35 files changed, 445 insertions, 206 deletions

diff --git a/src/common/logging/backend.cpp b/src/common/logging/backend.cpp
index 949384fd3..e40d117d6 100644
--- a/src/common/logging/backend.cpp
+++ b/src/common/logging/backend.cpp
@@ -18,6 +18,7 @@
 #include "common/fs/fs_paths.h"
 #include "common/fs/path_util.h"
 #include "common/literals.h"
+#include "common/thread.h"
 
 #include "common/logging/backend.h"
 #include "common/logging/log.h"
diff --git a/src/common/lru_cache.h b/src/common/lru_cache.h
new file mode 100644
index 000000000..365488ba5
--- /dev/null
+++ b/src/common/lru_cache.h
@@ -0,0 +1,140 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2+ or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <deque>
+#include <memory>
+#include <type_traits>
+
+#include "common/common_types.h"
+
+namespace Common {
+
+template <class Traits>
+class LeastRecentlyUsedCache {
+    using ObjectType = typename Traits::ObjectType;
+    using TickType = typename Traits::TickType;
+
+    struct Item {
+        ObjectType obj;
+        TickType tick;
+        Item* next{};
+        Item* prev{};
+    };
+
+public:
+    LeastRecentlyUsedCache() : first_item{}, last_item{} {}
+    ~LeastRecentlyUsedCache() = default;
+
+    size_t Insert(ObjectType obj, TickType tick) {
+        const auto new_id = Build();
+        auto& item = item_pool[new_id];
+        item.obj = obj;
+        item.tick = tick;
+        Attach(item);
+        return new_id;
+    }
+
+    void Touch(size_t id, TickType tick) {
+        auto& item = item_pool[id];
+        if (item.tick >= tick) {
+            return;
+        }
+        item.tick = tick;
+        if (&item == last_item) {
+            return;
+        }
+        Detach(item);
+        Attach(item);
+    }
+
+    void Free(size_t id) {
+        auto& item = item_pool[id];
+        Detach(item);
+        item.prev = nullptr;
+        item.next = nullptr;
+        free_items.push_back(id);
+    }
+
+    template <typename Func>
+    void ForEachItemBelow(TickType tick, Func&& func) {
+        static constexpr bool RETURNS_BOOL =
+            std::is_same_v<std::invoke_result<Func, ObjectType>, bool>;
+        Item* iterator = first_item;
+        while (iterator) {
+            if (static_cast<s64>(tick) - static_cast<s64>(iterator->tick) < 0) {
+                return;
+            }
+            Item* next = iterator->next;
+            if constexpr (RETURNS_BOOL) {
+                if (func(iterator->obj)) {
+                    return;
+                }
+            } else {
+                func(iterator->obj);
+            }
+            iterator = next;
+        }
+    }
+
+private:
+    size_t Build() {
+        if (free_items.empty()) {
+            const size_t item_id = item_pool.size();
+            auto& item = item_pool.emplace_back();
+            item.next = nullptr;
+            item.prev = nullptr;
+            return item_id;
+        }
+        const size_t item_id = free_items.front();
+        free_items.pop_front();
+        auto& item = item_pool[item_id];
+        item.next = nullptr;
+        item.prev = nullptr;
+        return item_id;
+    }
+
+    void Attach(Item& item) {
+        if (!first_item) {
+            first_item = &item;
+        }
+        if (!last_item) {
+            last_item = &item;
+        } else {
+            item.prev = last_item;
+            last_item->next = &item;
+            item.next = nullptr;
+            last_item = &item;
+        }
+    }
+
+    void Detach(Item& item) {
+        if (item.prev) {
+            item.prev->next = item.next;
+        }
+        if (item.next) {
+            item.next->prev = item.prev;
+        }
+        if (&item == first_item) {
+            first_item = item.next;
+            if (first_item) {
+                first_item->prev = nullptr;
+            }
+        }
+        if (&item == last_item) {
+            last_item = item.prev;
+            if (last_item) {
+                last_item->next = nullptr;
+            }
+        }
+    }
+
+    std::deque<Item> item_pool;
+    std::deque<size_t> free_items;
+    Item* first_item{};
+    Item* last_item{};
+};
+
+} // namespace Common
diff --git a/src/common/settings.cpp b/src/common/settings.cpp
index 996315999..fd3b639cd 100644
--- a/src/common/settings.cpp
+++ b/src/common/settings.cpp
@@ -59,7 +59,6 @@ void LogSettings() {
     log_setting("Renderer_UseVsync", values.use_vsync.GetValue());
     log_setting("Renderer_ShaderBackend", values.shader_backend.GetValue());
     log_setting("Renderer_UseAsynchronousShaders", values.use_asynchronous_shaders.GetValue());
-    log_setting("Renderer_UseGarbageCollection", values.use_caches_gc.GetValue());
     log_setting("Renderer_AnisotropicFilteringLevel", values.max_anisotropy.GetValue());
     log_setting("Audio_OutputEngine", values.sink_id.GetValue());
     log_setting("Audio_EnableAudioStretching", values.enable_audio_stretching.GetValue());
@@ -143,7 +142,6 @@ void RestoreGlobalState(bool is_powered_on) {
     values.shader_backend.SetGlobal(true);
     values.use_asynchronous_shaders.SetGlobal(true);
     values.use_fast_gpu_time.SetGlobal(true);
-    values.use_caches_gc.SetGlobal(true);
     values.bg_red.SetGlobal(true);
     values.bg_green.SetGlobal(true);
     values.bg_blue.SetGlobal(true);
diff --git a/src/common/settings.h b/src/common/settings.h
index 20769d310..ec4d381e8 100644
--- a/src/common/settings.h
+++ b/src/common/settings.h
@@ -475,7 +475,6 @@ struct Values {
                                                 ShaderBackend::SPIRV, "shader_backend"};
     Setting<bool> use_asynchronous_shaders{false, "use_asynchronous_shaders"};
     Setting<bool> use_fast_gpu_time{true, "use_fast_gpu_time"};
-    Setting<bool> use_caches_gc{false, "use_caches_gc"};
 
     Setting<u8> bg_red{0, "bg_red"};
     Setting<u8> bg_green{0, "bg_green"};
diff --git a/src/core/cpu_manager.cpp b/src/core/cpu_manager.cpp
index 7e195346b..77efcabf0 100644
--- a/src/core/cpu_manager.cpp
+++ b/src/core/cpu_manager.cpp
@@ -21,34 +21,25 @@ namespace Core {
 CpuManager::CpuManager(System& system_) : system{system_} {}
 CpuManager::~CpuManager() = default;
 
-void CpuManager::ThreadStart(CpuManager& cpu_manager, std::size_t core) {
-    cpu_manager.RunThread(core);
+void CpuManager::ThreadStart(std::stop_token stop_token, CpuManager& cpu_manager,
+                             std::size_t core) {
+    cpu_manager.RunThread(stop_token, core);
 }
 
 void CpuManager::Initialize() {
     running_mode = true;
     if (is_multicore) {
         for (std::size_t core = 0; core < Core::Hardware::NUM_CPU_CORES; core++) {
-            core_data[core].host_thread =
-                std::make_unique<std::thread>(ThreadStart, std::ref(*this), core);
+            core_data[core].host_thread = std::jthread(ThreadStart, std::ref(*this), core);
         }
     } else {
-        core_data[0].host_thread = std::make_unique<std::thread>(ThreadStart, std::ref(*this), 0);
+        core_data[0].host_thread = std::jthread(ThreadStart, std::ref(*this), 0);
     }
 }
 
 void CpuManager::Shutdown() {
     running_mode = false;
     Pause(false);
-    if (is_multicore) {
-        for (auto& data : core_data) {
-            data.host_thread->join();
-            data.host_thread.reset();
-        }
-    } else {
-        core_data[0].host_thread->join();
-        core_data[0].host_thread.reset();
-    }
 }
 
 std::function<void(void*)> CpuManager::GetGuestThreadStartFunc() {
@@ -317,7 +308,7 @@ void CpuManager::Pause(bool paused) {
     }
 }
 
