79 files changed, 2346 insertions, 1477 deletions

diff --git a/CMakeModules/GenerateSCMRev.cmake b/CMakeModules/GenerateSCMRev.cmake
index 09eabe2c7..21e03ae98 100644
--- a/CMakeModules/GenerateSCMRev.cmake
+++ b/CMakeModules/GenerateSCMRev.cmake
@@ -85,10 +85,12 @@ set(HASH_FILES
     "${VIDEO_CORE}/shader/decode/xmad.cpp"
     "${VIDEO_CORE}/shader/ast.cpp"
     "${VIDEO_CORE}/shader/ast.h"
-    "${VIDEO_CORE}/shader/control_flow.cpp"
-    "${VIDEO_CORE}/shader/control_flow.h"
     "${VIDEO_CORE}/shader/compiler_settings.cpp"
     "${VIDEO_CORE}/shader/compiler_settings.h"
+    "${VIDEO_CORE}/shader/const_buffer_locker.cpp"
+    "${VIDEO_CORE}/shader/const_buffer_locker.h"
+    "${VIDEO_CORE}/shader/control_flow.cpp"
+    "${VIDEO_CORE}/shader/control_flow.h"
     "${VIDEO_CORE}/shader/decode.cpp"
     "${VIDEO_CORE}/shader/expr.cpp"
     "${VIDEO_CORE}/shader/expr.h"
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 906c486fd..9c6f1c07c 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -74,10 +74,12 @@ add_custom_command(OUTPUT scm_rev.cpp
       "${VIDEO_CORE}/shader/decode/xmad.cpp"
       "${VIDEO_CORE}/shader/ast.cpp"
       "${VIDEO_CORE}/shader/ast.h"
-      "${VIDEO_CORE}/shader/control_flow.cpp"
-      "${VIDEO_CORE}/shader/control_flow.h"
       "${VIDEO_CORE}/shader/compiler_settings.cpp"
       "${VIDEO_CORE}/shader/compiler_settings.h"
+      "${VIDEO_CORE}/shader/const_buffer_locker.cpp"
+      "${VIDEO_CORE}/shader/const_buffer_locker.h"
+      "${VIDEO_CORE}/shader/control_flow.cpp"
+      "${VIDEO_CORE}/shader/control_flow.h"
       "${VIDEO_CORE}/shader/decode.cpp"
       "${VIDEO_CORE}/shader/expr.cpp"
       "${VIDEO_CORE}/shader/expr.h"
@@ -95,11 +97,11 @@ add_custom_command(OUTPUT scm_rev.cpp
 )
 
 add_library(common STATIC
+    algorithm.h
     alignment.h
     assert.h
     detached_tasks.cpp
     detached_tasks.h
-    binary_find.h
     bit_field.h
     bit_util.h
     cityhash.cpp
diff --git a/src/common/binary_find.h b/src/common/algorithm.h
index 5cc523bf9..e21b1373c 100644
--- a/src/common/binary_find.h
+++ b/src/common/algorithm.h
@@ -5,6 +5,12 @@
 #pragma once
 
 #include <algorithm>
+#include <functional>
+
+// Algorithms that operate on iterators, much like the <algorithm> header.
+//
+// Note: If the algorithm is not general-purpose and/or doesn't operate on iterators,
+//       it should probably not be placed within this header.
 
 namespace Common {
 
diff --git a/src/common/hash.h b/src/common/hash.h
index 40194d1ee..ebd4125e2 100644
--- a/src/common/hash.h
+++ b/src/common/hash.h
@@ -6,6 +6,8 @@
 
 #include <cstddef>
 #include <cstring>
+#include <utility>
+#include <boost/functional/hash.hpp>
 #include "common/cityhash.h"
 #include "common/common_types.h"
 
@@ -68,4 +70,13 @@ struct HashableStruct {
     }
 };
 
+struct PairHash {
+    template <class T1, class T2>
+    std::size_t operator()(const std::pair<T1, T2>& pair) const noexcept {
+        std::size_t seed = std::hash<T1>()(pair.first);
+        boost::hash_combine(seed, std::hash<T2>()(pair.second));
+        return seed;
+    }
+};
+
 } // namespace Common
diff --git a/src/core/core.cpp b/src/core/core.cpp
index d79045eea..eba17218a 100644
--- a/src/core/core.cpp
+++ b/src/core/core.cpp
@@ -112,8 +112,8 @@ FileSys::VirtualFile GetGameFileFromPath(const FileSys::VirtualFilesystem& vfs,
 }
 struct System::Impl {
     explicit Impl(System& system)
-        : kernel{system}, fs_controller{system}, cpu_core_manager{system},
-          applet_manager{system}, reporter{system} {}
+        : kernel{system}, fs_controller{system}, cpu_core_manager{system}, reporter{system},
+          applet_manager{system} {}
 
     Cpu& CurrentCpuCore() {
         return cpu_core_manager.GetCurrentCore();
@@ -240,22 +240,27 @@ struct System::Impl {
     }
 
     void Shutdown() {
-        // Log last frame performance stats
-        const auto perf_results = GetAndResetPerfStats();
-        telemetry_session->AddField(Telemetry::FieldType::Performance, "Shutdown_EmulationSpeed",
-                                    perf_results.emulation_speed * 100.0);
-        telemetry_session->AddField(Telemetry::FieldType::Performance, "Shutdown_Framerate",
-                                    perf_results.game_fps);
-        telemetry_session->AddField(Telemetry::FieldType::Performance, "Shutdown_Frametime",
-                                    perf_results.frametime * 1000.0);
-        telemetry_session->AddField(Telemetry::FieldType::Performance, "Mean_Frametime_MS",
-                                    perf_stats->GetMeanFrametime());
+        // Log last frame performance stats if game was loded
+        if (perf_stats) {
+            const auto perf_results = GetAndResetPerfStats();
+            telemetry_session->AddField(Telemetry::FieldType::Performance,
+                                        "Shutdown_EmulationSpeed",
+                                        perf_results.emulation_speed * 100.0);
+            telemetry_session->AddField(Telemetry::FieldType::Performance, "Shutdown_Framerate",
+                                        perf_results.game_fps);
+            telemetry_session->AddField(Telemetry::FieldType::Performance, "Shutdown_Frametime",
+                                        perf_results.frametime * 1000.0);
+            telemetry_session->AddField(Telemetry::FieldType::Performance, "Mean_Frametime_MS",
+                                        perf_stats->GetMeanFrametime());
+        }
 
         lm_manager.Flush();
 
         is_powered_on = false;
         exit_lock = false;
 
+        gpu_core->WaitIdle();
+
         // Shutdown emulation session
         renderer.reset();
         GDBStub::Shutdown();
diff --git a/src/core/hle/service/am/am.cpp b/src/core/hle/service/am/am.cpp
index 941ebc93a..3a32d5b41 100644
--- a/src/core/hle/service/am/am.cpp
+++ b/src/core/hle/service/am/am.cpp
@@ -1140,8 +1140,9 @@ void IApplicationFunctions::PopLaunchParameter(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_AM, "called, kind={:08X}", static_cast<u8>(kind));
 
     if (kind == LaunchParameterKind::ApplicationSpecific && !launch_popped_application_specific) {
-        const auto backend = BCAT::CreateBackendFromSettings(
-            [this](u64 tid) { return system.GetFileSystemController().GetBCATDirectory(tid); });
+        const auto backend = BCAT::CreateBackendFromSettings(system, [this](u64 tid) {
+            return system.GetFileSystemController().GetBCATDirectory(tid);
+        });
         const auto build_id_full = system.GetCurrentProcessBuildID();
         u64 build_id{};
         std::memcpy(&build_id, build_id_full.data(), sizeof(u64));
diff --git a/src/core/hle/service/apm/controller.cpp b/src/core/hle/service/apm/controller.cpp
index 073d0f6fa..25a886238 100644
--- a/src/core/hle/service/apm/controller.cpp
+++ b/src/core/hle/service/apm/controller.cpp
@@ -2,6 +2,10 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <algorithm>
+#include <array>
+#include <utility>
+
 #include "common/logging/log.h"
 #include "core/core_timing.h"
 #include "core/hle/service/apm/controller.h"
@@ -9,8 +13,7 @@
 
 namespace Service::APM {
 
-constexpr PerformanceConfiguration DEFAULT_PERFORMANCE_CONFIGURATION =
-    PerformanceConfiguration::Config7;
+constexpr auto DEFAULT_PERFORMANCE_CONFIGURATION = PerformanceConfiguration::Config7;
 
 Controller::Controller(Core::Timing::CoreTiming& core_timing)
     : core_timing{core_timing}, configs{
@@ -22,18 +25,35 @@ Controller::~Controller() = default;
 
 void Controller::SetPerformanceConfiguration(PerformanceMode mode,
                                              PerformanceConfiguration config) {
-    static const std::map<PerformanceConfiguration, u32> PCONFIG_TO_SPEED_MAP{
-        {PerformanceConfiguration::Config1, 1020},  {PerformanceConfiguration::Config2, 1020},
-        {PerformanceConfiguration::Config3, 1224},  {PerformanceConfiguration::Config4, 1020},
-        {PerformanceConfiguration::Config5, 1020},  {PerformanceConfiguration::Config6, 1224},
-        {PerformanceConfiguration::Config7, 1020},  {PerformanceConfiguration::Config8, 1020},
-        {PerformanceConfiguration::Config9, 1020},  {PerformanceConfiguration::Config10, 1020},
-        {PerformanceConfiguration::Config11, 1020}, {PerformanceConfiguration::Config12, 1020},
-        {PerformanceConfiguration::Config13, 1785}, {PerformanceConfiguration::Config14, 1785},
-        {PerformanceConfiguration::Config15, 1020}, {PerformanceConfiguration::Config16, 1020},
-    };
-
-    SetClockSpeed(PCONFIG_TO_SPEED_MAP.find(config)->second);
+    static constexpr std::array<std::pair<PerformanceConfiguration, u32>, 16> config_to_speed{{
+        {PerformanceConfiguration::Config1, 1020},
+        {PerformanceConfiguration::Config2, 1020},
+        {PerformanceConfiguration::Config3, 1224},
+        {PerformanceConfiguration::Config4, 1020},
+        {PerformanceConfiguration::Config5, 1020},
+        {PerformanceConfiguration::Config6, 1224},
+        {PerformanceConfiguration::Config7, 1020},
+        {PerformanceConfiguration::Config8, 1020},
+        {PerformanceConfiguration::Config9, 1020},
+        {PerformanceConfiguration::Config10, 1020},
+        {PerformanceConfiguration::Config11, 1020},
+        {PerformanceConfiguration::Config12, 1020},
+        {PerformanceConfiguration::Config13, 1785},
+        {PerformanceConfiguration::Config14, 1785},
+        {PerformanceConfiguration::Config15, 1020},
+        {PerformanceConfiguration::Config16, 1020},
+    }};
+
+    const auto iter = std::find_if(config_to_speed.cbegin(), config_to_speed.cend(),
+                                   [config](const auto& entry) { return entry.first == config; });
+
+    if (iter == config_to_speed.cend()) {
+        LOG_ERROR(Service_APM, "Invalid performance configuration value provided: {}",
+                  static_cast<u32>(config));
+        return;
+    }
+
+    SetClockSpeed(iter->second);
     configs.insert_or_assign(mode, config);
 }
 
@@ -48,7 +68,7 @@ void Controller::SetFromCpuBoostMode(CpuBoostMode mode) {
                                 BOOST_MODE_TO_CONFIG_MAP.at(static_cast<u32>(mode)));
 }
 
-PerformanceMode Controller::GetCurrentPerformanceMode() {
+PerformanceMode Controller::GetCurrentPerformanceMode() const {
     return Settings::values.use_docked_mode ? PerformanceMode::Docked : PerformanceMode::Handheld;
 }
 
diff --git a/src/core/hle/service/apm/controller.h b/src/core/hle/service/apm/controller.h
index 454caa6eb..af0c4cd34 100644
--- a/src/core/hle/service/apm/controller.h
+++ b/src/core/hle/service/apm/controller.h
@@ -56,7 +56,7 @@ public:
     void SetPerformanceConfiguration(PerformanceMode mode, PerformanceConfiguration config);
     void SetFromCpuBoostMode(CpuBoostMode mode);
 
-    PerformanceMode GetCurrentPerformanceMode();
+    PerformanceMode GetCurrentPerformanceMode() const;
     PerformanceConfiguration GetCurrentPerformanceConfiguration(PerformanceMode mode);
 
 private:
diff --git a/src/core/hle/service/bcat/backend/backend.cpp b/src/core/hle/service/bcat/backend/backend.cpp
index 9d6946bc5..b86fda29a 100644
--- a/src/core/hle/service/bcat/backend/backend.cpp
+++ b/src/core/hle/service/bcat/backend/backend.cpp
@@ -10,8 +10,8 @@
 
 namespace Service::BCAT {
 
-ProgressServiceBackend::ProgressServiceBackend(std::string_view event_name) {
-    auto& kernel{Core::System::GetInstance().Kernel()};
+ProgressServiceBackend::ProgressServiceBackend(Kernel::KernelCore& kernel,
+                                               std::string_view event_name) {
     event = Kernel::WritableEvent::CreateEventPair(
         kernel, Kernel::ResetType::Automatic,
         std::string("ProgressServiceBackend:UpdateEvent:").append(event_name));
diff --git a/src/core/hle/service/bcat/backend/backend.h b/src/core/hle/service/bcat/backend/backend.h
index 51dbd3316..ea4b16ad0 100644
--- a/src/core/hle/service/bcat/backend/backend.h
+++ b/src/core/hle/service/bcat/backend/backend.h
@@ -15,6 +15,14 @@
 #include "core/hle/kernel/writable_event.h"
 #include "core/hle/result.h"
 
+namespace Core {
+class System;
+}
+
+namespace Kernel {
+class KernelCore;
+}
+
 namespace Service::BCAT {
 
 struct DeliveryCacheProgressImpl;
@@ -88,7 +96,7 @@ public:
     void FinishDownload(ResultCode result);
 
 private:
-    explicit ProgressServiceBackend(std::string_view event_name);
+    explicit ProgressServiceBackend(Kernel::KernelCore& kernel, std::string_view event_name);
 
     Kernel::SharedPtr<Kernel::ReadableEvent> GetEvent() const;
     DeliveryCacheProgressImpl& GetImpl();
@@ -145,6 +153,6 @@ public:
     std::optional<std::vector<u8>> GetLaunchParameter(TitleIDVersion title) override;
 };
 
-std::unique_ptr<Backend> CreateBackendFromSettings(DirectoryGetter getter);
+std::unique_ptr<Backend> CreateBackendFromSettings(Core::System& system, DirectoryGetter getter);
 
 } // namespace Service::BCAT
diff --git a/src/core/hle/service/bcat/backend/boxcat.cpp b/src/core/hle/service/bcat/backend/boxcat.cpp
index 64022982b..918159e11 100644
--- a/src/core/hle/service/bcat/backend/boxcat.cpp
+++ b/src/core/hle/service/bcat/backend/boxcat.cpp
@@ -104,14 +104,15 @@ std::string GetZIPFilePath(u64 title_id) {
 
 // If the error is something the user should know about (build ID mismatch, bad client version),
 // display an error.
-void HandleDownloadDisplayResult(DownloadResult res) {
+void HandleDownloadDisplayResult(const AM::Applets::AppletManager& applet_manager,
+                                 DownloadResult res) {
     if (res == DownloadResult::Success || res == DownloadResult::NoResponse ||
         res == DownloadResult::GeneralWebError || res == DownloadResult::GeneralFSError ||
         res == DownloadResult::NoMatchTitleId || res == DownloadResult::InvalidContentType) {
         return;
     }
 
-    const auto& frontend{Core::System::GetInstance().GetAppletManager().GetAppletFrontendSet()};
+    const auto& frontend{applet_manager.GetAppletFrontendSet()};
     frontend.error->ShowCustomErrorText(
         ResultCode(-1), "There was an error while attempting to use Boxcat.",
         DOWNLOAD_RESULT_LOG_MESSAGES[static_cast<std::size_t>(res)], [] {});
@@ -264,12 +265,13 @@ private:
     u64 build_id;
 };
 
-Boxcat::Boxcat(DirectoryGetter getter) : Backend(std::move(getter)) {}
+Boxcat::Boxcat(AM::Applets::AppletManager& applet_manager_, DirectoryGetter getter)
+    : Backend(std::move(getter)), applet_manager{applet_manager_} {}
 
 Boxcat::~Boxcat() = default;
 
-void SynchronizeInternal(DirectoryGetter dir_getter, TitleIDVersion title,
-                         ProgressServiceBackend& progress,
+void SynchronizeInternal(AM::Applets::AppletManager& applet_manager, DirectoryGetter dir_getter,
+                         TitleIDVersion title, ProgressServiceBackend& progress,
                          std::optional<std::string> dir_name = {}) {
     progress.SetNeedHLELock(true);
 
@@ -295,7 +297,7 @@ void SynchronizeInternal(DirectoryGetter dir_getter, TitleIDVersion title,
             FileUtil::Delete(zip_path);
         }
 
-        HandleDownloadDisplayResult(res);
+        HandleDownloadDisplayResult(applet_manager, res);
         progress.FinishDownload(ERROR_GENERAL_BCAT_FAILURE);
         return;
     }
@@ -364,17 +366,24 @@ void SynchronizeInternal(DirectoryGetter dir_getter, TitleIDVersion title,
 
 bool Boxcat::Synchronize(TitleIDVersion title, ProgressServiceBackend& progress) {
     is_syncing.exchange(true);
-    std::thread([this, title, &progress] { SynchronizeInternal(dir_getter, title, progress); })
+
+    std::thread([this, title, &progress] {
+        SynchronizeInternal(applet_manager, dir_getter, title, progress);
+    })
         .detach();
+
     return true;
 }
 
 bool Boxcat::SynchronizeDirectory(TitleIDVersion title, std::string name,
                                   ProgressServiceBackend& progress) {
     is_syncing.exchange(true);
-    std::thread(
-        [this, title, name, &progress] { SynchronizeInternal(dir_getter, title, progress, name); })
+
+    std::thread([this, title, name, &progress] {
+        SynchronizeInternal(applet_manager, dir_getter, title, progress, name);
+    })
         .detach();
+
     return true;
 }
 
@@ -420,7 +429,7 @@ std::optional<std::vector<u8>> Boxcat::GetLaunchParameter(TitleIDVersion title)
                 FileUtil::Delete(path);
             }
 
-            HandleDownloadDisplayResult(res);
+            HandleDownloadDisplayResult(applet_manager, res);
             return std::nullopt;
         }
     }
diff --git a/src/core/hle/service/bcat/backend/boxcat.h b/src/core/hle/service/bcat/backend/boxcat.h
index 601151189..d65b42e58 100644
--- a/src/core/hle/service/bcat/backend/boxcat.h
+++ b/src/core/hle/service/bcat/backend/boxcat.h
@@ -9,6 +9,10 @@
 #include <optional>
 #include "core/hle/service/bcat/backend/backend.h"
 
+namespace Service::AM::Applets {
+class AppletManager;
+}
+
 namespace Service::BCAT {
 
 struct EventStatus {
@@ -20,12 +24,13 @@ struct EventStatus {
 /// Boxcat is yuzu's custom backend implementation of Nintendo's BCAT service. It is free to use and
 /// doesn't require a switch or nintendo account. The content is controlled by the yuzu team.
 class Boxcat final : public Backend {
-    friend void SynchronizeInternal(DirectoryGetter dir_getter, TitleIDVersion title,
+    friend void SynchronizeInternal(AM::Applets::AppletManager& applet_manager,
+                                    DirectoryGetter dir_getter, TitleIDVersion title,
                                     ProgressServiceBackend& progress,
                                     std::optional<std::string> dir_name);
 
 public:
-    explicit Boxcat(DirectoryGetter getter);
+    explicit Boxcat(AM::Applets::AppletManager& applet_manager_, DirectoryGetter getter);
     ~Boxcat() override;
 
     bool Synchronize(TitleIDVersion title, ProgressServiceBackend& progress) override;
@@ -53,6 +58,7 @@ private:
 
     class Client;
     std::unique_ptr<Client> client;
+    AM::Applets::AppletManager& applet_manager;
 };
 
 } // namespace Service::BCAT
diff --git a/src/core/hle/service/bcat/module.cpp b/src/core/hle/service/bcat/module.cpp
index 4e4aa758b..6d9d1527d 100644
--- a/src/core/hle/service/bcat/module.cpp
+++ b/src/core/hle/service/bcat/module.cpp
@@ -125,7 +125,11 @@ private:
 class IBcatService final : public ServiceFramework<IBcatService> {
 public:
     explicit IBcatService(Core::System& system_, Backend& backend_)
-        : ServiceFramework("IBcatService"), system{system_}, backend{backend_} {
+        : ServiceFramework("IBcatService"), system{system_}, backend{backend_},
+          progress{{
+              ProgressServiceBackend{system_.Kernel(), "Normal"},
+              ProgressServiceBackend{system_.Kernel(), "Directory"},
+          }} {
         // clang-format off
         static const FunctionInfo functions[] = {
             {10100, &IBcatService::RequestSyncDeliveryCache, "RequestSyncDeliveryCache"},
@@ -249,10 +253,7 @@ private:
     Core::System& system;
     Backend& backend;
 
-    std::array<ProgressServiceBackend, static_cast<std::size_t>(SyncType::Count)> progress{
-        ProgressServiceBackend{"Normal"},
-        ProgressServiceBackend{"Directory"},
-    };
+    std::array<ProgressServiceBackend, static_cast<std::size_t>(SyncType::Count)> progress;
 };
 
 void Module::Interface::CreateBcatService(Kernel::HLERequestContext& ctx) {
@@ -557,12 +558,12 @@ void Module::Interface::CreateDeliveryCacheStorageServiceWithApplicationId(
     rb.PushIpcInterface<IDeliveryCacheStorageService>(fsc.GetBCATDirectory(title_id));
 }
 
-std::unique_ptr<Backend> CreateBackendFromSettings(DirectoryGetter getter) {
-    const auto backend = Settings::values.bcat_backend;
-
+std::unique_ptr<Backend> CreateBackendFromSettings([[maybe_unused]] Core::System& system,
+                                                   DirectoryGetter getter) {
 #ifdef YUZU_ENABLE_BOXCAT
-    if (backend == "boxcat")
-        return std::make_unique<Boxcat>(std::move(getter));
+    if (Settings::values.bcat_backend == "boxcat") {
+        return std::make_unique<Boxcat>(system.GetAppletManager(), std::move(getter));
+    }
 #endif
 
     return std::make_unique<NullBackend>(std::move(getter));
@@ -571,7 +572,8 @@ std::unique_ptr<Backend> CreateBackendFromSettings(DirectoryGetter getter) {
 Module::Interface::Interface(Core::System& system_, std::shared_ptr<Module> module_,
                              FileSystem::FileSystemController& fsc_, const char* name)
     : ServiceFramework(name), fsc{fsc_}, module{std::move(module_)},
-      backend{CreateBackendFromSettings([&fsc_](u64 tid) { return fsc_.GetBCATDirectory(tid); })},
+      backend{CreateBackendFromSettings(system_,
+                                        [&fsc_](u64 tid) { return fsc_.GetBCATDirectory(tid); })},
       system{system_} {}
 
 Module::Interface::~Interface() = default;
diff --git a/src/core/hle/service/hid/controllers/npad.cpp b/src/core/hle/service/hid/controllers/npad.cpp
index a2b25a796..81bd2f3cb 100644
--- a/src/core/hle/service/hid/controllers/npad.cpp
+++ b/src/core/hle/service/hid/controllers/npad.cpp
@@ -583,36 +583,6 @@ bool Controller_NPad::SwapNpadAssignment(u32 npad_id_1, u32 npad_id_2) {
     return true;
 }
 
-bool Controller_NPad::IsControllerSupported(NPadControllerType controller) {
-    if (controller == NPadControllerType::Handheld) {
-        // Handheld is not even a supported type, lets stop here
-        if (std::find(supported_npad_id_types.begin(), supported_npad_id_types.end(),
-                      NPAD_HANDHELD) == supported_npad_id_types.end()) {
-            return false;
-        }
-        // Handheld should not be supported in docked mode
-        if (Settings::values.use_docked_mode) {
-            return false;
-        }
-    }
-    switch (controller) {
-    case NPadControllerType::ProController:
-        return style.pro_controller;
-    case NPadControllerType::Handheld:
-        return style.handheld;
-    case NPadControllerType::JoyDual:
-        return style.joycon_dual;
-    case NPadControllerType::JoyLeft:
-        return style.joycon_left;
-    case NPadControllerType::JoyRight:
-        return style.joycon_right;
-    case NPadControllerType::Pokeball:
-        return style.pokeball;
-    default:
-        return false;
-    }
-}
-
 Controller_NPad::LedPattern Controller_NPad::GetLedPattern(u32 npad_id) {
     if (npad_id == npad_id_list.back() || npad_id == npad_id_list[npad_id_list.size() - 2]) {
         // These are controllers without led patterns
@@ -659,25 +629,24 @@ void Controller_NPad::ClearAllConnectedControllers() {
 }
 
 void Controller_NPad::DisconnectAllConnectedControllers() {
-    std::for_each(connected_controllers.begin(), connected_controllers.end(),
-                  [](ControllerHolder& controller) { controller.is_connected = false; });
+    for (ControllerHolder& controller : connected_controllers) {
+        controller.is_connected = false;
+    }
 }
 
 void Controller_NPad::ConnectAllDisconnectedControllers() {
-    std::for_each(connected_controllers.begin(), connected_controllers.end(),
-                  [](ControllerHolder& controller) {
-                      if (controller.type != NPadControllerType::None && !controller.is_connected) {
-                          controller.is_connected = false;
-                      }
-                  });
+    for (ControllerHolder& controller : connected_controllers) {
+        if (controller.type != NPadControllerType::None && !controller.is_connected) {
+            controller.is_connected = true;
+        }
+    }
 }
 
 void Controller_NPad::ClearAllControllers() {
-    std::for_each(connected_controllers.begin(), connected_controllers.end(),
-                  [](ControllerHolder& controller) {
-                      controller.type = NPadControllerType::None;
-                      controller.is_connected = false;
-                  });
+    for (ControllerHolder& controller : connected_controllers) {
+        controller.type = NPadControllerType::None;
+        controller.is_connected = false;
+    }
 }
 
 u32 Controller_NPad::GetAndResetPressState() {
@@ -685,10 +654,10 @@ u32 Controller_NPad::GetAndResetPressState() {
 }
 
 bool Controller_NPad::IsControllerSupported(NPadControllerType controller) const {
-    const bool support_handheld =
-        std::find(supported_npad_id_types.begin(), supported_npad_id_types.end(), NPAD_HANDHELD) !=
-        supported_npad_id_types.end();
     if (controller == NPadControllerType::Handheld) {
+        const bool support_handheld =
+            std::find(supported_npad_id_types.begin(), supported_npad_id_types.end(),
+                      NPAD_HANDHELD) != supported_npad_id_types.end();
         // Handheld is not even a supported type, lets stop here
         if (!support_handheld) {
             return false;
@@ -700,6 +669,7 @@ bool Controller_NPad::IsControllerSupported(NPadControllerType controller) const
 
         return true;
     }
+
     if (std::any_of(supported_npad_id_types.begin(), supported_npad_id_types.end(),
                     [](u32 npad_id) { return npad_id <= MAX_NPAD_ID; })) {
         switch (controller) {
@@ -717,6 +687,7 @@ bool Controller_NPad::IsControllerSupported(NPadControllerType controller) const
             return false;
         }
     }
+
     return false;
 }
 
@@ -795,6 +766,7 @@ Controller_NPad::NPadControllerType Controller_NPad::DecideBestController(
         priority_list.push_back(NPadControllerType::JoyLeft);
         priority_list.push_back(NPadControllerType::JoyRight);
         priority_list.push_back(NPadControllerType::JoyDual);
+        break;
     }
 
     const auto iter = std::find_if(priority_list.begin(), priority_list.end(),
diff --git a/src/core/hle/service/hid/controllers/npad.h b/src/core/hle/service/hid/controllers/npad.h
index 1bc3d55d6..16c4caa1f 100644
--- a/src/core/hle/service/hid/controllers/npad.h
+++ b/src/core/hle/service/hid/controllers/npad.h
@@ -301,6 +301,11 @@ private:
         bool is_connected;
     };
 
+    void InitNewlyAddedControler(std::size_t controller_idx);
+    bool IsControllerSupported(NPadControllerType controller) const;
+    NPadControllerType DecideBestController(NPadControllerType priority) const;
+    void RequestPadStateUpdate(u32 npad_id);
+
     u32 press_state{};
 
     NPadType style{};
@@ -321,12 +326,7 @@ private:
     std::array<ControllerHolder, 10> connected_controllers{};
     bool can_controllers_vibrate{true};
 
-    void InitNewlyAddedControler(std::size_t controller_idx);
-    bool IsControllerSupported(NPadControllerType controller) const;
-    NPadControllerType DecideBestController(NPadControllerType priority) const;
-    void RequestPadStateUpdate(u32 npad_id);
     std::array<ControllerPad, 10> npad_pad_states{};
-    bool IsControllerSupported(NPadControllerType controller);
     bool is_in_lr_assignment_mode{false};
     Core::System& system;
 };
diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
index f764388bc..3f7b8e670 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
@@ -5,6 +5,7 @@
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "core/core.h"
+#include "core/core_timing.h"
 #include "core/hle/service/nvdrv/devices/nvdisp_disp0.h"
 #include "core/hle/service/nvdrv/devices/nvmap.h"
 #include "core/perf_stats.h"
@@ -38,7 +39,10 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u3
         transform, crop_rect};
 
     system.GetPerfStats().EndGameFrame();
+    system.GetPerfStats().EndSystemFrame();
     system.GPU().SwapBuffers(&framebuffer);
+    system.FrameLimiter().DoFrameLimiting(system.CoreTiming().GetGlobalTimeUs());
+    system.GetPerfStats().BeginSystemFrame();
 }
 
 } // namespace Service::Nvidia::Devices
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
index eb88fee1b..b27ee0502 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
@@ -63,16 +63,26 @@ u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>&
         return NvResult::BadParameter;
     }
 
+    u32 event_id = params.value & 0x00FF;
+
+    if (event_id >= MaxNvEvents) {
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::BadParameter;
+    }
+
+    auto event = events_interface.events[event_id];
     auto& gpu = system.GPU();
     // This is mostly to take into account unimplemented features. As synced
     // gpu is always synced.
     if (!gpu.IsAsync()) {
+        event.writable->Signal();
         return NvResult::Success;
     }
     auto lock = gpu.LockSync();
     const u32 current_syncpoint_value = gpu.GetSyncpointValue(params.syncpt_id);
     const s32 diff = current_syncpoint_value - params.threshold;
     if (diff >= 0) {
+        event.writable->Signal();
         params.value = current_syncpoint_value;
         std::memcpy(output.data(), &params, sizeof(params));
         return NvResult::Success;
@@ -88,27 +98,6 @@ u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>&
         return NvResult::Timeout;
     }
 
-    u32 event_id;
-    if (is_async) {
-        event_id = params.value & 0x00FF;
-        if (event_id >= MaxNvEvents) {
-            std::memcpy(output.data(), &params, sizeof(params));
-            return NvResult::BadParameter;
-        }
-    } else {
-        if (ctrl.fresh_call) {
-            const auto result = events_interface.GetFreeEvent();
-            if (result) {
-                event_id = *result;
-            } else {
-                LOG_CRITICAL(Service_NVDRV, "No Free Events available!");
-                event_id = params.value & 0x00FF;
-            }
-        } else {
-            event_id = ctrl.event_id;
-        }
-    }
-
     EventState status = events_interface.status[event_id];
     if (event_id < MaxNvEvents || status == EventState::Free || status == EventState::Registered) {
         events_interface.SetEventStatus(event_id, EventState::Waiting);
@@ -120,7 +109,7 @@ u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>&
             params.value = ((params.syncpt_id & 0xfff) << 16) | 0x10000000;
         }
         params.value |= event_id;
-        events_interface.events[event_id].writable->Clear();
+        event.writable->Clear();
         gpu.RegisterSyncptInterrupt(params.syncpt_id, target_value);
         if (!is_async && ctrl.fresh_call) {
             ctrl.must_delay = true;
diff --git a/src/core/hle/service/nvdrv/interface.cpp b/src/core/hle/service/nvdrv/interface.cpp
index 5e0c23602..68d139cfb 100644
--- a/src/core/hle/service/nvdrv/interface.cpp
+++ b/src/core/hle/service/nvdrv/interface.cpp
@@ -134,7 +134,9 @@ void NVDRV::QueryEvent(Kernel::HLERequestContext& ctx) {
     IPC::ResponseBuilder rb{ctx, 3, 1};
     rb.Push(RESULT_SUCCESS);
     if (event_id < MaxNvEvents) {
-        rb.PushCopyObjects(nvdrv->GetEvent(event_id));
+        auto event = nvdrv->GetEvent(event_id);
+        event->Clear();
+        rb.PushCopyObjects(event);
         rb.Push<u32>(NvResult::Success);
     } else {
         rb.Push<u32>(0);
diff --git a/src/core/hle/service/nvdrv/nvdrv.cpp b/src/core/hle/service/nvdrv/nvdrv.cpp
index 307a7e928..7bfb99e34 100644
--- a/src/core/hle/service/nvdrv/nvdrv.cpp
+++ b/src/core/hle/service/nvdrv/nvdrv.cpp
@@ -40,8 +40,8 @@ Module::Module(Core::System& system) {
     auto& kernel = system.Kernel();
     for (u32 i = 0; i < MaxNvEvents; i++) {
         std::string event_label = fmt::format("NVDRV::NvEvent_{}", i);
-        events_interface.events[i] = Kernel::WritableEvent::CreateEventPair(
-            kernel, Kernel::ResetType::Automatic, event_label);
+        events_interface.events[i] =
+            Kernel::WritableEvent::CreateEventPair(kernel, Kernel::ResetType::Manual, event_label);
         events_interface.status[i] = EventState::Free;
         events_interface.registered[i] = false;
     }
diff --git a/src/core/hle/service/nvflinger/buffer_queue.cpp b/src/core/hle/service/nvflinger/buffer_queue.cpp
index e1a07d3ee..55b68eb0c 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.cpp
+++ b/src/core/hle/service/nvflinger/buffer_queue.cpp
@@ -14,8 +14,8 @@
 
 namespace Service::NVFlinger {
 
-BufferQueue::BufferQueue(u32 id, u64 layer_id) : id(id), layer_id(layer_id) {
-    auto& kernel = Core::System::GetInstance().Kernel();
+BufferQueue::BufferQueue(Kernel::KernelCore& kernel, u32 id, u64 layer_id)
+    : id(id), layer_id(layer_id) {
     buffer_wait_event = Kernel::WritableEvent::CreateEventPair(kernel, Kernel::ResetType::Manual,
                                                                "BufferQueue NativeHandle");
 }
diff --git a/src/core/hle/service/nvflinger/buffer_queue.h b/src/core/hle/service/nvflinger/buffer_queue.h
index 356bedb81..8f9b18547 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.h
+++ b/src/core/hle/service/nvflinger/buffer_queue.h
@@ -15,6 +15,10 @@
 #include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nvdrv/nvdata.h"
 
+namespace Kernel {
+class KernelCore;
+}
+
 namespace Service::NVFlinger {
 
 struct IGBPBuffer {
@@ -44,7 +48,7 @@ public:
         NativeWindowFormat = 2,
     };
 
