15 files changed, 292 insertions, 82 deletions

diff --git a/src/core/file_sys/system_archive/ng_word.cpp b/src/core/file_sys/system_archive/ng_word.cpp
index d0acdbd49..f4443784d 100644
--- a/src/core/file_sys/system_archive/ng_word.cpp
+++ b/src/core/file_sys/system_archive/ng_word.cpp
@@ -26,7 +26,7 @@ constexpr std::array<u8, 30> WORD_TXT{
 VirtualDir NgWord1() {
     std::vector<VirtualFile> files(NgWord1Data::NUMBER_WORD_TXT_FILES);
 
-    for (std::size_t i = 0; i < NgWord1Data::NUMBER_WORD_TXT_FILES; ++i) {
+    for (std::size_t i = 0; i < files.size(); ++i) {
         files[i] = std::make_shared<ArrayVfsFile<NgWord1Data::WORD_TXT.size()>>(
             NgWord1Data::WORD_TXT, fmt::format("{}.txt", i));
     }
@@ -39,4 +39,43 @@ VirtualDir NgWord1() {
     return std::make_shared<VectorVfsDirectory>(files, std::vector<VirtualDir>{}, "data");
 }
 
+namespace NgWord2Data {
+
+constexpr std::size_t NUMBER_AC_NX_FILES = 0x10;
+
+// Should this archive replacement mysteriously not work on a future game, consider updating.
+constexpr std::array<u8, 4> VERSION_DAT{0x0, 0x0, 0x0, 0x15}; // 5.1.0 System Version
+
+constexpr std::array<u8, 0x2C> AC_NX_DATA{
+    0x1F, 0x8B, 0x08, 0x08, 0xD5, 0x2C, 0x09, 0x5C, 0x04, 0x00, 0x61, 0x63, 0x72, 0x61, 0x77,
+    0x00, 0xED, 0xC1, 0x01, 0x0D, 0x00, 0x00, 0x00, 0xC2, 0x20, 0xFB, 0xA7, 0xB6, 0xC7, 0x07,
+    0x0C, 0x00, 0x00, 0x00, 0xC8, 0x3B, 0x11, 0x00, 0x1C, 0xC7, 0x00, 0x10, 0x00, 0x00,
+}; // Deserializes to no bad words
+
+} // namespace NgWord2Data
+
+VirtualDir NgWord2() {
+    std::vector<VirtualFile> files(NgWord2Data::NUMBER_AC_NX_FILES * 3);
+
+    for (std::size_t i = 0; i < NgWord2Data::NUMBER_AC_NX_FILES; ++i) {
+        files[3 * i] = std::make_shared<ArrayVfsFile<NgWord2Data::AC_NX_DATA.size()>>(
+            NgWord2Data::AC_NX_DATA, fmt::format("ac_{}_b1_nx", i));
+        files[3 * i + 1] = std::make_shared<ArrayVfsFile<NgWord2Data::AC_NX_DATA.size()>>(
+            NgWord2Data::AC_NX_DATA, fmt::format("ac_{}_b2_nx", i));
+        files[3 * i + 2] = std::make_shared<ArrayVfsFile<NgWord2Data::AC_NX_DATA.size()>>(
+            NgWord2Data::AC_NX_DATA, fmt::format("ac_{}_not_b_nx", i));
+    }
+
+    files.push_back(std::make_shared<ArrayVfsFile<NgWord2Data::AC_NX_DATA.size()>>(
+        NgWord2Data::AC_NX_DATA, "ac_common_b1_nx"));
+    files.push_back(std::make_shared<ArrayVfsFile<NgWord2Data::AC_NX_DATA.size()>>(
+        NgWord2Data::AC_NX_DATA, "ac_common_b2_nx"));
+    files.push_back(std::make_shared<ArrayVfsFile<NgWord2Data::AC_NX_DATA.size()>>(
+        NgWord2Data::AC_NX_DATA, "ac_common_not_b_nx"));
+    files.push_back(std::make_shared<ArrayVfsFile<NgWord2Data::VERSION_DAT.size()>>(
+        NgWord2Data::VERSION_DAT, "version.dat"));
+
+    return std::make_shared<VectorVfsDirectory>(files, std::vector<VirtualDir>{}, "data");
+}
+
 } // namespace FileSys::SystemArchive
diff --git a/src/core/file_sys/system_archive/ng_word.h b/src/core/file_sys/system_archive/ng_word.h
index f4bc67344..cd81e0abb 100644
--- a/src/core/file_sys/system_archive/ng_word.h
+++ b/src/core/file_sys/system_archive/ng_word.h
@@ -9,5 +9,6 @@
 namespace FileSys::SystemArchive {
 
