return release early fences

Thanks, Debian.

Signed-off-by: crueter <crueter@eden-emu.dev>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3539

CMakeModules/CPMUtil.cmake

@@ -84,7 +84,11 @@ function(SystemPackageViable JSON_NAME)
     parse_object(${object})
 
     string(REPLACE " " ";" find_args "${find_args}")
-    find_package(${package} ${version} ${find_args} QUIET NO_POLICY_SCOPE)
+    if (${package}_FORCE_BUNDLED)
+        set(${package}_FOUND OFF)
+    else()
+        find_package(${package} ${version} ${find_args} QUIET NO_POLICY_SCOPE)
+    endif()
 
     set(${pkg}_VIABLE ${${package}_FOUND} PARENT_SCOPE)
     set(${pkg}_PACKAGE ${package} PARENT_SCOPE)

externals/cpmfile.json

@@ -150,7 +150,7 @@
         "package": "SDL2",
         "name": "SDL2",
         "repo": "crueter-ci/SDL2",
-        "version": "2.32.10-a65111bd2d",
+        "version": "2.32.10-cf5dabd6ea",
         "min_version": "2.26.4"
     },
     "catch2": {
@@ -184,7 +184,6 @@
         "repo": "libsdl-org/SDL",
         "tag": "release-%VERSION%",
         "hash": "d5622d6bb7266f7942a7b8ad43e8a22524893bf0c2ea1af91204838d9b78d32768843f6faa248757427b8404b8c6443776d4afa6b672cd8571a4e0c03a829383",
-        "key": "generic",
         "bundled": true,
         "git_version": "2.32.10",
         "skip_updates": true
@@ -194,7 +193,6 @@
         "repo": "libsdl-org/SDL",
         "sha": "cc016b0046",
         "hash": "b8d9873446cdb922387471df9968e078714683046674ef0d0edddf8e25da65a539a3bae83d635496b970237f90b07b36a69f8d7855d450de59311d6d6e8c3dbc",
-        "key": "steamdeck",
         "bundled": true,
         "skip_updates": "true"
     },

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

@@ -23,6 +23,7 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting {
     RENDERER_USE_DISK_SHADER_CACHE("use_disk_shader_cache"),
     RENDERER_FORCE_MAX_CLOCK("force_max_clock"),
     RENDERER_ASYNCHRONOUS_SHADERS("use_asynchronous_shaders"),
+    RENDERER_EARLY_RELEASE_FENCES("early_release_fences"),
     RENDERER_REACTIVE_FLUSHING("use_reactive_flushing"),
     ENABLE_BUFFER_HISTORY("enable_buffer_history"),
     SYNC_MEMORY_OPERATIONS("sync_memory_operations"),

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

@@ -679,6 +679,13 @@ abstract class SettingsItem(
                     descriptionId = R.string.renderer_asynchronous_shaders_description
                 )
             )
+            put(
+                SwitchSetting(
+                    BooleanSetting.RENDERER_EARLY_RELEASE_FENCES,
+                    titleId = R.string.renderer_early_release_fences,
+                    descriptionId = R.string.renderer_early_release_fences_description
+                )
+            )
             put(
                 SingleChoiceSetting(
                     IntSetting.FAST_GPU_TIME,

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

@@ -285,6 +285,7 @@ class SettingsFragmentPresenter(
             add(BooleanSetting.SKIP_CPU_INNER_INVALIDATION.key)
             add(BooleanSetting.FIX_BLOOM_EFFECTS.key)
             add(BooleanSetting.RENDERER_ASYNCHRONOUS_SHADERS.key)
+            add(BooleanSetting.RENDERER_EARLY_RELEASE_FENCES.key)
             add(SettingsItem.GPU_UNSWIZZLE_COMBINED)
 
             add(HeaderSetting(R.string.extensions))

