm = word & 0xF;
+
+    if (coproc == 15 && (word & 0x0F100010) == 0x0E000010) {
+        // MCR p15
+        const u32 opc1 = (word >> 21) & 7;
+        const u32 crn = (word >> 16) & 0xF;
+        const u32 opc2 = (word >> 5) & 7;
+        if (opc1 == 0 && crn == 7 && crm == 5 && opc2 == 4) {
+            // CP15_FLUSH_PREFETCH_BUFFER
+            return;
+        }
+        if (opc1 == 0 && crn == 7 && crm == 10 && (opc2 == 4 || opc2 == 5)) {
+            // CP15_DATA_SYNC_BARRIER, CP15_DATA_MEMORY_BARRIER
+            std::atomic_thread_fence(std::memory_order_seq_cst);
+            return;
+        }
+    } else if (coproc == 15 && (word & 0x0FF00000) == 0x0C500000) {
+        // MRRC p15
+        const u32 opc1 = (word >> 4) & 0xF;
+        const u32 rt = (word >> 12) & 0xF;
+        const u32 rt2 = (word >> 16) & 0xF;
+        if (opc1 == 0 && crm == 14) {
+            // CNTPCT
+            const u64 ticks = self.m_process->m_system.CoreTiming().GetClockTicks();
+            cpu->r[rt] = u32(ticks);
+            cpu->r[rt2] = u32(ticks >> 32);
+            return;
+        }
+    }
+
+    // TPIDRURW and TPIDRURO, and undefined for anything else.
+    snx_arm_coproc_user(cpu, word);
+    ASSERT(cpu->exit == SNX_ARM_OK && "unhandled coproc");
+}
+
+HaltReason ArmScarletNx32::RunThread(Kernel::KThread* thread) {
+    if (m_interrupt.exchange(false)) {
+        return HaltReason::BreakLoop;
+    }
+
+    // The block cache only covers the code present when it was made.
+    const u64 generation = m_process->CodeGeneration();
+    if (!m_module || generation != m_code_generation) {
+        snx_recompiled32_free(m_module);
+        m_module = m_process->MakeModule32();
+        m_code_generation = generation;
+    }
+
+    LoadCpu();
+    m_in_svc = false;
+    m_fault_scope.last_fault = ~u64{};
+    SetScarletNxFaultScope(&m_fault_scope);
+    if (sigsetjmp(m_fault_scope.env, 1) != 0) {
+        SetScarletNxFaultScope(nullptr);
+        LOG_CRITICAL(Core_ARM, "invalid memory access at {:#x}",
+                     m_fault_scope.fault_address);
+        SaveCpu();
+        return HaltReason::PrefetchAbort;
+    }
+    const int exit =
+        m_module ? snx_recompiled32_run(m_module, &m_cpu.cpu, &Svc, this) : SNX_ARM_BAD_ENTRY;
+    SetScarletNxFaultScope(nullptr);
+
+    SaveCpu();
+    return Halt(exit);
+}
+
+HaltReason ArmScarletNx32::Halt(int exit) {
+    if (m_in_svc) {
+        return HaltReason::SupervisorCall;
+    }
+    u32 const pc = m_cpu.cpu.r[15];
+    u32 const syndrome = m_cpu.cpu.syndrome;
+    switch (u32(exit)) {
+    case InterruptExit:
+        m_interrupt.store(false);
+        return HaltReason::BreakLoop;
+    case SNX_ARM_BREAKPOINT:
+        // As Dynarmic does without a debugger: report it and go on.
+        LOG_CRITICAL(Core_ARM, "BKPT #{:#x} at {:#x}", syndrome, pc);
+        return HaltReason::BreakLoop;
+    default:
+        LOG_CRITICAL(Core_ARM, "A32 exit {} at {:#x} (syndrome {:#x})", exit, pc, syndrome);
+        return HaltReason::PrefetchAbort;
+    }
+}
+
+HaltReason ArmScarletNx32::StepThread(Kernel::KThread* thread) {
+    // Never reached: the debugger selects Dynarmic.
+    LOG_ERROR(Core_ARM, "ScarletNX cannot single step");
+    return HaltReason::BreakLoop;
+}
+
+void ArmScarletNx32::LoadCpu() {
+    auto& cpu = m_cpu.cpu;
+    static_assert(sizeof(cpu.d) <= sizeof(m_ctx.v));
+    for (std::size_t i = 0; i < 16; i++)
+        cpu.r[i] = u32(m_ctx.r[i]);
+    cpu.cpsr = (m_ctx.pstate & ~ModeMask) | UserMode;
+    std::memcpy(cpu.d, m_ctx.v.data(), sizeof(cpu.d));
+    cpu.fpscr = FpsrFpcrToFpscr(m_ctx.fpsr, m_ctx.fpcr);
+    cpu.tpidrurw = u32(m_ctx.tpidr);
+    cpu.tpidruro = m_tpidrro;
+    cpu.exclusive_valid = 0;
+}
+
+void ArmScarletNx32::SaveCpu() {
+    const auto& cpu = m_cpu.cpu;
+    for (std::size_t i = 0; i < 16; i++)
+        m_ctx.r[i] = cpu.r[i];
+    m_ctx.fp = cpu.r[11];
+    m_ctx.sp = cpu.r[13];
+    m_ctx.lr = cpu.r[14];
+    m_ctx.pc = cpu.r[15];
+    m_ctx.pstate = cpu.cpsr;
+    std::memcpy(m_ctx.v.data(), cpu.d, sizeof(cpu.d));
+    const auto [fpsr, fpcr] = FpscrToFpsrFpcr(cpu.fpscr);
+    m_ctx.fpsr = fpsr;
+    m_ctx.fpcr = fpcr;
+    m_ctx.tpidr = cpu.tpidrurw;
+}
+
+void ArmScarletNx32::GetContext(Kernel::Svc::ThreadContext& ctx) const {
+    ctx = m_ctx;
+}
+
+void ArmScarletNx32::SetContext(const Kernel::Svc::ThreadContext& ctx) {
+    m_ctx = ctx;
+}
+
+void ArmScarletNx32::SetTpidrroEl0(u64 value) {
+    m_tpidrro = u32(value);
+}
+
+void ArmScarletNx32::GetSvcArguments(std::span<uint64_t, 8> args) const {
+    for (std::size_t i = 0; i < args.size(); i++)
+        args[i] = u32(m_ctx.r[i]);
+}
+
+void ArmScarletNx32::SetSvcArguments(std::span<const uint64_t, 8> args) {
+    for (std::size_t i = 0; i < args.size(); i++)
+        m_ctx.r[i] = u32(args[i]);
+}
+
+u32 ArmScarletNx32::GetSvcNumber() const {
+    return m_svc;
+}
+
+void ArmScarletNx32::SignalInterrupt(Kernel::KThread* thread) {
+    // The run loop checks the exit between blocks. A block's own exit takes precedence.
+    m_interrupt.store(true);
+    u32 expected = SNX_ARM_OK;
+    __atomic_compare_exchange_n(&m_cpu.cpu.exit, &expected, InterruptExit, false, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED);
+}
+
+void ArmScarletNx32::ClearInstructionCache() {
+    m_process->ClearInstructionCache();
+}
+
+void ArmScarletNx32::InvalidateCacheRange(u64 addr, std::size_t size) {
+    m_process->InvalidateCacheRange(addr, size);
+}
+
+} // namespace Core

src/core/arm/scarletnx/arm_scarletnx_32.h

@@ -0,0 +1,81 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <atomic>
+#include <memory>
+
+#include "core/arm/arm_interface.h"
+#include "core/arm/scarletnx/scarletnx_api.h"
+#include "core/arm/scarletnx/scarletnx_process.h"
+
+namespace Core {
+
+/// Runs A32 and Thumb code through ScarletNX's block compiler. Blocks return to the run loop, so
+/// an SVC ends the run with r15 past it. The block cache is not thread-safe, so each core has its
+/// own, rebuilt when the process' code changes.
+class ArmScarletNx32 final : public ArmInterface {
+public:
+    ArmScarletNx32(std::shared_ptr<ScarletNxProcess> process, std::size_t core_index);
+    ~ArmScarletNx32() override;
+
+    Architecture GetArchitecture() const override {
+        return Architecture::AArch32;
+    }
+
+    HaltReason RunThread(Kernel::KThread* thread) override;
+    HaltReason StepThread(Kernel::KThread* thread) override;
+
+    void GetContext(Kernel::Svc::ThreadContext& ctx) const override;
+    void SetContext(const Kernel::Svc::ThreadContext& ctx) override;
+    void SetTpidrroEl0(u64 value) override;
+
+    void GetSvcArguments(std::span<uint64_t, 8> args) const override;
+    void SetSvcArguments(std::span<const uint64_t, 8> args) override;
+    u32 GetSvcNumber() const override;
+
+    void SignalInterrupt(Kernel::KThread* thread) override;
+    void ClearInstructionCache() override;
+    void InvalidateCacheRange(u64 addr, std::size_t size) override;
+
+protected:
+    const Kernel::DebugWatchpoint* HaltedWatchpoint() const override {
+        return nullptr;
+    }
+
+    void RewindBreakpointInstruction() override {}
+
+private:
+    // The CPU first, so the coprocessor hook can find its interface.
+    struct Cpu32 {
+        SnxArmCPU cpu;
+        ArmScarletNx32* parent;
+    };
+
+    static int Svc(void* context, u32 imm, u64 number, const u64 arguments[6]);
+    static void Coprocessor(SnxArmCPU* cpu, u32 word);
+    HaltReason Halt(int exit);
+    void LoadCpu();
+    void SaveCpu();
+
+    std::shared_ptr<ScarletNxProcess> m_process;
+    std::size_t m_core_index{};
+
+    snx_recompiled32* m_module{};
+    u64 m_code_generation{~u64{}};
+
+    // The context the kernel reads and writes; m_cpu holds it while guest code runs.
+    Kernel::Svc::ThreadContext m_ctx{};
+    u32 m_tpidrro{};
+    Cpu32 m_cpu{};
+    SnxArmRt m_rt{};
+
+    bool m_in_svc{};
+    u32 m_svc{};
+    std::atomic<bool> m_interrupt{};
+
+    ScarletNxFaultScope m_fault_scope{};
+};
+
+} // namespace Core