-void CpuManager::RunThread(std::size_t core) {
+void CpuManager::RunThread(std::stop_token stop_token, std::size_t core) {
     /// Initialization
     system.RegisterCoreThread(core);
     std::string name;
@@ -361,6 +352,10 @@ void CpuManager::RunThread(std::size_t core) {
             return;
         }
 
+        if (stop_token.stop_requested()) {
+            break;
+        }
+
         auto current_thread = system.Kernel().CurrentScheduler()->GetCurrentThread();
         data.is_running = true;
         Common::Fiber::YieldTo(data.host_context, *current_thread->GetHostContext());
diff --git a/src/core/cpu_manager.h b/src/core/cpu_manager.h
index 140263b09..9d92d4af0 100644
--- a/src/core/cpu_manager.h
+++ b/src/core/cpu_manager.h
@@ -78,9 +78,9 @@ private:
     void SingleCoreRunSuspendThread();
     void SingleCorePause(bool paused);
 
-    static void ThreadStart(CpuManager& cpu_manager, std::size_t core);
+    static void ThreadStart(std::stop_token stop_token, CpuManager& cpu_manager, std::size_t core);
 
-    void RunThread(std::size_t core);
+    void RunThread(std::stop_token stop_token, std::size_t core);
 
     struct CoreData {
         std::shared_ptr<Common::Fiber> host_context;
@@ -89,7 +89,7 @@ private:
         std::atomic<bool> is_running;
         std::atomic<bool> is_paused;
         std::atomic<bool> initialized;
-        std::unique_ptr<std::thread> host_thread;
+        std::jthread host_thread;
     };
 
     std::atomic<bool> running_mode{};
diff --git a/src/core/hle/service/nvflinger/buffer_queue.cpp b/src/core/hle/service/nvflinger/buffer_queue.cpp
index 59ddf6298..b4c3a6099 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.cpp
+++ b/src/core/hle/service/nvflinger/buffer_queue.cpp
@@ -9,17 +9,20 @@
 #include "core/core.h"
 #include "core/hle/kernel/k_writable_event.h"
 #include "core/hle/kernel/kernel.h"
+#include "core/hle/service/kernel_helpers.h"
 #include "core/hle/service/nvflinger/buffer_queue.h"
 
 namespace Service::NVFlinger {
 
-BufferQueue::BufferQueue(Kernel::KernelCore& kernel, u32 id_, u64 layer_id_)
-    : id(id_), layer_id(layer_id_), buffer_wait_event{kernel} {
-    Kernel::KAutoObject::Create(std::addressof(buffer_wait_event));
-    buffer_wait_event.Initialize("BufferQueue:WaitEvent");
+BufferQueue::BufferQueue(Kernel::KernelCore& kernel, u32 id_, u64 layer_id_,
+                         KernelHelpers::ServiceContext& service_context_)
+    : id(id_), layer_id(layer_id_), service_context{service_context_} {
+    buffer_wait_event = service_context.CreateEvent("BufferQueue:WaitEvent");
 }
 
-BufferQueue::~BufferQueue() = default;
+BufferQueue::~BufferQueue() {
+    service_context.CloseEvent(buffer_wait_event);
+}
 
 void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer) {
     ASSERT(slot < buffer_slots);
@@ -41,7 +44,7 @@ void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer)
         .multi_fence = {},
     };
 
-    buffer_wait_event.GetWritableEvent().Signal();
+    buffer_wait_event->GetWritableEvent().Signal();
 }
 
 std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> BufferQueue::DequeueBuffer(u32 width,
@@ -119,7 +122,7 @@ void BufferQueue::CancelBuffer(u32 slot, const Service::Nvidia::MultiFence& mult
     }
     free_buffers_condition.notify_one();
 
-    buffer_wait_event.GetWritableEvent().Signal();
+    buffer_wait_event->GetWritableEvent().Signal();
 }
 
 std::optional<std::reference_wrapper<const BufferQueue::Buffer>> BufferQueue::AcquireBuffer() {
@@ -154,7 +157,7 @@ void BufferQueue::ReleaseBuffer(u32 slot) {
     }
     free_buffers_condition.notify_one();
 
-    buffer_wait_event.GetWritableEvent().Signal();
+    buffer_wait_event->GetWritableEvent().Signal();
 }
 
 void BufferQueue::Connect() {
@@ -169,7 +172,7 @@ void BufferQueue::Disconnect() {
         std::unique_lock lock{queue_sequence_mutex};
         queue_sequence.clear();
     }
-    buffer_wait_event.GetWritableEvent().Signal();
+    buffer_wait_event->GetWritableEvent().Signal();
     is_connect = false;
     free_buffers_condition.notify_one();
 }
@@ -189,11 +192,11 @@ u32 BufferQueue::Query(QueryType type) {
 }
 
 Kernel::KWritableEvent& BufferQueue::GetWritableBufferWaitEvent() {
-    return buffer_wait_event.GetWritableEvent();
+    return buffer_wait_event->GetWritableEvent();
 }
 
 Kernel::KReadableEvent& BufferQueue::GetBufferWaitEvent() {
-    return buffer_wait_event.GetReadableEvent();
+    return buffer_wait_event->GetReadableEvent();
 }
 
 } // namespace Service::NVFlinger
diff --git a/src/core/hle/service/nvflinger/buffer_queue.h b/src/core/hle/service/nvflinger/buffer_queue.h
index 61e337ac5..759247eb0 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.h
+++ b/src/core/hle/service/nvflinger/buffer_queue.h
@@ -24,6 +24,10 @@ class KReadableEvent;
 class KWritableEvent;
 } // namespace Kernel
 
+namespace Service::KernelHelpers {
+class ServiceContext;
+} // namespace Service::KernelHelpers
+
 namespace Service::NVFlinger {
 
 constexpr u32 buffer_slots = 0x40;
@@ -54,7 +58,8 @@ public:
         NativeWindowFormat = 2,
     };
 
-    explicit BufferQueue(Kernel::KernelCore& kernel, u32 id_, u64 layer_id_);
+    explicit BufferQueue(Kernel::KernelCore& kernel, u32 id_, u64 layer_id_,
+                         KernelHelpers::ServiceContext& service_context_);
     ~BufferQueue();
 
     enum class BufferTransformFlags : u32 {
@@ -130,12 +135,14 @@ private:
     std::list<u32> free_buffers;
     std::array<Buffer, buffer_slots> buffers;
     std::list<u32> queue_sequence;
-    Kernel::KEvent buffer_wait_event;
+    Kernel::KEvent* buffer_wait_event{};
 
     std::mutex free_buffers_mutex;
     std::condition_variable free_buffers_condition;
 
     std::mutex queue_sequence_mutex;
+
+    KernelHelpers::ServiceContext& service_context;
 };
 
 } // namespace Service::NVFlinger
diff --git a/src/core/hle/service/nvflinger/nvflinger.cpp b/src/core/hle/service/nvflinger/nvflinger.cpp
index 941748970..00bff8caf 100644
--- a/src/core/hle/service/nvflinger/nvflinger.cpp
+++ b/src/core/hle/service/nvflinger/nvflinger.cpp
@@ -61,12 +61,13 @@ void NVFlinger::SplitVSync() {
     }
 }
 
-NVFlinger::NVFlinger(Core::System& system_) : system(system_) {
-    displays.emplace_back(0, "Default", system);
-    displays.emplace_back(1, "External", system);
-    displays.emplace_back(2, "Edid", system);
-    displays.emplace_back(3, "Internal", system);
-    displays.emplace_back(4, "Null", system);
+NVFlinger::NVFlinger(Core::System& system_)
+    : system(system_), service_context(system_, "nvflinger") {
+    displays.emplace_back(0, "Default", service_context, system);
+    displays.emplace_back(1, "External", service_context, system);
+    displays.emplace_back(2, "Edid", service_context, system);
+    displays.emplace_back(3, "Internal", service_context, system);
+    displays.emplace_back(4, "Null", service_context, system);
     guard = std::make_shared<std::mutex>();
 