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

@@ -507,9 +507,11 @@
     <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="fix_bloom_effects">Fix Bloom Effects</string>
-    <string name="fix_bloom_effects_description">Reduces bloom blur in LA/EOW (Adreno 700), removes bloom in Burnout</string>
+    <string name="fix_bloom_effects_description">Reduces bloom blur in LA/EOW (Adreno 700), removes bloom in Burnout. Warning: may cause graphical artifacts in other games.</string>
     <string name="renderer_asynchronous_shaders">Use asynchronous shaders</string>
     <string name="renderer_asynchronous_shaders_description">Compiles shaders asynchronously. This may reduce stutters but may also introduce glitches.</string>
+    <string name="renderer_early_release_fences">Release Fences Early</string>
+    <string name="renderer_early_release_fences_description">Fixes crashes and freezes in some games, may cause issues with Unreal Engine games.</string>
     <string name="gpu_unswizzle_settings">GPU Unswizzle Settings</string>
     <string name="gpu_unswizzle_settings_description">Configure GPU-based texture unswizzling parameters or disable it entirely. Adjust these settings to balance performance and texture loading quality.</string>
     <string name="gpu_unswizzle_enable">Enable GPU Unswizzle</string>

src/common/settings.h

@@ -546,6 +546,16 @@ struct Values {
     SwitchableSetting<bool> use_asynchronous_shaders{linkage, false, "use_asynchronous_shaders",
                                                      Category::RendererHacks};
 
+#ifdef ANDROID
+    SwitchableSetting<bool> early_release_fences{linkage,
+                                                     false,
+                                                     "early_release_fences",
+                                                     Category::RendererAdvanced,
+                                                     Specialization::Default,
+                                                     true,
+                                                     true};
+#endif
+
     SwitchableSetting<GpuUnswizzleSize> gpu_unswizzle_texture_size{linkage,
                                                   GpuUnswizzleSize::Large,
                                                   "gpu_unswizzle_texture_size",
@@ -568,13 +578,7 @@ struct Values {
                                                   Category::RendererHacks};
 
     SwitchableSetting<ExtendedDynamicState> dyna_state{linkage,
-#if defined (_WIN32)
-                                           ExtendedDynamicState::EDS3,
-#elif defined (__FreeBSD__)
-                                           ExtendedDynamicState::EDS3,
-#elif defined (ANDROID)
-                                           ExtendedDynamicState::Disabled,
-#elif defined (__APPLE__)
+#if defined (ANDROID) || defined (__APPLE__)
                                            ExtendedDynamicState::Disabled,
 #else
                                            ExtendedDynamicState::EDS2,

src/core/arm/dynarmic/arm_dynarmic_32.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
@@ -38,10 +38,16 @@ u64 DynarmicCallbacks32::MemoryRead64(u32 vaddr) {
     CheckMemoryAccess(vaddr, 8, Kernel::DebugWatchpointType::Read);
     return m_memory.Read64(vaddr);
 }
+
 std::optional<u32> DynarmicCallbacks32::MemoryReadCode(u32 vaddr) {
     if (!m_memory.IsValidVirtualAddressRange(vaddr, sizeof(u32)))
         return std::nullopt;
-    return m_memory.Read32(vaddr);
+    auto const aligned_vaddr = vaddr & ~Core::Memory::YUZU_PAGEMASK;
+    if (last_code_addr != aligned_vaddr) {
+        m_memory.ReadBlock(aligned_vaddr, &cached_code_page, sizeof(cached_code_page));
+        last_code_addr = aligned_vaddr;
+    }
+    return cached_code_page.inst[(vaddr & Core::Memory::YUZU_PAGEMASK) / sizeof(u32)];
 }
 
 void DynarmicCallbacks32::MemoryWrite8(u32 vaddr, u8 value) {

src/core/arm/dynarmic/arm_dynarmic_32.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
@@ -7,6 +7,7 @@
 #pragma once
 
 #include <dynarmic/interface/A32/a32.h>
+#include <dynarmic/interface/code_page.h>
 
 #include "core/arm/arm_interface.h"
 #include "core/arm/dynarmic/dynarmic_exclusive_monitor.h"
@@ -49,6 +50,9 @@ public:
     u64 GetTicksRemaining() override;
     bool CheckMemoryAccess(u64 addr, u64 size, Kernel::DebugWatchpointType type);
     void ReturnException(u32 pc, Dynarmic::HaltReason hr);
+    //
+    Dynarmic::CodePage cached_code_page;
+    u64 last_code_addr = 0;
     ArmDynarmic32& m_parent;
     Core::Memory::Memory& m_memory;
     Kernel::KProcess* m_process{};