src/core/arm/scarletnx/arm_scarletnx_64.cpp

@@ -0,0 +1,261 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include <cstring>
+
+#include <sys/mman.h>
+
+#include "common/assert.h"
+#include "common/literals.h"
+#include "common/logging.h"
+#include "core/arm/scarletnx/arm_scarletnx_64.h"
+#include "core/hle/kernel/k_thread.h"
+#include "core/hle/kernel/k_process.h"
+
+namespace Core {
+
+namespace {
+
+using namespace Common::Literals;
+namespace ctx = boost::context::detail;
+
+// Reserved, not committed: guest calls nest on it, and recompiling a region runs on it too.
+constexpr std::size_t StackSize = 32_MiB;
+constexpr std::size_t GuardSize = 64_KiB;
+
+bool SameContext(const Kernel::Svc::ThreadContext a, const Kernel::Svc::ThreadContext b, bool in_svc) {
+    // An SVC returns its results in x0-x7.
+    for (std::size_t i = in_svc ? 8 : 0; i < a.r.size(); i++)
+        if (a.r[i] != b.r[i])
+            return false;
+    return a.fp == b.fp && a.lr == b.lr && a.sp == b.sp && a.pc == b.pc
+        && (a.pstate >> 28) == (b.pstate >> 28) && a.v == b.v && a.fpcr == b.fpcr
+        && a.fpsr == b.fpsr && a.tpidr == b.tpidr;
+}
+
+} // namespace
+
+const RtAPI* ArmScarletNx64::Runtime() {
+    // this incurs a cost each time we call Runtime() -- but we only ever call it once (ideally)
+    static const RtAPI rt = [] {
+        RtAPI api;
+        snx_runtime_defaults(&api);
+        api.svc = &Svc;
+        api.undef = &Undefined;
+        api.brk = &Breakpoint;
+        api.poll = &Poll;
+        api.miss = &Miss;
+        api.code_changed = &CodeChanged;
+        api.ticks = &ScarletNxTicks;
+        return api;
+    }();
+    return &rt;
+}
+
+ArmScarletNx64::ArmScarletNx64(std::shared_ptr<ScarletNxProcess> process, std::size_t core_index)
+    : ArmInterface{true}
+    , m_process{std::move(process)}, m_core_index{core_index}
+{
+    // can't use virtualbuffer...
+    void* const stack = mmap(nullptr, StackSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+    ASSERT_MSG(stack != MAP_FAILED, "cannot allocate a guest stack");
+    m_stack = static_cast<u8*>(stack);
+    mprotect(m_stack, GuardSize, PROT_NONE);
+
+    m_cpu.mem = m_process->m_arena;
+    m_cpu.rt = Runtime();
+    m_cpu.thread = this;
+
+    m_fault_scope.memory = &m_process->m_process.GetMemory();
+    m_fault_scope.arena = m_process->m_arena;
+    m_fault_scope.arena_size = ScarletNxProcess::ArenaSize;
+}
+
+ArmScarletNx64::~ArmScarletNx64() {
+    munmap(m_stack, StackSize);
+}
+
+void ArmScarletNx64::Svc(CPU* cpu, u32 imm) {
+    auto& self = From(cpu);
+    self.m_svc = imm;
+    self.Yield(HaltReason::SupervisorCall, true, true);
+}
+
+void ArmScarletNx64::Undefined(CPU* cpu, u64 pc, u32 word) {
+    // A reserved encoding rejected from inside an instruction, which then goes on.
+    if (cpu->pc != pc) {
+        LOG_ERROR(Core_ARM, "reserved encoding {:08X} at {:#x}", word, pc);
+        return;
+    }
+    LOG_CRITICAL(Core_ARM, "cannot execute {:08X} at {:#x}", word, pc);
+    From(cpu).Yield(HaltReason::PrefetchAbort, false, false);
+    UNREACHABLE();
+}
+
+void ArmScarletNx64::Breakpoint(CPU* cpu, u64 pc, u32 imm) {
+    // As Dynarmic does without a debugger: report it and go on.
+    LOG_CRITICAL(Core_ARM, "BRK #{:#x} at {:#x}", imm, pc);
+    cpu->pc = pc + 4;
+}
+
+void ArmScarletNx64::Poll(CPU* cpu) {
+    auto& self = From(cpu);
+    __atomic_store_n(&cpu->poll, 0, __ATOMIC_RELAXED);
+    if (self.m_interrupt.exchange(false))
+        self.Yield(HaltReason::BreakLoop, true, false);
+}
+
+void ArmScarletNx64::Miss(CPU* cpu, u64 pc) {
+    auto& process = *From(cpu).m_process;
+    const auto function = process.Lookup(pc);
+    if (!function) {
+        cpu->pc = pc;
+        Undefined(cpu, pc, 0);
+        return;
+    }
+    // The first region was recompiled by this lookup: dispatch through its table from now on.
+    if (snx_recompiled* const module = cpu->disp ? nullptr : process.FindModule(pc))
+        snx_recompiled_bind(module, cpu);
+    function(cpu, pc);
+}
+
+void ArmScarletNx64::CodeChanged(CPU* cpu, u64 addr) {
+    constexpr u64 ICacheLineSize = 64;
+    From(cpu).m_process->InvalidateCacheRange(addr & ~(ICacheLineSize - 1), ICacheLineSize);
+}
+
+void ArmScarletNx64::Entry(ctx::transfer_t transfer) {
+    auto* const self = static_cast<ArmScarletNx64*>(transfer.data);
+    self->m_host = transfer.fctx;
+    self->Execute();
+}
+
+void ArmScarletNx64::Execute() {
+    if (sigsetjmp(m_fault_scope.env, 1) != 0) {
+        LOG_CRITICAL(Core_ARM, "invalid memory access at {:#x}", m_fault_scope.fault_address);
+        Yield(HaltReason::PrefetchAbort, false, false);
+        UNREACHABLE();
+    }
+
+    // Every guest return to below the first frame comes back here.
+    while (true)
+        snx_lookup(&m_cpu, m_cpu.pc)(&m_cpu, m_cpu.pc);
+}
+
+void ArmScarletNx64::Yield(HaltReason reason, bool resumable, bool in_svc) {
+    m_halt = reason;
+    m_resumable = resumable;
+    m_suspended_in_svc = in_svc;
+    m_host = ctx::jump_fcontext(m_host, nullptr).fctx;
+}
+
+void ArmScarletNx64::ResetGuest() {
+    m_guest = ctx::make_fcontext(m_stack + StackSize, StackSize - GuardSize, &Entry);
+    if (snx_recompiled* const module = m_process->FindModule(m_cpu.pc))
+        snx_recompiled_bind(module, &m_cpu);
+}
+
+HaltReason ArmScarletNx64::RunThread(Kernel::KThread* thread) {
+    if (m_interrupt.exchange(false))
+        return HaltReason::BreakLoop;
+
+    bool const resume = m_resumable && m_suspended_thread_id == thread->GetThreadId() && SameContext(m_ctx, m_suspended_ctx, m_suspended_in_svc);
+    LoadCpu();
+    if (!resume)
+        ResetGuest();
+
+    m_resumable = false;
+    m_fault_scope.last_fault = ~u64{};
+    SetScarletNxFaultScope(&m_fault_scope);
+    m_guest = ctx::jump_fcontext(m_guest, this).fctx;
+    SetScarletNxFaultScope(nullptr);
+
+    SaveCpu();
+    if (m_resumable) {
+        m_suspended_thread_id = thread->GetThreadId();
+        m_suspended_ctx = m_ctx;
+    }
+    return m_halt;
+}
+
+HaltReason ArmScarletNx64::StepThread(Kernel::KThread* thread) {
+    // Never reached: the debugger selects Dynarmic.
+    LOG_ERROR(Core_ARM, "ScarletNX cannot single step");
+    return HaltReason::BreakLoop;
+}
+
+void ArmScarletNx64::LoadCpu() {
+    static_assert(sizeof(m_cpu.v) == sizeof(m_ctx.v));
+    for (std::size_t i = 0; i < m_ctx.r.size(); i++)
+        m_cpu.x[i] = m_ctx.r[i];
+    m_cpu.x[29] = m_ctx.fp;
+    m_cpu.x[30] = m_ctx.lr;
+    m_cpu.x[31] = m_ctx.sp;
+    m_cpu.pc = m_ctx.pc;
+    set_nzcv(&m_cpu, m_ctx.pstate >> 28);
+    std::memcpy(m_cpu.v, m_ctx.v.data(), sizeof(m_cpu.v));
+    m_cpu.fpcr = m_ctx.fpcr;
+    m_cpu.fpsr = m_ctx.fpsr;
+    m_cpu.tpidr = m_ctx.tpidr;
+    m_cpu.tpidrro = m_tpidrro;
+    m_cpu.excl_on = 0;
+}
+
+void ArmScarletNx64::SaveCpu() {
+    for (std::size_t i = 0; i < m_ctx.r.size(); i++) {
+        m_ctx.r[i] = m_cpu.x[i];
+    }
+    m_ctx.fp = m_cpu.x[29];
+    m_ctx.lr = m_cpu.x[30];
+    m_ctx.sp = m_cpu.x[31];
+    m_ctx.pc = m_cpu.pc;
+    m_ctx.pstate = get_nzcv(&m_cpu);
+    std::memcpy(m_ctx.v.data(), m_cpu.v, sizeof(m_cpu.v));
+    m_ctx.fpcr = m_cpu.fpcr;
+    m_ctx.fpsr = m_cpu.fpsr;
+    m_ctx.tpidr = m_cpu.tpidr;
+}
+
+void ArmScarletNx64::GetContext(Kernel::Svc::ThreadContext& ctx) const {
+    ctx = m_ctx;
+}
+
+void ArmScarletNx64::SetContext(const Kernel::Svc::ThreadContext& ctx) {
+    m_ctx = ctx;
+}
+
+void ArmScarletNx64::SetTpidrroEl0(u64 value) {
+    m_tpidrro = value;
+}
+
+void ArmScarletNx64::GetSvcArguments(std::span<uint64_t, 8> args) const {
+    for (std::size_t i = 0; i < args.size(); i++) {
+        args[i] = m_ctx.r[i];
+    }
+}
+
+void ArmScarletNx64::SetSvcArguments(std::span<const uint64_t, 8> args) {
+    for (std::size_t i = 0; i < args.size(); i++) {
+        m_ctx.r[i] = args[i];
+    }
+}
+
+u32 ArmScarletNx64::GetSvcNumber() const {
+    return m_svc;
+}
+
+void ArmScarletNx64::SignalInterrupt(Kernel::KThread* thread) {
+    // Generated code reads poll on backward branches and calls Poll, which yields.
+    m_interrupt.store(true);
+    __atomic_store_n(&m_cpu.poll, 1, __ATOMIC_RELAXED);
+}
+
+void ArmScarletNx64::ClearInstructionCache() {
+    m_process->ClearInstructionCache();
+}
+
+void ArmScarletNx64::InvalidateCacheRange(u64 addr, std::size_t size) {
+    m_process->InvalidateCacheRange(addr, size);
+}
+
+} // namespace Core

src/core/arm/scarletnx/arm_scarletnx_64.h

@@ -0,0 +1,100 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <atomic>
+#include <memory>
+
+#include <boost/context/detail/fcontext.hpp>
+
+#include "core/arm/arm_interface.h"
+#include "core/arm/scarletnx/scarletnx_api.h"
+#include "core/arm/scarletnx/scarletnx_process.h"
+
+namespace Core {
+
+/// Runs A64 code recompiled by ScarletNX.
+///
+/// A recompiled guest function is a host function that returns when the guest function returns,
+/// so guest calls nest on the host stack and an SVC cannot return from the generated code. Guest
+/// code therefore runs on a coroutine of this core, which the SVC and interrupt hooks suspend.
+/// When the kernel next runs the same thread with the same context, the coroutine resumes;
+/// otherwise its host frames are dropped and execution restarts at the new pc through dispatch.
+class ArmScarletNx64 final : public ArmInterface {
+public:
+    ArmScarletNx64(std::shared_ptr<ScarletNxProcess> process, std::size_t core_index);
+    ~ArmScarletNx64() override;
+
+    Architecture GetArchitecture() const override {
+        return Architecture::AArch64;
+    }
+
+    HaltReason RunThread(Kernel::KThread* thread) override;
+    HaltReason StepThread(Kernel::KThread* thread) override;
+
+    void GetContext(Kernel::Svc::ThreadContext& ctx) const override;
+    void SetContext(const Kernel::Svc::ThreadContext& ctx) override;
+    void SetTpidrroEl0(u64 value) override;
+
+    void GetSvcArguments(std::span<uint64_t, 8> args) const override;
+    void SetSvcArguments(std::span<const uint64_t, 8> args) override;
+    u32 GetSvcNumber() const override;
+
+    void SignalInterrupt(Kernel::KThread* thread) override;
+    void ClearInstructionCache() override;
+    void InvalidateCacheRange(u64 addr, std::size_t size) override;
+
+protected:
+    const Kernel::DebugWatchpoint* HaltedWatchpoint() const override {
+        return nullptr;
+    }
+
+    void RewindBreakpointInstruction() override {}
+
+private:
+    static ArmScarletNx64& From(CPU* cpu) {
+        return *static_cast<ArmScarletNx64*>(cpu->thread);
+    }
+    static const RtAPI* Runtime();
+
+    // Runtime hooks called by the generated code.
+    static void Svc(CPU* cpu, u32 imm);
+    static void Undefined(CPU* cpu, u64 pc, u32 word);
+    static void Breakpoint(CPU* cpu, u64 pc, u32 imm);
+    static void Poll(CPU* cpu);
+    static void Miss(CPU* cpu, u64 pc);
+    static void CodeChanged(CPU* cpu, u64 addr);
+
+    static void Entry(boost::context::detail::transfer_t transfer);
+    [[noreturn]] void Execute();
+    void Yield(HaltReason reason, bool resumable, bool in_svc);
+    void ResetGuest();
+    void LoadCpu();
+    void SaveCpu();
+
+    std::shared_ptr<ScarletNxProcess> m_process;
+    std::size_t m_core_index{};
+
+    // The context the kernel reads and writes; m_cpu holds it while guest code runs.
+    Kernel::Svc::ThreadContext m_ctx{};
+    u64 m_tpidrro{};
+    CPU m_cpu{};
+
+    HaltReason m_halt{};
+    u32 m_svc{};
+    std::atomic<bool> m_interrupt{};
+
+    // The coroutine, and what it was suspended in, when it can be resumed.
+    boost::context::detail::fcontext_t m_guest{};
+    boost::context::detail::fcontext_t m_host{};
+    u8* m_stack{};
+    bool m_resumable{};
+    bool m_suspended_in_svc{};
+    u64 m_suspended_thread_id{};
+    Kernel::Svc::ThreadContext m_suspended_ctx{};
+
+    ScarletNxFaultScope m_fault_scope{};
+};
+
+} // namespace Core

src/core/arm/scarletnx/scarletnx_api.h

@@ -0,0 +1,10 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+// snx_recompiler.h includes cpu.h and arm.h before its own extern "C" block, so the whole
+// header needs C linkage for the runtime library's functions to resolve.
+extern "C" {
+#include <scarletnx/snx_recompiler.h>
+}