     // Schedule the screen composition events
@@ -146,7 +147,7 @@ std::optional<u64> NVFlinger::CreateLayer(u64 display_id) {
 void NVFlinger::CreateLayerAtId(VI::Display& display, u64 layer_id) {
     const u32 buffer_queue_id = next_buffer_queue_id++;
     buffer_queues.emplace_back(
-        std::make_unique<BufferQueue>(system.Kernel(), buffer_queue_id, layer_id));
+        std::make_unique<BufferQueue>(system.Kernel(), buffer_queue_id, layer_id, service_context));
     display.CreateLayer(layer_id, *buffer_queues.back());
 }
 
diff --git a/src/core/hle/service/nvflinger/nvflinger.h b/src/core/hle/service/nvflinger/nvflinger.h
index d80fd07ef..6d84cafb4 100644
--- a/src/core/hle/service/nvflinger/nvflinger.h
+++ b/src/core/hle/service/nvflinger/nvflinger.h
@@ -15,6 +15,7 @@
 #include <vector>
 
 #include "common/common_types.h"
+#include "core/hle/service/kernel_helpers.h"
 
 namespace Common {
 class Event;
@@ -135,6 +136,8 @@ private:
     std::unique_ptr<std::thread> vsync_thread;
     std::unique_ptr<Common::Event> wait_event;
     std::atomic<bool> is_running{};
+
+    KernelHelpers::ServiceContext service_context;
 };
 
 } // namespace Service::NVFlinger
diff --git a/src/core/hle/service/vi/display/vi_display.cpp b/src/core/hle/service/vi/display/vi_display.cpp
index 0dd342dbf..b7705c02a 100644
--- a/src/core/hle/service/vi/display/vi_display.cpp
+++ b/src/core/hle/service/vi/display/vi_display.cpp
@@ -12,18 +12,21 @@
 #include "core/hle/kernel/k_event.h"
 #include "core/hle/kernel/k_readable_event.h"
 #include "core/hle/kernel/k_writable_event.h"
+#include "core/hle/service/kernel_helpers.h"
 #include "core/hle/service/vi/display/vi_display.h"
 #include "core/hle/service/vi/layer/vi_layer.h"
 
 namespace Service::VI {
 
-Display::Display(u64 id, std::string name_, Core::System& system)
-    : display_id{id}, name{std::move(name_)}, vsync_event{system.Kernel()} {
-    Kernel::KAutoObject::Create(std::addressof(vsync_event));
-    vsync_event.Initialize(fmt::format("Display VSync Event {}", id));
+Display::Display(u64 id, std::string name_, KernelHelpers::ServiceContext& service_context_,
+                 Core::System& system_)
+    : display_id{id}, name{std::move(name_)}, service_context{service_context_} {
+    vsync_event = service_context.CreateEvent(fmt::format("Display VSync Event {}", id));
 }
 
-Display::~Display() = default;
+Display::~Display() {
+    service_context.CloseEvent(vsync_event);
+}
 
 Layer& Display::GetLayer(std::size_t index) {
     return *layers.at(index);
@@ -34,11 +37,11 @@ const Layer& Display::GetLayer(std::size_t index) const {
 }
 
 Kernel::KReadableEvent& Display::GetVSyncEvent() {
-    return vsync_event.GetReadableEvent();
+    return vsync_event->GetReadableEvent();
 }
 
 void Display::SignalVSyncEvent() {
-    vsync_event.GetWritableEvent().Signal();
+    vsync_event->GetWritableEvent().Signal();
 }
 
 void Display::CreateLayer(u64 layer_id, NVFlinger::BufferQueue& buffer_queue) {
diff --git a/src/core/hle/service/vi/display/vi_display.h b/src/core/hle/service/vi/display/vi_display.h
index 166f2a4cc..0979fc421 100644
--- a/src/core/hle/service/vi/display/vi_display.h
+++ b/src/core/hle/service/vi/display/vi_display.h
@@ -18,6 +18,9 @@ class KEvent;
 namespace Service::NVFlinger {
 class BufferQueue;
 }
+namespace Service::KernelHelpers {
+class ServiceContext;
+} // namespace Service::KernelHelpers
 
 namespace Service::VI {
 
@@ -31,10 +34,13 @@ class Display {
 public:
     /// Constructs a display with a given unique ID and name.
     ///
-    /// @param id   The unique ID for this display.
+    /// @param id The unique ID for this display.
+    /// @param service_context_ The ServiceContext for the owning service.
     /// @param name_ The name for this display.
+    /// @param system_ The global system instance.
     ///
-    Display(u64 id, std::string name_, Core::System& system);
+    Display(u64 id, std::string name_, KernelHelpers::ServiceContext& service_context_,
+            Core::System& system_);
     ~Display();
 
     /// Gets the unique ID assigned to this display.
@@ -98,9 +104,10 @@ public:
 private:
     u64 display_id;
     std::string name;
+    KernelHelpers::ServiceContext& service_context;
 
     std::vector<std::shared_ptr<Layer>> layers;
-    Kernel::KEvent vsync_event;
+    Kernel::KEvent* vsync_event{};
 };
 
 } // namespace Service::VI
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
index 67df46499..68f360b3c 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
@@ -380,8 +380,9 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) {
             return ctx.OpBitcast(ctx.F32[1], ctx.OpISub(ctx.U32[1], index, base));
         }
     case IR::Attribute::FrontFace:
-        return ctx.OpSelect(ctx.U32[1], ctx.OpLoad(ctx.U1, ctx.front_face),
-                            ctx.Const(std::numeric_limits<u32>::max()), ctx.u32_zero_value);
+        return ctx.OpSelect(ctx.F32[1], ctx.OpLoad(ctx.U1, ctx.front_face),
+                            ctx.OpBitcast(ctx.F32[1], ctx.Const(std::numeric_limits<u32>::max())),
+                            ctx.f32_zero_value);
     case IR::Attribute::PointSpriteS:
         return ctx.OpLoad(ctx.F32[1],
                           ctx.OpAccessChain(ctx.input_f32, ctx.point_coord, ctx.u32_zero_value));
diff --git a/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp b/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp
index 8b3e0a15c..69eeaa3e6 100644
--- a/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp
+++ b/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp
@@ -20,6 +20,7 @@
 #include "shader_recompiler/frontend/maxwell/decode.h"
 #include "shader_recompiler/frontend/maxwell/structured_control_flow.h"
 #include "shader_recompiler/frontend/maxwell/translate/translate.h"
+#include "shader_recompiler/host_translate_info.h"
 #include "shader_recompiler/object_pool.h"
 
 namespace Shader::Maxwell {
@@ -652,7 +653,7 @@ class TranslatePass {
 public:
     TranslatePass(ObjectPool<IR::Inst>& inst_pool_, ObjectPool<IR::Block>& block_pool_,
                   ObjectPool<Statement>& stmt_pool_, Environment& env_, Statement& root_stmt,
-                  IR::AbstractSyntaxList& syntax_list_)
+                  IR::AbstractSyntaxList& syntax_list_, const HostTranslateInfo& host_info)
         : stmt_pool{stmt_pool_}, inst_pool{inst_pool_}, block_pool{block_pool_}, env{env_},
           syntax_list{syntax_list_} {
         Visit(root_stmt, nullptr, nullptr);
@@ -660,6 +661,9 @@ public:
         IR::Block& first_block{*syntax_list.front().data.block};
         IR::IREmitter ir(first_block, first_block.begin());
         ir.Prologue();
+        if (uses_demote_to_helper && host_info.needs_demote_reorder) {
+            DemoteCombinationPass();
+        }
     }
 