src/core/arm/dynarmic/arm_dynarmic_64.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
@@ -41,10 +41,17 @@ Dynarmic::A64::Vector DynarmicCallbacks64::MemoryRead128(u64 vaddr) {
     CheckMemoryAccess(vaddr, 16, Kernel::DebugWatchpointType::Read);
     return {m_memory.Read64(vaddr), m_memory.Read64(vaddr + 8)};
 }
+
 std::optional<u32> DynarmicCallbacks64::MemoryReadCode(u64 vaddr) {
     if (!m_memory.IsValidVirtualAddressRange(vaddr, sizeof(u32)))
         return std::nullopt;
-    return m_memory.Read32(vaddr);
+//    return m_memory.Read32(vaddr);
+    auto const aligned_vaddr = vaddr & ~Core::Memory::YUZU_PAGEMASK;
+    if (last_code_addr != aligned_vaddr) {
+        m_memory.ReadBlock(aligned_vaddr, &cached_code_page, sizeof(cached_code_page));
+        last_code_addr = aligned_vaddr;
+    }
+    return cached_code_page.inst[(vaddr & Core::Memory::YUZU_PAGEMASK) / sizeof(u32)];
 }
 
 void DynarmicCallbacks64::MemoryWrite8(u64 vaddr, u8 value) {

src/core/arm/dynarmic/arm_dynarmic_64.h

@@ -11,6 +11,7 @@
 #include <ankerl/unordered_dense.h>
 
 #include <dynarmic/interface/A64/a64.h>
+#include <dynarmic/interface/code_page.h>
 #include "common/common_types.h"
 #include "common/hash.h"
 #include "core/arm/arm_interface.h"
@@ -61,6 +62,8 @@ public:
     bool CheckMemoryAccess(u64 addr, u64 size, Kernel::DebugWatchpointType type);
     void ReturnException(u64 pc, Dynarmic::HaltReason hr);
 
+    Dynarmic::CodePage cached_code_page;
+    u64 last_code_addr = 0;
     ArmDynarmic64& m_parent;
     Core::Memory::Memory& m_memory;
     u64 m_tpidrro_el0{};

src/core/hle/service/jit/jit_context.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
@@ -10,6 +10,7 @@
 #include <boost/icl/interval_set.hpp>
 #include <dynarmic/interface/A64/a64.h>
 #include <dynarmic/interface/A64/config.h>
+#include <dynarmic/interface/code_page.h>
 
 #include "common/alignment.h"
 #include "common/common_funcs.h"
@@ -46,6 +47,15 @@ public:
         : memory{memory_}, local_memory{local_memory_},
           mapped_ranges{mapped_ranges_}, parent{parent_} {}
 
+    std::optional<std::uint32_t> MemoryReadCode(VAddr vaddr) override {
+        static_assert(Core::Memory::YUZU_PAGESIZE == Dynarmic::CODE_PAGE_SIZE);
+        auto const aligned_vaddr = vaddr & ~Core::Memory::YUZU_PAGEMASK;
+        if (last_code_addr != aligned_vaddr) {
+            cached_code_page = ReadMemory<Dynarmic::CodePage>(aligned_vaddr);
+            last_code_addr = aligned_vaddr;
+        }
+        return cached_code_page.inst[(vaddr & Core::Memory::YUZU_PAGEMASK) / sizeof(u32)];
+    }
     u8 MemoryRead8(u64 vaddr) override {
         return ReadMemory<u8>(vaddr);
     }
@@ -116,8 +126,7 @@ public:
         return 0;
     }
 
-    template <class T>
-    T ReadMemory(u64 vaddr) {
+    template<typename T> T ReadMemory(u64 vaddr) {
         T ret{};
         if (boost::icl::contains(mapped_ranges, vaddr)) {
             memory.ReadBlock(vaddr, &ret, sizeof(T));
@@ -146,6 +155,9 @@ private:
     std::vector<u8>& local_memory;
     IntervalSet& mapped_ranges;
     JITContextImpl& parent;
+
+    Dynarmic::CodePage cached_code_page;
+    u64 last_code_addr = 0;
 };
 
 class JITContextImpl {

src/core/memory.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: 2015 Citra Emulator Project
@@ -36,8 +36,7 @@
 