-    BufferQueue(u32 id, u64 layer_id);
+    explicit BufferQueue(Kernel::KernelCore& kernel, u32 id, u64 layer_id);
     ~BufferQueue();
 
     enum class BufferTransformFlags : u32 {
diff --git a/src/core/hle/service/nvflinger/nvflinger.cpp b/src/core/hle/service/nvflinger/nvflinger.cpp
index 2e4d707b9..cc9522aad 100644
--- a/src/core/hle/service/nvflinger/nvflinger.cpp
+++ b/src/core/hle/service/nvflinger/nvflinger.cpp
@@ -83,7 +83,7 @@ std::optional<u64> NVFlinger::CreateLayer(u64 display_id) {
 
     const u64 layer_id = next_layer_id++;
     const u32 buffer_queue_id = next_buffer_queue_id++;
-    buffer_queues.emplace_back(buffer_queue_id, layer_id);
+    buffer_queues.emplace_back(system.Kernel(), buffer_queue_id, layer_id);
     display->CreateLayer(layer_id, buffer_queues.back());
     return layer_id;
 }
@@ -187,14 +187,18 @@ void NVFlinger::Compose() {
         MicroProfileFlip();
 
         if (!buffer) {
-            // There was no queued buffer to draw, render previous frame
-            system.GetPerfStats().EndGameFrame();
-            system.GPU().SwapBuffers({});
             continue;
         }
 
         const auto& igbp_buffer = buffer->get().igbp_buffer;
 
+        const auto& gpu = system.GPU();
+        const auto& multi_fence = buffer->get().multi_fence;
+        for (u32 fence_id = 0; fence_id < multi_fence.num_fences; fence_id++) {
+            const auto& fence = multi_fence.fences[fence_id];
+            gpu.WaitFence(fence.id, fence.value);
+        }
+
         // Now send the buffer to the GPU for drawing.
         // TODO(Subv): Support more than just disp0. The display device selection is probably based
         // on which display we're drawing (Default, Internal, External, etc)
diff --git a/src/core/memory/cheat_engine.cpp b/src/core/memory/cheat_engine.cpp
index b56cb0627..10821d452 100644
--- a/src/core/memory/cheat_engine.cpp
+++ b/src/core/memory/cheat_engine.cpp
@@ -22,7 +22,7 @@ constexpr u32 KEYPAD_BITMASK = 0x3FFFFFF;
 
 StandardVmCallbacks::StandardVmCallbacks(const Core::System& system,
                                          const CheatProcessMetadata& metadata)
-    : system(system), metadata(metadata) {}
+    : metadata(metadata), system(system) {}
 
 StandardVmCallbacks::~StandardVmCallbacks() = default;
 
@@ -176,9 +176,8 @@ std::vector<CheatEntry> TextCheatParser::Parse(const Core::System& system,
 
 CheatEngine::CheatEngine(Core::System& system, std::vector<CheatEntry> cheats,
                          const std::array<u8, 0x20>& build_id)
-    : system{system}, core_timing{system.CoreTiming()}, vm{std::make_unique<StandardVmCallbacks>(
-                                                            system, metadata)},
-      cheats(std::move(cheats)) {
+    : vm{std::make_unique<StandardVmCallbacks>(system, metadata)},
+      cheats(std::move(cheats)), core_timing{system.CoreTiming()}, system{system} {
     metadata.main_nso_build_id = build_id;
 }
 
diff --git a/src/core/memory/dmnt_cheat_vm.cpp b/src/core/memory/dmnt_cheat_vm.cpp
index cc16d15a4..4f4fa5099 100644
--- a/src/core/memory/dmnt_cheat_vm.cpp
+++ b/src/core/memory/dmnt_cheat_vm.cpp
@@ -1133,8 +1133,8 @@ void DmntCheatVm::Execute(const CheatProcessMetadata& metadata) {
             case SaveRestoreRegisterOpType::ClearRegs:
             case SaveRestoreRegisterOpType::Restore:
             default:
-                src = registers.data();
-                dst = saved_values.data();
+                src = saved_values.data();
+                dst = registers.data();
                 break;
             }
             for (std::size_t i = 0; i < NumRegisters; i++) {
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index eaa694ff8..cb6eda1b8 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -6,6 +6,7 @@ add_library(video_core STATIC
     dma_pusher.h
     debug_utils/debug_utils.cpp
     debug_utils/debug_utils.h
+    engines/const_buffer_engine_interface.h
     engines/const_buffer_info.h
     engines/engine_upload.cpp
     engines/engine_upload.h
@@ -107,10 +108,12 @@ add_library(video_core STATIC
     shader/decode/other.cpp
     shader/ast.cpp
     shader/ast.h
-    shader/control_flow.cpp
-    shader/control_flow.h
     shader/compiler_settings.cpp
     shader/compiler_settings.h
+    shader/const_buffer_locker.cpp
+    shader/const_buffer_locker.h
+    shader/control_flow.cpp
+    shader/control_flow.h
     shader/decode.cpp
     shader/expr.cpp
     shader/expr.h
diff --git a/src/video_core/engines/const_buffer_engine_interface.h b/src/video_core/engines/const_buffer_engine_interface.h
new file mode 100644
index 000000000..ac27b6cbe
--- /dev/null
+++ b/src/video_core/engines/const_buffer_engine_interface.h
@@ -0,0 +1,119 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <type_traits>
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "video_core/engines/shader_bytecode.h"
+#include "video_core/textures/texture.h"
+
+namespace Tegra::Engines {
+
+enum class ShaderType : u32 {
+    Vertex = 0,
+    TesselationControl = 1,
+    TesselationEval = 2,
+    Geometry = 3,
+    Fragment = 4,
+    Compute = 5,
+};
+
+struct SamplerDescriptor {
+    union {
+        BitField<0, 20, Tegra::Shader::TextureType> texture_type;
+        BitField<20, 1, u32> is_array;
+        BitField<21, 1, u32> is_buffer;
+        BitField<22, 1, u32> is_shadow;
+        u32 raw{};
+    };
+
+    bool operator==(const SamplerDescriptor& rhs) const noexcept {
+        return raw == rhs.raw;
+    }
+
+    bool operator!=(const SamplerDescriptor& rhs) const noexcept {
+        return !operator==(rhs);
+    }
+
+    static SamplerDescriptor FromTicTexture(Tegra::Texture::TextureType tic_texture_type) {
+        SamplerDescriptor result;
+        switch (tic_texture_type) {
+        case Tegra::Texture::TextureType::Texture1D:
+            result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
+            result.is_array.Assign(0);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        case Tegra::Texture::TextureType::Texture2D:
+            result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
+            result.is_array.Assign(0);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        case Tegra::Texture::TextureType::Texture3D:
+            result.texture_type.Assign(Tegra::Shader::TextureType::Texture3D);
+            result.is_array.Assign(0);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        case Tegra::Texture::TextureType::TextureCubemap:
+            result.texture_type.Assign(Tegra::Shader::TextureType::TextureCube);
+            result.is_array.Assign(0);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        case Tegra::Texture::TextureType::Texture1DArray:
+            result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
+            result.is_array.Assign(1);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        case Tegra::Texture::TextureType::Texture2DArray:
+            result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
+            result.is_array.Assign(1);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        case Tegra::Texture::TextureType::Texture1DBuffer:
+            result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
+            result.is_array.Assign(0);
+            result.is_buffer.Assign(1);
+            result.is_shadow.Assign(0);
+            return result;
+        case Tegra::Texture::TextureType::Texture2DNoMipmap:
+            result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
+            result.is_array.Assign(0);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        case Tegra::Texture::TextureType::TextureCubeArray:
+            result.texture_type.Assign(Tegra::Shader::TextureType::TextureCube);
+            result.is_array.Assign(1);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        default:
+            result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
+            result.is_array.Assign(0);
+            result.is_buffer.Assign(0);
+            result.is_shadow.Assign(0);
+            return result;
+        }
+    }
+};
+static_assert(std::is_trivially_copyable_v<SamplerDescriptor>);
+
+class ConstBufferEngineInterface {
+public:
+    virtual ~ConstBufferEngineInterface() = default;
+    virtual u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const = 0;
+    virtual SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const = 0;
+    virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
+                                                    u64 offset) const = 0;
+    virtual u32 GetBoundBuffer() const = 0;
+};
+
+} // namespace Tegra::Engines
diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp
index 63d449135..91adef360 100644
--- a/src/video_core/engines/kepler_compute.cpp
+++ b/src/video_core/engines/kepler_compute.cpp
@@ -70,13 +70,31 @@ Texture::FullTextureInfo KeplerCompute::GetTextureInfo(const Texture::TextureHan
                                     GetTSCEntry(tex_handle.tsc_id)};
 }
 
-u32 KeplerCompute::AccessConstBuffer32(u64 const_buffer, u64 offset) const {
+u32 KeplerCompute::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const {
+    ASSERT(stage == ShaderType::Compute);
     const auto& buffer = launch_description.const_buffer_config[const_buffer];
     u32 result;
     std::memcpy(&result, memory_manager.GetPointer(buffer.Address() + offset), sizeof(u32));
     return result;
 }
 
+SamplerDescriptor KeplerCompute::AccessBoundSampler(ShaderType stage, u64 offset) const {
+    return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle));
+}
+
+SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 const_buffer,
+                                                       u64 offset) const {
+    ASSERT(stage == ShaderType::Compute);
+    const auto& tex_info_buffer = launch_description.const_buffer_config[const_buffer];
+    const GPUVAddr tex_info_address = tex_info_buffer.Address() + offset;
+
+    const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
+    const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle, offset);
+    SamplerDescriptor result = SamplerDescriptor::FromTicTexture(tex_info.tic.texture_type.Value());
+    result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
+    return result;
+}
+
 void KeplerCompute::ProcessLaunch() {
     const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address();
     memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description,
diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h
index 90cf650d2..8e7182727 100644
--- a/src/video_core/engines/kepler_compute.h
+++ b/src/video_core/engines/kepler_compute.h
@@ -10,6 +10,7 @@
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "video_core/engines/const_buffer_engine_interface.h"
 #include "video_core/engines/engine_upload.h"
 #include "video_core/gpu.h"
 #include "video_core/textures/texture.h"
@@ -37,7 +38,7 @@ namespace Tegra::Engines {
 #define KEPLER_COMPUTE_REG_INDEX(field_name)                                                       \
     (offsetof(Tegra::Engines::KeplerCompute::Regs, field_name) / sizeof(u32))
 
-class KeplerCompute final {
+class KeplerCompute final : public ConstBufferEngineInterface {
 public:
     explicit KeplerCompute(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
                            MemoryManager& memory_manager);
@@ -201,7 +202,16 @@ public:
     Texture::FullTextureInfo GetTextureInfo(const Texture::TextureHandle tex_handle,
                                             std::size_t offset) const;
 
-    u32 AccessConstBuffer32(u64 const_buffer, u64 offset) const;
+    u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
+
+    SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
+
+    SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
+                                            u64 offset) const override;
+
+    u32 GetBoundBuffer() const override {
+        return regs.tex_cb_index;
+    }
 
 private:
     Core::System& system;
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 7802fd808..514ed93fa 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -98,10 +98,10 @@ void Maxwell3D::InitializeRegisterDefaults() {
     mme_inline[MAXWELL3D_REG_INDEX(index_array.count)] = true;
 }
 
-#define DIRTY_REGS_POS(field_name) (offsetof(Maxwell3D::DirtyRegs, field_name))
+#define DIRTY_REGS_POS(field_name) static_cast<u8>(offsetof(Maxwell3D::DirtyRegs, field_name))
 
 void Maxwell3D::InitDirtySettings() {
-    const auto set_block = [this](const u32 start, const u32 range, const u8 position) {
+    const auto set_block = [this](std::size_t start, std::size_t range, u8 position) {
         const auto start_itr = dirty_pointers.begin() + start;
         const auto end_itr = start_itr + range;
         std::fill(start_itr, end_itr, position);
@@ -112,10 +112,10 @@ void Maxwell3D::InitDirtySettings() {
     constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
     constexpr u32 rt_start_reg = MAXWELL3D_REG_INDEX(rt);
     constexpr u32 rt_end_reg = rt_start_reg + registers_per_rt * 8;
-    u32 rt_dirty_reg = DIRTY_REGS_POS(render_target);
+    u8 rt_dirty_reg = DIRTY_REGS_POS(render_target);
     for (u32 rt_reg = rt_start_reg; rt_reg < rt_end_reg; rt_reg += registers_per_rt) {
         set_block(rt_reg, registers_per_rt, rt_dirty_reg);
-        rt_dirty_reg++;
+        ++rt_dirty_reg;
     }
     constexpr u32 depth_buffer_flag = DIRTY_REGS_POS(depth_buffer);
     dirty_pointers[MAXWELL3D_REG_INDEX(zeta_enable)] = depth_buffer_flag;
@@ -129,35 +129,35 @@ void Maxwell3D::InitDirtySettings() {
     constexpr u32 vertex_array_start = MAXWELL3D_REG_INDEX(vertex_array);
     constexpr u32 vertex_array_size = sizeof(regs.vertex_array[0]) / sizeof(u32);
     constexpr u32 vertex_array_end = vertex_array_start + vertex_array_size * Regs::NumVertexArrays;
-    u32 va_reg = DIRTY_REGS_POS(vertex_array);
-    u32 vi_reg = DIRTY_REGS_POS(vertex_instance);
+    u8 va_dirty_reg = DIRTY_REGS_POS(vertex_array);
+    u8 vi_dirty_reg = DIRTY_REGS_POS(vertex_instance);
     for (u32 vertex_reg = vertex_array_start; vertex_reg < vertex_array_end;
          vertex_reg += vertex_array_size) {
-        set_block(vertex_reg, 3, va_reg);
+        set_block(vertex_reg, 3, va_dirty_reg);
         // The divisor concerns vertex array instances
-        dirty_pointers[vertex_reg + 3] = vi_reg;
-        va_reg++;
-        vi_reg++;
+        dirty_pointers[static_cast<std::size_t>(vertex_reg) + 3] = vi_dirty_reg;
+        ++va_dirty_reg;
+        ++vi_dirty_reg;
     }
     constexpr u32 vertex_limit_start = MAXWELL3D_REG_INDEX(vertex_array_limit);
     constexpr u32 vertex_limit_size = sizeof(regs.vertex_array_limit[0]) / sizeof(u32);
     constexpr u32 vertex_limit_end = vertex_limit_start + vertex_limit_size * Regs::NumVertexArrays;
-    va_reg = DIRTY_REGS_POS(vertex_array);
+    va_dirty_reg = DIRTY_REGS_POS(vertex_array);
     for (u32 vertex_reg = vertex_limit_start; vertex_reg < vertex_limit_end;
          vertex_reg += vertex_limit_size) {
-        set_block(vertex_reg, vertex_limit_size, va_reg);
-        va_reg++;
+        set_block(vertex_reg, vertex_limit_size, va_dirty_reg);
+        va_dirty_reg++;
     }
     constexpr u32 vertex_instance_start = MAXWELL3D_REG_INDEX(instanced_arrays);
     constexpr u32 vertex_instance_size =
         sizeof(regs.instanced_arrays.is_instanced[0]) / sizeof(u32);
     constexpr u32 vertex_instance_end =
         vertex_instance_start + vertex_instance_size * Regs::NumVertexArrays;
-    vi_reg = DIRTY_REGS_POS(vertex_instance);
+    vi_dirty_reg = DIRTY_REGS_POS(vertex_instance);
     for (u32 vertex_reg = vertex_instance_start; vertex_reg < vertex_instance_end;
          vertex_reg += vertex_instance_size) {
-        set_block(vertex_reg, vertex_instance_size, vi_reg);
-        vi_reg++;
+        set_block(vertex_reg, vertex_instance_size, vi_dirty_reg);
+        vi_dirty_reg++;
     }
     set_block(MAXWELL3D_REG_INDEX(vertex_attrib_format), regs.vertex_attrib_format.size(),
               DIRTY_REGS_POS(vertex_attrib_format));
@@ -171,7 +171,7 @@ void Maxwell3D::InitDirtySettings() {
     // State
 
     // Viewport
-    constexpr u32 viewport_dirty_reg = DIRTY_REGS_POS(viewport);
+    constexpr u8 viewport_dirty_reg = DIRTY_REGS_POS(viewport);
     constexpr u32 viewport_start = MAXWELL3D_REG_INDEX(viewports);
     constexpr u32 viewport_size = sizeof(regs.viewports) / sizeof(u32);
     set_block(viewport_start, viewport_size, viewport_dirty_reg);
@@ -198,7 +198,7 @@ void Maxwell3D::InitDirtySettings() {
     set_block(primitive_restart_start, primitive_restart_size, DIRTY_REGS_POS(primitive_restart));
 
     // Depth Test
-    constexpr u32 depth_test_dirty_reg = DIRTY_REGS_POS(depth_test);
+    constexpr u8 depth_test_dirty_reg = DIRTY_REGS_POS(depth_test);
     dirty_pointers[MAXWELL3D_REG_INDEX(depth_test_enable)] = depth_test_dirty_reg;
     dirty_pointers[MAXWELL3D_REG_INDEX(depth_write_enabled)] = depth_test_dirty_reg;
     dirty_pointers[MAXWELL3D_REG_INDEX(depth_test_func)] = depth_test_dirty_reg;
@@ -223,12 +223,12 @@ void Maxwell3D::InitDirtySettings() {
     dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_mask)] = stencil_test_dirty_reg;
 
     // Color Mask
-    constexpr u32 color_mask_dirty_reg = DIRTY_REGS_POS(color_mask);
+    constexpr u8 color_mask_dirty_reg = DIRTY_REGS_POS(color_mask);
     dirty_pointers[MAXWELL3D_REG_INDEX(color_mask_common)] = color_mask_dirty_reg;
     set_block(MAXWELL3D_REG_INDEX(color_mask), sizeof(regs.color_mask) / sizeof(u32),
               color_mask_dirty_reg);
     // Blend State
-    constexpr u32 blend_state_dirty_reg = DIRTY_REGS_POS(blend_state);
+    constexpr u8 blend_state_dirty_reg = DIRTY_REGS_POS(blend_state);
     set_block(MAXWELL3D_REG_INDEX(blend_color), sizeof(regs.blend_color) / sizeof(u32),
               blend_state_dirty_reg);
     dirty_pointers[MAXWELL3D_REG_INDEX(independent_blend_enable)] = blend_state_dirty_reg;
@@ -237,12 +237,12 @@ void Maxwell3D::InitDirtySettings() {
               blend_state_dirty_reg);
 
     // Scissor State
-    constexpr u32 scissor_test_dirty_reg = DIRTY_REGS_POS(scissor_test);
+    constexpr u8 scissor_test_dirty_reg = DIRTY_REGS_POS(scissor_test);
     set_block(MAXWELL3D_REG_INDEX(scissor_test), sizeof(regs.scissor_test) / sizeof(u32),
               scissor_test_dirty_reg);
 
     // Polygon Offset
-    constexpr u32 polygon_offset_dirty_reg = DIRTY_REGS_POS(polygon_offset);
+    constexpr u8 polygon_offset_dirty_reg = DIRTY_REGS_POS(polygon_offset);
     dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_fill_enable)] = polygon_offset_dirty_reg;
     dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_line_enable)] = polygon_offset_dirty_reg;
     dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_point_enable)] = polygon_offset_dirty_reg;
@@ -251,7 +251,7 @@ void Maxwell3D::InitDirtySettings() {
     dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_clamp)] = polygon_offset_dirty_reg;
 
     // Depth bounds
-    constexpr u32 depth_bounds_values_dirty_reg = DIRTY_REGS_POS(depth_bounds_values);
+    constexpr u8 depth_bounds_values_dirty_reg = DIRTY_REGS_POS(depth_bounds_values);
     dirty_pointers[MAXWELL3D_REG_INDEX(depth_bounds[0])] = depth_bounds_values_dirty_reg;
     dirty_pointers[MAXWELL3D_REG_INDEX(depth_bounds[1])] = depth_bounds_values_dirty_reg;
 }
@@ -478,7 +478,7 @@ void Maxwell3D::CallMethodFromMME(const GPU::MethodCall& method_call) {
 }
 
 void Maxwell3D::FlushMMEInlineDraw() {
-    LOG_DEBUG(HW_GPU, "called, topology={}, count={}", static_cast<u32>(regs.draw.topology.Value()),
+    LOG_TRACE(HW_GPU, "called, topology={}, count={}", static_cast<u32>(regs.draw.topology.Value()),
               regs.vertex_buffer.count);
     ASSERT_MSG(!(regs.index_array.count && regs.vertex_buffer.count), "Both indexed and direct?");
     ASSERT(mme_draw.instance_count == mme_draw.gl_end_count);
@@ -846,7 +846,8 @@ void Maxwell3D::ProcessClearBuffers() {
     rasterizer.Clear();
 }
 
-u32 Maxwell3D::AccessConstBuffer32(Regs::ShaderStage stage, u64 const_buffer, u64 offset) const {
+u32 Maxwell3D::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const {
+    ASSERT(stage != ShaderType::Compute);
     const auto& shader_stage = state.shader_stages[static_cast<std::size_t>(stage)];
     const auto& buffer = shader_stage.const_buffers[const_buffer];
     u32 result;
@@ -854,4 +855,22 @@ u32 Maxwell3D::AccessConstBuffer32(Regs::ShaderStage stage, u64 const_buffer, u6
     return result;
 }
 
+SamplerDescriptor Maxwell3D::AccessBoundSampler(ShaderType stage, u64 offset) const {
+    return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle));
+}
+
+SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_buffer,
+                                                   u64 offset) const {
+    ASSERT(stage != ShaderType::Compute);
+    const auto& shader = state.shader_stages[static_cast<std::size_t>(stage)];
+    const auto& tex_info_buffer = shader.const_buffers[const_buffer];
+    const GPUVAddr tex_info_address = tex_info_buffer.address + offset;
+
+    const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
+    const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle, offset);
+    SamplerDescriptor result = SamplerDescriptor::FromTicTexture(tex_info.tic.texture_type.Value());
+    result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
+    return result;
+}
+
 } // namespace Tegra::Engines
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index e3f1047d5..987ad77b2 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -15,6 +15,7 @@
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/math_util.h"
+#include "video_core/engines/const_buffer_engine_interface.h"
 #include "video_core/engines/const_buffer_info.h"
 #include "video_core/engines/engine_upload.h"
 #include "video_core/gpu.h"
@@ -44,7 +45,7 @@ namespace Tegra::Engines {
 #define MAXWELL3D_REG_INDEX(field_name)                                                            \
     (offsetof(Tegra::Engines::Maxwell3D::Regs, field_name) / sizeof(u32))
 
-class Maxwell3D final {
+class Maxwell3D final : public ConstBufferEngineInterface {
 public:
     explicit Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
                        MemoryManager& memory_manager);
@@ -1165,6 +1166,8 @@ public:
 
     struct DirtyRegs {
         static constexpr std::size_t NUM_REGS = 256;
+        static_assert(NUM_REGS - 1 <= std::numeric_limits<u8>::max());
+
         union {
             struct {
                 bool null_dirty;
@@ -1257,7 +1260,16 @@ public:
     /// Returns the texture information for a specific texture in a specific shader stage.
     Texture::FullTextureInfo GetStageTexture(Regs::ShaderStage stage, std::size_t offset) const;
 
-    u32 AccessConstBuffer32(Regs::ShaderStage stage, u64 const_buffer, u64 offset) const;
+    u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
+
+    SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
+
+    SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
+                                            u64 offset) const override;
+
+    u32 GetBoundBuffer() const override {
+        return regs.tex_cb_index;
+    }
 
     /// Memory for macro code - it's undetermined how big this is, however 1MB is much larger than
     /// we've seen used.
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 7a6355ce2..d3d05a866 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -574,7 +574,7 @@ enum class ShuffleOperation : u64 {
 };
 
 union Instruction {
-    Instruction& operator=(const Instruction& instr) {
+    constexpr Instruction& operator=(const Instruction& instr) {
         value = instr.value;
         return *this;
     }
@@ -1760,22 +1760,22 @@ public:
 
     class Matcher {
     public:
-        Matcher(const char* const name, u16 mask, u16 expected, OpCode::Id id, OpCode::Type type)
+        constexpr Matcher(const char* const name, u16 mask, u16 expected, Id id, Type type)
             : name{name}, mask{mask}, expected{expected}, id{id}, type{type} {}
 
-        const char* GetName() const {
+        constexpr const char* GetName() const {
             return name;
         }
 
-        u16 GetMask() const {
+        constexpr u16 GetMask() const {
             return mask;
         }
 
-        Id GetId() const {
+        constexpr Id GetId() const {
             return id;
         }
 
-        Type GetType() const {
+        constexpr Type GetType() const {
             return type;
         }
 
@@ -1784,7 +1784,7 @@ public:
          * @param instruction The instruction to test
          * @returns true if the given instruction matches.
          */
-        bool Matches(u16 instruction) const {
+        constexpr bool Matches(u16 instruction) const {
             return (instruction & mask) == expected;
         }
 
@@ -1818,7 +1818,7 @@ private:
          * A '0' in a bitstring indicates that a zero must be present at that bit position.
          * A '1' in a bitstring indicates that a one must be present at that bit position.
          */
-        static auto GetMaskAndExpect(const char* const bitstring) {
+        static constexpr auto GetMaskAndExpect(const char* const bitstring) {
             u16 mask = 0, expect = 0;
             for (std::size_t i = 0; i < opcode_bitsize; i++) {
                 const std::size_t bit_position = opcode_bitsize - i - 1;
@@ -1835,15 +1835,15 @@ private:
                     break;
                 }
             }
-            return std::make_tuple(mask, expect);
+            return std::make_pair(mask, expect);
         }
 
     public:
         /// Creates a matcher that can match and parse instructions based on bitstring.
-        static auto GetMatcher(const char* const bitstring, OpCode::Id op, OpCode::Type type,
-                               const char* const name) {
-            const auto mask_expect = GetMaskAndExpect(bitstring);
-            return Matcher(name, std::get<0>(mask_expect), std::get<1>(mask_expect), op, type);
+        static constexpr auto GetMatcher(const char* const bitstring, Id op, Type type,
+                                         const char* const name) {
+            const auto [mask, expected] = GetMaskAndExpect(bitstring);
+            return Matcher(name, mask, expected, op, type);
         }
     };
 
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 76cfe8107..095660115 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include "common/assert.h"
+#include "common/microprofile.h"
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "core/memory.h"
@@ -17,6 +18,8 @@
 
 namespace Tegra {
 
+MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
+
 GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async)
     : system{system}, renderer{renderer}, is_async{is_async} {
     auto& rasterizer{renderer.Rasterizer()};
@@ -63,6 +66,16 @@ const DmaPusher& GPU::DmaPusher() const {
     return *dma_pusher;
 }
 
+void GPU::WaitFence(u32 syncpoint_id, u32 value) const {
+    // Synced GPU, is always in sync
+    if (!is_async) {
+        return;
+    }
+    MICROPROFILE_SCOPE(GPU_wait);
+    while (syncpoints[syncpoint_id].load(std::memory_order_relaxed) < value) {
+    }
+}
+
 void GPU::IncrementSyncPoint(const u32 syncpoint_id) {
     syncpoints[syncpoint_id]++;
     std::lock_guard lock{sync_mutex};
@@ -326,7 +339,7 @@ void GPU::ProcessSemaphoreTriggerMethod() {
         block.sequence = regs.semaphore_sequence;
         // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of
         // CoreTiming
-        block.timestamp = Core::System::GetInstance().CoreTiming().GetTicks();
+        block.timestamp = system.CoreTiming().GetTicks();
         memory_manager->WriteBlock(regs.semaphore_address.SemaphoreAddress(), &block,
                                    sizeof(block));
     } else {
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 29fa8e95b..dbca19f35 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -177,6 +177,12 @@ public:
     /// Returns a reference to the GPU DMA pusher.
     Tegra::DmaPusher& DmaPusher();
 
+    // Waits for the GPU to finish working
+    virtual void WaitIdle() const = 0;
+
+    /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame.
+    void WaitFence(u32 syncpoint_id, u32 value) const;
+
     void IncrementSyncPoint(u32 syncpoint_id);
 
     u32 GetSyncpointValue(u32 syncpoint_id) const;
diff --git a/src/video_core/gpu_asynch.cpp b/src/video_core/gpu_asynch.cpp
index f2a3a390e..04222d060 100644
--- a/src/video_core/gpu_asynch.cpp
+++ b/src/video_core/gpu_asynch.cpp
@@ -44,4 +44,8 @@ void GPUAsynch::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) con
     interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value);
 }
 
+void GPUAsynch::WaitIdle() const {
+    gpu_thread.WaitIdle();
+}
+
 } // namespace VideoCommon
diff --git a/src/video_core/gpu_asynch.h b/src/video_core/gpu_asynch.h
index a12f9bac4..1241ade1d 100644
--- a/src/video_core/gpu_asynch.h
+++ b/src/video_core/gpu_asynch.h
@@ -25,6 +25,7 @@ public:
     void FlushRegion(CacheAddr addr, u64 size) override;
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+    void WaitIdle() const override;
 
 protected:
     void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const override;
diff --git a/src/video_core/gpu_synch.h b/src/video_core/gpu_synch.h
index 5eb1c461c..c71baee89 100644
--- a/src/video_core/gpu_synch.h
+++ b/src/video_core/gpu_synch.h
@@ -24,6 +24,7 @@ public:
     void FlushRegion(CacheAddr addr, u64 size) override;
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+    void WaitIdle() const override {}
 
 protected:
     void TriggerCpuInterrupt([[maybe_unused]] u32 syncpoint_id,
diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index 5f039e4fd..758a37f14 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -5,8 +5,6 @@
 #include "common/assert.h"
 #include "common/microprofile.h"
 #include "core/core.h"
-#include "core/core_timing.h"
-#include "core/core_timing_util.h"
 #include "core/frontend/scope_acquire_window_context.h"
 #include "video_core/dma_pusher.h"
 #include "video_core/gpu.h"
@@ -68,14 +66,10 @@ ThreadManager::~ThreadManager() {
 
 void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_pusher) {
     thread = std::thread{RunThread, std::ref(renderer), std::ref(dma_pusher), std::ref(state)};
-    synchronization_event = system.CoreTiming().RegisterEvent(
-        "GPUThreadSynch", [this](u64 fence, s64) { state.WaitForSynchronization(fence); });
 }
 
 void ThreadManager::SubmitList(Tegra::CommandList&& entries) {
-    const u64 fence{PushCommand(SubmitListCommand(std::move(entries)))};
-    const s64 synchronization_ticks{Core::Timing::usToCycles(std::chrono::microseconds{9000})};
-    system.CoreTiming().ScheduleEvent(synchronization_ticks, synchronization_event, fence);
+    PushCommand(SubmitListCommand(std::move(entries)));
 }
 
 void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
@@ -96,16 +90,15 @@ void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     InvalidateRegion(addr, size);
 }
 
+void ThreadManager::WaitIdle() const {
+    while (state.last_fence > state.signaled_fence.load(std::memory_order_relaxed)) {
+    }
+}
+
 u64 ThreadManager::PushCommand(CommandData&& command_data) {
     const u64 fence{++state.last_fence};
     state.queue.Push(CommandDataContainer(std::move(command_data), fence));
     return fence;
 }
 
-MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
-void SynchState::WaitForSynchronization(u64 fence) {
-    while (signaled_fence.load() < fence)
-        ;
-}
-
 } // namespace VideoCommon::GPUThread
diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h
index 3ae0ec9f3..08dc96bb3 100644
--- a/src/video_core/gpu_thread.h
+++ b/src/video_core/gpu_thread.h
@@ -21,9 +21,6 @@ class DmaPusher;
 
 namespace Core {
 class System;
-namespace Timing {
-struct EventType;
-} // namespace Timing
 } // namespace Core
 
 namespace VideoCommon::GPUThread {
@@ -89,8 +86,6 @@ struct CommandDataContainer {
 struct SynchState final {
     std::atomic_bool is_running{true};
 
-    void WaitForSynchronization(u64 fence);
-
     using CommandQueue = Common::SPSCQueue<CommandDataContainer>;
     CommandQueue queue;
     u64 last_fence{};
@@ -121,6 +116,9 @@ public:
     /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size);
 
+    // Wait until the gpu thread is idle.
+    void WaitIdle() const;
+
 private:
     /// Pushes a command to be executed by the GPU thread
     u64 PushCommand(CommandData&& command_data);
@@ -128,7 +126,6 @@ private:
 private:
     SynchState state;
     Core::System& system;
-    Core::Timing::EventType* synchronization_event{};
     std::thread thread;
     std::thread::id thread_id;
 };
diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro_interpreter.cpp
index dbaeac6db..42031d80a 100644
--- a/src/video_core/macro_interpreter.cpp
+++ b/src/video_core/macro_interpreter.cpp
@@ -11,6 +11,77 @@
 MICROPROFILE_DEFINE(MacroInterp, "GPU", "Execute macro interpreter", MP_RGB(128, 128, 192));
 
 namespace Tegra {
+namespace {
+enum class Operation : u32 {
+    ALU = 0,
+    AddImmediate = 1,
+    ExtractInsert = 2,
+    ExtractShiftLeftImmediate = 3,
+    ExtractShiftLeftRegister = 4,
+    Read = 5,
+    Unused = 6, // This operation doesn't seem to be a valid encoding.
+    Branch = 7,
+};
+} // Anonymous namespace
+
+enum class MacroInterpreter::ALUOperation : u32 {
+    Add = 0,
+    AddWithCarry = 1,
+    Subtract = 2,
+    SubtractWithBorrow = 3,
+    // Operations 4-7 don't seem to be valid encodings.
+    Xor = 8,
+    Or = 9,
+    And = 10,
+    AndNot = 11,
+    Nand = 12
+};
+
+enum class MacroInterpreter::ResultOperation : u32 {
+    IgnoreAndFetch = 0,
+    Move = 1,
+    MoveAndSetMethod = 2,
+    FetchAndSend = 3,
+    MoveAndSend = 4,
+    FetchAndSetMethod = 5,
+    MoveAndSetMethodFetchAndSend = 6,
+    MoveAndSetMethodSend = 7
+};
+
+enum class MacroInterpreter::BranchCondition : u32 {
+    Zero = 0,
+    NotZero = 1,
+};
+
+union MacroInterpreter::Opcode {
+    u32 raw;
+    BitField<0, 3, Operation> operation;
+    BitField<4, 3, ResultOperation> result_operation;
+    BitField<4, 1, BranchCondition> branch_condition;
+    // If set on a branch, then the branch doesn't have a delay slot.
+    BitField<5, 1, u32> branch_annul;
+    BitField<7, 1, u32> is_exit;
+    BitField<8, 3, u32> dst;
+    BitField<11, 3, u32> src_a;
+    BitField<14, 3, u32> src_b;
+    // The signed immediate overlaps the second source operand and the alu operation.
+    BitField<14, 18, s32> immediate;
+
+    BitField<17, 5, ALUOperation> alu_operation;
+
+    // Bitfield instructions data
+    BitField<17, 5, u32> bf_src_bit;
+    BitField<22, 5, u32> bf_size;
+    BitField<27, 5, u32> bf_dst_bit;
+
+    u32 GetBitfieldMask() const {
+        return (1 << bf_size) - 1;
+    }
+
+    s32 GetBranchTarget() const {
+        return static_cast<s32>(immediate * sizeof(u32));
+    }
+};
 
 MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
 
diff --git a/src/video_core/macro_interpreter.h b/src/video_core/macro_interpreter.h
index 76b6a895b..631146d89 100644
--- a/src/video_core/macro_interpreter.h
+++ b/src/video_core/macro_interpreter.h
@@ -6,7 +6,6 @@
 