 VirtualDir NgWord1();
+VirtualDir NgWord2();
 
 } // namespace FileSys::SystemArchive
diff --git a/src/core/file_sys/system_archive/system_archive.cpp b/src/core/file_sys/system_archive/system_archive.cpp
index c9c40a07d..e3e79f40a 100644
--- a/src/core/file_sys/system_archive/system_archive.cpp
+++ b/src/core/file_sys/system_archive/system_archive.cpp
@@ -2,7 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <functional>
 #include "common/logging/log.h"
 #include "core/file_sys/romfs.h"
 #include "core/file_sys/system_archive/ng_word.h"
@@ -13,7 +12,7 @@ namespace FileSys::SystemArchive {
 constexpr u64 SYSTEM_ARCHIVE_BASE_TITLE_ID = 0x0100000000000800;
 constexpr std::size_t SYSTEM_ARCHIVE_COUNT = 0x28;
 
-using SystemArchiveSupplier = std::function<VirtualDir()>;
+using SystemArchiveSupplier = VirtualDir (*)();
 
 struct SystemArchiveDescriptor {
     u64 title_id;
@@ -21,7 +20,7 @@ struct SystemArchiveDescriptor {
     SystemArchiveSupplier supplier;
 };
 
-const std::array<SystemArchiveDescriptor, SYSTEM_ARCHIVE_COUNT> SYSTEM_ARCHIVES = {{
+constexpr std::array<SystemArchiveDescriptor, SYSTEM_ARCHIVE_COUNT> SYSTEM_ARCHIVES{{
     {0x0100000000000800, "CertStore", nullptr},
     {0x0100000000000801, "ErrorMessage", nullptr},
     {0x0100000000000802, "MiiModel", nullptr},
@@ -57,7 +56,7 @@ const std::array<SystemArchiveDescriptor, SYSTEM_ARCHIVE_COUNT> SYSTEM_ARCHIVES
     {0x0100000000000820, "PlatformConfigCopper", nullptr},
     {0x0100000000000821, "PlatformConfigHoag", nullptr},
     {0x0100000000000822, "ControllerFirmware", nullptr},
-    {0x0100000000000823, "NgWord2", nullptr},
+    {0x0100000000000823, "NgWord2", &NgWord2},
     {0x0100000000000824, "PlatformConfigIcosaMariko", nullptr},
     {0x0100000000000825, "ApplicationBlackList", nullptr},
     {0x0100000000000826, "RebootlessSystemUpdateVersion", nullptr},
diff --git a/src/core/hle/kernel/object.cpp b/src/core/hle/kernel/object.cpp
index bb1b68778..0ea851a74 100644
--- a/src/core/hle/kernel/object.cpp
+++ b/src/core/hle/kernel/object.cpp
@@ -15,6 +15,7 @@ bool Object::IsWaitable() const {
     switch (GetHandleType()) {
     case HandleType::ReadableEvent:
     case HandleType::Thread:
+    case HandleType::Process:
     case HandleType::Timer:
     case HandleType::ServerPort:
     case HandleType::ServerSession:
@@ -23,7 +24,6 @@ bool Object::IsWaitable() const {
     case HandleType::Unknown:
     case HandleType::WritableEvent:
     case HandleType::SharedMemory:
-    case HandleType::Process:
     case HandleType::AddressArbiter:
     case HandleType::ResourceLimit:
     case HandleType::ClientPort:
diff --git a/src/core/hle/kernel/process.cpp b/src/core/hle/kernel/process.cpp
index 4ecb8c926..211bf6686 100644
--- a/src/core/hle/kernel/process.cpp
+++ b/src/core/hle/kernel/process.cpp
@@ -9,6 +9,7 @@
 #include "common/logging/log.h"
 #include "core/core.h"
 #include "core/file_sys/program_metadata.h"
+#include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/resource_limit.h"
@@ -48,6 +49,21 @@ SharedPtr<ResourceLimit> Process::GetResourceLimit() const {
     return resource_limit;
 }
 