 private:
@@ -809,7 +813,14 @@ private:
             }
             case StatementType::Return: {
                 ensure_block();
-                IR::IREmitter{*current_block}.Epilogue();
+                IR::Block* return_block{block_pool.Create(inst_pool)};
+                IR::IREmitter{*return_block}.Epilogue();
+                current_block->AddBranch(return_block);
+
+                auto& merge{syntax_list.emplace_back()};
+                merge.type = IR::AbstractSyntaxNode::Type::Block;
+                merge.data.block = return_block;
+
                 current_block = nullptr;
                 syntax_list.emplace_back().type = IR::AbstractSyntaxNode::Type::Return;
                 break;
@@ -824,6 +835,7 @@ private:
                 auto& merge{syntax_list.emplace_back()};
                 merge.type = IR::AbstractSyntaxNode::Type::Block;
                 merge.data.block = demote_block;
+                uses_demote_to_helper = true;
                 break;
             }
             case StatementType::Unreachable: {
@@ -855,11 +867,117 @@ private:
         return block_pool.Create(inst_pool);
     }
 
+    void DemoteCombinationPass() {
+        using Type = IR::AbstractSyntaxNode::Type;
+        std::vector<IR::Block*> demote_blocks;
+        std::vector<IR::U1> demote_conds;
+        u32 num_epilogues{};
+        u32 branch_depth{};
+        for (const IR::AbstractSyntaxNode& node : syntax_list) {
+            if (node.type == Type::If) {
+                ++branch_depth;
+            }
+            if (node.type == Type::EndIf) {
+                --branch_depth;
+            }
+            if (node.type != Type::Block) {
+                continue;
+            }
+            if (branch_depth > 1) {
+                // Skip reordering nested demote branches.
+                continue;
+            }
+            for (const IR::Inst& inst : node.data.block->Instructions()) {
+                const IR::Opcode op{inst.GetOpcode()};
+                if (op == IR::Opcode::DemoteToHelperInvocation) {
+                    demote_blocks.push_back(node.data.block);
+                    break;
+                }
+                if (op == IR::Opcode::Epilogue) {
+                    ++num_epilogues;
+                }
+            }
+        }
+        if (demote_blocks.size() == 0) {
+            return;
+        }
+        if (num_epilogues > 1) {
+            LOG_DEBUG(Shader, "Combining demotes with more than one return is not implemented.");
+            return;
+        }
+        s64 last_iterator_offset{};
+        auto& asl{syntax_list};
+        for (const IR::Block* demote_block : demote_blocks) {
+            const auto start_it{asl.begin() + last_iterator_offset};
+            auto asl_it{std::find_if(start_it, asl.end(), [&](const IR::AbstractSyntaxNode& asn) {
+                return asn.type == Type::If && asn.data.if_node.body == demote_block;
+            })};
+            if (asl_it == asl.end()) {
+                // Demote without a conditional branch.
+                // No need to proceed since all fragment instances will be demoted regardless.
+                return;
+            }
+            const IR::Block* const end_if = asl_it->data.if_node.merge;
+            demote_conds.push_back(asl_it->data.if_node.cond);
+            last_iterator_offset = std::distance(asl.begin(), asl_it);
+
+            asl_it = asl.erase(asl_it);
+            asl_it = std::find_if(asl_it, asl.end(), [&](const IR::AbstractSyntaxNode& asn) {
+                return asn.type == Type::Block && asn.data.block == demote_block;
+            });
+
+            asl_it = asl.erase(asl_it);
+            asl_it = std::find_if(asl_it, asl.end(), [&](const IR::AbstractSyntaxNode& asn) {
+                return asn.type == Type::EndIf && asn.data.end_if.merge == end_if;
+            });
+            asl_it = asl.erase(asl_it);
+        }
+        const auto epilogue_func{[](const IR::AbstractSyntaxNode& asn) {
+            if (asn.type != Type::Block) {
+                return false;
+            }
+            for (const auto& inst : asn.data.block->Instructions()) {
+                if (inst.GetOpcode() == IR::Opcode::Epilogue) {
+                    return true;
+                }
+            }
+            return false;
+        }};
+        const auto reverse_it{std::find_if(asl.rbegin(), asl.rend(), epilogue_func)};
+        const auto return_block_it{(reverse_it + 1).base()};
+
+        IR::IREmitter ir{*(return_block_it - 1)->data.block};
+        IR::U1 cond(IR::Value(false));
+        for (const auto& demote_cond : demote_conds) {
+            cond = ir.LogicalOr(cond, demote_cond);
+        }
+        cond.Inst()->DestructiveAddUsage(1);
+
+        IR::AbstractSyntaxNode demote_if_node{};
+        demote_if_node.type = Type::If;
+        demote_if_node.data.if_node.cond = cond;
+        demote_if_node.data.if_node.body = demote_blocks[0];
+        demote_if_node.data.if_node.merge = return_block_it->data.block;
+
+        IR::AbstractSyntaxNode demote_node{};
+        demote_node.type = Type::Block;
+        demote_node.data.block = demote_blocks[0];
+
+        IR::AbstractSyntaxNode demote_endif_node{};
+        demote_endif_node.type = Type::EndIf;
+        demote_endif_node.data.end_if.merge = return_block_it->data.block;
+
+        asl.insert(return_block_it, demote_endif_node);
+        asl.insert(return_block_it, demote_node);
+        asl.insert(return_block_it, demote_if_node);
+    }
+
     ObjectPool<Statement>& stmt_pool;
     ObjectPool<IR::Inst>& inst_pool;
     ObjectPool<IR::Block>& block_pool;
     Environment& env;
     IR::AbstractSyntaxList& syntax_list;
+    bool uses_demote_to_helper{};
 
 // TODO: C++20 Remove this when all compilers support constexpr std::vector
 #if __cpp_lib_constexpr_vector >= 201907
@@ -871,12 +989,13 @@ private:
 } // Anonymous namespace
 
 IR::AbstractSyntaxList BuildASL(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Block>& block_pool,
-                                Environment& env, Flow::CFG& cfg) {
+                                Environment& env, Flow::CFG& cfg,
+                                const HostTranslateInfo& host_info) {
     ObjectPool<Statement> stmt_pool{64};
     GotoPass goto_pass{cfg, stmt_pool};
     Statement& root{goto_pass.RootStatement()};
     IR::AbstractSyntaxList syntax_list;
-    TranslatePass{inst_pool, block_pool, stmt_pool, env, root, syntax_list};
+    TranslatePass{inst_pool, block_pool, stmt_pool, env, root, syntax_list, host_info};
     return syntax_list;
 }
 
diff --git a/src/shader_recompiler/frontend/maxwell/structured_control_flow.h b/src/shader_recompiler/frontend/maxwell/structured_control_flow.h
index 88b083649..e38158da3 100644
--- a/src/shader_recompiler/frontend/maxwell/structured_control_flow.h
+++ b/src/shader_recompiler/frontend/maxwell/structured_control_flow.h
@@ -11,10 +11,13 @@
 #include "shader_recompiler/frontend/maxwell/control_flow.h"
 #include "shader_recompiler/object_pool.h"
 
-namespace Shader::Maxwell {
+namespace Shader {
+struct HostTranslateInfo;
+namespace Maxwell {
 
 [[nodiscard]] IR::AbstractSyntaxList BuildASL(ObjectPool<IR::Inst>& inst_pool,
                                               ObjectPool<IR::Block>& block_pool, Environment& env,
-                                              Flow::CFG& cfg);
+                                              Flow::CFG& cfg, const HostTranslateInfo& host_info);
 