 #include <array>
 #include <optional>
-#include <vector>
 
 #include "common/bit_field.h"
 #include "common/common_types.h"
@@ -28,75 +27,11 @@ public:
     void Execute(u32 offset, std::size_t num_parameters, const u32* parameters);
 
 private:
-    enum class Operation : u32 {
-        ALU = 0,
-        AddImmediate = 1,
-        ExtractInsert = 2,
-        ExtractShiftLeftImmediate = 3,
-        ExtractShiftLeftRegister = 4,
-        Read = 5,
-        Unused = 6, // This operation doesn't seem to be a valid encoding.
-        Branch = 7,
-    };
-
-    enum class ALUOperation : u32 {
-        Add = 0,
-        AddWithCarry = 1,
-        Subtract = 2,
-        SubtractWithBorrow = 3,
-        // Operations 4-7 don't seem to be valid encodings.
-        Xor = 8,
-        Or = 9,
-        And = 10,
-        AndNot = 11,
-        Nand = 12
-    };
-
-    enum class ResultOperation : u32 {
-        IgnoreAndFetch = 0,
-        Move = 1,
-        MoveAndSetMethod = 2,
-        FetchAndSend = 3,
-        MoveAndSend = 4,
-        FetchAndSetMethod = 5,
-        MoveAndSetMethodFetchAndSend = 6,
-        MoveAndSetMethodSend = 7
-    };
+    enum class ALUOperation : u32;
+    enum class BranchCondition : u32;
+    enum class ResultOperation : u32;
 
-    enum class BranchCondition : u32 {
-        Zero = 0,
-        NotZero = 1,
-    };
-
-    union Opcode {
-        u32 raw;
-        BitField<0, 3, Operation> operation;
-        BitField<4, 3, ResultOperation> result_operation;
-        BitField<4, 1, BranchCondition> branch_condition;
-        BitField<5, 1, u32>
-            branch_annul; // If set on a branch, then the branch doesn't have a delay slot.
-        BitField<7, 1, u32> is_exit;
-        BitField<8, 3, u32> dst;
-        BitField<11, 3, u32> src_a;
-        BitField<14, 3, u32> src_b;
-        // The signed immediate overlaps the second source operand and the alu operation.
-        BitField<14, 18, s32> immediate;
-
-        BitField<17, 5, ALUOperation> alu_operation;
-
-        // Bitfield instructions data
-        BitField<17, 5, u32> bf_src_bit;
-        BitField<22, 5, u32> bf_size;
-        BitField<27, 5, u32> bf_dst_bit;
-
-        u32 GetBitfieldMask() const {
-            return (1 << bf_size) - 1;
-        }
-
-        s32 GetBranchTarget() const {
-            return static_cast<s32>(immediate * sizeof(u32));
-        }
-    };
+    union Opcode;
 
     union MethodAddress {
         u32 raw;
@@ -149,9 +84,10 @@ private:
 
     Engines::Maxwell3D& maxwell3d;
 
-    u32 pc; ///< Current program counter
-    std::optional<u32>
-        delayed_pc; ///< Program counter to execute at after the delay slot is executed.
+    /// Current program counter
+    u32 pc;
+    /// Program counter to execute at after the delay slot is executed.
+    std::optional<u32> delayed_pc;
 
     static constexpr std::size_t NumMacroRegisters = 8;
 
diff --git a/src/video_core/morton.cpp b/src/video_core/morton.cpp
index ab71870ab..fe5f08ace 100644
--- a/src/video_core/morton.cpp
+++ b/src/video_core/morton.cpp
@@ -93,6 +93,7 @@ static constexpr ConversionArray morton_to_linear_fns = {
     MortonCopy<true, PixelFormat::DXT23_SRGB>,
     MortonCopy<true, PixelFormat::DXT45_SRGB>,
     MortonCopy<true, PixelFormat::BC7U_SRGB>,
+    MortonCopy<true, PixelFormat::R4G4B4A4U>,
     MortonCopy<true, PixelFormat::ASTC_2D_4X4_SRGB>,
     MortonCopy<true, PixelFormat::ASTC_2D_8X8_SRGB>,
     MortonCopy<true, PixelFormat::ASTC_2D_8X5_SRGB>,
@@ -101,6 +102,16 @@ static constexpr ConversionArray morton_to_linear_fns = {
     MortonCopy<true, PixelFormat::ASTC_2D_5X5_SRGB>,
     MortonCopy<true, PixelFormat::ASTC_2D_10X8>,
     MortonCopy<true, PixelFormat::ASTC_2D_10X8_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_6X6>,
+    MortonCopy<true, PixelFormat::ASTC_2D_6X6_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_10X10>,
+    MortonCopy<true, PixelFormat::ASTC_2D_10X10_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_12X12>,
+    MortonCopy<true, PixelFormat::ASTC_2D_12X12_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_8X6>,
+    MortonCopy<true, PixelFormat::ASTC_2D_8X6_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_6X5>,
+    MortonCopy<true, PixelFormat::ASTC_2D_6X5_SRGB>,
     MortonCopy<true, PixelFormat::Z32F>,
     MortonCopy<true, PixelFormat::Z16>,
     MortonCopy<true, PixelFormat::Z24S8>,
@@ -162,6 +173,17 @@ static constexpr ConversionArray linear_to_morton_fns = {
     MortonCopy<false, PixelFormat::DXT23_SRGB>,
     MortonCopy<false, PixelFormat::DXT45_SRGB>,
     MortonCopy<false, PixelFormat::BC7U_SRGB>,
+    MortonCopy<false, PixelFormat::R4G4B4A4U>,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
     nullptr,
     nullptr,
     nullptr,
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index a85f730a8..9431d64ac 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -348,6 +348,7 @@ static constexpr auto RangeFromInterval(Map& map, const Interval& interval) {
 }
 
 void RasterizerOpenGL::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {
+    std::lock_guard lock{pages_mutex};
     const u64 page_start{addr >> Memory::PAGE_BITS};
     const u64 page_end{(addr + size + Memory::PAGE_SIZE - 1) >> Memory::PAGE_BITS};
 
@@ -974,7 +975,8 @@ TextureBufferUsage RasterizerOpenGL::SetupDrawTextures(Maxwell::ShaderStage stag
             }
             const auto cbuf = entry.GetBindlessCBuf();
             Tegra::Texture::TextureHandle tex_handle;
-            tex_handle.raw = maxwell3d.AccessConstBuffer32(stage, cbuf.first, cbuf.second);
+            Tegra::Engines::ShaderType shader_type = static_cast<Tegra::Engines::ShaderType>(stage);
+            tex_handle.raw = maxwell3d.AccessConstBuffer32(shader_type, cbuf.first, cbuf.second);
             return maxwell3d.GetTextureInfo(tex_handle, entry.GetOffset());
         }();
 
@@ -1004,7 +1006,8 @@ TextureBufferUsage RasterizerOpenGL::SetupComputeTextures(const Shader& kernel)
             }
             const auto cbuf = entry.GetBindlessCBuf();
             Tegra::Texture::TextureHandle tex_handle;
-            tex_handle.raw = compute.AccessConstBuffer32(cbuf.first, cbuf.second);
+            tex_handle.raw = compute.AccessConstBuffer32(Tegra::Engines::ShaderType::Compute,
+                                                         cbuf.first, cbuf.second);
             return compute.GetTextureInfo(tex_handle, entry.GetOffset());
         }();
 
@@ -1049,7 +1052,8 @@ void RasterizerOpenGL::SetupComputeImages(const Shader& shader) {
             }
             const auto cbuf = entry.GetBindlessCBuf();
             Tegra::Texture::TextureHandle tex_handle;
-            tex_handle.raw = compute.AccessConstBuffer32(cbuf.first, cbuf.second);
+            tex_handle.raw = compute.AccessConstBuffer32(Tegra::Engines::ShaderType::Compute,
+                                                         cbuf.first, cbuf.second);
             return compute.GetTextureInfo(tex_handle, entry.GetOffset()).tic;
         }();
         SetupImage(bindpoint, tic, entry);
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index 9c10ebda3..c24a02d71 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -9,6 +9,7 @@
 #include <cstddef>
 #include <map>
 #include <memory>
+#include <mutex>
 #include <optional>
 #include <tuple>
 #include <utility>
@@ -230,6 +231,8 @@ private:
 
     using CachedPageMap = boost::icl::interval_map<u64, int>;
     CachedPageMap cached_pages;
+
+    std::mutex pages_mutex;
 };
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 42ca3b1bd..f1b89165d 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -3,13 +3,16 @@
 // Refer to the license.txt file included.
 
 #include <mutex>
+#include <optional>
+#include <string>
 #include <thread>
+#include <unordered_set>
 #include <boost/functional/hash.hpp>
 #include "common/assert.h"
-#include "common/hash.h"
 #include "common/scope_exit.h"
 #include "core/core.h"
 #include "core/frontend/emu_window.h"
+#include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
@@ -21,18 +24,20 @@
 
 namespace OpenGL {
 
+using Tegra::Engines::ShaderType;
+using VideoCommon::Shader::ConstBufferLocker;
 using VideoCommon::Shader::ProgramCode;
+using VideoCommon::Shader::ShaderIR;
+
+namespace {
 
 // One UBO is always reserved for emulation values on staged shaders
 constexpr u32 STAGE_RESERVED_UBOS = 1;
 
-struct UnspecializedShader {
-    std::string code;
-    GLShader::ShaderEntries entries;
-    ProgramType program_type;
-};
+constexpr u32 STAGE_MAIN_OFFSET = 10;
+constexpr u32 KERNEL_MAIN_OFFSET = 0;
 
-namespace {
+constexpr VideoCommon::Shader::CompilerSettings COMPILER_SETTINGS{};
 
 /// Gets the address for the specified shader stage program
 GPUVAddr GetShaderAddress(Core::System& system, Maxwell::ShaderProgram program) {
@@ -41,6 +46,39 @@ GPUVAddr GetShaderAddress(Core::System& system, Maxwell::ShaderProgram program)
     return gpu.regs.code_address.CodeAddress() + shader_config.offset;
 }
 
+/// Gets if the current instruction offset is a scheduler instruction
+constexpr bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
+    // Sched instructions appear once every 4 instructions.
+    constexpr std::size_t SchedPeriod = 4;
+    const std::size_t absolute_offset = offset - main_offset;
+    return (absolute_offset % SchedPeriod) == 0;
+}
+
+/// Calculates the size of a program stream
+std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
+    constexpr std::size_t start_offset = 10;
+    // This is the encoded version of BRA that jumps to itself. All Nvidia
+    // shaders end with one.
+    constexpr u64 self_jumping_branch = 0xE2400FFFFF07000FULL;
+    constexpr u64 mask = 0xFFFFFFFFFF7FFFFFULL;
+    std::size_t offset = start_offset;
+    while (offset < program.size()) {
+        const u64 instruction = program[offset];
+        if (!IsSchedInstruction(offset, start_offset)) {
+            if ((instruction & mask) == self_jumping_branch) {
+                // End on Maxwell's "nop" instruction
+                break;
+            }
+            if (instruction == 0) {
+                break;
+            }
+        }
+        offset++;
+    }
+    // The last instruction is included in the program size
+    return std::min(offset + 1, program.size());
+}
+
 /// Gets the shader program code from memory for the specified address
 ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr gpu_addr,
                           const u8* host_ptr) {
@@ -51,6 +89,7 @@ ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr g
     });
     memory_manager.ReadBlockUnsafe(gpu_addr, program_code.data(),
                                    program_code.size() * sizeof(u64));
+    program_code.resize(CalculateProgramSize(program_code));
     return program_code;
 }
 
@@ -71,14 +110,6 @@ constexpr GLenum GetShaderType(ProgramType program_type) {
     }
 }
 
-/// Gets if the current instruction offset is a scheduler instruction
-constexpr bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
-    // Sched instructions appear once every 4 instructions.
-    constexpr std::size_t SchedPeriod = 4;
-    const std::size_t absolute_offset = offset - main_offset;
-    return (absolute_offset % SchedPeriod) == 0;
-}
-
 /// Describes primitive behavior on geometry shaders
 constexpr std::tuple<const char*, const char*, u32> GetPrimitiveDescription(GLenum primitive_mode) {
     switch (primitive_mode) {
@@ -121,110 +152,142 @@ ProgramType GetProgramType(Maxwell::ShaderProgram program) {
     return {};
 }
 
-/// Calculates the size of a program stream
-std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
-    constexpr std::size_t start_offset = 10;
-    // This is the encoded version of BRA that jumps to itself. All Nvidia
-    // shaders end with one.
-    constexpr u64 self_jumping_branch = 0xE2400FFFFF07000FULL;
-    constexpr u64 mask = 0xFFFFFFFFFF7FFFFFULL;
-    std::size_t offset = start_offset;
-    std::size_t size = start_offset * sizeof(u64);
-    while (offset < program.size()) {
-        const u64 instruction = program[offset];
-        if (!IsSchedInstruction(offset, start_offset)) {
-            if ((instruction & mask) == self_jumping_branch) {
-                // End on Maxwell's "nop" instruction
-                break;
-            }
-            if (instruction == 0) {
-                break;
-            }
-        }
-        size += sizeof(u64);
-        offset++;
-    }
-    // The last instruction is included in the program size
-    return std::min(size + sizeof(u64), program.size() * sizeof(u64));
-}
-
 /// Hashes one (or two) program streams
 u64 GetUniqueIdentifier(ProgramType program_type, const ProgramCode& code,
-                        const ProgramCode& code_b, std::size_t size_a = 0, std::size_t size_b = 0) {
-    if (size_a == 0) {
-        size_a = CalculateProgramSize(code);
-    }
-    u64 unique_identifier = Common::CityHash64(reinterpret_cast<const char*>(code.data()), size_a);
-    if (program_type != ProgramType::VertexA) {
-        return unique_identifier;
-    }
-    // VertexA programs include two programs
-
-    std::size_t seed = 0;
-    boost::hash_combine(seed, unique_identifier);
-
-    if (size_b == 0) {
-        size_b = CalculateProgramSize(code_b);
+                        const ProgramCode& code_b) {
+    u64 unique_identifier = boost::hash_value(code);
+    if (program_type == ProgramType::VertexA) {
+        // VertexA programs include two programs
+        boost::hash_combine(unique_identifier, boost::hash_value(code_b));
     }
-    const u64 identifier_b =
-        Common::CityHash64(reinterpret_cast<const char*>(code_b.data()), size_b);
-    boost::hash_combine(seed, identifier_b);
-    return static_cast<u64>(seed);
+    return unique_identifier;
 }
 
 /// Creates an unspecialized program from code streams
-GLShader::ProgramResult CreateProgram(const Device& device, ProgramType program_type,
-                                      ProgramCode program_code, ProgramCode program_code_b) {
-    GLShader::ShaderSetup setup(program_code);
-    setup.program.size_a = CalculateProgramSize(program_code);
-    setup.program.size_b = 0;
-    if (program_type == ProgramType::VertexA) {
-        // VertexB is always enabled, so when VertexA is enabled, we have two vertex shaders.
-        // Conventional HW does not support this, so we combine VertexA and VertexB into one
-        // stage here.
-        setup.SetProgramB(program_code_b);
-        setup.program.size_b = CalculateProgramSize(program_code_b);
-    }
-    setup.program.unique_identifier = GetUniqueIdentifier(
-        program_type, program_code, program_code_b, setup.program.size_a, setup.program.size_b);
-
+std::string GenerateGLSL(const Device& device, ProgramType program_type, const ShaderIR& ir,
+                         const std::optional<ShaderIR>& ir_b) {
     switch (program_type) {
     case ProgramType::VertexA:
     case ProgramType::VertexB:
-        return GLShader::GenerateVertexShader(device, setup);
+        return GLShader::GenerateVertexShader(device, ir, ir_b ? &*ir_b : nullptr);
     case ProgramType::Geometry:
-        return GLShader::GenerateGeometryShader(device, setup);
+        return GLShader::GenerateGeometryShader(device, ir);
     case ProgramType::Fragment:
-        return GLShader::GenerateFragmentShader(device, setup);
+        return GLShader::GenerateFragmentShader(device, ir);
     case ProgramType::Compute:
-        return GLShader::GenerateComputeShader(device, setup);
+        return GLShader::GenerateComputeShader(device, ir);
     default:
         UNIMPLEMENTED_MSG("Unimplemented program_type={}", static_cast<u32>(program_type));
         return {};
     }
 }
 
-CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEntries& entries,
-                               ProgramType program_type, const ProgramVariant& variant,
-                               bool hint_retrievable = false) {
+constexpr const char* GetProgramTypeName(ProgramType program_type) {
+    switch (program_type) {
+    case ProgramType::VertexA:
+    case ProgramType::VertexB:
+        return "VS";
+    case ProgramType::TessellationControl:
+        return "TCS";
+    case ProgramType::TessellationEval:
+        return "TES";
+    case ProgramType::Geometry:
+        return "GS";
+    case ProgramType::Fragment:
+        return "FS";
+    case ProgramType::Compute:
+        return "CS";
+    }
+    return "UNK";
+}
+
+Tegra::Engines::ShaderType GetEnginesShaderType(ProgramType program_type) {
+    switch (program_type) {
+    case ProgramType::VertexA:
+    case ProgramType::VertexB:
+        return Tegra::Engines::ShaderType::Vertex;
+    case ProgramType::TessellationControl:
+        return Tegra::Engines::ShaderType::TesselationControl;
+    case ProgramType::TessellationEval:
+        return Tegra::Engines::ShaderType::TesselationEval;
+    case ProgramType::Geometry:
+        return Tegra::Engines::ShaderType::Geometry;
+    case ProgramType::Fragment:
+        return Tegra::Engines::ShaderType::Fragment;
+    case ProgramType::Compute:
+        return Tegra::Engines::ShaderType::Compute;
+    }
+    UNREACHABLE();
+    return {};
+}
+
+std::string GetShaderId(u64 unique_identifier, ProgramType program_type) {
+    return fmt::format("{}{:016X}", GetProgramTypeName(program_type), unique_identifier);
+}
+
+Tegra::Engines::ConstBufferEngineInterface& GetConstBufferEngineInterface(
+    Core::System& system, ProgramType program_type) {
+    if (program_type == ProgramType::Compute) {
+        return system.GPU().KeplerCompute();
+    } else {
+        return system.GPU().Maxwell3D();
+    }
+}
+
+std::unique_ptr<ConstBufferLocker> MakeLocker(Core::System& system, ProgramType program_type) {
+    return std::make_unique<ConstBufferLocker>(GetEnginesShaderType(program_type),
+                                               GetConstBufferEngineInterface(system, program_type));
+}
+
+void FillLocker(ConstBufferLocker& locker, const ShaderDiskCacheUsage& usage) {
+    for (const auto& key : usage.keys) {
+        const auto [buffer, offset] = key.first;
+        locker.InsertKey(buffer, offset, key.second);
+    }
+    for (const auto& [offset, sampler] : usage.bound_samplers) {
+        locker.InsertBoundSampler(offset, sampler);
+    }
+    for (const auto& [key, sampler] : usage.bindless_samplers) {
+        const auto [buffer, offset] = key;
+        locker.InsertBindlessSampler(buffer, offset, sampler);
+    }
+}
+
+CachedProgram BuildShader(const Device& device, u64 unique_identifier, ProgramType program_type,
+                          const ProgramCode& program_code, const ProgramCode& program_code_b,
+                          const ProgramVariant& variant, ConstBufferLocker& locker,
+                          bool hint_retrievable = false) {
+    LOG_INFO(Render_OpenGL, "called. {}", GetShaderId(unique_identifier, program_type));
+
+    const bool is_compute = program_type == ProgramType::Compute;
+    const u32 main_offset = is_compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET;
+    const ShaderIR ir(program_code, main_offset, COMPILER_SETTINGS, locker);
+    std::optional<ShaderIR> ir_b;
+    if (!program_code_b.empty()) {
+        ir_b.emplace(program_code_b, main_offset, COMPILER_SETTINGS, locker);
+    }
+    const auto entries = GLShader::GetEntries(ir);
+
     auto base_bindings{variant.base_bindings};
     const auto primitive_mode{variant.primitive_mode};
     const auto texture_buffer_usage{variant.texture_buffer_usage};
 
-    std::string source = R"(#version 430 core
+    std::string source = fmt::format(R"(// {}
+#version 430 core
 #extension GL_ARB_separate_shader_objects : enable
 #extension GL_ARB_shader_viewport_layer_array : enable
 #extension GL_EXT_shader_image_load_formatted : enable
 #extension GL_NV_gpu_shader5 : enable
 #extension GL_NV_shader_thread_group : enable
 #extension GL_NV_shader_thread_shuffle : enable
-)";
-    if (program_type == ProgramType::Compute) {
+)",
+                                     GetShaderId(unique_identifier, program_type));
+    if (is_compute) {
         source += "#extension GL_ARB_compute_variable_group_size : require\n";
     }
     source += '\n';
 
-    if (program_type != ProgramType::Compute) {
+    if (!is_compute) {
         source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
     }
 
@@ -268,7 +331,7 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
     }
 
     source += '\n';
-    source += code;
+    source += GenerateGLSL(device, program_type, ir, ir_b);
 
     OGLShader shader;
     shader.Create(source.c_str(), GetShaderType(program_type));
@@ -278,85 +341,97 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
     return program;
 }
 
-std::set<GLenum> GetSupportedFormats() {
-    std::set<GLenum> supported_formats;
-
+std::unordered_set<GLenum> GetSupportedFormats() {
     GLint num_formats{};
     glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats);
 
     std::vector<GLint> formats(num_formats);
     glGetIntegerv(GL_PROGRAM_BINARY_FORMATS, formats.data());
 
-    for (const GLint format : formats)
+    std::unordered_set<GLenum> supported_formats;
+    for (const GLint format : formats) {
         supported_formats.insert(static_cast<GLenum>(format));
+    }
     return supported_formats;
 }
 
 } // Anonymous namespace
 
 CachedShader::CachedShader(const ShaderParameters& params, ProgramType program_type,
-                           GLShader::ProgramResult result)
-    : RasterizerCacheObject{params.host_ptr}, cpu_addr{params.cpu_addr},
-      unique_identifier{params.unique_identifier}, program_type{program_type},
-      disk_cache{params.disk_cache}, precompiled_programs{params.precompiled_programs},
-      entries{result.second}, code{std::move(result.first)}, shader_length{entries.shader_length} {}
+                           GLShader::ShaderEntries entries, ProgramCode program_code,
+                           ProgramCode program_code_b)
+    : RasterizerCacheObject{params.host_ptr}, system{params.system},
+      disk_cache{params.disk_cache}, device{params.device}, cpu_addr{params.cpu_addr},
+      unique_identifier{params.unique_identifier}, program_type{program_type}, entries{entries},
+      program_code{std::move(program_code)}, program_code_b{std::move(program_code_b)} {
+    if (!params.precompiled_variants) {
+        return;
+    }
+    for (const auto& pair : *params.precompiled_variants) {
+        auto locker = MakeLocker(system, program_type);
+        const auto& usage = pair->first;
+        FillLocker(*locker, usage);
+
+        std::unique_ptr<LockerVariant>* locker_variant = nullptr;
+        const auto it =
+            std::find_if(locker_variants.begin(), locker_variants.end(), [&](const auto& variant) {
+                return variant->locker->HasEqualKeys(*locker);
+            });
+        if (it == locker_variants.end()) {
+            locker_variant = &locker_variants.emplace_back();
+            *locker_variant = std::make_unique<LockerVariant>();
+            locker_variant->get()->locker = std::move(locker);
+        } else {
+            locker_variant = &*it;
+        }
+        locker_variant->get()->programs.emplace(usage.variant, pair->second);
+    }
+}
 
 Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
                                            Maxwell::ShaderProgram program_type,
-                                           ProgramCode&& program_code,
-                                           ProgramCode&& program_code_b) {
-    const auto code_size{CalculateProgramSize(program_code)};
-    const auto code_size_b{CalculateProgramSize(program_code_b)};
-    auto result{
-        CreateProgram(params.device, GetProgramType(program_type), program_code, program_code_b)};
-    if (result.first.empty()) {
-        // TODO(Rodrigo): Unimplemented shader stages hit here, avoid using these for now
-        return {};
-    }
-
+                                           ProgramCode program_code, ProgramCode program_code_b) {
     params.disk_cache.SaveRaw(ShaderDiskCacheRaw(
-        params.unique_identifier, GetProgramType(program_type),
-        static_cast<u32>(code_size / sizeof(u64)), static_cast<u32>(code_size_b / sizeof(u64)),
-        std::move(program_code), std::move(program_code_b)));
-
+        params.unique_identifier, GetProgramType(program_type), program_code, program_code_b));
+
+    ConstBufferLocker locker(GetEnginesShaderType(GetProgramType(program_type)));
+    const ShaderIR ir(program_code, STAGE_MAIN_OFFSET, COMPILER_SETTINGS, locker);
+    // TODO(Rodrigo): Handle VertexA shaders
+    // std::optional<ShaderIR> ir_b;
+    // if (!program_code_b.empty()) {
+    //     ir_b.emplace(program_code_b, STAGE_MAIN_OFFSET);
+    // }
     return std::shared_ptr<CachedShader>(
-        new CachedShader(params, GetProgramType(program_type), std::move(result)));
+        new CachedShader(params, GetProgramType(program_type), GLShader::GetEntries(ir),
+                         std::move(program_code), std::move(program_code_b)));
 }
 
-Shader CachedShader::CreateStageFromCache(const ShaderParameters& params,
-                                          Maxwell::ShaderProgram program_type,
-                                          GLShader::ProgramResult result) {
-    return std::shared_ptr<CachedShader>(
-        new CachedShader(params, GetProgramType(program_type), std::move(result)));
-}
-
-Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode&& code) {
-    auto result{CreateProgram(params.device, ProgramType::Compute, code, {})};
-
-    const auto code_size{CalculateProgramSize(code)};
-    params.disk_cache.SaveRaw(ShaderDiskCacheRaw(params.unique_identifier, ProgramType::Compute,
-                                                 static_cast<u32>(code_size / sizeof(u64)), 0,
-                                                 std::move(code), {}));
+Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code) {
+    params.disk_cache.SaveRaw(
+        ShaderDiskCacheRaw(params.unique_identifier, ProgramType::Compute, code));
 
-    return std::shared_ptr<CachedShader>(
-        new CachedShader(params, ProgramType::Compute, std::move(result)));
+    ConstBufferLocker locker(Tegra::Engines::ShaderType::Compute);
+    const ShaderIR ir(code, KERNEL_MAIN_OFFSET, COMPILER_SETTINGS, locker);
+    return std::shared_ptr<CachedShader>(new CachedShader(
+        params, ProgramType::Compute, GLShader::GetEntries(ir), std::move(code), {}));
 }
 
-Shader CachedShader::CreateKernelFromCache(const ShaderParameters& params,
-                                           GLShader::ProgramResult result) {
-    return std::shared_ptr<CachedShader>(
-        new CachedShader(params, ProgramType::Compute, std::move(result)));
+Shader CachedShader::CreateFromCache(const ShaderParameters& params,
+                                     const UnspecializedShader& unspecialized) {
+    return std::shared_ptr<CachedShader>(new CachedShader(params, unspecialized.program_type,
+                                                          unspecialized.entries, unspecialized.code,
+                                                          unspecialized.code_b));
 }
 
 std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVariant& variant) {
-    const auto [entry, is_cache_miss] = programs.try_emplace(variant);
+    UpdateVariant();
+
+    const auto [entry, is_cache_miss] = curr_variant->programs.try_emplace(variant);
     auto& program = entry->second;
     if (is_cache_miss) {
-        program = TryLoadProgram(variant);
-        if (!program) {
-            program = SpecializeShader(code, entries, program_type, variant);
-            disk_cache.SaveUsage(GetUsage(variant));
-        }
+        program = BuildShader(device, unique_identifier, program_type, program_code, program_code_b,
+                              variant, *curr_variant->locker);
+        disk_cache.SaveUsage(GetUsage(variant, *curr_variant->locker));
 
         LabelGLObject(GL_PROGRAM, program->handle, cpu_addr);
     }
@@ -372,18 +447,33 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVar
     return {program->handle, base_bindings};
 }
 
-CachedProgram CachedShader::TryLoadProgram(const ProgramVariant& variant) const {
-    const auto found = precompiled_programs.find(GetUsage(variant));
-    if (found == precompiled_programs.end()) {
-        return {};
+void CachedShader::UpdateVariant() {
+    if (curr_variant && !curr_variant->locker->IsConsistent()) {
+        curr_variant = nullptr;
+    }
+    if (!curr_variant) {
+        for (auto& variant : locker_variants) {
+            if (variant->locker->IsConsistent()) {
+                curr_variant = variant.get();
+            }
+        }
+    }
+    if (!curr_variant) {
+        auto& new_variant = locker_variants.emplace_back();
+        new_variant = std::make_unique<LockerVariant>();
+        new_variant->locker = MakeLocker(system, program_type);
+        curr_variant = new_variant.get();
     }
-    return found->second;
 }
 
-ShaderDiskCacheUsage CachedShader::GetUsage(const ProgramVariant& variant) const {
+ShaderDiskCacheUsage CachedShader::GetUsage(const ProgramVariant& variant,
+                                            const ConstBufferLocker& locker) const {
     ShaderDiskCacheUsage usage;
     usage.unique_identifier = unique_identifier;
     usage.variant = variant;
+    usage.keys = locker.GetKeys();
+    usage.bound_samplers = locker.GetBoundSamplers();
+    usage.bindless_samplers = locker.GetBindlessSamplers();
     return usage;
 }
 
@@ -399,18 +489,15 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
         return;
     }
     const auto [raws, shader_usages] = *transferable;
-
-    auto [decompiled, dumps] = disk_cache.LoadPrecompiled();
-
-    const auto supported_formats{GetSupportedFormats()};
-    const auto unspecialized_shaders{
-        GenerateUnspecializedShaders(stop_loading, callback, raws, decompiled)};
-    if (stop_loading) {
+    if (!GenerateUnspecializedShaders(stop_loading, callback, raws) || stop_loading) {
         return;
     }
 
-    // Track if precompiled cache was altered during loading to know if we have to serialize the
-    // virtual precompiled cache file back to the hard drive
+    const auto dumps = disk_cache.LoadPrecompiled();
+    const auto supported_formats = GetSupportedFormats();
+
+    // Track if precompiled cache was altered during loading to know if we have to
+    // serialize the virtual precompiled cache file back to the hard drive
     bool precompiled_cache_altered = false;
 
     // Inform the frontend about shader build initialization
@@ -433,9 +520,6 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
                 return;
             }
             const auto& usage{shader_usages[i]};
-            LOG_INFO(Render_OpenGL, "Building shader {:016x} (index {} of {})",
-                     usage.unique_identifier, i, shader_usages.size());
-
             const auto& unspecialized{unspecialized_shaders.at(usage.unique_identifier)};
             const auto dump{dumps.find(usage)};
 
@@ -449,21 +533,28 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
                 }
             }
             if (!shader) {
-                shader = SpecializeShader(unspecialized.code, unspecialized.entries,
-                                          unspecialized.program_type, usage.variant, true);
+                auto locker{MakeLocker(system, unspecialized.program_type)};
+                FillLocker(*locker, usage);
+                shader = BuildShader(device, usage.unique_identifier, unspecialized.program_type,
+                                     unspecialized.code, unspecialized.code_b, usage.variant,
+                                     *locker, true);
             }
 
-            std::scoped_lock lock(mutex);
+            std::scoped_lock lock{mutex};
             if (callback) {
                 callback(VideoCore::LoadCallbackStage::Build, ++built_shaders,
                          shader_usages.size());
             }
 
             precompiled_programs.emplace(usage, std::move(shader));
+
+            // TODO(Rodrigo): Is there a better way to do this?
+            precompiled_variants[usage.unique_identifier].push_back(
+                precompiled_programs.find(usage));
         }
     };
 
-    const auto num_workers{static_cast<std::size_t>(std::thread::hardware_concurrency() + 1)};
+    const auto num_workers{static_cast<std::size_t>(std::thread::hardware_concurrency() + 1ULL)};
     const std::size_t bucket_size{shader_usages.size() / num_workers};
     std::vector<std::unique_ptr<Core::Frontend::GraphicsContext>> contexts(num_workers);
     std::vector<std::thread> threads(num_workers);
@@ -483,7 +574,6 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
     if (compilation_failed) {
         // Invalidate the precompiled cache if a shader dumped shader was rejected
         disk_cache.InvalidatePrecompiled();
-        dumps.clear();
         precompiled_cache_altered = true;
         return;
     }
@@ -491,8 +581,8 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
         return;
     }
 
-    // TODO(Rodrigo): Do state tracking for transferable shaders and do a dummy draw before
-    // precompiling them
+    // TODO(Rodrigo): Do state tracking for transferable shaders and do a dummy draw
+    // before precompiling them
 
     for (std::size_t i = 0; i < shader_usages.size(); ++i) {
         const auto& usage{shader_usages[i]};
@@ -508,9 +598,13 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
     }
 }
 
-CachedProgram ShaderCacheOpenGL::GeneratePrecompiledProgram(
-    const ShaderDiskCacheDump& dump, const std::set<GLenum>& supported_formats) {
+const PrecompiledVariants* ShaderCacheOpenGL::GetPrecompiledVariants(u64 unique_identifier) const {
+    const auto it = precompiled_variants.find(unique_identifier);
+    return it == precompiled_variants.end() ? nullptr : &it->second;
+}
 
+CachedProgram ShaderCacheOpenGL::GeneratePrecompiledProgram(
+    const ShaderDiskCacheDump& dump, const std::unordered_set<GLenum>& supported_formats) {
     if (supported_formats.find(dump.binary_format) == supported_formats.end()) {
         LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format - removing");
         return {};
@@ -532,56 +626,52 @@ CachedProgram ShaderCacheOpenGL::GeneratePrecompiledProgram(
     return shader;
 }
 
-std::unordered_map<u64, UnspecializedShader> ShaderCacheOpenGL::GenerateUnspecializedShaders(
+bool ShaderCacheOpenGL::GenerateUnspecializedShaders(
     const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback,
-    const std::vector<ShaderDiskCacheRaw>& raws,
-    const std::unordered_map<u64, ShaderDiskCacheDecompiled>& decompiled) {
-    std::unordered_map<u64, UnspecializedShader> unspecialized;
-
+    const std::vector<ShaderDiskCacheRaw>& raws) {
     if (callback) {
         callback(VideoCore::LoadCallbackStage::Decompile, 0, raws.size());
     }
 
     for (std::size_t i = 0; i < raws.size(); ++i) {
         if (stop_loading) {
-            return {};
+            return false;
         }
         const auto& raw{raws[i]};
         const u64 unique_identifier{raw.GetUniqueIdentifier()};
         const u64 calculated_hash{
             GetUniqueIdentifier(raw.GetProgramType(), raw.GetProgramCode(), raw.GetProgramCodeB())};
         if (unique_identifier != calculated_hash) {
-            LOG_ERROR(
-                Render_OpenGL,
-                "Invalid hash in entry={:016x} (obtained hash={:016x}) - removing shader cache",
-                raw.GetUniqueIdentifier(), calculated_hash);
+            LOG_ERROR(Render_OpenGL,
+                      "Invalid hash in entry={:016x} (obtained hash={:016x}) - "
+                      "removing shader cache",
+                      raw.GetUniqueIdentifier(), calculated_hash);
             disk_cache.InvalidateTransferable();
-            return {};
+            return false;
         }
 