+ResultCode Process::ClearSignalState() {
+    if (status == ProcessStatus::Exited) {
+        LOG_ERROR(Kernel, "called on a terminated process instance.");
+        return ERR_INVALID_STATE;
+    }
+
+    if (!is_signaled) {
+        LOG_ERROR(Kernel, "called on a process instance that isn't signaled.");
+        return ERR_INVALID_STATE;
+    }
+
+    is_signaled = false;
+    return RESULT_SUCCESS;
+}
+
 void Process::LoadFromMetadata(const FileSys::ProgramMetadata& metadata) {
     program_id = metadata.GetTitleID();
     is_64bit_process = metadata.Is64BitProgram();
@@ -137,13 +153,13 @@ void Process::Run(VAddr entry_point, s32 main_thread_priority, u32 stack_size) {
         .Unwrap();
 
     vm_manager.LogLayout();
-    status = ProcessStatus::Running;
+    ChangeStatus(ProcessStatus::Running);
 
     Kernel::SetupMainThread(kernel, entry_point, main_thread_priority, *this);
 }
 
 void Process::PrepareForTermination() {
-    status = ProcessStatus::Exited;
+    ChangeStatus(ProcessStatus::Exiting);
 
     const auto stop_threads = [this](const std::vector<SharedPtr<Thread>>& thread_list) {
         for (auto& thread : thread_list) {
@@ -167,6 +183,8 @@ void Process::PrepareForTermination() {
     stop_threads(system.Scheduler(1).GetThreadList());
     stop_threads(system.Scheduler(2).GetThreadList());
     stop_threads(system.Scheduler(3).GetThreadList());
+
+    ChangeStatus(ProcessStatus::Exited);
 }
 
 /**
@@ -265,7 +283,25 @@ ResultCode Process::UnmapMemory(VAddr dst_addr, VAddr /*src_addr*/, u64 size) {
     return vm_manager.UnmapRange(dst_addr, size);
 }
 
-Kernel::Process::Process(KernelCore& kernel) : Object{kernel} {}
+Kernel::Process::Process(KernelCore& kernel) : WaitObject{kernel} {}
 Kernel::Process::~Process() {}
 
+void Process::Acquire(Thread* thread) {
+    ASSERT_MSG(!ShouldWait(thread), "Object unavailable!");
+}
+
+bool Process::ShouldWait(Thread* thread) const {
+    return !is_signaled;
+}
+
+void Process::ChangeStatus(ProcessStatus new_status) {
+    if (status == new_status) {
+        return;
+    }
+
+    status = new_status;
+    is_signaled = true;
+    WakeupAllWaitingThreads();
+}
+
 } // namespace Kernel
diff --git a/src/core/hle/kernel/process.h b/src/core/hle/kernel/process.h
index 49345aa66..bcb9ac4b8 100644
--- a/src/core/hle/kernel/process.h
+++ b/src/core/hle/kernel/process.h
@@ -14,9 +14,10 @@
 #include "common/bit_field.h"
 #include "common/common_types.h"
 #include "core/hle/kernel/handle_table.h"
-#include "core/hle/kernel/object.h"
 #include "core/hle/kernel/thread.h"
 #include "core/hle/kernel/vm_manager.h"
+#include "core/hle/kernel/wait_object.h"
+#include "core/hle/result.h"
 
 namespace FileSys {
 class ProgramMetadata;
@@ -117,7 +118,7 @@ struct CodeSet final {
     VAddr entrypoint = 0;
 };
 
-class Process final : public Object {
+class Process final : public WaitObject {
 public:
     static constexpr std::size_t RANDOM_ENTROPY_SIZE = 4;
 