src/core/arm/scarletnx/scarletnx_process.cpp

@@ -0,0 +1,300 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include <algorithm>
+#include <mutex>
+
+#include <signal.h>
+
+#include "common/fs/path_util.h"
+#include "common/logging.h"
+#include "common/settings.h"
+#include "core/arm/scarletnx/arm_scarletnx_32.h"
+#include "core/arm/scarletnx/arm_scarletnx_64.h"
+#include "core/arm/scarletnx/scarletnx_api.h"
+#include "core/arm/scarletnx/scarletnx_process.h"
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/device_memory.h"
+#include "core/hle/kernel/k_process.h"
+#include "core/hle/kernel/kernel.h"
+#include "core/memory.h"
+
+namespace Core {
+
+namespace {
+
+// ScarletNX refuses dispatch tables spanning more than 1 GiB of guest code.
+constexpr u64 MaxModuleSpan = 1ULL << 30;
+constexpr u64 A32Limit = 1ULL << 32;
+
+std::atomic<Timing::CoreTiming*> g_core_timing{};
+
+// Needs be TLS because SNX expects like, more than 4 threads (host == guest threads)
+// and well -- quite impervious to "emulated" threads... maybe one day this can be removed
+thread_local ScarletNxFaultScope* g_fault_scope{};
+static struct {
+    struct sigaction old_sigsegv;
+    struct sigaction old_sigbus;
+    std::once_flag installed;
+} g_snx_data;
+
+bool Overlaps(const ScarletNxProcess::CodeRange& range, u64 addr, u64 size) {
+    return addr < range.addr + range.size && range.addr < addr + size;
+}
+
+void ChainSignal(int sig, siginfo_t* info, void* raw_context) {
+    const struct sigaction& old = sig == SIGSEGV ? g_snx_data.old_sigsegv : g_snx_data.old_sigbus;
+    if (old.sa_flags & SA_SIGINFO) {
+        old.sa_sigaction(sig, info, raw_context);
+    } else if (old.sa_handler == SIG_DFL) {
+        // The faulting instruction runs again and takes the default action.
+        signal(sig, SIG_DFL);
+    } else if (old.sa_handler != SIG_IGN) {
+        old.sa_handler(sig);
+    }
+}
+
+void FaultHandler(int sig, siginfo_t* info, void* raw_context) {
+    ScarletNxFaultScope* const scope = g_fault_scope;
+    u8* const addr = static_cast<u8*>(info->si_addr);
+    if (!scope || addr < scope->arena || addr >= scope->arena + scope->arena_size) {
+        ChainSignal(sig, info, raw_context);
+        return;
+    }
+
+    // Pages the rasterizer protects fault on access: invalidate them and retry. A second fault at
+    // the same addr means retrying will not help.
+    const u64 guest_address = u64(addr - scope->arena);
+    if (guest_address != scope->last_fault) {
+        scope->last_fault = guest_address;
+        if (scope->memory->InvalidateNCE(guest_address & ~Memory::YUZU_PAGEMASK, Memory::YUZU_PAGESIZE)) {
+            return;
+        }
+    }
+    scope->fault_address = guest_address;
+    siglongjmp(scope->env, 1);
+}
+
+} // namespace
+
+void InstallScarletNxFaultHandler() {
+    std::call_once(g_snx_data.installed, [] {
+        struct sigaction action {};
+        action.sa_flags = SA_SIGINFO;
+        action.sa_sigaction = &FaultHandler;
+        sigemptyset(&action.sa_mask);
+        sigaction(SIGSEGV, &action, &g_snx_data.old_sigsegv);
+        sigaction(SIGBUS, &action, &g_snx_data.old_sigbus);
+    });
+}
+
+void SetScarletNxFaultScope(ScarletNxFaultScope* scope) {
+    g_fault_scope = scope;
+}
+
+u64 ScarletNxTicks() {
+    return g_core_timing.load(std::memory_order_relaxed)->GetClockTicks();
+}
+
+std::unique_ptr<ArmInterface> MakeScarletNxInterface(std::shared_ptr<ScarletNxProcess> process, bool is_64bit, std::size_t core_index) {
+    if (is_64bit)
+        return std::make_unique<ArmScarletNx64>(std::move(process), core_index);
+    return std::make_unique<ArmScarletNx32>(std::move(process), core_index);
+}
+
+ScarletNxProcess::ScarletNxProcess(System& system, Kernel::KProcess& process)
+    : m_system{system}
+    , m_process{process}
+    , m_arena{system.DeviceMemory().buffer.VirtualBasePointer()}
+{
+    InstallScarletNxFaultHandler();
+    g_core_timing.store(&system.CoreTiming(), std::memory_order_relaxed);
+
+    m_recompiler = snx_recompiler_new(SNX_RECOMPILER_FLAG_LLVM);
+    ASSERT(m_recompiler && "cannot create the recompiler");
+
+    // Linked modules are kept by a hash of their code, so later launches skip the recompile.
+    const auto cache = Common::FS::GetEdenPath(Common::FS::EdenPath::CacheDir) / "scarletnx";
+    if (snx_recompiler_set_cache(m_recompiler, Common::FS::PathToUTF8String(cache).c_str()) != 0)
+        LOG_WARNING(Core_ARM, "no code cache: {}", snx_recompiler_error(m_recompiler));
+}
+
+ScarletNxProcess::~ScarletNxProcess() {
+    for (auto& mod : m_modules)
+        snx_recompiled_free(mod.mod);
+    snx_recompiler_free(m_recompiler);
+}
+
+bool ScarletNxProcess::ShouldUse(System& system, const Kernel::KProcess& process) {
+    if (Settings::values.cpu_backend.GetValue() != Settings::CpuBackend::ScarletNx)
+        return false;
+    auto const f = [](auto& s) {
+        LOG_WARNING(Core_ARM, "snx: {}", s);
+        return false;
+    };
+    if (!process.IsApplication())
+        return f("no qlaunch yet");
+    if (!system.Kernel().IsMulticore())
+        return f("single core mode needs cycle counting");
+    if (system.DebuggerEnabled())
+        return f("the debugger needs stepping and watchpoints");
+    /// ScarletNX expects flat memory, so it will only work with fastmem
+    if (!Settings::IsFastmemEnabled() || !system.DeviceMemory().buffer.VirtualBasePointer())
+        return f("guest memory must be mapped by fastmem");
+    if (snx_recompiler_abi() != SNX_ABI)
+        return f("the library was built with a different ABI");
+    if (!snx_recompiler_has(SNX_RECOMPILER_FLAG_LLVM))
+        return f("LLVM cannot target this host");
+    return true;
+}
+
+std::vector<ScarletNxProcess::CodeRange> ScarletNxProcess::GetCodeRanges() const {
+    std::vector<CodeRange> r;
+    const auto& page_table = m_process.GetPageTable().GetBasePageTable();
+    u64 addr = GetInteger(page_table.GetAddressSpaceStart());
+    while (true) {
+        Kernel::KMemoryInfo info;
+        Kernel::Svc::PageInfo page_info;
+        if (page_table.QueryInfo(&info, &page_info, addr).IsError())
+            break;
+        if (True(info.GetPermission() & Kernel::KMemoryPermission::UserExecute)) {
+            if (!r.empty() && r.back().addr + r.back().size == info.GetAddress()) {
+                r.back().size += info.GetSize();
+            } else {
+                r.push_back({info.GetAddress(), info.GetSize()});
+            }
+        }
+        // The block past the addr space ends at 0.
+        if (info.GetEndAddress() <= addr)
+            break;
+        addr = info.GetEndAddress();
+    }
+    return r;
+}
+
+const ScarletNxProcess::Module* ScarletNxProcess::FindModuleLocked(u64 pc) const {
+    for (const auto& mod : m_modules) {
+        if (mod.retired || pc < mod.lo || pc >= mod.hi)
+            continue;
+        for (const auto& range : mod.ranges)
+            if (pc >= range.addr && pc < range.addr + range.size)
+                return &mod;
+    }
+    return nullptr;
+}
+
+snx_recompiled* ScarletNxProcess::RecompileRegionLocked(u64 pc) {
+    const auto r = GetCodeRanges();
+    const auto it = std::ranges::find_if(r, [pc](const CodeRange& range) {
+        return pc >= range.addr && pc < range.addr + range.size;
+    });
+    if (it == r.end())
+        return nullptr;
+
+    const auto is_covered = [this](const CodeRange& range) {
+        return std::ranges::any_of(m_modules, [&](const Module& mod) {
+            return !mod.retired && std::ranges::any_of(mod.ranges, [&](const CodeRange& cr) {
+                return Overlaps(cr, range.addr, range.size);
+            });
+        });
+    };
+
+    // Recompile as many neighbouring regions as one dispatch table can cover.
+    auto first = size_t(it - r.begin());
+    auto last = first;
+    const auto end_of = [&](size_t i) { return r[i].addr + r[i].size; };
+    while (first > 0 && !is_covered(r[first - 1]) && end_of(last) - r[first - 1].addr <= MaxModuleSpan)
+        --first;
+    while (last + 1 < r.size() && !is_covered(r[last + 1])
+    && end_of(last + 1) - r[first].addr <= MaxModuleSpan)
+        ++last;
+
+    Module mod{};
+    mod.ranges.assign(r.begin() + first, r.begin() + last + 1);
+    mod.lo = r[first].addr;
+    mod.hi = end_of(last);
+
+    std::vector<snx_code_range> code;
+    u64 code_size = 0;
+    for (const auto& range : mod.ranges) {
+        code.push_back({range.addr, m_arena + range.addr, range.size});
+        code_size += range.size;
+    }
+
+    LOG_INFO(Core_ARM, "recompiling {} KiB of code at {:#x}-{:#x}", code_size / 1024, mod.lo, mod.hi);
+    mod.mod = snx_recompile_ranges(m_recompiler, code.data(), code.size(), &pc, 1);
+    if (mod.mod) {
+        LOG_INFO(Core_ARM, "recompiled {} functions", snx_recompiled_functions(mod.mod));
+    } else {
+        // Kept, so the region is not recompiled again on every miss.
+        LOG_ERROR(Core_ARM, "recompilation failed: {}", snx_recompiler_error(m_recompiler));
+    }
+
+    m_modules.push_back(std::move(mod));
+    return m_modules.back().mod;
+}
+
+ScarletNxProcess::SnxGuestFn ScarletNxProcess::Lookup(u64 pc) {
+    snx_recompiled* mod;
+    {
+        std::scoped_lock lk{m_mutex};
+        const Module* const known = FindModuleLocked(pc);
+        mod = known ? known->mod : RecompileRegionLocked(pc);
+    }
+    if (!mod)
+        return nullptr;
+    // Waits for the mod's JIT when discovery missed the code at pc.
+    return snx_recompiled_lookup(mod, pc);
+}
+
+snx_recompiled* ScarletNxProcess::FindModule(u64 pc) {
+    std::scoped_lock lk{m_mutex};
+    const Module* const mod = FindModuleLocked(pc);
+    return mod ? mod->mod : nullptr;
+}
+
+snx_recompiled32* ScarletNxProcess::MakeModule32() {
+    std::scoped_lock lk{m_mutex};
+    std::vector<snx_code_range> code;
+    for (const auto& range : GetCodeRanges()) {
+        if (range.addr >= A32Limit) {
+            continue;
+        }
+        code.push_back({range.addr, nullptr, std::min(range.size, A32Limit - range.addr)});
+    }
+    if (code.empty())
+        return nullptr;
+    snx_recompiled32* const mod = snx_recompile32_ranges(m_recompiler, code.data(), code.size());
+    if (!mod)
+        LOG_ERROR(Core_ARM, "cannot prepare A32 code: {}", snx_recompiler_error(m_recompiler));
+    return mod;
+}
+
+void ScarletNxProcess::InvalidateCacheRange(u64 addr, u64 size) {
+    std::scoped_lock lk{m_mutex};
+    for (auto& mod : m_modules) {
+        if (mod.retired || !std::ranges::any_of(mod.ranges, [&](const CodeRange& range) {
+            return Overlaps(range, addr, size);
+        }))
+            continue;
+        // Running code diverts out of the retired functions, and the region is recompiled from
+        // the new code when it is next reached.
+        if (mod.mod)
+            snx_recompiled_invalidate(mod.mod, addr, size, nullptr, nullptr);
+        mod.retired = true;
+    }
+    m_code_generation.fetch_add(1, std::memory_order_release);
+}
+
+void ScarletNxProcess::ClearInstructionCache() {
+    std::scoped_lock lk{m_mutex};
+    for (auto& mod : m_modules) {
+        if (!mod.retired && mod.mod)
+            snx_recompiled_invalidate(mod.mod, mod.lo, mod.hi - mod.lo, nullptr, nullptr);
+        mod.retired = true;
+    }
+    m_code_generation.fetch_add(1, std::memory_order_release);
+}
+
+} // namespace Core