-        GLShader::ProgramResult result;
-        if (const auto it = decompiled.find(unique_identifier); it != decompiled.end()) {
-            // If it's stored in the precompiled file, avoid decompiling it here
-            const auto& stored_decompiled{it->second};
-            result = {stored_decompiled.code, stored_decompiled.entries};
-        } else {
-            // Otherwise decompile the shader at boot and save the result to the decompiled file
-            result = CreateProgram(device, raw.GetProgramType(), raw.GetProgramCode(),
-                                   raw.GetProgramCodeB());
-            disk_cache.SaveDecompiled(unique_identifier, result.first, result.second);
-        }
-
-        precompiled_shaders.insert({unique_identifier, result});
-
-        unspecialized.insert(
-            {raw.GetUniqueIdentifier(),
-             {std::move(result.first), std::move(result.second), raw.GetProgramType()}});
+        const u32 main_offset =
+            raw.GetProgramType() == ProgramType::Compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET;
+        ConstBufferLocker locker(GetEnginesShaderType(raw.GetProgramType()));
+        const ShaderIR ir(raw.GetProgramCode(), main_offset, COMPILER_SETTINGS, locker);
+        // TODO(Rodrigo): Handle VertexA shaders
+        // std::optional<ShaderIR> ir_b;
+        // if (raw.HasProgramA()) {
+        //     ir_b.emplace(raw.GetProgramCodeB(), main_offset);
+        // }
+
+        UnspecializedShader unspecialized;
+        unspecialized.entries = GLShader::GetEntries(ir);
+        unspecialized.program_type = raw.GetProgramType();
+        unspecialized.code = raw.GetProgramCode();
+        unspecialized.code_b = raw.GetProgramCodeB();
+        unspecialized_shaders.emplace(raw.GetUniqueIdentifier(), unspecialized);
 
         if (callback) {
             callback(VideoCore::LoadCallbackStage::Decompile, i, raws.size());
         }
     }
-    return unspecialized;
+    return true;
 }
 
 Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
@@ -590,37 +680,35 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     }
 
     auto& memory_manager{system.GPU().MemoryManager()};
-    const GPUVAddr program_addr{GetShaderAddress(system, program)};
+    const GPUVAddr address{GetShaderAddress(system, program)};
 
     // Look up shader in the cache based on address
-    const auto host_ptr{memory_manager.GetPointer(program_addr)};
+    const auto host_ptr{memory_manager.GetPointer(address)};
     Shader shader{TryGet(host_ptr)};
     if (shader) {
         return last_shaders[static_cast<std::size_t>(program)] = shader;
     }
 
     // No shader found - create a new one
-    ProgramCode program_code{GetShaderCode(memory_manager, program_addr, host_ptr)};
-    ProgramCode program_code_b;
-    const bool is_program_a{program == Maxwell::ShaderProgram::VertexA};
-    if (is_program_a) {
-        const GPUVAddr program_addr_b{GetShaderAddress(system, Maxwell::ShaderProgram::VertexB)};
-        program_code_b = GetShaderCode(memory_manager, program_addr_b,
-                                       memory_manager.GetPointer(program_addr_b));
-    }
-
-    const auto unique_identifier =
-        GetUniqueIdentifier(GetProgramType(program), program_code, program_code_b);
-    const auto cpu_addr{*memory_manager.GpuToCpuAddress(program_addr)};
-    const ShaderParameters params{disk_cache, precompiled_programs, device, cpu_addr,
-                                  host_ptr,   unique_identifier};
-
-    const auto found = precompiled_shaders.find(unique_identifier);
-    if (found == precompiled_shaders.end()) {
-        shader = CachedShader::CreateStageFromMemory(params, program, std::move(program_code),
-                                                     std::move(program_code_b));
+    ProgramCode code{GetShaderCode(memory_manager, address, host_ptr)};
+    ProgramCode code_b;
+    if (program == Maxwell::ShaderProgram::VertexA) {
+        const GPUVAddr address_b{GetShaderAddress(system, Maxwell::ShaderProgram::VertexB)};
+        code_b = GetShaderCode(memory_manager, address_b, memory_manager.GetPointer(address_b));
+    }
+
+    const auto unique_identifier = GetUniqueIdentifier(GetProgramType(program), code, code_b);
+    const auto precompiled_variants = GetPrecompiledVariants(unique_identifier);
+    const auto cpu_addr{*memory_manager.GpuToCpuAddress(address)};
+    const ShaderParameters params{system,   disk_cache, precompiled_variants, device,
+                                  cpu_addr, host_ptr,   unique_identifier};
+
+    const auto found = unspecialized_shaders.find(unique_identifier);
+    if (found == unspecialized_shaders.end()) {
+        shader = CachedShader::CreateStageFromMemory(params, program, std::move(code),
+                                                     std::move(code_b));
     } else {
-        shader = CachedShader::CreateStageFromCache(params, program, found->second);
+        shader = CachedShader::CreateFromCache(params, found->second);
     }
     Register(shader);
 
@@ -638,15 +726,16 @@ Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
     // No kernel found - create a new one
     auto code{GetShaderCode(memory_manager, code_addr, host_ptr)};
     const auto unique_identifier{GetUniqueIdentifier(ProgramType::Compute, code, {})};
+    const auto precompiled_variants = GetPrecompiledVariants(unique_identifier);
     const auto cpu_addr{*memory_manager.GpuToCpuAddress(code_addr)};
-    const ShaderParameters params{disk_cache, precompiled_programs, device, cpu_addr,
-                                  host_ptr,   unique_identifier};
+    const ShaderParameters params{system,   disk_cache, precompiled_variants, device,
+                                  cpu_addr, host_ptr,   unique_identifier};
 
-    const auto found = precompiled_shaders.find(unique_identifier);
-    if (found == precompiled_shaders.end()) {
+    const auto found = unspecialized_shaders.find(unique_identifier);
+    if (found == unspecialized_shaders.end()) {
         kernel = CachedShader::CreateKernelFromMemory(params, std::move(code));
     } else {
-        kernel = CachedShader::CreateKernelFromCache(params, found->second);
+        kernel = CachedShader::CreateFromCache(params, found->second);
     }
 
     Register(kernel);
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h
index de195cc5d..6bd7c9cf1 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_cache.h
@@ -8,9 +8,10 @@
 #include <atomic>
 #include <bitset>
 #include <memory>
-#include <set>
+#include <string>
 #include <tuple>
 #include <unordered_map>
+#include <unordered_set>
 #include <vector>
 
 #include <glad/glad.h>
@@ -20,6 +21,8 @@
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
 #include "video_core/renderer_opengl/gl_shader_disk_cache.h"
+#include "video_core/shader/const_buffer_locker.h"
+#include "video_core/shader/shader_ir.h"
 
 namespace Core {
 class System;
@@ -40,11 +43,19 @@ using Shader = std::shared_ptr<CachedShader>;
 using CachedProgram = std::shared_ptr<OGLProgram>;
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 using PrecompiledPrograms = std::unordered_map<ShaderDiskCacheUsage, CachedProgram>;
-using PrecompiledShaders = std::unordered_map<u64, GLShader::ProgramResult>;
+using PrecompiledVariants = std::vector<PrecompiledPrograms::iterator>;
+
+struct UnspecializedShader {
+    GLShader::ShaderEntries entries;
+    ProgramType program_type;
+    ProgramCode code;
+    ProgramCode code_b;
+};
 
 struct ShaderParameters {
+    Core::System& system;
     ShaderDiskCacheOpenGL& disk_cache;
-    const PrecompiledPrograms& precompiled_programs;
+    const PrecompiledVariants* precompiled_variants;
     const Device& device;
     VAddr cpu_addr;
     u8* host_ptr;
@@ -55,23 +66,18 @@ class CachedShader final : public RasterizerCacheObject {
 public:
     static Shader CreateStageFromMemory(const ShaderParameters& params,
                                         Maxwell::ShaderProgram program_type,
-                                        ProgramCode&& program_code, ProgramCode&& program_code_b);
-
-    static Shader CreateStageFromCache(const ShaderParameters& params,
-                                       Maxwell::ShaderProgram program_type,
-                                       GLShader::ProgramResult result);
+                                        ProgramCode program_code, ProgramCode program_code_b);
+    static Shader CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code);
 
-    static Shader CreateKernelFromMemory(const ShaderParameters& params, ProgramCode&& code);
-
-    static Shader CreateKernelFromCache(const ShaderParameters& params,
-                                        GLShader::ProgramResult result);
+    static Shader CreateFromCache(const ShaderParameters& params,
+                                  const UnspecializedShader& unspecialized);
 
     VAddr GetCpuAddr() const override {
         return cpu_addr;
     }
 
     std::size_t GetSizeInBytes() const override {
-        return shader_length;
+        return program_code.size() * sizeof(u64);
     }
 
     /// Gets the shader entries for the shader
@@ -83,24 +89,36 @@ public:
     std::tuple<GLuint, BaseBindings> GetProgramHandle(const ProgramVariant& variant);
 
 private:
+    struct LockerVariant {
+        std::unique_ptr<VideoCommon::Shader::ConstBufferLocker> locker;
+        std::unordered_map<ProgramVariant, CachedProgram> programs;
+    };
+
     explicit CachedShader(const ShaderParameters& params, ProgramType program_type,
-                          GLShader::ProgramResult result);
+                          GLShader::ShaderEntries entries, ProgramCode program_code,
+                          ProgramCode program_code_b);
 
-    CachedProgram TryLoadProgram(const ProgramVariant& variant) const;
+    void UpdateVariant();
 
-    ShaderDiskCacheUsage GetUsage(const ProgramVariant& variant) const;
+    ShaderDiskCacheUsage GetUsage(const ProgramVariant& variant,
+                                  const VideoCommon::Shader::ConstBufferLocker& locker) const;
+
+    Core::System& system;
+    ShaderDiskCacheOpenGL& disk_cache;
+    const Device& device;
 
     VAddr cpu_addr{};
+
     u64 unique_identifier{};
     ProgramType program_type{};
-    ShaderDiskCacheOpenGL& disk_cache;
-    const PrecompiledPrograms& precompiled_programs;
 
     GLShader::ShaderEntries entries;
-    std::string code;
-    std::size_t shader_length{};
 
-    std::unordered_map<ProgramVariant, CachedProgram> programs;
+    ProgramCode program_code;
+    ProgramCode program_code_b;
+
+    LockerVariant* curr_variant = nullptr;
+    std::vector<std::unique_ptr<LockerVariant>> locker_variants;
 };
 
 class ShaderCacheOpenGL final : public RasterizerCache<Shader> {
@@ -123,21 +141,26 @@ protected:
     void FlushObjectInner(const Shader& object) override {}
 
 private:
-    std::unordered_map<u64, UnspecializedShader> GenerateUnspecializedShaders(
-        const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback,
-        const std::vector<ShaderDiskCacheRaw>& raws,
-        const std::unordered_map<u64, ShaderDiskCacheDecompiled>& decompiled);
+    bool GenerateUnspecializedShaders(const std::atomic_bool& stop_loading,
+                                      const VideoCore::DiskResourceLoadCallback& callback,
+                                      const std::vector<ShaderDiskCacheRaw>& raws);
 
     CachedProgram GeneratePrecompiledProgram(const ShaderDiskCacheDump& dump,
-                                             const std::set<GLenum>& supported_formats);
+                                             const std::unordered_set<GLenum>& supported_formats);
+
+    const PrecompiledVariants* GetPrecompiledVariants(u64 unique_identifier) const;
 
     Core::System& system;
     Core::Frontend::EmuWindow& emu_window;
     const Device& device;
+
     ShaderDiskCacheOpenGL disk_cache;
 
-    PrecompiledShaders precompiled_shaders;
     PrecompiledPrograms precompiled_programs;
+    std::unordered_map<u64, PrecompiledVariants> precompiled_variants;
+
+    std::unordered_map<u64, UnspecializedShader> unspecialized_shaders;
+
     std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
 };
 
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 6a610a3bc..030550c53 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -415,27 +415,6 @@ public:
         return code.GetResult();
     }
 
-    ShaderEntries GetShaderEntries() const {
-        ShaderEntries entries;
-        for (const auto& cbuf : ir.GetConstantBuffers()) {
-            entries.const_buffers.emplace_back(cbuf.second.GetMaxOffset(), cbuf.second.IsIndirect(),
-                                               cbuf.first);
-        }
-        for (const auto& sampler : ir.GetSamplers()) {
-            entries.samplers.emplace_back(sampler);
-        }
-        for (const auto& [offset, image] : ir.GetImages()) {
-            entries.images.emplace_back(image);
-        }
-        for (const auto& [base, usage] : ir.GetGlobalMemory()) {
-            entries.global_memory_entries.emplace_back(base.cbuf_index, base.cbuf_offset,
-                                                       usage.is_read, usage.is_written);
-        }
-        entries.clip_distances = ir.GetClipDistances();
-        entries.shader_length = ir.GetLength();
-        return entries;
-    }
-
 private:
     friend class ASTDecompiler;
     friend class ExprDecompiler;
@@ -1148,7 +1127,7 @@ private:
         for (const auto& variant : extras) {
             if (const auto argument = std::get_if<TextureArgument>(&variant)) {
                 expr += GenerateTextureArgument(*argument);
-            } else if (std::get_if<TextureAoffi>(&variant)) {
+            } else if (std::holds_alternative<TextureAoffi>(variant)) {
                 expr += GenerateTextureAoffi(meta->aoffi);
             } else {
                 UNREACHABLE();
@@ -1158,8 +1137,8 @@ private:
         return expr + ')';
     }
 
-    std::string GenerateTextureArgument(TextureArgument argument) {
-        const auto [type, operand] = argument;
+    std::string GenerateTextureArgument(const TextureArgument& argument) {
+        const auto& [type, operand] = argument;
         if (operand == nullptr) {
             return {};
         }
@@ -1235,7 +1214,7 @@ private:
 
     std::string BuildImageValues(Operation operation) {
         constexpr std::array constructors{"uint", "uvec2", "uvec3", "uvec4"};
-        const auto meta{std::get<MetaImage>(operation.GetMeta())};
+        const auto& meta{std::get<MetaImage>(operation.GetMeta())};
 
         const std::size_t values_count{meta.values.size()};
         std::string expr = fmt::format("{}(", constructors.at(values_count - 1));
@@ -1780,14 +1759,14 @@ private:
             return {"0", Type::Int};
         }
 
-        const auto meta{std::get<MetaImage>(operation.GetMeta())};
+        const auto& meta{std::get<MetaImage>(operation.GetMeta())};
         return {fmt::format("imageLoad({}, {}){}", GetImage(meta.image),
                             BuildIntegerCoordinates(operation), GetSwizzle(meta.element)),
                 Type::Uint};
     }
 
     Expression ImageStore(Operation operation) {
-        const auto meta{std::get<MetaImage>(operation.GetMeta())};
+        const auto& meta{std::get<MetaImage>(operation.GetMeta())};
         code.AddLine("imageStore({}, {}, {});", GetImage(meta.image),
                      BuildIntegerCoordinates(operation), BuildImageValues(operation));
         return {};
@@ -1795,7 +1774,7 @@ private:
 
     template <const std::string_view& opname>
     Expression AtomicImage(Operation operation) {
-        const auto meta{std::get<MetaImage>(operation.GetMeta())};
+        const auto& meta{std::get<MetaImage>(operation.GetMeta())};
         ASSERT(meta.values.size() == 1);
 
         return {fmt::format("imageAtomic{}({}, {}, {})", opname, GetImage(meta.image),
@@ -2246,7 +2225,7 @@ private:
         code.AddLine("#ifdef SAMPLER_{}_IS_BUFFER", sampler.GetIndex());
     }
 
-    std::string GetDeclarationWithSuffix(u32 index, const std::string& name) const {
+    std::string GetDeclarationWithSuffix(u32 index, std::string_view name) const {
         return fmt::format("{}_{}_{}", name, index, suffix);
     }
 
@@ -2271,17 +2250,15 @@ private:
     ShaderWriter code;
 };
 
-static constexpr std::string_view flow_var = "flow_var_";
-
 std::string GetFlowVariable(u32 i) {
-    return fmt::format("{}{}", flow_var, i);
+    return fmt::format("flow_var_{}", i);
 }
 
 class ExprDecompiler {
 public:
     explicit ExprDecompiler(GLSLDecompiler& decomp) : decomp{decomp} {}
 
-    void operator()(VideoCommon::Shader::ExprAnd& expr) {
+    void operator()(const ExprAnd& expr) {
         inner += "( ";
         std::visit(*this, *expr.operand1);
         inner += " && ";
@@ -2289,7 +2266,7 @@ public:
         inner += ')';
     }
 
-    void operator()(VideoCommon::Shader::ExprOr& expr) {
+    void operator()(const ExprOr& expr) {
         inner += "( ";
         std::visit(*this, *expr.operand1);
         inner += " || ";
@@ -2297,17 +2274,17 @@ public:
         inner += ')';
     }
 
-    void operator()(VideoCommon::Shader::ExprNot& expr) {
+    void operator()(const ExprNot& expr) {
         inner += '!';
         std::visit(*this, *expr.operand1);
     }
 
-    void operator()(VideoCommon::Shader::ExprPredicate& expr) {
+    void operator()(const ExprPredicate& expr) {
         const auto pred = static_cast<Tegra::Shader::Pred>(expr.predicate);
         inner += decomp.GetPredicate(pred);
     }
 
-    void operator()(VideoCommon::Shader::ExprCondCode& expr) {
+    void operator()(const ExprCondCode& expr) {
         const Node cc = decomp.ir.GetConditionCode(expr.cc);
         std::string target;
 
@@ -2316,10 +2293,13 @@ public:
             switch (index) {
             case Tegra::Shader::Pred::NeverExecute:
                 target = "false";
+                break;
             case Tegra::Shader::Pred::UnusedIndex:
                 target = "true";
+                break;
             default:
                 target = decomp.GetPredicate(index);
+                break;
             }
         } else if (const auto flag = std::get_if<InternalFlagNode>(&*cc)) {
             target = decomp.GetInternalFlag(flag->GetFlag());
@@ -2329,15 +2309,20 @@ public:
         inner += target;
     }
 
-    void operator()(VideoCommon::Shader::ExprVar& expr) {
+    void operator()(const ExprVar& expr) {
         inner += GetFlowVariable(expr.var_index);
     }
 
-    void operator()(VideoCommon::Shader::ExprBoolean& expr) {
+    void operator()(const ExprBoolean& expr) {
         inner += expr.value ? "true" : "false";
     }
 
-    std::string& GetResult() {
+    void operator()(VideoCommon::Shader::ExprGprEqual& expr) {
+        inner +=
+            "( ftou(" + decomp.GetRegister(expr.gpr) + ") == " + std::to_string(expr.value) + ')';
+    }
+
+    const std::string& GetResult() const {
         return inner;
     }
 
@@ -2350,7 +2335,7 @@ class ASTDecompiler {
 public:
     explicit ASTDecompiler(GLSLDecompiler& decomp) : decomp{decomp} {}
 
-    void operator()(VideoCommon::Shader::ASTProgram& ast) {
+    void operator()(const ASTProgram& ast) {
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
             Visit(current);
@@ -2358,7 +2343,7 @@ public:
         }
     }
 
-    void operator()(VideoCommon::Shader::ASTIfThen& ast) {
+    void operator()(const ASTIfThen& ast) {
         ExprDecompiler expr_parser{decomp};
         std::visit(expr_parser, *ast.condition);
         decomp.code.AddLine("if ({}) {{", expr_parser.GetResult());
@@ -2372,7 +2357,7 @@ public:
         decomp.code.AddLine("}}");
     }
 
-    void operator()(VideoCommon::Shader::ASTIfElse& ast) {
+    void operator()(const ASTIfElse& ast) {
         decomp.code.AddLine("else {{");
         decomp.code.scope++;
         ASTNode current = ast.nodes.GetFirst();
@@ -2384,29 +2369,29 @@ public:
         decomp.code.AddLine("}}");
     }
 
-    void operator()(VideoCommon::Shader::ASTBlockEncoded& ast) {
+    void operator()([[maybe_unused]] const ASTBlockEncoded& ast) {
         UNREACHABLE();
     }
 
-    void operator()(VideoCommon::Shader::ASTBlockDecoded& ast) {
+    void operator()(const ASTBlockDecoded& ast) {
         decomp.VisitBlock(ast.nodes);
     }
 
-    void operator()(VideoCommon::Shader::ASTVarSet& ast) {
+    void operator()(const ASTVarSet& ast) {
         ExprDecompiler expr_parser{decomp};
         std::visit(expr_parser, *ast.condition);
         decomp.code.AddLine("{} = {};", GetFlowVariable(ast.index), expr_parser.GetResult());
     }
 
-    void operator()(VideoCommon::Shader::ASTLabel& ast) {
+    void operator()(const ASTLabel& ast) {
         decomp.code.AddLine("// Label_{}:", ast.index);
     }
 
-    void operator()(VideoCommon::Shader::ASTGoto& ast) {
+    void operator()([[maybe_unused]] const ASTGoto& ast) {
         UNREACHABLE();
     }
 
-    void operator()(VideoCommon::Shader::ASTDoWhile& ast) {
+    void operator()(const ASTDoWhile& ast) {
         ExprDecompiler expr_parser{decomp};
         std::visit(expr_parser, *ast.condition);
         decomp.code.AddLine("do {{");
@@ -2420,7 +2405,7 @@ public:
         decomp.code.AddLine("}} while({});", expr_parser.GetResult());
     }
 
-    void operator()(VideoCommon::Shader::ASTReturn& ast) {
+    void operator()(const ASTReturn& ast) {
         const bool is_true = VideoCommon::Shader::ExprIsTrue(ast.condition);
         if (!is_true) {
             ExprDecompiler expr_parser{decomp};
@@ -2440,7 +2425,7 @@ public:
         }
     }
 
-    void operator()(VideoCommon::Shader::ASTBreak& ast) {
+    void operator()(const ASTBreak& ast) {
         const bool is_true = VideoCommon::Shader::ExprIsTrue(ast.condition);
         if (!is_true) {
             ExprDecompiler expr_parser{decomp};
@@ -2455,7 +2440,7 @@ public:
         }
     }
 
-    void Visit(VideoCommon::Shader::ASTNode& node) {
+    void Visit(const ASTNode& node) {
         std::visit(*this, *node->GetInnerData());
     }
 
@@ -2468,32 +2453,53 @@ void GLSLDecompiler::DecompileAST() {
     for (u32 i = 0; i < num_flow_variables; i++) {
         code.AddLine("bool {} = false;", GetFlowVariable(i));
     }
+
     ASTDecompiler decompiler{*this};
-    VideoCommon::Shader::ASTNode program = ir.GetASTProgram();
-    decompiler.Visit(program);
+    decompiler.Visit(ir.GetASTProgram());
 }
 
 } // Anonymous namespace
 
+ShaderEntries GetEntries(const VideoCommon::Shader::ShaderIR& ir) {
+    ShaderEntries entries;
+    for (const auto& cbuf : ir.GetConstantBuffers()) {
+        entries.const_buffers.emplace_back(cbuf.second.GetMaxOffset(), cbuf.second.IsIndirect(),
+                                           cbuf.first);
+    }
+    for (const auto& sampler : ir.GetSamplers()) {
+        entries.samplers.emplace_back(sampler);
+    }
+    for (const auto& [offset, image] : ir.GetImages()) {
+        entries.images.emplace_back(image);
+    }
+    for (const auto& [base, usage] : ir.GetGlobalMemory()) {
+        entries.global_memory_entries.emplace_back(base.cbuf_index, base.cbuf_offset, usage.is_read,
+                                                   usage.is_written);
+    }
+    entries.clip_distances = ir.GetClipDistances();
+    entries.shader_length = ir.GetLength();
+    return entries;
+}
+
 std::string GetCommonDeclarations() {
-    return fmt::format(
-        "#define ftoi floatBitsToInt\n"
-        "#define ftou floatBitsToUint\n"
-        "#define itof intBitsToFloat\n"
-        "#define utof uintBitsToFloat\n\n"
-        "bvec2 HalfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {{\n"
-        "    bvec2 is_nan1 = isnan(pair1);\n"
-        "    bvec2 is_nan2 = isnan(pair2);\n"
-        "    return bvec2(comparison.x || is_nan1.x || is_nan2.x, comparison.y || is_nan1.y || "
-        "is_nan2.y);\n"
-        "}}\n\n");
+    return R"(#define ftoi floatBitsToInt
+#define ftou floatBitsToUint
+#define itof intBitsToFloat
+#define utof uintBitsToFloat
+
+bvec2 HalfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {
+    bvec2 is_nan1 = isnan(pair1);
+    bvec2 is_nan2 = isnan(pair2);
+    return bvec2(comparison.x || is_nan1.x || is_nan2.x, comparison.y || is_nan1.y || is_nan2.y);
+}
+)";
 }
 
-ProgramResult Decompile(const Device& device, const ShaderIR& ir, ProgramType stage,
-                        const std::string& suffix) {
+std::string Decompile(const Device& device, const ShaderIR& ir, ProgramType stage,
+                      const std::string& suffix) {
     GLSLDecompiler decompiler(device, ir, stage, suffix);
     decompiler.Decompile();
-    return {decompiler.GetResult(), decompiler.GetShaderEntries()};
+    return decompiler.GetResult();
 }
 
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h
index e538dc001..fead2a51e 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.h
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h
@@ -34,10 +34,7 @@ enum class ProgramType : u32 {
 
 namespace OpenGL::GLShader {
 
-struct ShaderEntries;
-
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using ProgramResult = std::pair<std::string, ShaderEntries>;
 using SamplerEntry = VideoCommon::Shader::Sampler;
 using ImageEntry = VideoCommon::Shader::Image;
 
@@ -93,9 +90,11 @@ struct ShaderEntries {
     std::size_t shader_length{};
 };
 
+ShaderEntries GetEntries(const VideoCommon::Shader::ShaderIR& ir);
+
 std::string GetCommonDeclarations();
 
-ProgramResult Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
-                        ProgramType stage, const std::string& suffix);
+std::string Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
+                      ProgramType stage, const std::string& suffix);
 
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
index 74cc33476..184a565e6 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
@@ -22,6 +22,29 @@
 
 namespace OpenGL {
 
+using VideoCommon::Shader::BindlessSamplerMap;
+using VideoCommon::Shader::BoundSamplerMap;
+using VideoCommon::Shader::KeyMap;
+
+namespace {
+
+struct ConstBufferKey {
+    u32 cbuf;
+    u32 offset;
+    u32 value;
+};
+
+struct BoundSamplerKey {
+    u32 offset;
+    Tegra::Engines::SamplerDescriptor sampler;
+};
+
+struct BindlessSamplerKey {
+    u32 cbuf;
+    u32 offset;
+    Tegra::Engines::SamplerDescriptor sampler;
+};
+
 using ShaderCacheVersionHash = std::array<u8, 64>;
 
 enum class TransferableEntryKind : u32 {
@@ -29,18 +52,10 @@ enum class TransferableEntryKind : u32 {
     Usage,
 };
 
-enum class PrecompiledEntryKind : u32 {
-    Decompiled,
-    Dump,
-};
-
-constexpr u32 NativeVersion = 4;
+constexpr u32 NativeVersion = 5;
 
 // Making sure sizes doesn't change by accident
 static_assert(sizeof(BaseBindings) == 16);
-static_assert(sizeof(ShaderDiskCacheUsage) == 40);
-
-namespace {
 
 ShaderCacheVersionHash GetShaderCacheVersionHash() {
     ShaderCacheVersionHash hash{};
@@ -49,13 +64,11 @@ ShaderCacheVersionHash GetShaderCacheVersionHash() {
     return hash;
 }
 
-} // namespace
+} // Anonymous namespace
 
 ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
-                                       u32 program_code_size, u32 program_code_size_b,
                                        ProgramCode program_code, ProgramCode program_code_b)
     : unique_identifier{unique_identifier}, program_type{program_type},
-      program_code_size{program_code_size}, program_code_size_b{program_code_size_b},
       program_code{std::move(program_code)}, program_code_b{std::move(program_code_b)} {}
 
 ShaderDiskCacheRaw::ShaderDiskCacheRaw() = default;
@@ -90,15 +103,16 @@ bool ShaderDiskCacheRaw::Load(FileUtil::IOFile& file) {
 bool ShaderDiskCacheRaw::Save(FileUtil::IOFile& file) const {
     if (file.WriteObject(unique_identifier) != 1 ||
         file.WriteObject(static_cast<u32>(program_type)) != 1 ||
-        file.WriteObject(program_code_size) != 1 || file.WriteObject(program_code_size_b) != 1) {
+        file.WriteObject(static_cast<u32>(program_code.size())) != 1 ||
+        file.WriteObject(static_cast<u32>(program_code_b.size())) != 1) {
         return false;
     }
 
-    if (file.WriteArray(program_code.data(), program_code_size) != program_code_size)
+    if (file.WriteArray(program_code.data(), program_code.size()) != program_code.size())
         return false;
 
     if (HasProgramA() &&
-        file.WriteArray(program_code_b.data(), program_code_size_b) != program_code_size_b) {
+        file.WriteArray(program_code_b.data(), program_code_b.size()) != program_code_b.size()) {
         return false;
     }
     return true;
@@ -127,13 +141,13 @@ ShaderDiskCacheOpenGL::LoadTransferable() {
     u32 version{};
     if (file.ReadBytes(&version, sizeof(version)) != sizeof(version)) {
         LOG_ERROR(Render_OpenGL,
-                  "Failed to get transferable cache version for title id={} - skipping",
+                  "Failed to get transferable cache version for title id={}, skipping",
                   GetTitleID());
         return {};
     }
 
     if (version < NativeVersion) {
-        LOG_INFO(Render_OpenGL, "Transferable shader cache is old - removing");
+        LOG_INFO(Render_OpenGL, "Transferable shader cache is old, removing");
         file.Close();
         InvalidateTransferable();
         is_usable = true;
@@ -141,17 +155,18 @@ ShaderDiskCacheOpenGL::LoadTransferable() {
     }
     if (version > NativeVersion) {
         LOG_WARNING(Render_OpenGL, "Transferable shader cache was generated with a newer version "
-                                   "of the emulator - skipping");
+                                   "of the emulator, skipping");
         return {};
     }
 
     // Version is valid, load the shaders
+    constexpr const char error_loading[] = "Failed to load transferable raw entry, skipping";
     std::vector<ShaderDiskCacheRaw> raws;
     std::vector<ShaderDiskCacheUsage> usages;
     while (file.Tell() < file.GetSize()) {
         TransferableEntryKind kind{};
         if (file.ReadBytes(&kind, sizeof(u32)) != sizeof(u32)) {
-            LOG_ERROR(Render_OpenGL, "Failed to read transferable file - skipping");
+            LOG_ERROR(Render_OpenGL, "Failed to read transferable file, skipping");
             return {};
         }
 
@@ -159,7 +174,7 @@ ShaderDiskCacheOpenGL::LoadTransferable() {
         case TransferableEntryKind::Raw: {
             ShaderDiskCacheRaw entry;
             if (!entry.Load(file)) {
-                LOG_ERROR(Render_OpenGL, "Failed to load transferable raw entry - skipping");
+                LOG_ERROR(Render_OpenGL, error_loading);
                 return {};
             }
             transferable.insert({entry.GetUniqueIdentifier(), {}});
@@ -167,16 +182,45 @@ ShaderDiskCacheOpenGL::LoadTransferable() {
             break;
         }
         case TransferableEntryKind::Usage: {
-            ShaderDiskCacheUsage usage{};
-            if (file.ReadBytes(&usage, sizeof(usage)) != sizeof(usage)) {
-                LOG_ERROR(Render_OpenGL, "Failed to load transferable usage entry - skipping");
+            ShaderDiskCacheUsage usage;
+
+            u32 num_keys{};
+            u32 num_bound_samplers{};
+            u32 num_bindless_samplers{};
+            if (file.ReadArray(&usage.unique_identifier, 1) != 1 ||
+                file.ReadArray(&usage.variant, 1) != 1 || file.ReadArray(&num_keys, 1) != 1 ||
+                file.ReadArray(&num_bound_samplers, 1) != 1 ||
+                file.ReadArray(&num_bindless_samplers, 1) != 1) {
+                LOG_ERROR(Render_OpenGL, error_loading);
                 return {};
             }
+
+            std::vector<ConstBufferKey> keys(num_keys);
+            std::vector<BoundSamplerKey> bound_samplers(num_bound_samplers);
+            std::vector<BindlessSamplerKey> bindless_samplers(num_bindless_samplers);
+            if (file.ReadArray(keys.data(), keys.size()) != keys.size() ||
+                file.ReadArray(bound_samplers.data(), bound_samplers.size()) !=
+                    bound_samplers.size() ||
+                file.ReadArray(bindless_samplers.data(), bindless_samplers.size()) !=
+                    bindless_samplers.size()) {
+                LOG_ERROR(Render_OpenGL, error_loading);
+                return {};
+            }
+            for (const auto& key : keys) {
+                usage.keys.insert({{key.cbuf, key.offset}, key.value});
+            }
+            for (const auto& key : bound_samplers) {
+                usage.bound_samplers.emplace(key.offset, key.sampler);
+            }
+            for (const auto& key : bindless_samplers) {
+                usage.bindless_samplers.insert({{key.cbuf, key.offset}, key.sampler});
+            }
+
             usages.push_back(std::move(usage));
             break;
         }
         default:
-            LOG_ERROR(Render_OpenGL, "Unknown transferable shader cache entry kind={} - skipping",
+            LOG_ERROR(Render_OpenGL, "Unknown transferable shader cache entry kind={}, skipping",
                       static_cast<u32>(kind));
             return {};
         }
@@ -186,13 +230,14 @@ ShaderDiskCacheOpenGL::LoadTransferable() {
     return {{std::move(raws), std::move(usages)}};
 }
 
-std::pair<std::unordered_map<u64, ShaderDiskCacheDecompiled>, ShaderDumpsMap>
+std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>
 ShaderDiskCacheOpenGL::LoadPrecompiled() {
     if (!is_usable) {
         return {};
     }
 
-    FileUtil::IOFile file(GetPrecompiledPath(), "rb");
+    std::string path = GetPrecompiledPath();
+    FileUtil::IOFile file(path, "rb");
     if (!file.IsOpen()) {
         LOG_INFO(Render_OpenGL, "No precompiled shader cache found for game with title id={}",
                  GetTitleID());
@@ -202,7 +247,7 @@ ShaderDiskCacheOpenGL::LoadPrecompiled() {
     const auto result = LoadPrecompiledFile(file);
     if (!result) {
         LOG_INFO(Render_OpenGL,
-                 "Failed to load precompiled cache for game with title id={} - removing",
+                 "Failed to load precompiled cache for game with title id={}, removing",
                  GetTitleID());
         file.Close();
         InvalidatePrecompiled();
@@ -211,7 +256,7 @@ ShaderDiskCacheOpenGL::LoadPrecompiled() {
     return *result;
 }
 
-std::optional<std::pair<std::unordered_map<u64, ShaderDiskCacheDecompiled>, ShaderDumpsMap>>
+std::optional<std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>>
 ShaderDiskCacheOpenGL::LoadPrecompiledFile(FileUtil::IOFile& file) {
     // Read compressed file from disk and decompress to virtual precompiled cache file
     std::vector<u8> compressed(file.GetSize());
@@ -231,238 +276,56 @@ ShaderDiskCacheOpenGL::LoadPrecompiledFile(FileUtil::IOFile& file) {
         return {};
     }
 