@@ -212,6 +213,16 @@ public:
         return random_entropy.at(index);
     }
 
+    /// Clears the signaled state of the process if and only if it's signaled.
+    ///
+    /// @pre The process must not be already terminated. If this is called on a
+    ///      terminated process, then ERR_INVALID_STATE will be returned.
+    ///
+    /// @pre The process must be in a signaled state. If this is called on a
+    ///      process instance that is not signaled, ERR_INVALID_STATE will be
+    ///      returned.
+    ResultCode ClearSignalState();
+
     /**
      * Loads process-specifics configuration info with metadata provided
      * by an executable.
@@ -260,6 +271,17 @@ private:
     explicit Process(KernelCore& kernel);
     ~Process() override;
 
+    /// Checks if the specified thread should wait until this process is available.
+    bool ShouldWait(Thread* thread) const override;
+
+    /// Acquires/locks this process for the specified thread if it's available.
+    void Acquire(Thread* thread) override;
+
+    /// Changes the process status. If the status is different
+    /// from the current process status, then this will trigger
+    /// a process signal.
+    void ChangeStatus(ProcessStatus new_status);
+
     /// Memory manager for this process.
     Kernel::VMManager vm_manager;
 
@@ -305,6 +327,10 @@ private:
     /// specified by metadata provided to the process during loading.
     bool is_64bit_process = true;
 
+    /// Whether or not this process is signaled. This occurs
+    /// upon the process changing to a different state.
+    bool is_signaled = false;
+
     /// Total running time for the process in ticks.
     u64 total_process_running_time_ticks = 0;
 
diff --git a/src/core/hle/kernel/readable_event.cpp b/src/core/hle/kernel/readable_event.cpp
index 92e16b4e6..ba01f495c 100644
--- a/src/core/hle/kernel/readable_event.cpp
+++ b/src/core/hle/kernel/readable_event.cpp
@@ -4,10 +4,10 @@
 
 #include <algorithm>
 #include "common/assert.h"
+#include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/readable_event.h"
 #include "core/hle/kernel/thread.h"
-#include "core/hle/kernel/writable_event.h"
 
 namespace Kernel {
 
@@ -34,6 +34,16 @@ void ReadableEvent::Clear() {
     signaled = false;
 }
 
+ResultCode ReadableEvent::Reset() {
+    if (!signaled) {
+        return ERR_INVALID_STATE;
+    }
+
+    Clear();
+
+    return RESULT_SUCCESS;
+}
+
 void ReadableEvent::WakeupAllWaitingThreads() {
     WaitObject::WakeupAllWaitingThreads();
 
diff --git a/src/core/hle/kernel/readable_event.h b/src/core/hle/kernel/readable_event.h
index 867ff3051..80b3b0aba 100644
--- a/src/core/hle/kernel/readable_event.h
+++ b/src/core/hle/kernel/readable_event.h
@@ -7,6 +7,8 @@
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/wait_object.h"
 
+union ResultCode;
+
 namespace Kernel {
 
 class KernelCore;
@@ -39,8 +41,17 @@ public:
 
     void WakeupAllWaitingThreads() override;
 
+    /// Unconditionally clears the readable event's state.
     void Clear();
 
+    /// Clears the readable event's state if and only if it
+    /// has already been signaled.
+    ///
+    /// @pre The event must be in a signaled state. If this event
+    ///      is in an unsignaled state and this function is called,
+    ///      then ERR_INVALID_STATE will be returned.
+    ResultCode Reset();
+
 private:
     explicit ReadableEvent(KernelCore& kernel);
 
diff --git a/src/core/hle/kernel/svc.cpp b/src/core/hle/kernel/svc.cpp
index e6c77f9db..84df2040e 100644
--- a/src/core/hle/kernel/svc.cpp
+++ b/src/core/hle/kernel/svc.cpp
@@ -1433,17 +1433,24 @@ static ResultCode CloseHandle(Handle handle) {
     return handle_table.Close(handle);
 }
 
-/// Reset an event
+/// Clears the signaled state of an event or process.
 static ResultCode ResetSignal(Handle handle) {
     LOG_DEBUG(Kernel_SVC, "called handle 0x{:08X}", handle);
 
     const auto& handle_table = Core::CurrentProcess()->GetHandleTable();
+
     auto event = handle_table.Get<ReadableEvent>(handle);
+    if (event) {
+        return event->Reset();
+    }
 