src/core/arm/scarletnx/scarletnx_process.h

@@ -0,0 +1,104 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <atomic>
+#include <csetjmp>
+#include <memory>
+#include <mutex>
+#include <vector>
+
+#include "common/common_types.h"
+
+struct CPU;
+struct snx_recompiler;
+struct snx_recompiled;
+struct snx_recompiled32;
+
+namespace Kernel {
+class KProcess;
+}
+
+namespace Core::Memory {
+class Memory;
+}
+
+namespace Core {
+
+class ArmInterface;
+class System;
+
+class ScarletNxProcess {
+public:
+    using SnxGuestFn = void (*)(CPU*, u64);
+
+    // DeviceMemory reserves 39 bits of address space for the fastmem arena.
+    static constexpr u64 ArenaSize = 1ULL << 39;
+
+    struct CodeRange {
+        u64 addr;
+        u64 size;
+    };
+
+    ScarletNxProcess(System& system, Kernel::KProcess& process);
+    ~ScarletNxProcess();
+
+    static bool ShouldUse(System& system, const Kernel::KProcess& process);
+    SnxGuestFn Lookup(u64 pc);
+    snx_recompiled* FindModule(u64 pc);
+    snx_recompiled32* MakeModule32();
+
+    void InvalidateCacheRange(u64 addr, u64 size);
+    void ClearInstructionCache();
+
+    /// @brief general gate for when we are about to invalidate a given block aka. trigger regen
+    u64 CodeGeneration() const {
+        return m_code_generation.load(std::memory_order_acquire);
+    }
+
+    struct Module {
+        std::vector<CodeRange> ranges;
+        u64 lo;
+        u64 hi;
+        snx_recompiled* mod;
+        bool retired;
+    };
+
+    std::vector<CodeRange> GetCodeRanges() const;
+    const Module* FindModuleLocked(u64 pc) const;
+    snx_recompiled* RecompileRegionLocked(u64 pc);
+
+    std::vector<Module> m_modules;
+    std::atomic<u64> m_code_generation{};
+
+    // TODO: remove parent ref?
+    System& m_system;
+    Kernel::KProcess& m_process;
+    u8* m_arena{};
+    snx_recompiler* m_recompiler{};
+
+    // ScarletNX's recompiler is used by one thread at a time.
+    std::mutex m_mutex;
+};
+
+/// Recovers from host faults in guest memory on the thread running guest code: a fault on a
+/// page the rasterizer protects is resolved and retried, any other jumps to env.
+struct ScarletNxFaultScope {
+    sigjmp_buf env;
+    Memory::Memory* memory{};
+    u8* arena{};
+    u64 arena_size{};
+    u64 last_fault{~0ull};
+    u64 fault_address{};
+};
+
+void InstallScarletNxFaultHandler();
+void SetScarletNxFaultScope(ScarletNxFaultScope* scope);
+
+/// CNTPCT_EL0 of the emulated system.
+u64 ScarletNxTicks();
+
+std::unique_ptr<ArmInterface> MakeScarletNxInterface(std::shared_ptr<ScarletNxProcess> process, bool is_64bit, std::size_t core_index);
+
+} // namespace Core

src/core/core.cpp

@@ -4,6 +4,7 @@
 #include <array>
 #include <atomic>
 #include <memory>
+#include <unordered_map>
 #include <utility>
 
 #include "game_settings.h"
@@ -248,12 +249,30 @@ struct System::Impl {
         }
     }
 
-    void SetNVDECActive(bool is_nvdec_active) {
+    void NotifyNVDECChannelOpen(u64 process_id) {
-        nvdec_active = is_nvdec_active;
+        std::scoped_lock lock{nvdec_active_mutex};
+        ++nvdec_active_channels[process_id];
+    }
+
+    void NotifyNVDECChannelClose(u64 process_id) {
+        std::scoped_lock lock{nvdec_active_mutex};
+        const auto it = nvdec_active_channels.find(process_id);
+        if (it == nvdec_active_channels.end()) {
+            return;
+        }
+        if (--it->second == 0) {
+            nvdec_active_channels.erase(it);
+        }
     }
 
     bool GetNVDECActive() {
-        return nvdec_active;
+        std::scoped_lock lock{nvdec_active_mutex};
+        return !nvdec_active_channels.empty();
+    }
+
+    bool IsNVDECActiveForProcess(u64 process_id) {
+        std::scoped_lock lock{nvdec_active_mutex};
+        return nvdec_active_channels.contains(process_id);
     }
 
     void InitializeDebugger(System& system, u16 port) {
@@ -509,6 +528,8 @@ struct System::Impl {
 
     mutable std::mutex suspend_guard;
     std::mutex general_channel_mutex;
+    std::mutex nvdec_active_mutex;
+    std::unordered_map<u64, u32> nvdec_active_channels;
     std::atomic_bool is_paused{};
     std::atomic_bool is_shutting_down{};
     std::atomic_bool is_powered_on{};
@@ -516,7 +537,6 @@ struct System::Impl {
     bool extended_memory_layout : 1 = false;
     bool exit_locked : 1 = false;
     bool exit_requested : 1 = false;
-    bool nvdec_active : 1 = false;
 
     void EnsureGeneralChannelInitialized(System& system) {
         if (!general_channel_event) {
@@ -580,14 +600,22 @@ void System::UnstallApplication() {
     impl->UnstallApplication();
 }
 
-void System::SetNVDECActive(bool is_nvdec_active) {
+void System::NotifyNVDECChannelOpen(u64 process_id) {
-    impl->SetNVDECActive(is_nvdec_active);
+    impl->NotifyNVDECChannelOpen(process_id);
+}
+
+void System::NotifyNVDECChannelClose(u64 process_id) {
+    impl->NotifyNVDECChannelClose(process_id);
 }
 
 bool System::GetNVDECActive() {
     return impl->GetNVDECActive();
 }
 
+bool System::IsNVDECActiveForProcess(u64 process_id) {
+    return impl->IsNVDECActiveForProcess(process_id);
+}
+
 void System::InitializeDebugger() {
     impl->InitializeDebugger(*this, Settings::values.gdbstub_port.GetValue());
 }

src/core/core.h

@@ -191,8 +191,10 @@ public:
     std::unique_lock<std::mutex> StallApplication();
     void UnstallApplication();
 
-    void SetNVDECActive(bool is_nvdec_active);
+    void NotifyNVDECChannelOpen(u64 process_id);
+    void NotifyNVDECChannelClose(u64 process_id);
     [[nodiscard]] bool GetNVDECActive();
+    [[nodiscard]] bool IsNVDECActiveForProcess(u64 process_id);
 
     /**
      * Initialize the debugger.