-} // namespace Shader::Maxwell
+} // namespace Maxwell
+} // namespace Shader
diff --git a/src/shader_recompiler/frontend/maxwell/translate_program.cpp b/src/shader_recompiler/frontend/maxwell/translate_program.cpp
index c067d459c..012d55357 100644
--- a/src/shader_recompiler/frontend/maxwell/translate_program.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate_program.cpp
@@ -130,7 +130,7 @@ void AddNVNStorageBuffers(IR::Program& program) {
 IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Block>& block_pool,
                              Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info) {
     IR::Program program;
-    program.syntax_list = BuildASL(inst_pool, block_pool, env, cfg);
+    program.syntax_list = BuildASL(inst_pool, block_pool, env, cfg, host_info);
     program.blocks = GenerateBlocks(program.syntax_list);
     program.post_order_blocks = PostOrder(program.syntax_list.front());
     program.stage = env.ShaderStage();
diff --git a/src/shader_recompiler/host_translate_info.h b/src/shader_recompiler/host_translate_info.h
index 94a584219..96468b2e7 100644
--- a/src/shader_recompiler/host_translate_info.h
+++ b/src/shader_recompiler/host_translate_info.h
@@ -11,8 +11,9 @@ namespace Shader {
 
 /// Misc information about the host
 struct HostTranslateInfo {
-    bool support_float16{}; ///< True when the device supports 16-bit floats
-    bool support_int64{};   ///< True when the device supports 64-bit integers
+    bool support_float16{};      ///< True when the device supports 16-bit floats
+    bool support_int64{};        ///< True when the device supports 64-bit integers
+    bool needs_demote_reorder{}; ///< True when the device needs DemoteToHelperInvocation reordered
 };
 
 } // namespace Shader
diff --git a/src/video_core/buffer_cache/buffer_base.h b/src/video_core/buffer_cache/buffer_base.h
index c3318095c..be2113f5a 100644
--- a/src/video_core/buffer_cache/buffer_base.h
+++ b/src/video_core/buffer_cache/buffer_base.h
@@ -261,16 +261,6 @@ public:
         stream_score += score;
     }
 
-    /// Sets the new frame tick
-    void SetFrameTick(u64 new_frame_tick) noexcept {
-        frame_tick = new_frame_tick;
-    }
-
-    /// Returns the new frame tick
-    [[nodiscard]] u64 FrameTick() const noexcept {
-        return frame_tick;
-    }
-
     /// Returns the likeliness of this being a stream buffer
     [[nodiscard]] int StreamScore() const noexcept {
         return stream_score;
@@ -307,6 +297,14 @@ public:
         return words.size_bytes;
     }
 
+    size_t getLRUID() const noexcept {
+        return lru_id;
+    }
+
+    void setLRUID(size_t lru_id_) {
+        lru_id = lru_id_;
+    }
+
 private:
     template <Type type>
     u64* Array() noexcept {
@@ -603,9 +601,9 @@ private:
     RasterizerInterface* rasterizer = nullptr;
     VAddr cpu_addr = 0;
     Words words;
-    u64 frame_tick = 0;
     BufferFlagBits flags{};
     int stream_score = 0;
+    size_t lru_id = SIZE_MAX;
 };
 
 } // namespace VideoCommon
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 3b43554f9..7bfd57369 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -20,6 +20,7 @@
 #include "common/common_types.h"
 #include "common/div_ceil.h"
 #include "common/literals.h"
+#include "common/lru_cache.h"
 #include "common/microprofile.h"
 #include "common/scope_exit.h"
 #include "common/settings.h"
@@ -330,7 +331,7 @@ private:
     template <bool insert>
     void ChangeRegister(BufferId buffer_id);
 
-    void TouchBuffer(Buffer& buffer) const noexcept;
+    void TouchBuffer(Buffer& buffer, BufferId buffer_id) noexcept;
 
     bool SynchronizeBuffer(Buffer& buffer, VAddr cpu_addr, u32 size);
 
@@ -428,7 +429,11 @@ private:
     size_t immediate_buffer_capacity = 0;
     std::unique_ptr<u8[]> immediate_buffer_alloc;
 
-    typename SlotVector<Buffer>::Iterator deletion_iterator;
+    struct LRUItemParams {
+        using ObjectType = BufferId;
+        using TickType = u64;
+    };
+    Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
     u64 frame_tick = 0;
     u64 total_used_memory = 0;
 
@@ -445,7 +450,6 @@ BufferCache<P>::BufferCache(VideoCore::RasterizerInterface& rasterizer_,
       kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_} {
     // Ensure the first slot is used for the null buffer
     void(slot_buffers.insert(runtime, NullBufferParams{}));
-    deletion_iterator = slot_buffers.end();
     common_ranges.clear();
 }
 
@@ -454,20 +458,17 @@ void BufferCache<P>::RunGarbageCollector() {
     const bool aggressive_gc = total_used_memory >= CRITICAL_MEMORY;
     const u64 ticks_to_destroy = aggressive_gc ? 60 : 120;
     int num_iterations = aggressive_gc ? 64 : 32;
-    for (; num_iterations > 0; --num_iterations) {
-        if (deletion_iterator == slot_buffers.end()) {
-            deletion_iterator = slot_buffers.begin();
-        }
-        ++deletion_iterator;
-        if (deletion_iterator == slot_buffers.end()) {
-            break;
-        }
-        const auto [buffer_id, buffer] = *deletion_iterator;
-        if (buffer->FrameTick() + ticks_to_destroy < frame_tick) {
-            DownloadBufferMemory(*buffer);
-            DeleteBuffer(buffer_id);
+    const auto clean_up = [this, &num_iterations](BufferId buffer_id) {
+        if (num_iterations == 0) {
+            return true;
         }
-    }
+        --num_iterations;
+        auto& buffer = slot_buffers[buffer_id];
+        DownloadBufferMemory(buffer);
+        DeleteBuffer(buffer_id);
+        return false;
+    };
+    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up);
 }
 
 template <class P>
@@ -485,7 +486,7 @@ void BufferCache<P>::TickFrame() {
     const bool skip_preferred = hits * 256 < shots * 251;
     uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0;
 
-    if (Settings::values.use_caches_gc.GetValue() && total_used_memory >= EXPECTED_MEMORY) {
+    if (total_used_memory >= EXPECTED_MEMORY) {
         RunGarbageCollector();
     }
     ++frame_tick;
@@ -954,7 +955,7 @@ bool BufferCache<P>::IsRegionCpuModified(VAddr addr, size_t size) {
 template <class P>
 void BufferCache<P>::BindHostIndexBuffer() {
     Buffer& buffer = slot_buffers[index_buffer.buffer_id];
-    TouchBuffer(buffer);
+    TouchBuffer(buffer, index_buffer.buffer_id);
     const u32 offset = buffer.Offset(index_buffer.cpu_addr);
     const u32 size = index_buffer.size;
     SynchronizeBuffer(buffer, index_buffer.cpu_addr, size);
@@ -975,7 +976,7 @@ void BufferCache<P>::BindHostVertexBuffers() {
     for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
         const Binding& binding = vertex_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
-        TouchBuffer(buffer);
+        TouchBuffer(buffer, binding.buffer_id);
         SynchronizeBuffer(buffer, binding.cpu_addr, binding.size);
         if (!flags[Dirty::VertexBuffer0 + index]) {
             continue;
@@ -1011,7 +1012,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
     const VAddr cpu_addr = binding.cpu_addr;
     const u32 size = std::min(binding.size, (*uniform_buffer_sizes)[stage][index]);
     Buffer& buffer = slot_buffers[binding.buffer_id];
-    TouchBuffer(buffer);
+    TouchBuffer(buffer, binding.buffer_id);
     const bool use_fast_buffer = binding.buffer_id != NULL_BUFFER_ID &&
                                  size <= uniform_buffer_skip_cache_size &&
                                  !buffer.IsRegionGpuModified(cpu_addr, size);
@@ -1083,7 +1084,7 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
     ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) {
         const Binding& binding = storage_buffers[stage][index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
-        TouchBuffer(buffer);
+        TouchBuffer(buffer, binding.buffer_id);
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
@@ -1128,7 +1129,7 @@ void BufferCache<P>::BindHostTransformFeedbackBuffers() {
     for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) {
         const Binding& binding = transform_feedback_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
-        TouchBuffer(buffer);
+        TouchBuffer(buffer, binding.buffer_id);
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
@@ -1148,7 +1149,7 @@ void BufferCache<P>::BindHostComputeUniformBuffers() {
     ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) {
         const Binding& binding = compute_uniform_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
-        TouchBuffer(buffer);
+        TouchBuffer(buffer, binding.buffer_id);
         const u32 size = std::min(binding.size, (*compute_uniform_buffer_sizes)[index]);
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
@@ -1168,7 +1169,7 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
     ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) {
         const Binding& binding = compute_storage_buffers[index];
         Buffer& buffer = slot_buffers[binding.buffer_id];
-        TouchBuffer(buffer);
+        TouchBuffer(buffer, binding.buffer_id);
         const u32 size = binding.size;
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
@@ -1513,11 +1514,11 @@ BufferId BufferCache<P>::CreateBuffer(VAddr cpu_addr, u32 wanted_size) {
     const OverlapResult overlap = ResolveOverlaps(cpu_addr, wanted_size);
     const u32 size = static_cast<u32>(overlap.end - overlap.begin);
     const BufferId new_buffer_id = slot_buffers.insert(runtime, rasterizer, overlap.begin, size);
-    TouchBuffer(slot_buffers[new_buffer_id]);
     for (const BufferId overlap_id : overlap.ids) {
         JoinOverlap(new_buffer_id, overlap_id, !overlap.has_stream_leap);
     }
     Register(new_buffer_id);
+    TouchBuffer(slot_buffers[new_buffer_id], new_buffer_id);
     return new_buffer_id;
 }
 