-    ASSERT(event != nullptr);
+    auto process = handle_table.Get<Process>(handle);
+    if (process) {
+        return process->ClearSignalState();
+    }
 
-    event->Clear();
-    return RESULT_SUCCESS;
+    LOG_ERROR(Kernel_SVC, "Invalid handle (0x{:08X})", handle);
+    return ERR_INVALID_HANDLE;
 }
 
 /// Creates a TransferMemory object
diff --git a/src/core/hle/service/ldr/ldr.cpp b/src/core/hle/service/ldr/ldr.cpp
index ca119dd3a..453d90a22 100644
--- a/src/core/hle/service/ldr/ldr.cpp
+++ b/src/core/hle/service/ldr/ldr.cpp
@@ -335,10 +335,7 @@ public:
         vm_manager.ReprotectRange(*map_address + header.rw_offset, header.rw_size,
                                   Kernel::VMAPermission::ReadWrite);
 
-        Core::System::GetInstance().ArmInterface(0).ClearInstructionCache();
-        Core::System::GetInstance().ArmInterface(1).ClearInstructionCache();
-        Core::System::GetInstance().ArmInterface(2).ClearInstructionCache();
-        Core::System::GetInstance().ArmInterface(3).ClearInstructionCache();
+        Core::System::GetInstance().InvalidateCpuInstructionCaches();
 
         nro.insert_or_assign(*map_address, NROInfo{hash, nro_size + bss_size});
 
@@ -391,10 +388,7 @@ public:
                                      Kernel::MemoryState::ModuleCodeStatic) == RESULT_SUCCESS);
         ASSERT(process->UnmapMemory(mapped_addr, 0, nro_size) == RESULT_SUCCESS);
 
-        Core::System::GetInstance().ArmInterface(0).ClearInstructionCache();
-        Core::System::GetInstance().ArmInterface(1).ClearInstructionCache();
-        Core::System::GetInstance().ArmInterface(2).ClearInstructionCache();
-        Core::System::GetInstance().ArmInterface(3).ClearInstructionCache();
+        Core::System::GetInstance().InvalidateCpuInstructionCaches();
 
         nro.erase(iter);
         IPC::ResponseBuilder rb{ctx, 2};
diff --git a/src/core/memory.cpp b/src/core/memory.cpp
index 70abd856a..41fd2a6a0 100644
--- a/src/core/memory.cpp
+++ b/src/core/memory.cpp
@@ -53,6 +53,14 @@ void PageTable::Resize(std::size_t address_space_width_in_bits) {
 
     pointers.resize(num_page_table_entries);
     attributes.resize(num_page_table_entries);
+
+    // The default is a 39-bit address space, which causes an initial 1GB allocation size. If the
+    // vector size is subsequently decreased (via resize), the vector might not automatically
+    // actually reallocate/resize its underlying allocation, which wastes up to ~800 MB for
+    // 36-bit titles. Call shrink_to_fit to reduce capacity to what's actually in use.
+
+    pointers.shrink_to_fit();
+    attributes.shrink_to_fit();
 }
 
 static void MapPages(PageTable& page_table, VAddr base, u64 size, u8* memory, PageType type) {
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index b9faaf8e0..5ea094e64 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -1049,6 +1049,7 @@ union Instruction {
         BitField<49, 1, u64> nodep_flag;
         BitField<50, 3, u64> component_mask_selector;
         BitField<53, 4, u64> texture_info;
+        BitField<60, 1, u64> fp32_flag;
 
         TextureType GetTextureType() const {
             // The TEXS instruction has a weird encoding for the texture type.
@@ -1549,7 +1550,7 @@ private:
             INST("1110111011011---", Id::STG, Type::Memory, "STG"),
             INST("110000----111---", Id::TEX, Type::Memory, "TEX"),
             INST("1101111101001---", Id::TXQ, Type::Memory, "TXQ"),
-            INST("1101100---------", Id::TEXS, Type::Memory, "TEXS"),
+            INST("1101-00---------", Id::TEXS, Type::Memory, "TEXS"),
             INST("1101101---------", Id::TLDS, Type::Memory, "TLDS"),
             INST("110010----111---", Id::TLD4, Type::Memory, "TLD4"),
             INST("1101111100------", Id::TLD4S, Type::Memory, "TLD4S"),
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 9e93bd609..2b29fc45f 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -79,6 +79,26 @@ struct DrawParameters {
     }
 };
 