-    std::unordered_map<u64, ShaderDiskCacheDecompiled> decompiled;
     ShaderDumpsMap dumps;
     while (precompiled_cache_virtual_file_offset < precompiled_cache_virtual_file.GetSize()) {
-        PrecompiledEntryKind kind{};
-        if (!LoadObjectFromPrecompiled(kind)) {
+        u32 num_keys{};
+        u32 num_bound_samplers{};
+        u32 num_bindless_samplers{};
+        ShaderDiskCacheUsage usage;
+        if (!LoadObjectFromPrecompiled(usage.unique_identifier) ||
+            !LoadObjectFromPrecompiled(usage.variant) || !LoadObjectFromPrecompiled(num_keys) ||
+            !LoadObjectFromPrecompiled(num_bound_samplers) ||
+            !LoadObjectFromPrecompiled(num_bindless_samplers)) {
             return {};
         }
-
-        switch (kind) {
-        case PrecompiledEntryKind::Decompiled: {
-            u64 unique_identifier{};
-            if (!LoadObjectFromPrecompiled(unique_identifier)) {
-                return {};
-            }
-
-            auto entry = LoadDecompiledEntry();
-            if (!entry) {
-                return {};
-            }
-            decompiled.insert({unique_identifier, std::move(*entry)});
-            break;
-        }
-        case PrecompiledEntryKind::Dump: {
-            ShaderDiskCacheUsage usage;
-            if (!LoadObjectFromPrecompiled(usage)) {
-                return {};
-            }
-
-            ShaderDiskCacheDump dump;
-            if (!LoadObjectFromPrecompiled(dump.binary_format)) {
-                return {};
-            }
-
-            u32 binary_length{};
-            if (!LoadObjectFromPrecompiled(binary_length)) {
-                return {};
-            }
-
-            dump.binary.resize(binary_length);
-            if (!LoadArrayFromPrecompiled(dump.binary.data(), dump.binary.size())) {
-                return {};
-            }
-
-            dumps.insert({usage, dump});
-            break;
-        }
-        default:
+        std::vector<ConstBufferKey> keys(num_keys);
+        std::vector<BoundSamplerKey> bound_samplers(num_bound_samplers);
+        std::vector<BindlessSamplerKey> bindless_samplers(num_bindless_samplers);
+        if (!LoadArrayFromPrecompiled(keys.data(), keys.size()) ||
+            !LoadArrayFromPrecompiled(bound_samplers.data(), bound_samplers.size()) !=
+                bound_samplers.size() ||
+            !LoadArrayFromPrecompiled(bindless_samplers.data(), bindless_samplers.size()) !=
+                bindless_samplers.size()) {
             return {};
         }
-    }
-    return {{decompiled, dumps}};
-}
-
-std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEntry() {
-    u32 code_size{};
-    if (!LoadObjectFromPrecompiled(code_size)) {
-        return {};
-    }
-
-    std::string code(code_size, '\0');
-    if (!LoadArrayFromPrecompiled(code.data(), code.size())) {
-        return {};
-    }
-
-    ShaderDiskCacheDecompiled entry;
-    entry.code = std::move(code);
-
-    u32 const_buffers_count{};
-    if (!LoadObjectFromPrecompiled(const_buffers_count)) {
-        return {};
-    }
-
-    for (u32 i = 0; i < const_buffers_count; ++i) {
-        u32 max_offset{};
-        u32 index{};
-        bool is_indirect{};
-        if (!LoadObjectFromPrecompiled(max_offset) || !LoadObjectFromPrecompiled(index) ||
-            !LoadObjectFromPrecompiled(is_indirect)) {
-            return {};
+        for (const auto& key : keys) {
+            usage.keys.insert({{key.cbuf, key.offset}, key.value});
         }
-        entry.entries.const_buffers.emplace_back(max_offset, is_indirect, index);
-    }
-
-    u32 samplers_count{};
-    if (!LoadObjectFromPrecompiled(samplers_count)) {
-        return {};
-    }
-
-    for (u32 i = 0; i < samplers_count; ++i) {
-        u64 offset{};
-        u64 index{};
-        u32 type{};
-        bool is_array{};
-        bool is_shadow{};
-        bool is_bindless{};
-        if (!LoadObjectFromPrecompiled(offset) || !LoadObjectFromPrecompiled(index) ||
-            !LoadObjectFromPrecompiled(type) || !LoadObjectFromPrecompiled(is_array) ||
-            !LoadObjectFromPrecompiled(is_shadow) || !LoadObjectFromPrecompiled(is_bindless)) {
-            return {};
+        for (const auto& key : bound_samplers) {
+            usage.bound_samplers.emplace(key.offset, key.sampler);
         }
-        entry.entries.samplers.emplace_back(
-            static_cast<std::size_t>(offset), static_cast<std::size_t>(index),
-            static_cast<Tegra::Shader::TextureType>(type), is_array, is_shadow, is_bindless);
-    }
-
-    u32 images_count{};
-    if (!LoadObjectFromPrecompiled(images_count)) {
-        return {};
-    }
-    for (u32 i = 0; i < images_count; ++i) {
-        u64 offset{};
-        u64 index{};
-        u32 type{};
-        u8 is_bindless{};
-        u8 is_written{};
-        u8 is_read{};
-        u8 is_atomic{};
-        if (!LoadObjectFromPrecompiled(offset) || !LoadObjectFromPrecompiled(index) ||
-            !LoadObjectFromPrecompiled(type) || !LoadObjectFromPrecompiled(is_bindless) ||
-            !LoadObjectFromPrecompiled(is_written) || !LoadObjectFromPrecompiled(is_read) ||
-            !LoadObjectFromPrecompiled(is_atomic)) {
-            return {};
+        for (const auto& key : bindless_samplers) {
+            usage.bindless_samplers.insert({{key.cbuf, key.offset}, key.sampler});
         }
-        entry.entries.images.emplace_back(
-            static_cast<std::size_t>(offset), static_cast<std::size_t>(index),
-            static_cast<Tegra::Shader::ImageType>(type), is_bindless != 0, is_written != 0,
-            is_read != 0, is_atomic != 0);
-    }
 
-    u32 global_memory_count{};
-    if (!LoadObjectFromPrecompiled(global_memory_count)) {
-        return {};
-    }
-    for (u32 i = 0; i < global_memory_count; ++i) {
-        u32 cbuf_index{};
-        u32 cbuf_offset{};
-        bool is_read{};
-        bool is_written{};
-        if (!LoadObjectFromPrecompiled(cbuf_index) || !LoadObjectFromPrecompiled(cbuf_offset) ||
-            !LoadObjectFromPrecompiled(is_read) || !LoadObjectFromPrecompiled(is_written)) {
+        ShaderDiskCacheDump dump;
+        if (!LoadObjectFromPrecompiled(dump.binary_format)) {
             return {};
         }
-        entry.entries.global_memory_entries.emplace_back(cbuf_index, cbuf_offset, is_read,
-                                                         is_written);
-    }
 
-    for (auto& clip_distance : entry.entries.clip_distances) {
-        if (!LoadObjectFromPrecompiled(clip_distance)) {
+        u32 binary_length{};
+        if (!LoadObjectFromPrecompiled(binary_length)) {
             return {};
         }
-    }
-
-    u64 shader_length{};
-    if (!LoadObjectFromPrecompiled(shader_length)) {
-        return {};
-    }
-    entry.entries.shader_length = static_cast<std::size_t>(shader_length);
-
-    return entry;
-}
-
-bool ShaderDiskCacheOpenGL::SaveDecompiledFile(u64 unique_identifier, const std::string& code,
-                                               const GLShader::ShaderEntries& entries) {
-    if (!SaveObjectToPrecompiled(static_cast<u32>(PrecompiledEntryKind::Decompiled)) ||
-        !SaveObjectToPrecompiled(unique_identifier) ||
-        !SaveObjectToPrecompiled(static_cast<u32>(code.size())) ||
-        !SaveArrayToPrecompiled(code.data(), code.size())) {
-        return false;
-    }
-
-    if (!SaveObjectToPrecompiled(static_cast<u32>(entries.const_buffers.size()))) {
-        return false;
-    }
-    for (const auto& cbuf : entries.const_buffers) {
-        if (!SaveObjectToPrecompiled(static_cast<u32>(cbuf.GetMaxOffset())) ||
-            !SaveObjectToPrecompiled(static_cast<u32>(cbuf.GetIndex())) ||
-            !SaveObjectToPrecompiled(cbuf.IsIndirect())) {
-            return false;
-        }
-    }
-
-    if (!SaveObjectToPrecompiled(static_cast<u32>(entries.samplers.size()))) {
-        return false;
-    }
-    for (const auto& sampler : entries.samplers) {
-        if (!SaveObjectToPrecompiled(static_cast<u64>(sampler.GetOffset())) ||
-            !SaveObjectToPrecompiled(static_cast<u64>(sampler.GetIndex())) ||
-            !SaveObjectToPrecompiled(static_cast<u32>(sampler.GetType())) ||
-            !SaveObjectToPrecompiled(sampler.IsArray()) ||
-            !SaveObjectToPrecompiled(sampler.IsShadow()) ||
-            !SaveObjectToPrecompiled(sampler.IsBindless())) {
-            return false;
-        }
-    }
-
-    if (!SaveObjectToPrecompiled(static_cast<u32>(entries.images.size()))) {
-        return false;
-    }
-    for (const auto& image : entries.images) {
-        if (!SaveObjectToPrecompiled(static_cast<u64>(image.GetOffset())) ||
-            !SaveObjectToPrecompiled(static_cast<u64>(image.GetIndex())) ||
-            !SaveObjectToPrecompiled(static_cast<u32>(image.GetType())) ||
-            !SaveObjectToPrecompiled(static_cast<u8>(image.IsBindless() ? 1 : 0)) ||
-            !SaveObjectToPrecompiled(static_cast<u8>(image.IsWritten() ? 1 : 0)) ||
-            !SaveObjectToPrecompiled(static_cast<u8>(image.IsRead() ? 1 : 0)) ||
-            !SaveObjectToPrecompiled(static_cast<u8>(image.IsAtomic() ? 1 : 0))) {
-            return false;
-        }
-    }
 
-    if (!SaveObjectToPrecompiled(static_cast<u32>(entries.global_memory_entries.size()))) {
-        return false;
-    }
-    for (const auto& gmem : entries.global_memory_entries) {
-        if (!SaveObjectToPrecompiled(static_cast<u32>(gmem.GetCbufIndex())) ||
-            !SaveObjectToPrecompiled(static_cast<u32>(gmem.GetCbufOffset())) ||
-            !SaveObjectToPrecompiled(gmem.IsRead()) || !SaveObjectToPrecompiled(gmem.IsWritten())) {
-            return false;
-        }
-    }
-
-    for (const bool clip_distance : entries.clip_distances) {
-        if (!SaveObjectToPrecompiled(clip_distance)) {
-            return false;
+        dump.binary.resize(binary_length);
+        if (!LoadArrayFromPrecompiled(dump.binary.data(), dump.binary.size())) {
+            return {};
         }
-    }
 
-    if (!SaveObjectToPrecompiled(static_cast<u64>(entries.shader_length))) {
-        return false;
+        dumps.emplace(std::move(usage), dump);
     }
-
-    return true;
+    return dumps;
 }
 
 void ShaderDiskCacheOpenGL::InvalidateTransferable() {
@@ -494,10 +357,11 @@ void ShaderDiskCacheOpenGL::SaveRaw(const ShaderDiskCacheRaw& entry) {
     }
 
     FileUtil::IOFile file = AppendTransferableFile();
-    if (!file.IsOpen())
+    if (!file.IsOpen()) {
         return;
+    }
     if (file.WriteObject(TransferableEntryKind::Raw) != 1 || !entry.Save(file)) {
-        LOG_ERROR(Render_OpenGL, "Failed to save raw transferable cache entry - removing");
+        LOG_ERROR(Render_OpenGL, "Failed to save raw transferable cache entry, removing");
         file.Close();
         InvalidateTransferable();
         return;
@@ -523,29 +387,39 @@ void ShaderDiskCacheOpenGL::SaveUsage(const ShaderDiskCacheUsage& usage) {
     FileUtil::IOFile file = AppendTransferableFile();
     if (!file.IsOpen())
         return;
-
-    if (file.WriteObject(TransferableEntryKind::Usage) != 1 || file.WriteObject(usage) != 1) {
-        LOG_ERROR(Render_OpenGL, "Failed to save usage transferable cache entry - removing");
+    const auto Close = [&] {
+        LOG_ERROR(Render_OpenGL, "Failed to save usage transferable cache entry, removing");
         file.Close();
         InvalidateTransferable();
-        return;
-    }
-}
+    };
 
-void ShaderDiskCacheOpenGL::SaveDecompiled(u64 unique_identifier, const std::string& code,
-                                           const GLShader::ShaderEntries& entries) {
-    if (!is_usable) {
+    if (file.WriteObject(TransferableEntryKind::Usage) != 1 ||
+        file.WriteObject(usage.unique_identifier) != 1 || file.WriteObject(usage.variant) != 1 ||
+        file.WriteObject(static_cast<u32>(usage.keys.size())) != 1 ||
+        file.WriteObject(static_cast<u32>(usage.bound_samplers.size())) != 1 ||
+        file.WriteObject(static_cast<u32>(usage.bindless_samplers.size())) != 1) {
+        Close();
         return;
     }
-
-    if (precompiled_cache_virtual_file.GetSize() == 0) {
-        SavePrecompiledHeaderToVirtualPrecompiledCache();
+    for (const auto& [pair, value] : usage.keys) {
+        const auto [cbuf, offset] = pair;
+        if (file.WriteObject(ConstBufferKey{cbuf, offset, value}) != 1) {
+            Close();
+            return;
+        }
     }
-
-    if (!SaveDecompiledFile(unique_identifier, code, entries)) {
-        LOG_ERROR(Render_OpenGL,
-                  "Failed to save decompiled entry to the precompiled file - removing");
-        InvalidatePrecompiled();
+    for (const auto& [offset, sampler] : usage.bound_samplers) {
+        if (file.WriteObject(BoundSamplerKey{offset, sampler}) != 1) {
+            Close();
+            return;
+        }
+    }
+    for (const auto& [pair, sampler] : usage.bindless_samplers) {
+        const auto [cbuf, offset] = pair;
+        if (file.WriteObject(BindlessSamplerKey{cbuf, offset, sampler}) != 1) {
+            Close();
+            return;
+        }
     }
 }
 
@@ -554,6 +428,13 @@ void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint p
         return;
     }
 
+    // TODO(Rodrigo): This is a design smell. I shouldn't be having to manually write the header
+    // when writing the dump. This should be done the moment I get access to write to the virtual
+    // file.
+    if (precompiled_cache_virtual_file.GetSize() == 0) {
+        SavePrecompiledHeaderToVirtualPrecompiledCache();
+    }
+
     GLint binary_length{};
     glGetProgramiv(program, GL_PROGRAM_BINARY_LENGTH, &binary_length);
 
@@ -561,21 +442,51 @@ void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint p
     std::vector<u8> binary(binary_length);
     glGetProgramBinary(program, binary_length, nullptr, &binary_format, binary.data());
 
-    if (!SaveObjectToPrecompiled(static_cast<u32>(PrecompiledEntryKind::Dump)) ||
-        !SaveObjectToPrecompiled(usage) ||
-        !SaveObjectToPrecompiled(static_cast<u32>(binary_format)) ||
-        !SaveObjectToPrecompiled(static_cast<u32>(binary_length)) ||
-        !SaveArrayToPrecompiled(binary.data(), binary.size())) {
-        LOG_ERROR(Render_OpenGL, "Failed to save binary program file in shader={:016x} - removing",
+    const auto Close = [&] {
+        LOG_ERROR(Render_OpenGL, "Failed to save binary program file in shader={:016X}, removing",
                   usage.unique_identifier);
         InvalidatePrecompiled();
+    };
+
+    if (!SaveObjectToPrecompiled(usage.unique_identifier) ||
+        !SaveObjectToPrecompiled(usage.variant) ||
+        !SaveObjectToPrecompiled(static_cast<u32>(usage.keys.size())) ||
+        !SaveObjectToPrecompiled(static_cast<u32>(usage.bound_samplers.size())) ||
+        !SaveObjectToPrecompiled(static_cast<u32>(usage.bindless_samplers.size()))) {
+        Close();
         return;
     }
+    for (const auto& [pair, value] : usage.keys) {
+        const auto [cbuf, offset] = pair;
+        if (SaveObjectToPrecompiled(ConstBufferKey{cbuf, offset, value}) != 1) {
+            Close();
+            return;
+        }
+    }
+    for (const auto& [offset, sampler] : usage.bound_samplers) {
+        if (SaveObjectToPrecompiled(BoundSamplerKey{offset, sampler}) != 1) {
+            Close();
+            return;
+        }
+    }
+    for (const auto& [pair, sampler] : usage.bindless_samplers) {
+        const auto [cbuf, offset] = pair;
+        if (SaveObjectToPrecompiled(BindlessSamplerKey{cbuf, offset, sampler}) != 1) {
+            Close();
+            return;
+        }
+    }
+    if (!SaveObjectToPrecompiled(static_cast<u32>(binary_format)) ||
+        !SaveObjectToPrecompiled(static_cast<u32>(binary_length)) ||
+        !SaveArrayToPrecompiled(binary.data(), binary.size())) {
+        Close();
+    }
 }
 
 FileUtil::IOFile ShaderDiskCacheOpenGL::AppendTransferableFile() const {
-    if (!EnsureDirectories())
+    if (!EnsureDirectories()) {
         return {};
+    }
 
     const auto transferable_path{GetTransferablePath()};
     const bool existed = FileUtil::Exists(transferable_path);
@@ -607,8 +518,8 @@ void ShaderDiskCacheOpenGL::SavePrecompiledHeaderToVirtualPrecompiledCache() {
 
 void ShaderDiskCacheOpenGL::SaveVirtualPrecompiledFile() {
     precompiled_cache_virtual_file_offset = 0;
-    const std::vector<u8>& uncompressed = precompiled_cache_virtual_file.ReadAllBytes();
-    const std::vector<u8>& compressed =
+    const std::vector<u8> uncompressed = precompiled_cache_virtual_file.ReadAllBytes();
+    const std::vector<u8> compressed =
         Common::Compression::CompressDataZSTDDefault(uncompressed.data(), uncompressed.size());
 
     const auto precompiled_path{GetPrecompiledPath()};
diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.h b/src/video_core/renderer_opengl/gl_shader_disk_cache.h
index 9595bd71b..db23ada93 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.h
@@ -8,6 +8,7 @@
 #include <optional>
 #include <string>
 #include <tuple>
+#include <type_traits>
 #include <unordered_map>
 #include <unordered_set>
 #include <utility>
@@ -19,6 +20,7 @@
 #include "common/common_types.h"
 #include "core/file_sys/vfs_vector.h"
 #include "video_core/renderer_opengl/gl_shader_gen.h"
+#include "video_core/shader/const_buffer_locker.h"
 
 namespace Core {
 class System;
@@ -53,6 +55,7 @@ struct BaseBindings {
         return !operator==(rhs);
     }
 };
+static_assert(std::is_trivially_copyable_v<BaseBindings>);
 
 /// Describes the different variants a single program can be compiled.
 struct ProgramVariant {
@@ -70,13 +73,20 @@ struct ProgramVariant {
     }
 };
 
+static_assert(std::is_trivially_copyable_v<ProgramVariant>);
+
 /// Describes how a shader is used.
 struct ShaderDiskCacheUsage {
     u64 unique_identifier{};
     ProgramVariant variant;
+    VideoCommon::Shader::KeyMap keys;
+    VideoCommon::Shader::BoundSamplerMap bound_samplers;
+    VideoCommon::Shader::BindlessSamplerMap bindless_samplers;
 
     bool operator==(const ShaderDiskCacheUsage& rhs) const {
-        return std::tie(unique_identifier, variant) == std::tie(rhs.unique_identifier, rhs.variant);
+        return std::tie(unique_identifier, variant, keys, bound_samplers, bindless_samplers) ==
+               std::tie(rhs.unique_identifier, rhs.variant, rhs.keys, rhs.bound_samplers,
+                        rhs.bindless_samplers);
     }
 
     bool operator!=(const ShaderDiskCacheUsage& rhs) const {
@@ -123,8 +133,7 @@ namespace OpenGL {
 class ShaderDiskCacheRaw {
 public:
     explicit ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
-                                u32 program_code_size, u32 program_code_size_b,
-                                ProgramCode program_code, ProgramCode program_code_b);
+                                ProgramCode program_code, ProgramCode program_code_b = {});
     ShaderDiskCacheRaw();
     ~ShaderDiskCacheRaw();
 
@@ -155,22 +164,14 @@ public:
 private:
     u64 unique_identifier{};
     ProgramType program_type{};
-    u32 program_code_size{};
-    u32 program_code_size_b{};
 
     ProgramCode program_code;
     ProgramCode program_code_b;
 };
 
-/// Contains decompiled data from a shader
-struct ShaderDiskCacheDecompiled {
-    std::string code;
-    GLShader::ShaderEntries entries;
-};
-
 /// Contains an OpenGL dumped binary program
 struct ShaderDiskCacheDump {
-    GLenum binary_format;
+    GLenum binary_format{};
     std::vector<u8> binary;
 };
 
@@ -184,9 +185,7 @@ public:
     LoadTransferable();
 
     /// Loads current game's precompiled cache. Invalidates on failure.
-    std::pair<std::unordered_map<u64, ShaderDiskCacheDecompiled>,
-              std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>>
-    LoadPrecompiled();
+    std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump> LoadPrecompiled();
 
     /// Removes the transferable (and precompiled) cache file.
     void InvalidateTransferable();
@@ -200,10 +199,6 @@ public:
     /// Saves shader usage to the transferable file. Does not check for collisions.
     void SaveUsage(const ShaderDiskCacheUsage& usage);
 
-    /// Saves a decompiled entry to the precompiled file. Does not check for collisions.
-    void SaveDecompiled(u64 unique_identifier, const std::string& code,
-                        const GLShader::ShaderEntries& entries);
-
     /// Saves a dump entry to the precompiled file. Does not check for collisions.
     void SaveDump(const ShaderDiskCacheUsage& usage, GLuint program);
 
@@ -212,18 +207,9 @@ public:
 
 private:
     /// Loads the transferable cache. Returns empty on failure.
-    std::optional<std::pair<std::unordered_map<u64, ShaderDiskCacheDecompiled>,
-                            std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>>>
+    std::optional<std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>>
     LoadPrecompiledFile(FileUtil::IOFile& file);
 
-    /// Loads a decompiled cache entry from m_precompiled_cache_virtual_file. Returns empty on
-    /// failure.
-    std::optional<ShaderDiskCacheDecompiled> LoadDecompiledEntry();
-
-    /// Saves a decompiled entry to the passed file. Returns true on success.
-    bool SaveDecompiledFile(u64 unique_identifier, const std::string& code,
-                            const GLShader::ShaderEntries& entries);
-
     /// Opens current game's transferable file and write it's header if it doesn't exist
     FileUtil::IOFile AppendTransferableFile() const;
 
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp
index b5a43e79e..0e22eede9 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp
@@ -16,17 +16,8 @@ using VideoCommon::Shader::CompilerSettings;
 using VideoCommon::Shader::ProgramCode;
 using VideoCommon::Shader::ShaderIR;
 
-static constexpr u32 PROGRAM_OFFSET = 10;
-static constexpr u32 COMPUTE_OFFSET = 0;
-
-static constexpr CompilerSettings settings{CompileDepth::NoFlowStack, true};
-
-ProgramResult GenerateVertexShader(const Device& device, const ShaderSetup& setup) {
-    const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
-
-    std::string out = "// Shader Unique Id: VS" + id + "\n\n";
-    out += GetCommonDeclarations();
-
+std::string GenerateVertexShader(const Device& device, const ShaderIR& ir, const ShaderIR* ir_b) {
+    std::string out = GetCommonDeclarations();
     out += R"(
 layout (std140, binding = EMULATION_UBO_BINDING) uniform vs_config {
     vec4 viewport_flip;
@@ -34,17 +25,10 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform vs_config {
 };
 
 )";
-
-    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a, settings);
-    const auto stage = setup.IsDualProgram() ? ProgramType::VertexA : ProgramType::VertexB;
-    ProgramResult program = Decompile(device, program_ir, stage, "vertex");
-    out += program.first;
-
-    if (setup.IsDualProgram()) {
-        const ShaderIR program_ir_b(setup.program.code_b, PROGRAM_OFFSET, setup.program.size_b,
-                                    settings);
-        ProgramResult program_b = Decompile(device, program_ir_b, ProgramType::VertexB, "vertex_b");
-        out += program_b.first;
+    const auto stage = ir_b ? ProgramType::VertexA : ProgramType::VertexB;
+    out += Decompile(device, ir, stage, "vertex");
+    if (ir_b) {
+        out += Decompile(device, *ir_b, ProgramType::VertexB, "vertex_b");
     }
 
     out += R"(
@@ -52,7 +36,7 @@ void main() {
     execute_vertex();
 )";
 
-    if (setup.IsDualProgram()) {
+    if (ir_b) {
         out += "    execute_vertex_b();";
     }
 
@@ -66,17 +50,13 @@ void main() {
         // Viewport can be flipped, which is unsupported by glViewport
         gl_Position.xy *= viewport_flip.xy;
     }
-})";
-
-    return {std::move(out), std::move(program.second)};
+}
+)";
+    return out;
 }
 
-ProgramResult GenerateGeometryShader(const Device& device, const ShaderSetup& setup) {
-    const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
-
-    std::string out = "// Shader Unique Id: GS" + id + "\n\n";
-    out += GetCommonDeclarations();
-
+std::string GenerateGeometryShader(const Device& device, const ShaderIR& ir) {
+    std::string out = GetCommonDeclarations();
     out += R"(
 layout (std140, binding = EMULATION_UBO_BINDING) uniform gs_config {
     vec4 viewport_flip;
@@ -84,25 +64,18 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform gs_config {
 };
 
 )";
-
-    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a, settings);
-    ProgramResult program = Decompile(device, program_ir, ProgramType::Geometry, "geometry");
-    out += program.first;
+    out += Decompile(device, ir, ProgramType::Geometry, "geometry");
 
     out += R"(
 void main() {
     execute_geometry();
-};)";
-
-    return {std::move(out), std::move(program.second)};
+}
+)";
+    return out;
 }
 
-ProgramResult GenerateFragmentShader(const Device& device, const ShaderSetup& setup) {
-    const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
-
-    std::string out = "// Shader Unique Id: FS" + id + "\n\n";
-    out += GetCommonDeclarations();
-
+std::string GenerateFragmentShader(const Device& device, const ShaderIR& ir) {
+    std::string out = GetCommonDeclarations();
     out += R"(
 layout (location = 0) out vec4 FragColor0;
 layout (location = 1) out vec4 FragColor1;
@@ -119,36 +92,25 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform fs_config {
 };
 
 )";
-
-    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a, settings);
-    ProgramResult program = Decompile(device, program_ir, ProgramType::Fragment, "fragment");
-    out += program.first;
+    out += Decompile(device, ir, ProgramType::Fragment, "fragment");
 
     out += R"(
 void main() {
     execute_fragment();
 }
-
 )";
-    return {std::move(out), std::move(program.second)};
+    return out;
 }
 
-ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& setup) {
-    const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
-
-    std::string out = "// Shader Unique Id: CS" + id + "\n\n";
-    out += GetCommonDeclarations();
-
-    const ShaderIR program_ir(setup.program.code, COMPUTE_OFFSET, setup.program.size_a, settings);
-    ProgramResult program = Decompile(device, program_ir, ProgramType::Compute, "compute");
-    out += program.first;
-
+std::string GenerateComputeShader(const Device& device, const ShaderIR& ir) {
+    std::string out = GetCommonDeclarations();
+    out += Decompile(device, ir, ProgramType::Compute, "compute");
     out += R"(
 void main() {
     execute_compute();
 }
 )";
-    return {std::move(out), std::move(program.second)};
+    return out;
 }
 
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.h b/src/video_core/renderer_opengl/gl_shader_gen.h
index 3833e88ab..cba2be9f9 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.h
+++ b/src/video_core/renderer_opengl/gl_shader_gen.h
@@ -17,44 +17,18 @@ class Device;
 namespace OpenGL::GLShader {
 
 using VideoCommon::Shader::ProgramCode;
-
-struct ShaderSetup {
-    explicit ShaderSetup(ProgramCode program_code) {
-        program.code = std::move(program_code);
-    }
-
-    struct {
-        ProgramCode code;
-        ProgramCode code_b; // Used for dual vertex shaders
-        u64 unique_identifier;
-        std::size_t size_a;
-        std::size_t size_b;
-    } program;
-
-    /// Used in scenarios where we have a dual vertex shaders
-    void SetProgramB(ProgramCode program_b) {
-        program.code_b = std::move(program_b);
-        has_program_b = true;
-    }
-
-    bool IsDualProgram() const {
-        return has_program_b;
-    }
-
-private:
-    bool has_program_b{};
-};
+using VideoCommon::Shader::ShaderIR;
 
 /// Generates the GLSL vertex shader program source code for the given VS program
-ProgramResult GenerateVertexShader(const Device& device, const ShaderSetup& setup);
+std::string GenerateVertexShader(const Device& device, const ShaderIR& ir, const ShaderIR* ir_b);
 
 /// Generates the GLSL geometry shader program source code for the given GS program
-ProgramResult GenerateGeometryShader(const Device& device, const ShaderSetup& setup);
+std::string GenerateGeometryShader(const Device& device, const ShaderIR& ir);
 
 /// Generates the GLSL fragment shader program source code for the given FS program
-ProgramResult GenerateFragmentShader(const Device& device, const ShaderSetup& setup);
+std::string GenerateFragmentShader(const Device& device, const ShaderIR& ir);
 
 /// Generates the GLSL compute shader program source code for the given CS program
-ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& setup);
+std::string GenerateComputeShader(const Device& device, const ShaderIR& ir);
 
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 173b76c4e..2f9bfd7e4 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -111,7 +111,8 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format
     {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm,
      true}, // DXT45_SRGB
     {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm,
-     true},                                                                    // BC7U_SRGB
+     true},                                                                          // BC7U_SRGB
+    {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV, ComponentType::UNorm, false}, // R4G4B4A4U
     {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_4X4_SRGB
     {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X8_SRGB
     {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X5_SRGB
@@ -120,6 +121,16 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format
     {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_5X5_SRGB
     {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},        // ASTC_2D_10X8
     {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_10X8_SRGB
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},        // ASTC_2D_6X6
+    {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_6X6_SRGB
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},        // ASTC_2D_10X10
+    {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_10X10_SRGB
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},        // ASTC_2D_12X12
+    {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_12X12_SRGB
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},        // ASTC_2D_8X6
+    {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X6_SRGB
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},        // ASTC_2D_6X5
+    {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_6X5_SRGB
 
     // Depth formats
     {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT, ComponentType::Float, false}, // Z32F
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index 1e6ef66ab..4bbd17b12 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -102,8 +102,6 @@ RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::Syst
 RendererOpenGL::~RendererOpenGL() = default;
 
 void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-    system.GetPerfStats().EndSystemFrame();
-
     // Maintain the rasterizer's state as a priority
     OpenGLState prev_state = OpenGLState::GetCurState();
     state.AllDirty();
@@ -135,9 +133,6 @@ void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
 
     render_window.PollEvents();
 
-    system.FrameLimiter().DoFrameLimiting(system.CoreTiming().GetGlobalTimeUs());
-    system.GetPerfStats().BeginSystemFrame();
-
     // Restore the rasterizer state
     prev_state.AllDirty();
     prev_state.Apply();
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index 8bcd04221..42cf068b6 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -1648,32 +1648,32 @@ class ExprDecompiler {
 public:
     explicit ExprDecompiler(SPIRVDecompiler& decomp) : decomp{decomp} {}
 
-    Id operator()(VideoCommon::Shader::ExprAnd& expr) {
+    Id operator()(const ExprAnd& expr) {
         const Id type_def = decomp.GetTypeDefinition(Type::Bool);
         const Id op1 = Visit(expr.operand1);
         const Id op2 = Visit(expr.operand2);
         return decomp.Emit(decomp.OpLogicalAnd(type_def, op1, op2));
     }
 
-    Id operator()(VideoCommon::Shader::ExprOr& expr) {
+    Id operator()(const ExprOr& expr) {
         const Id type_def = decomp.GetTypeDefinition(Type::Bool);
         const Id op1 = Visit(expr.operand1);
         const Id op2 = Visit(expr.operand2);
         return decomp.Emit(decomp.OpLogicalOr(type_def, op1, op2));
     }
 
-    Id operator()(VideoCommon::Shader::ExprNot& expr) {
+    Id operator()(const ExprNot& expr) {
         const Id type_def = decomp.GetTypeDefinition(Type::Bool);
         const Id op1 = Visit(expr.operand1);
         return decomp.Emit(decomp.OpLogicalNot(type_def, op1));
     }
 
-    Id operator()(VideoCommon::Shader::ExprPredicate& expr) {
+    Id operator()(const ExprPredicate& expr) {
         const auto pred = static_cast<Tegra::Shader::Pred>(expr.predicate);
         return decomp.Emit(decomp.OpLoad(decomp.t_bool, decomp.predicates.at(pred)));
     }
 
-    Id operator()(VideoCommon::Shader::ExprCondCode& expr) {
+    Id operator()(const ExprCondCode& expr) {
         const Node cc = decomp.ir.GetConditionCode(expr.cc);
         Id target;
 
@@ -1682,10 +1682,13 @@ public:
             switch (index) {
             case Tegra::Shader::Pred::NeverExecute:
                 target = decomp.v_false;
+                break;
             case Tegra::Shader::Pred::UnusedIndex:
                 target = decomp.v_true;
+                break;
             default:
                 target = decomp.predicates.at(index);
+                break;
             }
         } else if (const auto flag = std::get_if<InternalFlagNode>(&*cc)) {
             target = decomp.internal_flags.at(static_cast<u32>(flag->GetFlag()));
@@ -1693,15 +1696,22 @@ public:
         return decomp.Emit(decomp.OpLoad(decomp.t_bool, target));
     }
 
-    Id operator()(VideoCommon::Shader::ExprVar& expr) {
+    Id operator()(const ExprVar& expr) {
         return decomp.Emit(decomp.OpLoad(decomp.t_bool, decomp.flow_variables.at(expr.var_index)));
     }
 
-    Id operator()(VideoCommon::Shader::ExprBoolean& expr) {
+    Id operator()(const ExprBoolean& expr) {
         return expr.value ? decomp.v_true : decomp.v_false;
     }
 
-    Id Visit(VideoCommon::Shader::Expr& node) {
+    Id operator()(const ExprGprEqual& expr) {
+        const Id target = decomp.Constant(decomp.t_uint, expr.value);
+        const Id gpr = decomp.BitcastTo<Type::Uint>(
+            decomp.Emit(decomp.OpLoad(decomp.t_float, decomp.registers.at(expr.gpr))));
+        return decomp.Emit(decomp.OpLogicalEqual(decomp.t_uint, gpr, target));
+    }
+
+    Id Visit(const Expr& node) {
         return std::visit(*this, *node);
     }
 
@@ -1713,7 +1723,7 @@ class ASTDecompiler {
 public:
     explicit ASTDecompiler(SPIRVDecompiler& decomp) : decomp{decomp} {}
 
-    void operator()(VideoCommon::Shader::ASTProgram& ast) {
+    void operator()(const ASTProgram& ast) {
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
             Visit(current);
@@ -1721,7 +1731,7 @@ public:
         }
     }
 
-    void operator()(VideoCommon::Shader::ASTIfThen& ast) {
+    void operator()(const ASTIfThen& ast) {
         ExprDecompiler expr_parser{decomp};
         const Id condition = expr_parser.Visit(ast.condition);
         const Id then_label = decomp.OpLabel();
@@ -1738,33 +1748,33 @@ public:
         decomp.Emit(endif_label);
     }
 
-    void operator()(VideoCommon::Shader::ASTIfElse& ast) {
+    void operator()([[maybe_unused]] const ASTIfElse& ast) {
         UNREACHABLE();
     }
 
-    void operator()(VideoCommon::Shader::ASTBlockEncoded& ast) {
+    void operator()([[maybe_unused]] const ASTBlockEncoded& ast) {
         UNREACHABLE();
     }
 
-    void operator()(VideoCommon::Shader::ASTBlockDecoded& ast) {
+    void operator()(const ASTBlockDecoded& ast) {
         decomp.VisitBasicBlock(ast.nodes);
     }
 
-    void operator()(VideoCommon::Shader::ASTVarSet& ast) {
+    void operator()(const ASTVarSet& ast) {
         ExprDecompiler expr_parser{decomp};
         const Id condition = expr_parser.Visit(ast.condition);
         decomp.Emit(decomp.OpStore(decomp.flow_variables.at(ast.index), condition));
     }
 
-    void operator()(VideoCommon::Shader::ASTLabel& ast) {
+    void operator()([[maybe_unused]] const ASTLabel& ast) {
         // Do nothing
     }
 
-    void operator()(VideoCommon::Shader::ASTGoto& ast) {
+    void operator()([[maybe_unused]] const ASTGoto& ast) {
         UNREACHABLE();
     }
 
-    void operator()(VideoCommon::Shader::ASTDoWhile& ast) {
+    void operator()(const ASTDoWhile& ast) {
         const Id loop_label = decomp.OpLabel();
         const Id endloop_label = decomp.OpLabel();
         const Id loop_start_block = decomp.OpLabel();
@@ -1787,7 +1797,7 @@ public:
         decomp.Emit(endloop_label);
     }
 
-    void operator()(VideoCommon::Shader::ASTReturn& ast) {
+    void operator()(const ASTReturn& ast) {
         if (!VideoCommon::Shader::ExprIsTrue(ast.condition)) {
             ExprDecompiler expr_parser{decomp};
             const Id condition = expr_parser.Visit(ast.condition);
@@ -1817,7 +1827,7 @@ public:
         }
     }
 
-    void operator()(VideoCommon::Shader::ASTBreak& ast) {
+    void operator()(const ASTBreak& ast) {
         if (!VideoCommon::Shader::ExprIsTrue(ast.condition)) {
             ExprDecompiler expr_parser{decomp};
             const Id condition = expr_parser.Visit(ast.condition);
@@ -1837,7 +1847,7 @@ public:
         }
     }
 
-    void Visit(VideoCommon::Shader::ASTNode& node) {
+    void Visit(const ASTNode& node) {
         std::visit(*this, *node->GetInnerData());
     }
 
@@ -1853,9 +1863,11 @@ void SPIRVDecompiler::DecompileAST() {
         Name(id, fmt::format("flow_var_{}", i));
         flow_variables.emplace(i, AddGlobalVariable(id));
     }
+
+    const ASTNode program = ir.GetASTProgram();
     ASTDecompiler decompiler{*this};
-    VideoCommon::Shader::ASTNode program = ir.GetASTProgram();
     decompiler.Visit(program);
+
     const Id next_block = OpLabel();
     Emit(OpBranch(next_block));
     Emit(next_block);
diff --git a/src/video_core/shader/ast.cpp b/src/video_core/shader/ast.cpp
index 436d45f4b..3f96d9076 100644
--- a/src/video_core/shader/ast.cpp
+++ b/src/video_core/shader/ast.cpp
@@ -3,6 +3,9 @@
 // Refer to the license.txt file included.
 