@@ -1534,12 +1535,14 @@ void BufferCache<P>::Unregister(BufferId buffer_id) {
 template <class P>
 template <bool insert>
 void BufferCache<P>::ChangeRegister(BufferId buffer_id) {
-    const Buffer& buffer = slot_buffers[buffer_id];
+    Buffer& buffer = slot_buffers[buffer_id];
     const auto size = buffer.SizeBytes();
     if (insert) {
         total_used_memory += Common::AlignUp(size, 1024);
+        buffer.setLRUID(lru_cache.Insert(buffer_id, frame_tick));
     } else {
         total_used_memory -= Common::AlignUp(size, 1024);
+        lru_cache.Free(buffer.getLRUID());
     }
     const VAddr cpu_addr_begin = buffer.CpuAddr();
     const VAddr cpu_addr_end = cpu_addr_begin + size;
@@ -1555,8 +1558,10 @@ void BufferCache<P>::ChangeRegister(BufferId buffer_id) {
 }
 
 template <class P>
-void BufferCache<P>::TouchBuffer(Buffer& buffer) const noexcept {
-    buffer.SetFrameTick(frame_tick);
+void BufferCache<P>::TouchBuffer(Buffer& buffer, BufferId buffer_id) noexcept {
+    if (buffer_id != NULL_BUFFER_ID) {
+        lru_cache.Touch(buffer.getLRUID(), frame_tick);
+    }
 }
 
 template <class P>
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 1aa43523a..7f4ca6282 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -475,10 +475,10 @@ public:
 
                 // These values are used by Nouveau and some games.
                 AddGL = 0x8006,
-                SubtractGL = 0x8007,
-                ReverseSubtractGL = 0x8008,
-                MinGL = 0x800a,
-                MaxGL = 0x800b
+                MinGL = 0x8007,
+                MaxGL = 0x8008,
+                SubtractGL = 0x800a,
+                ReverseSubtractGL = 0x800b
             };
 
             enum class Factor : u32 {
diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h
index ee992aed4..de9e41659 100644
--- a/src/video_core/renderer_opengl/gl_device.h
+++ b/src/video_core/renderer_opengl/gl_device.h
@@ -156,6 +156,10 @@ public:
         return shader_backend;
     }
 
+    bool IsAmd() const {
+        return vendor_name == "ATI Technologies Inc.";
+    }
+
 private:
     static bool TestVariableAoffi();
     static bool TestPreciseBug();
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 1f4dda17e..b0e14182e 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -219,6 +219,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
       host_info{
           .support_float16 = false,
           .support_int64 = device.HasShaderInt64(),
+          .needs_demote_reorder = device.IsAmd(),
       } {
     if (use_asynchronous_shaders) {
         workers = CreateWorkers();
diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
index 5c43b8acf..cb0580182 100644
--- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp
+++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
@@ -159,11 +159,13 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
 
         const VAddr framebuffer_addr = framebuffer.address + framebuffer.offset;
         const u8* const host_ptr = cpu_memory.GetPointer(framebuffer_addr);
-        const size_t size_bytes = GetSizeInBytes(framebuffer);
 
         // TODO(Rodrigo): Read this from HLE
         constexpr u32 block_height_log2 = 4;
         const u32 bytes_per_pixel = GetBytesPerPixel(framebuffer);
+        const u64 size_bytes{Tegra::Texture::CalculateSize(true, bytes_per_pixel,
+                                                           framebuffer.stride, framebuffer.height,
+                                                           1, block_height_log2, 0)};
         Tegra::Texture::UnswizzleTexture(
             mapped_span.subspan(image_offset, size_bytes), std::span(host_ptr, size_bytes),
             bytes_per_pixel, framebuffer.width, framebuffer.height, 1, block_height_log2, 0);
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index f316c4f92..31bfbcb06 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -325,6 +325,8 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw
     host_info = Shader::HostTranslateInfo{
         .support_float16 = device.IsFloat16Supported(),
         .support_int64 = device.IsShaderInt64Supported(),
+        .needs_demote_reorder = driver_id == VK_DRIVER_ID_AMD_PROPRIETARY_KHR ||
+                                driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE_KHR,
     };
 }
 
diff --git a/src/video_core/texture_cache/image_base.h b/src/video_core/texture_cache/image_base.h
index ff1feda9b..0c17a791b 100644
--- a/src/video_core/texture_cache/image_base.h
+++ b/src/video_core/texture_cache/image_base.h
@@ -80,7 +80,7 @@ struct ImageBase {
     VAddr cpu_addr_end = 0;
 
     u64 modification_tick = 0;
-    u64 frame_tick = 0;
+    size_t lru_index = SIZE_MAX;
 
     std::array<u32, MAX_MIP_LEVELS> mip_level_offsets{};
 
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index a087498ff..24b809242 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -5,7 +5,6 @@
 #pragma once
 
 #include "common/alignment.h"
-#include "common/settings.h"
 #include "video_core/dirty_flags.h"
 #include "video_core/texture_cache/samples_helper.h"
 #include "video_core/texture_cache/texture_cache_base.h"
@@ -43,8 +42,6 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
     void(slot_image_views.insert(runtime, NullImageParams{}));
     void(slot_samplers.insert(runtime, sampler_descriptor));
 