 namespace Core::Memory {
 
-static inline bool AddressSpaceContains(const Common::PageTable& table, const Common::ProcessAddress addr,
-                          const std::size_t size) {
+static inline bool AddressSpaceContains(const Common::PageTable& table, const Common::ProcessAddress addr, const std::size_t size) {
     const Common::ProcessAddress max_addr = 1ULL << table.GetAddressSpaceBits();
     return addr + size >= addr && addr + size <= max_addr;
 }

src/dynarmic/src/dynarmic/backend/x64/a32_interface.cpp

@@ -65,11 +65,10 @@ static Optimization::PolyfillOptions GenPolyfillOptions(const BlockOfCode& code)
 
 struct Jit::Impl {
     Impl(Jit* jit, A32::UserConfig conf) noexcept
-        : ir_block{LocationDescriptor(0, PSR(0), FPSCR(0), false)}
-        , conf(std::move(conf))
-        , block_of_code(GenRunCodeCallbacks(conf.callbacks, &GetCurrentBlockThunk, this, conf), JitStateInfo{jit_state}, conf.code_cache_size, GenRCP(conf))
+        : block_of_code(GenRunCodeCallbacks(conf.callbacks, &GetCurrentBlockThunk, this, conf), JitStateInfo{jit_state}, conf.code_cache_size, GenRCP(conf))
         , emitter(block_of_code, conf, jit)
         , polyfill_options(GenPolyfillOptions(block_of_code))
+        , conf(std::move(conf))
         , jit_interface(jit)
     {}
 
@@ -204,8 +203,7 @@ private:
     }
 
     A32EmitX64::BlockDescriptor GetBasicBlock(IR::LocationDescriptor descriptor) {
-        auto block = emitter.GetBasicBlock(descriptor);
-        if (block)
+        if (auto block = emitter.GetBasicBlock(descriptor))
             return *block;
 
         constexpr size_t MINIMUM_REMAINING_CODESIZE = 1 * 1024 * 1024;
@@ -215,8 +213,10 @@ private:
         }
         block_of_code.EnsureMemoryCommitted(MINIMUM_REMAINING_CODESIZE);
 
-        ir_block.Reset(descriptor);
-        A32::Translate(ir_block, A32::LocationDescriptor{descriptor}, conf.callbacks, {conf.arch_version, conf.define_unpredictable_behaviour, conf.hook_hint_instructions});
+        // LocationDescriptor ctor() does important ops (like tflags) do not skip
+        auto const arch_descriptor = A32::LocationDescriptor{descriptor};
+        ir_block.Reset(arch_descriptor);
+        A32::Translate(ir_block, arch_descriptor, conf.callbacks, {conf.arch_version, conf.define_unpredictable_behaviour, conf.hook_hint_instructions});
         Optimization::Optimize(ir_block, conf, polyfill_options);
         return emitter.Emit(ir_block);
     }
@@ -243,12 +243,13 @@ private:
         }
     }
 
-    IR::Block ir_block;
-    const A32::UserConfig conf;
+    IR::Block ir_block = {LocationDescriptor(0, PSR(0), FPSCR(0), false)};
     A32JitState jit_state;
     BlockOfCode block_of_code;
     A32EmitX64 emitter;
     Optimization::PolyfillOptions polyfill_options;
+    // Keep it here, you don't wanna mess with the fuckery that's initializer lists
+    const A32::UserConfig conf;
     Jit* jit_interface;
 
     // Requests made during execution to invalidate the cache are queued up here.

src/dynarmic/src/dynarmic/backend/x64/a64_interface.cpp

@@ -62,8 +62,7 @@ static Optimization::PolyfillOptions GenPolyfillOptions(const BlockOfCode& code)
 struct Jit::Impl final {
 public:
     Impl(Jit* jit, UserConfig conf)
-        : ir_block{LocationDescriptor(0, FP::FPCR(0), false)}
-        , conf(conf)
+        : conf(conf)
         , block_of_code(GenRunCodeCallbacks(conf.callbacks, &GetCurrentBlockThunk, this, conf), JitStateInfo{jit_state}, conf.code_cache_size, GenRCP(conf))
         , emitter(block_of_code, conf, jit)
         , polyfill_options(GenPolyfillOptions(block_of_code))
@@ -257,8 +256,8 @@ private:
         return GetBlock(A64::LocationDescriptor{GetCurrentLocation()}.SetSingleStepping(true));
     }
 