+struct FramebufferCacheKey {
+    bool is_single_buffer = false;
+    bool stencil_enable = false;
+
+    std::array<GLenum, Maxwell::NumRenderTargets> color_attachments{};
+    std::array<GLuint, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> colors{};
+    u32 colors_count = 0;
+
+    GLuint zeta = 0;
+
+    auto Tie() const {
+        return std::tie(is_single_buffer, stencil_enable, color_attachments, colors, colors_count,
+                        zeta);
+    }
+
+    bool operator<(const FramebufferCacheKey& rhs) const {
+        return Tie() < rhs.Tie();
+    }
+};
+
 RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo& info)
     : res_cache{*this}, shader_cache{*this}, emu_window{window}, screen_info{info},
       buffer_cache(*this, STREAM_BUFFER_SIZE) {
@@ -90,9 +110,6 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo
 
     OpenGLState::ApplyDefaultState();
 
-    // Create render framebuffer
-    framebuffer.Create();
-
     shader_program_manager = std::make_unique<GLShader::ProgramManager>();
     state.draw.shader_program = 0;
     state.Apply();
@@ -361,6 +378,44 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
     SyncClipEnabled(clip_distances);
 }
 
+void RasterizerOpenGL::SetupCachedFramebuffer(const FramebufferCacheKey& fbkey,
+                                              OpenGLState& current_state) {
+    const auto [entry, is_cache_miss] = framebuffer_cache.try_emplace(fbkey);
+    auto& framebuffer = entry->second;
+
+    if (is_cache_miss)
+        framebuffer.Create();
+
+    current_state.draw.draw_framebuffer = framebuffer.handle;
+    current_state.ApplyFramebufferState();
+
+    if (!is_cache_miss)
+        return;
+
+    if (fbkey.is_single_buffer) {
+        if (fbkey.color_attachments[0] != GL_NONE) {
+            glFramebufferTexture(GL_DRAW_FRAMEBUFFER, fbkey.color_attachments[0], fbkey.colors[0],
+                                 0);
+        }
+        glDrawBuffer(fbkey.color_attachments[0]);
+    } else {
+        for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
+            if (fbkey.colors[index]) {
+                glFramebufferTexture(GL_DRAW_FRAMEBUFFER,
+                                     GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index),
+                                     fbkey.colors[index], 0);
+            }
+        }
+        glDrawBuffers(fbkey.colors_count, fbkey.color_attachments.data());
+    }
+
+    if (fbkey.zeta) {
+        GLenum zeta_attachment =
+            fbkey.stencil_enable ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT;
+        glFramebufferTexture(GL_DRAW_FRAMEBUFFER, zeta_attachment, fbkey.zeta, 0);
+    }
+}
+
 std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
     const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
 
@@ -444,10 +499,10 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
     UNIMPLEMENTED_IF(regs.rt_separate_frag_data != 0);
 
     // Bind the framebuffer surfaces
-    current_state.draw.draw_framebuffer = framebuffer.handle;
-    current_state.ApplyFramebufferState();
     current_state.framebuffer_srgb.enabled = regs.framebuffer_srgb != 0;
 
+    FramebufferCacheKey fbkey;
+
     if (using_color_fb) {
         if (single_color_target) {
             // Used when just a single color attachment is enabled, e.g. for clearing a color buffer
@@ -463,14 +518,12 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
                 state.framebuffer_srgb.enabled |= color_surface->GetSurfaceParams().srgb_conversion;
             }
 