-    deletion_iterator = slot_images.begin();
-
     if constexpr (HAS_DEVICE_MEMORY_INFO) {
         const auto device_memory = runtime.GetDeviceLocalMemory();
         const u64 possible_expected_memory = (device_memory * 3) / 10;
@@ -64,70 +61,38 @@ template <class P>
 void TextureCache<P>::RunGarbageCollector() {
     const bool high_priority_mode = total_used_memory >= expected_memory;
     const bool aggressive_mode = total_used_memory >= critical_memory;
-    const u64 ticks_to_destroy = high_priority_mode ? 60 : 100;
-    int num_iterations = aggressive_mode ? 256 : (high_priority_mode ? 128 : 64);
-    for (; num_iterations > 0; --num_iterations) {
-        if (deletion_iterator == slot_images.end()) {
-            deletion_iterator = slot_images.begin();
-            if (deletion_iterator == slot_images.end()) {
-                break;
-            }
+    const u64 ticks_to_destroy = aggressive_mode ? 10ULL : high_priority_mode ? 25ULL : 100ULL;
+    size_t num_iterations = aggressive_mode ? 10000 : (high_priority_mode ? 100 : 5);
+    const auto clean_up = [this, &num_iterations, high_priority_mode](ImageId image_id) {
+        if (num_iterations == 0) {
+            return true;
         }
-        auto [image_id, image_tmp] = *deletion_iterator;
-        Image* image = image_tmp; // fix clang error.
-        const bool is_alias = True(image->flags & ImageFlagBits::Alias);
-        const bool is_bad_overlap = True(image->flags & ImageFlagBits::BadOverlap);
-        const bool must_download = image->IsSafeDownload();
-        bool should_care = is_bad_overlap || is_alias || (high_priority_mode && !must_download);
-        const u64 ticks_needed =
-            is_bad_overlap
-                ? ticks_to_destroy >> 4
-                : ((should_care && aggressive_mode) ? ticks_to_destroy >> 1 : ticks_to_destroy);
-        should_care |= aggressive_mode;
-        if (should_care && image->frame_tick + ticks_needed < frame_tick) {
-            if (is_bad_overlap) {
-                const bool overlap_check = std::ranges::all_of(
-                    image->overlapping_images, [&, image](const ImageId& overlap_id) {
-                        auto& overlap = slot_images[overlap_id];
-                        return overlap.frame_tick >= image->frame_tick;
-                    });
-                if (!overlap_check) {
-                    ++deletion_iterator;
-                    continue;
-                }
-            }
-            if (!is_bad_overlap && must_download) {
-                const bool alias_check = std::ranges::none_of(
-                    image->aliased_images, [&, image](const AliasedImage& alias) {
-                        auto& alias_image = slot_images[alias.id];
-                        return (alias_image.frame_tick < image->frame_tick) ||
-                               (alias_image.modification_tick < image->modification_tick);
-                    });
-
-                if (alias_check) {
-                    auto map = runtime.DownloadStagingBuffer(image->unswizzled_size_bytes);
-                    const auto copies = FullDownloadCopies(image->info);
-                    image->DownloadMemory(map, copies);
-                    runtime.Finish();
-                    SwizzleImage(gpu_memory, image->gpu_addr, image->info, copies, map.mapped_span);
-                }
-            }
-            if (True(image->flags & ImageFlagBits::Tracked)) {
-                UntrackImage(*image, image_id);
-            }
-            UnregisterImage(image_id);
-            DeleteImage(image_id);
-            if (is_bad_overlap) {
-                ++num_iterations;
-            }
+        --num_iterations;
+        auto& image = slot_images[image_id];
+        const bool must_download = image.IsSafeDownload();
+        if (!high_priority_mode && must_download) {
+            return false;
         }
-        ++deletion_iterator;
-    }
+        if (must_download) {
+            auto map = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes);
+            const auto copies = FullDownloadCopies(image.info);
+            image.DownloadMemory(map, copies);
+            runtime.Finish();
+            SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span);
+        }
+        if (True(image.flags & ImageFlagBits::Tracked)) {
+            UntrackImage(image, image_id);
+        }
+        UnregisterImage(image_id);
+        DeleteImage(image_id);
+        return false;
+    };
+    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up);
 }
 
 template <class P>
 void TextureCache<P>::TickFrame() {
-    if (Settings::values.use_caches_gc.GetValue() && total_used_memory > minimum_memory) {
+    if (total_used_memory > minimum_memory) {
         RunGarbageCollector();
     }
     sentenced_images.Tick();
@@ -1078,6 +1043,8 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
         tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
     }
     total_used_memory += Common::AlignUp(tentative_size, 1024);
+    image.lru_index = lru_cache.Insert(image_id, frame_tick);
+
     ForEachGPUPage(image.gpu_addr, image.guest_size_bytes,
                    [this, image_id](u64 page) { gpu_page_table[page].push_back(image_id); });
     if (False(image.flags & ImageFlagBits::Sparse)) {
@@ -1115,6 +1082,7 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
         tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
     }
     total_used_memory -= Common::AlignUp(tentative_size, 1024);
+    lru_cache.Free(image.lru_index);
     const auto& clear_page_table =
         [this, image_id](
             u64 page,
@@ -1384,7 +1352,7 @@ void TextureCache<P>::PrepareImage(ImageId image_id, bool is_modification, bool
     if (is_modification) {
         MarkModification(image);
     }
-    image.frame_tick = frame_tick;
+    lru_cache.Touch(image.lru_index, frame_tick);
 }
 
 template <class P>
diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h
index e4ae351cb..d7528ed24 100644
--- a/src/video_core/texture_cache/texture_cache_base.h
+++ b/src/video_core/texture_cache/texture_cache_base.h
@@ -14,6 +14,7 @@
 
 #include "common/common_types.h"
 #include "common/literals.h"
+#include "common/lru_cache.h"
 #include "video_core/compatible_formats.h"
 #include "video_core/delayed_destruction_ring.h"
 #include "video_core/engines/fermi_2d.h"
@@ -370,6 +371,12 @@ private:
     std::vector<ImageId> uncommitted_downloads;
     std::queue<std::vector<ImageId>> committed_downloads;
 
+    struct LRUItemParams {
+        using ObjectType = ImageId;
+        using TickType = u64;
+    };
+    Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
+
     static constexpr size_t TICKS_TO_DESTROY = 6;
     DelayedDestructionRing<Image, TICKS_TO_DESTROY> sentenced_images;
     DelayedDestructionRing<ImageView, TICKS_TO_DESTROY> sentenced_image_view;
@@ -379,7 +386,6 @@ private:
 
     u64 modification_tick = 0;
     u64 frame_tick = 0;
-    typename SlotVector<Image>::Iterator deletion_iterator;
 };
 
 } // namespace VideoCommon
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index c010b9353..24e943e4c 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -63,14 +63,6 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32
                 const u32 unswizzled_offset =
                     slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL;
 
-                if (const auto offset = (TO_LINEAR ? unswizzled_offset : swizzled_offset);
-                    offset >= input.size()) {
-                    // TODO(Rodrigo): This is an out of bounds access that should never happen. To
-                    // avoid crashing the emulator, break.
-                    ASSERT_MSG(false, "offset {} exceeds input size {}!", offset, input.size());
-                    break;
-                }
-
                 u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset];
                 const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset];
 
diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp
index 377795326..85d292bcc 100644
--- a/src/yuzu/configuration/config.cpp
+++ b/src/yuzu/configuration/config.cpp
@@ -818,7 +818,6 @@ void Config::ReadRendererValues() {
     ReadGlobalSetting(Settings::values.shader_backend);
     ReadGlobalSetting(Settings::values.use_asynchronous_shaders);
     ReadGlobalSetting(Settings::values.use_fast_gpu_time);
-    ReadGlobalSetting(Settings::values.use_caches_gc);
     ReadGlobalSetting(Settings::values.bg_red);
     ReadGlobalSetting(Settings::values.bg_green);
     ReadGlobalSetting(Settings::values.bg_blue);
@@ -1359,7 +1358,6 @@ void Config::SaveRendererValues() {
                  Settings::values.shader_backend.UsingGlobal());
     WriteGlobalSetting(Settings::values.use_asynchronous_shaders);
     WriteGlobalSetting(Settings::values.use_fast_gpu_time);
-    WriteGlobalSetting(Settings::values.use_caches_gc);
     WriteGlobalSetting(Settings::values.bg_red);
     WriteGlobalSetting(Settings::values.bg_green);
     WriteGlobalSetting(Settings::values.bg_blue);
diff --git a/src/yuzu/configuration/configure_graphics.ui b/src/yuzu/configuration/configure_graphics.ui
index 099ddbb7c..43f1887d1 100644
--- a/src/yuzu/configuration/configure_graphics.ui
+++ b/src/yuzu/configuration/configure_graphics.ui
@@ -156,7 +156,7 @@
         <item>
          <widget class="QCheckBox" name="use_disk_shader_cache">
           <property name="text">
-           <string>Use disk shader cache</string>
+           <string>Use disk pipeline cache</string>
           </property>
          </widget>
         </item>
diff --git a/src/yuzu/configuration/configure_graphics_advanced.cpp b/src/yuzu/configuration/configure_graphics_advanced.cpp
index a31b8e192..bfd464061 100644
--- a/src/yuzu/configuration/configure_graphics_advanced.cpp
+++ b/src/yuzu/configuration/configure_graphics_advanced.cpp
@@ -28,7 +28,6 @@ void ConfigureGraphicsAdvanced::SetConfiguration() {
 