-    CodePtr GetBlock(IR::LocationDescriptor current_location) {
-        if (auto block = emitter.GetBasicBlock(current_location))
+    CodePtr GetBlock(IR::LocationDescriptor descriptor) {
+        if (auto block = emitter.GetBasicBlock(descriptor))
             return block->entrypoint;
 
         constexpr size_t MINIMUM_REMAINING_CODESIZE = 1 * 1024 * 1024;
@@ -271,8 +270,10 @@ private:
 
         // JIT Compile
         const auto get_code = [this](u64 vaddr) { return conf.callbacks->MemoryReadCode(vaddr); };
-        ir_block.Reset(current_location);
-        A64::Translate(ir_block, A64::LocationDescriptor{current_location}, get_code, {conf.define_unpredictable_behaviour, conf.wall_clock_cntpct});
+        // LocationDescriptor ctor() does important ops (like tflags) do not skip
+        auto const arch_descriptor = A64::LocationDescriptor{descriptor};
+        ir_block.Reset(arch_descriptor);
+        A64::Translate(ir_block, arch_descriptor, get_code, {conf.define_unpredictable_behaviour, conf.wall_clock_cntpct});
         Optimization::Optimize(ir_block, conf, polyfill_options);
         return emitter.Emit(ir_block).entrypoint;
     }
@@ -299,7 +300,7 @@ private:
         }
     }
 
-    IR::Block ir_block;
+    IR::Block ir_block = {LocationDescriptor(0, FP::FPCR(0), false)};
     const UserConfig conf;
     A64JitState jit_state;
     BlockOfCode block_of_code;

src/dynarmic/src/dynarmic/interface/code_page.h

@@ -0,0 +1,17 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <cstddef>
+#include <cstdint>
+
+namespace Dynarmic {
+
+/// @brief Smallest valid page (may change for Apple?)
+constexpr inline uint64_t CODE_PAGE_SIZE = 4096;
+struct CodePage {
+    uint32_t inst[CODE_PAGE_SIZE / sizeof(uint32_t)];
+};
+
+}

src/dynarmic/src/dynarmic/ir/basic_block.cpp

@@ -22,13 +22,10 @@
 
 namespace Dynarmic::IR {
 
-Block::Block(const LocationDescriptor& location)
-    : location{location},
-    end_location{location},
-    cond{Cond::AL}
-{
-
-}
+Block::Block(LocationDescriptor location) noexcept
+    : location{location}
+    , end_location{location}
+{}
 
 /// Prepends a new instruction to this basic block before the insertion point,
 /// handling any allocations necessary to do so.
@@ -60,6 +57,21 @@ Block::iterator Block::PrependNewInst(iterator insertion_point, Opcode opcode, s
     return instructions.insert_before(insertion_point, inst);
 }
 
+void Block::Reset(LocationDescriptor location_) noexcept {
+    mcl::intrusive_list<IR::Inst> tmp = {};
+    instructions.swap(tmp);
+    inlined_inst.clear();
+    pooled_inst.clear();
+    cond_failed.reset();
+    location = location_;
+    end_location = location_;
+    cond = Cond::AL;
+    terminal = Term::Invalid{};
+    cond_failed_cycle_count = 0;
+    cycle_count = 0;
+    ASSERT(instructions.size() == 0);
+}
+
 static std::string TerminalToString(const Terminal& terminal_variant) noexcept {
     struct : boost::static_visitor<std::string> {
         std::string operator()(const Term::Invalid&) const {
@@ -123,11 +135,11 @@ std::string DumpBlock(const IR::Block& block) noexcept {
         case Type::A32ExtReg: return A32::ExtRegToString(arg.GetA32ExtRegRef());
         case Type::A64Reg: return A64::RegToString(arg.GetA64RegRef());
         case Type::A64Vec: return A64::VecToString(arg.GetA64VecRef());
-        case Type::CoprocInfo: return fmt::format("#<coproc>");
-        case Type::NZCVFlags: return fmt::format("#<NZCV flags>");
-        case Type::Cond: return fmt::format("#<cond={}>", A32::CondToString(arg.GetCond()));
-        case Type::Table: return fmt::format("#<table>");
-        case Type::AccType: return fmt::format("#<acc-type={}>", u32(arg.GetAccType()));
+        case Type::CoprocInfo: return fmt::format("$coproc{}", arg.GetCoprocInfo()[0]);
+        case Type::NZCVFlags: return fmt::format("$nzcv");
+        case Type::Cond: return fmt::format("$cond={}", A32::CondToString(arg.GetCond()));
+        case Type::Table: return fmt::format("$table");
+        case Type::AccType: return fmt::format("$acc-type={}", u32(arg.GetAccType()));
         default: return fmt::format("<unknown immediate type {}>", arg.GetType());
         }
     };