 #include <string>
+#include <string_view>
+
+#include <fmt/format.h>
 
 #include "common/assert.h"
 #include "common/common_types.h"
@@ -225,11 +228,16 @@ public:
         inner += expr.value ? "true" : "false";
     }
 
+    void operator()(const ExprGprEqual& expr) {
+        inner += "( gpr_" + std::to_string(expr.gpr) + " == " + std::to_string(expr.value) + ')';
+    }
+
     const std::string& GetResult() const {
         return inner;
     }
 
-    std::string inner{};
+private:
+    std::string inner;
 };
 
 class ASTPrinter {
@@ -249,7 +257,7 @@ public:
     void operator()(const ASTIfThen& ast) {
         ExprPrinter expr_parser{};
         std::visit(expr_parser, *ast.condition);
-        inner += Ident() + "if (" + expr_parser.GetResult() + ") {\n";
+        inner += fmt::format("{}if ({}) {{\n", Indent(), expr_parser.GetResult());
         scope++;
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
@@ -257,11 +265,13 @@ public:
             current = current->GetNext();
         }
         scope--;
-        inner += Ident() + "}\n";
+        inner += fmt::format("{}}}\n", Indent());
     }
 
     void operator()(const ASTIfElse& ast) {
-        inner += Ident() + "else {\n";
+        inner += Indent();
+        inner += "else {\n";
+
         scope++;
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
@@ -269,40 +279,41 @@ public:
             current = current->GetNext();
         }
         scope--;
-        inner += Ident() + "}\n";
+
+        inner += Indent();
+        inner += "}\n";
     }
 
     void operator()(const ASTBlockEncoded& ast) {
-        inner += Ident() + "Block(" + std::to_string(ast.start) + ", " + std::to_string(ast.end) +
-                 ");\n";
+        inner += fmt::format("{}Block({}, {});\n", Indent(), ast.start, ast.end);
     }
 
-    void operator()(const ASTBlockDecoded& ast) {
-        inner += Ident() + "Block;\n";
+    void operator()([[maybe_unused]] const ASTBlockDecoded& ast) {
+        inner += Indent();
+        inner += "Block;\n";
     }
 
     void operator()(const ASTVarSet& ast) {
         ExprPrinter expr_parser{};
         std::visit(expr_parser, *ast.condition);
-        inner +=
-            Ident() + "V" + std::to_string(ast.index) + " := " + expr_parser.GetResult() + ";\n";
+        inner += fmt::format("{}V{} := {};\n", Indent(), ast.index, expr_parser.GetResult());
     }
 
     void operator()(const ASTLabel& ast) {
-        inner += "Label_" + std::to_string(ast.index) + ":\n";
+        inner += fmt::format("Label_{}:\n", ast.index);
     }
 
     void operator()(const ASTGoto& ast) {
         ExprPrinter expr_parser{};
         std::visit(expr_parser, *ast.condition);
-        inner += Ident() + "(" + expr_parser.GetResult() + ") -> goto Label_" +
-                 std::to_string(ast.label) + ";\n";
+        inner +=
+            fmt::format("{}({}) -> goto Label_{};\n", Indent(), expr_parser.GetResult(), ast.label);
     }
 
     void operator()(const ASTDoWhile& ast) {
         ExprPrinter expr_parser{};
         std::visit(expr_parser, *ast.condition);
-        inner += Ident() + "do {\n";
+        inner += fmt::format("{}do {{\n", Indent());
         scope++;
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
@@ -310,32 +321,23 @@ public:
             current = current->GetNext();
         }
         scope--;
-        inner += Ident() + "} while (" + expr_parser.GetResult() + ");\n";
+        inner += fmt::format("{}}} while ({});\n", Indent(), expr_parser.GetResult());
     }
 
     void operator()(const ASTReturn& ast) {
         ExprPrinter expr_parser{};
         std::visit(expr_parser, *ast.condition);
-        inner += Ident() + "(" + expr_parser.GetResult() + ") -> " +
-                 (ast.kills ? "discard" : "exit") + ";\n";
+        inner += fmt::format("{}({}) -> {};\n", Indent(), expr_parser.GetResult(),
+                             ast.kills ? "discard" : "exit");
     }
 
     void operator()(const ASTBreak& ast) {
         ExprPrinter expr_parser{};
         std::visit(expr_parser, *ast.condition);
-        inner += Ident() + "(" + expr_parser.GetResult() + ") -> break;\n";
+        inner += fmt::format("{}({}) -> break;\n", Indent(), expr_parser.GetResult());
     }
 
-    std::string& Ident() {
-        if (memo_scope == scope) {
-            return tabs_memo;
-        }
-        tabs_memo = tabs.substr(0, scope * 2);
-        memo_scope = scope;
-        return tabs_memo;
-    }
-
-    void Visit(ASTNode& node) {
+    void Visit(const ASTNode& node) {
         std::visit(*this, *node->GetInnerData());
     }
 
@@ -344,16 +346,29 @@ public:
     }
 
 private:
+    std::string_view Indent() {
+        if (space_segment_scope == scope) {
+            return space_segment;
+        }
+
+        // Ensure that we don't exceed our view.
+        ASSERT(scope * 2 < spaces.size());
+
+        space_segment = spaces.substr(0, scope * 2);
+        space_segment_scope = scope;
+        return space_segment;
+    }
+
     std::string inner{};
-    u32 scope{};
+    std::string_view space_segment;
 
-    std::string tabs_memo{};
-    u32 memo_scope{};
+    u32 scope{};
+    u32 space_segment_scope{};
 
-    static constexpr std::string_view tabs{"                                    "};
+    static constexpr std::string_view spaces{"                                    "};
 };
 
-std::string ASTManager::Print() {
+std::string ASTManager::Print() const {
     ASTPrinter printer{};
     printer.Visit(main_node);
     return printer.GetResult();
@@ -549,13 +564,13 @@ bool ASTManager::DirectlyRelated(const ASTNode& first, const ASTNode& second) co
     return min->GetParent() == max->GetParent();
 }
 
-void ASTManager::ShowCurrentState(std::string_view state) {
+void ASTManager::ShowCurrentState(std::string_view state) const {
     LOG_CRITICAL(HW_GPU, "\nState {}:\n\n{}\n", state, Print());
     SanityCheck();
 }
 
-void ASTManager::SanityCheck() {
-    for (auto& label : labels) {
+void ASTManager::SanityCheck() const {
+    for (const auto& label : labels) {
         if (!label->GetParent()) {
             LOG_CRITICAL(HW_GPU, "Sanity Check Failed");
         }
diff --git a/src/video_core/shader/ast.h b/src/video_core/shader/ast.h
index d7bf11821..a2f0044ba 100644
--- a/src/video_core/shader/ast.h
+++ b/src/video_core/shader/ast.h
@@ -328,13 +328,13 @@ public:
 
     void InsertReturn(Expr condition, bool kills);
 
-    std::string Print();
+    std::string Print() const;
 
     void Decompile();
 
-    void ShowCurrentState(std::string_view state);
+    void ShowCurrentState(std::string_view state) const;
 
-    void SanityCheck();
+    void SanityCheck() const;
 
     void Clear();
 
diff --git a/src/video_core/shader/const_buffer_locker.cpp b/src/video_core/shader/const_buffer_locker.cpp
new file mode 100644
index 000000000..fe467608e
--- /dev/null
+++ b/src/video_core/shader/const_buffer_locker.cpp
@@ -0,0 +1,110 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <algorithm>
+#include <memory>
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/shader/const_buffer_locker.h"
+
+namespace VideoCommon::Shader {
+
+using Tegra::Engines::SamplerDescriptor;
+
+ConstBufferLocker::ConstBufferLocker(Tegra::Engines::ShaderType shader_stage)
+    : stage{shader_stage} {}
+
+ConstBufferLocker::ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
+                                     Tegra::Engines::ConstBufferEngineInterface& engine)
+    : stage{shader_stage}, engine{&engine} {}
+
+ConstBufferLocker::~ConstBufferLocker() = default;
+
+std::optional<u32> ConstBufferLocker::ObtainKey(u32 buffer, u32 offset) {
+    const std::pair<u32, u32> key = {buffer, offset};
+    const auto iter = keys.find(key);
+    if (iter != keys.end()) {
+        return iter->second;
+    }
+    if (!engine) {
+        return std::nullopt;
+    }
+    const u32 value = engine->AccessConstBuffer32(stage, buffer, offset);
+    keys.emplace(key, value);
+    return value;
+}
+
+std::optional<SamplerDescriptor> ConstBufferLocker::ObtainBoundSampler(u32 offset) {
+    const u32 key = offset;
+    const auto iter = bound_samplers.find(key);
+    if (iter != bound_samplers.end()) {
+        return iter->second;
+    }
+    if (!engine) {
+        return std::nullopt;
+    }
+    const SamplerDescriptor value = engine->AccessBoundSampler(stage, offset);
+    bound_samplers.emplace(key, value);
+    return value;
+}
+
+std::optional<Tegra::Engines::SamplerDescriptor> ConstBufferLocker::ObtainBindlessSampler(
+    u32 buffer, u32 offset) {
+    const std::pair key = {buffer, offset};
+    const auto iter = bindless_samplers.find(key);
+    if (iter != bindless_samplers.end()) {
+        return iter->second;
+    }
+    if (!engine) {
+        return std::nullopt;
+    }
+    const SamplerDescriptor value = engine->AccessBindlessSampler(stage, buffer, offset);
+    bindless_samplers.emplace(key, value);
+    return value;
+}
+
+void ConstBufferLocker::InsertKey(u32 buffer, u32 offset, u32 value) {
+    keys.insert_or_assign({buffer, offset}, value);
+}
+
+void ConstBufferLocker::InsertBoundSampler(u32 offset, SamplerDescriptor sampler) {
+    bound_samplers.insert_or_assign(offset, sampler);
+}
+
+void ConstBufferLocker::InsertBindlessSampler(u32 buffer, u32 offset, SamplerDescriptor sampler) {
+    bindless_samplers.insert_or_assign({buffer, offset}, sampler);
+}
+
+bool ConstBufferLocker::IsConsistent() const {
+    if (!engine) {
+        return false;
+    }
+    return std::all_of(keys.begin(), keys.end(),
+                       [this](const auto& pair) {
+                           const auto [cbuf, offset] = pair.first;
+                           const auto value = pair.second;
+                           return value == engine->AccessConstBuffer32(stage, cbuf, offset);
+                       }) &&
+           std::all_of(bound_samplers.begin(), bound_samplers.end(),
+                       [this](const auto& sampler) {
+                           const auto [key, value] = sampler;
+                           return value == engine->AccessBoundSampler(stage, key);
+                       }) &&
+           std::all_of(bindless_samplers.begin(), bindless_samplers.end(),
+                       [this](const auto& sampler) {
+                           const auto [cbuf, offset] = sampler.first;
+                           const auto value = sampler.second;
+                           return value == engine->AccessBindlessSampler(stage, cbuf, offset);
+                       });
+}
+
+bool ConstBufferLocker::HasEqualKeys(const ConstBufferLocker& rhs) const {
+    return keys == rhs.keys && bound_samplers == rhs.bound_samplers &&
+           bindless_samplers == rhs.bindless_samplers;
+}
+
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/const_buffer_locker.h b/src/video_core/shader/const_buffer_locker.h
new file mode 100644
index 000000000..600e2f3c3
--- /dev/null
+++ b/src/video_core/shader/const_buffer_locker.h
@@ -0,0 +1,80 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <unordered_map>
+#include "common/common_types.h"
+#include "common/hash.h"
+#include "video_core/engines/const_buffer_engine_interface.h"
+
+namespace VideoCommon::Shader {
+
+using KeyMap = std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>;
+using BoundSamplerMap = std::unordered_map<u32, Tegra::Engines::SamplerDescriptor>;
+using BindlessSamplerMap =
+    std::unordered_map<std::pair<u32, u32>, Tegra::Engines::SamplerDescriptor, Common::PairHash>;
+
+/**
+ * The ConstBufferLocker is a class use to interface the 3D and compute engines with the shader
+ * compiler. with it, the shader can obtain required data from GPU state and store it for disk
+ * shader compilation.
+ **/
+class ConstBufferLocker {
+public:
+    explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage);
+
+    explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
+                               Tegra::Engines::ConstBufferEngineInterface& engine);
+
+    ~ConstBufferLocker();
+
+    /// Retrieves a key from the locker, if it's registered, it will give the registered value, if
+    /// not it will obtain it from maxwell3d and register it.
+    std::optional<u32> ObtainKey(u32 buffer, u32 offset);
+
+    std::optional<Tegra::Engines::SamplerDescriptor> ObtainBoundSampler(u32 offset);
+
+    std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset);
+
+    /// Inserts a key.
+    void InsertKey(u32 buffer, u32 offset, u32 value);
+
+    /// Inserts a bound sampler key.
+    void InsertBoundSampler(u32 offset, Tegra::Engines::SamplerDescriptor sampler);
+
+    /// Inserts a bindless sampler key.
+    void InsertBindlessSampler(u32 buffer, u32 offset, Tegra::Engines::SamplerDescriptor sampler);
+
+    /// Checks keys and samplers against engine's current const buffers. Returns true if they are
+    /// the same value, false otherwise;
+    bool IsConsistent() const;
+
+    /// Returns true if the keys are equal to the other ones in the locker.
+    bool HasEqualKeys(const ConstBufferLocker& rhs) const;
+
+    /// Gives an getter to the const buffer keys in the database.
+    const KeyMap& GetKeys() const {
+        return keys;
+    }
+
+    /// Gets samplers database.
+    const BoundSamplerMap& GetBoundSamplers() const {
+        return bound_samplers;
+    }
+
+    /// Gets bindless samplers database.
+    const BindlessSamplerMap& GetBindlessSamplers() const {
+        return bindless_samplers;
+    }
+
+private:
+    const Tegra::Engines::ShaderType stage;
+    Tegra::Engines::ConstBufferEngineInterface* engine = nullptr;
+    KeyMap keys;
+    BoundSamplerMap bound_samplers;
+    BindlessSamplerMap bindless_samplers;
+};
+
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/control_flow.cpp b/src/video_core/shader/control_flow.cpp
index 268d1aed0..d47c63d9f 100644
--- a/src/video_core/shader/control_flow.cpp
+++ b/src/video_core/shader/control_flow.cpp
@@ -35,14 +35,20 @@ struct BlockStack {
     std::stack<u32> pbk_stack{};
 };
 
-struct BlockBranchInfo {
-    Condition condition{};
-    s32 address{exit_branch};
-    bool kill{};
-    bool is_sync{};
-    bool is_brk{};
-    bool ignore{};
-};
+template <typename T, typename... Args>
+BlockBranchInfo MakeBranchInfo(Args&&... args) {
+    static_assert(std::is_convertible_v<T, BranchData>);
+    return std::make_shared<BranchData>(T(std::forward<Args>(args)...));
+}
+
+bool BlockBranchIsIgnored(BlockBranchInfo first) {
+    bool ignore = false;
+    if (std::holds_alternative<SingleBranch>(*first)) {
+        const auto branch = std::get_if<SingleBranch>(first.get());
+        ignore = branch->ignore;
+    }
+    return ignore;
+}
 
 struct BlockInfo {
     u32 start{};
@@ -56,10 +62,11 @@ struct BlockInfo {
 };
 
 struct CFGRebuildState {
-    explicit CFGRebuildState(const ProgramCode& program_code, const std::size_t program_size,
-                             const u32 start)
-        : start{start}, program_code{program_code}, program_size{program_size} {}
+    explicit CFGRebuildState(const ProgramCode& program_code, u32 start, ConstBufferLocker& locker)
+        : program_code{program_code}, start{start}, locker{locker} {}
 
+    const ProgramCode& program_code;
+    ConstBufferLocker& locker;
     u32 start{};
     std::vector<BlockInfo> block_info{};
     std::list<u32> inspect_queries{};
@@ -69,8 +76,6 @@ struct CFGRebuildState {
     std::map<u32, u32> ssy_labels{};
     std::map<u32, u32> pbk_labels{};
     std::unordered_map<u32, BlockStack> stacks{};
-    const ProgramCode& program_code;
-    const std::size_t program_size;
     ASTManager* manager;
 };
 
@@ -124,10 +129,116 @@ enum class ParseResult : u32 {
     AbnormalFlow,
 };
 
+struct BranchIndirectInfo {
+    u32 buffer{};
+    u32 offset{};
+    u32 entries{};
+    s32 relative_position{};
+};
+
+std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState& state,
+                                                          u32 start_address, u32 current_position) {
+    const u32 shader_start = state.start;
+    u32 pos = current_position;
+    BranchIndirectInfo result{};
+    u64 track_register = 0;
+
+    // Step 0 Get BRX Info
+    const Instruction instr = {state.program_code[pos]};
+    const auto opcode = OpCode::Decode(instr);
+    if (opcode->get().GetId() != OpCode::Id::BRX) {
+        return std::nullopt;
+    }
+    if (instr.brx.constant_buffer != 0) {
+        return std::nullopt;
+    }
+    track_register = instr.gpr8.Value();
+    result.relative_position = instr.brx.GetBranchExtend();
+    pos--;
+    bool found_track = false;
+
+    // Step 1 Track LDC
+    while (pos >= shader_start) {
+        if (IsSchedInstruction(pos, shader_start)) {
+            pos--;
+            continue;
+        }
+        const Instruction instr = {state.program_code[pos]};
+        const auto opcode = OpCode::Decode(instr);
+        if (opcode->get().GetId() == OpCode::Id::LD_C) {
+            if (instr.gpr0.Value() == track_register &&
+                instr.ld_c.type.Value() == Tegra::Shader::UniformType::Single) {
+                result.buffer = instr.cbuf36.index.Value();
+                result.offset = static_cast<u32>(instr.cbuf36.GetOffset());
+                track_register = instr.gpr8.Value();
+                pos--;
+                found_track = true;
+                break;
+            }
+        }
+        pos--;
+    }
+
+    if (!found_track) {
+        return std::nullopt;
+    }
+    found_track = false;
+
+    // Step 2 Track SHL
+    while (pos >= shader_start) {
+        if (IsSchedInstruction(pos, shader_start)) {
+            pos--;
+            continue;
+        }
+        const Instruction instr = state.program_code[pos];
+        const auto opcode = OpCode::Decode(instr);
+        if (opcode->get().GetId() == OpCode::Id::SHL_IMM) {
+            if (instr.gpr0.Value() == track_register) {
+                track_register = instr.gpr8.Value();
+                pos--;
+                found_track = true;
+                break;
+            }
+        }
+        pos--;
+    }
+
+    if (!found_track) {
+        return std::nullopt;
+    }
+    found_track = false;
+
+    // Step 3 Track IMNMX
+    while (pos >= shader_start) {
+        if (IsSchedInstruction(pos, shader_start)) {
+            pos--;
+            continue;
+        }
+        const Instruction instr = state.program_code[pos];
+        const auto opcode = OpCode::Decode(instr);
+        if (opcode->get().GetId() == OpCode::Id::IMNMX_IMM) {
+            if (instr.gpr0.Value() == track_register) {
+                track_register = instr.gpr8.Value();
+                result.entries = instr.alu.GetSignedImm20_20() + 1;
+                pos--;
+                found_track = true;
+                break;
+            }
+        }
+        pos--;
+    }
+
+    if (!found_track) {
+        return std::nullopt;
+    }
+    return result;
+}
+
 std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address) {
     u32 offset = static_cast<u32>(address);
-    const u32 end_address = static_cast<u32>(state.program_size / sizeof(Instruction));
+    const u32 end_address = static_cast<u32>(state.program_code.size());
     ParseInfo parse_info{};
+    SingleBranch single_branch{};
 
     const auto insert_label = [](CFGRebuildState& state, u32 address) {
         const auto pair = state.labels.emplace(address);
@@ -140,13 +251,14 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
         if (offset >= end_address) {
             // ASSERT_OR_EXECUTE can't be used, as it ignores the break
             ASSERT_MSG(false, "Shader passed the current limit!");
-            parse_info.branch_info.address = exit_branch;
-            parse_info.branch_info.ignore = false;
+
+            single_branch.address = exit_branch;
+            single_branch.ignore = false;
             break;
         }
         if (state.registered.count(offset) != 0) {
-            parse_info.branch_info.address = offset;
-            parse_info.branch_info.ignore = true;
+            single_branch.address = offset;
+            single_branch.ignore = true;
             break;
         }
         if (IsSchedInstruction(offset, state.start)) {
@@ -163,24 +275,26 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
         switch (opcode->get().GetId()) {
         case OpCode::Id::EXIT: {
             const auto pred_index = static_cast<u32>(instr.pred.pred_index);
-            parse_info.branch_info.condition.predicate =
-                GetPredicate(pred_index, instr.negate_pred != 0);
-            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+            single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
+            if (single_branch.condition.predicate == Pred::NeverExecute) {
                 offset++;
                 continue;
             }
             const ConditionCode cc = instr.flow_condition_code;
-            parse_info.branch_info.condition.cc = cc;
+            single_branch.condition.cc = cc;
             if (cc == ConditionCode::F) {
                 offset++;
                 continue;
             }
-            parse_info.branch_info.address = exit_branch;
-            parse_info.branch_info.kill = false;
-            parse_info.branch_info.is_sync = false;
-            parse_info.branch_info.is_brk = false;
-            parse_info.branch_info.ignore = false;
+            single_branch.address = exit_branch;
+            single_branch.kill = false;
+            single_branch.is_sync = false;
+            single_branch.is_brk = false;
+            single_branch.ignore = false;
             parse_info.end_address = offset;
+            parse_info.branch_info = MakeBranchInfo<SingleBranch>(
+                single_branch.condition, single_branch.address, single_branch.kill,
+                single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
 
             return {ParseResult::ControlCaught, parse_info};
         }
@@ -189,99 +303,107 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
                 return {ParseResult::AbnormalFlow, parse_info};
             }
             const auto pred_index = static_cast<u32>(instr.pred.pred_index);
-            parse_info.branch_info.condition.predicate =
-                GetPredicate(pred_index, instr.negate_pred != 0);
-            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+            single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
+            if (single_branch.condition.predicate == Pred::NeverExecute) {
                 offset++;
                 continue;
             }
             const ConditionCode cc = instr.flow_condition_code;
-            parse_info.branch_info.condition.cc = cc;
+            single_branch.condition.cc = cc;
             if (cc == ConditionCode::F) {
                 offset++;
                 continue;
             }
             const u32 branch_offset = offset + instr.bra.GetBranchTarget();
             if (branch_offset == 0) {
-                parse_info.branch_info.address = exit_branch;
+                single_branch.address = exit_branch;
             } else {
-                parse_info.branch_info.address = branch_offset;
+                single_branch.address = branch_offset;
             }
             insert_label(state, branch_offset);
-            parse_info.branch_info.kill = false;
-            parse_info.branch_info.is_sync = false;
-            parse_info.branch_info.is_brk = false;
-            parse_info.branch_info.ignore = false;
+            single_branch.kill = false;
+            single_branch.is_sync = false;
+            single_branch.is_brk = false;
+            single_branch.ignore = false;
             parse_info.end_address = offset;
+            parse_info.branch_info = MakeBranchInfo<SingleBranch>(
+                single_branch.condition, single_branch.address, single_branch.kill,
+                single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
 
             return {ParseResult::ControlCaught, parse_info};
         }
         case OpCode::Id::SYNC: {
             const auto pred_index = static_cast<u32>(instr.pred.pred_index);
-            parse_info.branch_info.condition.predicate =
-                GetPredicate(pred_index, instr.negate_pred != 0);
-            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+            single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
+            if (single_branch.condition.predicate == Pred::NeverExecute) {
                 offset++;
                 continue;
             }
             const ConditionCode cc = instr.flow_condition_code;
-            parse_info.branch_info.condition.cc = cc;
+            single_branch.condition.cc = cc;
             if (cc == ConditionCode::F) {
                 offset++;
                 continue;
             }
-            parse_info.branch_info.address = unassigned_branch;
-            parse_info.branch_info.kill = false;
-            parse_info.branch_info.is_sync = true;
-            parse_info.branch_info.is_brk = false;
-            parse_info.branch_info.ignore = false;
+            single_branch.address = unassigned_branch;
+            single_branch.kill = false;
+            single_branch.is_sync = true;
+            single_branch.is_brk = false;
+            single_branch.ignore = false;
             parse_info.end_address = offset;
+            parse_info.branch_info = MakeBranchInfo<SingleBranch>(
+                single_branch.condition, single_branch.address, single_branch.kill,
+                single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
 
             return {ParseResult::ControlCaught, parse_info};
         }
         case OpCode::Id::BRK: {
             const auto pred_index = static_cast<u32>(instr.pred.pred_index);
-            parse_info.branch_info.condition.predicate =
-                GetPredicate(pred_index, instr.negate_pred != 0);
-            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+            single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
+            if (single_branch.condition.predicate == Pred::NeverExecute) {
                 offset++;
                 continue;
             }
             const ConditionCode cc = instr.flow_condition_code;
-            parse_info.branch_info.condition.cc = cc;
+            single_branch.condition.cc = cc;
             if (cc == ConditionCode::F) {
                 offset++;
                 continue;
             }
-            parse_info.branch_info.address = unassigned_branch;
-            parse_info.branch_info.kill = false;
-            parse_info.branch_info.is_sync = false;
-            parse_info.branch_info.is_brk = true;
-            parse_info.branch_info.ignore = false;
+            single_branch.address = unassigned_branch;
+            single_branch.kill = false;
+            single_branch.is_sync = false;
+            single_branch.is_brk = true;
+            single_branch.ignore = false;
             parse_info.end_address = offset;
+            parse_info.branch_info = MakeBranchInfo<SingleBranch>(
+                single_branch.condition, single_branch.address, single_branch.kill,
+                single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
 
             return {ParseResult::ControlCaught, parse_info};
         }
         case OpCode::Id::KIL: {
             const auto pred_index = static_cast<u32>(instr.pred.pred_index);
-            parse_info.branch_info.condition.predicate =
-                GetPredicate(pred_index, instr.negate_pred != 0);
-            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+            single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
+            if (single_branch.condition.predicate == Pred::NeverExecute) {
                 offset++;
                 continue;
             }
             const ConditionCode cc = instr.flow_condition_code;
-            parse_info.branch_info.condition.cc = cc;
+            single_branch.condition.cc = cc;
             if (cc == ConditionCode::F) {
                 offset++;
                 continue;
             }
-            parse_info.branch_info.address = exit_branch;
-            parse_info.branch_info.kill = true;
-            parse_info.branch_info.is_sync = false;
-            parse_info.branch_info.is_brk = false;
-            parse_info.branch_info.ignore = false;
+            single_branch.address = exit_branch;
+            single_branch.kill = true;
+            single_branch.is_sync = false;
+            single_branch.is_brk = false;
+            single_branch.ignore = false;
             parse_info.end_address = offset;
+            parse_info.branch_info = MakeBranchInfo<SingleBranch>(
+                single_branch.condition, single_branch.address, single_branch.kill,
+                single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
 
             return {ParseResult::ControlCaught, parse_info};
         }
@@ -298,6 +420,29 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
             break;
         }
         case OpCode::Id::BRX: {
+            auto tmp = TrackBranchIndirectInfo(state, address, offset);
+            if (tmp) {
+                auto result = *tmp;
+                std::vector<CaseBranch> branches{};
+                s32 pc_target = offset + result.relative_position;
+                for (u32 i = 0; i < result.entries; i++) {
+                    auto k = state.locker.ObtainKey(result.buffer, result.offset + i * 4);
+                    if (!k) {
+                        return {ParseResult::AbnormalFlow, parse_info};
+                    }
+                    u32 value = *k;
+                    u32 target = static_cast<u32>((value >> 3) + pc_target);
+                    insert_label(state, target);
+                    branches.emplace_back(value, target);
+                }
+                parse_info.end_address = offset;
+                parse_info.branch_info = MakeBranchInfo<MultiBranch>(
+                    static_cast<u32>(instr.gpr8.Value()), std::move(branches));
+
+                return {ParseResult::ControlCaught, parse_info};
+            } else {
+                LOG_WARNING(HW_GPU, "BRX Track Unsuccesful");
+            }
             return {ParseResult::AbnormalFlow, parse_info};
         }
         default:
@@ -306,10 +451,13 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
 
         offset++;
     }
-    parse_info.branch_info.kill = false;
-    parse_info.branch_info.is_sync = false;
-    parse_info.branch_info.is_brk = false;
+    single_branch.kill = false;
+    single_branch.is_sync = false;
+    single_branch.is_brk = false;
     parse_info.end_address = offset - 1;
+    parse_info.branch_info = MakeBranchInfo<SingleBranch>(
+        single_branch.condition, single_branch.address, single_branch.kill, single_branch.is_sync,
+        single_branch.is_brk, single_branch.ignore);
     return {ParseResult::BlockEnd, parse_info};
 }
 
@@ -333,9 +481,10 @@ bool TryInspectAddress(CFGRebuildState& state) {
         BlockInfo& current_block = state.block_info[block_index];
         current_block.end = address - 1;
         new_block.branch = current_block.branch;
-        BlockBranchInfo forward_branch{};
-        forward_branch.address = address;
-        forward_branch.ignore = true;
+        BlockBranchInfo forward_branch = MakeBranchInfo<SingleBranch>();
+        const auto branch = std::get_if<SingleBranch>(forward_branch.get());
+        branch->address = address;
+        branch->ignore = true;
         current_block.branch = forward_branch;
         return true;
     }
@@ -350,12 +499,15 @@ bool TryInspectAddress(CFGRebuildState& state) {
 
     BlockInfo& block_info = CreateBlockInfo(state, address, parse_info.end_address);
     block_info.branch = parse_info.branch_info;
-    if (parse_info.branch_info.condition.IsUnconditional()) {
+    if (std::holds_alternative<SingleBranch>(*block_info.branch)) {
+        const auto branch = std::get_if<SingleBranch>(block_info.branch.get());
+        if (branch->condition.IsUnconditional()) {
+            return true;
+        }
+        const u32 fallthrough_address = parse_info.end_address + 1;
+        state.inspect_queries.push_front(fallthrough_address);
         return true;
     }
-
-    const u32 fallthrough_address = parse_info.end_address + 1;
-    state.inspect_queries.push_front(fallthrough_address);
     return true;
 }
 
@@ -393,31 +545,42 @@ bool TryQuery(CFGRebuildState& state) {
     state.queries.pop_front();
     gather_labels(q2.ssy_stack, state.ssy_labels, block);
     gather_labels(q2.pbk_stack, state.pbk_labels, block);
-    if (!block.branch.condition.IsUnconditional()) {
-        q2.address = block.end + 1;
-        state.queries.push_back(q2);
-    }
+    if (std::holds_alternative<SingleBranch>(*block.branch)) {
+        const auto branch = std::get_if<SingleBranch>(block.branch.get());
+        if (!branch->condition.IsUnconditional()) {
+            q2.address = block.end + 1;
+            state.queries.push_back(q2);
+        }
 
-    Query conditional_query{q2};
-    if (block.branch.is_sync) {
-        if (block.branch.address == unassigned_branch) {
-            block.branch.address = conditional_query.ssy_stack.top();
+        Query conditional_query{q2};
+        if (branch->is_sync) {
+            if (branch->address == unassigned_branch) {
+                branch->address = conditional_query.ssy_stack.top();
+            }
+            conditional_query.ssy_stack.pop();
         }
-        conditional_query.ssy_stack.pop();
-    }
-    if (block.branch.is_brk) {
-        if (block.branch.address == unassigned_branch) {
-            block.branch.address = conditional_query.pbk_stack.top();
+        if (branch->is_brk) {
+            if (branch->address == unassigned_branch) {
+                branch->address = conditional_query.pbk_stack.top();
+            }
+            conditional_query.pbk_stack.pop();
         }
-        conditional_query.pbk_stack.pop();
+        conditional_query.address = branch->address;
+        state.queries.push_back(std::move(conditional_query));
+        return true;
+    }
+    const auto multi_branch = std::get_if<MultiBranch>(block.branch.get());
+    for (const auto& branch_case : multi_branch->branches) {
+        Query conditional_query{q2};
+        conditional_query.address = branch_case.address;
+        state.queries.push_back(std::move(conditional_query));
     }
-    conditional_query.address = block.branch.address;
-    state.queries.push_back(std::move(conditional_query));
     return true;
 }
+
 } // Anonymous namespace
 