     ui->use_vsync->setChecked(Settings::values.use_vsync.GetValue());
     ui->use_asynchronous_shaders->setChecked(Settings::values.use_asynchronous_shaders.GetValue());
-    ui->use_caches_gc->setChecked(Settings::values.use_caches_gc.GetValue());
     ui->use_fast_gpu_time->setChecked(Settings::values.use_fast_gpu_time.GetValue());
 
     if (Settings::IsConfiguringGlobal()) {
@@ -55,8 +54,6 @@ void ConfigureGraphicsAdvanced::ApplyConfiguration() {
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_asynchronous_shaders,
                                              ui->use_asynchronous_shaders,
                                              use_asynchronous_shaders);
-    ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_caches_gc, ui->use_caches_gc,
-                                             use_caches_gc);
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_fast_gpu_time,
                                              ui->use_fast_gpu_time, use_fast_gpu_time);
 }
@@ -81,7 +78,6 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
         ui->use_asynchronous_shaders->setEnabled(
             Settings::values.use_asynchronous_shaders.UsingGlobal());
         ui->use_fast_gpu_time->setEnabled(Settings::values.use_fast_gpu_time.UsingGlobal());
-        ui->use_caches_gc->setEnabled(Settings::values.use_caches_gc.UsingGlobal());
         ui->anisotropic_filtering_combobox->setEnabled(
             Settings::values.max_anisotropy.UsingGlobal());
 
@@ -94,8 +90,6 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
                                             use_asynchronous_shaders);
     ConfigurationShared::SetColoredTristate(ui->use_fast_gpu_time,
                                             Settings::values.use_fast_gpu_time, use_fast_gpu_time);
-    ConfigurationShared::SetColoredTristate(ui->use_caches_gc, Settings::values.use_caches_gc,
-                                            use_caches_gc);
     ConfigurationShared::SetColoredComboBox(
         ui->gpu_accuracy, ui->label_gpu_accuracy,
         static_cast<int>(Settings::values.gpu_accuracy.GetValue(true)));
diff --git a/src/yuzu/configuration/configure_graphics_advanced.h b/src/yuzu/configuration/configure_graphics_advanced.h
index 7356e6916..13ba4ff6b 100644
--- a/src/yuzu/configuration/configure_graphics_advanced.h
+++ b/src/yuzu/configuration/configure_graphics_advanced.h
@@ -37,5 +37,4 @@ private:
     ConfigurationShared::CheckState use_vsync;
     ConfigurationShared::CheckState use_asynchronous_shaders;
     ConfigurationShared::CheckState use_fast_gpu_time;
-    ConfigurationShared::CheckState use_caches_gc;
 };
diff --git a/src/yuzu/configuration/configure_graphics_advanced.ui b/src/yuzu/configuration/configure_graphics_advanced.ui
index 4fe6b86ae..b91abc2f0 100644
--- a/src/yuzu/configuration/configure_graphics_advanced.ui
+++ b/src/yuzu/configuration/configure_graphics_advanced.ui
@@ -82,7 +82,7 @@
            <string>Enables asynchronous shader compilation, which may reduce shader stutter. This feature is experimental.</string>
           </property>
           <property name="text">
-           <string>Use asynchronous shader building (hack)</string>
+           <string>Use asynchronous shader building (Hack)</string>
           </property>
          </widget>
         </item>
@@ -92,17 +92,7 @@
             <string>Enables Fast GPU Time. This option will force most games to run at their highest native resolution.</string>
           </property>
           <property name="text">
-           <string>Use Fast GPU Time (hack)</string>
-          </property>
-         </widget>
-        </item>
-        <item>
-         <widget class="QCheckBox" name="use_caches_gc">
-          <property name="toolTip">
-           <string>Enables garbage collection for the GPU caches, this will try to keep VRAM within 3-4 GB by flushing the least used textures/buffers. May cause issues in a few games.</string>
-          </property>
-          <property name="text">
-           <string>Enable GPU cache garbage collection (experimental)</string>
+           <string>Use Fast GPU Time (Hack)</string>
           </property>
          </widget>
         </item>
diff --git a/src/yuzu/game_list.cpp b/src/yuzu/game_list.cpp
index e97804220..f9d949e75 100644
--- a/src/yuzu/game_list.cpp
+++ b/src/yuzu/game_list.cpp
@@ -515,16 +515,16 @@ void GameList::AddGamePopup(QMenu& context_menu, u64 program_id, const std::stri
     QAction* open_save_location = context_menu.addAction(tr("Open Save Data Location"));
     QAction* open_mod_location = context_menu.addAction(tr("Open Mod Data Location"));
     QAction* open_transferable_shader_cache =
-        context_menu.addAction(tr("Open Transferable Shader Cache"));
+        context_menu.addAction(tr("Open Transferable Pipeline Cache"));
     context_menu.addSeparator();
     QMenu* remove_menu = context_menu.addMenu(tr("Remove"));
     QAction* remove_update = remove_menu->addAction(tr("Remove Installed Update"));
     QAction* remove_dlc = remove_menu->addAction(tr("Remove All Installed DLC"));
     QAction* remove_custom_config = remove_menu->addAction(tr("Remove Custom Configuration"));
-    QAction* remove_gl_shader_cache = remove_menu->addAction(tr("Remove OpenGL Shader Cache"));
-    QAction* remove_vk_shader_cache = remove_menu->addAction(tr("Remove Vulkan Shader Cache"));
+    QAction* remove_gl_shader_cache = remove_menu->addAction(tr("Remove OpenGL Pipeline Cache"));
+    QAction* remove_vk_shader_cache = remove_menu->addAction(tr("Remove Vulkan Pipeline Cache"));
     remove_menu->addSeparator();
-    QAction* remove_shader_cache = remove_menu->addAction(tr("Remove All Shader Caches"));
+    QAction* remove_shader_cache = remove_menu->addAction(tr("Remove All Pipeline Caches"));
     QAction* remove_all_content = remove_menu->addAction(tr("Remove All Installed Contents"));
     QMenu* dump_romfs_menu = context_menu.addMenu(tr("Dump RomFS"));
     QAction* dump_romfs = dump_romfs_menu->addAction(tr("Dump RomFS"));
diff --git a/src/yuzu_cmd/config.cpp b/src/yuzu_cmd/config.cpp
index 4f14be524..757dd1ea0 100644
--- a/src/yuzu_cmd/config.cpp
+++ b/src/yuzu_cmd/config.cpp
@@ -468,7 +468,6 @@ void Config::ReadValues() {
     ReadSetting("Renderer", Settings::values.use_nvdec_emulation);
     ReadSetting("Renderer", Settings::values.accelerate_astc);
     ReadSetting("Renderer", Settings::values.use_fast_gpu_time);
-    ReadSetting("Renderer", Settings::values.use_caches_gc);
 
     ReadSetting("Renderer", Settings::values.bg_red);
     ReadSetting("Renderer", Settings::values.bg_green);