src/dynarmic/src/dynarmic/ir/basic_block.h

@@ -43,7 +43,7 @@ public:
     using reverse_iterator = instruction_list_type::reverse_iterator;
     using const_reverse_iterator = instruction_list_type::const_reverse_iterator;
 
-    explicit Block(const LocationDescriptor& location);
+    Block(LocationDescriptor location) noexcept;
     ~Block() = default;
     Block(const Block&) = delete;
     Block& operator=(const Block&) = delete;
@@ -58,6 +58,7 @@ public:
         return PrependNewInst(instructions.end(), opcode, args);
     }
     iterator PrependNewInst(iterator insertion_point, Opcode op, std::initializer_list<Value> args) noexcept;
+    void Reset(LocationDescriptor location_) noexcept;
 
     /// Gets a mutable reference to the instruction list for this basic block.
     inline instruction_list_type& Instructions() noexcept {
@@ -140,18 +141,6 @@ public:
         return cycle_count;
     }
 
-    inline void Reset(LocationDescriptor location_) noexcept {
-        inlined_inst.clear();
-        pooled_inst.clear();
-        cond_failed.reset();
-        location = location_;
-        terminal = Term::Invalid{};
-        cond_failed_cycle_count = 0;
-        cycle_count = 0;
-        instruction_list_type tmp{};
-        instructions.swap(tmp);
-    }
-
     /// "Hot cache" for small blocks so we don't call global allocator
     boost::container::static_vector<Inst, 30> inlined_inst;
     /// List of instructions in this block.
@@ -165,7 +154,7 @@ public:
     /// Description of the end location of this block
     LocationDescriptor end_location;
     /// Conditional to pass in order to execute this block
-    Cond cond;
+    Cond cond = Cond::AL;
     /// Terminal instruction of this block.
     Terminal terminal = Term::Invalid{};
     /// Number of cycles this block takes to execute if the conditional fails.

src/dynarmic/src/dynarmic/ir/opt_passes.cpp

@@ -1042,6 +1042,11 @@ static void FoldZeroExtendXToLong(IR::Inst& inst) {
 static void ConstantPropagation(IR::Block& block) {
     for (auto& inst : block.instructions) {
         auto const opcode = inst.GetOpcode();
+        // skip NZCV so we dont end up discarding side effects :)
+        // TODO(lizzie): hey stupid maybe fix the A64 codegen for folded constants AND
+        // redirect the mfer properly?!??! just saying :)
+        if (IR::MayGetNZCVFromOp(opcode) && inst.GetAssociatedPseudoOperation(IR::Opcode::GetNZCVFromOp))
+            continue;
         switch (opcode) {
         case Op::LeastSignificantWord:
             FoldLeastSignificantWord(inst);

src/video_core/fence_manager.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
@@ -72,15 +72,26 @@ public:
     }
 
     void SignalFence(std::function<void()>&& func) {
-        if constexpr (!can_async_check) {
-            TryReleasePendingFences<false>();
-        }
+#ifdef ANDROID
+        const bool early_release_fences = Settings::values.early_release_fences.GetValue();
+#else
+        constexpr bool early_release_fences = false;
+#endif
         const bool should_flush = ShouldFlush();
         const bool delay_fence = Settings::IsGPULevelHigh() || (Settings::IsGPULevelMedium() && should_flush);
+        if (!can_async_check || (!delay_fence && early_release_fences)) {
+            TryReleasePendingFences<false>();
+        }
         CommitAsyncFlushes();
         TFence new_fence = CreateFence(!should_flush);
-        if constexpr (can_async_check) {
-            guard.lock();
+        if (!early_release_fences){
+            if constexpr (can_async_check) {
+                guard.lock();
+            }
+        } else {
+            if (delay_fence) {
+                guard.lock();
+            }
         }
         if (delay_fence) {
             uncommitted_operations.emplace_back(std::move(func));
@@ -94,9 +105,16 @@ public:
         if (should_flush) {
             rasterizer.FlushCommands();
         }
-        if constexpr (can_async_check) {
-            guard.unlock();
-            cv.notify_all();
+        if (!early_release_fences){
+            if constexpr (can_async_check) {
+                guard.unlock();
+                cv.notify_all();
+            }
+        } else {
+            if (delay_fence) {
+                guard.unlock();
+                cv.notify_all();
+            }
         }
         rasterizer.InvalidateGPUCache();
     }