-void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch) {
+void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch_info) {
     const auto get_expr = ([&](const Condition& cond) -> Expr {
         Expr result{};
         if (cond.cc != ConditionCode::T) {
@@ -444,15 +607,24 @@ void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch) {
         }
         return MakeExpr<ExprBoolean>(true);
     });
-    if (branch.address < 0) {
-        if (branch.kill) {
-            mm.InsertReturn(get_expr(branch.condition), true);
+    if (std::holds_alternative<SingleBranch>(*branch_info)) {
+        const auto branch = std::get_if<SingleBranch>(branch_info.get());
+        if (branch->address < 0) {
+            if (branch->kill) {
+                mm.InsertReturn(get_expr(branch->condition), true);
+                return;
+            }
+            mm.InsertReturn(get_expr(branch->condition), false);
             return;
         }
-        mm.InsertReturn(get_expr(branch.condition), false);
+        mm.InsertGoto(get_expr(branch->condition), branch->address);
         return;
     }
-    mm.InsertGoto(get_expr(branch.condition), branch.address);
+    const auto multi_branch = std::get_if<MultiBranch>(branch_info.get());
+    for (const auto& branch_case : multi_branch->branches) {
+        mm.InsertGoto(MakeExpr<ExprGprEqual>(multi_branch->gpr, branch_case.cmp_value),
+                      branch_case.address);
+    }
 }
 
 void DecompileShader(CFGRebuildState& state) {
@@ -464,25 +636,26 @@ void DecompileShader(CFGRebuildState& state) {
         if (state.labels.count(block.start) != 0) {
             state.manager->InsertLabel(block.start);
         }
-        u32 end = block.branch.ignore ? block.end + 1 : block.end;
+        const bool ignore = BlockBranchIsIgnored(block.branch);
+        u32 end = ignore ? block.end + 1 : block.end;
         state.manager->InsertBlock(block.start, end);
-        if (!block.branch.ignore) {
+        if (!ignore) {
             InsertBranch(*state.manager, block.branch);
         }
     }
     state.manager->Decompile();
 }
 
-std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 program_size,
-                                                u32 start_address,
-                                                const CompilerSettings& settings) {
+std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
+                                                const CompilerSettings& settings,
+                                                ConstBufferLocker& locker) {
     auto result_out = std::make_unique<ShaderCharacteristics>();
     if (settings.depth == CompileDepth::BruteForce) {
         result_out->settings.depth = CompileDepth::BruteForce;
         return result_out;
     }
 
-    CFGRebuildState state{program_code, program_size, start_address};
+    CFGRebuildState state{program_code, start_address, locker};
     // Inspect Code and generate blocks
     state.labels.clear();
     state.labels.emplace(start_address);
@@ -547,11 +720,9 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
         ShaderBlock new_block{};
         new_block.start = block.start;
         new_block.end = block.end;
-        new_block.ignore_branch = block.branch.ignore;
+        new_block.ignore_branch = BlockBranchIsIgnored(block.branch);
         if (!new_block.ignore_branch) {
-            new_block.branch.cond = block.branch.condition;
-            new_block.branch.kills = block.branch.kill;
-            new_block.branch.address = block.branch.address;
+            new_block.branch = block.branch;
         }
         result_out->end = std::max(result_out->end, block.end);
         result_out->blocks.push_back(new_block);
diff --git a/src/video_core/shader/control_flow.h b/src/video_core/shader/control_flow.h
index 74e54a5c7..5304998b9 100644
--- a/src/video_core/shader/control_flow.h
+++ b/src/video_core/shader/control_flow.h
@@ -7,6 +7,7 @@
 #include <list>
 #include <optional>
 #include <set>
+#include <variant>
 
 #include "video_core/engines/shader_bytecode.h"
 #include "video_core/shader/ast.h"
@@ -37,29 +38,61 @@ struct Condition {
     }
 };
 
-struct ShaderBlock {
-    struct Branch {
-        Condition cond{};
-        bool kills{};
-        s32 address{};
+class SingleBranch {
+public:
+    SingleBranch() = default;
+    SingleBranch(Condition condition, s32 address, bool kill, bool is_sync, bool is_brk,
+                 bool ignore)
+        : condition{condition}, address{address}, kill{kill}, is_sync{is_sync}, is_brk{is_brk},
+          ignore{ignore} {}
+
+    bool operator==(const SingleBranch& b) const {
+        return std::tie(condition, address, kill, is_sync, is_brk, ignore) ==
+               std::tie(b.condition, b.address, b.kill, b.is_sync, b.is_brk, b.ignore);
+    }
+
+    bool operator!=(const SingleBranch& b) const {
+        return !operator==(b);
+    }
+
+    Condition condition{};
+    s32 address{exit_branch};
+    bool kill{};
+    bool is_sync{};
+    bool is_brk{};
+    bool ignore{};
+};
 
-        bool operator==(const Branch& b) const {
-            return std::tie(cond, kills, address) == std::tie(b.cond, b.kills, b.address);
-        }
+struct CaseBranch {
+    CaseBranch(u32 cmp_value, u32 address) : cmp_value{cmp_value}, address{address} {}
+    u32 cmp_value;
+    u32 address;
+};
+
+class MultiBranch {
+public:
+    MultiBranch(u32 gpr, std::vector<CaseBranch>&& branches)
+        : gpr{gpr}, branches{std::move(branches)} {}
+
+    u32 gpr{};
+    std::vector<CaseBranch> branches{};
+};
+
+using BranchData = std::variant<SingleBranch, MultiBranch>;
+using BlockBranchInfo = std::shared_ptr<BranchData>;
 
-        bool operator!=(const Branch& b) const {
-            return !operator==(b);
-        }
-    };
+bool BlockBranchInfoAreEqual(BlockBranchInfo first, BlockBranchInfo second);
 
+struct ShaderBlock {
     u32 start{};
     u32 end{};
     bool ignore_branch{};
-    Branch branch{};
+    BlockBranchInfo branch{};
 
     bool operator==(const ShaderBlock& sb) const {
-        return std::tie(start, end, ignore_branch, branch) ==
-               std::tie(sb.start, sb.end, sb.ignore_branch, sb.branch);
+        return std::tie(start, end, ignore_branch) ==
+                   std::tie(sb.start, sb.end, sb.ignore_branch) &&
+               BlockBranchInfoAreEqual(branch, sb.branch);
     }
 
     bool operator!=(const ShaderBlock& sb) const {
@@ -76,8 +109,8 @@ struct ShaderCharacteristics {
     CompilerSettings settings{};
 };
 
-std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 program_size,
-                                                u32 start_address,
-                                                const CompilerSettings& settings);
+std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
+                                                const CompilerSettings& settings,
+                                                ConstBufferLocker& locker);
 
 } // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode.cpp b/src/video_core/shader/decode.cpp
index 2626b1616..21fb9cb83 100644
--- a/src/video_core/shader/decode.cpp
+++ b/src/video_core/shader/decode.cpp
@@ -33,7 +33,7 @@ constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
     return (absolute_offset % SchedPeriod) == 0;
 }
 
-} // namespace
+} // Anonymous namespace
 
 class ASTDecoder {
 public:
@@ -102,7 +102,7 @@ void ShaderIR::Decode() {
     std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
 
     decompiled = false;
-    auto info = ScanFlow(program_code, program_size, main_offset, settings);
+    auto info = ScanFlow(program_code, main_offset, settings, locker);
     auto& shader_info = *info;
     coverage_begin = shader_info.start;
     coverage_end = shader_info.end;
@@ -155,7 +155,7 @@ void ShaderIR::Decode() {
         [[fallthrough]];
     case CompileDepth::BruteForce: {
         coverage_begin = main_offset;
-        const u32 shader_end = static_cast<u32>(program_size / sizeof(u64));
+        const std::size_t shader_end = program_code.size();
         coverage_end = shader_end;
         for (u32 label = main_offset; label < shader_end; label++) {
             basic_blocks.insert({label, DecodeRange(label, label + 1)});
@@ -198,24 +198,39 @@ void ShaderIR::InsertControlFlow(NodeBlock& bb, const ShaderBlock& block) {
         }
         return result;
     };
-    if (block.branch.address < 0) {
-        if (block.branch.kills) {
-            Node n = Operation(OperationCode::Discard);
-            n = apply_conditions(block.branch.cond, n);
+    if (std::holds_alternative<SingleBranch>(*block.branch)) {
+        auto branch = std::get_if<SingleBranch>(block.branch.get());
+        if (branch->address < 0) {
+            if (branch->kill) {
+                Node n = Operation(OperationCode::Discard);
+                n = apply_conditions(branch->condition, n);
+                bb.push_back(n);
+                global_code.push_back(n);
+                return;
+            }
+            Node n = Operation(OperationCode::Exit);
+            n = apply_conditions(branch->condition, n);
             bb.push_back(n);
             global_code.push_back(n);
             return;
         }
-        Node n = Operation(OperationCode::Exit);
-        n = apply_conditions(block.branch.cond, n);
+        Node n = Operation(OperationCode::Branch, Immediate(branch->address));
+        n = apply_conditions(branch->condition, n);
         bb.push_back(n);
         global_code.push_back(n);
         return;
     }
-    Node n = Operation(OperationCode::Branch, Immediate(block.branch.address));
-    n = apply_conditions(block.branch.cond, n);
-    bb.push_back(n);
-    global_code.push_back(n);
+    auto multi_branch = std::get_if<MultiBranch>(block.branch.get());
+    Node op_a = GetRegister(multi_branch->gpr);
+    for (auto& branch_case : multi_branch->branches) {
+        Node n = Operation(OperationCode::Branch, Immediate(branch_case.address));
+        Node op_b = Immediate(branch_case.cmp_value);
+        Node condition =
+            GetPredicateComparisonInteger(Tegra::Shader::PredCondition::Equal, false, op_a, op_b);
+        auto result = Conditional(condition, {n});
+        bb.push_back(result);
+        global_code.push_back(result);
+    }
 }
 
 u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
diff --git a/src/video_core/shader/decode/arithmetic_integer.cpp b/src/video_core/shader/decode/arithmetic_integer.cpp
index b73f6536e..a33d242e9 100644
--- a/src/video_core/shader/decode/arithmetic_integer.cpp
+++ b/src/video_core/shader/decode/arithmetic_integer.cpp
@@ -144,7 +144,7 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
     case OpCode::Id::ICMP_IMM: {
         const Node zero = Immediate(0);
 
-        const auto [op_b, test] = [&]() -> std::pair<Node, Node> {
+        const auto [op_rhs, test] = [&]() -> std::pair<Node, Node> {
             switch (opcode->get().GetId()) {
             case OpCode::Id::ICMP_CR:
                 return {GetConstBuffer(instr.cbuf34.index, instr.cbuf34.offset),
@@ -161,10 +161,10 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
                 return {zero, zero};
             }
         }();
-        const Node op_a = GetRegister(instr.gpr8);
+        const Node op_lhs = GetRegister(instr.gpr8);
         const Node comparison =
             GetPredicateComparisonInteger(instr.icmp.cond, instr.icmp.is_signed != 0, test, zero);
-        SetRegister(bb, instr.gpr0, Operation(OperationCode::Select, comparison, op_a, op_b));
+        SetRegister(bb, instr.gpr0, Operation(OperationCode::Select, comparison, op_lhs, op_rhs));
         break;
     }
     case OpCode::Id::LOP_C:
diff --git a/src/video_core/shader/decode/image.cpp b/src/video_core/shader/decode/image.cpp
index 95ec1cdd9..b02d2cb95 100644
--- a/src/video_core/shader/decode/image.cpp
+++ b/src/video_core/shader/decode/image.cpp
@@ -144,8 +144,8 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
 
 Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type) {
     const auto offset{static_cast<std::size_t>(image.index.Value())};
-    if (const auto image = TryUseExistingImage(offset, type)) {
-        return *image;
+    if (const auto existing_image = TryUseExistingImage(offset, type)) {
+        return *existing_image;
     }
 
     const std::size_t next_index{used_images.size()};
diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index 7923d4d69..335d78146 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -166,9 +166,17 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }();
 
         const auto [real_address_base, base_address, descriptor] =
-            TrackAndGetGlobalMemory(bb, instr, false);
+            TrackGlobalMemory(bb, instr, false);
 
         const u32 count = GetUniformTypeElementsCount(type);
+        if (!real_address_base || !base_address) {
+            // Tracking failed, load zeroes.
+            for (u32 i = 0; i < count; ++i) {
+                SetRegister(bb, instr.gpr0.Value() + i, Immediate(0.0f));
+            }
+            break;
+        }
+
         for (u32 i = 0; i < count; ++i) {
             const Node it_offset = Immediate(i * 4);
             const Node real_address =
@@ -260,22 +268,19 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }();
 
         const auto [real_address_base, base_address, descriptor] =
-            TrackAndGetGlobalMemory(bb, instr, true);
-
-        // Encode in temporary registers like this: real_base_address, {registers_to_be_written...}
-        SetTemporary(bb, 0, real_address_base);
+            TrackGlobalMemory(bb, instr, true);
+        if (!real_address_base || !base_address) {
+            // Tracking failed, skip the store.
+            break;
+        }
 
         const u32 count = GetUniformTypeElementsCount(type);
         for (u32 i = 0; i < count; ++i) {
-            SetTemporary(bb, i + 1, GetRegister(instr.gpr0.Value() + i));
-        }
-        for (u32 i = 0; i < count; ++i) {
             const Node it_offset = Immediate(i * 4);
-            const Node real_address =
-                Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
+            const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);
             const Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
-
-            bb.push_back(Operation(OperationCode::Assign, gmem, GetTemporary(i + 1)));
+            const Node value = GetRegister(instr.gpr0.Value() + i);
+            bb.push_back(Operation(OperationCode::Assign, gmem, value));
         }
         break;
     }
@@ -301,15 +306,17 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
     return pc;
 }
 
-std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackAndGetGlobalMemory(NodeBlock& bb,
-                                                                           Instruction instr,
-                                                                           bool is_write) {
+std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackGlobalMemory(NodeBlock& bb,
+                                                                     Instruction instr,
+                                                                     bool is_write) {
     const auto addr_register{GetRegister(instr.gmem.gpr)};
     const auto immediate_offset{static_cast<u32>(instr.gmem.offset)};
 
     const auto [base_address, index, offset] =
         TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()));
-    ASSERT(base_address != nullptr);
+    ASSERT_OR_EXECUTE_MSG(base_address != nullptr,
+                          { return std::make_tuple(nullptr, nullptr, GlobalMemoryBase{}); },
+                          "Global memory tracking failed");
 
     bb.push_back(Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", index, offset)));
 
diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp
index d46e0f823..116b95f76 100644
--- a/src/video_core/shader/decode/other.cpp
+++ b/src/video_core/shader/decode/other.cpp
@@ -67,7 +67,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::MOV_SYS: {
-        const Node value = [&]() {
+        const Node value = [this, instr] {
             switch (instr.sys20) {
             case SystemVariable::Ydirection:
                 return Operation(OperationCode::YNegate);
diff --git a/src/video_core/shader/decode/shift.cpp b/src/video_core/shader/decode/shift.cpp
index f6ee68a54..d419e9c45 100644
--- a/src/video_core/shader/decode/shift.cpp
+++ b/src/video_core/shader/decode/shift.cpp
@@ -18,7 +18,7 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
     const auto opcode = OpCode::Decode(instr);
 
     Node op_a = GetRegister(instr.gpr8);
-    Node op_b = [&]() {
+    Node op_b = [this, instr] {
         if (instr.is_b_imm) {
             return Immediate(instr.alu.GetSignedImm20_20());
         } else if (instr.is_b_gpr) {
diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp
index 0b934a069..d61e656b7 100644
--- a/src/video_core/shader/decode/texture.cpp
+++ b/src/video_core/shader/decode/texture.cpp
@@ -141,7 +141,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
         const Node component = Immediate(static_cast<u32>(instr.tld4s.component));
 
         const auto& sampler =
-            GetSampler(instr.sampler, TextureType::Texture2D, false, depth_compare);
+            GetSampler(instr.sampler, {{TextureType::Texture2D, false, depth_compare}});
 
         Node4 values;
         for (u32 element = 0; element < values.size(); ++element) {
@@ -150,7 +150,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
             values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
         }
 
-        WriteTexsInstructionFloat(bb, instr, values);
+        WriteTexsInstructionFloat(bb, instr, values, true);
         break;
     }
     case OpCode::Id::TXQ_B:
@@ -165,10 +165,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
         // Sadly, not all texture instructions specify the type of texture their sampler
         // uses. This must be fixed at a later instance.
         const auto& sampler =
-            is_bindless
-                ? GetBindlessSampler(instr.gpr8, Tegra::Shader::TextureType::Texture2D, false,
-                                     false)
-                : GetSampler(instr.sampler, Tegra::Shader::TextureType::Texture2D, false, false);
+            is_bindless ? GetBindlessSampler(instr.gpr8, {}) : GetSampler(instr.sampler, {});
 
         u32 indexer = 0;
         switch (instr.txq.query_type) {
@@ -207,9 +204,9 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
 
         auto texture_type = instr.tmml.texture_type.Value();
         const bool is_array = instr.tmml.array != 0;
-        const auto& sampler = is_bindless
-                                  ? GetBindlessSampler(instr.gpr20, texture_type, is_array, false)
-                                  : GetSampler(instr.sampler, texture_type, is_array, false);
+        const auto& sampler =
+            is_bindless ? GetBindlessSampler(instr.gpr20, {{texture_type, is_array, false}})
+                        : GetSampler(instr.sampler, {{texture_type, is_array, false}});
 
         std::vector<Node> coords;
 
@@ -285,9 +282,26 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
     return pc;
 }
 
-const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, TextureType type,
-                                    bool is_array, bool is_shadow) {
-    const auto offset = static_cast<std::size_t>(sampler.index.Value());
+const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
+                                    std::optional<SamplerInfo> sampler_info) {
+    const auto offset = static_cast<u32>(sampler.index.Value());
+
+    Tegra::Shader::TextureType type;
+    bool is_array;
+    bool is_shadow;
+    if (sampler_info) {
+        type = sampler_info->type;
+        is_array = sampler_info->is_array;
+        is_shadow = sampler_info->is_shadow;
+    } else if (auto sampler = locker.ObtainBoundSampler(offset); sampler) {
+        type = sampler->texture_type.Value();
+        is_array = sampler->is_array.Value() != 0;
+        is_shadow = sampler->is_shadow.Value() != 0;
+    } else {
+        type = Tegra::Shader::TextureType::Texture2D;
+        is_array = false;
+        is_shadow = false;
+    }
 
     // If this sampler has already been used, return the existing mapping.
     const auto itr =
@@ -303,15 +317,31 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, Textu
     const std::size_t next_index = used_samplers.size();
     const Sampler entry{offset, next_index, type, is_array, is_shadow};
     return *used_samplers.emplace(entry).first;
-}
+} // namespace VideoCommon::Shader
 
-const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg, TextureType type,
-                                            bool is_array, bool is_shadow) {
+const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg,
+                                            std::optional<SamplerInfo> sampler_info) {
     const Node sampler_register = GetRegister(reg);
     const auto [base_sampler, cbuf_index, cbuf_offset] =
         TrackCbuf(sampler_register, global_code, static_cast<s64>(global_code.size()));
     ASSERT(base_sampler != nullptr);
     const auto cbuf_key = (static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset);
+    Tegra::Shader::TextureType type;
+    bool is_array;
+    bool is_shadow;
+    if (sampler_info) {
+        type = sampler_info->type;
+        is_array = sampler_info->is_array;
+        is_shadow = sampler_info->is_shadow;
+    } else if (auto sampler = locker.ObtainBindlessSampler(cbuf_index, cbuf_offset); sampler) {
+        type = sampler->texture_type.Value();
+        is_array = sampler->is_array.Value() != 0;
+        is_shadow = sampler->is_shadow.Value() != 0;
+    } else {
+        type = Tegra::Shader::TextureType::Texture2D;
+        is_array = false;
+        is_shadow = false;
+    }
 
     // If this sampler has already been used, return the existing mapping.
     const auto itr =
@@ -344,14 +374,14 @@ void ShaderIR::WriteTexInstructionFloat(NodeBlock& bb, Instruction instr, const
     }
 }
 
-void ShaderIR::WriteTexsInstructionFloat(NodeBlock& bb, Instruction instr,
-                                         const Node4& components) {
+void ShaderIR::WriteTexsInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components,
+                                         bool ignore_mask) {
     // TEXS has two destination registers and a swizzle. The first two elements in the swizzle
     // go into gpr0+0 and gpr0+1, and the rest goes into gpr28+0 and gpr28+1
 
     u32 dest_elem = 0;
     for (u32 component = 0; component < 4; ++component) {
-        if (!instr.texs.IsComponentEnabled(component))
+        if (!instr.texs.IsComponentEnabled(component) && !ignore_mask)
             continue;
         SetTemporary(bb, dest_elem++, components[component]);
     }
@@ -411,9 +441,9 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
                              (texture_type == TextureType::TextureCube && is_array && is_shadow),
                          "This method is not supported.");
 
-    const auto& sampler = is_bindless
-                              ? GetBindlessSampler(*bindless_reg, texture_type, is_array, is_shadow)
-                              : GetSampler(instr.sampler, texture_type, is_array, is_shadow);
+    const auto& sampler =
+        is_bindless ? GetBindlessSampler(*bindless_reg, {{texture_type, is_array, is_shadow}})
+                    : GetSampler(instr.sampler, {{texture_type, is_array, is_shadow}});
 
     const bool lod_needed = process_mode == TextureProcessMode::LZ ||
                             process_mode == TextureProcessMode::LL ||
@@ -577,7 +607,7 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
         dc = GetRegister(parameter_register++);
     }
 
-    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, depth_compare);
+    const auto& sampler = GetSampler(instr.sampler, {{texture_type, is_array, depth_compare}});
 
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
@@ -610,7 +640,7 @@ Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
     // const Node aoffi_register{is_aoffi ? GetRegister(gpr20_cursor++) : nullptr};
     // const Node multisample{is_multisample ? GetRegister(gpr20_cursor++) : nullptr};
 
-    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, false);
+    const auto& sampler = GetSampler(instr.sampler, {{texture_type, is_array, false}});
 
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
@@ -646,7 +676,7 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is
     // When lod is used always is in gpr20
     const Node lod = lod_enabled ? GetRegister(instr.gpr20) : Immediate(0);
 
-    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, false);
+    const auto& sampler = GetSampler(instr.sampler, {{texture_type, is_array, false}});
 
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
diff --git a/src/video_core/shader/decode/video.cpp b/src/video_core/shader/decode/video.cpp
index 97fc6f9b1..b047cf870 100644
--- a/src/video_core/shader/decode/video.cpp
+++ b/src/video_core/shader/decode/video.cpp
@@ -23,7 +23,7 @@ u32 ShaderIR::DecodeVideo(NodeBlock& bb, u32 pc) {
     const Node op_a =
         GetVideoOperand(GetRegister(instr.gpr8), instr.video.is_byte_chunk_a, instr.video.signed_a,
                         instr.video.type_a, instr.video.byte_height_a);
-    const Node op_b = [&]() {
+    const Node op_b = [this, instr] {
         if (instr.video.use_register_b) {
             return GetVideoOperand(GetRegister(instr.gpr20), instr.video.is_byte_chunk_b,
                                    instr.video.signed_b, instr.video.type_b,
diff --git a/src/video_core/shader/decode/warp.cpp b/src/video_core/shader/decode/warp.cpp
index a8e481b3c..fa8a250cc 100644
--- a/src/video_core/shader/decode/warp.cpp
+++ b/src/video_core/shader/decode/warp.cpp
@@ -46,9 +46,10 @@ u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::SHFL: {
-        Node mask = instr.shfl.is_mask_imm ? Immediate(static_cast<u32>(instr.shfl.mask_imm))
-                                           : GetRegister(instr.gpr39);
-        Node width = [&] {
+        Node width = [this, instr] {
+            Node mask = instr.shfl.is_mask_imm ? Immediate(static_cast<u32>(instr.shfl.mask_imm))
+                                               : GetRegister(instr.gpr39);
+
             // Convert the obscure SHFL mask back into GL_NV_shader_thread_shuffle's width. This has
             // been done reversing Nvidia's math. It won't work on all cases due to SHFL having
             // different parameters that don't properly map to GLSL's interface, but it should work
diff --git a/src/video_core/shader/expr.h b/src/video_core/shader/expr.h
index d3dcd00ec..4e8264367 100644
--- a/src/video_core/shader/expr.h
+++ b/src/video_core/shader/expr.h
@@ -17,13 +17,14 @@ using Tegra::Shader::Pred;
 class ExprAnd;
 class ExprBoolean;
 class ExprCondCode;
+class ExprGprEqual;
 class ExprNot;
 class ExprOr;
 class ExprPredicate;
 class ExprVar;
 
-using ExprData =
-    std::variant<ExprVar, ExprCondCode, ExprPredicate, ExprNot, ExprOr, ExprAnd, ExprBoolean>;
+using ExprData = std::variant<ExprVar, ExprCondCode, ExprPredicate, ExprNot, ExprOr, ExprAnd,
+                              ExprBoolean, ExprGprEqual>;
 using Expr = std::shared_ptr<ExprData>;
 
 class ExprAnd final {
@@ -118,6 +119,22 @@ public:
     bool value;
 };
 
+class ExprGprEqual final {
+public:
+    ExprGprEqual(u32 gpr, u32 value) : gpr{gpr}, value{value} {}
+
+    bool operator==(const ExprGprEqual& b) const {
+        return gpr == b.gpr && value == b.value;
+    }
+
+    bool operator!=(const ExprGprEqual& b) const {
+        return !operator==(b);
+    }
+
+    u32 gpr;
+    u32 value;
+};
+
 template <typename T, typename... Args>
 Expr MakeExpr(Args&&... args) {
     static_assert(std::is_convertible_v<T, ExprData>);
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index 338bab17c..447fb5c1d 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -410,7 +410,7 @@ public:
     explicit OperationNode(OperationCode code) : OperationNode(code, Meta{}) {}
 
     explicit OperationNode(OperationCode code, Meta meta)
-        : OperationNode(code, meta, std::vector<Node>{}) {}
+        : OperationNode(code, std::move(meta), std::vector<Node>{}) {}
 
     explicit OperationNode(OperationCode code, std::vector<Node> operands)
         : OperationNode(code, Meta{}, std::move(operands)) {}
diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp
index c1f2b88c8..1d9825c76 100644
--- a/src/video_core/shader/shader_ir.cpp
+++ b/src/video_core/shader/shader_ir.cpp
@@ -2,8 +2,9 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <algorithm>
+#include <array>
 #include <cmath>
-#include <unordered_map>
 
 #include "common/assert.h"
 #include "common/common_types.h"
@@ -22,10 +23,9 @@ using Tegra::Shader::PredCondition;
 using Tegra::Shader::PredOperation;
 using Tegra::Shader::Register;
 
-ShaderIR::ShaderIR(const ProgramCode& program_code, u32 main_offset, const std::size_t size,
-                   CompilerSettings settings)
-    : program_code{program_code}, main_offset{main_offset}, program_size{size}, basic_blocks{},
-      program_manager{true, true}, settings{settings} {
+ShaderIR::ShaderIR(const ProgramCode& program_code, u32 main_offset, CompilerSettings settings,
+                   ConstBufferLocker& locker)
+    : program_code{program_code}, main_offset{main_offset}, settings{settings}, locker{locker} {
     Decode();
 }
 
@@ -271,21 +271,24 @@ Node ShaderIR::GetSaturatedHalfFloat(Node value, bool saturate) {
 }
 
 Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, Node op_b) {
-    const std::unordered_map<PredCondition, OperationCode> PredicateComparisonTable = {
-        {PredCondition::LessThan, OperationCode::LogicalFLessThan},
-        {PredCondition::Equal, OperationCode::LogicalFEqual},
-        {PredCondition::LessEqual, OperationCode::LogicalFLessEqual},
-        {PredCondition::GreaterThan, OperationCode::LogicalFGreaterThan},
-        {PredCondition::NotEqual, OperationCode::LogicalFNotEqual},
-        {PredCondition::GreaterEqual, OperationCode::LogicalFGreaterEqual},
-        {PredCondition::LessThanWithNan, OperationCode::LogicalFLessThan},
-        {PredCondition::NotEqualWithNan, OperationCode::LogicalFNotEqual},
-        {PredCondition::LessEqualWithNan, OperationCode::LogicalFLessEqual},
-        {PredCondition::GreaterThanWithNan, OperationCode::LogicalFGreaterThan},
-        {PredCondition::GreaterEqualWithNan, OperationCode::LogicalFGreaterEqual}};
-
-    const auto comparison{PredicateComparisonTable.find(condition)};
-    UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonTable.end(),
+    static constexpr std::array comparison_table{
+        std::pair{PredCondition::LessThan, OperationCode::LogicalFLessThan},
+        std::pair{PredCondition::Equal, OperationCode::LogicalFEqual},
+        std::pair{PredCondition::LessEqual, OperationCode::LogicalFLessEqual},
+        std::pair{PredCondition::GreaterThan, OperationCode::LogicalFGreaterThan},
+        std::pair{PredCondition::NotEqual, OperationCode::LogicalFNotEqual},
+        std::pair{PredCondition::GreaterEqual, OperationCode::LogicalFGreaterEqual},
+        std::pair{PredCondition::LessThanWithNan, OperationCode::LogicalFLessThan},
+        std::pair{PredCondition::NotEqualWithNan, OperationCode::LogicalFNotEqual},
+        std::pair{PredCondition::LessEqualWithNan, OperationCode::LogicalFLessEqual},
+        std::pair{PredCondition::GreaterThanWithNan, OperationCode::LogicalFGreaterThan},
+        std::pair{PredCondition::GreaterEqualWithNan, OperationCode::LogicalFGreaterEqual},
+    };
+
+    const auto comparison =
+        std::find_if(comparison_table.cbegin(), comparison_table.cend(),
+                     [condition](const auto entry) { return condition == entry.first; });
+    UNIMPLEMENTED_IF_MSG(comparison == comparison_table.cend(),
                          "Unknown predicate comparison operation");
 
     Node predicate = Operation(comparison->second, NO_PRECISE, op_a, op_b);
@@ -306,21 +309,24 @@ Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, N
 
 Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_signed, Node op_a,
                                              Node op_b) {
-    const std::unordered_map<PredCondition, OperationCode> PredicateComparisonTable = {
-        {PredCondition::LessThan, OperationCode::LogicalILessThan},
-        {PredCondition::Equal, OperationCode::LogicalIEqual},
-        {PredCondition::LessEqual, OperationCode::LogicalILessEqual},
-        {PredCondition::GreaterThan, OperationCode::LogicalIGreaterThan},
-        {PredCondition::NotEqual, OperationCode::LogicalINotEqual},
-        {PredCondition::GreaterEqual, OperationCode::LogicalIGreaterEqual},
-        {PredCondition::LessThanWithNan, OperationCode::LogicalILessThan},
-        {PredCondition::NotEqualWithNan, OperationCode::LogicalINotEqual},
-        {PredCondition::LessEqualWithNan, OperationCode::LogicalILessEqual},
-        {PredCondition::GreaterThanWithNan, OperationCode::LogicalIGreaterThan},
-        {PredCondition::GreaterEqualWithNan, OperationCode::LogicalIGreaterEqual}};
-
-    const auto comparison{PredicateComparisonTable.find(condition)};
-    UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonTable.end(),
+    static constexpr std::array comparison_table{
+        std::pair{PredCondition::LessThan, OperationCode::LogicalILessThan},
+        std::pair{PredCondition::Equal, OperationCode::LogicalIEqual},
+        std::pair{PredCondition::LessEqual, OperationCode::LogicalILessEqual},
+        std::pair{PredCondition::GreaterThan, OperationCode::LogicalIGreaterThan},
+        std::pair{PredCondition::NotEqual, OperationCode::LogicalINotEqual},
+        std::pair{PredCondition::GreaterEqual, OperationCode::LogicalIGreaterEqual},
+        std::pair{PredCondition::LessThanWithNan, OperationCode::LogicalILessThan},
+        std::pair{PredCondition::NotEqualWithNan, OperationCode::LogicalINotEqual},
+        std::pair{PredCondition::LessEqualWithNan, OperationCode::LogicalILessEqual},
+        std::pair{PredCondition::GreaterThanWithNan, OperationCode::LogicalIGreaterThan},
+        std::pair{PredCondition::GreaterEqualWithNan, OperationCode::LogicalIGreaterEqual},
+    };
+
+    const auto comparison =
+        std::find_if(comparison_table.cbegin(), comparison_table.cend(),
+                     [condition](const auto entry) { return condition == entry.first; });
+    UNIMPLEMENTED_IF_MSG(comparison == comparison_table.cend(),
                          "Unknown predicate comparison operation");
 
     Node predicate = SignedOperation(comparison->second, is_signed, NO_PRECISE, std::move(op_a),
@@ -337,36 +343,43 @@ Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_si
 
 Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, Node op_a,
                                           Node op_b) {
-    const std::unordered_map<PredCondition, OperationCode> PredicateComparisonTable = {
-        {PredCondition::LessThan, OperationCode::Logical2HLessThan},
-        {PredCondition::Equal, OperationCode::Logical2HEqual},
-        {PredCondition::LessEqual, OperationCode::Logical2HLessEqual},
-        {PredCondition::GreaterThan, OperationCode::Logical2HGreaterThan},
-        {PredCondition::NotEqual, OperationCode::Logical2HNotEqual},
-        {PredCondition::GreaterEqual, OperationCode::Logical2HGreaterEqual},
-        {PredCondition::LessThanWithNan, OperationCode::Logical2HLessThanWithNan},
-        {PredCondition::NotEqualWithNan, OperationCode::Logical2HNotEqualWithNan},
-        {PredCondition::LessEqualWithNan, OperationCode::Logical2HLessEqualWithNan},
-        {PredCondition::GreaterThanWithNan, OperationCode::Logical2HGreaterThanWithNan},
-        {PredCondition::GreaterEqualWithNan, OperationCode::Logical2HGreaterEqualWithNan}};
-
-    const auto comparison{PredicateComparisonTable.find(condition)};
-    UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonTable.end(),
+    static constexpr std::array comparison_table{
+        std::pair{PredCondition::LessThan, OperationCode::Logical2HLessThan},
+        std::pair{PredCondition::Equal, OperationCode::Logical2HEqual},
+        std::pair{PredCondition::LessEqual, OperationCode::Logical2HLessEqual},
+        std::pair{PredCondition::GreaterThan, OperationCode::Logical2HGreaterThan},
+        std::pair{PredCondition::NotEqual, OperationCode::Logical2HNotEqual},
+        std::pair{PredCondition::GreaterEqual, OperationCode::Logical2HGreaterEqual},
+        std::pair{PredCondition::LessThanWithNan, OperationCode::Logical2HLessThanWithNan},
+        std::pair{PredCondition::NotEqualWithNan, OperationCode::Logical2HNotEqualWithNan},
+        std::pair{PredCondition::LessEqualWithNan, OperationCode::Logical2HLessEqualWithNan},
+        std::pair{PredCondition::GreaterThanWithNan, OperationCode::Logical2HGreaterThanWithNan},
+        std::pair{PredCondition::GreaterEqualWithNan, OperationCode::Logical2HGreaterEqualWithNan},
+    };
+
+    const auto comparison =
+        std::find_if(comparison_table.cbegin(), comparison_table.cend(),
+                     [condition](const auto entry) { return condition == entry.first; });
+    UNIMPLEMENTED_IF_MSG(comparison == comparison_table.cend(),
                          "Unknown predicate comparison operation");
 
     return Operation(comparison->second, NO_PRECISE, std::move(op_a), std::move(op_b));
 }
 
 OperationCode ShaderIR::GetPredicateCombiner(PredOperation operation) {
-    const std::unordered_map<PredOperation, OperationCode> PredicateOperationTable = {
-        {PredOperation::And, OperationCode::LogicalAnd},
-        {PredOperation::Or, OperationCode::LogicalOr},
-        {PredOperation::Xor, OperationCode::LogicalXor},
+    static constexpr std::array operation_table{
+        OperationCode::LogicalAnd,
+        OperationCode::LogicalOr,
+        OperationCode::LogicalXor,
     };
 