src/core/crypto/ncz_decompressor.cpp

@@ -1,430 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-#include <algorithm>
-#include <array>
-#include <cstring>
-#include <memory>
-#include <mutex>
-#include <optional>
-#include <span>
-#include <string>
-#include <vector>
-
-#include <zstd.h>
-
-#include "common/common_types.h"
-#include "common/logging.h"
-#include "common/zstd_compression.h"
-#include "core/crypto/aes_util.h"
-#include "core/crypto/key_manager.h"
-#include "core/crypto/ncz_decompressor.h"
-#include "core/file_sys/vfs/vfs.h"
-
-namespace Core::Crypto {
-
-namespace {
-
-constexpr u64 HeaderSize = 0x4000;
-constexpr std::size_t SectionWire = 0x40;
-constexpr std::size_t BlockFixed = 0x18;
-constexpr std::size_t Chunk = 0x100000;
-// Upper bound on NCZ section entries (64 bytes each, so the table is at most 4 MiB). A base NCA
-// has a handful of sections, but nsz stores an update's patch (BKTR) data one entry per
-// subsection, which is thousands of them.
-constexpr u64 MaxSectionCount = 0x10000;
-constexpr std::array<u8, 8> SectionsMagic{'N', 'C', 'Z', 'S', 'E', 'C', 'T', 'N'};
-constexpr std::array<u8, 8> BlockMagic{'N', 'C', 'Z', 'B', 'L', 'O', 'C', 'K'};
-
-struct Section {
-    s64 offset, size;
-    u64 crypto_type;
-    Key128 key, counter;
-    bool Ctr() const { return crypto_type == 3 || crypto_type == 4; }
-};
-
-struct BlockHeader {
-    u32 exponent;
-    u64 decompressed_size;
-    std::vector<u32> sizes;
-};
-
-u64 Le(const u8* p, std::size_t n) {
-    u64 v = 0;
-    for (std::size_t i = n; i-- > 0;) v = (v << 8) | p[i];
-    return v;
-}
-
-// Mirrors nsz AESCTR.seek(): upper 8 bytes = stored counter, lower 8 = absolute NCA offset / 16.
-void SetIv(Key128& iv, const Key128& base, u64 off) {
-    iv = base;
-    u64 blk = off >> 4;
-    for (std::size_t i = 0; i < 8; ++i) { iv[15 - i] = static_cast<u8>(blk & 0xFF); blk >>= 8; }
-}
-
-// Parse header + section table + optional block header. Returns payload_start.
-std::optional<std::pair<std::vector<Section>, std::optional<BlockHeader>>> ParseHeader(
-    const FileSys::VirtualFile& f, u64& payload_start) {
-    const u64 fsize = f->GetSize();
-    if (fsize < HeaderSize + 16) return std::nullopt;
-
-    const auto head = f->ReadBytes(16, HeaderSize);
-    if (head.size() != 16 || !std::equal(head.begin(), head.begin() + 8, SectionsMagic.begin()))
-        return std::nullopt;
-    const u64 count = Le(head.data() + 8, 8);
-    if (count == 0 || count > MaxSectionCount || count * SectionWire > fsize - HeaderSize - 16)
-        return std::nullopt;
-
-    const auto table = f->ReadBytes(count * SectionWire, HeaderSize + 16);
-    if (table.size() != count * SectionWire) return std::nullopt;
-
-    std::vector<Section> sections;
-    sections.reserve(count + 1);
-    for (u64 i = 0; i < count; ++i) {
-        const u8* e = table.data() + i * SectionWire;
-        Section s{};
-        s.offset = static_cast<s64>(Le(e, 8));
-        s.size = static_cast<s64>(Le(e + 8, 8));
-        s.crypto_type = Le(e + 16, 8);
-        std::memcpy(s.key.data(), e + 32, 16);
-        std::memcpy(s.counter.data(), e + 48, 16);
-        sections.push_back(s);
-    }
-
-    const u64 block_start = HeaderSize + 16 + count * SectionWire;
-    const auto fixed = f->ReadBytes(BlockFixed, block_start);
-    std::optional<BlockHeader> bh;
-    if (fixed.size() == BlockFixed &&
-        std::equal(fixed.begin(), fixed.begin() + 8, BlockMagic.begin())) {
-        BlockHeader b{};
-        b.exponent = fixed[11];
-        const s32 n = static_cast<s32>(Le(fixed.data() + 12, 4));
-        if (n <= 0) return std::nullopt;
-        b.decompressed_size = Le(fixed.data() + 16, 8);
-        const auto sizes = f->ReadBytes(static_cast<u64>(n) * 4, block_start + BlockFixed);
-        if (sizes.size() != static_cast<u64>(n) * 4) return std::nullopt;
-        b.sizes.resize(n);
-        for (s32 i = 0; i < n; ++i) b.sizes[i] = static_cast<u32>(Le(sizes.data() + i * 4, 4));
-        payload_start = block_start + BlockFixed + static_cast<u64>(n) * 4;
-        bh = std::move(b);
-    } else {
-        payload_start = block_start;
-    }
-
-    if (!sections.empty() && sections[0].offset > static_cast<s64>(HeaderSize)) {
-        Section fake{};
-        fake.offset = static_cast<s64>(HeaderSize);
-        fake.size = sections[0].offset - static_cast<s64>(HeaderSize);
-        sections.insert(sections.begin(), fake);
-    }
-    return std::make_pair(std::move(sections), std::move(bh));
-}
-
-// Sequential reader over independently-compressed blocks, one block cached (like nsz_rs).
-class BlockReader {
-public:
-    BlockReader(FileSys::VirtualFile f, const BlockHeader& bh, u64 payload_start)
-        : source{std::move(f)}, block_size{1ULL << bh.exponent}, sizes{bh.sizes},
-          decompressed_size{bh.decompressed_size} {
-        if (bh.exponent < 14 || bh.exponent > 32 || sizes.empty())
-            throw std::runtime_error("bad NCZBLOCK header");
-        u64 off = payload_start;
-        offsets.reserve(sizes.size());
-        for (u32 s : sizes) { offsets.push_back(off); off += s; }
-    }
-
-    // Fills `out` from `position`; returns bytes written.
-    std::size_t ReadAt(u64 position, std::span<u8> out) {
-        std::size_t done = 0;
-        while (done < out.size() && position < decompressed_size) {
-            const u64 id = position / block_size;
-            if (id >= sizes.size()) break;
-            if (cache_id != id && !Load(id)) break;
-            const u64 in = position - id * block_size;
-            const std::size_t n = static_cast<std::size_t>(
-                std::min<u64>(out.size() - done, cache.size() - in));
-            std::memcpy(out.data() + done, cache.data() + in, n);
-            done += n;
-            position += n;
-        }
-        return done;
-    }
-
-private:
-    bool Load(u64 id) {
-        u64 want = block_size;
-        if (id + 1 == sizes.size()) {
-            const u64 rem = decompressed_size % block_size;
-            if (rem > 0) want = rem; // avoid the 0-size bug from the original nsz
-        }
-        try { cache.resize(static_cast<std::size_t>(want)); }
-        catch (const std::bad_alloc&) { return false; }
-        const u32 csize = sizes[id];
-        if (csize < want) {
-            const auto comp = source->ReadBytes(csize, offsets[id]);
-            if (comp.size() != csize ||
-                Common::Compression::DecompressDataZSTD(comp, cache) != want) return false;
-        } else if (source->Read(cache.data(), cache.size(), offsets[id]) != cache.size()) {
-            return false;
-        }
-        cache_id = id;
-        return true;
-    }
-
-    FileSys::VirtualFile source;
-    u64 block_size;
-    std::vector<u64> offsets;
-    std::vector<u32> sizes;
-    u64 decompressed_size;
-    u64 cache_id = ~0ULL;
-    std::vector<u8> cache;
-};
-
-class BlockNczFile final : public FileSys::VfsFile {
-public:
-    BlockNczFile(std::string name_, std::vector<u8> hdr_, std::vector<Section> sections_,
-                 std::unique_ptr<BlockReader> body_)
-        : name{std::move(name_)}, hdr{std::move(hdr_)}, sections{std::move(sections_)},
-          body{std::move(body_)} {
-        nca_size = HeaderSize;
-        for (const auto& s : sections)
-            nca_size = std::max<u64>(nca_size, static_cast<u64>(s.offset + s.size));
-        ciphers.reserve(sections.size());
-        for (const auto& s : sections)
-            ciphers.push_back(s.Ctr() ? std::make_unique<AESCipher<Key128>>(s.key, Mode::CTR)
-                                      : nullptr);
-    }
-
-    std::string GetName() const override { return name; }
-    std::size_t GetSize() const override { return static_cast<std::size_t>(nca_size); }
-    bool Resize(std::size_t) override { return false; }
-    FileSys::VirtualDir GetContainingDirectory() const override { return nullptr; }
-    bool IsWritable() const override { return false; }
-    bool IsReadable() const override { return true; }
-
-    std::size_t Read(u8* data, std::size_t len, std::size_t off) const override {
-        if (off >= nca_size) return 0;
-        len = static_cast<std::size_t>(std::min<u64>(len, nca_size - off));
-        const u64 end = off + len;
-        u64 pos = off;
-        if (pos < HeaderSize) {
-            const u64 n = std::min<u64>(end, HeaderSize) - pos;
-            std::memcpy(data, hdr.data() + pos, static_cast<std::size_t>(n));
-            pos += n;
-        }
-        std::scoped_lock lock{mutex};
-        for (std::size_t i = 0; i < sections.size() && pos < end; ++i) {
-            const auto& s = sections[i];
-            const u64 begin = std::max<u64>(static_cast<u64>(s.offset), HeaderSize);
-            const u64 stop = std::max<u64>(static_cast<u64>(s.offset + s.size), begin);
-            if (stop <= pos) continue;
-            if (pos < begin) {
-                const u64 g = std::min(end, begin);
-                std::memset(data + (pos - off), 0, static_cast<std::size_t>(g - pos));
-                pos = g;
-                if (pos >= end) break;
-            }
-            const u64 pe = std::min(end, stop);
-            const std::span<u8> piece(data + (pos - off), static_cast<std::size_t>(pe - pos));
-            if (body->ReadAt(pos - HeaderSize, piece) != piece.size())
-                return static_cast<std::size_t>(pos - off);
-            if (ciphers[i]) {
-                Key128 iv{};
-                SetIv(iv, s.counter, pos);
-                ciphers[i]->SetIV(iv);
-                ciphers[i]->Transcode(piece.data(), piece.size(), piece.data(), Op::Encrypt);
-            }
-            pos = pe;
-        }
-        return len;
-    }
-
-    std::size_t Write(const u8*, std::size_t, std::size_t) override { return 0; }
-    bool Rename(std::string_view) override { return false; }
-
-private:
-    const std::string name;
-    const std::vector<u8> hdr;
-    const std::vector<Section> sections;
-    const std::unique_ptr<BlockReader> body;
-    std::vector<std::unique_ptr<AESCipher<Key128>>> ciphers;
-    mutable std::mutex mutex;
-    u64 nca_size{};
-};
-
-// Solid NCZ: one zstd stream over the whole body. Decoded incrementally into memory, up to the
-// requested offset, so sequential install reads are O(n) instead of O(n^2).
-class SolidNczFile final : public FileSys::VfsFile {
-public:
-    SolidNczFile(std::string name_, FileSys::VirtualFile source_,
-                 std::vector<Section> sections_, u64 payload_start_)
-        : name{std::move(name_)}, source{std::move(source_)}, sections{std::move(sections_)},
-          payload_start{payload_start_} {
-        const auto raw_header = source->ReadBytes(HeaderSize, 0);
-        data.assign(raw_header.begin(), raw_header.end());
-        frontier = HeaderSize;
-    }
-
-    ~SolidNczFile() override {
-        if (stream != nullptr) ZSTD_freeDStream(stream);
-    }
-
-    std::string GetName() const override { return name; }
-    std::size_t GetSize() const override {
-        u64 size = HeaderSize;
-        for (const auto& s : sections)
-            size = std::max<u64>(size, static_cast<u64>(s.offset + s.size));
-        return static_cast<std::size_t>(size);
-    }
-    bool Resize(std::size_t) override { return false; }
-    FileSys::VirtualDir GetContainingDirectory() const override { return nullptr; }
-    bool IsWritable() const override { return false; }
-    bool IsReadable() const override { return true; }
-
-    std::size_t Read(u8* out, std::size_t len, std::size_t off) const override {
-        std::scoped_lock lock{mutex};
-        const u64 nca_size = GetSize();
-        if (off >= nca_size) return 0;
-        len = static_cast<std::size_t>(std::min<u64>(len, nca_size - off));
-        const u64 needed = off + len;
-        if (needed > frontier && !DecodeTo(needed)) return 0;
-        std::memcpy(out, data.data() + off, len);
-        return len;
-    }
-
-    std::size_t Write(const u8*, std::size_t, std::size_t) override { return 0; }
-    bool Rename(std::string_view) override { return false; }
-
-private:
-    bool DecodeTo(u64 needed) const {
-        if (stream == nullptr) {
-            stream = ZSTD_createDStream();
-            if (stream == nullptr) return false;
-            ZSTD_DCtx_setParameter(stream, ZSTD_d_windowLogMax, 31);
-            read_pos = payload_start;
-        }
-
-        const u64 file_size = source->GetSize();
-        std::vector<u8> out_buf(Chunk);
-
-        std::size_t index = 0;
-        for (; index < sections.size(); ++index) {
-            const auto& s = sections[index];
-            const u64 begin = std::max<u64>(static_cast<u64>(s.offset), HeaderSize);
-            const u64 end = std::max<u64>(static_cast<u64>(s.offset) + static_cast<u64>(s.size),
-                                          begin);
-            if (end > frontier) break;
-        }
-
-        for (; index < sections.size() && frontier < needed; ++index) {
-            const auto& s = sections[index];
-            const u64 begin = std::max<u64>(static_cast<u64>(s.offset), HeaderSize);
-            const u64 end = std::max<u64>(static_cast<u64>(s.offset) + static_cast<u64>(s.size),
-                                          begin);
-            if (end <= frontier) continue;
-            u64 pos = std::max<u64>(begin, frontier);
-
-            std::optional<AESCipher<Key128>> cipher;
-            if (s.Ctr()) cipher.emplace(s.key, Mode::CTR);
-
-            while (pos < end && frontier < needed) {
-                const u64 want = std::min<u64>(end - pos, out_buf.size());
-                std::size_t produced = 0;
-                while (produced < want) {
-                    if (in.pos >= in.size) {
-                        const u64 remaining = file_size - read_pos;
-                        const std::size_t chunk = static_cast<std::size_t>(
-                            std::min<u64>(Chunk, remaining));
-                        if (chunk == 0) { LOG_ERROR(Crypto, "Solid NCZ: truncated"); return false; }
-                        const auto bytes = source->ReadBytes(chunk, read_pos);
-                        if (bytes.empty()) { LOG_ERROR(Crypto, "Solid NCZ: truncated"); return false; }
-                        read_pos += bytes.size();
-                        in_buf.assign(bytes.begin(), bytes.end());
-                        in.src = in_buf.data();
-                        in.size = in_buf.size();
-                        in.pos = 0;
-                    }
-                    ZSTD_outBuffer outb{out_buf.data() + produced, want - produced, 0};
-                    const std::size_t res = ZSTD_decompressStream(stream, &outb, &in);
-                    if (ZSTD_isError(res)) {
-                        LOG_ERROR(Crypto, "Solid NCZ: zstd error");
-                        return false;
-                    }
-                    produced += outb.pos;
-                }
-                if (cipher) {
-                    Key128 iv{};
-                    SetIv(iv, s.counter, pos);
-                    cipher->SetIV(iv);
-                    cipher->Transcode(out_buf.data(), produced, out_buf.data(), Op::Encrypt);
-                }
-                data.insert(data.end(), out_buf.begin(), out_buf.begin() + produced);
-                pos += produced;
-                frontier = pos;
-            }
-        }
-        return frontier >= needed;
-    }
-
-    const std::string name;
-    const FileSys::VirtualFile source;
-    const std::vector<Section> sections;
-    const u64 payload_start;
-    mutable std::mutex mutex;
-    mutable std::vector<u8> data;
-    mutable ZSTD_DStream* stream = nullptr;
-    mutable u64 read_pos = 0;
-    mutable u64 frontier = 0;
-    mutable std::vector<u8> in_buf;
-    mutable ZSTD_inBuffer in{nullptr, 0, 0};
-};
-
-// Shared opener: block path always; solid only when allow_solid is true.
-FileSys::VirtualFile OpenNcz(const FileSys::VirtualFile& f, std::string_view name,
-                             bool allow_solid) {
-    if (!f) return nullptr;
-    u64 ps = 0;
-    auto parsed = ParseHeader(f, ps);
-    if (!parsed) return nullptr;
-    auto& [sections, block] = *parsed;
-
-    if (block) {
-        auto hdr = f->ReadBytes(HeaderSize, 0);
-        if (hdr.size() != HeaderSize) return nullptr;
-        std::unique_ptr<BlockReader> body;
-        try { body = std::make_unique<BlockReader>(f, *block, ps); }
-        catch (const std::exception& e) {
-            LOG_ERROR(Crypto, "Failed to open NCZBLOCK: {}", e.what());
-            return nullptr;
-        }
-        return std::make_shared<BlockNczFile>(std::string{name}, std::move(hdr),
-                                              std::move(sections), std::move(body));
-    }
-
-    if (allow_solid) {
-        return std::make_shared<SolidNczFile>(std::string{name}, f, std::move(sections), ps);
-    }
-
-    LOG_DEBUG(Crypto, "Solid NCZ {} skipped (only NCZBLOCK is supported for play)", name);
-    return nullptr;
-}
-
-} // Anonymous namespace
-
-bool IsBlockCompressedNCZ(const FileSys::VirtualFile& f) {
-    if (!f) return false;
-    u64 ps = 0;
-    const auto p = ParseHeader(f, ps);
-    return p && p->second.has_value();
-}
-
-FileSys::VirtualFile DecompressNCZ(const FileSys::VirtualFile& f, std::string_view name) {
-    return OpenNcz(f, name, /*allow_solid=*/false);
-}
-
-FileSys::VirtualFile DecompressNCZForInstall(const FileSys::VirtualFile& f,
-                                             std::string_view name) {
-    return OpenNcz(f, name, /*allow_solid=*/true);
-}
-
-} // namespace Core::Crypto

src/core/crypto/ncz_decompressor.h

@@ -1,35 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-#pragma once
-
-#include <string_view>
-
-#include "core/file_sys/vfs/vfs_types.h"
-
-namespace Core::Crypto {
-
-/**
- * Opens an NCZBLOCK-compressed NCA (nsz -B / .xcz block format) as a VirtualFile backed by
- * on-demand block decoding. Nothing is written to disk; each read decodes only the blocks it
- * covers and re-encrypts CTR sections. Solid NCZ is not supported.
- *
- * @param ncz_file    The raw NCZ payload (e.g. "<id>.ncz" entry from an NSZ/XCZ PFS0).
- * @param output_name Name to give the returned VirtualFile (e.g. "<id>.nca").
- *
- * @return a VirtualFile for the NCA, or nullptr if the NCZ is malformed or not NCZBLOCK.
- */
-[[nodiscard]] FileSys::VirtualFile DecompressNCZ(const FileSys::VirtualFile& ncz_file,
-                                                 std::string_view output_name);
-
-/**
- * Returns true if the given file is an NCZ payload in block-compressed (NCZBLOCK) form,
- * i.e. one that DecompressNCZ() can open. Solid NCZ returns false.
- */
-[[nodiscard]] bool IsBlockCompressedNCZ(const FileSys::VirtualFile& ncz_file);
-
-// Install to NAND: same as DecompressNCZ, but solid NCZ is decoded on demand too.
-[[nodiscard]] FileSys::VirtualFile DecompressNCZForInstall(
-    const FileSys::VirtualFile& ncz_file, std::string_view output_name);
-
-} // namespace Core::Crypto

src/core/file_sys/content_archive.cpp

@@ -45,7 +45,6 @@ NCA::NCA(VirtualFile file_, const NCA* base_nca, bool allow_missing_base)
     // Ensure we have the proper key area keys to continue.
     const u8 master_key_id = MasterKeyIdForKeyGeneration(reader->GetKeyGeneration());
     if (!keys.HasKey(Core::Crypto::S128KeyType::KeyArea, master_key_id, reader->GetKeyIndex())) {
-        LOG_ERROR(Loader, "Cannot decrypt {}: keys for gen 0x{:02X}+ required.", file->GetName(), reader->GetKeyGeneration());
         status = Loader::ResultStatus::ErrorMissingKeyAreaKey;
         return;
     }

src/core/file_sys/registered_cache.cpp

@@ -16,7 +16,6 @@
 #include "common/random.h"
 #include "common/string_util.h"
 #include "core/crypto/key_manager.h"
-#include "core/crypto/ncz_decompressor.h"
 #include "core/file_sys/card_image.h"
 #include "core/file_sys/common_funcs.h"
 #include "core/file_sys/content_archive.h"
@@ -159,25 +158,6 @@ static std::shared_ptr<NSP> OpenContainerAsNsp(const VirtualFile& file, Loader::
     return nullptr;
 }
 