src/video_core/renderer_vulkan/vk_descriptor_pool.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
@@ -20,7 +20,7 @@
 namespace Vulkan {
 
 // Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines
-constexpr size_t SETS_GROW_RATE = 32; //test difference between 16 and 32
+constexpr size_t SETS_GROW_RATE = 16;
 constexpr s32 SCORE_THRESHOLD = 3;
 
 struct DescriptorBank {
@@ -29,12 +29,9 @@ struct DescriptorBank {
 };
 
 bool DescriptorBankInfo::IsSuperset(const DescriptorBankInfo& subset) const noexcept {
-    return uniform_buffers >= subset.uniform_buffers &&
-           storage_buffers >= subset.storage_buffers &&
-           texture_buffers >= subset.texture_buffers &&
-           image_buffers >= subset.image_buffers &&
-           textures >= subset.textures &&
-           images >= subset.images;
+    return uniform_buffers >= subset.uniform_buffers && storage_buffers >= subset.storage_buffers &&
+           texture_buffers >= subset.texture_buffers && image_buffers >= subset.image_buffers &&
+           textures >= subset.textures && images >= subset.images;
 }
 
 template <typename Descriptors>
@@ -48,19 +45,6 @@ static u32 Accumulate(const Descriptors& descriptors) {
 
 static DescriptorBankInfo MakeBankInfo(std::span<const Shader::Info> infos) {
     DescriptorBankInfo bank;
-    if (infos.size() == 1) {
-        const auto& info = infos.front();
-        const auto acc = [](const auto& ds){ u32 c=0; for (const auto& d: ds) c+=d.count; return c; };
-        bank.uniform_buffers += acc(info.constant_buffer_descriptors);
-        bank.storage_buffers += acc(info.storage_buffers_descriptors);
-        bank.texture_buffers += acc(info.texture_buffer_descriptors);
-        bank.image_buffers   += acc(info.image_buffer_descriptors);
-        bank.textures        += acc(info.texture_descriptors);
-        bank.images          += acc(info.image_descriptors);
-        bank.score = bank.uniform_buffers + bank.storage_buffers + bank.texture_buffers +
-        bank.image_buffers + bank.textures + bank.images;
-        return bank;
-    }
     for (const Shader::Info& info : infos) {
         bank.uniform_buffers += Accumulate(info.constant_buffer_descriptors);
         bank.storage_buffers += Accumulate(info.storage_buffers_descriptors);
@@ -118,22 +102,13 @@ void DescriptorAllocator::Allocate(size_t begin, size_t end) {
 }
 
 vk::DescriptorSets DescriptorAllocator::AllocateDescriptors(size_t count) {
-    std::array<VkDescriptorSetLayout, 64> stack{};
-    const VkDescriptorSetLayout* p_layouts = nullptr;
-    std::vector<VkDescriptorSetLayout> heap;
-    if (count <= stack.size()) {
-        stack.fill(layout);
-        p_layouts = stack.data();
-    } else {
-        heap.assign(count, layout);
-        p_layouts = heap.data();
-    }
+    const std::vector<VkDescriptorSetLayout> layouts(count, layout);
     VkDescriptorSetAllocateInfo allocate_info{
         .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
         .pNext = nullptr,
         .descriptorPool = *bank->pools.back(),
         .descriptorSetCount = static_cast<u32>(count),
-        .pSetLayouts = p_layouts,
+        .pSetLayouts = layouts.data(),
     };
     vk::DescriptorSets new_sets = bank->pools.back().Allocate(allocate_info);
     if (!new_sets.IsOutOfPoolMemory()) {
@@ -171,58 +146,21 @@ DescriptorAllocator DescriptorPool::Allocator(VkDescriptorSetLayout layout,
 }
 