-    const auto op = PredicateOperationTable.find(operation);
-    UNIMPLEMENTED_IF_MSG(op == PredicateOperationTable.end(), "Unknown predicate operation");
-    return op->second;
+    const auto index = static_cast<std::size_t>(operation);
+    if (index >= operation_table.size()) {
+        UNIMPLEMENTED_MSG("Unknown predicate operation.");
+        return {};
+    }
+
+    return operation_table[index];
 }
 
 Node ShaderIR::GetConditionCode(Tegra::Shader::ConditionCode cc) const {
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index 105981d67..1fd44bde1 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -17,6 +17,7 @@
 #include "video_core/engines/shader_header.h"
 #include "video_core/shader/ast.h"
 #include "video_core/shader/compiler_settings.h"
+#include "video_core/shader/const_buffer_locker.h"
 #include "video_core/shader/node.h"
 
 namespace VideoCommon::Shader {
@@ -66,8 +67,8 @@ struct GlobalMemoryUsage {
 
 class ShaderIR final {
 public:
-    explicit ShaderIR(const ProgramCode& program_code, u32 main_offset, std::size_t size,
-                      CompilerSettings settings);
+    explicit ShaderIR(const ProgramCode& program_code, u32 main_offset, CompilerSettings settings,
+                      ConstBufferLocker& locker);
     ~ShaderIR();
 
     const std::map<u32, NodeBlock>& GetBasicBlocks() const {
@@ -172,6 +173,13 @@ public:
 
 private:
     friend class ASTDecoder;
+
+    struct SamplerInfo {
+        Tegra::Shader::TextureType type;
+        bool is_array;
+        bool is_shadow;
+    };
+
     void Decode();
 
     NodeBlock DecodeRange(u32 begin, u32 end);
@@ -296,12 +304,11 @@ private:
 
     /// Accesses a texture sampler
     const Sampler& GetSampler(const Tegra::Shader::Sampler& sampler,
-                              Tegra::Shader::TextureType type, bool is_array, bool is_shadow);
+                              std::optional<SamplerInfo> sampler_info);
 
     // Accesses a texture sampler for a bindless texture.
     const Sampler& GetBindlessSampler(const Tegra::Shader::Register& reg,
-                                      Tegra::Shader::TextureType type, bool is_array,
-                                      bool is_shadow);
+                                      std::optional<SamplerInfo> sampler_info);
 
     /// Accesses an image.
     Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type);
@@ -322,7 +329,7 @@ private:
                                   const Node4& components);
 
     void WriteTexsInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
-                                   const Node4& components);
+                                   const Node4& components, bool ignore_mask = false);
     void WriteTexsInstructionHalfFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
                                        const Node4& components);
 
@@ -371,12 +378,15 @@ private:
     std::pair<Node, s64> TrackRegister(const GprNode* tracked, const NodeBlock& code,
                                        s64 cursor) const;
 
-    std::tuple<Node, Node, GlobalMemoryBase> TrackAndGetGlobalMemory(
-        NodeBlock& bb, Tegra::Shader::Instruction instr, bool is_write);
+    std::tuple<Node, Node, GlobalMemoryBase> TrackGlobalMemory(NodeBlock& bb,
+                                                               Tegra::Shader::Instruction instr,
+                                                               bool is_write);
 
     const ProgramCode& program_code;
     const u32 main_offset;
-    const std::size_t program_size;
+    const CompilerSettings settings;
+    ConstBufferLocker& locker;
+
     bool decompiled{};
     bool disable_flow_stack{};
 
@@ -385,8 +395,7 @@ private:
 
     std::map<u32, NodeBlock> basic_blocks;
     NodeBlock global_code;
-    ASTManager program_manager;
-    CompilerSettings settings{};
+    ASTManager program_manager{true, true};
 
     std::set<u32> used_registers;
     std::set<Tegra::Shader::Pred> used_predicates;
diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp
index 250afc6d6..9a3c05288 100644
--- a/src/video_core/surface.cpp
+++ b/src/video_core/surface.cpp
@@ -212,6 +212,14 @@ PixelFormat PixelFormatFromTextureFormat(Tegra::Texture::TextureFormat format,
             break;
         }
         break;
+    case Tegra::Texture::TextureFormat::A4B4G4R4:
+        switch (component_type) {
+        case Tegra::Texture::ComponentType::UNORM:
+            return PixelFormat::R4G4B4A4U;
+        default:
+            break;
+        }
+        break;
     case Tegra::Texture::TextureFormat::R8:
         switch (component_type) {
         case Tegra::Texture::ComponentType::UNORM:
@@ -252,6 +260,7 @@ PixelFormat PixelFormatFromTextureFormat(Tegra::Texture::TextureFormat format,
         default:
             break;
         }
+        break;
     case Tegra::Texture::TextureFormat::R32_G32_B32_A32:
         switch (component_type) {
         case Tegra::Texture::ComponentType::FLOAT:
@@ -350,6 +359,16 @@ PixelFormat PixelFormatFromTextureFormat(Tegra::Texture::TextureFormat format,
         return is_srgb ? PixelFormat::ASTC_2D_8X5_SRGB : PixelFormat::ASTC_2D_8X5;
     case Tegra::Texture::TextureFormat::ASTC_2D_10X8:
         return is_srgb ? PixelFormat::ASTC_2D_10X8_SRGB : PixelFormat::ASTC_2D_10X8;
+    case Tegra::Texture::TextureFormat::ASTC_2D_6X6:
+        return is_srgb ? PixelFormat::ASTC_2D_6X6_SRGB : PixelFormat::ASTC_2D_6X6;
+    case Tegra::Texture::TextureFormat::ASTC_2D_10X10:
+        return is_srgb ? PixelFormat::ASTC_2D_10X10_SRGB : PixelFormat::ASTC_2D_10X10;
+    case Tegra::Texture::TextureFormat::ASTC_2D_12X12:
+        return is_srgb ? PixelFormat::ASTC_2D_12X12_SRGB : PixelFormat::ASTC_2D_12X12;
+    case Tegra::Texture::TextureFormat::ASTC_2D_8X6:
+        return is_srgb ? PixelFormat::ASTC_2D_8X6_SRGB : PixelFormat::ASTC_2D_8X6;
+    case Tegra::Texture::TextureFormat::ASTC_2D_6X5:
+        return is_srgb ? PixelFormat::ASTC_2D_6X5_SRGB : PixelFormat::ASTC_2D_6X5;
     case Tegra::Texture::TextureFormat::R16_G16:
         switch (component_type) {
         case Tegra::Texture::ComponentType::FLOAT:
@@ -510,6 +529,16 @@ bool IsPixelFormatASTC(PixelFormat format) {
     case PixelFormat::ASTC_2D_8X5_SRGB:
     case PixelFormat::ASTC_2D_10X8:
     case PixelFormat::ASTC_2D_10X8_SRGB:
+    case PixelFormat::ASTC_2D_6X6:
+    case PixelFormat::ASTC_2D_6X6_SRGB:
+    case PixelFormat::ASTC_2D_10X10:
+    case PixelFormat::ASTC_2D_10X10_SRGB:
+    case PixelFormat::ASTC_2D_12X12:
+    case PixelFormat::ASTC_2D_12X12_SRGB:
+    case PixelFormat::ASTC_2D_8X6:
+    case PixelFormat::ASTC_2D_8X6_SRGB:
+    case PixelFormat::ASTC_2D_6X5:
+    case PixelFormat::ASTC_2D_6X5_SRGB:
         return true;
     default:
         return false;
@@ -530,6 +559,11 @@ bool IsPixelFormatSRGB(PixelFormat format) {
     case PixelFormat::ASTC_2D_5X4_SRGB:
     case PixelFormat::ASTC_2D_5X5_SRGB:
     case PixelFormat::ASTC_2D_10X8_SRGB:
+    case PixelFormat::ASTC_2D_6X6_SRGB:
+    case PixelFormat::ASTC_2D_10X10_SRGB:
+    case PixelFormat::ASTC_2D_12X12_SRGB:
+    case PixelFormat::ASTC_2D_8X6_SRGB:
+    case PixelFormat::ASTC_2D_6X5_SRGB:
         return true;
     default:
         return false;
diff --git a/src/video_core/surface.h b/src/video_core/surface.h
index 1e1c432a5..97668f802 100644
--- a/src/video_core/surface.h
+++ b/src/video_core/surface.h
@@ -67,27 +67,38 @@ enum class PixelFormat {
     DXT23_SRGB = 49,
     DXT45_SRGB = 50,
     BC7U_SRGB = 51,
-    ASTC_2D_4X4_SRGB = 52,
-    ASTC_2D_8X8_SRGB = 53,
-    ASTC_2D_8X5_SRGB = 54,
-    ASTC_2D_5X4_SRGB = 55,
-    ASTC_2D_5X5 = 56,
-    ASTC_2D_5X5_SRGB = 57,
-    ASTC_2D_10X8 = 58,
-    ASTC_2D_10X8_SRGB = 59,
+    R4G4B4A4U = 52,
+    ASTC_2D_4X4_SRGB = 53,
+    ASTC_2D_8X8_SRGB = 54,
+    ASTC_2D_8X5_SRGB = 55,
+    ASTC_2D_5X4_SRGB = 56,
+    ASTC_2D_5X5 = 57,
+    ASTC_2D_5X5_SRGB = 58,
+    ASTC_2D_10X8 = 59,
+    ASTC_2D_10X8_SRGB = 60,
+    ASTC_2D_6X6 = 61,
+    ASTC_2D_6X6_SRGB = 62,
+    ASTC_2D_10X10 = 63,
+    ASTC_2D_10X10_SRGB = 64,
+    ASTC_2D_12X12 = 65,
+    ASTC_2D_12X12_SRGB = 66,
+    ASTC_2D_8X6 = 67,
+    ASTC_2D_8X6_SRGB = 68,
+    ASTC_2D_6X5 = 69,
+    ASTC_2D_6X5_SRGB = 70,
 
     MaxColorFormat,
 
     // Depth formats
-    Z32F = 60,
-    Z16 = 61,
+    Z32F = 71,
+    Z16 = 72,
 
     MaxDepthFormat,
 
     // DepthStencil formats
-    Z24S8 = 62,
-    S8Z24 = 63,
-    Z32FS8 = 64,
+    Z24S8 = 73,
+    S8Z24 = 74,
+    Z32FS8 = 75,
 
     MaxDepthStencilFormat,
 
@@ -177,6 +188,7 @@ constexpr std::array<u32, MaxPixelFormat> compression_factor_shift_table = {{
     2, // DXT23_SRGB
     2, // DXT45_SRGB
     2, // BC7U_SRGB
+    0, // R4G4B4A4U
     2, // ASTC_2D_4X4_SRGB
     2, // ASTC_2D_8X8_SRGB
     2, // ASTC_2D_8X5_SRGB
@@ -185,6 +197,16 @@ constexpr std::array<u32, MaxPixelFormat> compression_factor_shift_table = {{
     2, // ASTC_2D_5X5_SRGB
     2, // ASTC_2D_10X8
     2, // ASTC_2D_10X8_SRGB
+    2, // ASTC_2D_6X6
+    2, // ASTC_2D_6X6_SRGB
+    2, // ASTC_2D_10X10
+    2, // ASTC_2D_10X10_SRGB
+    2, // ASTC_2D_12X12
+    2, // ASTC_2D_12X12_SRGB
+    2, // ASTC_2D_8X6
+    2, // ASTC_2D_8X6_SRGB
+    2, // ASTC_2D_6X5
+    2, // ASTC_2D_6X5_SRGB
     0, // Z32F
     0, // Z16
     0, // Z24S8
@@ -261,6 +283,7 @@ constexpr std::array<u32, MaxPixelFormat> block_width_table = {{
     4,  // DXT23_SRGB
     4,  // DXT45_SRGB
     4,  // BC7U_SRGB
+    1,  // R4G4B4A4U
     4,  // ASTC_2D_4X4_SRGB
     8,  // ASTC_2D_8X8_SRGB
     8,  // ASTC_2D_8X5_SRGB
@@ -269,6 +292,16 @@ constexpr std::array<u32, MaxPixelFormat> block_width_table = {{
     5,  // ASTC_2D_5X5_SRGB
     10, // ASTC_2D_10X8
     10, // ASTC_2D_10X8_SRGB
+    6,  // ASTC_2D_6X6
+    6,  // ASTC_2D_6X6_SRGB
+    10, // ASTC_2D_10X10
+    10, // ASTC_2D_10X10_SRGB
+    12, // ASTC_2D_12X12
+    12, // ASTC_2D_12X12_SRGB
+    8,  // ASTC_2D_8X6
+    8,  // ASTC_2D_8X6_SRGB
+    6,  // ASTC_2D_6X5
+    6,  // ASTC_2D_6X5_SRGB
     1,  // Z32F
     1,  // Z16
     1,  // Z24S8
@@ -285,71 +318,82 @@ static constexpr u32 GetDefaultBlockWidth(PixelFormat format) {
 }
 
 constexpr std::array<u32, MaxPixelFormat> block_height_table = {{
-    1, // ABGR8U
-    1, // ABGR8S
-    1, // ABGR8UI
-    1, // B5G6R5U
-    1, // A2B10G10R10U
-    1, // A1B5G5R5U
-    1, // R8U
-    1, // R8UI
-    1, // RGBA16F
-    1, // RGBA16U
-    1, // RGBA16UI
-    1, // R11FG11FB10F
-    1, // RGBA32UI
-    4, // DXT1
-    4, // DXT23
-    4, // DXT45
-    4, // DXN1
-    4, // DXN2UNORM
-    4, // DXN2SNORM
-    4, // BC7U
-    4, // BC6H_UF16
-    4, // BC6H_SF16
-    4, // ASTC_2D_4X4
-    1, // BGRA8
-    1, // RGBA32F
-    1, // RG32F
-    1, // R32F
-    1, // R16F
-    1, // R16U
-    1, // R16S
-    1, // R16UI
-    1, // R16I
-    1, // RG16
-    1, // RG16F
-    1, // RG16UI
-    1, // RG16I
-    1, // RG16S
-    1, // RGB32F
-    1, // RGBA8_SRGB
-    1, // RG8U
-    1, // RG8S
-    1, // RG32UI
-    1, // RGBX16F
-    1, // R32UI
-    8, // ASTC_2D_8X8
-    5, // ASTC_2D_8X5
-    4, // ASTC_2D_5X4
-    1, // BGRA8_SRGB
-    4, // DXT1_SRGB
-    4, // DXT23_SRGB
-    4, // DXT45_SRGB
-    4, // BC7U_SRGB
-    4, // ASTC_2D_4X4_SRGB
-    8, // ASTC_2D_8X8_SRGB
-    5, // ASTC_2D_8X5_SRGB
-    4, // ASTC_2D_5X4_SRGB
-    5, // ASTC_2D_5X5
-    5, // ASTC_2D_5X5_SRGB
-    8, // ASTC_2D_10X8
-    8, // ASTC_2D_10X8_SRGB
-    1, // Z32F
-    1, // Z16
-    1, // Z24S8
-    1, // S8Z24
-    1, // Z32FS8
+    1,  // ABGR8U
+    1,  // ABGR8S
+    1,  // ABGR8UI
+    1,  // B5G6R5U
+    1,  // A2B10G10R10U
+    1,  // A1B5G5R5U
+    1,  // R8U
+    1,  // R8UI
+    1,  // RGBA16F
+    1,  // RGBA16U
+    1,  // RGBA16UI
+    1,  // R11FG11FB10F
+    1,  // RGBA32UI
+    4,  // DXT1
+    4,  // DXT23
+    4,  // DXT45
+    4,  // DXN1
+    4,  // DXN2UNORM
+    4,  // DXN2SNORM
+    4,  // BC7U
+    4,  // BC6H_UF16
+    4,  // BC6H_SF16
+    4,  // ASTC_2D_4X4
+    1,  // BGRA8
+    1,  // RGBA32F
+    1,  // RG32F
+    1,  // R32F
+    1,  // R16F
+    1,  // R16U
+    1,  // R16S
+    1,  // R16UI
+    1,  // R16I
+    1,  // RG16
+    1,  // RG16F
+    1,  // RG16UI
+    1,  // RG16I
+    1,  // RG16S
+    1,  // RGB32F
+    1,  // RGBA8_SRGB
+    1,  // RG8U
+    1,  // RG8S
+    1,  // RG32UI
+    1,  // RGBX16F
+    1,  // R32UI
+    8,  // ASTC_2D_8X8
+    5,  // ASTC_2D_8X5
+    4,  // ASTC_2D_5X4
+    1,  // BGRA8_SRGB
+    4,  // DXT1_SRGB
+    4,  // DXT23_SRGB
+    4,  // DXT45_SRGB
+    4,  // BC7U_SRGB
+    1,  // R4G4B4A4U
+    4,  // ASTC_2D_4X4_SRGB
+    8,  // ASTC_2D_8X8_SRGB
+    5,  // ASTC_2D_8X5_SRGB
+    4,  // ASTC_2D_5X4_SRGB
+    5,  // ASTC_2D_5X5
+    5,  // ASTC_2D_5X5_SRGB
+    8,  // ASTC_2D_10X8
+    8,  // ASTC_2D_10X8_SRGB
+    6,  // ASTC_2D_6X6
+    6,  // ASTC_2D_6X6_SRGB
+    10, // ASTC_2D_10X10
+    10, // ASTC_2D_10X10_SRGB
+    12, // ASTC_2D_12X12
+    12, // ASTC_2D_12X12_SRGB
+    6,  // ASTC_2D_8X6
+    6,  // ASTC_2D_8X6_SRGB
+    5,  // ASTC_2D_6X5
+    5,  // ASTC_2D_6X5_SRGB
+    1,  // Z32F
+    1,  // Z16
+    1,  // Z24S8
+    1,  // S8Z24
+    1,  // Z32FS8
 }};
 
 static constexpr u32 GetDefaultBlockHeight(PixelFormat format) {
@@ -413,6 +457,7 @@ constexpr std::array<u32, MaxPixelFormat> bpp_table = {{
     128, // DXT23_SRGB
     128, // DXT45_SRGB
     128, // BC7U
+    16,  // R4G4B4A4U
     128, // ASTC_2D_4X4_SRGB
     128, // ASTC_2D_8X8_SRGB
     128, // ASTC_2D_8X5_SRGB
@@ -421,6 +466,16 @@ constexpr std::array<u32, MaxPixelFormat> bpp_table = {{
     128, // ASTC_2D_5X5_SRGB
     128, // ASTC_2D_10X8
     128, // ASTC_2D_10X8_SRGB
+    128, // ASTC_2D_6X6
+    128, // ASTC_2D_6X6_SRGB
+    128, // ASTC_2D_10X10
+    128, // ASTC_2D_10X10_SRGB
+    128, // ASTC_2D_12X12
+    128, // ASTC_2D_12X12_SRGB
+    128, // ASTC_2D_8X6
+    128, // ASTC_2D_8X6_SRGB
+    128, // ASTC_2D_6X5
+    128, // ASTC_2D_6X5_SRGB
     32,  // Z32F
     16,  // Z16
     32,  // Z24S8
@@ -504,6 +559,7 @@ constexpr std::array<SurfaceCompression, MaxPixelFormat> compression_type_table
     SurfaceCompression::Compressed, // DXT23_SRGB
     SurfaceCompression::Compressed, // DXT45_SRGB
     SurfaceCompression::Compressed, // BC7U_SRGB
+    SurfaceCompression::None,       // R4G4B4A4U
     SurfaceCompression::Converted,  // ASTC_2D_4X4_SRGB
     SurfaceCompression::Converted,  // ASTC_2D_8X8_SRGB
     SurfaceCompression::Converted,  // ASTC_2D_8X5_SRGB
@@ -512,6 +568,16 @@ constexpr std::array<SurfaceCompression, MaxPixelFormat> compression_type_table
     SurfaceCompression::Converted,  // ASTC_2D_5X5_SRGB
     SurfaceCompression::Converted,  // ASTC_2D_10X8
     SurfaceCompression::Converted,  // ASTC_2D_10X8_SRGB
+    SurfaceCompression::Converted,  // ASTC_2D_6X6
+    SurfaceCompression::Converted,  // ASTC_2D_6X6_SRGB
+    SurfaceCompression::Converted,  // ASTC_2D_10X10
+    SurfaceCompression::Converted,  // ASTC_2D_10X10_SRGB
+    SurfaceCompression::Converted,  // ASTC_2D_12X12
+    SurfaceCompression::Converted,  // ASTC_2D_12X12_SRGB
+    SurfaceCompression::Converted,  // ASTC_2D_8X6
+    SurfaceCompression::Converted,  // ASTC_2D_8X6_SRGB
+    SurfaceCompression::Converted,  // ASTC_2D_6X5
+    SurfaceCompression::Converted,  // ASTC_2D_6X5_SRGB
     SurfaceCompression::None,       // Z32F
     SurfaceCompression::None,       // Z16
     SurfaceCompression::None,       // Z24S8
diff --git a/src/video_core/texture_cache/surface_base.cpp b/src/video_core/texture_cache/surface_base.cpp
index 683c49207..829268b4c 100644
--- a/src/video_core/texture_cache/surface_base.cpp
+++ b/src/video_core/texture_cache/surface_base.cpp
@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "common/algorithm.h"
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/microprofile.h"
diff --git a/src/video_core/texture_cache/surface_base.h b/src/video_core/texture_cache/surface_base.h
index 5e497e49f..1bed82898 100644
--- a/src/video_core/texture_cache/surface_base.h
+++ b/src/video_core/texture_cache/surface_base.h
@@ -4,12 +4,11 @@
 
 #pragma once
 
-#include <algorithm>
+#include <optional>
+#include <tuple>
 #include <unordered_map>
 #include <vector>
 
-#include "common/assert.h"
-#include "common/binary_find.h"
 #include "common/common_types.h"
 #include "video_core/gpu.h"
 #include "video_core/morton.h"
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index ca2da8f97..6a92b22d3 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -62,10 +62,10 @@ public:
         }
     }
 
-    /***
-     * `Guard` guarantees that rendertargets don't unregister themselves if the
+    /**
+     * Guarantees that rendertargets don't unregister themselves if the
      * collide. Protection is currently only done on 3D slices.
-     ***/
+     */
     void GuardRenderTargets(bool new_guard) {
         guard_render_targets = new_guard;
     }
@@ -287,7 +287,7 @@ protected:
                            const Tegra::Engines::Fermi2D::Config& copy_config) = 0;
 
     // Depending on the backend, a buffer copy can be slow as it means deoptimizing the texture
-    // and reading it from a sepparate buffer.
+    // and reading it from a separate buffer.
     virtual void BufferCopy(TSurface& src_surface, TSurface& dst_surface) = 0;
 
     void ManageRenderTargetUnregister(TSurface& surface) {
@@ -386,12 +386,13 @@ private:
     };
 
     /**
-     * `PickStrategy` takes care of selecting a proper strategy to deal with a texture recycle.
-     * @param overlaps, the overlapping surfaces registered in the cache.
-     * @param params, the paremeters on the new surface.
-     * @param gpu_addr, the starting address of the new surface.
-     * @param untopological, tells the recycler that the texture has no way to match the overlaps
-     * due to topological reasons.
+     * Takes care of selecting a proper strategy to deal with a texture recycle.
+     *
+     * @param overlaps      The overlapping surfaces registered in the cache.
+     * @param params        The parameters on the new surface.
+     * @param gpu_addr      The starting address of the new surface.
+     * @param untopological Indicates to the recycler that the texture has no way
+     *                      to match the overlaps due to topological reasons.
      **/
     RecycleStrategy PickStrategy(std::vector<TSurface>& overlaps, const SurfaceParams& params,
                                  const GPUVAddr gpu_addr, const MatchTopologyResult untopological) {
@@ -402,7 +403,7 @@ private:
         if (params.block_depth > 1 || params.target == SurfaceTarget::Texture3D) {
             return RecycleStrategy::Flush;
         }
-        for (auto s : overlaps) {
+        for (const auto& s : overlaps) {
             const auto& s_params = s->GetSurfaceParams();
             if (s_params.block_depth > 1 || s_params.target == SurfaceTarget::Texture3D) {
                 return RecycleStrategy::Flush;
@@ -419,16 +420,19 @@ private:
     }
 
     /**
-     *  `RecycleSurface` es a method we use to decide what to do with textures we can't resolve in
-     *the cache It has 2 implemented strategies: Ignore and Flush. Ignore just unregisters all the
-     *overlaps and loads the new texture. Flush, flushes all the overlaps into memory and loads the
-     *new surface from that data.
-     * @param overlaps, the overlapping surfaces registered in the cache.
-     * @param params, the paremeters on the new surface.
-     * @param gpu_addr, the starting address of the new surface.
-     * @param preserve_contents, tells if the new surface should be loaded from meory or left blank
-     * @param untopological, tells the recycler that the texture has no way to match the overlaps
-     * due to topological reasons.
+     * Used to decide what to do with textures we can't resolve in the cache It has 2 implemented
+     * strategies: Ignore and Flush.
+     *
+     * - Ignore: Just unregisters all the overlaps and loads the new texture.
+     * - Flush: Flushes all the overlaps into memory and loads the new surface from that data.
+     *
+     * @param overlaps          The overlapping surfaces registered in the cache.
+     * @param params            The parameters for the new surface.
+     * @param gpu_addr          The starting address of the new surface.
+     * @param preserve_contents Indicates that the new surface should be loaded from memory or left
+     *                          blank.
+     * @param untopological     Indicates to the recycler that the texture has no way to match the
+     *                          overlaps due to topological reasons.
      **/
     std::pair<TSurface, TView> RecycleSurface(std::vector<TSurface>& overlaps,
                                               const SurfaceParams& params, const GPUVAddr gpu_addr,
@@ -465,10 +469,12 @@ private:
     }
 
     /**
-     * `RebuildSurface` this method takes a single surface and recreates into another that
-     * may differ in format, target or width alingment.
-     * @param current_surface, the registered surface in the cache which we want to convert.
-     * @param params, the new surface params which we'll use to recreate the surface.
+     * Takes a single surface and recreates into another that may differ in
+     * format, target or width alignment.
+     *
+     * @param current_surface The registered surface in the cache which we want to convert.
+     * @param params          The new surface params which we'll use to recreate the surface.
+     * @param is_render       Whether or not the surface is a render target.
      **/
     std::pair<TSurface, TView> RebuildSurface(TSurface current_surface, const SurfaceParams& params,
                                               bool is_render) {
@@ -502,12 +508,14 @@ private:
     }
 
     /**
-     * `ManageStructuralMatch` this method takes a single surface and checks with the new surface's
-     * params if it's an exact match, we return the main view of the registered surface. If it's
-     * formats don't match, we rebuild the surface. We call this last method a `Mirage`. If formats
+     * Takes a single surface and checks with the new surface's params if it's an exact
+     * match, we return the main view of the registered surface. If its formats don't
+     * match, we rebuild the surface. We call this last method a `Mirage`. If formats
      * match but the targets don't, we create an overview View of the registered surface.
-     * @param current_surface, the registered surface in the cache which we want to convert.
-     * @param params, the new surface params which we want to check.
+     *
+     * @param current_surface The registered surface in the cache which we want to convert.
+     * @param params          The new surface params which we want to check.
+     * @param is_render       Whether or not the surface is a render target.
      **/
     std::pair<TSurface, TView> ManageStructuralMatch(TSurface current_surface,
                                                      const SurfaceParams& params, bool is_render) {
@@ -529,13 +537,14 @@ private:
     }
 
     /**
-     * `TryReconstructSurface` unlike `RebuildSurface` where we know the registered surface
-     * matches the candidate in some way, we got no guarantess here. We try to see if the overlaps
-     * are sublayers/mipmaps of the new surface, if they all match we end up recreating a surface
-     * for them, else we return nothing.
-     * @param overlaps, the overlapping surfaces registered in the cache.
-     * @param params, the paremeters on the new surface.
-     * @param gpu_addr, the starting address of the new surface.
+     * Unlike RebuildSurface where we know whether or not registered surfaces match the candidate
+     * in some way, we have no guarantees here. We try to see if the overlaps are sublayers/mipmaps
+     * of the new surface, if they all match we end up recreating a surface for them,
+     * else we return nothing.
+     *
+     * @param overlaps The overlapping surfaces registered in the cache.
+     * @param params   The parameters on the new surface.
+     * @param gpu_addr The starting address of the new surface.
      **/
     std::optional<std::pair<TSurface, TView>> TryReconstructSurface(std::vector<TSurface>& overlaps,
                                                                     const SurfaceParams& params,
@@ -575,7 +584,7 @@ private:
         } else if (Settings::values.use_accurate_gpu_emulation && passed_tests != overlaps.size()) {
             return {};
         }
-        for (auto surface : overlaps) {
+        for (const auto& surface : overlaps) {
             Unregister(surface);
         }
         new_surface->MarkAsModified(modified, Tick());
@@ -584,19 +593,27 @@ private:
     }
 
     /**
-     * `GetSurface` gets the starting address and parameters of a candidate surface and tries
-     * to find a matching surface within the cache. This is done in 3 big steps. The first is to
-     * check the 1st Level Cache in order to find an exact match, if we fail, we move to step 2.
-     * Step 2 is checking if there are any overlaps at all, if none, we just load the texture from
-     * memory else we move to step 3. Step 3 consists on figuring the relationship between the
-     * candidate texture and the overlaps. We divide the scenarios depending if there's 1 or many
-     * overlaps. If there's many, we just try to reconstruct a new surface out of them based on the
-     * candidate's parameters, if we fail, we recycle. When there's only 1 overlap then we have to
-     * check if the candidate is a view (layer/mipmap) of the overlap or if the registered surface
-     * is a mipmap/layer of the candidate. In this last case we reconstruct a new surface.
-     * @param gpu_addr, the starting address of the candidate surface.
-     * @param params, the paremeters on the candidate surface.
-     * @param preserve_contents, tells if the new surface should be loaded from meory or left blank.
+     * Gets the starting address and parameters of a candidate surface and tries
+     * to find a matching surface within the cache. This is done in 3 big steps:
+     *
+     * 1. Check the 1st Level Cache in order to find an exact match, if we fail, we move to step 2.
+     *
+     * 2. Check if there are any overlaps at all, if there are none, we just load the texture from
+     *    memory else we move to step 3.
+     *
+     * 3. Consists of figuring out the relationship between the candidate texture and the
+     *    overlaps. We divide the scenarios depending if there's 1 or many overlaps. If
+     *    there's many, we just try to reconstruct a new surface out of them based on the
+     *    candidate's parameters, if we fail, we recycle. When there's only 1 overlap then we
+     *    have to check if the candidate is a view (layer/mipmap) of the overlap or if the
+     *    registered surface is a mipmap/layer of the candidate. In this last case we reconstruct
+     *    a new surface.
+     *
+     * @param gpu_addr          The starting address of the candidate surface.
+     * @param params            The parameters on the candidate surface.
+     * @param preserve_contents Indicates that the new surface should be loaded from memory or
+     *                          left blank.
+     * @param is_render         Whether or not the surface is a render target.
      **/
     std::pair<TSurface, TView> GetSurface(const GPUVAddr gpu_addr, const SurfaceParams& params,
                                           bool preserve_contents, bool is_render) {
@@ -651,7 +668,7 @@ private:
         // Step 3
         // Now we need to figure the relationship between the texture and its overlaps
         // we do a topological test to ensure we can find some relationship. If it fails
-        // inmediatly recycle the texture
+        // immediately recycle the texture
         for (const auto& surface : overlaps) {
             const auto topological_result = surface->MatchesTopology(params);
             if (topological_result != MatchTopologyResult::FullMatch) {
@@ -720,12 +737,13 @@ private:
     }
 
     /**
-     * `DeduceSurface` gets the starting address and parameters of a candidate surface and tries
-     * to find a matching surface within the cache that's similar to it. If there are many textures
+     * Gets the starting address and parameters of a candidate surface and tries to find a
+     * matching surface within the cache that's similar to it. If there are many textures
      * or the texture found if entirely incompatible, it will fail. If no texture is found, the
      * blit will be unsuccessful.
-     * @param gpu_addr, the starting address of the candidate surface.
-     * @param params, the paremeters on the candidate surface.
+     *
+     * @param gpu_addr The starting address of the candidate surface.
+     * @param params   The parameters on the candidate surface.
      **/
     Deduction DeduceSurface(const GPUVAddr gpu_addr, const SurfaceParams& params) {
         const auto host_ptr{system.GPU().MemoryManager().GetPointer(gpu_addr)};
@@ -777,11 +795,14 @@ private:
     }
 
     /**
-     * `DeduceBestBlit` gets the a source and destination starting address and parameters,
+     * Gets the a source and destination starting address and parameters,
      * and tries to deduce if they are supposed to be depth textures. If so, their
      * parameters are modified and fixed into so.
-     * @param gpu_addr, the starting address of the candidate surface.
-     * @param params, the parameters on the candidate surface.
+     *
+     * @param src_params   The parameters of the candidate surface.
+     * @param dst_params   The parameters of the destination surface.
+     * @param src_gpu_addr The starting address of the candidate surface.
+     * @param dst_gpu_addr The starting address of the destination surface.
      **/
     void DeduceBestBlit(SurfaceParams& src_params, SurfaceParams& dst_params,
                         const GPUVAddr src_gpu_addr, const GPUVAddr dst_gpu_addr) {
diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp
index a9b8f69af..58b608a36 100644
--- a/src/video_core/textures/astc.cpp
+++ b/src/video_core/textures/astc.cpp
@@ -422,7 +422,7 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) {
     TexelWeightParams params;
 
     // Read the entire block mode all at once
-    uint16_t modeBits = strm.ReadBits(11);
+    uint16_t modeBits = static_cast<uint16_t>(strm.ReadBits(11));
 
     // Does this match the void extent block mode?
     if ((modeBits & 0x01FF) == 0x1FC) {
@@ -625,10 +625,10 @@ static void FillVoidExtentLDR(InputBitStream& strm, uint32_t* const outBuf, uint
     }
 
     // Decode the RGBA components and renormalize them to the range [0, 255]
-    uint16_t r = strm.ReadBits(16);
-    uint16_t g = strm.ReadBits(16);
-    uint16_t b = strm.ReadBits(16);
-    uint16_t a = strm.ReadBits(16);
+    uint16_t r = static_cast<uint16_t>(strm.ReadBits(16));
+    uint16_t g = static_cast<uint16_t>(strm.ReadBits(16));
+    uint16_t b = static_cast<uint16_t>(strm.ReadBits(16));
+    uint16_t a = static_cast<uint16_t>(strm.ReadBits(16));
 
     uint32_t rgba = (r >> 8) | (g & 0xFF00) | (static_cast<uint32_t>(b) & 0xFF00) << 8 |
                     (static_cast<uint32_t>(a) & 0xFF00) << 16;
@@ -681,9 +681,10 @@ protected:
 
 public:
     Pixel() = default;
-    Pixel(ChannelType a, ChannelType r, ChannelType g, ChannelType b, unsigned bitDepth = 8)
+    Pixel(uint32_t a, uint32_t r, uint32_t g, uint32_t b, unsigned bitDepth = 8)
         : m_BitDepth{uint8_t(bitDepth), uint8_t(bitDepth), uint8_t(bitDepth), uint8_t(bitDepth)},
-          color{a, r, g, b} {}
+          color{static_cast<ChannelType>(a), static_cast<ChannelType>(r),
+                static_cast<ChannelType>(g), static_cast<ChannelType>(b)} {}
 
     // Changes the depth of each pixel. This scales the values to
     // the appropriate bit depth by either truncating the least