-// Checks that every .ncz entry in the container is NCZBLOCK. Solid NCZ has no random access,
-// and the whole point of this build is to reject it at scan time so such files never appear
-// in the game list. Containers without any .ncz (plain NSP/XCI) pass through.
-static bool ContainerIsFullyBlockCompressed(const std::shared_ptr<NSP>& nsp) {
-    if (!nsp) {
-        return false;
-    }
-    for (const auto& entry : nsp->GetFiles()) {
-        const auto name = entry->GetName();
-        if (name.size() < 4 || name.substr(name.size() - 4) != ".ncz") {
-            continue;
-        }
-        if (!Core::Crypto::IsBlockCompressedNCZ(entry)) {
-            return false; // solid NCZ -> reject the whole container
-        }
-    }
-    return true; // no .ncz at all, or all block-compressed
-}
-
 template <typename Callback>
 bool ForEachContainerEntry(const std::shared_ptr<NSP>& nsp, bool only_content,
                            std::optional<u64> base_program_id, Callback&& on_entry) {
@@ -831,14 +811,7 @@ std::vector<ContentProviderEntry> RegisteredCache::ListEntriesFilter(
 }
 
 static std::shared_ptr<NCA> GetNCAFromNSPForID(const NSP& nsp, const NcaID& id) {
-    const auto id_string = Common::HexToString(id, false);
+    auto file = nsp.GetFile(fmt::format("{}.nca", Common::HexToString(id, false)));
-    auto file = nsp.GetFile(fmt::format("{}.nca", id_string));
-    if (file == nullptr) {
-        if (auto ncz_file = nsp.GetFile(fmt::format("{}.ncz", id_string))) {
-            file = Core::Crypto::DecompressNCZForInstall(
-                ncz_file, fmt::format("{}.nca", id_string));
-        }
-    }
     if (file == nullptr) {
         return nullptr;
     }
@@ -1300,12 +1273,6 @@ bool ManualContentProvider::AddEntriesFromContainer(VirtualFile file, bool only_
         return false;
     }
 
-    // Reject solid NCZ (.nsz without -B): only block-compressed NCZ is supported.
-    if (!ContainerIsFullyBlockCompressed(nsp)) {
-        LOG_DEBUG(Service_FS, "Skipping solid-compressed container: {}", file->GetName());
-        return false;
-    }
-
     return ForEachContainerEntry(
         nsp, only_content, base_program_id,
         [this](TitleType title_type, ContentRecordType content_type, u64 title_id,
@@ -1448,9 +1415,8 @@ void ExternalContentProvider::ScanDirectory(const VirtualDir& dir) {
         }
 
         const auto extension = Common::ToLower(filename.substr(dot_pos + 1));
-        // .xcz is an XCI container, but its updates only work through the NSP/PFS0 path
+
-        // (DecompressNCZ), not ProcessXCI. So route .xcz here on purpose.
+        if (extension == "nsp") {
-        if (extension == "nsp" || extension == "nsz" || extension == "xcz") {
             ProcessNSP(file);
         } else if (extension == "xci") {
             ProcessXCI(file);
@@ -1468,12 +1434,6 @@ void ExternalContentProvider::ProcessNSP(const VirtualFile& file) {
         return;
     }
 
-    // Reject solid NCZ (.nsz without -B): Eden only supports block-compressed NCZ.
-    if (!ContainerIsFullyBlockCompressed(nsp)) {
-        LOG_DEBUG(Service_FS, "Skipping solid-compressed container: {}", file->GetName());
-        return;
-    }
-
     LOG_DEBUG(Service_FS, "Processing NSP file: {}", file->GetName());
     AddExternalEntriesFromContainer(nsp, entries, versions, multi_version_entries);
 }

src/core/file_sys/submission_package.cpp

@@ -12,7 +12,6 @@
 #include "common/hex_util.h"
 #include "common/logging.h"
 #include "core/crypto/key_manager.h"
-#include "core/crypto/ncz_decompressor.h"
 #include "core/file_sys/common_funcs.h"
 #include "core/file_sys/content_archive.h"
 #include "core/file_sys/nca_metadata.h"
@@ -247,17 +246,6 @@ void NSP::ReadNCAs(const std::vector<VirtualFile>& files) {
                 const auto id_string = Common::HexToString(rec.nca_id, false);
                 auto next_file = pfs->GetFile(fmt::format("{}.nca", id_string));
 
-                if (next_file == nullptr) {
-                    auto ncz_file = pfs->GetFile(fmt::format("{}.ncz", id_string));
-                    if (ncz_file != nullptr) {
-                        next_file = Core::Crypto::DecompressNCZ(ncz_file, fmt::format("{}.nca", id_string));
-                        if (next_file == nullptr) {
-                            LOG_DEBUG(Service_FS, "Skipping solid or unsupported NCZ content {}.ncz", id_string);
-                            continue;
-                        }
-                    }
-                }
-
                 if (next_file == nullptr) {
                     if (rec.type != ContentRecordType::DeltaFragment) {
                         LOG_WARNING(Service_FS,

src/core/hle/kernel/k_process.cpp

@@ -23,6 +23,9 @@
 #ifdef HAS_NCE
 #include "core/arm/nce/arm_nce.h"
 #endif
+#ifdef HAS_SCARLETNX
+#include "core/arm/scarletnx/scarletnx_process.h"
+#endif
 
 namespace Kernel {
 
@@ -1311,6 +1314,13 @@ void KProcess::InitializeInterfaces(KernelCore& kernel) {
         for (size_t i = 0; i < Core::Hardware::NUM_CPU_CORES; i++)
             m_arm_interfaces[i] = std::make_unique<Core::ArmNce>(kernel.System(), true, i);
     } else
+#endif
+#ifdef HAS_SCARLETNX
+    if (Core::ScarletNxProcess::ShouldUse(kernel.System(), *this)) {
+        auto scarletnx = std::make_shared<Core::ScarletNxProcess>(kernel.System(), *this);
+        for (size_t i = 0; i < Core::Hardware::NUM_CPU_CORES; i++)
+            m_arm_interfaces[i] = Core::MakeScarletNxInterface(scarletnx, this->Is64Bit(), i);
+    } else
 #endif
         if (this->Is64Bit()) {
         for (size_t i = 0; i < Core::Hardware::NUM_CPU_CORES; i++) {

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

@@ -4,10 +4,6 @@
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <algorithm>
-#include <chrono>
-#include <thread>
-
 #include "core/file_sys/errors.h"
 #include "core/hle/service/cmif_serialization.h"
 #include "core/hle/service/filesystem/fsp/fs_i_storage.h"
@@ -35,14 +31,6 @@ Result IStorage::Read(
     R_UNLESS(length >= 0, FileSys::ResultInvalidSize);
     R_UNLESS(offset >= 0, FileSys::ResultInvalidOffset);
 
-    const auto now = std::chrono::steady_clock::now();
-    if (now < next_read_tick) {
-        std::this_thread::sleep_for(next_read_tick - now);
-    }
-    next_read_tick = std::chrono::steady_clock::now() +
-                     std::chrono::nanoseconds{std::min(length, MaxChargedReadSize) *
-                                              1'000'000'000 / ReadBytesPerSecond};
-
     // Read the data from the Storage backend
     backend->Read(out_bytes.data(), length, offset);
 

src/core/hle/service/filesystem/fsp/fs_i_storage.h

@@ -6,8 +6,6 @@
 
 #pragma once
 
-#include <chrono>
-
 #include "core/file_sys/vfs/vfs.h"
 #include "core/hle/service/cmif_types.h"
 #include "core/hle/service/filesystem/filesystem.h"
@@ -22,10 +20,6 @@ public:
 private:
     FileSys::VirtualFile backend;
 
-    static constexpr s64 ReadBytesPerSecond = 40'000'000;
-    static constexpr s64 MaxChargedReadSize = 64 * 1024;
-    std::chrono::steady_clock::time_point next_read_tick{};
-
     Result Read(
         OutBuffer<BufferAttr_HipcMapAlias | BufferAttr_HipcMapTransferAllowsNonSecure> out_bytes,
         s64 offset, s64 length);

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

@@ -8,6 +8,7 @@
 #include "common/assert.h"
 #include "common/logging.h"
 #include "core/core.h"
+#include "core/hle/kernel/k_process.h"
 #include "core/hle/service/nvdrv/core/container.h"
 #include "core/hle/service/nvdrv/devices/ioctl_serialization.h"
 #include "core/hle/service/nvdrv/devices/nvhost_nvdec.h"
@@ -71,17 +72,23 @@ NvResult nvhost_nvdec::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> in
 
 void nvhost_nvdec::OnOpen(NvCore::SessionId session_id, DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream started");
-    system.SetNVDECActive(true);
     sessions[fd] = session_id;
+    if (const auto* session = core.GetSession(session_id);
+        session != nullptr && session->process != nullptr) {
+        system.NotifyNVDECChannelOpen(session->process->GetId());
+    }
     host1x.StartDevice(fd, Tegra::Host1x::ChannelType::NvDec, channel_syncpoint);
 }
 
 void nvhost_nvdec::OnClose(DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream ended");
     host1x.StopDevice(fd, Tegra::Host1x::ChannelType::NvDec);
-    system.SetNVDECActive(false);
     auto it = sessions.find(fd);
     if (it != sessions.end()) {
+        if (const auto* session = core.GetSession(it->second);
+            session != nullptr && session->process != nullptr) {
+            system.NotifyNVDECChannelClose(session->process->GetId());
+        }
         sessions.erase(it);
     }
 }

src/core/hle/service/service.cpp

@@ -4,12 +4,16 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <chrono>
 #include <fmt/ranges.h>
+#include <string_view>
+#include <thread>
 #include "common/assert.h"
 #include "common/logging.h"
 #include "common/settings.h"
 #include "core/core.h"
 #include "core/hle/ipc.h"
+#include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/service/ipc_helpers.h"
 #include "core/hle/service/service.h"
@@ -33,6 +37,7 @@ ServiceFrameworkBase::ServiceFrameworkBase(Core::System& system_, const char* se
     : SessionRequestHandler(system_.Kernel(), service_name_)
     , system{system_}
     , service_name{service_name_}
+    , is_i_storage{std::string_view{service_name_} == "IStorage"}
     , handler_invoker{handler_invoker_}
     , max_sessions{max_sessions_}
 {}
@@ -77,13 +82,22 @@ void ServiceFrameworkBase::ReportUnimplementedFunction(HLERequestContext& ctx,
 }
 
 void ServiceFrameworkBase::InvokeRequest(HLERequestContext& ctx) {
-    auto it = handlers.find(ctx.GetCommand());
+    const auto command = ctx.GetCommand();
+    auto it = handlers.find(command);
+    const bool is_cmd_read = command == 0;
     FunctionInfoBase const* info = it == handlers.end() ? nullptr : &it->second;
     if (info == nullptr || info->handler_callback == nullptr)
         return ReportUnimplementedFunction(ctx, info);
 
     LOG_TRACE(Service, "{}", MakeFunctionString(info->name, GetServiceName(), ctx.CommandBuffer()));
     handler_invoker(this, info->handler_callback, ctx);
+
+    if (is_i_storage && is_cmd_read) {
+        const auto* const process = ctx.GetThread().GetOwnerProcess();
+        if (process != nullptr && system.IsNVDECActiveForProcess(process->GetId())) {
+            std::this_thread::sleep_for(std::chrono::microseconds{600});
+        }
+    }
 }
 
 void ServiceFrameworkBase::InvokeRequestTipc(HLERequestContext& ctx) {

src/core/hle/service/service.h

@@ -107,6 +107,8 @@ protected:
     Core::System& system;
     /// Identifier string used to connect to the service.
     const char* service_name;
+    /// Whether this is the IStorage service.
+    const bool is_i_storage;
     /// Function used to safely up-cast pointers to the derived class before invoking a handler.
     InvokerFn* handler_invoker;
     /// Maximum number of concurrent sessions that this service can handle.

src/core/loader/loader.cpp

@@ -82,29 +82,25 @@ std::shared_ptr<FileSys::NSP> OpenContainerAsNsp(FileSys::VirtualFile file, File
         return nullptr;
     }
 
-    if (type == FileType::NSP || type == FileType::NSZ) {
+    if (type == FileType::NSP) {
         auto nsp = std::make_shared<FileSys::NSP>(file, program_id, program_index);
         return nsp->GetStatus() == ResultStatus::Success ? nsp : nullptr;
     }
 