 DescriptorBank& DescriptorPool::Bank(const DescriptorBankInfo& reqs) {
-    {
-        std::scoped_lock lk(cache_mutex);
-        DescriptorBank* best = nullptr; u64 best_stamp = 0;
-        for (const auto& e : cache_) {
-            if (!e.bank) continue;
-            if (std::abs(e.info.score - reqs.score) < SCORE_THRESHOLD && e.info.IsSuperset(reqs)) {
-                if (e.stamp >= best_stamp) { best_stamp = e.stamp; best = e.bank; }
-            }
-        }
-        if (best) return *best;
-    }
     std::shared_lock read_lock{banks_mutex};
     const auto it = std::ranges::find_if(bank_infos, [&reqs](const DescriptorBankInfo& bank) {
         return std::abs(bank.score - reqs.score) < SCORE_THRESHOLD && bank.IsSuperset(reqs);
     });
     if (it != bank_infos.end()) {
-        DescriptorBank& found = *banks[std::distance(bank_infos.begin(), it)].get();
-        read_lock.unlock();
-        // update cache
-        std::scoped_lock lk(cache_mutex);
-        size_t victim = 0; u64 oldest = UINT64_MAX;
-        for (size_t i=0;i<cache_.size();++i) if (cache_[i].stamp < oldest) { oldest = cache_[i].stamp; victim = i; }
-             cache_[victim] = CacheEntry{found.info, &found, ++cache_tick_};
-        return found;
+        return *banks[std::distance(bank_infos.begin(), it)].get();
     }
     read_lock.unlock();
 
     std::unique_lock write_lock{banks_mutex};
-    auto it2 = std::ranges::find_if(bank_infos, [&reqs](const DescriptorBankInfo& bank) {
-          return std::abs(bank.score - reqs.score) < SCORE_THRESHOLD && bank.IsSuperset(reqs);
-    });
-    if (it2 != bank_infos.end()) {
-        DescriptorBank& found = *banks[std::distance(bank_infos.begin(), it2)].get();
-        // update cache
-        std::scoped_lock lk(cache_mutex);
-        size_t victim = 0; u64 oldest = UINT64_MAX;
-        for (size_t i=0;i<cache_.size();++i) if (cache_[i].stamp < oldest) { oldest = cache_[i].stamp; victim = i; }
-        cache_[victim] = CacheEntry{found.info, &found, ++cache_tick_};
-        return found;
-    }
     bank_infos.push_back(reqs);
 
     auto& bank = *banks.emplace_back(std::make_unique<DescriptorBank>());
     bank.info = reqs;
     AllocatePool(device, bank);
-    // update cache
-    {
-         std::scoped_lock lk(cache_mutex);
-         size_t victim = 0; u64 oldest = UINT64_MAX;
-         for (size_t i=0;i<cache_.size();++i) if (cache_[i].stamp < oldest) { oldest = cache_[i].stamp; victim = i; }
-         cache_[victim] = CacheEntry{bank.info, &bank, ++cache_tick_};
-    }
     return bank;
 }
 

src/video_core/renderer_vulkan/vk_descriptor_pool.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
 
@@ -9,8 +6,7 @@
 #include <shared_mutex>
 #include <span>
 #include <vector>
-#include <array>
-#include <mutex>
+
 #include "shader_recompiler/shader_info.h"
 #include "video_core/renderer_vulkan/vk_resource_pool.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
@@ -79,14 +75,6 @@ public:
 
 private:
     DescriptorBank& Bank(const DescriptorBankInfo& reqs);
-    struct CacheEntry {
-        DescriptorBankInfo info{};
-        DescriptorBank* bank{nullptr};
-        u64 stamp{0};
-    };
-    std::mutex cache_mutex{};
-    std::array<CacheEntry, 8> cache_{}; //test and then adjust
-    u64 cache_tick_{0};
 
     const Device& device;
     MasterSemaphore& master_semaphore;

src/video_core/renderer_vulkan/vk_master_semaphore.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
@@ -243,7 +243,11 @@ void MasterSemaphore::WaitThread(std::stop_token token) {
             free_queue.push_front(std::move(fence));
             gpu_tick.store(host_tick, std::memory_order_release);
         }
+#ifdef ANDROID
         free_cv.notify_all();
+#else
+        free_cv.notify_one();
+#endif
     }
 }