-            glFramebufferTexture2D(
-                GL_DRAW_FRAMEBUFFER,
-                GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target), GL_TEXTURE_2D,
-                color_surface != nullptr ? color_surface->Texture().handle : 0, 0);
-            glDrawBuffer(GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target));
+            fbkey.is_single_buffer = true;
+            fbkey.color_attachments[0] =
+                GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target);
+            fbkey.colors[0] = color_surface != nullptr ? color_surface->Texture().handle : 0;
         } else {
             // Multiple color attachments are enabled
-            std::array<GLenum, Maxwell::NumRenderTargets> buffers;
             for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
                 Surface color_surface = res_cache.GetColorBufferSurface(index, preserve_contents);
 
@@ -485,22 +538,17 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
                         color_surface->GetSurfaceParams().srgb_conversion;
                 }
 
-                buffers[index] = GL_COLOR_ATTACHMENT0 + regs.rt_control.GetMap(index);
-                glFramebufferTexture2D(
-                    GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index),
-                    GL_TEXTURE_2D, color_surface != nullptr ? color_surface->Texture().handle : 0,
-                    0);
+                fbkey.color_attachments[index] =
+                    GL_COLOR_ATTACHMENT0 + regs.rt_control.GetMap(index);
+                fbkey.colors[index] =
+                    color_surface != nullptr ? color_surface->Texture().handle : 0;
             }
-            glDrawBuffers(regs.rt_control.count, buffers.data());
+            fbkey.is_single_buffer = false;
+            fbkey.colors_count = regs.rt_control.count;
         }
     } else {
-        // No color attachments are enabled - zero out all of them
-        for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
-            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER,
-                                   GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index), GL_TEXTURE_2D,
-                                   0, 0);
-        }
-        glDrawBuffer(GL_NONE);
+        // No color attachments are enabled - leave them as zero
+        fbkey.is_single_buffer = true;
     }
 
     if (depth_surface) {
@@ -508,22 +556,12 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
         // the shader doesn't actually write to it.
         depth_surface->MarkAsModified(true, res_cache);
 
-        if (regs.stencil_enable) {
-            // Attach both depth and stencil
-            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
-                                   depth_surface->Texture().handle, 0);
-        } else {
-            // Attach depth
-            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
-                                   depth_surface->Texture().handle, 0);
-            // Clear stencil attachment
-            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
-        }
-    } else {
-        // Clear both depth and stencil attachment
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
-                               0);
+        fbkey.zeta = depth_surface->Texture().handle;
+        fbkey.stencil_enable = regs.stencil_enable;
     }
+
+    SetupCachedFramebuffer(fbkey, current_state);
+
     SyncViewport(current_state);
 }
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index 988fa3e27..8a891ffc7 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -40,6 +40,7 @@ namespace OpenGL {
 
 struct ScreenInfo;
 struct DrawParameters;
+struct FramebufferCacheKey;
 
 class RasterizerOpenGL : public VideoCore::RasterizerInterface {
 public:
@@ -195,11 +196,12 @@ private:
              OGLVertexArray>
         vertex_array_cache;
 
+    std::map<FramebufferCacheKey, OGLFramebuffer> framebuffer_cache;
+
     std::array<SamplerInfo, Tegra::Engines::Maxwell3D::Regs::NumTextureSamplers> texture_samplers;
 
     static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024;
     OGLBufferCache buffer_cache;
-    OGLFramebuffer framebuffer;
     PrimitiveAssembler primitive_assembler{buffer_cache};
     GLint uniform_buffer_alignment;
 
@@ -214,6 +216,8 @@ private:
 
     void SetupShaders(GLenum primitive_mode);
 
+    void SetupCachedFramebuffer(const FramebufferCacheKey& fbkey, OpenGLState& current_state);
+
     enum class AccelDraw { Disabled, Arrays, Indexed };
     AccelDraw accelerate_draw = AccelDraw::Disabled;
 
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 8d68156bf..4fc09cac6 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -50,6 +50,14 @@ public:
     using std::runtime_error::runtime_error;
 };
 
+/// Generates code to use for a swizzle operation.
+static std::string GetSwizzle(u64 elem) {
+    ASSERT(elem <= 3);
+    std::string swizzle = ".";
+    swizzle += "xyzw"[elem];
+    return swizzle;
+}
+
 /// Translate topology
 static std::string GetTopologyName(Tegra::Shader::OutputTopology topology) {
     switch (topology) {
@@ -1004,14 +1012,6 @@ private:
         }
     }
 