-    if (type == FileType::XCI || type == FileType::XCZ) {
+    if (type == FileType::XCI) {
         FileSys::XCI xci{file, program_id, program_index};
-        if (xci.GetStatus() == ResultStatus::Success) {
+        if (xci.GetStatus() != ResultStatus::Success) {
-            auto secure_nsp = xci.GetSecurePartitionNSP();
-            if (secure_nsp != nullptr && secure_nsp->GetStatus() == ResultStatus::Success) {
-                return secure_nsp;
-            }
-        }
-
-        // PFS0-shaped .xcz (NSZ-style container) does not parse as XCI; fall back to NSP.
-        if (type == FileType::XCZ) {
-            auto nsp = std::make_shared<FileSys::NSP>(file, program_id, program_index);
-            return nsp->GetStatus() == ResultStatus::Success ? nsp : nullptr;
-        }
-
             return nullptr;
         }
 
+        auto secure_nsp = xci.GetSecurePartitionNSP();
+        if (secure_nsp == nullptr || secure_nsp->GetStatus() != ResultStatus::Success) {
+            return nullptr;
+        }
+
+        return secure_nsp;
+    }
+
     return nullptr;
 }
 
@@ -148,8 +144,7 @@ FileType IdentifyFile(FileSys::VirtualFile file) {
 }
 
 bool IsContainerType(FileType type) {
-    return type == FileType::NSP || type == FileType::NSZ || type == FileType::XCI ||
+    return type == FileType::NSP || type == FileType::XCI;
-           type == FileType::XCZ;
 }
 
 bool IsBootableGameContainer(FileSys::VirtualFile file, FileType type, u64 program_id,
@@ -185,12 +180,8 @@ FileType GuessFromFilename(const std::string& name) {
         return FileType::NCA;
     else if (extension == "xci")
         return FileType::XCI;
-    else if (extension == "xcz")
-        return FileType::XCZ;
     else if (extension == "nsp")
         return FileType::NSP;
-    else if (extension == "nsz")
-        return FileType::NSZ;
     else if (extension == "kip")
         return FileType::KIP;
     return FileType::Unknown;
@@ -206,14 +197,10 @@ std::string GetFileTypeString(FileType type) {
         return "NCA";
     case FileType::XCI:
         return "XCI";
-    case FileType::XCZ:
-        return "XCZ";
     case FileType::NAX:
         return "NAX";
     case FileType::NSP:
         return "NSP";
-    case FileType::NSZ:
-        return "NSZ";
     case FileType::KIP:
         return "KIP";
     case FileType::DeconstructedRomDirectory:
@@ -339,29 +326,12 @@ static std::unique_ptr<AppLoader> GetFileLoader(Core::System& system, FileSys::V
                                                system.GetContentProvider(), program_id,
                                                program_index);
 
-    // NCZ-compressed XCI container. XCI-shaped .xcz goes to the XCI loader, while
-    // PFS0-shaped (NSZ-style) .xcz containers fall back to the NSP loader.
-    case FileType::XCZ: {
-        FileSys::XCI probe{file, program_id, program_index};
-        if (probe.GetStatus() == ResultStatus::Success &&
-            probe.GetSecurePartitionNSP() != nullptr) {
-            return std::make_unique<AppLoader_XCI>(std::move(file),
-                                                   system.GetFileSystemController(),
-                                                   system.GetContentProvider(), program_id,
-                                                   program_index);
-        }
-        return std::make_unique<AppLoader_NSP>(std::move(file), system.GetFileSystemController(),
-                                               system.GetContentProvider(), program_id,
-                                               program_index);
-    }
-
     // NX NAX (NintendoAesXts) file format.
     case FileType::NAX:
         return std::make_unique<AppLoader_NAX>(std::move(file));
 
     // NX NSP (Nintendo Submission Package) file format
     case FileType::NSP:
-    case FileType::NSZ:
         return std::make_unique<AppLoader_NSP>(std::move(file), system.GetFileSystemController(),
                                                system.GetContentProvider(), program_id,
                                                program_index);

src/core/loader/loader.h

@@ -42,9 +42,7 @@ enum class FileType {
     NRO,
     NCA,
     NSP,
-    NSZ,
     XCI,
-    XCZ,
     NAX,
     KIP,
     DeconstructedRomDirectory,

src/core/loader/nsp.cpp

@@ -56,17 +56,6 @@ AppLoader_NSP::AppLoader_NSP(FileSys::VirtualFile file_,
 AppLoader_NSP::~AppLoader_NSP() = default;
 
 FileType AppLoader_NSP::IdentifyType(const FileSys::VirtualFile& nsp_file) {
-    // .nsz is NSP-compatible; return NSZ so the game list shows the right type.
-    if (nsp_file->GetName().ends_with(".nsz")) {
-        return FileType::NSZ;
-    }
-
-    // PFS0-shaped .xcz containers are handled by this loader too; report XCZ so the game
-    // list does not show them as NSP. XCI-shaped .xcz is identified by AppLoader_XCI.
-    if (nsp_file->GetName().ends_with(".xcz")) {
-        return FileType::XCZ;
-    }
-
     const FileSys::NSP nsp(nsp_file);
 
     if (nsp.GetStatus() != ResultStatus::Success) {

src/frontend_common/content_manager.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: 2024 yuzu Emulator Project
@@ -163,9 +163,7 @@ inline InstallResult InstallNSP(Core::System& system, FileSys::VfsFilesystem& vf
 
     std::shared_ptr<FileSys::NSP> nsp;
     FileSys::VirtualFile file = vfs.OpenFile(filename, FileSys::OpenMode::Read);
-    const auto lower_name = boost::to_lower_copy(file->GetName());
+    if (boost::to_lower_copy(file->GetName()).ends_with("nsp")) {
-
-    if (lower_name.ends_with("nsp") || lower_name.ends_with("nsz")) {
         nsp = std::make_shared<FileSys::NSP>(file);
         if (nsp->IsExtractedType()) {
             return InstallResult::Failure;

src/qt_common/config/shared_translation.cpp

@@ -194,6 +194,10 @@ std::unique_ptr<TranslationMap> InitializeTranslations(QObject* parent) {
     INSERT(Settings, skip_cpu_inner_invalidation, tr("Skip CPU Inner Invalidation"),
            tr("Skips certain cache invalidations during memory updates, reducing CPU usage and "
               "improving latency. This may cause soft-crashes."));
+    INSERT(Settings, nce_invalidation_gpu_readback,
+           tr("GPU Readback Before NCE Invalidation"),
+           tr("Flushes GPU buffer data back to guest memory before NCE invalidates the whole "
+              "guest page, preserving it while the CPU store proceeds."));
     INSERT(Settings, vsync_mode, tr("VSync Mode:"),
            tr("FIFO (VSync) does not drop frames or exhibit tearing but is limited by the screen "
               "refresh rate.\nFIFO Relaxed allows tearing as it recovers from a slow down.\n"
@@ -445,7 +449,12 @@ std::unique_ptr<ComboboxTranslationMap> ComboboxEnumeration(QObject* parent) {
     translations->insert({Settings::EnumMetadata<Settings::CpuBackend>::Index(),
                           {
                               PAIR(CpuBackend, Dynarmic, tr("Dynarmic")),
+#ifdef HAS_NCE
                               PAIR(CpuBackend, Nce, tr("NCE")),
+#endif
+#ifdef HAS_SCARLETNX
+                              PAIR(CpuBackend, ScarletNx, tr("ScarletNX")),
+#endif
                           }});
     translations->insert({Settings::EnumMetadata<Settings::FullscreenMode>::Index(),
                           {

src/qt_common/game_list/worker.cpp

@@ -360,8 +360,7 @@ void GameListWorker::ScanFileSystem(ScanTarget target, const std::string& dir_pa
             }
 
             if (target == ScanTarget::PopulateGameList &&
-                (file_type == Loader::FileType::XCI || file_type == Loader::FileType::XCZ ||
+                (file_type == Loader::FileType::XCI || file_type == Loader::FileType::NSP) &&
-                 file_type == Loader::FileType::NSP || file_type == Loader::FileType::NSZ) &&
                 !Loader::IsBootableGameContainer(file, file_type)) {
                 return true;
             }
@@ -376,9 +375,7 @@ void GameListWorker::ScanFileSystem(ScanTarget target, const std::string& dir_pa
                                        program_id, file);
                 } else if (Settings::values.ext_content_from_game_dirs.GetValue() &&
                            (file_type == Loader::FileType::XCI ||
-                            file_type == Loader::FileType::XCZ ||
+                            file_type == Loader::FileType::NSP)) {
-                            file_type == Loader::FileType::NSP ||
-                            file_type == Loader::FileType::NSZ)) {
                     void(provider->AddEntriesFromContainer(file));
                 }
             } else {
@@ -405,8 +402,7 @@ void GameListWorker::ScanFileSystem(ScanTarget target, const std::string& dir_pa
                 };
 
                 if (res2 == Loader::ResultStatus::Success && program_ids.size() > 1 &&
-                    (file_type == Loader::FileType::XCI || file_type == Loader::FileType::XCZ ||
+                    (file_type == Loader::FileType::XCI || file_type == Loader::FileType::NSP)) {
-                     file_type == Loader::FileType::NSP || file_type == Loader::FileType::NSZ)) {
                     for (const auto id : program_ids) {
                         // dravee suggested this, only viable way to
                         // not show sub-games in qlaunch for now.

src/qt_common/qt_common.cpp

@@ -57,8 +57,7 @@ std::unique_ptr<EmuThread> emu_thread = nullptr;
 
 const QStringList supported_file_extensions = {QStringLiteral("nro"), QStringLiteral("nso"),
                                                QStringLiteral("nca"), QStringLiteral("xci"),
-                                               QStringLiteral("xcz"), QStringLiteral("nsp"),
+                                               QStringLiteral("nsp"), QStringLiteral("kip")};
-                                               QStringLiteral("nsz"), QStringLiteral("kip")};
 
 Core::Frontend::WindowSystemType GetWindowSystemType() {
     // Determine WSI type based on Qt platform.

src/shader_recompiler/frontend/ir/breadth_first_search.h

@@ -1,12 +1,8 @@
-// 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
 
 #pragma once
 
-#include <functional>
 #include <optional>
 #include <type_traits>
 #include <queue>
@@ -17,10 +13,9 @@
 
 namespace Shader::IR {
 
-template <typename Pred, typename Resolve = std::identity>
+template <typename Pred>
-auto BreadthFirstSearch(Value value, Pred&& pred, Resolve&& resolve = {})
+auto BreadthFirstSearch(const Value& value, Pred&& pred)
     -> std::invoke_result_t<Pred, const Inst*> {
-    value = resolve(value);
     if (value.IsImmediate()) {
         // Nothing to do with immediates
         return std::nullopt;
@@ -41,7 +36,7 @@ auto BreadthFirstSearch(Value value, Pred&& pred, Resolve&& resolve = {})
         }
         // Visit the right most arguments first
         for (size_t arg = inst->NumArgs(); arg--;) {
-            const Value arg_value{resolve(inst->Arg(arg))};
+            const Value arg_value{inst->Arg(arg)};
             if (arg_value.IsImmediate()) {
                 continue;
             }

src/shader_recompiler/ir_opt/texture_pass.cpp

@@ -31,60 +31,6 @@ struct TextureInst {
 
 using TextureInstVector = boost::container::small_vector<TextureInst, 24>;
 
-const IR::Block* ConditionalHeader(const IR::Block* body) {
-    const auto predecessors{body->ImmPredecessors()};
-    const auto successors{body->ImmSuccessors()};
-    if (predecessors.size() != 1 || successors.size() != 1) return nullptr;
-    const IR::Block* const header{predecessors[0]};
-    const auto targets{header->ImmSuccessors()};
-    if (targets.size() != 2 || targets[0] != body || targets[1] != successors[0] ||
-        targets[1]->ImmPredecessors().size() != 2 || header->empty() ||
-        header->back().GetOpcode() != IR::Opcode::ConditionRef) {
-        return nullptr;
-    }
-    return header;
-}
-
-std::pair<IR::Value, bool> Condition(IR::Value value) {
-    bool negated{};
-    while (const IR::Inst* inst = value.TryInstRecursive()) {
-        if (inst->GetOpcode() == IR::Opcode::LogicalNot) negated = !negated;
-        else if (inst->GetOpcode() != IR::Opcode::ConditionRef) break;
-        value = inst->Arg(0);
-    }
-    return {value.Resolve(), negated};
-}
-
-IR::Value ResolveConditionalPhi(IR::Value value, const IR::Block* block) {
-    if (!value.Resolve().IsPhi() || !(block = ConditionalHeader(block))) return value;
-    const auto [condition, negated]{Condition(block->back().Arg(0))};
-    boost::container::small_vector<const IR::Block*, 16> visited;
-    while (value.Resolve().IsPhi() && std::ranges::find(visited, block) == visited.end()) {
-        visited.push_back(block);
-        const IR::Inst* const phi{value.InstRecursive()};
-        const auto predecessors{block->ImmPredecessors()};
-        if (predecessors.size() == 1) {
-            block = predecessors[0];
-            continue;
-        }
-        if (predecessors.size() != 2 || phi->NumArgs() != 2) break;
-        const IR::Block* body{predecessors[0]};
-        const IR::Block* header{ConditionalHeader(body)};
-        if (header != predecessors[1]) {
-            body = predecessors[1];
-            header = ConditionalHeader(body);
-            if (header != predecessors[0]) break;
-        }
-        const auto [other_condition, other_negated]{Condition(header->back().Arg(0))};
-        if (condition == other_condition && (phi->PhiBlock(0) == header || phi->PhiBlock(0) == body)) {
-            const IR::Block* const incoming{negated == other_negated ? body : header};
-            value = phi->Arg(phi->PhiBlock(0) == incoming ? 0 : 1);
-        }
-        block = header;
-    }
-    return value;
-}
-
 constexpr u32 DESCRIPTOR_SIZE = 8;
 constexpr u32 DESCRIPTOR_SIZE_SHIFT = u32(std::countr_zero(DESCRIPTOR_SIZE));
 constexpr u32 DESCRIPTOR_MAX_COUNT = 1024;
@@ -353,23 +299,23 @@ static inline bool IsTexturePixelFormatIntegerCached(Environment& env,
 }
 