-    /// Generates code to use for a swizzle operation.
-    static std::string GetSwizzle(u64 elem) {
-        ASSERT(elem <= 3);
-        std::string swizzle = ".";
-        swizzle += "xyzw"[elem];
-        return swizzle;
-    }
-
     ShaderWriter& shader;
     ShaderWriter& declarations;
     std::vector<GLSLRegister> regs;
@@ -1343,7 +1343,7 @@ private:
         regs.SetRegisterToInteger(dest, true, 0, result, 1, 1);
     }
 
-    void WriteTexsInstruction(const Instruction& instr, const std::string& texture) {
+    void WriteTexsInstructionFloat(const Instruction& instr, const std::string& texture) {
         // 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
 
@@ -1368,6 +1368,38 @@ private:
         }
     }
 
+    void WriteTexsInstructionHalfFloat(const Instruction& instr, const std::string& texture) {
+        // TEXS.F16 destionation registers are packed in two registers in pairs (just like any half
+        // float instruction).
+
+        std::array<std::string, 4> components;
+        u32 written_components = 0;
+
+        for (u32 component = 0; component < 4; ++component) {
+            if (!instr.texs.IsComponentEnabled(component))
+                continue;
+            components[written_components++] = texture + GetSwizzle(component);
+        }
+        if (written_components == 0)
+            return;
+
+        const auto BuildComponent = [&](std::string low, std::string high, bool high_enabled) {
+            return "vec2(" + low + ", " + (high_enabled ? high : "0") + ')';
+        };
+
+        regs.SetRegisterToHalfFloat(
+            instr.gpr0, 0, BuildComponent(components[0], components[1], written_components > 1),
+            Tegra::Shader::HalfMerge::H0_H1, 1, 1);
+
+        if (written_components > 2) {
+            ASSERT(instr.texs.HasTwoDestinations());
+            regs.SetRegisterToHalfFloat(
+                instr.gpr28, 0,
+                BuildComponent(components[2], components[3], written_components > 3),
+                Tegra::Shader::HalfMerge::H0_H1, 1, 1);
+        }
+    }
+
     static u32 TextureCoordinates(Tegra::Shader::TextureType texture_type) {
         switch (texture_type) {
         case Tegra::Shader::TextureType::Texture1D:
@@ -2766,24 +2798,27 @@ private:
                 const bool depth_compare =
                     instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC);
                 const auto process_mode = instr.texs.GetTextureProcessMode();
+
                 UNIMPLEMENTED_IF_MSG(instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP),
                                      "NODEP is not implemented");
 
                 const auto scope = shader.Scope();
 
-                const auto [coord, texture] =
+                auto [coord, texture] =
                     GetTEXSCode(instr, texture_type, process_mode, depth_compare, is_array);
 
                 shader.AddLine(coord);
 
-                if (!depth_compare) {
-                    shader.AddLine("vec4 texture_tmp = " + texture + ';');
+                if (depth_compare) {
+                    texture = "vec4(" + texture + ')';
+                }
+                shader.AddLine("vec4 texture_tmp = " + texture + ';');
 
+                if (instr.texs.fp32_flag) {
+                    WriteTexsInstructionFloat(instr, "texture_tmp");
                 } else {
-                    shader.AddLine("vec4 texture_tmp = vec4(" + texture + ");");
+                    WriteTexsInstructionHalfFloat(instr, "texture_tmp");
                 }
-
-                WriteTexsInstruction(instr, "texture_tmp");
                 break;
             }
             case OpCode::Id::TLDS: {
@@ -2842,7 +2877,7 @@ private:
                     }
                 }();
 
-                WriteTexsInstruction(instr, texture);
+                WriteTexsInstructionFloat(instr, texture);
                 break;
             }
             case OpCode::Id::TLD4: {
@@ -2940,7 +2975,8 @@ private:
                 if (depth_compare) {
                     texture = "vec4(" + texture + ')';
                 }
-                WriteTexsInstruction(instr, texture);
+
+                WriteTexsInstructionFloat(instr, texture);
                 break;
             }
             case OpCode::Id::TXQ: {