 
-std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env, const HostTranslateInfo& host_info, const IR::Block* block);
+std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env, const HostTranslateInfo& host_info);
-static inline std::optional<ConstBufferAddr> TrackCached(const IR::Value& v, Environment& env, const HostTranslateInfo& host_info, const IR::Block* block) {
+static inline std::optional<ConstBufferAddr> TrackCached(const IR::Value& v, Environment& env, const HostTranslateInfo& host_info) {
-    if (const IR::Inst* key = v.TryInstRecursive()) {
+    if (const IR::Inst* key = v.InstRecursive()) {
         if (auto it = env.track_cache.find(key); it != env.track_cache.end()) return it->second;
-        auto found = Track(v, env, host_info, block);
+        auto found = Track(v, env, host_info);
         if (found) env.track_cache.emplace(key, *found);
         return found;
     }
-    return Track(v, env, host_info, block);
+    return Track(v, env, host_info);
 }
 
-std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env, const HostTranslateInfo& host_info, const IR::Block* block);
+std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env, const HostTranslateInfo& host_info);
 
-std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env, const HostTranslateInfo& host_info, const IR::Block* block) {
+std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env, const HostTranslateInfo& host_info) {
-    return IR::BreadthFirstSearch(value, [&env, &host_info, block](const IR::Inst* inst) {
+    return IR::BreadthFirstSearch(value, [&env, &host_info](const IR::Inst* inst) {
-        return TryGetConstBuffer(inst, env, host_info, block);
+        return TryGetConstBuffer(inst, env, host_info);
-    }, [&](IR::Value operand) { return ResolveConditionalPhi(operand, block); });
+    });
 }
 
 std::optional<u32> TryGetConstant(IR::Value& value, Environment& env) {
@@ -393,13 +339,13 @@ std::optional<u32> TryGetConstant(IR::Value& value, Environment& env) {
     return ReadCbufCached(env, index_number, offset_number);
 }
 
-std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env, const HostTranslateInfo& host_info, const IR::Block* block) {
+std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env, const HostTranslateInfo& host_info) {
     switch (inst->GetOpcode()) {
     default:
         return std::nullopt;
     case IR::Opcode::BitwiseOr32: {
-        std::optional lhs{TrackCached(inst->Arg(0), env, host_info, block)};
+        std::optional lhs{TrackCached(inst->Arg(0), env, host_info)};
-        std::optional rhs{TrackCached(inst->Arg(1), env, host_info, block)};
+        std::optional rhs{TrackCached(inst->Arg(1), env, host_info)};
         if (!lhs || !rhs) {
             return std::nullopt;
         }
@@ -429,7 +375,7 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environme
         if (!shift.IsImmediate()) {
             return std::nullopt;
         }
-        std::optional lhs{TrackCached(inst->Arg(0), env, host_info, block)};
+        std::optional lhs{TrackCached(inst->Arg(0), env, host_info)};
         if (lhs) {
             lhs->shift_left = shift.U32();
         }
@@ -457,7 +403,7 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environme
                 return std::nullopt;
             } while (false);
         }
-        std::optional lhs{TrackCached(op1, env, host_info, block)};
+        std::optional lhs{TrackCached(op1, env, host_info)};
         if (lhs) {
             lhs->shift_left = static_cast<u32>(std::countr_zero(op2.U32()));
         }
@@ -523,7 +469,7 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environme
 TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst, const HostTranslateInfo& host_info) {
     ConstBufferAddr addr;
     if (IsBindless(inst)) {
-        const std::optional<ConstBufferAddr> track_addr{TrackCached(inst.Arg(0), env, host_info, block)};
+        const std::optional<ConstBufferAddr> track_addr{TrackCached(inst.Arg(0), env, host_info)};
 
         if (!track_addr) {
             throw NotImplementedException("Failed to track bindless texture constant buffer");
@@ -725,7 +671,6 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo
     env.track_cache.clear();
     TextureInstVector to_replace;
     for (IR::Block* const block : program.post_order_blocks) {
-        env.track_cache.clear();
         for (IR::Inst& inst : block->Instructions()) {
             if (!IsTextureInstruction(inst)) {
                 continue;

src/video_core/buffer_cache/buffer_cache.h

@@ -1738,13 +1738,6 @@ void BufferCache<P>::TouchBuffer(Buffer& buffer, BufferId buffer_id) noexcept {
 
 template <class P>
 bool BufferCache<P>::SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 size) {
-#ifdef __ANDROID__
-    const bool use_optimized_vertex_buffers = Settings::values.use_optimized_vertex_buffers.GetValue();
-#else
-    constexpr bool use_optimized_vertex_buffers = true;
-#endif
-
-    if (use_optimized_vertex_buffers) {
     upload_copies.clear();
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
@@ -1775,27 +1768,6 @@ bool BufferCache<P>::SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 si
     UploadMemory(buffer, total_size_bytes, largest_copy, copies_span);
     any_buffer_uploaded = true;
     return false;
-    } else {
-        boost::container::small_vector<BufferCopy, 4> copies;
-        u64 total_size_bytes = 0;
-        u64 largest_copy = 0;
-        DAddr buffer_start = buffer.CpuAddr();
-        memory_tracker.ForEachUploadRange(device_addr, size, [&](u64 device_addr_out, u64 range_size) {
-            copies.push_back(BufferCopy{
-                .src_offset = total_size_bytes,
-                .dst_offset = device_addr_out - buffer_start,
-                .size = range_size,
-            });
-            total_size_bytes += range_size;
-            largest_copy = (std::max)(largest_copy, range_size);
-        });
-        if (total_size_bytes == 0) {
-            return true;
-        }
-        const std::span<BufferCopy> copies_span(copies.data(), copies.size());
-        UploadMemory(buffer, total_size_bytes, largest_copy, copies_span);
-        return false;
-    }
 }
 
 template <class P>

src/video_core/gpu.cpp

@@ -227,18 +227,18 @@ struct GPU::Impl {
 
     /// Notify rasterizer that any caches of the specified region should be invalidated
     void InvalidateRegion(DAddr addr, u64 size, bool preserve_gpu_writes) {
-        if (preserve_gpu_writes) {
+        if (Settings::values.nce_invalidation_gpu_readback.GetValue()) {
             VideoCore::RasterizerInterface* rasterizer = renderer->ReadRasterizer();
-            if (!rasterizer->TryCacheInvalidation(addr, size)) {
+            if (preserve_gpu_writes && rasterizer->MustFlushRegion(addr, size, VideoCommon::CacheType::BufferCache)) {
                 const u64 fence = RequestSyncOperation([rasterizer, addr, size] {
                     rasterizer->FlushRegion(addr, size, VideoCommon::CacheType::BufferCache);
                     rasterizer->OnCacheInvalidation(addr, size);
                 });
                 gpu_thread.TickGPU(is_async);
                 WaitForSyncOperation(fence);
-            }
                 return;
             }
+        }
         gpu_thread.InvalidateRegion(addr, size);
     }
 

src/video_core/rasterizer_interface.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
 
@@ -114,12 +111,6 @@ public:
     /// Notify rasterizer that any caches of the specified region are desync with guest
     virtual void OnCacheInvalidation(PAddr addr, u64 size) = 0;
 
-    virtual bool TryCacheInvalidation(DAddr addr, u64 size) {
-        if (MustFlushRegion(addr, size, VideoCommon::CacheType::BufferCache)) return false;
-        OnCacheInvalidation(addr, size);
-        return true;
-    }
-
     virtual bool OnCPUWrite(PAddr addr, u64 size) = 0;
 
     /// Sync memory between guest and host.

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -414,11 +414,6 @@ GraphicsPipeline* ShaderCache::BuiltPipeline(GraphicsPipeline* pipeline) const n
     if (!use_asynchronous_shaders) {
         return pipeline;
     }
-    // If something is using depth, we can assume that games are not rendering anything which
-    // will be used one time.
-    if (maxwell3d->regs.zeta_enable) {
-        return nullptr;
-    }
     // If games are using a small index count, we can assume these are full screen quads.
     // Usually these shaders are only used once for building textures so we can assume they
     // can't be built async

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -447,9 +447,7 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
                                        driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE ||
                                        driver_id == VK_DRIVER_ID_MESA_RADV ||
                                        driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS ||
-                                       driver_id == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA ||
+                                       driver_id == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA,
-                                       driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY ||
-                                       driver_id == VK_DRIVER_ID_MESA_TURNIP,
 
         .has_broken_spirv_clamp = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS,
         .has_broken_spirv_position_input = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY,
@@ -789,11 +787,6 @@ GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const
     if (!use_asynchronous_shaders) {
         return pipeline;
     }
-    // If something is using depth, we can assume that games are not rendering anything which
-    // will be used one time.
-    if (maxwell3d->regs.zeta_enable) {
-        return nullptr;
-    }
     // If games are using a small index count, we can assume these are full screen quads.
     // Usually these shaders are only used once for building textures so we can assume they
     // can't be built async

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -798,18 +798,6 @@ void RasterizerVulkan::OnCacheInvalidation(DAddr addr, u64 size) {
     pipeline_cache.InvalidateRegion(addr, size);
 }
 
-bool RasterizerVulkan::TryCacheInvalidation(DAddr addr, u64 size) {
-    if (addr == 0 || size == 0) return true;
-    {
-        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
-        if (buffer_cache.IsRegionGpuModified(addr, size)) return false;
-        texture_cache.WriteMemory(addr, size);
-        buffer_cache.WriteMemory(addr, size);
-    }
-    pipeline_cache.InvalidateRegion(addr, size);
-    return true;
-}
-
 void RasterizerVulkan::InvalidateGPUCache() {
     gpu.InvalidateGPUCache();
 }

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -106,7 +106,6 @@ public:
                           VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
     void InnerInvalidation(std::span<const std::pair<DAddr, std::size_t>> sequences) override;
     void OnCacheInvalidation(DAddr addr, u64 size) override;
-    bool TryCacheInvalidation(DAddr addr, u64 size) override;
     bool OnCPUWrite(DAddr addr, u64 size) override;
     void InvalidateGPUCache() override;
     void UnmapMemory(DAddr addr, u64 size) override;

src/yuzu/configuration/configure_cpu.cpp

@@ -34,7 +34,7 @@ ConfigureCpu::ConfigureCpu(const Core::System& system_,
     connect(backend_combobox, qOverload<int>(&QComboBox::currentIndexChanged), this,
             &ConfigureCpu::UpdateGroup);
 
-#ifdef HAS_NCE
+#if defined(HAS_NCE) || defined(HAS_SCARLETNX)
     ui->backend_group->setVisible(true);
 #endif
 }

src/yuzu/main_window.cpp

@@ -2797,9 +2797,9 @@ void MainWindow::IncrementInstallProgress() {
 
 void MainWindow::OnMenuInstallToNAND() {
     const QString file_filter =
-        tr("Installable Switch File (*.nca *.nsp *.nsz *.xci *.xcz);;Nintendo Content Archive "
+        tr("Installable Switch File (*.nca *.nsp *.xci);;Nintendo Content Archive "
-           "(*.nca);;Nintendo Submission Package (*.nsp *.nsz);;NX Cartridge "
+           "(*.nca);;Nintendo Submission Package (*.nsp);;NX Cartridge "
-           "Image (*.xci *.xcz)");
+           "Image (*.xci)");
 
     QStringList filenames = QFileDialog::getOpenFileNames(
         this, tr("Install Files"), QString::fromStdString(UISettings::values.roms_path),
@@ -2856,10 +2856,7 @@ void MainWindow::OnMenuInstallToNAND() {
         QFuture<ContentManager::InstallResult> future;
         ContentManager::InstallResult result;
 
-        if (file.endsWith(QStringLiteral("nsp"), Qt::CaseInsensitive) ||
+        if (file.endsWith(QStringLiteral("nsp"), Qt::CaseInsensitive)) {
-            file.endsWith(QStringLiteral("nsz"), Qt::CaseInsensitive) ||
-            file.endsWith(QStringLiteral("xci"), Qt::CaseInsensitive) ||
-            file.endsWith(QStringLiteral("xcz"), Qt::CaseInsensitive)) {
             const auto progress_callback = [this](size_t size, size_t progress) {
                 emit UpdateInstallProgress();
                 if (install_progress->wasCanceled()) {